Aldh1 Expression and Activity Increase During Tumor Evolution in Sarcoma Cancer Stem Cell Populations

Tumors evolve from initial tumorigenic events into increasingly aggressive behaviors in a process usually driven by subpopulations of cancer stem cells (CSCs). Mesenchymal stromal/stem cells (MSCs) may act as the cell-of-origin for sarcomas, and CSCs that present MSC features have been identified in sarcomas due to their ability to grow as self-renewed floating spheres (tumorspheres). Accordingly, we previously developed sarcoma models using human MSCs transformed with relevant oncogenic events. To study the evolution/emergence of CSC subpopulations during tumor progression, we compared the tumorigenic properties of bulk adherent cultures and tumorsphere-forming subpopulations both in the sarcoma cell-of-origin models (transformed MSCs) and in their corresponding tumor xenograft-derived cells. Tumor formation assays showed that the tumorsphere cultures from xenograft-derived cells, but not from the cell-of-origin models, were enriched in CSCs, providing evidence of the emergence of bona fide CSCs subpopulations during tumor progression. Relevant CSC-related factors, such as ALDH1 and SOX2, were increasingly upregulated in CSCs during tumor progression, and importantly, the increased levels and activity of ALDH1 in these subpopulations were associated with enhanced tumorigenicity. In addition to being a CSC marker, our findings indicate that ALDH1 could also be useful for tracking the malignant potential of CSC subpopulations during sarcoma evolution.

We first study how sphere-formation ability evolves during tumor evolution by comparing data obtained from these cell-of-origin models with data obtained from several xenograft-derived cell lines that are able to recapitulate the originally formed sarcoma (T-4H-FC#1, T-5H-GFP#1 and T-5H-FC#1) 38,40 (Fig. 1A; Table S1). All cell-of-origin models [MSC-4H and MSC-5H cell types (MSC-XH)] and tumor cell lines [T-4H and T-5H cell types (T-XH)] can be serially expanded as self-renewed spheres with similar efficiency (Fig. 1B), although T-XH-derived spheres were much larger than those formed by MSC-XH cells ( Fig. 1C and Figure S1). By performing time-lapse microscopy, we could monitor the sphere-formation process. Interestingly, we observed clonal division of T-5H-FC#1 cells combined with aggregation of forming spheres into bigger clusters ( Fig. 1D; Figure  S2; Video S1; Video S2). Similarly, in MSC-5H-FC cells, we observed both clonal tumorsphere formation ( Fig. 1E; Figure S3A; Video S3) and aggregation in the initial steps of the tumorsphere-formation process ( Figure S3B; Video S3). These data indicate that aggregation occurs in low density multi-cell sphere cultures even when media is supplemented with methylcellulose to reduce the mobility of the cells. In any case, a high percentage of spheres were initiated by the clonal division of a single cell and not by the aggregation of two or more cells (Video S1; Video S2; Video S3).
To further confirm the existence of cells that are able to form clonal spheres in these sarcoma models and to estimate their frequency, we performed limiting dilution assays (LDA) to detect tumorsphere formation from 1000, 100, 10 and 1 cell (Fig. 1E). Single-cell assays showed that a high percentage of cells (between 23.0% and 37.9%) were indeed able to initiate clonal growth. Sphere-forming frequency (SFF) calculated using ELDA software was also notably high in all cell types.
CSC subpopulations isolated from sarcomas have been reported to exhibit differentiation potential to MSC related lineages 7,12,41 . We previously found that this collection of sequentially mutated MSCs lost their adipogenic potential during the transformation process, and MSC-4H, MSC-5H, T-5H and T-4H cells (regardless the expression of FUS-CHOP) displayed an impaired pattern of differentiation in which most cells of the culture presented a small amount of lipid droplets in their cytoplasm. In addition, MSC-4H and MSC-5H cells retained their full ability to differentiate toward the osteogenic lineage 40 . We found that tumorspheres derived from all cell types display high osteogenic and low adipogenic potential similar to that observed in the corresponding bulk adherent cultures ( Figure S4). Given that the blockage of the adipogenic differentiation pathways is a hallmark of liposarcoma development 42 , this finding is in line with a liposarcoma-forming potential of these cells when combined with other liposarcoma instructive mutations such as the expression of FUS-CHOP.
Tumorsphere cultures from tumor-derived cells but not from the cells of origin are highly enriched in CSCs. Transplantation assays into immunocompromised mice is the current "gold standard" for identifying CSCs 1 . To check whether tumorsphere cultures of both a cell-of-origin model and their derived cell lines are enriched in tumor-initiating cells able to reproduce the original tumor, we inoculated cells from both adherent and tumorsphere cultures into NOD-SCID mice. We found that the sphere-forming subpopulation of T5H-FC#1 cells more efficiently induces tumor formation in immunodeficient mice than the bulk adherent Scientific RepoRts | 6:27878 | DOI: 10.1038/srep27878 cultures. Thus, significant differences in tumor growth were observed as early as 12 days after cell inoculation, and a 5-fold difference in tumor volume was evidenced at day 22 ( Fig. 2A,B). Otherwise, MSC-5H-FC-formed spheres need a much larger latency period to induce tumor formation, and adherent cultures showed only a slight and nonsignificant delay in tumor formation when compared with the corresponding tumorsphere cultures ( Fig. 2A,B). Importantly, tumorsphere cultures of both MSC-5H-FC and T-5H-FC#1 cells were able to reproduce all the features of the human MRCLS (Fig. 2C), confirming that i) MSC-5H-FC cells act as cell-of-origin for this type of tumor as we previously shown 40 , and ii) T-5H-FC#1 cultures contain CSCs that are able to reproduce all the heterogeneity of the original tumor.
To confirm and quantify the enrichment of CSCs in the tumorsphere cultures of T5H-FC#1 cells, we performed LDA assays comparing adherent and tumorsphere cultures. As shown above, tumorsphere cultures produce a significantly faster tumor growth than adherent cultures at all the assayed cell dilutions (Fig. 2D). Given that adherent cultures also contain CSCs, tumor formation were eventually observed in all cases. Therefore, we used data from the day when the tumorsphere and adherent series start to show significant differences (day 14)  The number of mice that grew tumors at day 14 and total number of inoculated mice for each condition is indicated. TIF was calculated using ELDA software. Error bars represent the standard deviation and asterisks indicate a statistically significant difference between the tumorspheres and adherent cell conditions (* p < 0.05, * * p < 0.001; two-sided Student's t-test). In limiting dilutions assays Pr (>chiSq) value is indicated. formation was detected in the case of adherent cultures. Therefore, an ELDA analysis showed a 22-fold enrichment in tumor-initiating frequency (TIF) in tumorsphere cultures (1 in 5572) versus adherent cultures (1 in 125186) (Fig. 2E). These experiments suggest that tumorsphere cultures from the T-5H-FC#1 sarcoma-derived cell line are highly enriched in CSCs. On the other hand, their cell-of-origin cell line (MSC-5H-FC), although able to grow as floating spheres (most likely due to their intrinsic properties and not only by their tumorigenic status), could not be significantly enriched in subpopulations with increased tumorigenicity by tumorsphere culture.

Increased expression of SOX2 and ALDH1 in CSC subpopulations during tumor progression.
According to gene expression experiments, a western blotting analysis confirmed that SOX2 and ALDH1A1 protein levels were upregulated in T-5H-FC#1 tumorspheres at a greater level than in MSC-5H-FC tumorspheres ( Fig. 4A). Moreover, the immunofluorescence analysis of adherent cultures and tumorspheres that were allowed to attach to the substrate before fixation showed that T-5H-FC#1 adherent and tumorsphere cultures presented a significantly higher percentage of SOX2-positive stained nuclei than the respective nuclei of MSC-5H-FC cultures, with the nuclei of T-5H-FC#1 tumorspheres displaying the higher levels ( Fig. 4B,C). Similar results were obtained after an immunofluorescence analysis of ALDH1A1 expression (cytosolic + nuclear) (Fig. 4D,E). Furthermore, in situ simultaneous immunofluorescent staining of SOX2 and ALDHA1 confirmed that T-5H-FC#1 tumorspheres displayed a higher proportion of cells presenting nuclear expression of SOX2 and a higher expression of cytosolic ALDH1A1 compared with MSC-5H-FC tumorspheres (Fig. 4F). Nevertheless, both populations are not totally overlapping, and there are subsets of cells expressing high levels of nuclear SOX2 but not high ALDH1A1 (Fig. 4F, blue arrows) and vice versa (white arrows).
To correlate the increased expression of ALDH1 with its enzymatic activity, we performed an ALDEFLUOR assay with adherent and tumorsphere cultures. Because this assay is based on the generation of a green fluorescent compound, we used the original MSC-5H cells (MSC-5H-O) and a cell line derived from a MSC-5H-O-generated xenograft (T-5H-O), neither transduced with GFP-expressing lentiviral vectors. We first checked the protein levels of ALDH1A1 and ALDH1A3, another ALDH1 isoform that has been reported to contribute to ALDEFLUOR activity 43 . These two isoforms were upregulated at greater levels in T-5H-O tumorspheres than in MSC-5H-O tumorspheres (Fig. 4G), as was seen when comparing MSC-5H-FC and T-5H-FC#1 cells. Similarly, the ALDEFLUOR assay showed that T-5H-O cells displayed higher levels of activity than MSC-5H-O cells in both adherent and especially tumorsphere cultures (Fig. 4H,I).
To study the contributions of ALDH1A1 and ALDH1A3 to the ALDEFLUOR activity, we performed siRNA knockdown of both isoforms in T-5H-O cells. Western blotting confirmed similar knockdown efficiency of both proteins (Fig. 5A). ALDEFLUOR assays showed that ALDH1A1 and ALDH1A3 depletion inhibited the activity by 30% and 70%, respectively, indicating that both isoforms contribute to the ALDEFLUOR activity, although ALDH1A3 seems to be the largest contributor (Fig. 5B,C).
Altogether, these results show that SOX2, ALDH1A1 and ALDH1A3 expression, together with their associated ALDEFLUOR activities are progressively enhanced in CSC subpopulations during sarcoma progression toward more aggressive phenotypes.

ALDH1 high cells displayed increased tumorigenic properties.
To determine whether the cells presenting high ALDH1 activity are also enriched with tumor-propagating properties, we used the ALDEFLUOR assay to isolate ALDH1 high and ALDH1 low by flow cytometry (Fig. 6A). As expected, the ALDH1 high fraction of T-5H-O cells retained most of the expression of ALDH1A1 and ALDH1A3 proteins. Likewise, SOX2 expression was highly increased in the ALDH1 high fraction (Fig. 6B). Similar results were observed after the analysis of ALDH1 high and ALDH1 low populations in MSC-5H-O cells ( Figure S5A). On the other hand, KLF4, which was upregulated in tumorsphere cultures of MSC-5H-FC cells, was not enriched in ALDH high sorted populations of MSC-5H-O or T-5H-O cells ( Figure S5B), similar to previous observations in osteosarcoma samples 44 .
Notably, the ALDH1 high fraction in T-5H-O cells presented a higher ability to form colonies in soft agar, a surrogate in vitro transformation assay (Fig. 6C). In contrast, ALDH1 high and ALDH1 low populations showed similar abilities to form serially passaged tumorspheres (results from two separate experiments shown in Fig. 6D and Figure S6A). To further analyze the self-renewal and tumorigenic properties of both populations, second-passage tumorsphere cultures from ALDH1 high and ALDH1 low cells were disaggregated and assayed form colony formation in soft agar. Again, ALDH1 high cells displayed enhanced cell growth in these conditions ( Fig. 6E and Figure  S6B). Moreover, the ALDEFLUOR assay performed with the cells recovered from the in vitro transformation Scientific RepoRts | 6:27878 | DOI: 10.1038/srep27878 assay reveled that ALDH1 high cells maintain high levels of ALDH1 activity while the ALDH1 low cells were moderately enriched in ALDH1 activity after tumorsphere culture ( Fig. 6F and Figure S6C).
Finally, we found that as few as 400 ALDH1 high cells were sufficient to develop tumors in vivo with 100% of incidence (n = 6). The immunohistological analysis showed that these tumors resembled the original SCS histology and that only a small portion of tumor cells (< 5%) expressed high levels of ALDH1A1 (Fig. 6G). These results indicate that high activity of ALDH1 is strongly related to increased tumorigenic properties and that ALDH1 high cells behave as true CSC subpopulations able to regenerate all the tumor cell subpopulations.

Discussion
To gain insights about the evolution of CSC subpopulations during tumor progression in sarcomas, we compared the tumorsphere-forming subpopulations derived from MSC-XH cells (cell-of-origin models) vs. those derived from their corresponding tumor xenograft-derived T-XH cells (Fig. 1A) 38,40 . We observed that both MSC-XH and T-XH cells formed clonal tumorspheres with very high efficiency. In this regard, it is likely that the intrinsic self-renewal properties of the MSCs 45,46 acting as cell-of-origin for sarcomas could result in higher frequencies of tumorsphere formation upon tumoral transformation. In fact, high frequencies of tumorsphere formation have previously been reported in some cases of sarcoma 12 . Notably, certain cautions have been raised about the use of tumorsphere cultures to enrich CSC subpopulations in some cases. Thus, using imaging approaches, neurospheres were observed to frequently aggregate even at low densities 47 . Similarly, using time-lapse microscopy, we showed that in low-density multi-cell tumorsphere cultures, aggregation occurs even in the presence of methylcellulose. In any case, our experiments also demonstrate that clonal proliferation takes place before and after sphere aggregation both in MSC-5H-FC and T-5H-FC#1 cells. Importantly, we found remarkable differences in the ability of these clonal/aggregated tumorsphere cultures from these two cell types to effectively enrich in tumor-initiating populations (Table 1). On the one hand, tumorsphere cultures from the T-5H-FC#1 sarcoma-derived cell line are highly enriched in CSCs as seen by their highly increased capacity to initiate tumor formation in vivo. On the  By analyzing the expression of several genes involved in the CSC phenotype in MSC-5H-FC and T-5H-FC#1 adherent and tumorsphere cultures, we found a group of genes highly enriched in T-5H-FC#1 vs. MSC-5H-FC tumorsphere cultures that are therefore progressively increased in tumorsphere-forming subpopulations during sarcoma progression. Among these genes, ALDH1A1 and SOX2 were the most highly upregulated in T-5H-FC#1 tumorspheres and may be meaningful indicators of CSC progression in sarcomagenesis and also useful for the prospective isolation of CSCs. Both factors were also upregulated at protein levels and their ALDEFLUOR activities were equally enhanced in T-5H vs. MSC-5H tumorspheres (Table 1). In line with this finding, the upregulation of the mRNA levels of ALDH1A1 23 and SOX2 12,18,19,23 were already detected in tumorsphere cultures of sarcomas.
It is important to note that apart from ALDH1A1, other members of the ALDH superfamily, especially ALDH1A3, seem to contribute to ALDEFLUOR activity 43,48 . We confirmed that, similar to ALDH1A1, ALDH1A3 was upregulated in tumorsphere cultures. Likewise, cells with high ALDEFLUOR activity retain most of the ALDH1A1 and ALDH1A3 expression, and both isoforms contribute importantly to ALDEFLUOR activity in our models. In any case, we could not discard a potential role for other members of the ALDH family. ALDH1 activity has been used to isolate CSC subpopulations in different types of sarcomas 23,[27][28][29][30][31] . These studies found that subpopulations with high activity of ALDH1 showed increased expression of pluripotency markers like SOX2, enhanced ability to grow as tumorspheres, increased tumorigenicity and strong chemo-resistance. In line with these findings, we found that SOX2 expression was highly increased in ALDH1 high cells. However, we observed that both ALDH1 high and ALDH1 low populations are able to grow as tumorspheres, indicating that other markers or combinations of markers are needed to discriminate tumorsphere-forming populations. In any case, ALDH1 activity was always enriched in these growth conditions and ALDH1 high cells possess high self-renewal potential as indicated by their ability to grow as serially passaged spheres that maintain high ALDH1 activity. Importantly, we also found that ALDH1 activity in T-5H cells was associated with enhanced tumorigenic properties. Therefore, our findings correspond to previous studies suggesting that ALDH1 seems to constitute a valuable CSC marker in sarcomas. Moreover, we present for the first time evidence that ALDH1 expression and activity is increased during tumor evolution in CSC subpopulations (T-5H vs. MSC-5H tumorsphere cultures), suggesting that the level of ALDH1 could be used as an indicator of the evolution of the CSCs' malignant potential.
Similar to ALDH1, our results show that SOX2 is upregulated in CSC subpopulations during tumor evolution. Importantly, the depletion of SOX2 has been reported to induce a significant decrease in the tumor-initiating capability of ALDH1 high cells in melanoma 37 . We found that SOX2 was highly enriched in ALDH high populations, although certain level of SOX2 expression was detected in ALDH low cells. In addition, immunofluorescence analysis of tumorspheres that are enriched in ALDEFLUOR activity showed cells expressing different levels of SOX2. Altogether, these data suggest that the selection of SOX2 + /ALDH high cells could result in a better discrimination of CSC subpopulations. As commented before, live cells with high ALDH1 activity can be easily sorted using the ALDEFLUOR assay. Unfortunately, the sorting of live cells based on the level of SOX2 expression cannot yet be achieved despite some promising technical advances 49 . Once this technology becomes available, it would be able to separate subpopulations presenting high activity or expression of both ALDH1 and SOX2 and could potentially represent tumor-propagating subpopulations better than the cell subset solely selected by ALDH1 activity.
Beside ALDH1A1 and SOX2, other molecules such as DKK1 and NOTCH1 were notably upregulated in CSCs during tumor progression. Notably, the WNT-antagonist DKK1 has been proposed to enhance pro-tumorigenic properties in osteosarcoma, in part through the upregulation of ALDH1A1 50 . Likewise, NOTCH signaling has been associated with ALDH activity and increased metastatic potential in osteosarcoma cells 51 .
In conclusion, our model of sarcomagenesis initiated from transformed BM-MSCs, allowing us to study the evolution of the tumor from the cell-of-origin toward increasingly aggressive xenograft-derived cells. By comparing the tumorigenic properties of the tumorsphere-forming populations derived from both populations, we evidenced the emergence of bona fide CSC subpopulations during tumor progression. Several factors related with the CSC phenotype, such as ALDH1A1, SOX2, DKK1 and NOTCH1, were increasingly upregulated in the emerging CSC subpopulations during tumor progression. Importantly, the increased levels of ALDH1 expression and activity in these subpopulations was associated with enhanced tumorigenicity, thus confirming the suitability of this molecule to be used as a CSC marker for sarcomas and also suggesting that ALDH1 could be useful for tracking the malignant potential of CSC subpopulations during tumor evolution and treatment.

Methods
Cell types. Human BM-MSCs sequentially mutated with up to 6 oncogenic events, and tumor lines derived from transformed BM-MSC-induced xenografts were previously generated and characterized (Table S1) [38][39][40] . The identity of transformed human BM-MSCs has been authenticated by a Short Tandem Repeats analysis during the last 5 months. All the cell types were cultured as previously described 40,52 . All experimental protocols have been performed in accordance with institutional review board guidelines and were approved by the Institutional Ethics Committee of the Hospital Universitario Central de Asturias. All samples from human origin were obtained upon signed informed consent.
Tumorsphere culture. Cells lines were plated at a density of 5,000 cells per well in 6-well plates ( After 10 days, the number of wells presenting spheres were counted and the sphere-forming frequency (SFF) was calculated using the ELDA software 53 . Tumorsphere formation was monitored using a Zeiss Cell Observer Live Imaging microscope (Zeiss, Thornwood, NY) coupled with a CO2 and temperature-maintenance system. Time-lapse images were acquired every 8 hours during 6 days using a Zeiss AxioCam MRc camera.

Analysis of CSC-related genes using RT-PCR arrays. The Human Cancer Stem Cells RT2 Profiler PCR
Array (PAHS-176-Z; SABiosciences, Qiagen Iberia, Madrid, Spain) was used to analyze the expression of 84 genes linked to CSCs properties according to the manufacturer instructions. Only RNA samples with RNA integrity ≥ 9 were used in the analysis as determined using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). The RT-qPCR reactions were performed for three independent experiments of each condition in a StepOnePlus Real Time PCR system (Life Technologies, Carlsbad, CA) using the RT 2 First Strand Kit and the RT 2 qPCR SYBR Green/ROX MasterMix (SABiosciences). The PCR conditions included an initial denaturation step at 95 °C for 10 minutes, followed by 40 cycles of a denaturation step at 95 °C for 15 seconds and an annealing/extension step at 60 °C for 1 minute. A final dissociation curve was generated to verify that a single product was amplified. Reactions in the absence of template and in the absence of enzyme were also included as negative controls. A complete data set, including gene information and experimentally obtained C t values, is presented in Table S2. The RT-qPCR raw data were analyzed using the PCR Array Data analysis template (SABiosciences) and software available online at the SABiosciences website: www.sabiosciences.com/pcr/arrayanalysis.php, which provides a statistical analysis of data. Genes presenting amplification Ct values > 35 in both untreated and treated cells were discarded. Relative expression values of the different genes were calculated from the threshold cycle (Ct) following the DDCt method using ACTB, B2M, GAPDH, HPRT1 and RPLP0 as reference genes. Tumor volumes for all mice in each xenograft-treatment group were averaged to obtain the mean tumor volume for the corresponding group. Student's t-test was performed to determine the statistical significance between control and treated groups. Animals were sacrificed when the tumors formed from tumorsphere cultures reached approximately 1,000 mm 3 . Relative tumor-initiating frequency (TIF) was calculated using the data obtained at the day that differences in tumor volume between tumorsphere and adherent groups became statistically significant (day 14 after inoculation) using the ELDA software 53 . Upon removal, tumor samples were fixed in formol, embedded in paraffin, cut into 4-μ m sections, and stained with hematoxylin and eosin (H&E) and anti-ALDH1A1 [(ab105920), 1:400 dilution] from Abcam (Cambridge, UK) as previously described 54 . All experimental protocols were carried out in accordance with the institutional guidelines of the University of Oviedo and were approved by the Animal Research Ethical Committee of the University of Oviedo prior to the study.
Western blot. Whole cell protein extraction and western blot analysis were performed as previously described 55 . The antibodies used in Western blot analysis were as follows: anti-ALDH1A1 [(ab105920) at room temperature (RT) and permeabilized in PBS containing 0.2% triton X-100 (Sigma) for 5 min at 4 °C. Cells were then blocked with PBS containing 1% BSA and 0.05% Tween 20 (PBT) for 5 min at RT and incubated with 1:200 diluted mouse monoclonal anti-ALDH1A1 (ab105920 from Abcam,) or 1:250 diluted rabbit monoclonal anti-SOX2 (PA1-094 from Thermo Fisher) for 24 hours, washed 3 times with PBS and incubated with 1:300 anti-mouse alexa fluor 555 (A-21422 from Thermo Fisher) or 1:300 anti-rabbit alexa fluor 555 (A-21428 from Thermo Fisher) for 1 hour in the dark. Antibody dilutions and washes after incubations were conducted in PBT. Coverslips were finally mounted in Vectashield mounting medium with DAPI (Vector Laboratories Inc.; Burlingame, Ca).
In Confocal section and projection images collected with identical exposure times were obtained using an Ultra-Spectral TCS-SP2-AOBS confocal microscope (Leica, Wetzlar, Germany). Quantification of positive nuclei (SOX2) or nuclei + cytoplasm (ALDH1A1) in 2D experiments were performed in 10-12 randomly selected fields using Confocal Uniovi Image J software 1.5.1 (University of Oviedo, Spain; http://www.sct.uniovi.es/confocaluniovi) based in the Image J software (National Institute of Health, Bethesda, MD).
Aldefluor assay. ALDH activity was determined using the activated Aldefluor TM reagent, a fluorescent non-toxic substrate for ALDH1 able to freely diffuse into intact and viable cells (Stem Cells Technologies, Grenoble, France). 1 × 10 6 cells were suspended in 1 ml of Aldefluor assay buffer containing the ALDH1 substrate (Bodipy-Aminoacetaldehyde) and incubated for 45 min at 37 °C. As a reference control, the cells were suspended in buffer containing the substrate in the presence of diethylaminobenzaldehyde (DEAB; 20 μ M for the experiments described in Fig. 4 and 10 μ M for the rest of the experiments), a specific ALDH1 enzyme inhibitor. Of note, the lower concentration of DEAB failed to fully inhibit high ALDEFLUOR activities (e.g., those observed in tumorsphere cultures of T-5H-O cells) as previously reported 48 . In any case, gates were established to include less than 1% positive cells in the DEAB controls of the condition displaying less ALDEFLUOR activity in adherent or tumorsphere cultures and were maintained in all conditions. Cells were incubated with 0.5 μ g/ml propidium iodide for 15 min, and cells positive for this staining (dead cells) were excluded from the analysis ( Figure S7). The brightly fluorescent ALDH1-expressing cells (ALDH1 high ) were detected and sorted using the green fluorescence channel (520-540 nm) of a MoFlo XDP flow cytometry (Beckman Coulter, Brea, CA). Soft agar colony formation assay. A soft agar colony formation assay was carried out using the CytoSelect TM 96-Well Cell Transformation Assay Kit (Cell Biolabs Inc, San Francisco, CA) as described 57 . Quantification of anchorage-independent cell growth was performed after agar layer solubilization and cell recovering according to the manufacturer's instructions. Cell viability was analyzed using the Cell Proliferation Reagent WST-1 (Roche, Mannheim, Germany). Briefly, recovered cells were plated in 24-well plates, left to attach to the plastic substrate (4-6 hours) and incubated for 1 hour at 37 °C with 18 μ l/ml of WST-1 reagent. The color change produced by the cleavage of the WST-1 reagent by mitochondrial dehydrogenases was measured by reading the absorbance at 440 nm with the use of a Synergy HT plate reader (BioTek, Winooski, VT). The surviving fraction at each treatment was calculated as a percentage of untreated cells. Two independent experiments were performed in triplicate.