Hepatoblastoma Cancer Stem Cells Express PD-L1, Reveal Plasticity and Can Emerge upon Chemotherapy

Simple Summary Cancer stem cells are thought to cause a poor response to chemotherapies. The aim of our study was to explore the still unknown biology of hepatoblastoma cancer stem cells that are essential for tumorigenesis. With our investigations, we aimed to gain more insight into cancer stem cell characteristics and to provide directions for new therapeutic approaches to treat refractory hepatoblastoma. We showed that hepatoblastoma cancer stem cells express PD-L1, a factor which helps tumors escape immune attacks. Furthermore, we detected cancer stem cell progeny evolving from non-cancer stem cells. Finally, we revealed that another subset of cancer stem cells is induced during chemotherapy. Our findings give a possible explanation why chemotherapies fail in certain hepatoblastoma cases and why new therapeutic approaches should consider the plasticity of hepatoblastoma cancer stem cells. Abstract The biology of cancer stem cells (CSCs) of pediatric cancers, such as hepatoblastoma, is sparsely explored. This is mainly due to the very immature nature of these tumors, which complicates the distinction of CSCs from the other tumor cells. Previously, we identified a CSC population in hepatoblastoma cell lines expressing the CSC markers CD34 and CD90, cell surface Vimentin (csVimentin) and binding of OV-6. In this study, we detected the co-expression of the immune escape factor PD-L1 in the CSC population, whereas the other tumor cells remained negative. FACS data revealed that non-CSCs give rise to CSCs, reflecting plasticity of CSCs and non-CSCs in hepatoblastoma as seen in other tumors. When we treated cells with cisplatin and decitabine, a new CD34+/lowOV-6lowCD90+ population emerged that lacked csVimentin and PD-L1 expression. Expression analyses showed that this new CSC subset shared similar pluripotency and EMT features with the already-known CSCs. FACS results further revealed that this subset is also generated from non-CSCs. In conclusion, we showed that hepatoblastoma CSCs express PD-L1 and that the biology of hepatoblastoma CSCs is of a plastic nature. Chemotherapeutic treatment leads to another CSC subset, which is highly chemoresistant and could be responsible for a poor prognosis after postoperative chemotherapy.


Introduction
Hepatoblastoma is an embryonal tumor that is usually diagnosed before the age of 3 years [1,2]. The International Pediatric Liver Consensus Classification subdivides the tumors by histology in fetal, embryonal, cholangioblastic, macrotrabecular and small-cell undifferentiated (SCUD) hepatoblastoma [3]. These subtypes can be highly heterogeneous with closely intermixed histological components resembling different developmental stages [4][5][6][7]. Complete surgical resection is the mainstay of treatment. The anatomical tumor location, determined by the PRE-and POST-TEXT (Pre-and post-treatment extend

Fluorescence-Activated Cell Sorting
The HuH6 cells were simultaneously stained with a BV421-conjugated anti-CD34 antibody (clone 581, BD Horizon, Franklin Lakes, NJ, USA) and an APC-conjugated OV-6 antibody (R&D Systems, Biotechne, Wiesbaden, Germany). The CD34 and OV-6 doublenegative cells were sorted in a BD Aria Fusion cytometer. The sorting gates were established using cells stained with isotype controls (Brilliant Violet 421 mouse IgG1 isotype control and APC mouse IgG1 isotype control, Biolegend, Koblenz, Germany). After the sorting, the collected CD34 − OV-6 − and CD34 + OV-6 + cells were re-analyzed for CD34 expression and OV-6 binding using the cytometer.

RNA Extraction and Transcription Analysis Using Real-Time PCR
As previously described [16], RNA was extracted from the cells using the Extractme Total RNA Kit (blirt S.A., Gdansk, Poland) according to the manufacturer's protocol. cDNA was synthesized with the iScript cDNA Synthesis Kit (Bio-Rad Laboratories, Inc., Hercules, CA, USA). Real-time PCR analyses were performed using the iTaq Universal SYBR Green Supermix (Bio-Rad Laboratories, Inc., Hercules, CA, USA) with a Stratagene Mx3005P machine (Agilent, Santa Clara, CA, USA). qPCR data were calculated using the mean of two experimental replicates, and all qPCR experiments were repeated at least 3 times. Dissociation curves were generated to confirm the amplification of a single PCR product. All the quantifications were normalized to an endogenous ACTB control and calculated using the 2 −∆∆Ct method. All the primers used spanned introns and are listed in Supplementary Table S1 [27][28][29][30]. Primers were designed using Primer3Plus or retrieved from publications.

Statistical Analysis
The results are presented in diagrams as means and standard deviations. A nonparametric Dunn's multiple comparisons test was performed for comparisons between several groups. The p-values < 0.05 were considered statistically significant.

Results
3.1. CD34 + OV-6 + CD90 + csVimentin + Hepatoblastoma Cancer Stem Cells Are PD-L1 Positive As previously shown [16], we identified a CSC population which co-expressed the cell surface markers CD34 and CD90 and cell surface Vimentin (csVimentin). Additionally, the CSCs were stained by the OV-6 antibody (i.e., the mouse monoclonal antibody OV-6), which recognizes a common epitope in Keratin 14 and Keratin 19 [31]. This population could be distinguished from the rest of the tumor cells, the non-CSCs, which were almost Recent studies observed elevated levels of the immune escape factor PD-L1 on the CSCs of other tumor types. Therefore, we analyzed the hepatoblastoma cell lines, HuH6 and HepG2, for a potential PD-L1 expression using flow cytometry. We detected similar expression levels when compared to CD34, CD90, csVimentin and OV-6 binding. Simultaneous staining with α-CD34 and OV-6 antibodies revealed that PD-L1 was expressed on the same cells as CD34 and the OV-6 antigens and, hence, on the same cells as CD90 and csVimentin as shown in Figure 1 (the corresponding CD34/OV-6 dot blots are shown in Supplementary Figure S1). In conclusion, the CSC subset could be extended using a fifth cell surface marker to become CD34 + OV-6 + CD90 + csVimentin + PD-L1 + .

Statistical Analysis
The results are presented in diagrams as means and standard deviations. A non-parametric Dunn's multiple comparisons test was performed for comparisons between several groups. The p-values < 0.05 were considered statistically significant.

CD34 + OV-6 + CD90 + csVimentin + Hepatoblastoma Cancer Stem Cells Are PD-L1 Positive
As previously shown [16], we identified a CSC population which co-expressed the cell surface markers CD34 and CD90 and cell surface Vimentin (csVimentin). Additionally, the CSCs were stained by the OV-6 antibody (i.e., the mouse monoclonal antibody OV-6), which recognizes a common epitope in Keratin 14 and Keratin 19 [31]. This population could be distinguished from the rest of the tumor cells, the non-CSCs, which were almost entirely CD34 -OV-6 -CD90 -csVimentin -.
Recent studies observed elevated levels of the immune escape factor PD-L1 on the CSCs of other tumor types. Therefore, we analyzed the hepatoblastoma cell lines, HuH6 and HepG2, for a potential PD-L1 expression using flow cytometry. We detected similar expression levels when compared to CD34, CD90, csVimentin and OV-6 binding. Simultaneous staining with α-CD34 and OV-6 antibodies revealed that PD-L1 was expressed on the same cells as CD34 and the OV-6 antigens and, hence, on the same cells as CD90 and csVimentin as shown in Figure 1 (the corresponding CD34/OV-6 dot blots are shown in Supplementary Figure S1). In conclusion, the CSC subset could be extended using a fifth cell surface marker to become CD34 + OV-6 + CD90 + csVimentin + PD-L1 + .  3.2. CD34 − OV-6 − Non-CSCs Give Rise to CD34 + OV-6 + CD90 + csVimentin + PD-L1 + CSCs To elucidate the source of the CSCs, we performed a fluorescence-activated cell sorting (FACS) to obtain CD34 and OV-6 double-negative non-CSCs. Therefore, we stained HuH6 cells with α-CD34 and OV-6 antibodies and sorted for CD34 − OV-6 − cells, as shown in Figures 2A, S2 and S6. We immediately plated the sorted and unsorted cells out in 6-well plates. After 48, 72 and 96 h of cultivation, we analyzed the cells using flow cytometry for CD34 expression and OV-6 binding. As CD34 + OV-6 + cells co-express CD90, csVimentin and PD-L1, CD34 and OV-6 analyses were considered sufficient to detect CD34 + OV-6 + CD90 + csVimentin + PD-L1 + cells. After 48 h, we observed CD34 + OV-6 + CSCs in the CD34 − OV-6 − sample with an average of 6.3% compared to the unsorted control sample with an average of 8.9%, as shown in Figure 2B,C. Interestingly, a certain percentage of cells in the CD34 − OV-6 − sample showed various CD34 low and OV-6 low expression levels. After 72 h, the CD34 − OV-6 − sample had even higher numbers of CD34 + OV-6 + cells, which were almost as high as in the unsorted cells (8.1 versus 9.6%, respectively). A larger difference was apparent after 96 h, when the numbers in the unsorted sample rose to about 17.8%, whereas the numbers in the CD34 − OV-6 − sample only rose to 10.5% on average. 3.2. CD34 -OV-6 -Non-CSCs Give Rise to CD34 + OV-6 + CD90 + csVimentin + PD-L1 + CSCs To elucidate the source of the CSCs, we performed a fluorescence-activated cell sorting (FACS) to obtain CD34 and OV-6 double-negative non-CSCs. Therefore, we stained HuH6 cells with α-CD34 and OV-6 antibodies and sorted for CD34 -OV-6 -cells, as shown in Figures 2A, S2 and S6. We immediately plated the sorted and unsorted cells out in 6well plates. After 48, 72 and 96 h of cultivation, we analyzed the cells using flow cytometry for CD34 expression and OV-6 binding. As CD34 + OV-6 + cells co-express CD90, csVimentin

A CD34 +/low OV-6 low Population Emerges after Treatment with Cisplatin
As CSCs are resistant to chemotherapeutic treatments, we treated HuH6 and HepG2 cells with cisplatin (1, 2.5, 5 and 7.5 µg/mL) for 72 h and analyzed the surviving cells for CD34 expression and OV-6 binding using flow cytometry ( Figure 3A). The dot plots of both cell lines revealed two CD34 + OV-6 + populations: a CD34 +/high OV-6 high population (green line), which was the previously described CSCs, and a new CD34 +/low OV-6 low (red line) population. The new CD34 +/low OV-6 low population became more visible with higher cisplatin concentrations (5 µg/mL for HuH6 and 2.5 µg/mL for HepG2). The number of the CD34 +/low OV-6 low cells strongly rose with higher cisplatin concentrations from 3.3% to 56% on average in HuH6 cells and from 1.1% to 46.6% in HepG2 cells. However, the percentage of the CD34 +/high OV-6 high cells only increased within the initial cisplatin concentrations (1 and 2.5 µg/mL) from 4.7% to 16.7% in HuH6 cells and from 2.6% to 8.6% in HepG2 cells, which showed that the CD34 +/low OV-6 low cells were responsible for the total increase of CD34 + OV-6 + cells ( Figure 3B). In conclusion, we observed a new CD34 +/low OV-6 low subset which seems highly chemoresistant and can be distinguished from the previously identified CSC subset (CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + cells). The histograms depict the ratio of CD34 +/low OV-6 low (low, black bars) and CD34 +/high OV-6 high (high, grey bars) in each cisplatin concentration along with the total CD34 + OV-6 + population (grey line). The columns represent the means with error bars depicting the standard deviation from the mean.

Treatment with Cisplatin and Decitabine in Combination
Results in Increased Numbers of CD34 +/low OV-6 low Cells, Chemoresistant CD34 +/low OV-6 low Cells Are CD90 Positive but csVimentin and PD-L1 Negative Higher concentrations of cisplatin (5 and 7.5 µg/mL) unmask the CD34 +/low OV-6 low population but result in only a few viable cells [16]. Therefore, we searched for another method to obtain a reasonable amount of these cells to investigate their profile. In previous attempts to specifically target CSCs, we treated the cells with decitabine and observed a slight yet promising inhibitory effect on our subset of CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + cells. However, when we added this agent to a cisplatin treatment, the number of cells in the subset did not decrease, but, more interestingly, the number of the newly discovered CD34 +/low OV-6 low CSC subset strongly increased compared to cisplatin mono-treated cells (Supplementary Figure S3). Therefore, we decided that treatment with the hypomethylating agent decitabine would obtain a measurable amount of CD34 +/low OV-6 low cells with the use of moderate cisplatin concentrations. We treated HuH6 with 100nM decitabine and 2 µg/mL cisplatin and HepG2 with 250 nM decitabine and 3 µg/mL cisplatin for 72 h and analyzed the surviving cells for CD34, CD90, csVimentin and PD-L1 expression and OV-6 binding using flow cytometry (Figures 4 and S4). The results again showed two different CD34 + OV-6 + cell populations with CD34 +/low OV-6 low cells ( Figure 4, red solid line) as the majority. Furthermore, we detected CD90 expression on the CD34 +/low OV-6 low cells but did not detect csVimentin or PD-L1 expression, which differs from the CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + CSC subset (green dotted line, also shown in Figure 1). We refer to this new subset as CD34 +/low OV-6 low CD90 + cells.

Treatment with Cisplatin and Decitabine in Combination Results in Increased Numbers of CD34 +/low OV-6 low Cells. Chemoresistant CD34 +/low OV-6 low Cells Are CD90 Positive but csVimentin and PD-L1 Negative
Higher concentrations of cisplatin (5 and 7.5µg/mL) unmask the CD34 +/low OV-6 low population but result in only a few viable cells [16]. Therefore, we searched for another method to obtain a reasonable amount of these cells to investigate their profile. In previous attempts to specifically target CSCs, we treated the cells with decitabine and observed a slight yet promising inhibitory effect on our subset of CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + cells. However, when we added this agent to a cisplatin treatment, the number of cells in the subset did not decrease, but, more interestingly, the number of the newly discovered CD34 +/low OV-6 low CSC subset strongly increased compared to cisplatin mono-treated cells (Supplementary Figure S3). Therefore, we decided that treatment with the hypomethylating agent decitabine would obtain a measurable amount of CD34 +/low OV-6 low cells with the use of moderate cisplatin concentrations. We treated HuH6 with 100nM decitabine and 2µg/mL cisplatin and HepG2 with 250 nM decitabine and 3µg/mL cisplatin for 72 h and analyzed the surviving cells for CD34, CD90, csVimentin and PD-L1 expression and OV-6 binding using flow cytometry (Figures 4 and S4). The results again showed two different CD34 + OV-6 + cell populations with CD34 +/low OV-6 low cells ( Figure 4, red solid line) as the majority. Furthermore, we detected CD90 expression on the CD34 +/low OV-6 low cells but did not detect csVimentin or PD-L1 expression, which differs from the CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + CSC subset (green dotted line, also shown in Figure 1). We refer to this new subset as CD34 +/low OV-6 low CD90 + cells.  3.5. CD34 +/low OV-6 low CD90 + Cells Are Not More Pluripotent and Do Not Show Higher EMT Features Than CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + Cells We investigated the pluripotency and EMT profile of the CD34 +/low OV-6 low CD90 + cells and compared the results to the already examined CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + CSCs. HuH6 cells were treated with 100 nM decitabine and 2 µg/mL cisplatin and HepG2 cells were treated with 250 nM decitabine and 3 µg/mL cisplatin for 72 h. Both sets of treated cells were then enriched for CD34-positive cells using MACS (magnetic-activated cell sorting). As shown in Figure 5A for HuH6 cells, the CD34 enriched fraction (CD34+, green bar) of the control cells had an average of 61.7% CD34 + cells compared to the unsorted control cells, which had an average of 20.3% CD34 + cells (white bar). The CD34+ fraction of the treated cells (DAC+Cis, red bar) had an average of 72.6% CD34 + cells compared to the unsorted treated cells (grey bar), which had an average of 36.7% CD34 + cells. For HepG2 cells, as shown in Figure 5B, the CD34+ fraction (green bar) of the control cells had an average of 65.2% CD34 + cells compared to the unsorted control cells (white bar) with 17.9%. The CD34+ fraction of the treated cells (red bar) had an average of 73.2% CD34 + cells compared to the unsorted treated sample (grey bar) with 38.2%.
We performed qPCR analyses to evaluate the expression of the pluripotency factors Oct4 and Nanog; the CSC markers CD34, KRT14 (one of the OV-6 antigens) and CD90; the EMT transcription factors SNAI1 and Twist1; the proto-oncogene c-myc; and either EpCAM, which is expressed in premature liver cells, or Albumin, which is expressed in mature hepatocytes. For HuH6 cells, the qPCR analyses ( Figure 5E-M) revealed a significant increase in Oct4, Nanog, CD34 and KRT14 for the CD34+ sorted control fraction when compared to the unsorted control and/or the treated DAC+Cis cells. Furthermore, a significant increase in Oct4, CD34 and Twist1 was observed for the CD34+ sorted DAC+Cis fraction compared to the unsorted control and/or the DAC+Cis cells. However, none of the analyzed factors showed a significant difference between the CD34+ sorted fractions of the control and the DAC+Cis cells. For HepG2 cells, the qPCR results ( Figure 5N-V) showed a significant increase in Oct4, Nanog, CD34, KRT14, CD90, SNAI1, Twist 1 and c-myc for the CD34+ sorted control fraction when compared to the unsorted control and/or DAC+Cis cells. The expression of Oct4, Nanog, CD34, KRT14, CD90, SNAI1, Twist1 and c-myc was significantly increased in the CD34+ sorted DAC+Cis fraction compared to the unsorted control and/or the DAC+Cis cells. Again, no significant difference for any of the factors was detected between the CD34+ sorted fractions except for Albumin, which was increased in the CD34+ sorted control fraction. In conclusion, the CD34 +/low OV-6 low CD90 + cells presented in the CD34+ sorted DAC+Cis sample seemed to have increased levels of pluripotency and EMT factors but did not differ significantly in their pluripotency or EMT levels compared to the CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + cells. Thus, CD34 +/low OV-6 low CD90 + cells represent another subset of hepatoblastoma CSCs. 3.6. CD34 − OV-6 − Non-CSCs Give Rise to CD34 +/low OV-6 low CD90 + Cells When Treated with Cisplatin and Decitabine We also explored the origin of the CD34 +/low OV-6 low CD90 + cells. For this analysis, we used FACS to sort for CD34 − OV-6 − cells and then plated the cells out in 6-well plates as shown in Figures 2A and S2. After 24 h, we treated the sorted CD34 − OV-6 − and unsorted cells with 100 nM decitabine and 2 µg/mL cisplatin. Again, we analyzed the cells for CD34 expression and OV-6 binding after 24, 48 and 72 h of treatment using flow cytometry ( Figures 6A and S6). Forty-eight hours after sorting and 24 h after treatment, the numbers of CD34 +/high OV6 high CD90 + csVimentin + PD-L1 + cells were almost identical between the unsorted and CD34 − OV-6 − samples (9.2 versus 9.8% CD34 +/high OV6 high CD90 + csVimentin + PD-L1 + cells, respectively) ( Figure 6B). Throughout the time measurements, both samples showed comparable numbers of CD34 +/high OV6 high CD90 + csVimentin + PD-L1 + cells, which slightly decreased to 8.1% in the unsorted sample versus 8.9% in the CD34 − OV-6 − sample. shown in Figures 2A and S2. After 24 h, we treated the sorted CD34 -OV-6 -and unsorted cells with 100 nM decitabine and 2µ g/mL cisplatin. Again, we analyzed the cells for CD34 expression and OV-6 binding after 24, 48 and 72 h of treatment using flow cytometry (Figures 6A and S6). Forty-eight hours after sorting and 24 h after treatment, the numbers of CD34 +/high OV6 high CD90 + csVimentin + PD-L1 + cells were almost identical between the unsorted and CD34 -OV-6 -samples (9.2 versus 9.8% CD34 +/high OV6 high CD90 + csVimentin + PD-L1 + cells, respectively) ( Figure 6B). Throughout the time measurements, both samples showed comparable numbers of CD34 +/high OV6 high CD90 + csVimentin + PD-L1 + cells, which slightly decreased to 8.1% in the unsorted sample versus 8.9% in the CD34 -OV-6 -sample.   On the contrary, the number of the CD34 +/low OV-6 low CD90 + cells clearly increased in both samples from 6.6 to 19.5% in the unsorted sample and from 14.1 to 24.8% in the CD34 − OV-6 − sample ( Figure 6C). This result suggests that CD34 +/low OV-6 low CD90 + cells can derive from CD34 − OV-6 − non-CSCs and indicates that this CSC subset is indeed more chemoresistant than the CD34 +/high OV6 high CD90 + csVimentin + PD-L1 + CSC subset.

Discussion
It is well known that CSCs use several tools to escape attacks from the immune system. For example, CSCs can decrease the expression of molecules crucial for antigen presentation to T-cell receptors [32]. With PD-L1 expression, which is increased on the CSCs of certain cancers, the cells can suppress the effector functions of tumor-infiltrating lymphocytes [21][22][23][24]. We analyzed PD-L1 in hepatoblastoma cell lines and found that our previously investigated CSC population of CD34 +/high OV-6 high CD90 + csVimentin + cells express PD-L1, whereas non-CSC tumor cells were negative for PD-L1. A few studies have investigated PD-L1 in pediatric cancers, and most detected no or a weak PD-L1 expression on tumor specimens [33][34][35][36]. Only studies on pediatric gliomas revealed a higher detection rate of PD-L1 [37,38]. As shown by others in adult studies and in our present data, PD-L1 is preferentially expressed in CSCs, which represent a minority of the tumor cells. The general low to weak PD-L1 detection in pediatric cancer studies could be explained by the possibility that PD-L1 expression was analyzed in gross tumor specimens and not specifically in the CSC population.
PD-L1 expression analyses of pediatric cancers can be used to assess the potential efficacy of anti-PD-L1 antibody therapy. This therapy has already shown successful treatment results for various adult tumors, but the efficacy is still under investigation for pediatric cancers. Monotherapy trials with only a few pediatric patients have shown rather disappointing results and, consequently, combinatorial strategies were suggested for treatment and are currently being investigated [39][40][41][42]. Previous PD-L1 expression analysis studies only included a few hepatoblastoma specimens. Therefore, further investigations with a focus on hepatoblastoma are needed to clarify the significance of the PD-L1/PD-1 pathway and its possible blockade in this pediatric tumor. Evaluating the PD-L1 expression in refractory SCU hepatoblastomas, which have very immature tumor cells, would provide particularly important information.
The origin and (self-)renewal of CSCs either follow a rigid hierarchical organization in a unidirectional fashion, as shown in glioblastoma, or reveal a certain plasticity in their hierarchy, as shown for colorectal CSCs [43][44][45]. Our FACS sorting analyses detected new CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + cells which derived from non-CSCs. We conclude that hepatoblastoma CSCs reflect the plasticity model, which is bidirectional not unidirectional.
To investigate changes in CSC behavior, we treated the tumor cells with cisplatin and observed that a new CD34 + OV-6 + population emerged on a lower expression level. It is possible that this subset may have existed before cisplatin treatment because CD90enriched populations included this population under untreated conditions. Cisplatin treatment might only unmask the CD34 + OV-6 + population through the vast cell death of non-CSCs, which would allow the percentage among the viable cell population to increase. It is interesting that this CD34 +/low OV-6 low subset outnumbered the already known CSC subset at higher cisplatin concentrations, which indicates an increased chemoresistance. However, it could also be possible that the high numbers were due to an increased proliferation rate. Expression analyses of CD90, csVimentin and PD-L1 showed that the CD34 +/low OV-6 low subset expresses CD90 but lacks csVimentin and PD-L1, which differs from the already investigated CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + CSC population. Concerning the pluripotency and EMT status, we observed a comparable behavior of the CD34 +/low OV-6 low CD90 + population. Therefore, we propose that this cell subset is another CSC population in hepatoblastoma.
As already mentioned, the combinatorial use of cisplatin with decitabine revealed unexpected observations. Previous studies using decitabine observed a reduction of bladder and breast cancer stem cells [46,47]. However, we found that the numbers of CD34 +/low OV-6 low CD90 + CSCs were tremendously increased when decitabine was added to cisplatin. These contrasting results could imply that hypomethylation mediated by this DNA-Methyltransferase-inhibitor does not lead to the predicted effects in pediatric cancers, such as hepatoblastoma. Further evaluation of these trends is needed.
The detection of CD34 +/low OV-6 low CD90 + CSCs using cisplatin treatment raised several questions about the source of this population. Were these CSCs the daughter cells of CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + cells, which are the dominant CSC population under non-treated conditions? Our FACS results revealed that the CSCs were not daughter cells. When treated with cisplatin and decitabine, non-CSCs also gave rise to CD34 +/low OV-6 low CD90 + cells. This observation is supported by several studies showing that chemo-and radiotherapies induce an increased proliferation rate in pre-existing quiescent CSCs and also trigger non-CSCs to undergo de-differentiation and produce therapy-induced CSCs in breast and non-small cell lung cancer [12,[48][49][50][51][52]. One can assume that this plasticity can be also seen in various other tumor types.
Nonetheless, the possibility that CD34 +/low OV-6 low CD90 + cells can also derive from CD34 +/high OV-6 high CD90 + csVimentin + PD-L1 + cells, and vice versa, is of great importance. Additionally, since chemotherapy applies exogenous stress to the cells, it would be interesting to determine if other types of stress, such as hypoxia or nutrient deprivation, can also cause the emergence of CD34 +/low OV-6 low CD90 + CSCs.
Altogether, our in vitro findings show that the plasticity of hepatoblastoma CSCs may complicate new therapeutic approaches. Indeed, our results suggest that targeting intrinsic CSC features will not inhibit differentiated tumor cells from producing a new CSC progeny.

Conclusions
In summary, our findings suggest that CSC biology in hepatoblastoma is plastic and that several different CSC subsets may co-exist, as shown for other cancers [48,[53][54][55]. Given the fact that CSCs were induced by cisplatin treatment, our data reinforce the idea that chemotherapy is a two-edged sword. On the one hand, the vast tumor mass will be reduced. On the other hand, residual cells could convert into new CSCs that are highly chemoresistant and give rise to differentiated daughter cells, which may lead to tumor regrowth. The newly identified CSC population of CD34 +/low OV-6 low CD90 + cells in this study lacks PD-L1 and csVimentin expression, thus eliminating the treatment target for anti-PD-L1 or anti-csVimentin antibodies [56]. This chemotherapy-induced CSC subpopulation may, therefore, be a treatment-resistant population of new hepatoblastoma CSCs [52,57,58]. Despite the difference in PD-L1 and csVimentin expression, both CSC subsets share CD34 and CD90 expression along with OV-6 binding [6,7]. This should be further analyzed as possible targets for hepatoblastoma CSCs. For example, target factors for chimeric antigen receptor (CAR) T-cell approaches is a new immune therapy and is currently under investigation for neuroblastoma in clinical trials [59][60][61][62].

Data Availability Statement:
The data presented in this study are available upon request from the corresponding author.