Inducing Differentiation of Premalignant Hepatic Cells as a Novel Therapeutic Strategy in Hepatocarcinoma

Hepatocellular carcinoma (HCC) is resistant to chemotherapeutic regimens and is the second leading cause of cancer-associated deaths worldwide (1). In 28% to 50% of all HCC markers of hepatic progenitors and tumor-initiating cells (TIC) are expressed. Although their origin is still under debate, targeting TICs represents a promising therapeutic approach.

In this work, we took advantage of the characterized Cullin3 (Cul3) and Cul3/p53 knockout (ko) mouse model, in which loss of Cul3 leads to massive expansion of hepatic progenitors. Importantly, loss of Cul3 expression was detected in a large series of human liver cancers and correlated with tumor de-differentiation (2).

Cul3 is a multiprotein ubiquitin ligase, which controls the degradation of several proteins, including cyclin E (3). In vivo, hepatic Cul3 ko progenitors accumulate cyclin E and subsequent DNA damage, which triggers the p53 dependent activation of the senescence checkpoint to block transformation. Simultaneous loss of p53 and Cul3 causes transformation of the differentiating hepatic progenitors into TICs (2).

Based on the accumulation of DNA damage Cul3/p53 ko cells differentiate into premalignant cells and into TICs. Premalignant Cul3/p53 ko cells are CD34⁺ and express the liver progenitor marker alpha-1-fetoprotein (AFP). Cells that differentiate downregulate CD34 accumulate DNA damage and are transformed into TICs upon loss of p53. Thus, TICs are CD34⁻ express hepatic stem cell markers and tumor-initiating markers such as CD133 and CK19 (2, 4, 5). Cells isolated from livers of Cul3 ko and Cul3/p53 ko mouse models mimic the described in vivo phenotype (2). We, therefore, used these cells to identify differences in signaling pathways regulated by chemokines.

Cul3 ko and Cul3/p53 ko cells were described previously (2). HepT1 and Huh6 were provided by J.F. HepG2 and Huh7 cells were obtained from ATCC.

Authentication: HuH-7; JCRB 0403, HEP-G2; DSMZ ACC 180. HepT1, Huh6 were identified by their morphology and marker expression. All cell lines were treated with BM Cyclin (Roche) and tested for mycoplasma using DAPI and nested PCR.

Cells were seeded at a density of 2.5 × 10⁵ cells per 6 cm plate and treated for 48 to 72 hours. Concentrations of all used chemicals were determined by titration and were as follows: 2 μg/mL doxorubicin, 7.5 μmol/L Rad001 (Novartis AG), and 0.013041 pg/cell/mL mCXCL1 (based on Luminex-analysis; Tocris) for Cul3/p53 ko. For hHCC cells, concentrations were as follows: doxorubicin 2 μg/mL, 7.5 μmol/L Rad001 (Novartis AG), and 0.015 pg/cell/mL of hCXCL8 (Tocris). After doxorubicin treatment prior to harvesting, cells were washed twice with PBS to remove free doxorubicin and dead cells. Cells were trypsinized, collected in 0.1% FCS/PBS, and used for analysis.

Cells were seeded at a density of 3.4 × 10⁴ cells/well on coverslips [12-well plate (BD Transduction Lab), coated with 0.1% gelatin]. Fourteen-hour after incubation with mCXCL1 and Rad001, cells were stained for CD34 (BioLegend, 1:100). After incubation for 30 minutes, cells were washed and an anti-rat Alexa Fluor 555 (Invitrogen, dilution 1:200) antibody was used to detect CD34⁺ cells. Cells were re-incubated in the appropriate medium for 8.5 hours and stained for CD34 receptor, which was detected using an anti-rat-FITC antibody (Dianova, dilution 1:200). As this labeling is a transient labeling because the Cy3-Fluorophore is degraded over time, after fixation, permeabilization and DAPI staining quantification of cells that internalized the stained CD34-Cy3 receptor was done using Zeiss and Leica microscopes (Apotome, DM5000).

CD34 stain was performed as described above. After fixation, cells were permeabilized with 0.1% Triton X-100 and incubated 30 minutes at room temperature with Aurora B antibody (Abcam) at 1:100 dilution. Then, coverslips were washed and incubated with anti-rabbit FITC antibody? (Dianova, dilution 1:200) for 30 minutes at room temperature. Cells were covered after washing and counterstained with DAPI as described above. Cells were imaged using a Leica DM 5000 microscope. Cells in which Aurora B staining was located to the cytokinetic furrow were analyzed for the expression of CD34 marker.

Cells were harvested and resuspended in flow buffer (PBS, 1% BSA, 1 mmol/L EDTA). After three washes, and blockade of unspecific binding sites (2% BSA/PBS), cells were stained with primary antibodies. Measurements were done (80,000–500,000 live cells) on FACSCalibur and LSRII (BD Biosciences) flow cytometer (488 nm, 635 nm) equipped with four fluorescence photomultiplier filters.

RNA isolation of treated or doxorubicin sorted cells was performed buffer according to the manufacturer's instructions (PeqGold; Roche). The removal of DNA from RNA was performed using DNAse (Life Technologies) according to the instructions. cDNA synthesis was performed using “RevertAid H Minus Strand cDNA Synthesis Kit” (Invitrogen) according to the instructions.

The detection of target gene mRNA levels was performed by qRT-PCR analysis with an ABI Prism 7000 Sequence Detection System (Applied Biosystems) and qPCR Maxima Mastermix (Fermentas) as previously described. Primer sequences can be obtained upon request.

Statistical analysis was performed with SigmaStat and GraphPad Prism 4 (2004). Additional and detailed methods are described in the Supplementary Information.

Tumor cells secrete inflammatory cytokines that affect cell signaling, self-renewal, and treatment resistance (6). To determine whether Cul3/p53 ko cells show a different secretory profile in comparison to Cul3 ko (control) cells, we performed a comparative ELISA-based Luminex assay on the supernatants of untreated cells. Cul3/p53 ko cells secreted significantly higher amounts of mCXCL1 (Fig. 1A). To determine the effect of increased mCXCL1 level, cells treated with mCXCL1 were analyzed by qRT-PCR analysis. An upregulation of self-renewal and stem cell maintenance-associated genes such as sox2, bmi1, and nanog (7) was detected (Fig. 1B). Expression of D site of albumin promoter binding protein (dbp), a liver-enriched transcriptional protein (8) was found to be downregulated, suggesting a dedifferentiated state (Fig. 1B). Furthermore, a flow-cytometry analysis of Cul3/p53 ko cells treated with mCXCL1 was performed and the expression of stem cell–related markers such as CD34, Sca-1, c-kit, and CD44 (4) was determined. We found that only the CD34⁺ subpopulation accumulated significantly under the mCXCL1 treatment (Fig. 1C and D and Supplementary Fig. S1A). The CD34⁺ cells expressed liver-associated markers (albumin, dbp, hn-factors) at lower levels than adult hepatocytes (Supplementary Fig. S1B, data represented as absolute Ct value normalized to its housekeeping GAPDH); and increased levels of hepatic stem cell markers such as CD44, CK14, CK18, CK19, as well as Sca-1 (Supplementary Fig. S1B; refs. 4, 9). We continued to examine the cellular state of CD34⁺ cells in more detail by comparing CD34⁺ to CD34⁻ cells using qRT-PCR analysis: CD34⁻ cells showed a decreased expression of bmi1 and oct4 as well an increased expression of albumin (Supplementary Fig. S1C). These results confirmed our previous studies showing that CD34⁻ cells are more differentiated (2). Furthermore, not more than 1,000 CD34⁺ cells sufficed to promote tumor growth with tendency toward decreased tumor incidence of 62.5% as compared to 75% in CD34⁻ cells (Supplementary Fig. S1D).

Because mCXCL1 has been described as a mitogen for CD133⁺ liver TICs (10), we further investigated the impact of mCXCL1 on the proliferation of CD34⁺ cells. We measured bromdesoxyuridin (BrdUrd) uptake by flow cytometry. CD34⁻ cells showed a significantly increased proliferation upon mCXCL1 stimulation (Fig. 1E). In contrast, BrdUrd uptake of CD34⁺ cells remained unchanged. These results suggested that the accumulation of CD34⁺ cells was not caused by increased proliferation.

The elevated numbers of CD34⁺ cells could also be caused by a different distribution of CD34 on dividing cells. Based on this hypothesis, we would expect a depolarization of CD34. The dividing cell would produce two CD34⁺ daughter cells upon mCXCL1 treatment instead of one CD34⁺ and one CD34⁻ daughter cell. Mitotic cells were identified by a combination of CD34 and Aurora B staining. We could not detect any changes in distribution of CD34 on dividing cells upon mCXCL1 treatment (Fig. 1F).

To summarize, mCXCL1 exerts opposing cellular functions on cells of different cellular states. Although CD34⁻ TICs (CD133⁺, CK19^+/; ref. 2) show increased proliferation CD34⁺ cells accumulate in their more undifferentiated cellular stage under mCXCL1 treatment.

We developed a single-cell approach based on an immunofluorescence cell-tracking assay to determine why CD34⁺ cells accumulated under mCXCL1 treatment. This assay is based on the labeling of CD34⁺ cells by the internalization of the Cy3 stained CD34 receptor and a second staining of the CD34 receptor using FITC-tagged antibodies (Fig. 2A and Supplementary Fig. S2A and S2B). The method, however, did not allow to determine the de novo synthesis of CD34 receptor. We quantified those cells showing intracellular red-fluorescence and determined re-exposure of the FITC-labeled CD34 receptor after 9 hours. Cells that showed an intracellular red fluorescent staining (1st CD34 stain) and also an extracellular FITC staining (2^nd CD34 stain) kept their stem cell phenotype. Single intracellular red stained cells (1st CD34 stain) differentiated. Quantification (Fig. 2B) revealed that mCXCL1 treatment led to a significant increase of undifferentiated cells of up to 16% (Fig. 2C).

Immunofluorescence staining was performed using anti-CK14 or anti-CK19 antibodies instead of anti-CD34 antibody in the second staining. CK14 is expressed on both CD34⁺ and CD34⁻ populations (2). CK19 is predominantly expressed on differentiating CD34⁻ cells (2). A continuous expression of CK14 on CD34⁺ cells and a significant decrease in CD34⁺/CK19⁺ cells upon mCXCL1 treatment was found (Fig. 2D and E).

Thus, we concluded that mCXCL1 promotes the maintenance of the cellular state of CD34⁺ premalignant stem/progenitor cells by preventing their differentiation.

To understand how maintenance of the premalignant stem/progenitor state of CD34⁺ cells is induced by mCXCL1, the activation of signaling molecules downstream of the mCXCL1 receptors CXCR1 and CXCR2 was analyzed (11). Both receptors activate a broad array of intracellular signal transduction pathways including mTORC1, p38MAPK, PI3K-, and ERK signaling (12). Using a proteome profiler, an increased phosphorylation of AKT-2/-3 (13), CREB (14), JNK-3 (15), and p70-S6K (14) was found (Supplementary Fig. S3A). Because mTORC1 signaling represents a central signaling molecule within the signaling pathways of the identified kinases, the phosphorylation of the downstream located 4EBP-1 as an indicator of mTORC1 activity using Western blotting was determined (16). A slight but significant change in p4EBP-1 levels under mCXCL1 treatment was detected (Fig. 3A). The latter results could be confirmed by flow cytometry analysis showing that only a small subpopulation of proliferating Cul3/p53 ko cells are positive for p4EBP-1 (Fig. 3B), whereas adult hepatocytes do not show a phosphorylation of 4EBP-1 (Supplementary Fig. S3B).

As part of a functional analysis, the percentage of CD34⁺ cells under mTORC1 blockade (Rad001) was detected and found to be significantly decreased in comparison to untreated cells (Fig. 3C). mTORC1 blockade was confirmed by Western blotting against phospho-S6 ribosomal protein (pS6RP) and 4EBP-1 (Fig. 3C). Propidium iodide stainings showed that the decrease of CD34⁺ cells was not caused by increased cell death under Rad001 (Supplementary Fig. S3C). We also tracked single CD34⁺ cells (compare to Fig. 2) and found a significant decrease of double positive (CD34Cy3/FITC) cells upon mTORC1 blockade (Fig. 3D). The differentiating effect could not be rescued by simultaneous treatment with mCXCL1, indicating that mCXCL1 and mTORC1 act in the same signaling cascade (Fig. 3E and Supplementary Fig. S3D).

Thus, we concluded that mCXCL1 mediates stem cell maintenance via mTORC1 activation.

To test if an association between mCXCL1 and stem cell maintenance could also be detected in vivo, a HCC mouse model was analyzed. In this mouse model, tumorigenesis is based on an increased expression of lymphotoxins (LTα and LTβ) and downstream target genes, mimicking the inflammatory signature of virus-infection induced (hepatitis B and hepatitis C) chronic liver inflammation (17). As the mice show an increased expression of mCXCL1 (17), we asked whether a correlation in the mTORC1 kinase activation in early hepatic progenitor cells of AlbLTαβ mice could be found. CD34⁺ cells are located in the unaffected liver tissue in close proximity to the tumor or in the tumor tissue (Fig. 3F). Quantification of CD34⁺/pS6RP⁺ cells (Fig. 3G) indicated a significant increase in mTORC1 kinase activity in CD34⁺.

Therefore, the inflammatory-driven HCC model supported our in vitro findings, demonstrating that increased mCXCL1 levels correlate with the activity of the mTORC1 complex in CD34⁺ cells.

Next, the importance of the mCXCL1/mTORC1 cascade in chemotherapy resistance was analyzed. Tumorigenic hepatic progenitors were treated with the chemotherapeutic drug doxorubicin. Doxorubicin-treated cells displayed higher pS6RP levels than untreated cells (Fig. 4A). Flow cytometry of doxorubicin-treated cells revealed three different populations (doxorubicin negative, intermediate, and positive cells) identified by their specific PE (doxorubicin) intensity (Fig. 4B). Cells in these populations differed in their forward scatter (size) and sideward scatter (granularity; data not shown). We sorted three doxorubicin populations and measured their cell size. Doxorubicin negative (PE⁻) cells showed a significantly smaller cell size (Fig. 4C), which could be indicative for an undifferentiated cellular state. Flow cytometry analysis demonstrated a significant increase of CD34⁺/p4EBP-1⁺ population among doxorubicin intermediate and negative cells (Fig. 4D), which correlated with an overall accumulation of p4EBP-1 [doxorubicin negative: 3.1% ± 1.6%; doxorubicin intermediate: 26% ± 8% (n = 4)].

The interference with the mCXCL1–mTORC1 pathway by activating with mCXCL1 or by blocking it, with the mTORC1 inhibitor Rad001, significantly increased (mCXCL1) or decreased (Rad001) the number of doxorubicin-resistant cells as measured by flow cytometry (Fig. 4E and Supplementary Fig. S4A). In addition, mCXCL1 treatment significantly increased the percentage of CD34⁺ doxorubicin negative cells (Fig. 4F).

Expression analysis in three enriched doxorubicin populations revealed an increase of mCXCL1 expression that correlated with an enhanced expression of nanog, bmi1, and sox2 (Fig. 4G, also compare Fig. 1B). Furthermore, we performed cell-cycle analysis using DAPI and found that doxorubicin-escaping cells accumulated in G₁-phase of the cell cycle (Fig. 4H and Supplementary Fig. S4B).

In concordance with our previous data we concluded that chemotherapy resistance is mediated by the mCXCL1–mTORC1 axis, a phenotype that is associated with a nonproliferate cellular state.

To test whether these findings might be of clinical relevance, four different hHCC cell lines with different genetic background were analyzed. Expression analysis of cells treated with doxorubicin showed an upregulation of hCXCL8, the human homolog of mCXCL1 (Fig. 5A). Furthermore, doxorubicin negative and intermediate cells also accumulated in G₁ phase (Fig. 5B) of the cell cycle. p4EBP-1 accumulation could also be detected in the doxorubicin negative and intermediate population in three out of four cell lines (Fig. 5C and Supplementary Fig. S5A and S5B).

Finally, mTORC1 was blocked and the percentage of doxorubicin negative cells was analyzed. Blockade of mTORC1 signaling under doxorubicin treatment led to decreased numbers of escaping cells, however, not in Huh6 cells (Fig. 5D and Supplementary Fig. S5B). Interestingly, the decrease in resistant cells correlated with an increased autofluorescence of doxorubicin (Fig. 5E) indicative for a reduced ability to escape the chemotherapeutic treatment.

In human HCC, the accumulation of stem cell markers and TIC markers such as CD44, CD90, and CD133 correlate with the de-differentiation of the tumor (18), of which CD90 and CD44 accumulated under doxorubicin treatment in Huh7 cells (Supplementary Fig. S6A). CD133 and CD44 have been also described as hepatic (19) and precancerous markers (20).

Interestingly, the percentage of CD44⁺ cells in Huh7 cells correlated with the increased hCXCL8 levels in comparison to HepG2 cells (Supplementary Fig. S6B and S6C). They induced the observed phenotype (Supplementary Fig. S6D). Upon doxorubicin treatment and mTORC1 blockade, mainly CD44⁺ cells showed a significant decrease (Fig. 6A and B and Supplementary Fig. S6E).

To underpin the induction of differentiation as a possible therapeutic approach, we hypothesized that a combinational approach is more efficient than a chemotherapeutic treatment alone. Thus, Huh7 tumors were established and treated using doxorubicin, Rad001, or a combination of both. Single treatment slowed down the tumor growth. The combination therapy significantly reduced tumor growth (Fig. 6C). Comparing hematoxylin and eosin stainings of control or single and combinational treated tumors revealed increase in necrotic tissue (Fig. 6D). Therefore, Western blotting was performed to compare CD44 and AFP in untreated to treated samples and a correlation of these markers with p4EBP-1 intensity was found (Fig. 6E, compare also Fig. 6D).

We concluded that the mCXCL1/hCXCL8–mTORC1 signaling pathway mediates chemotherapy resistance by activation of stem cell maintenance correlated with nonproliferative cellular state. This approach is of translational importance, as mTORC1 blockade under chemotherapy treatment significantly reduces the percentage of resistant cells and is likely to improve the survival of cancer patients.

HCC is characterized by the presence of TICs, whose origin is still under debate (5, 21, 22). Here, we elucidated a new resistance mechanism specific to premalignant hepatic cells. We found a significant increase in mCXCL1 secretion in Cul3/p53 ko cells in comparison to their nontumorigenic counterpart (Cul3 ko cells). mCXCL1 is the mouse homolog of hCXCL8 (23). Increased hCXCL8 levels correlate with poor prognosis (24), increased angiogenesis, growth, and self-renewal of TICs (10). In our work, we underpin the important function of hCXCL8 and its proposed use as prognostic factor. The establishment of a single cell tracking assay, a method that does not address de-differentiation, allowed us to determine that mCXCL1/hCXCL8 activates cellular quiescence and blocks the differentiation in dependence of the mTORC1 kinase. Other chemokines whose pathways are crosslinked to mTORC1 (25), might also support the here described resistance phenotype. Our data point to a parallel mechanism to the recently published G_Alert phase of adult stem cells (26). Upon cellular stress such as an exposure to chemotherapy, cells enter a G_Alert phase and maintain stemness by well-known pluripotency factors (7, 27, 28).

Interestingly, mTORC1 blockade did not only deplete surviving cells but significantly reduced the percentage of doxorubicin intermediate cells, indicating that mTORC1 activity is the most important to achieve resistance.

Our study implies a hierarchical order of TICs in hHCC. However, detailed studies of markers and characteristics of TICs and their relation have not been performed.

The importance of our study is underlined by the finding that mTORC1 activity has recently been shown to also mediate sorafenib resistance (29). Another study showed that increased expression of pluripotency factors is also induced by cyclophosphamide, a drug that blocks angiogenesis (30). These studies point to more general resistance phenotype and are in line with our data.

Although further studies have to be performed to confirm resistance mechanisms using different mouse models, the results of our study underline the importance of combinational therapeutic approaches.

No potential conflicts of interest were disclosed.

Conception and design: B. Wolf, U. Kossatz-Boehlert

Development of methodology: B. Wolf, S. Hauser, U. Kossatz-Boehlert

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Wolf, K. Krieg, C. Falk, H. Keppeler, T. Biedermann, E. Schmid, S. Warmann, J. Fuchs, D. Thiele, M. Nieser, L. Schöls, S. Hauser, M. Ringelhahn, U. Kossatz-Boehlert

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B. Wolf, M. Avci-Adali, M. Heikenwalder, U. Kossatz-Boehlert

Writing, review, and/or revision of the manuscript: B. Wolf, K. Breuhahn, T. Biedermann, S. Warmann, M. Nieser, M. Avci-Adali, L. Schöls, T. Yevsa, M. Heikenwalder, U. Kossatz-Boehlert

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Keppeler, T. Biedermann, E. Schmid, S. Warmann, J. Fuchs, S. Vetter, D. Thiele, M. Nieser, M. Avci-Adali, Yu. Skokowa, L. Schöls, M. Heikenwalder, U. Kossatz-Boehlert

Study supervision: J. Fuchs, U. Kossatz-Boehlert

We thank N. Malek for critically reading the manuscript and discussing the data. The authors thank M. Schwab for support, helpful discussion, and critically reading the manuscript. We thank Cornelia Grimmel from the Core Facility of the Interdisciplinary Center for Clinical Research of the University Hospital of Tübingen.

This work was supported by a grant from the German Research Foundation (KO 3999/1-2 to U. Kossatz-Boehlert). Heiknwälder Mathias was supported by the ERC starting Grant “LiverCancerMech,” an ERC Consolidator Grant “HepatoMetaboPath,” the Stiftung für “Biomedizinische Forschung” (Hofschneider foundation), and the Preclinical Comprehensive Cancer Center (PCCC) Helmholtz Alliance. T. Yevsa is funded by the German Research Foundation (DFG, YE 151/2-1).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.