Background
Melanoma is a malignancy that arises from uncontrolled proliferation and metastasis of neoplastic melanocytes. Normally, melanocytes are located in the most basal epidermal layer, attached to a collagen-IV rich basement membrane, which separates epidermal and dermal compartments of the skin [
1,
2]. Malignant melanoma is the most lethal form of skin cancer [
3]. Its incidence continues to increase each year and is currently responsible for more than 80% of deaths from skin cancer [
4]. The majority of melanoma mutations are C > T transitions caused by ultraviolet light irradiation [
5], which mostly affect the mitogen-activated protein kinase pathway [
6‐
9]. When diagnosed in its early ‘non-tumorigenic’ stages, resection of the lesion results in very high survival rates [
10]. In this period, which is also termed as radial growth phase [
11], pigmented patches of skin (nevi) increase laterally in size and become palpable, but melanoma cells typically still reside within the epidermis and are not metastasis competent [
12]. Nonetheless, already at this point, they affect cellular behavior in their local environment. For example, neoplastic melanocytes induce hyperproliferation and impair differentiation of keratinocytes [
13]. Once further mutations have mediated metastasis competence, the lesion becomes ‘tumorigenic’ and enters the vertical growth phase, during which the morphology of nevi often switches from plaque to balloon-like [
12]. Melanoma cells lead to breakdown of the basement membrane, massively invade the dermal and hypodermal compartments and metastasize to distant organs. Surgery is then no longer sufficient and the disease becomes much more challenging to treat [
3,
14,
15]. Treatment options for late stage melanomas include kinase inhibitors and immunotherapies like the BRAF inhibitor vemurafenib and the anti-cytotoxic T-lymphocyte antigen-4 antibody ipilimumab [
16‐
18]. However, monotherapy is unlikely to yield a long-term benefit due to multi-drug resistance and, therefore, combination therapies with different targeted and immunotherapies as well as standard chemotherapeutics are being evaluated [
19‐
22]. In melanoma cells, ATP-binding cassette (ABC) transporters, in particular of type ABCB5, were found to mediate resistance to the chemotherapeutics doxorubicin and temozolomide [
23,
24]. Although ABCB5 is present in several human tissues, it is highly abundant in melanocyte progenitors, melanoma cell lines, and melanoma biopsies [
23,
25‐
28]. Furthermore, its expression correlates with tumor progression and metastasis competence [
29].
For the reasons of simplicity, convenience, and cost, in vitro studies on melanoma are often performed in 2D-cell culture assays. However, gene expression is significantly different between 2D and 3D melanoma cultures, likely affecting the signaling exerted by and the sensitivity to drugs of melanoma cells [
30]. Furthermore, the interactions between different cell types of tumor and stroma are difficult to model in 2D. Thus, to better mimic the in vivo situation, different 3D-cell culture approaches with several degrees of complexity have been developed, including spheroids, tumorospheres, human skin equivalents, and melanoma-on-chips assays [
31] as well as xenografts of human melanoma spheroids in rodent recipients [
32]. Such formats are very useful for basic and applied melanoma research, but the currently existing models are either composed of only melanoma cells or they are so complex that the behavior of individual cell types is difficult to understand, and often they are then hard to establish and expensive. In the present work, we describe a novel, simple spheroid-based melanoma model composed of fibroblasts, keratinocytes, and melanoma cells. It allows to track cellular behavior in a cell-type specific manner and recapitulates different characteristics of early melanoma stages. The different cell types arranged into a collagen-IV rich fibroblast core, a ring of keratinocytes, and groups of highly proliferating melanoma cells on the outside. Some melanoma cells were also regularly found to invade the fibroblast core. While in the absence of melanoma cells the keratinocyte ring stratified into central basal-like and peripheral, more differentiated cells, addition of melanoma cells clearly reduced keratinocyte differentiation. Treatment with the cytostatic drug, docetaxel, which has been primarily tested for combination therapy of melanoma [
33‐
35], restored keratinocyte differentiation and ablated external melanoma cells. The few remaining external melanoma cells, however, showed a significantly increased amount of ABCB5-immunoreactivity.
Methods
Cell culture
The human fibroblast cell line CCD-1137Sk (ATCC® CRL-2703™) was cultured in Iscove’s Modified Dulbecco’s Medium (IMDM), with L-Glutamine, supplemented with 10% fetal bovine serum (Sigma), and 1% Penicillin Streptomycin (Capricorn). The human keratinocyte cell line HaCaT (CLS order no. 300493) and the human malignant melanoma cell line SK-MEL-28 (CLS order no. 300337) were cultured in Dulbecco’s Modified Eagle Medium (DMEM) High Glucose (4.5 g/l), with L-Glutamine, with Sodium Pyruvate (Capricorn) supplemented with 10% fetal bovine serum, and 1% Penicillin Streptomycin. Cells were maintained at 37 °C in 5% CO2. Cell lines were obtained in 2016 and repeatedly authenticated by phenotypic analysis, including expression of collagen IV for CCD-1137Sk, establishment of a CK10/CK14 gradient in 3D for HaCaT, and high proliferation rate for SK-MEL-28. Mycoplasma tests using the MycoAlert™ Mycoplasma Detection Kit (Lonza) were routinely performed to ensure mycoplasma-free cell cultures.
3D spheroid cultures and docetaxel treatment
Spheroids were prepared using 96- and 384-well cell repellent plates (Greiner). For mono-culture spheroids, fibroblasts (10,000 cells/well) and HaCaT cells (20,000 cells/well) were seeded. For skin bi-cultures, 10,000 cells of each, fibroblasts and keratinocytes, were used per well, and HaCaT cells were added three days after formation of the fibroblast core. Mono- and bi-cultures were cultured for seven days. For tri-culture spheroids, fibroblasts (10,000 cells/well) were seeded. After three days, HaCaT (10,000 cells/well) and SK-MEL-28 cells (2500 cells/well) were added simultaneously. To distinguish between the different cell lines, CellTracker Fluorescent Probes (Life Technologies) were used. Before adding cells to the 3D co-culture, HaCaT cells were labeled with CellTracker Red CMPTX dye (Life Technologies, C34552) and SK-MEL-28 cells were labeled with CellTracker Green CMFDA (Life Technologies, C2925), each for a time period of 45 min according to the CellTracker manuals. Another two days later, tri-culture spheroids were treated with 100 nM docetaxel or 0.01 ‰ of DMSO as control for 15, 24, 48, and 72 h, respectively. Stock solutions (10 mM) of docetaxel (Sigma) were prepared in dimethylsulfoxide (DMSO). After treatments, spheroids were normally fixed and immunostained as described below. For some experiments, spheroids were transferred to 3D agarose molds (Sigma, Z764051) on day five after seeding in cell repellent plates. Treatment with DMSO or 100 nM docetaxel for 72 h, as well as fixation and cryosectioning were then carried out in the molds.
Immunofluorescence
Immunostaining of spheroids used the following steps. Spheroids were collected in an Eppendorf tube, washed once with PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na
2HPO
4 × 2 H
2O, 2 mM KH
2PO
4, pH 7.4), and fixed with 4% wt/vol paraformaldehyde in PBS at room temperature for 30 min. Then, spheroids were incubated overnight at 4 °C in 15% sucrose in PBS, followed by an incubation overnight at 4 °C in 25% sucrose (Roth, 4621.1) in PBS, before they were embedded in OCT (Leica Biosystems). A CM-1950 cryostat (Leica Biosystems) was used for preparing 10-μm-thick sections. 3D molds were washed once with PBS and fixed with 4% wt/vol paraformaldehyde in PBS at room temperature for 30 min. Next, molds were embedded in OCT and cut with a cryostat into 20-μm-thick sections. All sections were permeabilized with 0.1% Triton X-100 (Roth, 3051.4) in PBS, blocked with 3% BSA (Roth, 8076.3) in PBS, and stained with rabbit anti-Ki67 (Merck, AB9260), rabbit anti-cleaved caspase 3 (CAS3)(Cell Signaling, 9661), rabbit anti-cytokeratin 10 (CK10)(Thermo Fisher Scientific, PA5–32459), rabbit anti-collagen IV (Rockland, 600–401-106S), mouse anti-cytokeratin 14 (CK14)(Merck, MAB3232), or mouse anti-ABCB5 (3C2-1D12 [
29]; and Thermo Fisher Scientific, MA5–17026) antibodies, followed by goat anti-rabbit Alexa Fluor 647 (Invitrogen, A21246), goat anti-mouse Alexa Fluor 555 (Invitrogen, A21424), or donkey anti-mouse Alexa Fluor 647 (Invitrogen, A31571) secondary antibody labeling. Nuclei were stained with Dapi (Sigma, 10,236,276,001). Finally, sections were washed with PBS and mounted with Mowiol (Roth, 0713.2) for confocal microscopy (SP8, Leica).
Statistical analysis
Images were composed using Adobe Illustrator (Adobe Systems Software) and ImageJ. All numeric data were handled using Microsoft Excel 2013 and were subsequently incorporated into the Adobe Illustrator composite. Quantitative analysis of Ki67-, CAS3-, and CK10-positive cells was performed using ImageJ. Graphs are presented as mean ± SEM and statistically analyzed using one-way ANOVA with post-hoc Tukey HSD Calculator or Student t-test. P-values are indicated as * < 0.05, ** < 0.01.
Discussion
3D in vitro models of melanoma are increasingly used to study drug efficacy and mode of action as well as drug combinations. Compared to classical two-dimensional cell cultures, 3D models are thought to better represent a series of parameters that are critical for cancer cell behavior, including substrate stiffness, cell-cell interactions, distribution of oxygen and waste products, as well as drug diffusion [
36]. Currently existing 3D models are mostly tuned for either simplicity and high-throughput, complexity and similarity to the in vivo situation, or personalized medicine [
31,
32,
36‐
43]. Here, we aimed to set up an early-stage 3D melanoma model that would allow to investigate several relevant drug-induced processes in a quantitative and cell-type specific manner. Yet, it should be also fast, easy, and robust in performance. The solution presented in this study is an easy to handle spheroid-based model composed of melanoma cells and major cellular components of a stroma-like environment, i.e. human fibroblasts and keratinocytes. To avoid batch-to-batch variability and to render the system cost effective, we opted for the established cell lines SK-MEL-28, HaCaT, and CCD-1137Sk. Within these constraints, the model was found to reliably mimic melanoma cell invasion into the dermal compartment, drug-induced selection of ABCB5-expressing melanoma cells, and loss of melanoma-induced keratinocyte differentiation.
As for the latter, optimal differentiation of keratinocytes in vitro leading to human skin equivalents (HSE), typically requires the use of primary cells and multifactorial external control, including pH and Ca
2+ gradients as well as air lift [
38,
41,
42]. Given that these operations are time consuming and difficult to generate in high numbers, we avoided such complex maneuvers and allowed HaCaT cells to automatically stratify on top of a fibroblast core. Compared to HSE models, we observed a partial differentiation pattern that included stratification into lower and upper strata expressing either CK14 or CK10, respectively, but lacked a cornified layer. However, while HSE models often come with generation times of several weeks [
44], the spheroid-based tri-culture was ready for use after only seven days. It is arguable, whether the observed stratification really reflects differentiation or if pre-differentiated cells migrated to the outer regions of the spheroid. Yet, we observed an interesting related feature that is also known from human melanoma. Indeed, melanoma cells were reported to influence the differentiation pattern of human epidermal keratinocytes in vivo, i.e. that it leads to a loss of CK10 in hyperplastic regions [
13]. Consistent with this, we found that melanoma cells also decreased CK10 expression by HaCaT cells in our tri-culture spheroids (Figs.
2 and
6). Notably, such loss of CK10 expression mostly occurred in direct vicinity of melanoma cells and keratinocyte differentiation was restored upon treatment with docetaxel which led to apoptosis of external melanoma cells (Fig.
6).
Another interesting feature of the present model was the division of melanoma cells into two populations, i.e. external and internal. The finding of internal SK-MEL-28 cells suggested their invasion into the fibroblast core and this would fit to the fact that this cell line is from the metastatic phase of melanoma [
37] and known to rapidly migrate downwards through the skin [
45]. In general, it would be interesting to further explore these cultures as a simple test system for antimigratory effects of diverse drugs. Apart from these future prospects, the identification of two melanoma cell pools was also interesting, because both pools showed differential behavior in at least three characteristics. First, external melanoma cells, which were located on the outside of the spheroids and thus in direct contact to keratinocytes, tended to form growing aggregates. Conversely, internal melanoma cells, which were found in the fibroblast core, were typically solitary and did not coalesce (Fig.
2). Apart from the first day after adding the keratinocyte-melanoma cell mixture to the fibroblast core, melanoma cells were hardly ever found in the HaCaT ring but always in the fibroblast core (Additional file
5: Figure S5). The second clear difference between internal and external melanoma cells was their response to docetaxel. While external cells massively went into apoptosis and became loose, internal melanoma cells remained apparently unaffected. Their numbers were stable even after 72 h of treatment and the relative amounts of apoptotic and proliferating cells was unaltered. It would be interesting to know whether such differential behavior was due to limited access of the drug to the spheroid core or rather due to cell-specific differences. For example, it could be that only drug-resistant cell subpopulations, which are frequently observed in malignant melanoma [
46], were able to invade the fibroblast core or whether some cellular signaling within the core would have led to drug insensitivity. In any case, it was intriguing to observe that docetaxel changed proliferation and apoptosis apparently only in melanoma cells but not in keratinocytes or fibroblasts. A third difference between external and internal melanoma cells was related to their expression of the ATP-dependent transporter protein, ABCB5. Based on ABCB5 immunofluorescence signal intensity profiles, docetaxel led to higher ABCB5 immunofluorescence signals in the external but not the internal melanoma cells (Fig.
7 and Additional file
4: Figure S4). We do not think that internal melanoma cells would have been unable to increase ABCB5 expression, because – contrary to our expectation – they typically showed lower ABCB5 signals than external ones before treatment. Thus, it could again be that internal melanoma cells either represented a special subpopulation of cells, or they were not exposed to sufficient amounts of the drug, or their local microenvironment impaired such drug-induced changes in gene expression. However, the observed effect of docetaxel on ABCB5 signals in external melanoma cells is compatible with either, an up-regulation of ABCB5 in weakly expressing cells or selection of strongly expressing cells. In general, our results fit nicely to previous studies, which reported that ABCB5 expression is enhanced in malignant melanoma [
47], that it has a functional role in tumor growth [
48], and that chemotherapy leads to the selection of ABCB5-expressing cells [
24].
For the present study, docetaxel was used as a test substance. Although mitogen-activated protein kinase pathway inhibitors and immunotherapies against the immune checkpoints cytotoxic T lymphocyte-associated antigen and programmed death 1 have largely replaced classical alkylating and cytostatic chemotherapeutics as first-line treatment [
20,
22], the mitotic inhibitor paclitaxel and its derivative docetaxel [
49] are being considered as adjuvant treatments [
19,
21,
22] and explored for use in novel formulations (see e.g. [
34,
35,
50,
51]). Given that our model in its current version lacks immune cells, immunotherapies were not in the focus of this study. While the addition of T-cells and other immune cell components to future adaptations of the present 3D tri-culture might be valuable ideas to follow, we here concentrated on the effects of a classical agent on melanoma cells and on their chemoresistance features. With respect to effective drug concentrations, significant effects on external SK-MEL-28 survival were observed after 48 h at 100 nM of docetaxel. In comparison, SK-MEL-28 cells cultured in 2D appeared much more susceptible to docetaxel treatment (Additional file
2: Figure S2). This is in agreement with previous studies, which found a maximal effect of docetaxel at around 10–20 nM on different 2D melanoma cell cultures [
52] and a generally altered sensitivity of cells grown in 2D versus 3D [
53‐
55].
In comparison with the present study, other three-dimensional melanoma spheroid models using, for example, the liquid overlay method [
56] are only composed of one cell type, the melanoma cells. Thus, they do not aim to represent the stromal environment of a tumor. On the other hand, HSE models are often generated by seeding primary fibroblasts in collagen type I followed by simultaneous seeding of primary keratinocytes together with melanoma cells [
57] or by seeding melanoma cells with primary fibroblasts to embed both cell types in the collagen type I matrix [
44]. This method spontaneously forms melanoma nests. Therefore, numbers and sizes of such nests might vary between individual skin reconstructs. As a consequence, it is often difficult to quantitatively validate these models and to predict therapeutic impacts. Conversely, the present tri-culture spheroid model always formed very similar sizes of spheroids with a highly reproducible arrangement of the different cell types allowing reliable quantification of cellular drug effects. Next, skin-on-a-chip models can be performed under a controlled perfusion of growth factors or nutrients [
58]. This cannot be realized in a static spheroid-based system as presented here. Using a skin-on-a-chip platform, Abaci and co-workers demonstrated that the cancer drug, doxorubicin, may have direct toxic effects on keratinocyte proliferation and differentiation [
59]. However, this platform is not suitable for high-throughput screening. For this purpose, simple spheroid models might be more appropriate. Given that the tri-culture spheroid model is composed of both, stroma and tumor cells, it is also possible to test general cell toxicity of a drug by evaluating the effect on surrounding non-transformed cells [
60].
On a technical note, the docetaxel treatment of tricultures led to a consistent loss of external melanoma cells. However, this was only true when the cultures were transferred after treatment from the spheroid formation plate into another container for washing (Fig.
5). If docetaxel treatment, washing, and embedding were carried out altogether without any transfer, presumably all – dead and alive – cells were still present in the immediate vicinity of the spheroids. Although it cannot be completely excluded that the observed difference in cell numbers was due to a distinct effect of the drug in the agarose mold versus the plastic plate, the most straight forward explanation appears to be that many of the external melanoma cells became loose upon drug treatment and were lost during the transfer from one container to the next due to mechanical shear force (see Additional file
1: Figure S1B for schematic illustration). This finding might be of general interest, because similar mechanisms of treatment-induced cell loss could possibly also occur in other spheroid or organoid models. If phenotypic quantitative analysis of either culture size, cell number, or fraction of apoptotic or proliferating cells is used, this effect could easily lead to erroneous data interpretation. Clearly, further investigation in that direction would be useful.