We have characterized growth, differentiation and genome-wide mRNA expression patterns for a large panel of normal, non-transformed and prostate cell lines in Matrigel, covering all classic and many novel PrCa cell lines 
. The development of miniaturized and cost-effective 3D models enabled us to monitor growth, maturation, invasion and motility of prostaspheres in real-time and high resolution, by combined live cell and confocal microscopy. These models will facilitate higher-throughput compound screens in 3D, allowing quantitative measurement of growth, size, shape, cellular dynamics and morphology of acinar structures. Recent research activities have mainly focused on the role of stem/progenitor cell populations in spheroids 
, reviewed in 
. With very few exceptions 
, these studies refer to prostaspheres cultured under anchorage-independent conditions, lacking any contact to ECM 
. In contrast, our differentiation-related models showed essentially no enrichment of stem cell markers. It is clear and expected that lrECM primarily supports differentiation, but we were surprised that Matrigel is able to trigger normal-like epithelial differentiation programs even in PrCa cell lines that have been in vitro
culture for over three decades. This essentially confirms the concepts formulated by Mina Bissell two decades ago, that context and in particular tumor environment matters and may powerfully override 
malignant genotypes. However, our experimental data show that repression of the tumorigenic phenotype may also be only temporarily.
The specific aim of this study was a detailed analysis of various different modes of growth, migration and invasion of normal and prostate cancer cells, and the identification of small-molecule inhibitors that may specifically block invasive behavior. This is the first study describing the dynamic reversion of polarized epithelial spheroids into invasive cells, and gene co-expression networks associated with this transformation. While cell invasion and motility are traditionally analyzed by Boyden chamber, transwell or two-dimensional would-healing assays, our system provides a unique system to monitor and modulate invasive processes in an organotypic environment. Characterization of altered gene expression in spheroids and particularly invasive cells confirmed the importance of AKT and PI3-Kinase pathways in mammosphere- or prostasphere growth 
. However, AKT and PI3K pathways were shown to be particularly critical for invasion: Most drugs targeting these pathways effectively blocked aggressive invasion processes, but were less potent in 2D conditions, and often minimally affected growth and branching of normal cells. In contrast, mTOR, IGF1R and JAK/STAT pathways appeared to be primarily important for growth, branching and differentiation of both normal and tumor cells, regardless of the cell culture conditions, ECM and the microenvironment. Induction of JAK/STAT signaling, as reflected by the expression of many interferon-inducible proteins, may represent a general feature of migratory cells, and was observed in both branching and malignant invasive cells. Inflammation-related pathways 
seemed less relevant for either growth or invasion. Compounds inhibiting the NFκB pathway were largely ineffective, in line with the observation of reduced expression of NFκB1, IKKα and increase of NFκB inhibitors IκBα, IκBε and IκBζ in maturing spheroids. Furthermore, although expression of pro-inflammatory chemokines (CCL2, IL-1α, TNFα) was induced in spheroid formation, compounds targeting the corresponding receptors proved ineffective. Most drugs inhibiting cell cycle progression/mitosis, p38 and p42/44 MAP kinases, or matrix metalloproteinase's (MMPs) were also ineffective against invasion, with the exception of WAY-170523, a specific inhibitor of MMP13 
The pattern of invasion observed in aggressive PC-3 and PC-3M cells can be best described as streaming or chain migration 
, and only occasionally single cells move by themselves. Invading cells transiently form and resolve cell-cell contacts, while moving along a common track through the ECM. The simultaneous induction of integrins such as ITGB2, ITGB4 and ITGA10, a panel of collagens and many other extracellular proteins indicates the importance of dynamic cell-matrix adhesion and attachment forces in this type of invasion. The over-expression of some of these markers in high grade PrCa may indicate that similar mechanisms and genes also play a role in vivo. Furthermore, dynamic actin polymerization-depolymerization cycles and Rho/Rac-mediated control of cell protrusion may be required for propelling migratory cells 
. Collective chain invasion is very different from the sheet- or tube-like movement observed in branching acinar morphogenesis of normal cells - a hallmark of normal organ development - and generally more dynamic. It is also different from amoeboid or gliding patterns of movement more commonly observed in 2D cultures.
The re-expression of epithelial markers such as laminin-5 
, and the tight-junction protein Cx43 (GJP1) in invading cells is contradicting some previous reports in prostate 
, breast and ovarian cancers 
, but it is consistent with the dynamic formation and resolution of cell-cell contacts in streaming invasion. Specific laminins may be required for lubrication and maintenance of tracks utilized as channels for invasion through the ECM. Guiding cells, referred to as “guerilla cells”, may provide overall orientation and direction (reviewed in 
). The question whether fibroblasts may serve as guide cells remains to be elucidated 
. In our models, guide cells can be identified by sharp, elongated and spindle-like filopodia, formed prior to the onset of invasion.
In addition to the re-expression of epithelial markers in invasive cells, streaming invasion is not considered a characteristic for mesenchymal cells or epithelial cells that have undergone an EMT. These are traditionally thought to migrate as single cells in a fibroblast-like fashion. Although an EMT genotype was indicated by the expression of mesenchymal markers, we were not able to define a clear mesenchymal, invasion-related phenotype. Furthermore, the invasive cells lacked prominent stem-cell related expression signatures 
and did not acquire properties of CSCs (e.g. CD44+/CD24−). In contrast, expression of mesenchymal markers was a common feature in many cell lines and not causally related to malignant transformation nor invasiveness 
. Mesenchymal markers are detected in branching (RWPE-1), round (DU145) and all stellate (RWPE-2/w99, ALVA31, PC-3, PC-3M), but not in mass-phenotype spheroids (e.g. LNCaP, 22rV1) with a prominent luminal phenotype. Round, early stage PC-3 and PC-3M spheroids expressed mesenchymal markers Vimentin and Fibronectin, which remained at the same expression levels even after the invasive conversion. Vimentin was co-expressed with epithelial markers such as cytokeratins 5 and 14 or E-cadherin in round spheroids, which did not interfere with epithelial polarization and differentiation (DU145, PC-3). Nuclear translocation of β-catenin and associated Wnt pathway induction, another hallmark of EMT 
, were not observed in invading cells. Of the classic E-box binding transcription factors associated with EMT, only expression of TWIST1 
and ZEB1 correlated with the invasive potential of cell lines. None of these genes were further induced upon cell invasion. Surprisingly, Slug (SNAI2) expression was repressed during invasion, but strongly expressed in normal spheroids–suggesting a role in epithelial differentiation instead of EMT.
EMT as a developmental mechanism could be involved in normal developmental processes and invasive cancers alike, and likely represents a bidirectional process (EMT versus MET) 
. In cancers, EMT might simply be a sign of increased tumor cell plasticity, rather than a key mechanism that provides invasive properties per se. Meta-stable and phenotypic flexible cancer cells, having undergone an EMT, are still capable of epithelial differentiation. This may be particularly relevant for the survival of micro-metastases in the blood stream, successful tissue colonization, and the formation of distant metastases 
. It is interesting to note that despite the lack of both E-cadherin and alpha catenin, PC-3 cells are still able to form epithelial cell-cell contacts, apparently using alternative mechanisms which may not be a specialty restricted to this cell line. Further investigation of dynamic transformation of epithelial into invasive cells (and the reverse) may provide more general insights into these mechanisms, and the putative role of EMT. Recent reports confirm a possible function of EMT in mixed sheet- and chain migration patterns for various cell types 
. Expression of invasion-associated markers and pathways (AKT, PI3 kinase, STAT1), identified in our in vitro models, will be further investigated in clinical tumor samples, with a focus on high grade, metastasizing and invasive cancers.
In summary, our experimental systems facilitate the investigation of polarized epithelial structures or spheroids which mimic morphology, biochemistry, and invasive processes of tumors in vitro
. We and others 
have shown that breast- and PrCa cell lines in 3D are representative for many questions relevant to tumor cell biology, rather poorly addressed in monolayer cell cultures. These 3D models can be useful and more reliable for cancer drug discovery and target identification, particularly if reproducibility and quantification of the relevant assays are properly addressed.
Our models provide comparatively low cost, high throughput in vitro tools for cancer research and drug discovery, allowing complex cell biology questions to be explored experimentally, and may partly reduce or replace animal xenograft models. 3D models could therefore serve as an intermediate decision-making step in the pre-clinical drug development pipeline, linking large scale high-throughput compound screens for lead identification and increasingly expensive validation studies based on animal xenografts.