Introduction
A key step in the development of epithelial tumours is the development of a metastatic invasive phenotype that allows tumour cells to penetrate into surrounding tissues. This process involves signals provided both by the neoplastic tumour cells and by the surrounding stroma1. Although the contribution of stromal vasculature and immune cell influx to tumour progression has been recognized for some time, increasing attention is now being paid to cancer-associated fibroblasts (CAFs) — activated fibroblasts associated with malignant tumours2. CAFs act to re-model the ECM by synthesizing elevated levels of ECM molecules and ECM-degrading proteases; they also physically attach to the ECM molecules and apply contractile force. On page 1392 of this issue, Sahai and colleagues use a three-dimensional (3D) organotypic culture model to show that CAFs also enable multicellular invasion of squamous cell carcinoma (SCC) cells, remodelling the ECM to create channels that the SCC cells follow to penetrate the ECM3.
The authors used a 3D model in which a dense gel composed of fibrillar collagen I and Matrigel, which contains primarily laminin-1 and collagen IV, was supported in media such that the upper surface of the gel was exposed to air. SCC cells cultured on top of this gel were highly motile but lacked the ability to invade the ECM; however, when SCC cells were co-cultured with CAFs isolated from oral or vulval squamous cell carcinomas, the SCC cells easily penetrated the ECM. Differential labelling of CAFs and SCC cells revealed that invasion occurred in tandem, with the CAFs preceding the invading chains of SCC cells. SCC cell invasiveness required physical, non-diffusible signals provided by the CAFs, as CAF-conditioned medium did not stimulate SCC invasion and CAF-induced invasion was blocked when a thin layer of ECM separated them from the SCC cells. However, SCC cell invasion did not require direct physical contact with CAFs: when CAFs were allowed to interact with the ECM but were removed prior to addition of SCC cells, invasion was maintained. Transmission electron microscopy revealed that the leading CAFs deformed the matrix to generate cleared channels that were lined with ECM molecules which SCC cells utilized to invade the ECM. Biochemical studies revealed that there was clear separation of labour between the leading CAFs and the following SCC cells: CAF-invasion required the action of matrix metalloproteinases (MMPs), but SCC cells could follow CAF tracks even when MMPs were completely inhibited. Strikingly, blocking the ability of CAFs to physically rearrange the ECM, either through disruption of integrins or by inhibition of Rho/ROCK, had little effect on CAF invasion, but did prevent subsequent invasion by SCC cells. These results demonstrate that generation of invasion channels requires both proteolytic cleavage and structural rearrangement of the ECM (Fig. 1).
Figure 1: Generation of ECM channels for multicellular invasion.
Leading cells (purple) physically interact with ECM molecules (green) to advance forward, use MMPs to degrade those molecules that impede their progress, and then structurally modify the ECM to create an oriented track for subsequent cells (yellow) to follow.
Full size image (77 KB)Additional insight into how leading cells produce invasion-competent channels was provided by a recent study from Friedl and colleagues4. Building on their previous work showing that protease-dependent migration of tumour cells through 3D collagen lattices led to the production of channels in the ECM5, the authors used time-resolved confocal microscopy to dissect the protease-dependent invasion mechanism. They found that collective invasion of HT-1080 fibrosarcoma cells into the fibrillar collagen matrix required the function of both 
1-integrins and MMPs. Microscopic analysis of invading HT-1080 cells expressing fluorescently tagged cell surface membrane type-1 MMP (MT1–MMP) revealed that the anterior region of the cell contained localized 1-integrin and filamentous actin, but lacked active proteolysis, whereas the central region of the cell contained the proteolytic zone. Live, moving cells were found to extend leading pseudopods that pulled on the collagen fibres to propel the cell forward; in these cells, collagenolysis was confined to fibres that crossed the cell body in a perpendicular fashion. Careful analysis of time-lapse confocal images revealed that the cleaved collagen fibre ends remained bound to the cell surface and became displaced forward to form an oriented scaffold that was subsequently followed by chains of cells (Fig. 1). When cell clusters were embedded in the ECM, the leading cells produced collagen-lined degradation tracks that were reinforced and then enlarged by subsequent cell passages of up to four cells in juxtaposition. Treatment with protease inhibitors or knockdown of MT1–MMP did not affect single cell migration, but did block the formation of channels that could be used for multicellular strand invasion.
Together, these two studies advocate a model for multicellular invasion of tumour cells in which leading cells pave an ECM-lined, directionally oriented pathway through which subsequent clusters of tumour cells follow. This model of fibroblast-induced invasion may also provide insight into the mechanisms driving tissue invasion processes observed in other model systems (Fig. 2). For example, animal studies in which normal human breast epithelial cells are cultured in cleared mammary fat pads or under the kidney capsule demonstrate that reconstitution of mammary ductal morphogenesis requires the addition of both breast epithelial cells and fibroblasts6, 7. Similarly, primary breast carcinoma cells cultured in fibrillar collagen form spherical colonies, but co-cultivation with activated fibroblasts facilitates invasion of multilayered cords of carcinoma cells8. Although fibroblasts have been most commonly investigated as the leaders of invasion, other cell types have also been seen to function in this way: recent studies in mouse models of SMAD4-deficient intestinal tumours have shown that bone marrow-derived cells are recruited to the tumour front, where they stimulate invasion through a MMP-dependent process9.
Figure 2: Leading cells in morphogenic processes.
(a) CAF-lead clusters of SCC cells3, 5. (b) CAFs stimulate invasion of breast carcinoma cells8. (c) Normal fibroblasts stimulate branching morphogenesis of ductal epithelial cells6. (d) Bone-marrow-derived cells stimulate colon carcinoma invasion9. Leading cells are shown in purple, and following cells are shown in yellow.
Full size image (32 KB)Sophisticated as the elegant co-culture 3D models discussed here are, they are still simplifications of tumour progression in vivo. In the study by Sahai and colleagues3, it is clear that CAFs specifically act as leading cells, as collective cell migration never occurs when a SCC cell is the leader. However, in living tumours, the lineage of the cells undertaking a particular functional role cannot always be determined with confidence as cells can exhibit phenomena of phenotypic plasticity and epithelial-mesenchymal transition (EMT). EMT is best characterized for its role in embryonic development, where it facilitates tissue morphogenesis by allowing epithelial cells to adopt a migratory, invasive phenotype; however, accumulating evidence suggests that EMT-related events may facilitate tumour invasion and metastasis10. CAF-like cells which are derived from the neoplastic carcinoma by EMT have been identified in breast cancer biopsies11. This is a cell type that may greatly enhance tumour malignancy through its ability to promote multicellular invasion8. A complete trans-differentiation may not be necessary for a leading cell to be produced, as partial or transient EMT processes have been implicated in morphogenic collective cell invasion by epithelial cells12. This process is reminiscent of the study by Wolf et al. in which the same cell type acted as both leader and follower4. The challenge for the future will be to dissect how the processes of multicellular invasion revealed in these new 3D models can give insight into tissue morphogenesis and cancer metastasis processes in vivo.