Using loss-of-function screens in 3D co-cultures of primary fibroblasts from 28 breast cancer patients with a breast carcinoma cell line, we have identified paracrine, secreted factors that regulate carcinoma cell mitogenesis. This screen, which was made feasible by a recently developed microfluidic culture platform 
, uncovered a functional patient-to-patient heterogeneity of paracrine mediators. Despite this heterogeneity, FGF-2, HB-EGF, Heparanase-1 and MT1-MMP, emerged as paracrine growth stimulators active in co-cultures with most CAF samples. IGF-1 and TGF-β1 were required for growth stimulation by CAF from low-grade but not high-grade carcinomas, indicating a grade-dependency of paracrine signaling pathways.
To limit the number of variables and focus on the patient-to-patient variability of the fibroblasts, we used one cell line as indicator to represent the carcinoma component in the co-cultures. T47D cells are considered relatively well differentiated and fall into the common category of luminal type, ER, PR positive and Her-2 negative 
. Naturally, the paracrine factors identified by the functional screen depend on the expression of growth factor receptors by T47D cells and the functionality of signaling pathways downstream of these receptors. For example, T47D cells express only low levels of the receptor tyrosine kinase Met 
, which explains why the neutralizing antibody targeting its cognate ligand HGF had little effect in co-culture with CAF. We plan to expand the co-culture screens to include other breast carcinoma subtypes and primary, stroma-matched patient-derived carcinoma cells.
In agreement with our previous work on immortalized mammary fibroblasts, FGF-2 was identified as important paracrine growth promoting factor in CAF from all but three carcinomas. siRNA expression silencing confirmed that FGF-2 originates from the fibroblasts but it is unclear whether the growth factor acts on the carcinoma cells or modulates the fibroblasts in an autocrine manner. A direct stimulatory effect of FGF-2 on T47D cells has been reported by us 
and others 
, however, it is likely that an indirect modulation of CAF behavior by FGF-2 also plays a role. Approximately 10% of all and 16–23% of luminal type B breast carcinomas show FGFR1 gene amplification; one of the most common focal amplifications observed in breast cancer 
. Amplification is tightly linked to FGFR1 overexpression in patient samples and forced overexpression of FGFR1 drives resistance to hormonal therapy in vitro 
SDF-1 is well established as an important fibroblast-derived paracrine factor that promotes breast carcinoma cell growth in vitro and in vivo by direct paracrine stimulation of carcinoma cells and by stimulating angiogenesis 
. Heparanase-1, an endoglucoronidase, cleaves heparan sulfate glycosaminoglycans in the extracellular matrix and at the cell surface into smaller fragments 
. In human breast carcinoma, heparanase-1 expression is associated with larger tumor size and lymph node metastasis 
. Several modes of action have been proposed: Heparanase-1 releases growth factors from their heparan sulfate storage sites in the ECM and generates smaller, bioactive heparan sulfate fragments, which may enhance growth factor – receptor interactions at the cell surface. Heparanase-1 also enhances proteolytic shedding of the proteoglycan syndecan-1 from the cell surface (see below) 
. Our experiments also reveal TGF-β1 as carcinoma growth promoter in co-cultures of CAF from low-grade carcinomas, which reflect the molecule’s complex and context-dependent function in cancer 
Hotary et al. showed that the enzyme MT1-MMP stimulates the growth of carcinoma cells embedded in 3D collagen but not in 2D monolayer culture - an activity that requires collagen degradation 
. In human breast carcinomas, MT1-MMP is mostly stroma-derived 
. We have recently shown that MT1-MMP cleaves syndecan-1, a proteoglycan induced in stromal fibroblasts by neighboring carcinoma cells in vivo and in vitro 
. Sdc1 ectodomain, thus released from the fibroblast cell surface, can act as paracrine growth stimulator in concert with FGF-2 and SDF-1 
. Paradoxically, in co-culture with NF, MT1-MMP appears to act as a growth inhibitor of carcinoma cells. A similar tumor suppressor activity has been proposed for MMP 3, 8, 9 and 12 
. The simplest explanation for the opposite effects seen with CAF and NF would be that the fibroblast types produce different substrates. This hypothesis is plausible considering that a differential proteomics screen has described a large and diverse group of potential MT1-MMP substrates – many of which are candidate paracrine signaling molecules 
. These findings indicate that similar to TGF-β1, MT1-MMP possesses dual functions as growth suppressor and stimulator and that these activities are critically regulated by the micro-environmental context. The paradoxical activities of MT1-MMP also offer an attractive explanation for the failure of MMP inhibitors in clinical trials. Two clinical trials had to be aborted because the MMP inhibitor led to accelerated tumor progression 
One surprising finding of our study is the fact that loss of function of any single one of these factors (FGF-2, HB-EGF, heparanase-1, SDF-1 or TGF-β1) reverses the growth advantage imparted by CAF. This apparent simultaneous dependency on multiple paracrine factors suggests a complex network of interactions between stromal fibroblasts and carcinoma cells. This apparent addiction to stromal signals creates cautious optimism that a therapeutic disruption of these pathways might retard breast carcinoma growth. A microfluidic-based high-throughput co-culture assay platform as described here, could be used to identify the critical factors in individual patients and customize stroma-targeted therapy.
The stromal tumor compartment in breast carcinomas is characterized by a remarkable inter-individual heterogeneity of gene expression 
. The stromal expression signatures can be used to cluster individual tumors into distinct subclasses, which stratify patients into prognostic groups 
. Our study demonstrates that this heterogeneity extends to functional activities of stromal fibroblasts. The fibroblast diversity may be the result of differences in stromal cell composition or due to differences in paracrine induction (e.g. trans-differentiation) by adjacent carcinoma cells 
. Differences in stromal fibroblast composition may be caused by variations in the recruitment of fibroblast precursors (e.g. local recruitment of resident fibroblasts vs. marrow-derived mesenchymal stem cells) or by selection of fibroblast subtypes in the tumor microenvironment. Unexpectedly, we observed a trend towards a loss of functional heterogeneity of mammary fibroblast as tumors develop and then progress into a more malignant type. This observation is consistent with our previous finding that global gene expression is more variable in NF than in CAF 
In summary, functional screens of paracrine fibroblast-carcinoma signaling networks may provide us with the understanding necessary to design rational, stroma-targeted therapies which disrupt these signaling pathways. The present study represents a step in this direction.