A direct clinical application of these findings is the development of a high-specificity invasion-sensing biomarker product detecting coordinated overexpression of a few top-ranked genes, such as COL11A1, INHBA, and THBS2, as shown in the scatter plots of Figure 2. A positive result in seemingly low-stage primary tumors will indicate that the disease has obtained the stromal signature and thus has already reached an invasive stage. As described above, the same product can also be used to predict resistance to neoadjuvant chemotherapy.
Of course, the most significant clinical application would be to develop metastasis-inhibiting therapeutics using targets deduced from the biological knowledge provided by the MAF signature. Our top ranked genes strongly suggest that they are produced by myofibroblasts or myofibroblast-like cells activated by activin A-induced TGF-β signaling and leading to some form of altered proteolysis , which results in extracellular matrix remodeling. Supporting this hypothesis are the facts that both INHBA and THBS2 are involved in TGF-β signaling: Activin A (INHBA homodimer) is a TGF-β superfamily member (ligand) and THBS2 inhibits activation of TGF-β by THBS1, which is also present in the MAF signature. Remarkably, activin A is already known to facilitate fibroblast-mediated collagen gel contraction . The role of gene LRRC15 (aka LIB) appears important but unclear, though it has already been recognized as promoting migration through the extracellular matrix . Versican (VCAN) is an extracellular matrix proteoglycan already known to play a role in metastasis, while MMP11 is one of several matrix metalloproteinases involved in the breakdown of extracellular matrix.
Although each of the MAF signature molecules could serve as a potential therapeutic target, the hypothesis that activin A signaling is at the heart of the MAF mechanism immediately suggests that follistatin (activin-binding protein) could serve as a metastasis inhibitor, which is exactly what recent research  indicates. Specifically, lung cancer cell lines transfected with follistatin and injected into immunodeficient mice markedly inhibited metastasis compared with non-transfected cell lines, but the authors of the study recognize that the role of follistatin in cancer metastasis is totally unknown . Our work provides an explanation and suggests that the same could be true for other cancers as well. Further support is provided by the fact that follistatin virtually abolished the fibroblast-mediated collagen gel contraction mentioned earlier .
There are several reasons that the core MAF signature has not yet been discovered as a multi-cancer metastasis-associated signature. First, it is essential to identify the correct phenotypic staging threshold recognizing that the signature only exists in a subset of tumors that exceed that particular stage. Indeed, if the threshold in breast cancer was put between stage I and stage II, or between stage II and stage III, rather than between in situ and stage I, the signature would not be apparent. Second, each cancer type may have its own additional features accompanying the MAF signature. For example, in ovarian cancer it is accompanied by sharp downregulation of genes COLEC11, PEG3 and TSPAN8, which is not the case in other cancers. Indeed, the main contribution of our work is the identification of the common multi-cancer "core" signature, from which a universal metastasis-associated biological mechanism can be identified. Third, the MAF signature may be reversible, perhaps as a result of the disappearance of many of the stromal cells in the mature desmoplastic stroma when it is replaced by "acellular" matrix . The presence of the signature in high-stage samples may even paradoxically be associated with longer survival if its reversal is required for further distant metastases (see below).
An important topic of further research is the determination of the precise biological event of interaction of cancer cells with the microenvironment that gives rise to the stromal MAF signature and associated invasiveness. Because of the recognized similarities with the mechanisms of wound healing , it is likely that this event uses existing wound healing response pathways. For example, it appears to occur very early in breast cancer, late in ovarian cancer, and never in glioblastoma (which is reasonable, because glioblastoma metastasizes extremely rarely). The late appearance of the MAF signature in ovarian cancer and its presence in omental metastases can be explained by the fact that ovarian cancer initially progresses by disseminating locally across mesothelial surfaces and that, contrary to hematogenously metastasizing tumors, initial metastasis is probably carried by the physiological movement of peritoneal fluid to the peritoneum and omentum .
Several of the top-ranked genes in the MAF signature (such as thrombospondins, decorin, INHBA itself) are known to be potent anti-angiogenesis mediators. The reversal of the MAF signature would thus facilitate further metastatic dissemination to distant sites. In other words, (a) the desmoplastic MAF signature and (b) angiogenesis, are two independent biological events. The former appears to be based on activin A signaling, as several of the MAF proteins in addition to INHBA are also known inhibitors of the standard TGF-β ligand. The reversal of the MAF signature would allow the standard ligand to take over in TGF-β signaling, and may thus facilitate further metastasis. These observations provide explanations for the seemingly contradictory observed roles of TGF-β signaling inhibiting early cancer but facilitating metastasis.
The possible reversibility of the MAF signature leads to the intriguing hypothesis that perhaps all metastases have, at some point temporarily been there, which explains why we only observe it in a subset of them. This would be particularly exciting, because in that case any metastasis-inhibiting therapeutic intervention targeting the MAF mechanism would be widely applicable to low-stage tumors.