The remarkable ability of a single physiological fluid—serum—to promote the growth and survival of diverse normal and cancer cells in culture suggests that there may be a conserved, programmed response to the molecular signals that serum provides. In vivo, serum as a physiological signal has a very specific meaning: cells encounter serum—the soluble fraction of coagulated blood—only in the context of a local injury. In virtually any tissue, a rapid, concerted multicellular response, with distinct physiological exigencies that evolve over minutes, hours, and days, is required to preserve the integrity of the tissue and often the survival of the organism. In response to a wound, many of the normal differentiated characteristics of the cells in the wounded tissue are temporarily set aside in favor of an emergency response. In wound repair, as in cancer, cells that ordinarily divide infrequently are induced to proliferate rapidly, extracellular matrix and connective tissues are invaded and remodeled, epithelial cells and stromal cells migrate, and new blood vessels are recruited. In all these respects, a wound response—and the characteristic physiological response to serum—would appear to provide a highly favorable milieu for cancer progression.
We defined a stereotyped genomic expression response of fibroblasts to serum, which reflects many features of the physiology of wound healing. When we examined the expression of these genes in human tumors, we found strong evidence that a wound-like phenotype was variably present in many common human cancers (including many that are not known to be preceded by chronic wounds) and was a remarkably powerful predictor of metastasis and death in several different carcinomas.
The proposed link between the fibroblast serum response signature and cancer progression raises many questions for additional studies. Perhaps most importantly, our results do not allow us to distinguish whether the wound-like phenotype has a functionally important role in tumor progression or merely serves as a marker for the underlying propensity of a cancer to progress and metastasize. However, at least three genes induced in the fibroblast serum response, PLAUR
, and MIF
, have been previously shown to increase cancer invasiveness or angiogenesis in animal xenograft models; each of these three genes has also been shown to play an important role in wound healing (Akiri et al. 2003
; Nishihira et al. 2003
; Sidenius and Blasi 2003
). Thus, we are inclined to believe that coordinate induction of a wound-healing program in carcinomas contributes to tumor invasion and metastasis.
Several potential mechanisms might contribute to the wound-like gene expression pattern in cancers. In some cancers, ongoing local tissue injury, resulting from growth and dysfunctional behavior of the tumor cells, could continuously trigger a normal wound-healing response. The classic observation of deposited fibrin products in human tumors is consistent with this model (Dvorak 1986
). Inflammatory cells, presumably recruited by tissue disorder, may amplify the wound response and contribute to tumor invasion in part by expression of metalloproteinases (Coussens et al. 2000
; Daniel et al. 2003
). The wound response might also be initiated directly by signals from the tumor cells (Fukumura et al. 1998
), whose ability to activate an inappropriate wound-healing response—favorable to cell proliferation, invasion, and angiogenesis—might be strongly selected during cancer progression. The possibility that stromal cells might play a primary role in promoting a wound-like phenotype in some cancers is raised by studies showing that tumor-associated fibroblasts can enhance tumor engraftment and metastasis in animal models (Elenbaas and Weinberg 2001
) and the demonstration in some cancers of genotypic abnormalities in tumor-associated fibroblasts (Kurose et al. 2002
). Heterotopic interaction experiments, genetic models, and cell-culture models should enable these and other possible mechanisms to be investigated.
Our results illustrate the power of using gene expression data from specific cells or physiological and genetic manipulations to build an interpretive framework for the complex gene expression profiles of clinical samples (Lamb et al. 2003
). Several prognostic models based on gene expression patterns have previously been identified from systematic DNA microarray profiles of gene expression in human cancers. Some of these prognostic gene expression profiles appear to reflect the developmental lineage of the cancer cells (Alizadeh et al. 2000
; Sorlie et al. 2001
; Pomeroy et al. 2002
), some appear to reflect the activity of specific molecular determinants of tumor behavior (e.g., the activity of PLA2G2A
in gastric cancer [Leung et al. 2002
]), while still others represent the mechanistically agnostic results of machine-assisted learning (van 't Veer et al. 2002
; Ramaswamy et al. 2003
). Although they serve to identify many of the same tumors with unfavorable prognosis, the genes that define the fibroblast CSR overlap minimally with the genes previously used to predict outcome in the same cancers. For example, the fibroblast CSR involves only 20 out of 456 genes in an “intrinsic gene list” that can serve to segregate breast cancers into prognostically distinct groups (Perou et al. 2000
) and four out of 128 genes that define the general metastasis signature reported by Ramaswamy et al. (2003
). Only 11 genes are in common between the 231 gene van't Veer poor prognosis signature for breast cancer (van 't Veer et al. 2002
) and the fibroblast CSR genes. The prognostic power of these different sets of genes illustrates the multidimensional variation in the gene expression programs in cancers and the complex interplay of many distinct genetic and physiological factors in determining the distinctive biology of each individual tumor. Our success in discovering a significant new determinant of cancer progression, using previously published and publicly available data, illustrates the richness of the data as a continuing source for future discoveries and the importance of unrestricted access to published research data (Roberts et al. 2001
The signals and regulatory systems that normally initiate, sustain, and eventually shut down the physiological response to a wound remain to be identified and understood. Identification of the molecular control mechanisms in this pathway may pave the way to new cancer therapies or chemopreventative agents. For example, cyclooxygenase 2 is strongly induced in the response of fibroblasts to serum (Iyer et al. 1999
), and platelet-derived growth factor is one of the principal molecular signals and mitogenic factors in serum. Platelet-derived growth factor receptor and cyclooxygenase 2 are inhibited by imatinib mesylate and nonsteroidal anti-inflammatory agents, respectively—two drugs with established efficacy in treating or preventing cancer (Bergers et al. 2003
; Huls et al. 2003
). Whether these or other small molecules might derive significant activity against cancer from their ability to inhibit a dysregulated wound-healing response will be an important question for future investigation.