Pathways and processes that are part of normal homeostasis, such as wound healing and TGF-β signaling, become tumor promoting in the presence of initiated cells [32
]. While this has been proven experimentally, the role of these processes in predicting human cancer outcomes has been less well studied in observational contexts. We previously observed that wound healing signatures can be detected in more than 90% of tumor adjacent tissues [7
], and in the current study, we demonstrate that extratumoral wound response can be of two types: Active and Inactive. Active cancer-adjacent tissue shows features of active EMT or cellular dedifferentiation and was associated with poor survival. Inactive extratumoral tissue maintains higher expression of cellular adhesion genes and shows lower expression levels of EMT-related transcription factors. Our preliminary estimate of the hazard ratio (HR) associated with Active (compared to Inactive) microenvironment was approximately 2.5, which is substantial considering that established genomic phenotypes such as p53 mutation status [34
] and Cyclin E overexpression [35
] have HRs near two. Acknowledging the limitations of this study, which included heterogeneously treated patients of various stages, the magnitude of this preliminary effect estimate suggests that the association merits further investigation for clinical relevance and lends credence to the influence of tumor microenvironment on breast cancer outcome.
The clinical relevance of our gene expression subtypes appeared, in our study, to be restricted to ER-positive patients. It is common for tumor-derived gene expression signatures to have prognostic value only among ER-positive tumors, which may be attributable to relatively uniform, poor survival among ER-negative patients. A recent paper simultaneously reviewed multiple gene expression signatures and showed that very few tumor signatures had value in predicting ER-negative breast cancer survival [25
]. However, it may also be the case that there are specific biological interactions between ER-positive tumors and their microenvironments. For example, TGF-β is a well-established paracrine mediator of breast cancer aggressiveness [36
], and consistent with our findings, TGF-β signatures are prognostic solely among ER-positive cancers [38
]. This example is particularly relevant given that Active microenvironments are correlated with a TGF-β signature.
Increased TGF-β signaling along with other biological characteristics of the Active microenvironment subtype suggested dedifferentiation or EMT-like gene expression, which may seem surprising given that EMT is a developmental program. However, EMT is known to be activated in normal, adult tissues undergoing wound healing (referred to as type 2 EMT, in contrast to type 1 EMT that occurs during development) [26
]. In tumors and cancer cells, EMT-associated expression has been observed [39
] (referred to as type 3 EMT), and has recently been linked with the claudin-low subtype and with resistance to therapy [21
]. In our study, we observed that claudin-low gene expression features are more common (more than twice as likely to occur) in cancer-adjacent tissues than in breast tumors, and occur independent of whether the adjacent tumor is claudin-low. In fact, claudin-low gene expression in the adjacent tissue appears to be independent of tumor subtype altogether. This suggests that the Active signature may be an endogenous response, dependent upon patient biology and not tumor biology. It also appears that the signature can occur early in disease pathogenesis, being present in some DCIS patients in the current study.
A recent study has suggested that claudin-low features may also occur (though much less frequently) in normal tissues from non-diseased women [43
]. In that study, claudin-low features were associated with high risk of breast cancer. Future studies should assess whether the prevalence of this subtype differs according to disease progression (DCIS versus Invasive), or other patient characteristics, such as age, smoking history or race. In one study of esophageal cancers, it has been hypothesized that EMT could be artificially induced in surgery [44
], consistent with the idea that EMT is induced during wounding. However, tumors displaying this signature were exposed to ischemic conditions for at least four hours. In collections at UNC, more than 95% of specimens are snap frozen within two hours of devascularization, with the majority snap frozen in under an hour. Thus induction of ischemic conditions is unlikely to explain our observation of EMT-like gene expression.
In support of an exogenous or tumor-specific induction of EMT-like gene expression, EMT marker expression has been previously reported as a highly localized response to cancer progression. Trujillo et al.
] observed high expression of SNAIL1 and TGFβ in the breast epithelium adjacent to tumors. Interestingly, across the five patients evaluated in that study, these markers were most highly expressed in the peritumoral region (< 1 cm) and expression was much lower at a distance of 5 cm from the tumor margin. In our analyses, many pairs of peritumoral and remote tissue had pervasive expression of an EMT-like signature, both near (< 2 cm) and remote from the tumor (> 2 cm). Differences between the number of markers (hundreds of genes versus few markers) and differences in the sensitivity of RNA-based and protein-based analyses may account for differences between our study and the report by Trujillo et al.
However, future work should continue to evaluate the role of distance from tumor in modifying these biological processes. If there is a wide geographic range of the tumor-effects on adjacent tissue, this has potential implications for surgery strategies. One might speculate that patients for whom a widespread gene expression alteration is present might benefit from mastectomy. If a promoting wound healing signature or Active signature persists in the lumpectomy bed, it raises the question of whether this could account for the higher rate of recurrences in this region after breast conserving therapy. If this hypothesis is to be advanced, it will be important to evaluate whether Active/Inactive signatures persists after breast conserving therapy.
Other studies of stroma-derived signatures have supported both exogenous and endogenous origins for stromal response [46
]. Whether the signature is tumor-dependent or a host factor varies signature by signature. In our findings, estrogen response signatures in extratumoral tissue were correlated with tumor ER status. However, neither the tumor ER status nor the estrogen responsiveness of the adjacent tissue were correlated with the extratumoral subtypes we identified, suggesting that the extratumoral phenotypes are not driven by hormonal exposures or endogenous hormone response. These results are important because breast tumor biology is so strongly driven by estrogen response and because previous reports have shown that estrogen positive tumor features are reflected in the adjacent normal tissue of these patients [13
]. However, while our Active and Inactive subtypes are independent of hormone receptor status of the adjacent tumor, we did recapitulate the finding of Graham et al.
] that extratumoral estrogen responsiveness mirrors the ER expression phenotype of the adjacent tissue, despite the fact that we did not micro-dissect the epithelium from the stroma. Thus, we are able to report that even signatures such as estrogen responsiveness are detectable in whole (not micro-dissected) tissue. The use of whole tissue facilitates microarray analysis of a much larger number of samples (more than 90 in the current study) facilitated the assessment of clinical associations and inter-and intra-individual variation that would not have been possible with smaller sample sizes that have been used in micro-dissection studies [1
The higher prevalence of Active subtype in our study of women with DCIS and cancer (compared to the study of non-diseased high risk patients) suggests that either the signature is associated with risk or that initial activation of the signature requires the presence of tumor or another early benign lesion [7
]. Further research to determine temporality and distinguish risk and response is needed. It will also be important to determine whether there are some extratumoral signatures that are dictated by tumor characteristics, to fully explore the role of the extratumoral tissue in disease progression. For example, grade is known to induce widespread stromal alterations, but the specific molecular signatures associated with high grade tumors are not well studied. Likewise, extratumoral microenvironments that are common to basal-like breast cancers and could account for higher risk of loco-regional recurrence could provide novel insights about the biology of basal-like breast cancer progression.