Unifying model of TGFβ signaling in breast epithelium. Implications for cancer treatment using TGFβ pathway antagonists. The model assumes the existence of a primitive multipotent stem cell population that is ER-negative, PR-negative, CD44⁺, CD24^lo, ESA⁺, KRT5⁺, TGFbR2⁺ and possibly express mesenchymal markers. When these cells divide, they give rise either to phenotypically identical daughter stem cells or to precursors of differentiated progeny (luminal, basal, or myoepithelial). Luminal cells are epithelioid (ESA⁺, KRT8/18⁺, CDH1⁺), express ER and PR, are CD44^-, CD24⁺, and the TGFBR2 gene is transcriptionally silent (TGFBR2^-). The basal and myoepithelial progeny express basal cytokeratins (KRT5/6/14⁺) or myoepithelial markers (αSMA, p63), lack ESA (ESA^-), and appear to be enriched for CD44⁺, CD24^lo cells. Whether these cells do or do not continue to express TGFBR2 is unclear. One of the fundamental determinants of the proportions of these different subpopulations in normal or cancer tissue is the ratio of stem cell divisions that give rise to daughter stem cells versus differentiated progeny [174]. We propose that exposure to TGFβ leads to enrichment of the cell population with multipotent CD44⁺, CD24^lo clonogenic/tumor initiating cells that express mesenchymal markers (EMT) [60], while treatment with a TGFβ receptor kinase inhibitor leads to enrichment with CD44^-, CD24⁺, TGFBR2^- nonclonogenic differentiated luminal cells [78]. Thus, we would like to propose the idea that TGFβ signaling is a key determinant of the fate of stem cell progeny, and responsible for maintenance of the stem cell pool in the normal mammary gland. Moreover, this model suggests the hypothesis that the constitutive activation of TGFβ signaling associated with invasive and metastatic breast cancer might contribute to steadily enriching the cancer stem cell pool, thereby indirectly enhancing the potential for metastasis and poor patient survival. In this context, it is noteworthy that EMT of normal as well as malignant human mammary epithelial cells appears to be tightly associated with the stem cell state [60]. Superimposed on to these population dynamics at the tissue level is the role of systemic estrogens and progestins in regulating TGFβ signaling (see section “Extensive cross-talk between estrogen- and TGFβ signaling pathways”). The ER/PR-positive luminal cells respond to estrogen by producing growth factors and lowering TGFβ production. These conditions would favor the commitment of stem cell daughter cells to become more differentiated progenitor cells. Moreover, because the TGFBR2 gene is transcriptionally silenced in these luminal daughter cells, this would further favor their clonal expansion. Conversely, anti-estrogens induce TGFβ production, which, even if it inhibits the rate of stem cell proliferation, would favor their self-renewal over commitment to differentiation. Once cancer stem cells break through the barrier of TGFβ-mediated growth arrest, the presence of active TGFβ would select for continued expansion of the cancer stem cell pool, and, consequently, metastasis. This model predicts that treatment of basal-like cancers with TGFβ antagonists may have anti-tumor effects by converting the putative stem cell population into a luminal, non-proliferating (and, perhaps, non-metastatic), phenotype (mesenchymal-to-epithelial transition, MET). Conversely, these agents might antagonize the therapeutic effects of anti-estrogens in estrogen-dependent luminal cancers. However, once these cells become estrogen-independent because they escape from TGFβ-mediated growth arrest, our model provides a rationale for including TGFβ pathway antagonists in the treatment program. Each of these predictions is testable in model systems as well as in clinical trials

A The TGFβ/Smad signaling pathway is highly conserved from fruitfly (left) to man (right). Currently available TGFβ pathway antagonists are shown in red next to their molecular targets. AP-12009, GC-1008 and LY2157299 are currently undergoing early phase clinical trials. Preclinical studies of TGFβ antagonists in cancer have recently been reviewed in [127, 187, 237- 239]. These studies have provided convincing evidence that targeting the TGFβ pathway can inhibit tumor growth and metastasis in at least some cancer models. These findings demonstrate that the neoplastic phenotype can be critically dependent on TGFβ signaling, so-called pathway addiction. Besides cancer, preclinical studies of chemical TGFβ receptor antagonists have demonstrated therapeutic activity in animal models of idiopathic pulmonary fibrosis [240], liver disease [241, 242], renal fibrosis [243] and Marfan syndrome [244]. B TGFβ superfamily of ligands and associated signaling receptors. The two major subgroups of ligands activate either the Smad2/3 or the Smad1/5 pathways that control two largely non-overlapping genetic programs

Association between TGFβ pathway gene expression and TGFβ target gene response signature (TBRS) and human breast cancer subsets. A We first posed the question whether or not expression of genes involved in the canonical TGFβ signaling pathway were associated with particular breast cancer subsets as defined by Alexe et al. [179]. Gene expression profiles from 200 breast cancer specimens described by Wang et al. [146] and from 47 breast cancer specimens reported by Richardson et al. [245] were examined. The samples in these datasets were first classified into the six subtypes of breast cancer identified based on a nearest centroid approach as described by Alexe et al. [179]. Expression data were preprocessed and normalized as previously described (MAS 5.0 normalization followed by max median compression of probeset IDs to unigene symbols, followed by standard normalization of genes across samples; integration of the two standard normalized datasets was performed by applying the Distance Weighted Discrimination (DWD) approach) [179]. The identification of differential expression of TGFβ-pathway genes in breast cancer subtypes versus normal samples was performed based on a non-parametric (Wilcoxon) t-test with multiple re-sampling to correct for biased distribution sizes. Boxplots of gene expression intensities across breast cancer subtypes and normal samples are shown. TGFB1 was significantly up-regulated in the HER2 + (I) and LA subsets (P < 0.01), the BA1 subset (P = 0.03) and the HER2 + (NI) (P = 0.04). TGFB2 was not significantly up- or down regulated in any of the subtypes. TGFB3 was significantly down-regulated in the basal subtypes, but not significantly up-regulated in any other subtype. TGFBR1 (ALK5) was up-regulated in BA1 (P = 0.04), both HER2 + subclusters (HER2 + (I) P = 0.03; HER2 + (NI) P = 0.013). TGFBR2 was significantly down-regulated in each breast cancer subtype, most notably in the LA, BA2, and HER2 + (NI) subclusters, which are most strongly associated with the TBRS (see B). B We then posed the question whether the 153-gene TBRS described by Padua et al. [147] (TBRS^MSKCC) as well as a similar 92-gene signature developed in our own laboratory [67] (TBRS^CINJ) were associated with any particular breast cancer subsets, as defined by Alexe et al. [179], using Gene Set Enrichment Analysis (GSEA) [246, 247]. Given a list of genes, ranked by the correlation of their genome-wide expression profiles with one of several phenotypes, GSEA seeks to estimate the significance of the over-representation of an independently defined set of genes, S, in the highly correlated or anti-correlated genes in the list. To evaluate this degree of `enrichment' the GSEA method calculates an Enrichment Score (ES) by walking down the list, increasing a cumulative sum when a gene is in S and decreasing it if a gene is not in S. The size of the increment depends on the genephenotype correlation. The ES is the maximum deviation from zero of the cumulative sum and can be interpreted as a weighted Kolmogorov-Smirnov statistic. The genes in the gene set S that ▶ appear in the ranked list before the point where the running sum achieves the ES are called the leading-edge subset and are particularly important in evaluating the results of GSEA analysis. The significance of a gene set's ES is estimated by an empirical phenotype-based permutation test procedure. When an entire database of gene sets is scored, an adjustment must be made to the resulting P-values to account for multiple hypotheses testing. GSEA normalizes the ES for each gene set to account for the variation in set sizes, yielding a normalized enrichment score (NES), and calculates a false discovery rate (FDR) corresponding to each NES. The FDR gives an estimate of the probability that a set with a given NES represents a false positive finding; it is computed by comparing the tails of the observed and permutation-computed null distributions for the NES. For each tumor, a number between -1 and +1 was derived, indicating the likelihood that TGFβ signaling is active in that tumor. Each bar represents an individual tumor specimen. Three different breast cancer expression profile data sets were analyzed. These included the GSE_2034 data of 286 specimens described by Wang et al. [146], the GSE_4992 data of 249 specimens described by Ivshina et al. [181] and the GSE_7390 data on 165 specimens reported by Desmedt et al. [180]. Based on the GSEA scores, the TBRS was significantly associated with the poorprognosis BA2 and HER2(NI) subclusters, as well as with the good prognosis LA subset. Moreover, results obtained using the TBRS^CINJ signature [67] closely matched those obtained with the TBRS^MSKCC [147] (Correlation coefficient ~0.80). C The relationships between TBRS^MSKCC and overall survival were explored on patients with lymph node negative disease, who did not receive adjuvant systemic therapy. Cases from Wang et al. [147] and Desmedt et al. [180] were pooled for this analysis. Red lines: TBRS^MSKCC-negative; Blue lines: TBRS^MSKCC-positive; Grey lines: Combined groups. The expression of a TBRS was associated with poor survival. However, in subset analysis, TBRS was not associated with a difference in overall survival in the luminal A subset, although there was trend towards a negative effect in the luminal B subset. Similar trends were found between TBRS expression and survival in both two basal-like cancer subsets. Patients with a HER2(I) cancer had a significantly better outcome than those with a HER2(NI) cancer. This result validates the results previously reported by Alexe et al. [179]. However, among the HER2(NI) cases, the expression of a TBRS predicted for extremely poor survival, while survival of patients whose tumors did not express the TBRS was essentially indistinguishable from those with HER(I) disease. These results suggest that activation of the TGFβ pathway largely accounts for the poor prognosis of patient with HER2(NI) breast cancer, and that this particular subset may benefit the most from treatment with TGFβ pathway antagonists