Introduction
Genomic studies have established four major breast cancer intrinsic subtypes (luminal A, Luminal B, HER2-enriched, basal-like) and a normal breast-like group that show significant differences in incidence, survival and response to therapy [
1‐
3]. However, as gene expression studies evolve, further subclassification of breast tumors into new molecular entities is expected to occur. In 2007, we identified a new molecular subtype, referred to as claudin-low, using 13 samples [
5]. These distinct tumors were found in both human and murine breast tumor data sets and were characterized by the low gene expression of tight junction proteins claudin 3, 4 and 7 and E-cadherin, a calcium-dependent cell-cell adhesion glycoprotein. More recently, a tumor initiating cell (TIC) genomic signature derived from CD44
+/CD24
-/low-sorted cells and mammospheres obtained from primary human breast tumors was found to be exclusively enriched by gene expression in the claudin-low subtype [
6,
7], and the expression of this CD44
+/CD24
-/low/claudin-low profile increased in posttreatment samples after neoadjuvant chemotherapy or hormone therapy [
7]. Overall, these studies suggest that the claudin-low tumor subtype lacks common epithelial cell features and is enriched for TIC features.
In this study, we comprehensively characterize the claudin-low subtype using an updated human tumor database and multiple independent data sets and present the pathological and chemotherapy response characteristics of this subtype of "triple negative" breast cancers. In contrast to the basal-like subtype, we provide evidence that claudin-low tumors are more enriched in epithelial-to-mesenchymal transition (EMT) features, immune system responses, and stem cell-associated biological processes. The molecular characterization of the claudin-low intrinsic subtype in tumors and cell lines reveals a breast cancer differentiation hierarchy that resembles the normal epithelial mammary developmental cascade.
Discussion
Here, claudin-low tumors were comprehensively characterized, and many important biological and clinical features were identified. Specifically, we addressed four topics for claudin-low tumors including (1) molecular features, (2) clinical and histological characteristics, (3) relation to established breast cancer cell lines and genetically engineered mouse models and (4) differentiation status based on analyses of purified normal mammary epithelial cell subpopulations.
Molecular characterization of the claudin-low subtype reveals that these tumors are significantly enriched in EMT and stem cell-like features while showing a low expression of luminal and proliferation-associated genes. Among these molecular characteristics, EMT and stem cell features have recently been linked to one another [
18,
33,
42,
43]. Indeed, expression of EMT-inducing transcription factors SNAI1 [
33] or TWIST1 [
33] or repression of E-cadherin [
43] in mammary epithelial cells increases the number of stem cells, and these and other EMT-inducing transcription factors such as ZEB2 and TWIST2, as well as the mesenchymal marker vimentin, are expressed at higher levels in CD44
+CD24
-/low stem cell-like cells than in more differentiated epithelial CD44
-CD24
+ cells [
18,
33]. Consistent with this finding, we observed a high mRNA expression of known transcriptional repressors of E-cadherin such as SNAI1, SNAI2, TWIST1, TWIST2, ZEB1 and ZEB2, and other EMT-inducing factors such as hypoxia-inducible factor-1a in claudin-low tumors [
31] (Figure
1b, Figure S2 in Additional file
1). Thus, our data suggest that claudin-low tumors, compared with the other intrinsic breast tumor subtypes, are the most enriched for stem cell and/or TIC features, and on the basis of our vimentin immunofluorescence staining, it appears that these mesenchymal features are present within epithelial cells, which is a feature not seen in normal breast tissues.
Acquisition of EMT and/or stem cell-like biological processes has been associated with therapeutic resistance [
7,
43,
44]. We observed that claudin-low tumors do show a lower pCR rate than basal-like tumors (Figure
3a); however, the pCR rate of claudin-low tumors was roughly equivalent to that of the HER2-enriched subtype (without anti-HER2 therapies) and much higher than luminal A or luminal B tumors. Thus, as has been described for basal-like tumors [
4], claudin-low tumors show some chemotherapy sensitivity, yet patients with these tumors still have poor survival outcomes overall (Figure
3b). A potential explanation for this similar scenario of basal-like and claudin-low tumors is that chemoresistant cells with TIC or mesenchymal properties are present at diagnosis in these two tumor subtypes as suggested by our immunofluorescence dual staining (Figure S9 in Additional file
1). This is also in concordance with a previous immunohistochemical study of 491 breast tumors where high expression of mesenchymal markers (i.e., vimentin, N-cadherin) and low expression of CDH1 were found almost exclusively in the triple-negative subgroup of tumors [
45]. However, our treatment response data suggest that these tumor cells with mesenchymal properties within basal-like and claudin-low subtypes might not have the same treatment sensitivity to anthracycline/taxane-based chemotherapy. Thus, further studies are needed to better characterize the treatment sensitivity of claudin-low and basal-like tumors to specific chemotherapeutics and/or targeted therapies. The claudin-low nine-cell line centroid predictor developed here will assist immediately in identifying the claudin-low subtype and its possible predictive value in any neoadjuvant clinical trial with associated microarray data. However, we acknowledge a potential caveat of the nine-cell line claudin-low predictor, which is that tumors with high stromal content might also be identified as claudin-low. It is possible that the signature set of genes that are high in claudin-low tumors (and cell lines) are also high in nonepithelial cells, including fibroblasts and other mesenchyme-derived cells. Thus, we cannot rule out the possibility that some of the claudin-low tumors identified in this study are tumors with low epithelial and high myofibroblast content. It is also possible that this signature is one that can occur within epithelial cells, within stromal cells, or both. Special attention to the percentage of tumor cellularity of the sample being analyzed and/or strategies that can differentiate tumor cells with mesenchymal properties (i.e., immunoflourescence assays) from normal or tumor-associated fibroblasts with mesenchymal properties are needed for the further evaluation of this signature. Finally, from a translational point of view, it is interesting to note that the publicly available NCI-60
in vitro drug-screening database includes six breast cancer cell lines, four of which are claudin-low (BT549, MDA-MB-231, MDA-MB-435 and Hs578T) and two of which are luminal (MCF-7 and T47D). Among them, MDA-MB-435 cells have been shown to have melanoma characteristics [
46], which is still a controversial topic [
47]. Nonetheless, there is a need to develop better screening programs of drug sensitivity in breast cancer cell lines that resemble the basal-like subtype, as this subtype is missing from the NCI-60 set.
Invasive ductal, metaplastic and medullary or medullary-like claudin-low carcinomas share important biological relationships as defined by gene expression, suggesting that yet to be discovered common oncogenic changes might exist. Metaplastic and medullary carcinomas both have a high incidence of methylation of BRCA1 [
48,
49], and ~50% of breast tumors from BRCA1 mutation carriers show medullary-like features [
50]. In addition, MDA-MB-436 and SUM1315MO2 claudin-low cell lines have mutations in BRCA1 [
51]. Moreover, we have shown that BRCA1 mutant basal-like SUM149PT cell line has a small subpopulation of cells with mesenchymal/claudin-low-like features, and that these cells give rise to the basal-like cells that dominate these cultures. These data suggest that BRCA1 deficiency, which has been implicated in the differentiation of MaSC or bipotent progenitors into ER-positive luminal cells [
52], might also contribute to the development or progression of undifferentiated claudin-low tumors and cell lines.
Although we have not performed functional tumor cell repopulating assays on human claudin-low tumors to show their enrichment for TICs because of the low incidence of these tumors (i.e., ~7 to 14%), there is, however, evidence that the claudin-low cell lines identified here show stem cell properties and may be highly enriched for TICs. For example, Charafe-Jauffret
et al. [
53] reported that in addition to having EMT features and high expression of stem cell markers such as ALDH1, many of these cell lines contain functional TICs. This is in concordance with another report [
54] that showed that MDA-MB-231, SUM159PT and SUM1315MO2 have a high proportion (>90%) of CD44
+/CD24
-/low cells, and that the CD44
+/CD24
-/low subpopulation obtained from these cell lines was capable of self-renewal, forming tumors in nonobese diabetic severe combined immunodeficient mice, and were more resistant to chemotherapy.
Lim
et al. [
24] delineated a human mammary epithelial hierarchy by performing cell sorting on the basis of two cell surface markers (CD49f and EpCAM) and a series of
in vitro and
in vivo experiments, including gene expression profiling of different subpopulations of the normal breast. Using their microarray data, we developed a genomic differentiation predictor that classifies breast tumors on the basis of their differentiation status along a continuous MaSC → pL → mL epithelial hierarchy. We observed that the information provided by the differentiation status adds prognostic value even when considered with intrinsic subtype and the classical clinical variables. However, as developmental studies further characterize the normal mammary differentiation hierarchy, approaches such as the one reported here can be improved. For example, much less is known about other cell types in the normal breast, such as the myoepithelial progenitors and other potential intermediate progenitors, which may be responsible for the development of other rare breast cancer subtypes such as medullary carcinomas. Finally, a similar genomic approach based on FACS data coming from other developmental studies such as the ones by Lim
et al. [
24] or Raouf
et al. [
23] might prove useful in leukemia [
55] or other solid tumors [
56], where similar differentiation hierarchies have been identified, and thus this differentiation predictor algorithm may show benefit in cancers other than breast cancer.
Integration of the claudin-low tumor subtype together with the known intrinsic subtypes delineates a differentiation hierarchy that resembles the normal epithelial development. These data point to different cells of origin for each intrinsic subtype, or different stages of developmental arrest for each subtype with a common cell type of transformation, or some combination of the two as different processes may be occurring for each different subtype. Indeed, Lim
et al. [
24] suggested that the potential cell of origin of the basal-like subtype in BRCA1 carriers might be the pL instead of the MaSC. Alternatively, as suggested by our
in vitro analyses of the SUM149PT cell line, BRCA1-mutated basal-like tumors might arise from transformation of a MaSC that is similar to claudin-low tumors or cell lines, but the claudin-low tumors stay arrested in this undifferentiated state, while MaSC or claudin-low cells in basal-like tumors are able to divide asymmetrically and give off differentiated progeny that then arrest at the pL state [
57]. The therapeutic implication of the claudin-low subtype will require additional retrospective and prospective evaluations, but what does appear clearer is that the intrinsic subtypes of breast cancer may be reflective of distinct stages of mammary epithelial cell development and that the claudin-low tumors (and cell lines) show the least differentiated phenotype.
Competing interests
CMP is a major stockholder of BioClassifier LLC and co-founder and managing partner of University Genomics. CMP and JSP have filed a patent on the PAM50 assay (University of North Carolina) and on intrinsic subtyping (University of Utah).
Authors' contributions
AP, JSP, OK and CMP contributed to experimental design. AP, JSP, OK, CL, JIH and XH were responsible for performing experiments. AP, JSP, OK and CF contributed to data analysis. AP and CMP contributed to manuscript preparation.