Tumor initiating but differentiated luminal-like breast cancer cells are highly invasive in the absence of basal-like activity

Abstract

The majority of human breast cancers exhibit luminal epithelial differentiation. However, most aggressive behavior, including invasion and purported cancer stem cell activity, are considered characteristics of basal-like cells. We asked the following questions: Must luminal-like breast cancer cells become basal-like to initiate tumors or to invade? Could luminally differentiated cells within a basally initiated hierarchy also be tumorigenic? To answer these questions, we used rare and mutually exclusive lineage markers to isolate subsets of luminal-like and basal-like cells from human breast tumors. We enriched for populations with or without prominent basal-like traits from individual tumors or single cell cloning from cell lines and recovered cells with a luminal-like phenotype. Tumor cells with basal-like traits mimicked phenotypic and functional behavior associated with stem cells assessed by gene expression, mammosphere formation and lineage markers. Luminal-like cells without basal-like traits, surprisingly, were fully capable of initiating invasive tumors in NOD SCID gamma (NSG) mice. In fact, these phenotypically pure luminal-like cells generated larger and more invasive tumors than their basal-like counterparts. The tumorigenicity and invasive potential of the luminal-like cancer cells relied strongly on the expression of the gene GCNT1, which encodes a key glycosyltransferase controlling O-glycan branching. These findings demonstrate that basal-like cells, as defined currently, are not a requirement for breast tumor aggressiveness, and that within a single tumor there are multiple “stem-like” cells with tumorigenic potential casting some doubt on the hypothesis of hierarchical or differentiative loss of tumorigenicity.

The two most frequent subtypes of human breast cancer are the luminal and the basal-like, named after their resemblance to the two major lineages in the normal human breast (1). Luminal differentiation independently of estrogen receptor expression, typically includes MUC1 and simple epithelial keratins such as K19 (MUC1⁺/K19⁺/ER^+/−), whereas basal-like activity is marked by expression of basal keratins K5, K14, and K17, and/or the transcription factor p63, on an estrogen receptor-low or -negative background (K5⁺/K14⁺/K17⁺/p63⁺/ER^lo/-) (2–4). Despite the tempting extrapolation to luminal vs. basal cells-of-origin for luminal and basal-like breast cancer, respectively, there is increasing evidence to suggest that both subtypes originate from the luminal epithelial lineage (5–7). However, once cancer is established, the only cancer stem cells so far described are essentially basal-like, and the majority of luminal cells within the tumors have been characterized as more differentiated and less malignant than basal cells (8, 9). Adding more complexity is the report that metastatic lesions are enriched in luminal cells (9). A popular explanation for these contradictions is the concept of tumor cell plasticity, i.e., the possibility that differentiated luminal cells must acquire basal-like traits to become malignant (10–14). We set out to examine whether the above explanation always must be true or whether differentiated luminal-like breast cancer cells within a basally initiated hierarchy could be aggressive and stem-like in their own right.

To address these questions, we used two mutually exclusive markers, milk mucin (MM) and CD271, which identify subtypes of cells with either luminal-like differentiation or basal-like activity, respectively. MM was detected by the M18 antibody, which recognizes branched glycans (15), whereas CD271/p75^NTR was detected by the ME20.4 antibody (16). We show that frankly differentiated luminal-like cells without acquiring appreciable basal-like traits can be aggressive and invasive when serially transplanted into NOD SCID gamma (NSG) mice or tested for invasiveness in Boyden chambers. Furthermore, that luminal-like cells derived from a stem-like, basal hierarchy cannot only be tumorigenic, but that they can also be more aggressive than their progenitors.

Results

CD271 and MM, Two Distinct Differentiation Markers of Normal Human Breast, Identify Distinct Subsets of Cells in Primary Tumors and Cell Lines.

In a search for distinct and rare candidates in both basal and luminal compartments, we used multicolor imaging of normal breast tissue, primary breast carcinomas and established cell lines stained with antibodies against a panel of markers, from which we selected two rare markers, CD271 (p75^NTR) and MM (branched glycans, here abbreviated MM for ‘milk mucin’). These two markers were selected carefully based on their lineage specificity in normal breast, cell surface location, trypsin insensitivity and expression in cultured cells, features that made them ideal for cell sorting and cloning applications (Fig. 1A Inset). Our initial staining of 53 biopsies of primary breast tumors revealed that CD271⁺ cells were present in distinct single cells or small foci in 28 of 53 (53%) biopsies, of which 20 of 28 also contained distinct and nonoverlapping populations of MM⁺ cells (Table S1); 14 of 53 (26%) tumor biopsies exhibited foci of MM⁺ cells only, and 11 of 53 (21%) showed neither CD271 nor MM staining. The above set of positive populations were present in ER⁺/PR⁺/ErbB2^lo, ER^±/PR⁻/ErbB2^hi, and ER⁻/PR⁻/ErbB2^lo subtypes of breast cancers and the corresponding sample cell lines chosen for further study (MCF7, BT474, and BT549) (Fig. 1). Multicolor imaging revealed that whereas MM was part of a MUC1⁺/ K19⁺ compartment, CD271 often costained with p63, K17 and K5 albeit with the latter two exhibiting a more widespread pattern of staining in general, but usually in an essentially ER-and MM-negative background (Fig. S1). Based on these data, we used CD271 as the most stringent criterion for basal-like activity, with p63 as an additional, facultative marker.

Fig. 1.

CD271 and MM, two distinct differentiation markers of normal human breast identify distinct subsets of cells in primary tumors and cell lines. (A–F) CD271 and MM staining define separate populations in cryostat sections of tumors (A, C, and E) and in cultures of breast cancer cell lines (B, D, and F) regardless of the subtype. Tumors and established cell lines representative for ER⁺/PR⁺/ErbB2^lo (A and B), ER^±/PR⁻/ErbB2^hi (C and D), and ER⁻/PR⁻/ErbB2^lo (E and F) breast tumors were stained with MM (red), CD271 (green), and nuclei (blue). (A Inset) Staining of normal breast tissue for comparison. Irrespective of subtype, there was no overlap in staining between MM⁺ and CD271⁺. (Scale bar, 25 μm.)

CD271⁺-Derived Clones Contain MM⁺ Populations, but Freshly Isolated MM⁺ Clones Do Not Contain CD271⁺.

To determine the relation between the two populations, we combined prospective isolation of cells with single cell cloning. FACS analysis of the MCF7 cell line readily allowed us to separate cells into CD271⁺ and MM⁺ gates (Fig. 2A). Single-cell cloning efficiency was identical between the gates that allowed delineation of CD271⁺ and MM⁺ cells (10–15%), and the expansion of single cells into “clones” required at least a month. That all clones indeed originated from single cells was verified by genomic profiling, and that these underwent essentially the same number of cell divisions before analysis was determined by evaluation of growth curves (Fig. S2 A and B). Although interclonal variation was recorded, cultures derived from the CD271⁺ clones recapitulated a pattern of heterogeneity similar to that of the unsorted population, whereas cultures derived from MM⁺ clones remained essentially negative for CD271 (Fig. 2A). The first generation MM⁺ clones also were negative for other basal-like activity measured including p63, CD44v6, Maspin, and K17 (Fig. S2C). Moreover, that CD271⁺ drove both a basal-like and a luminal-like component, whereas MM⁺ drove only differentiated luminal-like self-renewal at least in initial passages, was demonstrated with additional cultures derived from prospectively isolated primary tumor-derived cells, as well as clones of MCF7 and BT474, or FACS-sorted cells from BT549 (Fig. 2B).

Fig. 2.

CD271⁺-derived clones contain MM⁺ populations, but freshly isolated MM⁺ clones do not contain CD271⁺. (A) Representative FACS profiles of a tumor cell line (MCF7) stained with CD271 and MM. Note the l-shaped FACS profile of the parent population including CD271⁺ (green shade), MM⁺ (red shade) and double-negative CD271⁻/MM⁻ (no shade). Reanalysis of parent-derived clones (encircled) reveal that the CD271⁺-derived clone regenerates both cell types, whereas the MM⁺-derived clone does not generate CD271⁺. (B) Frequency of CD271⁺ (green) and MM⁺ (red) cells after reanalysis of CD271⁺- and MM⁺-derived cells enriched or cloned from primary tumor (STC1) or cell lines. MCF7 and BT474 cell lines could be cloned as single cells. The CD271⁺ clones consistently gave rise to mixed populations whereas the MM⁺ clones remained restricted. The data are reported as mean ± SEM (error bar).

As a proof of principle that pure MM⁺ luminal-like cells could arise directly from CD271⁺ cells rather than always being the result of aberrations sustained over length of time the cancer cells have been in culture, we recovered a second round of single cell clones from the newly derived MM⁺ cells from a freshly isolated CD271⁺ clone and reanalyzed the resulting progeny for the presence of MM⁺ and CD271⁺ cells. Indeed, pure MM⁺ luminal-like clones emerged in 4 out of 14 clones, confirming that intraclonal unidirectional differentiation does occur even in cancer cell lines. Also, whereas the CD271⁻/MM⁻ phenotype was always the most frequent, these were not necessarily similar to one another by the measures we used to assess clonal phenotypes, indicating that there is much additional heterogeneity within this CD271⁻/MM⁻ group that remains to be characterized (Fig. S2D). The CD271⁻/MM⁻ is not currently ripe for further interrogation because we have yet to find tractable and rare markers within this group. Whereas the data in the literature indicate that luminal cells with a basal-like component exhibit lineage marker dynamics in favor of a basal-initiated hierarchy, our data show that this is not always a requirement for clonal expansion.

CD271⁺ and MM⁺ Clones Are Distinct Populations by Gene and miRNA Expressions, as Well as by Mammosphere Formation, with only the CD271⁺ Exhibiting “Stem-Like Activity.”

We compared the global gene and miRNA expressions of different clones we isolated, using microarray analyses. The cultures derived from CD271⁺ clones expressed a number of stem cell markers not present in cultures from MM⁺ clones (Fig. 3A). This was confirmed by quantitative real-time PCR (qRT-PCR) and extended by additional markers (Fig. 3B). Several members of the canonical Wnt signaling pathway, shown to be associated with stem cell renewal (17), and a subset of genes that were previously shown to be up-regulated in CD44⁺ cells (9) were high in the CD271⁺ clones (Fig. S3). Expression of the miRNAs 205, 221, and 222, which correlate positively with maintenance of mammary epithelial progenitor cells in mice (18, 19) and ER negativity in breast cancer (20), were strongly up-regulated in the CD271⁺-derived cultures by miRNA microarrays (Fig. 3C Left); this result was confirmed by qRT-PCR (Fig. 3C Right). We measured progenitor activity using the mammosphere assay, which purportedly measures “stemness” (21): CD271⁺ cells had enhanced mammosphere-forming capacity compared with MM⁺ cells (Fig. 3D). Thus, for all practical purposes only cells with a basal-like component exhibited stem-like characteristics.

Fig. 3.

CD271⁺ and MM⁺ clones are distinct populations by gene and miRNA expressions, and mammosphere formation with only the CD271⁺ exhibiting stem cell properties. (A) Heat maps of selective gene sets from clustered data are shown depicting high expression of specific stem-cell makers (arrows) in CD271⁺ clones compared with MM⁺ clones of MCF7. Yellow = highest expression; blue = lowest expression, black = average expression; gray = missing data. (B) Differential gene expression verified and expanded by qRT-PCR suggests stem-like properties of CD271⁺ as opposed to MM⁺ cells. (C) miRNAs differentially expressed between CD271⁺ and MM⁺ clones of MCF7. Expression was determined by microarray analysis (Left) and confirmed by qRT-PCR (Right). The blue to green dots indicate that the values are higher in CD271, whereas orange to red indicate that the values are higher in the MM cells. The miRNAs that show at least twofold change between the two populations are presented. Note that miRs 205, 221 and 222 generally associated with mammary progenitor activity were highly expressed in the CD271⁺ clones. (D) CD271⁺ and MM⁺ populations have distinct mammosphere-forming capacity. Mammospheres were quantified as numbers of mammospheres per well (tumors) or frequency of wells with mammosphere formation (cell lines). **P < 0.01 compared with MM⁺.

Luminal-Like Cells Within the Hierarchy Are Tumor Initiating and Invasive in the Absence of Basal-Like Traits.

To determine how purified differentiated luminal-like cells function in vivo, CD271⁺ and MM⁺ cells from both primary tumors and cell lines were xenografted into NSG mice (details of the additional nine primary tumors used is in Table S2). All populations (including CD271⁻/MM⁻) formed tumors from relatively low numbers of xenografted cells in NSG mice (Fig. 4A). A basal-like-initiated hierarchy was functional in our in vivo assay, because CD271⁺ cells, but not MM⁺ cells (with or without basal keratins), were able to recapitulate the heterogeneity of the original, presorted population. This recapitulation was observed irrespective of whether the xenografted cells were derived from primary tumors or tumor cell lines (Fig. 4B). Surprisingly, however, we observed that the MM⁺-derived tumors grew substantially larger than the CD271⁺-derived tumors and appeared invasive (Fig. 4C). We confirmed the invasiveness of each of the populations using the Matrigel-coated transwell filter assay, and the MM⁺ populations proved by far the most invasive (Fig. 4D). We conclude that although there appears to be a differentiation hierarchy, this hierarchy does not confer progressive loss of tumor-initiating capacity and invasiveness. These properties do not diminish with differentiation; if anything, the opposite is true.

Fig. 4.

Luminal-like cells within the hierarchy are tumor initiating and invasive in the absence of basal-like traits. (A) Summary of tumor formation in NSG mice. Tumor formation within 10–15 wk after injection into NSG mice of cells from a xenografted tumor (PT13), from short-term cultured samples (STC1, STC2), or from a tumor cell line (MCF7) all separated by flow cytometry into CD271⁺ (green), MM⁺ (red), and CD271⁻/MM⁻ (blue) populations. All cell populations readily formed tumors. Asterisks indicate tumor formation by serial transplantation. (B) Multicolor imaging of cryostat sections from CD271⁺- and MM⁺-derived xenografts stained with CD271 (green), MM (red), and nuclei (blue). Note that the hierarchical organization of tumor phenotype is maintained also in vivo. (Scale bar, 50 μm.) (C) Tumor volume of CD271⁺- and MM⁺-derived xenografts (second passage PT13 and STC1, first passage SCT2, second passage MCF7 clones) as measured 10–15 wk after injection of 10³–10⁴ cells. For STC1 the CD271⁺ cells were substituted with MM⁻/EpCAM⁺ cells due to lack of sufficient numbers of bona fide CD271⁺ cells. Dot plots indicate individual tumor volumes (Mean indicated by solid line). *P < 0.05 compared with CD271⁺. (D) Cells sorted by flow cytometry into CD271⁺ and MM⁺ from primary tumors (1 × 10⁴ cells; n = 5) and tumor cell lines (1 × 10⁵ cells of MCF7 clones or BT474 clones; 5 × 10⁴ cells of FACS-sorted BT549; n = 3 each cell line) were incubated in a modified Boyden chamber assay with thinly layered Matrigel-coated Transwell chambers. For some of primary tumors (PT1, PT7, PT9) or for BT549, CD271⁺ cells were replaced with MM⁻/EpCAM⁺ or MM⁻/CD271⁻ due to insufficient numbers of CD271⁺ cells. Arrow indicates MM⁺ cell invasion in vitro after GCNT1 silencing. Silenced cells were less invasive than the control cells. *P < 0.05; **P < 0.01 compared with CD271⁺. (E) Schematic representation of a GCNT1 silencing strategy in MM⁺ cells. The core 2 β6GlcNAcT1 encoded by GCNT1 controls branching of O-glycans, here assumed to represent the MM epitope. Short hairpin RNA targeting of GCNT1 (shGCNT1) inhibits expression of the MM epitope. (F) Kaplan–Meier graphs depict the incidence of palpable tumor development over time after s.c. injection of 10⁷ cells from the CD271⁺ or MM⁺ clones of MCF7 in nude mice, showing that GCNT1 silencing (n = 3) inhibits tumor formation of MM⁺ clone to the level of CD271⁺ clone in nude mice. MM⁺ cells are much more aggressively tumorigenic in nude mice. Five out of six injections developed tumors with MM⁺ cells.

To understand the mechanism responsible for the MM⁺ cells display the robust malignant phenotype, we investigated the role of expression of the branched glycan specific to MM cells. This glycan is detected by MAb M18, an antibody that has previously been identified as a differentiation marker of breast epithelium (22). The primary gene controlling branching of O-glycans in breast epithelium is the core 2 synthase (core 2 β6GlcNAcT1) encoded by GCNT1. We therefore tested whether GCNT1 may play a role in the aggressiveness of the MM⁺ subpopulation (Fig. 4E and Fig. S4A). Knockdown of GCNT1 did not interfere with expression of MUC1 as revealed by staining with the antibody 115D8 (Fig. S4B). GCNT1 knockdown in cultures derived from a MM⁺ clone of MCF7 dramatically delayed tumor formation in nude mice to a level comparable to the delay observed for cultures derived from CD271⁺ clones (Fig. 4F); in addition, there was a significant reduction in the invasive capacity as measured by the Matrigel-coated transwell filter assay (arrow in Fig. 4D). These data support the surprising conclusion that aggressiveness of MM⁺ tumors is dependent on specific changes in glycosylation.

CD271⁺ and MM⁺ Gene Signatures Predict Poor Clinical Outcome in Breast Cancer.

To determine the clinical implications of these results, we analyzed the gene signatures for the CD271⁺ and MM⁺ tumors. We selected a set of 1,025 probes that were among the top genes identified by differential expression between CD271⁺ and MM⁺ clones using significance analysis of microarrays (SAM). (Fig. 5A; probes and genes listed in Dataset S1). In three independent datasets, we found that both CD271⁺ or MM⁺ gene signatures predicted poor, relapse-free survival from breast cancer (P < 0.05) (Fig. 5B and Fig. S5 A and B). Remarkably, the relapse-free survival curves are almost identical whether one uses the CD271⁺ or the MM⁺ signatures despite the fact that the two signatures are not related. As control, a specific gene set of highly expressed genes in parental MCF7, different from the CD271⁺ and MM⁺ signatures, was not predictive of outcome in the same cohorts. Thus, both CD271- and MM-derived gene expression signatures show prognostic value.

Fig. 5.

CD271⁺ and MM⁺ gene signatures predict poor clinical outcome in breast cancer. (A) Scaled-down representation of the 1025 probe cluster, illustrating statistically differentially expressed genes between four MM⁺ and three CD271⁺ clones of MCF7. 591 probes that were higher expressed in MM⁺ cells compared with CD271⁺ cells, are referred to as the ‘MM⁺ gene signature’, whereas 434 probes that were higher expressed in CD271⁺ cells compared with MM⁺ cells are referred to as ‘CD271⁺ gene signature’. (B) Disease-specific survival for patients separated in two groups based on expression of CD271⁺ or MM⁺ gene signature. Expression of the gene signatures was analyzed in three independent breast cancer microarray datasets: UNC dataset with 243 patients, Miller dataset with 237 patients, and MicMa dataset with 107 patients. Kaplan–Meier curves and log rank test show that both gene signatures are able to identify patients with higher rates of breast cancer recurrence.

Discussion

We have prospectively isolated two distinct populations of malignant cells, one basal-like and one luminal-like, forming a differentiation hierarchy in primary breast tumors as well as in diverse breast cancer cell lines. Both the population with the basal-like and the luminal-like markers are tumor-initiating and invasive by all of the criteria measured. These data could be interpreted either in light of the cancer stem cell (CSC) model or the clonal evolution model (for review, see ref. 23). Our finding of a stem-like subpopulation, which by all measures appears to recapitulate the heterogeneity of the original population, would be expected within a CSC framework. Such prediction is a hallmark of the CSC hypothesis (8). However, our results differ dramatically from the CSC model in that all of the populations tested within the differentiation hierarchy are tumor initiating at relatively low cell numbers in the more permissive NSG mice. One explanation for this apparent contradiction could be the possibility that differentiated cancer cells potentially convert into tumor-initiating CSCs in a reversible manner (11). Breast CSCs are thought to be generally basal-like (9, 14), implying that tumor formation would have to be initiated by cells with basal-like activity.

By applying two narrowly expressed, mutually exclusive markers, MM and CD271, not used previously for breast CSC analysis, here we provide evidence compatible with a third, “fusion” model. CD271 has been described as a unique marker of melanoma stem cells in humans (24). However, its first description in breast attributed it as a marker of myoepithelial cells (16). More recent studies show that its overexpression strongly increases resistance to anti-tumoral TRAIL treatment (25). The M18 antibody was shown to recognize branched glycans (15) and to stain apical membranes of luminal epithelial cells in the normal human breast (22), but further specificity of its affinity remained to be investigated. Here we identify a linear connection between cells expressing these two nonoverlapping markers suggestive of at least a rudimentary differentiation lineage hierarchy in breast cancers—and surprisingly also in breast cancer cell lines. We show, however, that differentiation within this hierarchy not only does not decrease tumor-initiating capacity as predicted by the CSC model, but instead, the expanded sets of derivative cells appear to function as additional candidate cancer-initiating cells driving aggressive tumors. However, these cells are distinct from canonical stem cells in that they neither have undergone a detectable epithelial-to-mesenchymal transition, nor have they regressed to any particular dedifferentiated or basal-like state. These data require a revision to the existing hypotheses of breast cancer progression that can account for the observed contributions of multiple cell populations to tumor initiation.

Another unexpected finding in the present study is that basal-like- or luminal-like- differentiation within subpopulations appears to not be limited to any particular subtype of breast cancer. In fact, we found rare basal-like cells even in bona fide primary luminal breast tumors or more surprisingly in breast cancer cell lines, such as MCF7, and found MM⁺ luminally differentiated cells in bona fide primary basal tumors or basal tumor cell lines such as BT549. This could be interpreted either in favor of the classical view of cancer as a caricature of normal tissue renewal (26), or as transformation-induced basal/luminal-like reprogramming (27). The first interpretation would have to rely on the assumption that breast cancer originates from normal stem cells. Candidate human breast stem cells have been reported repeatedly to be basal-primed by several of the markers used in the present study to characterize basal-like activity (9, 28). It was shown recently in mice that adult tissue homeostasis does not rely on classical stem cells, but rather on self-duplication of lineage-restricted progenitors (29). However, in contrast to mice, humans contain a substantial amount of K5⁺/K14⁺/K17⁺ luminal breast epithelial cells, which is why basal-like in the human breast is better defined by the concomitant expression of p63 and CD271. Nevertheless, if breast epithelial multipotency were limited to fetal life also in humans, and basal-like activity in its most stringent CD271⁺ sense reflected cell-of-origin in cancer, it would imply that the earliest seeds of human breast cancer must exist in utero. Although increased susceptibility to breast cancer has been linked to the intrauterine origin (30), the preponderance of data that suggest a luminal origin of most breast cancers favors the second option for generation of basal-like activity in breast cancers, i.e., postnatal or postpuberty reprogramming (5, 6). This possibility includes that luminal cells should occasionally acquire a basal-like phenotype upon malignant transformation. We have shown previously that under certain culture conditions a small minority can convert to basal-like cells (31). With the recent demonstration of in vivo self-duplication in the mouse mammary gland in mind (29), it may be that culture conditions may unmask latent (or cryptic) potentials in luminal cells. Such distinctive expression of gene expression potentials between cells in vivo and cells in culture was summarized long ago and continues to be reported (32, 33). In fact, it all comes down to microenvironmental cues and context because, even in vivo, mouse mammary gland luminal progenitors generate basal cells if exposed to a cleared fat pad (34, 35).

Finally, the most intriguing finding of our investigation, apart from the fact that multiple cell types are tumor initiating, is the observation that both invasion and tumorigenicity of luminal-like breast cancer and breast cancer cell lines depended on GCNT1 expression, which is required for elaboration of the MM-specific branched glycan recognized by antibody M18 (15). There is increasing appreciation that glycans play an important role in malignant behavior. Indeed, GCNT2 has been associated with breast cancer metastasis, and blocking its expression was shown to abrogate migration and invasion (36).

In conclusion, our data implicate multiple types of tumor-initiating cells in breast cancers camouflaged in different phenotypic cloaks, including differentiated luminal-like cells. This particular finding is of profound clinical importance and strongly suggests the need for combinatorial therapies targeting multiple cell types in our search to prevent cancer recurrence or curb tumor aggressiveness while sparing the host.

Materials and Methods

Human Tissue Samples and Mice.

Human tissue included fifteen reduction mammoplasty samples and eighty-one mastectomy biopsies. Mice included BALB/cA nude mice as well as NSG mice. See SI Materials and Methods.

Human Breast Cancer Cell Lines.

Cell lines included three established cell lines (MCF7, BT474, and BT549) and two primary derived cell lines defined as cultures expanded within the limits of a normal finite life span (HMT-3909 and L56Br-C1). See SI Materials and Methods.

Cell Isolation, Staining, and Sorting.

Tumor tissue was enzymatically digested by collagenase and trypsin/EDTA into single cells before FACS analysis. Single cell clones were generated from established cell lines, and >50 clones were reanalyzed for multipotency. See SI Materials and Methods.

Antibodies and Immunostaining.

Cryostat sections of normal or tumor tissue as well as cultured cells were stained by immunoperoxidase or immunofluorescence. Tumors were classified as ER⁺/PR⁺/ErbB2^lo, ER^±/PR⁻/ErbB2^hi, or ER⁻/PR⁻/ErbB2^lo phenotype based on staining with ER, PR, ErbB2, K5, and K17. See SI Materials and Methods.

In Vitro Cellular Assays.

See SI Materials and Methods.

In Vivo Transplantation.

Limited dilution of MM⁺ or CD271⁺ MCF7 clones in a suspension of 50% Matrigel (BD Biosciences, cat. no. 356231) in DME/F12 medium with 10% (vol/vol) FCS was performed in NSG mice for 8–10 wk. MM⁻/CD271⁻ cells of MCF7 were tested by two inoculations of 10² cells. For serial transplantation, MM⁺ or CD271⁺ cells isolated by FACS from an initial inoculation of 10² MM⁺ or CD271⁺ MCF7 clones were retransplanted in NSG mice with 10³ cells. Tumor growth was monitored for 10 wk, and tumor volume was measured at the time of sacrifice and calculated by the ellipsoid volume (37). For serial transplantation, cells were isolated by FACS, transplanted into NSG mice and monitored for up to 15 wk. See SI Materials and Methods.

RNA Isolation and mRNA/miRNA Expression Analysis.

See SI Materials and Methods.

Microarray Experiments.

See SI Materials and Methods.

GCNT1 Silencing.

See SI Materials and Methods.

Data Analysis and Statistics.

Genes significantly differentially expressed between MM⁺ and CD271⁺ cells were identified using two-class, unpaired SAM and 1025 genes were confirmed by DEDS. Survival analysis was performed by Kaplan–Meier analysis with log-rank test. Nonparametric Mann–Whitney tests were performed to measure the differences in mammosphere-forming capacity, tumor formation, and invasiveness between CD271⁺ and MM⁺ cells. See SI Materials and Methods.