The Heisenberg uncertainty principle has a great impact on medical research by drawing our attention to the bias introduced by our experimental tools. In a recent issue of Breast Cancer Research, Keller and colleagues [1] report an example of this principle: sustained propagation of large numbers of cells, through the establishment of cell lines, disrupts the normal balance between differentiated cells and their progenitors, as observed in fresh biological specimens. The work of these authors contributes another piece in a contentious field that combines tissue morphology and immunohistochemical phenotypes [2, 3], molecular classification of breast cancer tissues [4], and cell biological assays aimed at the tumor-initiating cell (TIC) phenotype [5]. Sorting cells according to their respective cell surface markers, CD44+/CD24-/low, results in the enrichment of TIC activities, including mammospheres [6] and transplantation efficiency in mouse xenografts [7]. Establishing xenograft growth could be the product of several system-specific selections other than breast progenitor phenotypes. However, further molecular profiling of these cell populations - in which CD44+/CD24-/low-sorted cells expressed low levels of luminal differentiation markers (such as MUC1, CD24, or CDH1) and elevated levels of epithelial-mesenchymal transition markers (such as VIM, collagens, TWIST1, SNAI1/2, and Zeb1/2) - indicated a link between epithelial-mesenchymal transition, TIC, and basal-like [6, 8] or claudin-low [9, 10]-specific breast cancer molecular subtypes. More recently, however, a more comprehensive interrogation of pluripotent self-renewal identified a population high for CD24, or luminal progenitors [9, 1113], capable of giving rise to mesenchymal or basal-like tumors, at least in the context of BrCa1 mutations. Given the variability of single markers within single individuals, the different sensitivities each cell biological assay presents with, and the consistency across other genes (which are more likely to be drivers of the phenotypes rather than effective surrogate markers), the more recent work presents compelling evidence that, admittedly, neither CD24 nor CD44 populations are homogenous or contain pure differentiated of progenitor populations, respectively. A hypothetical linear sequential differentiation track that would flip back and forth with respect to CD24 expression and appear as a hidden subpopulation in a majority of cells from another differentiation stage could explain this discrepancy. A more comprehensive whole-genome mRNA profiling analysis of the relatedness between luminal progenitors (CD49fhi/EpCAM+), stem cells (CD49fhi/EpCAM-), and CD44+/CD24-/low populations is necessary to assess this hypothesis. This point is emphasized by the fact that stem cell marker ALDH1 [14] is expressed by only some of the cells in either fractions described above yet ALDH1+ cells exhibit the greatest TIC capacity.

The underlying hypothesis assumes that, within the dynamic steady state of breast epithelial maintenance, self-renewal, and differentiation (as it responds to lactation-related breast augmentation) and the successive involution, cancer arises from specific intermediary states and somehow maintains the molecular profile of its cell of origin [15]. It is indeed remarkable that breast cancer cell lines can be subdivided into the same molecular subtypes as primary cancer [16]. However, these observations are good in only first-degree approximation since molecular profiling of tissue whole mass cannot capture the incredible heterogeneity of cancer populations [17]. Although cellular heterogeneity severely hinders our ability to assign stem cell phenotype and markers on the single-cell level, recent advances in expression profiling of single cells [18] may shed more light on this mystery. Nevertheless, population analysis still informs our understanding of TIC markers. By employing fluorescence-activated cell sorting (FACS) and carefully appraising each marker, Keller and colleagues [1] find that CD44 is a relatively promiscuous marker whereas EpCAM, CD24, and CD49f demonstrate extensive heterogeneity within cultured populations of the investigated cell lines. Interpreting these results in terms of self-renewal and morphological phenotypes (such as mammosphere and xenograft growth efficiency) or mesenchymal appearance further demonstrated the complexity of differentiation states, as judged by a handful of markers. The authors benchmark the progenitor cell population by assuming that the overall self-renewal phenotype of a given cell line's mixed population should correlate with the abundance of the particular population allegedly capable of the self-renewal. Further support to their cell fraction-self-renewal assignment is still needed from direct cell population subfractionation by using FACS similar to that performed by Al-Hajj and colleagues [7].

Stem cell 'purification' may gain insight from a biochemical purification scheme, in which successive fractionation results in diminishing yields and increasing specific activity. For example, consider the purification of mitosis-promoting factor (MPF). For a long time, conflicting reports claimed that MPF depends on or is attenuated by phosphatase activity. Ultimately, it was recognized that the activity was dependent, in a sequential fashion, on both the kinase activity of MPF and the phosphatase activity of CDC25. It is agreed that, in normal tissue, progenitors are regulated by signals from their respective niche. However, assays for the activity of self-renewal, which not only mimic the niche more effectively but potentially involve mixing back the isolated cells with other cell populations at the onset of the assay (much to the same effect as mixing homogenous basal transcription factors in an in vitro reconstituted transcription reaction), have not yet been employed. Consequently, as was the case for MPF, it is possible that all current reports rely on mixed populations. In support of this notion, the claudin-low candidate TIC-like cancers are also elevated with leukocyte infiltrate signature [10], which could reflect the role of heterotypic interactions that regulate progenitor phenotype in vivo, but not in current model systems. Such rigorous reconstitution experiments, with trackable cell progeny, may offer new handles with which to control, rather than monitor, breast stem cells.