Cancer and Metastasis Reviews

, Volume 27, Issue 3, pp 459–470

Cancer stem cells: markers or biomarkers?


    • Department of Radiation OncologyThe University of Texas M.D. Anderson Cancer Center
  • Erik P. Sulman
    • Department of Radiation OncologyThe University of Texas M.D. Anderson Cancer Center

DOI: 10.1007/s10555-008-9130-2

Cite this article as:
Woodward, W.A. & Sulman, E.P. Cancer Metastasis Rev (2008) 27: 459. doi:10.1007/s10555-008-9130-2



The lineages assumed by stem cells during hematopoiesis can be identified by the pattern of protein markers present on the surface of cells at different stages of differentiation. Specific antibodies directed at these markers have facilitated the isolation of hematopoietic stem cells by flow cytometry.


Similarly, stem cells in solid organs also can be identified using cell surface markers. In addition, solid tumors have recently been found to contain small proportions of cells that are capable of proliferation, self-renewal, and differentiation into the various cell types seen in the bulk tumor. Of particular concern, these tumor-initiating cells (termed cancer stem cells when multipotency and self-renewal have been demonstrated) often display characteristics of treatment resistance, particularly to ionizing radiation. Thus, it is important to be able to identify these cells in order to better understand the mechanisms of resistance, and to be able to predict outcome and response to treatment. This depends, of course, on identifying markers that can be used to identify the cells, and for some solid tumors, a specific pattern of cell surface markers is emerging. In breast cancer, for example, the tumor-initiating cells have a characteristic \({\text{Lin}}^ - {\text{CD}}44^ + {\text{CD2}}4^{{ - \mathord{\left/ {\vphantom { - {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} {\text{ESA}}^{\text{ + }} \) antigenic pattern. In cells derived from some high-grade gliomas, expression of CD133 on the cell surface appears to select for a population of tumor-initiating, treatment resistant cells.


Because multiple markers, typically examined on single cells using flow cytometry, are used routinely to identify the subpopulation of tumor-initiating cells, and because the number of these cells is small, the challenge remains to detect them in clinical samples and to determine their ability to predict outcome and/or response to treatment, the hallmarks of established biomarkers.


Breast cancerFlow cytometryColorectal cancerProstate cancerPancreatic cancerReviewCancer stem cellsBiomarkersRadiation therapyProgenitors

1 Introduction

The field of solid tumor stem cell biology has re-emerged at the forefront of clinical oncology in recent years due largely to the identification of new, prospectively identified stem/progenitor cell markers. Numerous putative markers are currently under investigation and considerable work is being done to identify new ones. However, there are many challenges and considerations that researchers and clinicians must contend with to bring this field into the clinic and impact patient outcomes.

This process is beset by two main challenges: how robust are the criteria used to validate the identified cell as a stem cell, and to what degree, under what conditions, and in which patients are identification of these cells reproducible? Even if an ideal, reproducible marker is identified, the leap from marker to biomarker will not be easy. This reflects in part the very different technologies used in stem cell biology and for the detection of biomarkers. For the former, fluorescent-activated cell sorting (FACS) analysis using several different antibodies for both positive and negative selection and precise gates is usually required to isolate stem and progenitor cells. In contrast, biomarkers are usually analyzed using a single antibody by immunohistochemistry on a paraffin embedded tissue section. Furthermore, many of these cell surface molecules are both alternatively spliced and contain numerous post-translational modifications, such as glycosylation, so it is critical to use antibodies directed against appropriate epitopes and the correct isoforms to identify stem and progenitor cells. A stem cell biomarker will need to support the principle that the fate of the stem cell population determines response to treatment or outcomes from cancer and be reproducible and measurable in patient samples; a tall order considering the relatively few general biomarkers that have been successfully integrated into clinical oncology after many years of research [13].

According to the most stringent criteria, a marker deserving of the moniker “stem cell marker” should reliably identify a multipotent single cell capable of recapitulating the heterogeneity of the tumor from which it was derived. It should also be capable of self-renewal; that is, it should have the ability to divide and give rise to an exact replica of itself. It should have limitless replicative potential when needed, although the cell may be quiescent in the non-pathologic state. Unfortunately none of these criteria are easily measured or observed in vivo, thus creating the need for a surrogate marker or set of markers to identify these cells. Further complicating this effort is the reality that studying cells under ex vivo conditions may indeed change the profile of the markers [4]. In addition, the expected rarity of these cells requires sensitive techniques to measure them. To date, there are few if any single markers that fulfill all of these criteria in human solid tumor stem cell biology. Many markers have been examined that fulfill some of these criteria and have therefore been dubbed “progenitor markers.” These markers identify a subcategory of cells the stem cell hierarchy that at this point have unclear clinical significance. To the stem cell purist, finding the progenitor is not the Holy Grail. Clinically, however, it is possible that some tumors may in fact result from self-renewing mutations in progenitors, which could make progenitor markers a meaningful biomarker for these tumors. To date the least well-studied potential source of biomarkers in the stem cell field is the stem cell niche [57]. Because interactions between cancer stem cells and the niche or the stroma are widely believed to be critical components of the stem cell story, it is plausible that markers in the stromal environment or on circulating stromal or endothelial cells are as important as the markers on the cancer cells themselves. Indeed, noting that the cell adhesion molecules reported to represent significant stem cell markers in the murine mammary gland are not critical to mammary gland development has led to the proposal that these adhesion molecules may, in fact, be markers of stem cells adherent to the stem cell niche which are less easily digested during isolation [8].

Meticulous research in the area of hematopoetic stem cell markers has led to marker delineation of hematopoetic stem cells and progenitors in every lineage [9]. This work has in turn provided the paradigm for the identification of solid organ stem cell markers, predominantly in the breast (normal murine mammary gland [10, 11] and human breast cancer [12]) initially, but now expanding rapidly into other tumor sites. However, some of these markers though useful, are not amenable for use in human studies or the study of patient samples and will not be discussed here. These include cells that exclude the Hoechst dye (side population cells) which have been reported to be enriched for progenitor cells and have been identified in human breast cancer cell lines [1316], but can only be identified using a labor intensive assay that must be done in live cells, a technique that is not practical for large scale clinical application. Similarly long-term label-chase experiments using BrdU are useful for marking quiescent cells, but are not practical for clinical application. The markers we will describe are generally cell surface antigens detectable by immunologic techniques such as flow cytometry and immunohistochemistry.

In this review we will highlight the progress being made in the identification of cancer stem cell makers across disease sites, and illustrate the progress and problems in establishing these markers as “biomarkers,” i.e. surrogate markers of a clinical endpoint. The most significant body of work has been done in breast and brain tumors, and these will be examined in more detail. Of particular note, investigators have shown that cancer stem cells and progenitors in tumors of both sites are relatively resistant to radiation [1618] and the implications of these findings in terms of novel therapeutics will also be discussed.

2 Solid tumor stem cell hierarchy: progress from the normal mouse mammary gland

In the normal mouse mammary gland, transplantation of retrovirally labeled mammary gland cells into the cleared mammary fatpad convincingly demonstrated the existence of a mammary gland stem cell capable of recapitulating all components of the functional gland and of self-renewal [19]. Additional work identified several mammary gland progenitors that could be isolated in vitro [20]. Through a series of transplantation experiments using markers identified in other sites (Table 1), Stingl et al. and Shackleton et al. [10] have shown a hierarchy in the normal mouse mammary gland where the most primitive stem cell, which is capable of both self-renewal and of repopulating the mammary gland from a single cell, is the \({\text{Lin}}^ - {\text{CD24}}^{\text{ + }} {\text{CD29}}^{{\text{hi}}} \) cell. This population overlaps with \({\text{Lin}}^{\text{ - }} {\text{CD24}}^{\text{ + }} {\text{CD49f}}^{{\text{ + + }}} \) cells [10] although single cell outgrowth potential of the CD49f+ population has not been established. Stingl et al. further examined the utility of CD49f finding the \({\text{Lin}}^{\text{ - }} {\text{CD}}24^{{\text{med}}} {\text{CD49f}}^{{\text{high}}} \) population contains 1 mammary gland repopulating unit (MRU) cell per 60 cells from FVB mice and per 90 cells from C57Bl/6 mice [11]. Looking at a single marker, CD24, Sleeman et al. [21] reported the highest outgrowth potential was exhibited by a population of CD24 mammary gland cells they termed CD24lo, largely consistent with the Stingl population (CD24med) and the single cell phenotype (CD24+). Using additional markers Sleeman et al have characterized the MRU CD24lo population as consistent with a basal/myoepithelial cell [21]. Similar work explicitly examining the expression of estrogen receptor (ER)-alpha, progesterone receptor (PR), and ErbB2 on these cells confirmed that the \({\text{Lin}}^{\text{ - }} {\text{CD}}24^{{\text{med}}} {\text{CD49f}}^{{\text{high}}} \) population is an ER-negative, basal-like population which instead expresses the epidermal growth factor receptor (EGFR) [22].
Table 1

Common marker information




Heat Stable Antigen, luminal


Beta1 integrin


Hyaluronic acid receptor


Alpha6 integrin


Epithelial surface antigen (ESA), EpCam


Hematopoetic Marker


Endothelial marker


Hematopoetic Marker


Stromal marker




PROCR, protein C receptor


Activated leukocyte cell adhesion molecule (ALCAM)


Intermediate filament

Stingl et al. and Shackleton et al. additionally showed that cells from the mammary gland capable of forming colonies are lower in the hierarchy than the mammary gland repopulating units and are \({\text{Lin}}^{\text{ - }} {\text{CD}}24^{{\text{high}}} {\text{CD49f}}^{{\text{low}}} \). Sleeman et al. [23] reported similar findings regarding colony formation; that is, they found that CD24highprominin cells had the highest colony forming capability compared with \({\text{CD24}}^{{\text{high}}} {\text{prominin}}^{\text{ + }} \) and \({\text{CD24}}^{{\text{low}}} {\text{prominin}}^{\text{ - }} \) cells. Liao et al have further characterized the colony forming cells from the mouse mammary gland using the mammosphere assay (analogous to the neurosphere assay described below) adapted by Dontu et al. [24] to propagate mammary progenitor cells in vitro [4]. They observed that the mammosphere forming activity resides predominantly in the CD24+ fraction with increased efficiency in the CD24high population and that CD24 cells had virtually no mammosphere forming activity. In addition, the inclusion of two new markers from the hematopoetic stem cell field, endoglin and prion protein (PrP) define a population enriched in mammosphere forming capability \(\left( {{\text{CD24}}^{\text{ + }} {\text{PrP}}^{\text{ + }} {\text{endoglin}}^{\text{ + }} {\text{CD49f}}^{\text{ + }} } \right)\) [4], although corresponding increased outgrowth potential of these mammospheres was not demonstrated. Interestingly, in contrast to these results generating mammospheres from freshly digested tissue, when pre-cultured cells were placed into mammosphere culture the ability to form mammospheres resided in the \({\text{CD24}}^{\text{ - }} {\text{PrP}}^{\text{ - }} \) fraction. Further, the pre-cultured \({\text{CD24}}^{\text{ - }} {\text{PrP}}^{\text{ - }} \) fraction contained the ability to generate mammary ductal tree outgrowths [4] (serial transplantation/self-renewal was not reported), highlighting the complexity in assessing stem cell markers and the critical need to compare studies within the context and conditions in which they were performed.

The success in identifying stem cell markers in the normal murine mammary gland is very much dependent on the availability of a reproducible in vivo assay that can test multipotency and self renewal, namely mammary fatpad outgrowth experiments. Until recently, similar assays were not available for use with human cells [25], and as such the corollary experiments using the above markers identified in mouse have not been rigorously carried out in humans. Although mammospheres grown from both normal and pathologic human breast tissue have been characterized [24, 26], this has not been done explicitly in the context of the above markers, CD24, CD29, and CD49f. Data from the human breast cancer cell line MCF-7 demonstrated selective enrichment of \({\text{CD24}}^{\text{ + }} {\text{CD29}}^{\text{ + }} \) cells after irradiation at clinically relevant doses [16], however the significance of these findings is difficult to interpret in the absence of in vivo functional data from human \({\text{CD24}}^{\text{ + }} {\text{CD29}}^{\text{ + }} \) cells. In general, work to clarify the role of \({\text{CD24}}^{\text{ + }} {\text{CD29}}^{\text{ + }} \)or \({\text{CD24}}^{{\text{lo}}} {\text{CD49f}}^{\text{ + }} \) cells in human breast cancer is still lacking and needs to be done.

3 Human breast cancer stem cells/tumor initiating cells (\({\text{Lin}}^{\text{ - }} {\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} {\text{ESA}}^{\text{ + }} \))

The existence of human breast cancer tumor-initiating cells was strongly bolstered by the landmark paper by Al-Hajj et al. [12] demonstrating the prospective identification of a population of human breast tumor initiating cells capable of recapitulating the phenotype of the human tumors from which they were derived when injected into the cleared mammary fatpad of a mouse. Al-Hajj et al reported that as few as 200 \({\text{Lin}}^{\text{ - }} {\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} {\text{ESA}}^{\text{ + }} \) cells derived from pleural effusions of patients with metastatic breast cancer were capable of regenerating tumors in contrast to thousands of cells lacking this phenotype that did not give rise to tumors. This work has, as expected, galvanized the field and, in spite of several limitations, has spawned the most literature to date leading towards a meaningful biomarker. However, a potential drawback of the experiments reported by Al-Hajj et al. is that the cells were isolated from a small number of patients with cancers of different histologies and treatments and were handled differently ex vivo, which has given rise to questions whether data from this small diverse sample are applicable to all patients [27]. Others have questioned whether the tumor growth of human cells in a mouse fatpad identifies functional stem cells or merely cells able to overcome the engraftment incompatibility that exists when injecting human cells into a mouse fatpad of immunocompromised mice [28]. Nevertheless, comparison of the results of a gene expression analysis of the \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) cells from these cases with results for cells from normal mammary epithelium yielded a gene expression signature that predicted distant-metastasis free survival and overall survival in breast cancer as well as three other tumor types [29]. In preclinical studies, Phillips et al demonstrated that MCF7 cells with the \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) phenotype were relatively resistant to radiation, generating fewer reactive oxygen species and decreased evidence of DNA damage in response to radiation [18]. In total, the \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) phenotype exhibited several of the characteristics expected of a meaningful “cancer stem cell” biomarker when assayed by FACS analysis of freshly isolated cells. Cells expressing this phenotype from human metastatic pleural effusions are tumor-initiating, gene profiles from these cells predict for outcome in multiple tumor types, and in cell culture, cells expressing this phenotype are resistant to radiation. Preliminary studies in primary mammosphere culture from human pleural effusion cells suggest these findings of resistance to radiation in cell lines are clinically relevant and testable in fresh primary human material grown as mammospheres (Fig. 1).
Fig. 1

Fluid obtained after therapeutic thoracentesis was collected from a patient with metastatic breast cancer on an IRB approved protocol. Total cells were placed into non-adherent, growth factor mammosphere culture (Dontu et al. [24]) to select for in vitro passage of human breast cancer progenitor cells. Cytopathologic examination of cultured material confirmed malignant cells similar in pathologic appearance to the primary tumor. Primary mammospheres (P0) and secondary mammospheres (P1) (trypsinized mammospheres re-plated as single cells and allowed to regenerate new mammospheres) were irradiated. Cells were dissociated and examined using flow cytometry for the tumor initiating marker phenotype described by Al-Hajj et al. [12]. Samples contain a greater percentage of tumor-initiating cells after irradiation similar to results seen in cell line experiments [18]

Archived pathology specimens are clearly the most readily available human tumor samples, and antibody-based biomarkers amenable to immunohistochemical assays are desirable, but not easily converted from flow cytometry studies such as those described above. Efforts to correlate findings from an immunohistochemical analysis of \({\text{CD44}}^{\text{ + }} {\text{24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) expression have demonstrated the challenge in using a multi-marker biomarker. Abraham et al examined the \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) expression in paraffin embedded tissue from 136 patients with breast cancer and found no correlation with outcome. Specifically, the percentage of \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) cells ranged from 0–40% in the normal breast and from 0% to 80% in tumor tissues [30]. As with CD29 and CD49f, CD44 and CD24 are both prevalent in normal tissues, making the detection of multiple markers critical, but technically challenging. One potential solution, Shipitsin et al used the markers previously described to perform gene expression profiling on CD44+ cells from ascites fluid and pleural effusions and identified a unique marker, PROCR, that is present on 100% of CD44+ cells but not on leukocytes and myofibroblasts which also express CD44 [31]. They further reported near mutually exclusive expression on CD44 and CD24 in a small subset of digested primary invasive breast cancer tissues as well as minimal ER-alpha and ErbB2 expression on CD44+ cells, similar to the murine data using CD29 and CD49f and CD24 [31]. These data may be encouraging for the hope for a single marker to identify stem/progenitor cells.

In the absence of a single marker that can be applied to immunohistochemistry, Balic et al. employed spectral imaging in conjunction with double marker immunohistochemistry to examine the simultaneous expression of CD44 and CD24 on cytokeratin-positive, disseminated tumor cells in the bone marrow pf patients with early stage breast cancer. They reported that this was a technically feasible approach in these samples, and detected \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) cells in all 50 samples with a median prevalence of 66% [32]. The identification of this tumor-initiating phenotype on all cytokeratin positive disseminated tumor cells in the bone marrow raises interesting questions about the potential role for the detection of disseminated or circulating tumor cells as surrogate stem cell biomarkers. Circulating/disseminated tumor cells (C/DTCs) are tumor/epithelial cells in the blood or bone marrow of patients with breast cancer. In patients with metastatic breast cancer, the presence of more than five of these cells in 7.5 ml of peripheral blood predicts for overall survival [33]. These cells are detected by the presence of the epithelial cell marker CD326 (aka ESA or Ep-CAM), can be found in up to 30% of patients without known metastatic disease appreciated on standard staging studies even after systemic chemotherapy [3436] and may predict for response to treatment [37]. CTCs can be assayed in a standardized FDA approved assay and quantitated. Importantly, the work flow involved in the collection, storage, and analysis in large randomized trials has been successfully demonstrated in Europe [34]. It has been suggested that the comparatively low frequency of CTCs found in the bone marrow of early stage patients treated in a North American phase III trial may have been the result of delayed shipping or suboptimal storage of specimens, highlighting the importance of optimal sample processing if efforts to incorporate CTCs into clinical practice are to be successful. In our own practice we have found that 75–90% of disseminated tumor cells in the bone marrow are \({\text{CD44}}^{\text{ + }} {\text{24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \), and interestingly similar \({\text{CD24}}^{\text{ + }} {\text{CD29}}^{\text{ + }} \)flow cytometry profiles similar to that seen in the normal mouse mammary gland have been found when analyzing the CD326+ cells in the bone marrow of patients with early stage breast cancer (Fig. 2, ARC-MD*, unpublished data).
Fig. 2

Disseminated human tumor cells (CD326+) in the bone marrow of a patient with non-metastatic breast cancer demonstrate similar distribution of expression as mouse mammary epithelial stem cells. Left: Published flow cytometry profile of mouse mammary epithelial cells capable of recapitulating the functional mammary gland from a single cell (reprinted with permission (Shackleton et al. [10]). Right Cells from bone marrow aspirate obtained on an IRB approved protocol from an early stage breast cancer patient prior to undergoing definitive surgery. Cells were enriched for CD326+ cells using magnetic bead sorting and analyzed by flow cytometry for lineage negative cells (unpublished data from the Advanced Research Center for Micrometastatic Disease, ARC-MD)

4 Pancreas

The tumor-initiating studies in breast cancer have now served as the springboard for similar studies in pancreatic carcinoma. In one effort, Li et al. isolated tumor cells from ten patients with adenocarcinoma of the pancreas (eight with non-metastatic tumors, two with metastatic tumors). Prior treatment not reported and injected increasingly smaller numbers of prospectively identified populations of cells into the peritoneal cavity of non-obese diabetic/severe combined immunodeficient (NOD/SCID) mice using the CD44, ESA, and CD24 markers [38]. These investigators found that the tumor initiating capability resided in the \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{\text{ + }} {\text{ESA}}^{\text{ + }} \) fraction, with 50% vs. 0% tumor initiation when 100 cells were injected. Tumors were histologically similar to the original primary tumors and were serially passaged four times, thus demonstrating self-renewal. It is not clear if the difference in CD24 expression between tumorigenic breast cancer cells and pancreatic cancer cells is a function of flow cytometry gating, a question rasied by the finding that the tumorigenic potential resides predominantly in the low/intermediate population found in the \({\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) population in the work by Al-Hajj et al. but in the CD24+ population in the work by Li, or if this represents a difference in stem cell origin, i.e. luminal versus basal, in these organs. Findings made in studies of other organ systems have also been extrapolated to pancreatic cancer. Specifically, prostate stem cell antigen (PSCA, discussed below), a putative stem cell marker in prostate cancer [3942], CD133 from central nervous system tumors (discussed below) [43], ABCG2, the molecular pump that confers the side population phenotype [43], and numerous embryonic and hematopoetic stem cell markers have also been examined in pancreatic cancer [39, 44].

5 Prostate

Data supporting the existence of stem cells in the prostate mirror data obtained in studies of the mammary gland, which is not surprising given the similarities between these structures. That is, like the breast, the prostate is a hormone-responsive gland that can undergo remarkable regression and regeneration in response to hormonal regulation [45, 46]. Indeed, the transplantation of freshly isolated mouse and human prostatic tissue can regenerate defined prostatic glands [47, 48], and label-retention studies have localized quiescent putative stem cells in the normal gland to the proximal prostatic tubules [49]. Candidate prostate stem cells appear to be derived from the basal compartment and do not express androgen receptors [50, 51]. In mouse prostate and mammary gland cells expressing the hematopoietic stem cell markers, stem cell antigen-1 (Sca-1, a member of the Ly6 family with no known human homologue) has been demonstrated to be enriched in stem/progenitor cells with high regenerative potential [15, 20, 52, 53]. Interestingly, over 60% of the Sca-1-expressing prostate cells co-express CD49f [52], and the Sca-1-expressing cells appear to initiate murine prostate tumors [53]. Although Sca-1 lacks a human homologue, extensive work has been done on PSCA, a homologue of the Ly-6/Thy-1 family of cell surface antigens, expressed by a majority of human prostate cancers as well as bladder and transitional cell tumors [54, 55]. Immunohistochemical analysis of 246 patient samples further demonstrated a significant correlation between PSCA intensity and adverse prognostic features, such as a high Gleason score and the presence of extra-organ disease [56]. Of particular note, immunotherapy targeted against PSCA in human prostate cancer xenograft mouse models has been found to inhibit subcutaneous and orthotopic xenograft tumors in a dose-dependent manner. Furthermore, administration of anti-PSCA monoclonal antibodies led to retardation of established orthotopic tumor growth and inhibition of metastasis to distant sites, resulting in a significant prolongation in the survival of tumor-bearing mice [57]. Thus, in prostate and possibly breast cancer, Sca-1 family members may have potential as progenitor-cell biomarkers, although the MRUs were Sca-1lo.

Other markers whose potential has been evaluated in other disease sites, including CD44, α2β1and CD133 [5862], have also been examined in prostate carcinoma as candidate stem cell markers. Uniquely, Lui et al. demonstrated not only that CD44+ cells show characteristics of progenitor cells, but also that these and the terminally differentiated luminal CD57+ cells can be induced to express markers of differentiation when co-cultured with stromal cells. These data highlight the critical role of cancer cell-stromal interactions in cell functional and in marker identification [62].

Recently the prospective isolation of self-renewing cells from human prostate tumors has been reported [59]. Collins et al report that approximately 0.1% of prostate tumor cells express the CD44+ ,\(\alpha _{\text{2}} \beta _1^{{\text{hi}}} {\text{CD133}}^{\text{ + }} \) phenotype and that these cells maintained in long-term culture possess the capacity for self-renewal and multipotent differentiation in vitro. The in vivo tumor-initiating capacity in humans remains to be reported, however, although Patrawala et al. [58] have demonstrated that the tumor-initiating capacity resides in the CD44+ cells from human xenograft samples. While there is little or no data regarding the potential predictive or prognostic value of CD44 positivity alone or in combination with other markers in human prostate cancer, it is interesting that prostate cancer is one of the disease sites for which the invasiveness gene signature identified in \({\text{CD44}}^{\text{ + }} {\text{CD24}}^{{{\text{ - }} \mathord{\left/ {\vphantom {{\text{ - }} {{\text{lo}}}}} \right. \kern-\nulldelimiterspace} {{\text{lo}}}}} \) human breast cancer cells predicts for overall survival [29].

6 Squamous cell carcinomas of the head and neck

In samples of tumors from patients with head and neck cancer the CD44+ fraction has likewise been shown to contain the tumorigenic fraction of cells capable of recapitulating in vivo tumors in two immunodeficient mouse models [63]. As with similar studies in breast cancer, however, this work is limited by the scarcity of patient samples and the fact that the material examined comes from patients with a range of head and neck primaries treated with varying regimens. Nevertheless, Prince et al were able to show that LinCD44+ cells are capable of tumor initiation and self-renewal and are predominantly basal cells that co-localize with cells positive for the stem cell survival factor bmi-1, as shown by immunohistochemical analysis [63]. Substantially more clinical data regarding CD44 as a biomarker are available in head and neck patients compared to other tumor sites. Soluble CD44 isoforms from serum have been correlated to stage, and levels were significantly higher in samples pre-treatment compared to post-treatment and compared to healthy controls [64, 65]. Similar work comparing CD44 isoforms from oral rinses in 102 head and neck cancer patients to 69 controls revealed high specificity for CD44 (75–88% measured by enzyme linked immunosorbant assay) and high sensitivity when combined with CD44 methylation status [66]. Recently a novel CD44 isoform, CD44v3 was demonstrated in human head and neck cancers and cell lines. Its expression was 4.5 times higher in tumor human tumor than in true normal specimens [67]. Although phase I clinical trials of an antibody against CD44v6 yielded discouraging results, and were discontinued because of problems with excessive skin toxicity [68], in vitro studies showing the inhibitory effects of the CD44 ligand hyaluronin as well as signaling through the epidermal growth factor receptor (EGFR) which co-localizes with CD44 through hyaluronin interactions [69] may provide less toxic therapeutic targets for clinical trials.

7 Central nervous system

The concept of a common progenitor cell of the constituent cellular components of the mammalian central nervous system is of relatively recent advent. Neurons and glia have been historically believed to be derived from distinct sets of precursors [70]. However, work in invertebrates suggested that this may not be the case and that a common progenitor does exist [71]. However, the long-held view that no new nerve cells formed postnatally contradicted the possibility of germinal centers containing neural stem cells (NSCs) within the brain. The identification of neural regeneration in adult vertebrates, demonstrated in the telecephalon of birds, led to the end of this established view [72, 73]. In fact, regions containing germinal centers of neurogenesis in mammals have been identified in the subgranular layer of the hippocampal dentate gyrus and in the subventricular zone (SVZ) of the lateral ventricles [7476]. Indeed, a persistent population of NSCs in the SVZ has now been characterized in mammals, including humans [77, 78]. In addition, three cell types within the SVZ have been reported, which together describe a lineage for mature astroglia [79].

The model that has been devised to describe the NSCs is as follows [80]. First, the neuroblasts of the SVZ, which is separated from the lateral ventricle by an ependymal cell layer, migrate as chains of type A cells oriented in parallel to the ventricles through tubes composed of processes of type B cells. Type C cells are distributed within the chains and divide rapidly to produce the type A cells. The slow-growing type B cells are the precursors to the type C cells [80]. The type A cells bare an antigenic profile positive for the polysialylated neural adhesion cell molecule (PSA-NCAM), the neuronal marker Tuj1 (β-III tubulin), and nestin and negative for the glial-fibrillary acidic protein (GFAP) and vimentin. In contrast, type B cells are positive for GFAP, vimentin, and nestin but negative for Tuj1 and PSA-NCAM. Type C cells are positive only for nestin. In their study of the migration of neurons to the olfactory bulb, Doetsch et al. [81] demonstrated that it is the type B cells of the SVZ that represent the true NSCs, capable of self-renewal and differentiation into neurons and glia.

Several lines of evidence have led to the proposal that the NSCs of the SVZ also serve as the origin of brain tumor cancer stem cells [81]. For example, the deregulation of pathways responsible for G1 cell cycle arrest in astrocytes was observed to lead to a loss of senescence of astrocytes and genomic changes consistent with glioma tumorigenesis [82]. Further, inactivation of both p53 and NF1 in mice was found to lead to the formation of malignant gliomas [83]. Finally, the activation of Ras and Akt in NSCs in a mouse model was noted to lead to the formation of glioblastomas (GBMs) [84]. These collective data suggest that NSCs are a likely origin for brain tumor cancer stem cells, particularly for astrocytic tumors, the most lethal of which is GBM.

NSCs have been isolated from normal mammalian brains by the neurosphere assay, which measures both self-renewal and proliferation capabilities and permits the identification differentiation lineages in vitro [85, 86]. In this method, tissue is disrupted into single cells and cultured in the presence of epidermal growth factor and basic fibroblast growth factor until the non-adherent cells form three-dimensional spheres that are enriched for NSCs [86]. Cancer stem cells were first isolated from GBMs by Ignatova et al. [88], who use a modified neurosphere assay to isolate tumor clones from cortical glial tumors and compared them to normal brain-derived NSCs, which showed that the cancer stem cells expressed many NSC markers but not markers of normal lineage differentiation pathways. In another study in which a panel of 14 pediatric brain tumors was examined, consisting of eight medulloblastomas, three World Health Organization grade 1 astrocytomas, one ependymoma, and one ganglioma, Singh et al. [89] utilized the neurosphere assay to identify a population of tumor precursor cells. Those cells that were capable of self-renewal, proliferation, and differentiation proved to be positive for CD133 (AC133, Prominin-1). In contrast, the CD133 cells neither formed neurospheres nor did they express undifferentiated cell markers such as nestin. In another study, the CD133+ cells derived from GBMs were capable of initiating brain tumors in vivo in NOD/SCID mice, which allowed their characterization as legitimate cancer stem cells [87]. Indeed, the tumors that formed from these cells recapitulated the histologic characteristics of GBMs [88, 89], including invasion, micro-vascular proliferation, and pseudo-pallisading necrosis, as well as the expression profile and genotype of the tumors from which they were derived [90]. This was not the case for cells from medulloblastoma-derived neurospheres, for which in vivo tumor initiation has net yet been demonstrated [88, 91]. This suggests that different conditions are needed for the isolation and proliferation of medulloblastoma cancer stem cells in vitro compared with glioma cancer stem cells [88].

Gene expression profiling of 103 ependymomas identified a signature similar to that of embryonic radial glial cells (RGCs), and neurosphere assays of cells from ependymomas identified a small subpopulation of RGC-like cells believed to represent the cancer stem cells for this tumor type [92]. These RGC-like cells were found to have a cell-surface antigen profile similar to that of the NSCs from the SVZ but with additional RGC-specific markers. Specifically, they were found to stain positively for CD133 and nestin as well as the radial glial markers RC2 and brain lipid-binding protein. The selection of CD133+ cells and their transplantation into nude mice resulted in the formation of tumors with characteristic histologic and cytologic features of ependymomas, such as moderate cellularity, a predominantly monomorphic nuclear morphology, and occasional pseudorosettes; these cells also recapitulated the gene expression signature of the parent tumor [92].

In the case of GBM-derived CSCs, Bao et al. [9] observed that CD133 selects for a population of cells that exhibit radiation resistance [17]. In particular, they found that CD133+ cells exhibited increased clonogenic survival, decreased apoptosis, and upregulation of the double-strand DNA repair pathway following ionizing radiation. In in vitro cell mixing experiments with fluorescently labeled CD133+ and CD133 cells, Bao et al. further showed that radiation induced a preferential proliferation of CD133+ cells. In addition, irradiated CD133+ cells retained their in vivo tumor-initiating capacity, in contrast to the behavior of CD133 cells (in the specific case of a CD133 population that was capable of forming tumors when not irradiated). CD133+ cells also showed greater upregulation of the ATM-mediated DNA repair pathway compared with their negative counterparts [17]. The fact that a population of CD133 cells was capable of tumor initiation suggests that additional markers are needed to better characterize GBM-derived cancer stem cells.

The identity of additional markers of glioma cancer stem cells may come from the wealth of expression profiling data available for these tumors [9396]. In this regard, Phillips et al. [98] described a mesenchymal/angiogenic gene signature in high-grade gliomas that is prognostic for poor survival. It is possible that markers from this expression signature are specific to cancer stem cells, which have been shown to promote angiogenesis through a vascular endothelial growth factor-dependent mechanism [97]. Expression profiling of neurosphere-derived cells from GBMs before and after serum-induced differentiation in vitro may also serve as a source of novel GBM cancer stem cell markers [90]. In this regard, Piccirillo et al. [100] demonstrated that CD133+ cells express bone morphogenic proteins (BMP) and receptors. The in vitro treatment of GBM-derived cells with BMP4 decreased the CD133 levels and abolished tumorigenicity, suggesting not only a therapeutic strategy but also that BMPs and BMP receptors are markers of cancer stem cells. Ultimately, it is likely that a panel of markers will be required to best assess the presence of true cancer stem cells in tumors. An additional goal, given that most of the work to date has been accomplished using fluorescently labeled markers on fresh tumor specimens, is to develop reliable immunohistochemistry-based markers suitable for tissue sections.

8 Colorectal carcinoma

Two recent studies reported simultaneously have identified a small population of tumor-initiating cells that are positive for CD133 in colorectal cancer specimens [98, 99]. In the study of O’Brien et al. [98], a NOD/SCID renal capsule xenograft model was developed and implanted with a panel of colorectal cancer cell lines. Based on the data for CD133 in other solid tumors, these researchers hypothesized that the cancer stem cells would be CD133+. Indeed, tumors formed in only one of 47 mice injected with 2.5 × 105 CD133cells (the highest number used). In contrast, tumors formed in 45 of 49 mice injected with as few as 1,000 CD133+ cells. Limiting dilution to as few as 100 CD133+ cells, however, resulted in tumor formation in only one of four mice [98]. Ricci-Vitiani et al. [99] observed similar results for the CD133+ cells in a subcutaneous model [99]. In both cases, CD133 was reliably detected by immunohistochemistry in small groups of cells in tumor sections.

In a more comprehensive analysis of solid tumor cancer stem cell markers in colorectal cancer, Dalerba et al. [100] examined the tumorigenicity of cells expressing EpCAM,, CD44, CD166, and CD133. These authors demonstrated that in vivo engraftment from colorectal xenografts was established in six of six transplants of 200–500 \({\text{EpCAM}}^{{\text{high}}} {\text{CD44}}^{\text{ + }} \) cells. Tumors generated in these experiments were histologically and phenotypically similar to the patient starting material. In matched samples of freshly digested normal and primary tumor epithelial, the \({\text{EpCAM}}^{{\text{high}}} {\text{CD44}}^{\text{ + }} \) population was consistently larger in tumor samples. Systematic evaluation of previously described markers revealed CD49f was detectable on most tumor cells but was higher on CD44+ cells. CD133+ expression was variable with some tumors scoring homogenously negative both in the primary and in serially transplanted tumors from CD133 primaries. CD44+ cells were a clearly identifiable smaller subset of CD133+ cells in tumors that contained CD133+ cells suggesting \({\text{EpCAM}}^{{\text{high}}} {\text{CD44}}^{\text{ + }} \) cells may be a more selected tumorigenic subset of CD133+ cells. Aldehyde dehydrogenase enzymatic activity was also examined, and while useful for isolation of tumorigenic cells from some tumors, this was inconsistent among xenografts. Lastly these authors identified a new marker, CD166 which was differentially expressed in both xenografts and primary tumors, and the authors propose based on tumorgenicity experiments from two freshly digested human primary colorectal tumors that colorectal cancer stem cells are enriched in the EpCAMhigh, \({\text{CD44}}^{\text{ + }} {\text{CD166}}^{\text{ + }} \) population [100].

9 Conclusions

Clearly the work being done to identify biomarker surrogates for cancer stem cells is growing exponentially, but despite this, no single marker in any site has emerged as the definitive solution. The explosion of new data in this exciting field makes it impossible here to review all of the new findings to date. If indeed cancer stem cells are mediators of recurrence, methods to identify and target these cells will represent a significant advance in cancer therapy. Data from several tumor sites now suggest that tumor-initiating cells can be resistant to therapy, and thus far, all of the studies examining the resistance of stem or progenitor cells to radiation therapy indicate that there are selective signaling pathways unique to the stem cell/progenitor cell population that may be explicitly targeted to sensitize these cells [1618, 101, 102]. Given the data in head and neck cancer suggesting EGFR may interact with or be a co-regulator of a stem/progenitor cell pathway, the recent demonstration in a phase III randomized trial that a targeted anti-EGFR therapy improved outcomes when added to radiation therapy for the treatment of head and neck cancer [103] is encouraging. A clear connection between radiation resistance of cancer stem cells in head and neck cancer and the EGFR signaling pathway would provide strong evidence that the cancer stem cell is indeed the true target for radiation, and for the clinical relevance and potential of cancer stem cell directed therapies overall.

The challenge ahead is to incorporate the most promising of these markers into clinical trials. The answers to which marker, what method, what material, and which patients are critical to the success of such trials and the development of novel therapies. Thus far, prevailing observations across disease sites are that cancer stem cells are often basal cells, devoid of markers of differentiation such as hormone receptors, and can exhibit similar adhesion molecule profiles. Commonalities among tumor-initiating cell surface markers have facilitated tumor-initiating cell identification in multiple tumor sites; however, the impact of tissue digestion on marker specificity must be evaluated in order to resolve the inability to identify meaningful markers to be used on intact tissue sections or archived tissue. Cell surface molecules while ideal targets for FACS analysis, may not provide the optimal targets for IHC.

In the meantime, the CTC population is an appealing substitute as these cells need not be subjected to digestion for evaluation, and appear to express the markers representative of tumor initiating cells, hypothetically because they are indeed seeking a niche in which to sustain new disease foci. Prior and ongoing studies have established the feasibility of collecting CTCs in the randomized trial setting, and technology is emerging to facilitate systematic examination and quantitation of these cells. Recently, the US Food and Drug Administration (FDA) approved an assay for the detection of circulating tumor cells in peripheral blood using a semi-automated system, the CellSearch™ system, and other approaches not yet approved by the FDA are under investigation. The presence of the tumor-initiating cell surface markers on disseminated cytokeratin positive cells in the bone marrow of women with early stage breast cancer is provocative evidence that this phenotype not only marks the cells capable of initiating primary in situ tumors but also cells capable of forming metastatic lesions. As such, these cells circulating in the blood or disseminated in the bone marrow may prove to be cancer stem cell biomarkers and further study of this population may provide new insights into cancer stem cell targets and mechanisms of resistance. Although still in the early stages, this is an exciting time with significant potential to improve cancer outcomes through the emergence of novel agents both to target direct and to radiosensitize cancer stem cells. Particularly in this time of rapid discovery and enormous enthusiasm, care must be taken to move forward with rigor in study design, sample collection, sample processing and reporting [104106] to avoid rejecting a potentially critical advance in the treatment of solid tumors: cancer stem cell targeted therapy and biomarkers.


We thank Jeff Rosen and Betty Notzon for critical review of this manuscript. We also acknowledge contribution of the collaborative work of the Advanced Research Center for Micrometastatic Disease at The University of Texas M.D. Anderson Cancer Center, in particular Massimo Cristofanilli, James Reuben, Anthony Lucci, Savitri Krishnamurthy, Wendy A. Woodward, Li Li, and Hui Gao.

Copyright information

© Springer Science+Business Media, LLC 2008