Author contributions: P. M.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing; C.A.D.: collection and/or assembly of data, data analysis and interpretation, final approval of manuscript; D.P.: collection and/or assembly of data, final approval of manuscript; R.A.: collection and/or assembly of data, collection and/or assembly of data, final approval of manuscript; M.G.: collection and/or assembly of data, final approval of manuscript; M.J.: collection and/or assembly of data, final approval of manuscript; L.H.: provision of study material or patients, final approval of manuscript; L.P.: collection and/or assembly of data, final approval of manuscript; A.L.: collection and/or assembly of data, final approval of manuscript; S.G.: collection and/or assembly of data, final approval of manuscript; C.A.G.: conception and design, financial support, provision of study material or patients, final approval of manuscript; P.W.K.L.: conception and design, financial support, data analysis and interpretation, manuscript writing.
First published online in STEM CELLS EXPRESS November 23, 2010.
Disclosure of potential conflicts of interest is found at the end of this article.
Cancer stem cells (CSCs) are proposed to initiate cancer and propagate metastasis. Breast CSCs identified by aldehyde dehydrogenase (ALDH) activity are highly tumorigenic in xenograft models. However, in patient breast tumor immunohistological studies, where CSCs are identified by expression of ALDH isoform ALDH1A1, CSC prevalence is not correlative with metastasis, raising some doubt as to the role of CSCs in cancer. We characterized the expression of all 19 ALDH isoforms in patient breast tumor CSCs and breast cancer cell lines by total genome microarray expression analysis, immunofluorescence protein expression studies, and quantitative polymerase chain reaction. These studies revealed that ALDH activity of patient breast tumor CSCs and cell lines correlates best with expression of another isoform, ALDH1A3, not ALDH1A1. We performed shRNA knockdown experiments of the various ALDH isoforms and found that only ALDH1A3 knockdown uniformly reduced ALDH activity of breast cancer cells. Immunohistological studies with fixed patient breast tumor samples revealed that ALDH1A3 expression in patient breast tumors correlates significantly with tumor grade, metastasis, and cancer stage. Our results, therefore, identify ALDH1A3 as a novel CSC marker with potential clinical prognostic applicability, and demonstrate a clear correlation between CSC prevalence and the development of metastatic breast cancer. STEM CELLS 2011;29:32–45
Tumors of breast and other cancers are composed of a heterotypic population of cells. Increasing evidence suggests that only a subset of these cells have tumor-propagating ability [1, 2]. In comparison with the bulk of tumor cells, a relatively small number of these tumor-propagating cells, often referred to as cancer stem cells (CSCs), can form tumors in immunocompromised mice. CSCs are discretely isolatable based on the expression of certain markers. It has been proposed that CSCs have some of the properties of normal stem cells, as tumors that arise from CSCs are heterogeneous in nature, consisting of differentiated non-CSCs and self-renewing tumorigenic CSCs. This has led to the contentious and much debated hypothesis that cancer originates from mutated stem cells. Regardless of the origin of these cells, it is now generally accepted that there are distinct, isolatable tumor populations in most cancers, which possess varying degrees of tumor-propagating ability. It has been suggested that the prevalence of these cells in patient tumors could correlate with the relative aggressiveness of the cancer, and therefore, CSCs are potential prognostic indicators in patients with cancer .
Breast CSCs were originally isolated based on cell surface marker expression, that is, CD24−/lowCD44+ . More recently, “functional” markers that are based on stem cell characteristics are being investigated for their potential use in the isolation of breast CSCs. Using this approach, Ginestier et al. isolated breast CSCs applying the aldefluor assay (Stemcell Technologies), an enzyme-based assay that detects aldehyde dehydrogenase (ALDH) activity. The aldefluor assay was originally designed to isolate viable hematopoietic stem cells . The assay is thought to specifically detect ALDH isoform ALDH1A1 activity.
One of 19 ALDH isoforms expressed in humans, ALDH1A1 is a detoxifying enzyme responsible for oxidizing aldehydes to carboxylic acids . It is predominantly expressed in the epithelium of testis, brain, eye, liver, kidney, and is also found in high levels in hematopoietic and neural stem cells [4–6]. This enzyme is thought to play a role in the differentiation of hematopoietic and neural stem cells via the oxidation of retinal to retinoic acid . Retinoic acid activates nuclear retinoic acid receptors (RARs) and RARs subsequently regulate the transcription of genes with retinoic acid response elements. Furthermore, ALDH1A1 is known to metabolize and detoxify chemotherapeutics like cyclophosphamide , and is therefore thought to contribute to the innate chemotherapeutic resistance properties of hematopoietic stem cells. Using the aldefluor assay, Ginestier et al.  showed that aldefluor+ breast cancer cells were highly tumorigenic and had the renewal/differentiation properties of CSCs.
CSC quantification is a proposed prognostic indicator, however, translating this into clinic use requires immunohistological methods for identification of CSCs in fixed tumor tissue, and in this respect, the data is less convincing [2, 9–16]. Specifically, expression of the ALDH1A1 isoform correlated with higher tumor grade but not the important prognostic indicators of cancer stage and metastasis . In contrast, for a rare highly aggressive breast cancer subtype, inflammatory breast cancer, Charafe-Jauffret et al.  found a significant correlation with ALDH1A1 expression and development of metastasis and worse outcome. Despite the positive correlation with this rare form of breast cancer, others have failed to show a significant correlation with ALDH1A1 prevalence and higher tumor grade, metastasis, therapeutic resistance, or patient outcome with breast cancer in general [12–14, 18]. This has led to the suggestion that identifying other novel CSC markers could lead more prognostic relevance, and has resulted in some scepticism as to the validity of the existing identified markers .
In this study, we used microarray gene expression analysis and immunofluorescence studies of patient tumor samples and RNAi knockdown of breast cancer cell lines to show that the aldefluor activity of breast CSCs is primarily due to isoform ALDH1A3 and not ALDH1A1. Immunohistological analysis of fixed patient breast tissues also showed good correlation between ALDH1A3 levels and metastatic disease, linking the development of metastatic breast cancer with increased CSC numbers.
MATERIALS AND METHODS
Expansion of Primary Human Breast Cancer Tumors by Mammary Fat Pad Implantation in Non-obese Diabetic Severe Combined Immunodeficiency (NOD/SCID) Mice
Core tumor samples of invasive ductal carcinoma were obtained from patients with breast cancer at the time of their surgery (patient consent given and ethics approval obtained from Research Ethics Board). Pieces were implanted into the upper mammary fat pad of 8-week-old female NOD/SCID mice purchased from Charles River (Wilmington, MA) along with a 17β-estradiol pellet (Innovative research of America, Sarasota, FL). Postimplantation, if a palpable tumor grew in the mice; it was harvested and above procedure repeated for passaging of the tissue.
Aldefluor Assay of Tumor Cells Performed by Fluorescence-Activated Cell Sorting
Postexpansion in the mammary fat pad of NOD/SCID mice, tumor tissue was harvested from euthanized mice and digested with a 225 U/ml of collagenase III (Bioshop, Burlington, Canada) and red blood cells lysed. Cells were stained with anti-H2Kd (mouse histocompatibility class I) conjugated to alexafluor 647 nm (ebioscience, San Diego, CA) to eliminate contamination of noncancer cells of mouse origin and viability dye 7-aminoactinomycin D (7AAD) (BD Pharmingin, Mississauga, Canada) to discard dead cells. Side scatter and forward scatter were used to eliminate debris. The BAAA Aldefluor assay substrate, with or without the addition of diethylaminobenzaldehye (DEAB) ALDH inhibitor was added to identify and isolate aldefluor+ cells, as per the manufacturer's instructions (Stemcell Technologies, Vancouver, Canada). Desired cell populations were isolated using a FACSAria flow cytometer (Becton Dickinson, Mississauga, Canada). Of note, the standard purity check could not be performed postsort as the sorted cells were no longer in the aldefluor buffer but in sheath fluid. In the absence of the aldefluor buffer, which blocks ABC transporters, the fluorescent substrate is not retained in the cells. In lieu of the purity test, we performed tumorigenicity assays (described in the following section) to confirm that aldefluor-positive populations were at least 10-fold more tumorigenic than the aldefluor-negative populations.
Alternatively, in the rare instances that sufficient tumor sample was obtained from the breast cancer patient surgery, the tumor tissue was digested and labeled for the isolation of live aldefluor+ or aldefluor− tumor cells without prior expansion in the mouse. In these instances, the cell recovery was low, limiting the experimental application. As such, total isolated cell populations were frozen in Trizol RNA isolation solution (Invitrogen, Burlington, Canada) and samples sent to Miltenyi biotech (Auburn, CA) for one-color Aligent whole genome expression microarray analysis.
Tumorigenicity Studies with Aldefluor+ and Aldefluor− Tumor Cell Isolated Populations
Isolated aldefluor+ and aldefluor− tumor cells from mouse passaged tumors were injected into the mammary fat pads (fat pad number four) of 8-week-old NOD/SCID mice that had been implanted with a 17β-estradiol pellet 1 week prior. Either 3,500 or 35,000 tumor cells with 50,000 immortalized stromal fibroblasts (half of which were gamma irradiated with 4 Gy, a gift from Dr. Weinberg, MIT) were admixed 1:1 with matrigel (BD Bioscience, Mississauga, Canada) and were injected into the mice. Mice were monitored weekly for the development of tumors for up to 3 months.
Immunofluorescence of Aldefluor-Sorted Tumor Cell Populations
Aldelfuor+ and aldefluor− isolated tumor cells were cytospun on to slides for immunofluorescence analysis. Slides were labeled with anti-human ALDH1A1 (ab51028), ALDH2 (ab70917), ALDH5A1 (ab65469), ALDH6A1 (ab65471), ALDH7a1 (ab51029) purchased from Abcam, Cambridge, MA, or ALDH1A3 (Abgent, San Diego, CA, AP7847a), ALDH4A1 (Abnova, Walnut, CA, 1A12-A5), or ALDH18A1 (Sigma-Aldrich, Oakville, Canada, HPA012604) followed by species-specific alexafluor 488 nm-conjugated secondary antibody (Invitrogen). Nuclei were labeled with To-Pro-3 (Invitrogen) and slides were mounted with mounting media for fluorescence (Vector Laboratories Inc., Burlingame, CA). Images were captured with a Zeiss LSM 510 laser scanning confocal microscope.
Cell lines, shRNA Molecular Constructs, and Reagents
MDA-MB-468, SKBR3, MDA-MB-435, BT-20, MCF7, T47D, and MDA-MB-231 breast cancer cell lines (American Type Culture Collection, ATCC, Manassas, VA) and the Phoenix packaging cell line (Dr. Nolan, Stanford University) were cultivated in Dulbecco's Modified Eagle's Medium, 10% fetal bovine serum (Invitrogen), supplemented with glutamine and sodium pyruvate (Invitrogen).
To generate ALDH isoform knockdowns, retroviral vector pSMP (Open Biosystems, Huntsville, AL) with either the shRNAmir scramble sequence or shRNAmir sequences specific to each ALDH isoform (supporting information Table 1) were transfected into Phoenix cells following standard procedures. The retroviral supernatants were applied to cultured MDA-MB-468, SKBR3, and MDA-MB-435 cells and stable transfectants were selected with puromycin (Sigma Aldrich).
Quantitative RT-PCR and Western Blotting
To detect ALDH isoform levels in the control cells and respective ALDH knockdown cells, RNA was extracted with TRIzol (Invitrogen) and was reverse transcribed using the Superscript II reverse transcriptase kit (Invitrogen) as per manufacturer's instructions. Quantifast SYBR real time polymerase chain reaction (RT-PCR) kit (Qiagen, Mississauga, Canada) with gene-specific primers (supporting information Table 2) was used as per manufacturer's instructions. Standard curves were generated to calculate relative level of mRNA compared with internal standard control glyceraldehyde 3-phosphate dehydrogenase (GAPDH).
Western blotting of cell lysates performed using the same ALDH isoform-specific antibodies described earlier, followed by secondary species-specific horseradish peroxidase-conjugated goat anti-rabbit IgG (Jackson Immunoresearch, West Grove, PA) were used to detect ALDH isoforms protein levels. β-actin was detected using specific mouse monoclonal antibody (Santa Cruz Biotechnologies, Santa Cruz, CA). Immunoreactive proteins were detected by enhanced chemiluminescence (GE Healthcare, Baie d'Urfe, Canada) and visualized with a Typhoon 9400 imager.
Tumor Tissue Histological Analysis by Immunofluorescence Microscopy
Tumor core biopsy tissue taken postsurgery from consenting patients under the age of 80, who were diagnosed with breast cancer since 2007 at the Queen Elizabeth II Health Sciences Centre (QEII HSC) in Halifax, Nova Scotia, Canada, were formalin fixed and paraffin embedded. Staff pathologists at the QEII HSC conducted standard pathological assessment of tumors (supporting information Table 3) and the panel of patients was blinded in all subsequent studies. Sequential sections of 5 μm were cut and mounted on microscope slides. After antigen retrieval and blocking, slides were stained with the above described ALDH isoform-specific primary antibodies and/or anti-human CD24 (mouse IgM, Labvision, Fremont, CA) and anti-human CD44 (mouse IgG2a, Labvision). Secondary antibodies specific to the species or to the primary antibody Ig subclass for dual labeling were conjugated to either Alexafluor 488 or 546 (Invitrogen). Nuclei were stained with To-Pro-3 (Invitrogen). Washed slides were mounted with vectashield mounting media (Vector Laboratories, Inc) and images were captured with a Zeiss LSM 510 laser scanning confocal microscope.
To quantify the number of positive tumor cells in each sample, multiple thin sections were stained to ensure that representative areas of the entire tissue were assessed. Five to nine random images were taken of each thin section applied to a slide, to ensure representative area of the entire thin section was captured. To estimate the percentages of positive tumor cells, a grid of 16 equal squares was applied to each image. The total positive and negative cells in five preselected squares were counted for all the processed slides. The final positive cell number per tumor tissue was calculated based on the average percentage over all the images taken of multiple thin sections per tumor tissue sample.
Breast CSCs Identified by the Aldefluor Positivity are Tumorigenic in Mice, but ALDH1A1 Is a Relatively Poor Marker for Metastatic Cancers
In our studies with breast CSCs, we first confirmed our sorted aldefluor+ breast tumor cells from patient-derived breast tumors that had been passaged in mice (Fig. 1A) had enhanced tumorigenicity compared with isolated aldefluor− tumor cells. As few as 3,500 aldefluor+ tumor cells initiated tumors in the mice, whereas as many as 35,000 aldefluor− tumor cells did not (n = 3, Fig. 1B), confirming the positive correlation between ALDH activity and tumorigenicity.
We then performed immunohistological studies on a panel of fixed breast cancer patient tumor thin sections (supporting information Table 3) with the currently proposed breast CSC markers. We detected between 0% and 40% ALDH1A1+ tumor cells, with a median of 6.9%, in 47 patient tumor samples. When we separated the patient samples based on clinical prognostic indicators, ALDH1A1 expression failed to correlate with tumor grade, proximal metastasis, tumor size, or cancer stage (Fig. 1C, 1D), in agreement with earlier studies [13, 14]. Although we did note a nonsignificant correlation between higher tumor grade and increased ALDH1A1 expression, similar to another study , that could potentially be significant if the patient sample size were increased.
Importantly, CD44, a previously described CSC marker  that has also been identified, independently of the CSC studies, as a predictor of tumor grade [19–21], was comparatively a much better predictor of tumor grade than ALDH1A1 (Fig. 1C, 1D). Further, CD44 alone was a better prognostic indicator than ALDH1A1 and CD44 combined, again reflecting the poor prognostic potential of ALDH1A1. We also stained the same panel of tumor sample for the prevalence of CD24−/CD44+, CD24+/CD44+, and CD24+ tumor cells and found that CD24−/CD44+ was similar to CD44 alone in terms of positive correlation with tumor grade (supporting information Fig. 1).
Although we were able to show that patient tumor cells isolated based on increased aldefluor activity were tumorigenic, we were unable to show that ALDH1A1 expression, thought to be responsible for the aldefluor activity of CSCs, was correlative with the aggressive nature of patient tumors. These observations suggest the use of ALDH1A1 as a readout of aldefluor positivity, and hence, tumorigenicity may not be entirely accurate.
Aldefluor Activity of Breast CSCs Correlates Better with the Expression of ALDH Isoforms Other Than ALDH1A1
To determine if aldefluor positivity could be due to other ALDH isoforms, we carried out total genome microarray expression analysis on four patient tumor CSC and non-CSC populations isolated based on aldefluor+ and aldefluor− expression. These samples were either directly from biopsies of the primary and/or metastatic tumor or postexpansion of the primary tumor sample in the mammary fat pad of NOD/SCID mice (performed to obtain sufficient cell numbers). Figure 2A shows that all four samples contained aldefluor+ cells (as identified by their sensitivity to the competitive ALDH inhibitor, DEAB). Interestingly, microarray analysis revealed that only one of the four tumor tissue samples had significantly increased ALDH1A1 expression in the aldefluor+ cells (Fig. 2B). Of the 18 other ALDH isoforms expressed in humans, we were able to calculate the fold increase from only one of four and two of four patient sample for ALDH isoforms 1A3 and 1A2, respectively (the signal intensity for the ALDH+, ALDH−, or both ALDH+ and ALDH− samples was not positive or significant above the background to allow for statistically relevant fold difference calculation). Regardless of the missing data, in this expanded analysis, we detected an increased expression of a few of the ALDH isoforms. In particular, ALDH isoforms 1A3, 2, 4A1, 5A1, 6A1, 7A1, and 18A1 (Fig. 2B) were increased significantly in at least one patient sample. The raw data from which the fold increases were calculated are listed in supporting information Table 4.
We next performed immunofluorescence protein expression analysis on the microarray-implicated ALDH isoforms. To obtain sufficient cell numbers for these experiments, tumor biopsy samples from three different patients were first amplified by passaging in the mammary fat pads of mice. The resulting tumors were sorted for aldefluor+ and aldefluor− tumor cells (Fig. 3A). On a side note, we observed an increase in the percentage of aldefluor+ cells in the progressive passages of the patient tumor 3 (compare passage 8 of the tumor in Fig. 2A and passage 11 in Fig. 3A). There appears to be an in vivo selection for aldefluor+ cells in this particular tumor.
The sorted cells were then fixed on slides and stained for expression of the ALDH isoforms implicated from the microarray analysis (ALDH1A1, 1A3, 2, 4A1, 5A1, 6A1, 7A1, and 18A1). We verified the working concentration of the antibodies and noted their subcellular specificity (cyoplasmic vs. mitochondrial staining) by taking high-magnification images of SKBR3 breast cancer cells stained with the antibodies (supporting information Fig. 2). Of the sorted tumor cells, ALDH1A1 expression was only detected significantly in the aldefluor+ population of patient 3 (Fig. 3B, representative images; supporting information Table 5, percentage positive stained cells). In contrast, ALDH1A3 (and to a lesser degree, isoforms 4A1, 6A1, and 7A1) was selectively expressed in the aldefluor+ tumor populations (Fig. 3B, representative images; supporting information Table 5, percentage positive stained cells). We also noted that we detected expression of ALDH1A1, 1A3, 4A1, and 6A1 in the aldefluor+ cytospins of patient 3, despite not detecting upregulated mRNA in the ALDH+ cells of patient 3 for these isoforms in the microarray data (Fig. 2B), suggesting that ALDH mRNA levels does not necessarily reflect protein levels. Regardless, this result, in combination with the microarray gene expression data, suggests that other ALDH isoforms (in particular ALDH1A3) could potentially be responsible, at least partially, for the aldefluor activity of breast CSCs previously attributed to only ALDH1A1.
Aldefluor Activity in Breast Cancer Cell Lines Is Primarily Due To ALDH1A3 Activity
Next, we compared the ALDH expression profile and aldefluor activity of seven breast cancer cell lines. Breast cancer cell lines show a variety of aldefluor activity, which correlates with their molecular subtype . In some cell lines, it is possible to isolate both aldefluor+ and aldefluor− cells, where the aldefluor+ population are more tumorigenic . Other cell lines are 100% aldefluor+ (e.g., SKBR3 cells) and some have 0%–1% aldefluor+ cells (T47D, MCF7, and MDA-MB-231 cell lines) . Similarly, we found SKBR3 cells to have high aldefluor activity and MCF7, T47D, MDA-MB-231 cell lines to have very low aldefluor activity (Fig. 4A). In addition, we show the previously uncharacterized MDA-MB-468 breast cancer cell line to have the most aldefluor activity and MDA-MB-435 to have moderate levels of aldefluor activity.
To determine if expression level of ALDH1A1 or any other ALDH isoform correlated with the aldefluor activity of the breast cancer cell lines, we isolated the mRNA from each cell line and performed quantitative polymerase chain reaction (QPCR) with isoform-specific primers. Plotting the mean fluorescent units generated from the aldefluor assay of each cell line versus the amount of each ALDH isoform transcript (normalized to GAPDH), we found that of the 19 isoforms, only levels of the ALDH1A3, and to a lesser degree ALDH2, correlated well with the aldefluor activity of the cell lines (Fig. 4B shows the result for the isoforms implicated from the microarray data, supporting information Fig. 3 shows the remaining isoforms). Of note, ALDH1A1 mRNA levels correlate poorly with the aldefluor activity reported in each cell line, although we did detect a high level of ALDH1A1 mRNA in the BT-20 cell line, potentially causing the low ALDH activity of this particular cell line. However, as whole, this result further supports our contention that aldefluor activity in breast CSCs is not due to the activity of ALDH1A1 alone, but likely one or more of the other isoforms (in particular ALDH1A3).
To identify the ALDH isoform(s) that is/are responsible for the aldefluor activity of breast cancer cells, we performed shRNA knockdown of each microarray-implicated ALDH isoform on three aldefluor-positive breast cancer cell lines (MDA-MB-468, SKBR3, and MDA-MB-435). Reduction in mRNA levels of a specific ALDH isoform was quantified by QPCR (Fig. 5C) and then correlated with reduction in ALDH activity as measured by the aldefluor assay (Fig. 5A, 5C). Western blotting confirmed the knockdown efficiency of ALDH1A1, 1A3, and 2 by the shRNA sequences and the specificity of the ALDH antibodies (Fig. 5B and supporting information Fig. 4). The results in Figure 5 show that reduction of ALDH1A1 transcript to less than 20% by three different ALDH1A1-specific shRNA sequences did not significantly reduce aldefluor activity in the three cell lines. In contrast, reduction of ALDH1A3 transcript levels to 5%–60% by three different ALDH1A3-specific shRNA sequences significantly reduced aldefluor activity in all three cell lines. Of the other isoform knockdowns, only ALDH2 knockdown reduced aldefluor activity in one of the three cell lines (SKBR3 cells).
When we added 15 μM DEAB, (which only partially inhibited the aldefluor activity of MDA-MB-468, SKBR3, and MDA-MB-435 cells, Fig. 4A), the fluorescence of the ALDH1A3 and ALDH2 (SKBR3 only) knockdowns was further inhibited compared with the shRNA scramble control cells (Fig. 5A). This effect is due to the reduced level of 1A3 (and ALDH2 for SKBR3 cells) in the knockdowns, and as such, 15 μM of DEAB is more effective at inhibiting ALDH activity (Fig. 5A).
Overall, this result shows that ALDH1A3 is of primary importance for aldefluor activity in the breast cancer cells, agreeing with the correlative result using breast CSCs isolated from patient tumor samples (Fig. 3). Moreover, it shows that aldefluor activity in breast cancer can be due to more than one ALDH isoforms, because in the SKBR3 cell line, ALDH2 also contributes to aldefluor positivity.
ALDH1A3 Expression Predicts Metastasis in Breast Cancer
An important corollary of CSC's presumed role in cancer is that they are more abundant in metastatic cancers. To our knowledge, this has yet to be shown for breast cancer in general. In fact, the prevalence of CSCs in breast cancer as defined using currently known markers often does not correlate with metastatic disease. We postulate that if aldefluor activity is also an indicator of a cancer's metastatic potential, then ALDH1A3 (and potentially other implicated isoforms) should be prevalent not only in higher grade/stage breast cancers but also in metastatic cancers.
To test this hypothesis, we stained the same panel of 47 archived breast cancer patient samples (Fig. 1C, 1D and supporting information Table 3) with specific ALDH isoform antibodies for ALDH1A3 (Fig. 6), ALDH2 (Fig. 7), ALDH4 (supporting information Fig. 5), or ALDH6A1 (supporting information Fig. 6). We found that ALDH1A3 expression in the breast tumor samples was the best predictor of aggressive disease, as its expression correlated significantly with higher grade tumors, proximal metastasis, and higher cancer stage (Fig. 6B). Although we did not detect a significant correlation with tumor size, we noted a nonsignificant trend of larger tumors (≥2 cm) having greater percentages of ALDH1A3+ tumor cells (Fig. 6B, lower left panel). Staining the samples for ALDH1A3+CD44+ cells also revealed a significant correlation with tumor grade, metastatic disease, and cancer stage (Fig. 6). However, these correlations were not better than staining the tumor samples for ALDH1A3+ cells alone.
For the first time to our knowledge, we also show that ALDH2 could be a potential marker for breast cancer, where higher tumor grade correlates significantly with increased number of ALDH2-positive tumor cells (Fig. 7). However, we were unable to detect a trend neither with increased ALDH2 positivity and metastatic disease nor with higher cancer stage or tumor size, thereby limiting its future applicability. Interestingly, ALDH4A1 expression did not correlate with the clinical assessments of breast cancer (supporting information Fig. 5), but ALDH6A1 expression did correlate with metastatic disease (supporting information Fig. 6). This could be due to a hitherto unknown function of ALDH6A1 in breast cancer or merely coincidental expression of the isoform detected in aldefluor+ breast CSCs (Figs. 2 and 3).
As the identification of ALDH activity as the hallmark of highly tumorigenic, therapeutic-resistant CSC population of breast tumors, it has been postulated that ALDH/CSC prevalence could be used to predict patient outcome. As a whole, however, the previously published immunohistological evidence for ALDH1A1 being responsible for ALDH activity (and hence serving as a functional CSC marker) does not lend strong support to the hypothesized potential role of CSCs in mitigating metastasis [2, 11–14, 16–18]. We also found that ALDH1A1 prevalence did not predict cancer stage or early metastasis (Fig. 1C, 1D). This disconnect has both confounded researchers and led to questions regarding the reliability of ALDH as a CSC marker .
Our unexpected result that ALDH1A1 is not uniformly overexpressed in the aldefluor+ tumor populations (Fig. 2) led us to hypothesize that other ALDH isoform(s) could be contributing to the increased ALDH activity of breast CSCs. Interestingly, other groups working on normal stem cell model systems (characterization of mouse hematopoietic and nervous stem cells  and mouse pancreatic progenitor cells ) have previously presented data that ALDH isozymes other than ALDH1A1 isoform may be responsible for the aldefluor activity of the stem/progenitor cells of certain normal tissues. Our data, in combination with this previous work, suggests that the assumed role of ALDH1A1 in causing the ALDH activity of both cancer and normal stem cells in general should be re-evaluated.
The 19 human ALDH isoforms share some amino acid sequence similarity and certain functions. However, they differ in tissue distribution, substrate specificity, and cytoplasmic localization. Interestingly, the ALDH1A3 isoform, revealed in the present study to be the primary cause of ALDH activity in breast cancer, has been previously reported to be localized to normal breast tissue . Rexer et al.  isolated the enzyme responsible for retinoic acid synthesis in normal breast tissue and called the newly discovered ALDH isoform ALDH6 (later more commonly known as ALDH1A3). Earlier, the same group found a lack of retinoic acid synthesis activity in five of six breast cancer cell lines tested (MCF7, T47D, MDA-MB-231, ED, and EK cells) , similar to our finding of corresponding low levels of aldefluor activity in MCF7, T47D, and MDA-MB-231 cells (Fig. 4). Intriguingly, the sixth cell line tested by Mira et al.,  MDA-MB-468, was described as an anomaly, having high levels of retinoic acid synthesis, much like our result of high aldefluor activity found in this cell line (Fig. 4). The lack of expression of ALDH1A3 in MCF7 cells, compared with normal breast tissue, led Rexer et al.  to postulate that deficiency of this enzyme has a potential role in breast cancer. High levels of ALDH1A3 in MDA-MB-468 cells was assumed to be the exception, whereas cell lines like MCF7 and T47D cells were presumed to represent the breast cancer norm [25, 26]. In contrast, the sum of our data led us to a different conclusion that ALDH1A3 is a marker for a subpopulation of highly tumorigenic breast cancer cells, that is, CSCs (Fig. 3). It is interesting to note in this regard that Kock et al.  described ALDH1A3 as one of five potential candidate genes linked to mammary tumorigenesis in a spontaneous mouse breast cancer model, heterozygous for p53.
It is perhaps not surprising that one of the ALDH isoforms most similar to ALDH1A1 would be revealed to be the predominant isoform contributing to aldefluor activity in breast cancer. ALDH1A1 and 1A3 share 70% amino acid similarity and, along with ALDH1A2 and ALDH8A1, oxidize retinaldehydes. What was unexpected was the finding that ALDH2, which is not known to oxidize retinaldehyde, also contributed to the aldefluor activity of one of the three aldefluor-positive breast cancer cell lines.
With our discovery that ALDH1A3 is the primary cause of aldefluor activity in breast cancer, and its specific expression is a marker for breast CSCs, we hypothesized that ALDH1A3 upregulation could be associated with more aggressive metastatic cancers. This was indeed found to be the case, revealing for the first time that the expression of a CSC marker is associated with metastatic breast cancer, not specified to one breast cancer subtype (Fig. 6).
It has been suggested that combining the known breast CSC markers would lead to the isolation of the most tumorigenic cancer cells. Indeed, Ginestier et al.  showed that aldefluor+CD24-CD44+ cells were the most tumorigenic tumor cells, with as few as 20 cells being sufficient to induce tumors in immunocompromised mice. This would suggest that combining the markers in immunohistological studies with fixed patient tumors could reveal the strongest correlation between increased prevalence of CSCs and worse prognosis. Our study reveals that the aldefluor+ component of the combination can be represented by ALDH1A3 as a strong correlation exists between increased prevalence of ALDH1A3+/CD44+ cells and higher tumor grade, cancer stage, and metastatic disease (Fig. 6). However, the ALDH1A3+/CD44+ combination is not more significant than ALDH1A3 alone in our study. It remains to be seen whether data from a larger panel of archived tumor samples would favor the combined over the single marker approach.
Our results also have implications on the development of chemotherapeutic resistance in recurrent breast cancer. CSCs originating from various cancers are comparatively resistant to commonly used chemotherapeutics [28–33]. There are several suggested mechanisms for the apparent resistance of CSCs to current anticancer therapies. First, Bao et al.  reported that resistance of glioblastoma CSCs to irradiation is due to increased activation of the DNA damage checkpoint response. Second, the observation that CSCs can be isolated by Hoechst stain exclusion is compatible with enhanced expression in these cells of transporters that efflux the stain . These transporters also efflux chemotherapeutic drugs, a common cause of chemotherapeutic resistance . Finally, with the isolation of CSCs based on increased ALDH activity, it is possible that chemotherapeutic resistance is due to an ALDH-specific activity that metabolizes chemotherapeutics such as cyclophosphamide . In support of this, CSC enrichment was observed in colorectal cancer xenograft tumors after cyclophosphamide treatment in an ALDH-dependant manner . In view of our present findings, it will be of interest to investigate the role of ALDH1A3 in resistance to chemotherapeutics like cyclophosphamide that is commonly used in breast cancer treatment.
Additionally, ALDH activity mediated primarily by ALDH1A3, may have a functional role in the progression of breast cancer, beyond its use as a CSC and prognostic marker. Recently, Ginestier et al.  showed that tumor sphere formation and expression of genes involved in self-renewal and differentiation could be altered by the addition of chemical retinoic acid signaling inducers or inhibitors in breast cancer cell lines.
Finally, as it was found that CSCs from solid tumors can be isolated based on aldefluor activity , others have also isolated CSCs using the aldefluor assay from cancers of the lung, liver, bone, colon, pancreatic, ovarian, head and neck, and prostate [38–46]. The aldefluor activity specific for the CSCs of these cancers has been attributed to ALDH1A1 and as such prognostic studies have also been targeted on this isoform. In light of our results, it would be worthwhile to assess the potential contribution of ALDH1A3 and the other ALDH isoforms to the aldefluor activity of the CSCs of these cancers as it may have important prognostic revelations in these cancers as well.
This work was supported by operating grants from the Canadian Breast Cancer Foundation, Atlantic Chapter and Cancer Care Nova Scotia to P.W.K.L., P.M., and C.A.G. We thank Pat Colp, Alexander Edgar, Dale Mollenkoph, and Karen Inglis for their technical assistance on the project.
Disclosure of Potential Conflicts of Interest
The authors indicate no potential conflicts of interest.