A-C. ALDH1 staining of normal breast epithelium. ALDH1-positive cells (red cytoplasmic staining) were in a luminal location, bridging across the lumen, probably at branching points of side-ducts (arrows). D. ALDH1 staining in mammospheres. Only 1-5 cells/mammosphere showed positive staining for ALDH1, (approximately 5% of the total population). E-F. Immunofluorescence of normal breast epithelium. E. Double staining with CK18 (red) and ALDH1 (green). Composite image (merge) showed absence of overlap between CK18 positive cells (mature luminal cells) and ALDH1-positive cells (arrow). F. Double staining with SMA (green) and ALDH1 (red). Composite image (merge) showed absence of overlap between SMA-positive cells (mature myoepithelial cells) and ALDH1-positive cells (arrow). G. Immunofluorescence of mammosphere sections. Double staining with CK5/6 (green) and ALDH1 (red). Composite image (merge) showed that only few ALDH1-positive cells displayed an exclusive red signal (arrow) whereas all the CK5/6-positive cells (asterisk) displayed a hybrid signal (yellow) corresponding to cells positive for ALDH1 and CK5/6. H. Double staining with CK14 (green) and ALDH1 (red). Composite image (merge) showed that most of the ALDH1-postive cells displayed an exclusive red signal (asterisk) whereas all the CK14 positive cells (arrow) displayed a hybrid signal (yellow) corresponding to cells positive for ALDH1 and CK14. All nuclei were counterstained with DAPI.

A. Table showing the number of outgrowths generated in NOD/scid mouse fat pads by ALDEFLUOR-positive, ALDEFLUOR-negative, and Unseparated cells. B-J. Hematoxylin and eosin staining of ducts generated by ALDEEFLUOR-positive cells (B, E, H), ALDEFLUOR-negative cells (C, F, I), Unseparated cells (D, G, J). The number of cells injected is indicated on the left side (25,000 cells in B –D, 5,000 cells in E-G, 500 cells in H-J). We observed formation of ducts in the fat pads injected with 25,000 and 5,000 ALDEFLUOR-positive (B,E) or Unseparated cells (D,G). Only residual Matrigel and mouse tissue were observed in all the other fat pads (C, F, H-J). K-L. Evaluation of the number of ducts generated by each population (ALDEFLUOR-positive, ALDEFLUOR-negative, Unseparated). Only the ALDEFLUOR-positive and Unseparated cells produced outgrowths. The ALDEFLUOR-positive cells produced significantly more ducts than the Unseparated cells (p<0.05). (K, L)

A-D. Evidence of the human origin of the epithelial ducts. Positive staining with a specific human antibody (anti-ESA) (A, B) which does not cross-react with mouse tissue (C, D) confirms the human origin of the ductal structure (red staining). E-I. Cell composition of ducts generated by ALDEFLUOR-positive cells. The ducts had a luminal layer (stained with anti-CK18, green signal) (and a myoepithelial layer (stained with SMA, red signal). Double staining with luminal-like cytokeratin (CK18; green signal) and basal-like cytokeratins (CK14, CK17, CK5/6, red signal) demonstrated a partial overlap between the luminal cells and the basal cells (yellow signal in the composite image, merge) suggesting a lineage evolution during duct formation. Double staining with CK14 (green signal) and SMA (red signal) showed cells positive for both markers (yellow signal) on the composite image (merge) (I). All nuclei were counterstained with DAPI.

Cells were immunostained with a CD24–PE antibody, a CD44–APC antibody and antibodies for lineage markers labeled PE-Cy5, and subsequently stained with ALDEFLUOR. Cells were first gated based on viability and lin^- markers, which represented 12.3±1.1% of the total population. Cells of mouse origin were also eliminated from the analysis. The four cell subpopulation defined by the ALDEFLUOR and CD44+/CD24-/lin- phenotypes were separated by FACS. A-B. The percentages shown in the diagram shows the representation of the cell sub-populations in the total tumor cell population and the overlap between the ALDEFLUOR phenotype and the CD24^-/CD44⁺/lin^- phenotype. C-F Tumorigenicity of the cell populations defined by the ALDEFLUOR and CD44-/CD24+/lin- phenotypes was tested using the xenotransplantation model described. Experiments were performed in triplicate. The unseparated cells generated tumors when implanted in numbers higher than 500 cells (C). The ALDEFLUOR-negative CD44+/CD24-/lin- cells were not tumorigenic, even at 50,000 cells/fat pad (D). The ALDEFLUOR-positive/CD44+/CD24-/lin- cells generated tumors from as few as 20 cells (E). The ALDEFLUOR-positive/lin-/NonCD44+/CD24- cells generated tumors when implanted in numbers higher than 1500 cells (F).

A-D. Example of ALDH1 staining in breast cancer. Only two of the 577 tumors analyzed were fully positive for ALDH1 (A). Representative examples of breast tumor cores positive for ALDH1 with 5-10% ALDH1-positive cells detected (B-C). Example of a tumor core with no detectable ALDH1 staining (D). E-F. Kaplan-Meier plot of patient overall survival: Survival differed significantly according to ALDH1 expression. Patients with tumors positive for ALDH1 staining (green curve) had a poor prognosis compared to patients with tumors negative for ALDH1 staining (blue curve). Similar results were observed in the U.M. set composed of 136 patients (p=0.0459) (E) and I.P.C. set composed of 341 patients (p=0.000675) (F). G. Cox multivariate analysis of overall survival for patients from I.P.C. set. When compared with known prognostic factors, ALDH1 status was an independent factor of prognosis, as was Ki-67 status, tumor size, SBR grade.

A-B. Representative flow cytometry analysis of ALDH activity in cells derived from a human breast tumor, orthotopically xenotransplanted in NOD/scid mice. The ALDEFLUOR assay was performed as described above. In addition mouse of cell origin were eliminated from the analysis (see Supplementary methods 1 and Supplementary Figure 5). (A, B) All the ALDEFLUOR analyses on human breast tumor cells were first gated on PI negative cells (viable cells) which represented 73.6±1.8% (Mean ± SDEV, n=43) of the total population. C-G Only the ALDEFLUOR-positive population was tumorigenic. C. The ALDEFLUOR-positive population was capable of regenerating the phenotypic heterogeneity of the initial tumor after a passage in NOD/scid mice. D. Tumor growth curves were plotted for the numbers of cells injected (50,000 cells; 5,000 cells; 500 cells) and for each population (ALDEFLUOR-positive, ALDEFLUOR-negative, unseparated). Tumor growth kinetics correlated with the latency and size of tumor formation and the number of ALDEFLUOR-positive cells. E. Representative tumor grown in NOD/scid mouse at the ALDEFLUOR-positive cells’ injection site (5,000 cells injected). No tumor was detected at the ALDEFLUOR-negative cells’ injection site (5,000 cells injected). F-G. H & E staining of ALDEFLUOR-positive cells’ injection site, revealing presence of tumor cells (F). The ALDEFLUOR-negative cells’ injection site contained only residual Matrigel, apoptotic cells and mouse tissue (G). All the data presented in this figure were generated by analysis of the MC1 tumor. Similar results were obtained for three other tumors, generated from different patients (UM1, UM2, and UM3) tested (Supplementary Figure 4).

A-B. Representative FACS analysis of normal breast epithelial cells using the ALDEFLUOR assay. Cells incubated with ALDEFLUOR substrate (BAAA) and the specific inhibitor of ALDH, DEAB, were used to establish the baseline fluorescence of these cells (R1) and to define the ALDEFLUOR-positive region (R2) (A). Incubation of cells with ALDEFLUOR substrate in the absence of DEAB induces a shift in BAAA fluorescence defining the ALDEFLUOR-positive population (B). In all experiments cells were first gated on PI negative cells (viable cells) which represented 93.4 ± 2.4% (Mean ± SDEV, n= 31) of the total population. C-E. ALDEFLUOR-positive cells sorted from fresh reduction mammoplasties were enriched in sphere initiating cells(C) with 451 ± 42 mammospheres (Mean ± SDEV, n= 6, derived from 3 different patients) generated by 10,000 cells plated, versus 50 ± 8 mammospheres (Mean ± SDEV, n= 6) generated by 10,000 unseparated cells (E). ALDEFLUOR-negative cells failed to grow in suspension (D-E). ALDEFLUOR-positive cells and unseparated cells were capable of self-renewal in vitro, as shown by similar mammosphere-initiating capacity in three passages (E). F-J. Evaluation of the differentiation potential of ALDEFLUOR-positive and ALDEFLUOR-negative cells. Sorted cells were grown in differentiating conditions for 12 days and stained by IHC with lineage-specific markers (ESA, CD10). The ALDEFLUOR-positive population generated 237±15 mixed colonies/1000 cells plated (67.2 ± 3.5% bi-lineage colonies) (ESA+ cells stained in brown and CD10+ stained in purple) (F), 11± 1 myoepithelial colonies/1000 cells plated (2.9 ± 0.5%) (CD10+) (G), and 108±25 luminal colonies/1000 cells plated (30.6 ± 5.4%) (ESA+) (H). The ALDEFLUOR-negative population produced 72±10 luminal colonies/1000 cells plated (90.8 ± 3.1%) (ESA+) (H), and only 7±2 mixed colonies/1000 cells plated (9.1±1.3%) (I). Data represent Means ± SDEV, n= 6, derived from 3 different patients. J. ALDEFLUOR-positive and ALDEFLUOR-negative cells grown in differentiating conditions were collected for flow cytometry analysis of lineage markers (ESA, CD10). ALDEFLUOR-positive cells generated uncommitted progeny (15.3±3.2%, CD10-/ESA-; 21.2±1.5%, CD10+/ESA+), luminal cells (63.2 ± 4.1%, CD10-/ESA+) and myoepithelial cells (2.1±0.3%, CD10+/ESA-), whereas ALDEFLUOR-negative cells generated predominantly luminal cells (93.5±3.4%, CD10-/ESA+). Data represent Means ± SDEV, n=3.