Selection of Tumorigenic Melanoma Cells Using ALDH

The frequency of human melanoma initiating cells in NOD/SCID mice

To determine the range in MIC frequency in melanomas from different patients, cell suspensions were xenografted subcutaneously into NOD/SCID mice at various doses and tumor growth was followed over time (Figure 1a). Only one patient sample failed to grow after transplantation. Xenografted tumors recapitulated the original patient tumor with similar histopathologic features (Figure 1b). The frequency of MICs in freshly obtained specimens from individual patients showed a 100-fold range, from 1 in 18,000 to 1 in 1,851,000 cells (Table 1). Thus, in all patients only a fraction of tumor cells initiated further tumor growth in NOD/SCID mice, and the size of the fraction varied greatly.

The time to first palpability was different in xenografts from different patients. In order to use data from all cell doses, a Cox regression model using cell doses as strata was used to associate time to palpability with MIC frequency. The hazards ratio for shortened time to palpability was 1.24 per 1 MIC increase in 150,000 cells (95% confidence interval (CI) 1.08 to 1.142, P=0.017). Thus, there was a significant association between increased MIC frequency and decreased time to first palpability (Figure 1c).

For the 13 patients examined (Table 1, Supplementary Table S1 online), there was an inverse correlation between MIC frequency and metastasis-free survival (time between diagnosis of the primary melanoma and diagnosis of metastatic melanoma). Spearman’s correlation coefficient was −0.6035 (P=0.038), such that the higher the MIC frequency, the shorter the time to metastasis (Supplementary Figure S1 online).

In general, the melanoma samples effaced the lymph node and were sheets of melanoma cells microscopically. Cell counts under the microscope excluded lymphocytes (by size and morphology), which constituted 5–15% of cells. To examine whether it was likely that proportions of contaminating cells contributed to the results obtained above by affecting xenograft growth, we studied the number of CD45⁺ and CD31⁺ human cells in primary lesions and xenografts. For PM 41 (MIC frequency 1 in 18,000), PM 28 (1 in 20,000), PM 37 (1 in 161,000) PM 22 (1 in 710,000), and PM 47 (1 in >2,000,000), there was no evidence for a correlation between the number of contaminating cells (by immunoperoxidase for human-specific CD45 and CD31 and MIC frequency (Supplementary Figure S2 online). In the corresponding xenografts, neither CD45⁺ nor CD31⁺ human cells were found (Supplementary Figure S3 online). Thus, these data do not provide evidence that contaminating cells contributed significantly to the 100-fold difference in MIC frequency.

The population of human melanoma cells with high ALDH activity is enriched in MICs in NOD/SCID mice

We next determined whether cells marked by high ALDH activity (ALDH^hiSSC^lo) were tumorigenic cells. Freshly obtained melanoma cells were sorted for ALDH activity using Aldefluor (Stemco Biomedical, Durham, NC) (Figure 2a). Diethylaminobenzaldehyde, an ALDH inhibitor, was the negative control. The selected ALDH^hiSSC^lo population was 2% of live cells (Figure 2a). Dead cells were excluded using 7-aminoactinomycin D. Lymphocytes formed a smaller, less complex population than the melanoma cells and could be excluded on the forward scatter/side scatter (FSC/SSC) plot. Remaining hematopoietic cells were excluded using CD45 (Figure 2a), such that the ALDH⁺ population was found to be over 99.5% CD45⁻CD31⁻, and the ALDH⁻ population was over 95% CD45⁻CD31⁻ (not shown). ALDH^hiSSC^lo cells from a melanoma cell line (WM-266-4) were enriched for MICs 5.7-fold over UNF cells so that 1 in 65 cells (95% CI 1 in 43 to 1 in 97 cells) versus 1 in 370 UNF cells (95% CI 1 in 190 to 1 in 710) was a MIC (Table 2a). ALDH^hiSSC^lo cells from freshly collected patient melanomas (n=3) or human melanomas grown for one passage in the NOD/SCID mouse (n=6) were also enriched in MICs 16.8-fold, such that 1 in 21,000 cells (95% CI 1 in 14,000 to 1 in 30,000) versus 1 in 353,000 UNF cells (95% CI 1 in 246,000 to 1 in 507,000) was a MIC (Table 2b). Xenograft melanomas that developed from ALDH^hiSSC^lo cells contained both ALDH^hi and ALDH⁻ cells by FACS (Figure 2b). Furthermore, 5,000 ALDH^hiSSC^lo cells secondarily transplanted into NOD/SCID mice could once again produce a xenograft melanoma. These studies provide evidence that the ALDH^hiSSC^lo population contains the tumorigenic cells, can self-renew, and produce a range of progeny in the NOD/SCID mouse.

The population of human melanoma cells that expresses CD133 is enriched in MICs in NOD/SCID mice

The CD133-positive population was examined for MIC enrichment. By flow cytometry, CD133⁺ cells constituted 0.01–0.2% of cells (Figure 2c) and 2.9–29% of CD133⁺ were ALDH^hi. AutoMACS was used to isolate CD133⁺ melanoma cells for limiting dilution analysis from three fresh melanoma samples. CD133⁺ cells were enriched for MICs 2.9-fold over UNF cells, such that 1 in 43,000 cells (95% CI 1 in 20,000 to 1 in 90,000) was a MIC (Table 2c). CD133⁻ cells in large doses (800,000) contained MICs in 12 out of 12 samples, suggesting that there may be a distinct population of MICs that is CD133⁻. This is reminiscent of findings in glioblastoma (Beier et al., 2007). Given the lesser enrichment with CD133 versus ALDH, we did not pursue this marker further.

Serial transplantation and MIC frequency in NOD/SCID mice

To determine whether serial transplantation altered MIC frequency, we transplanted tumors with a range of MIC frequencies through six serial transplantations. After 2–3 transplantations, tumors characterized by low MIC frequency achieved high MIC frequency and had shorter time to first palpability (Figure 2d). Tumors characterized by high MIC frequency maintained high MIC frequency through serial transplantations, and exhibited decreased time to palpability after 2–3 serial transplantations (Figure 2d).

The population of human melanoma cells that expresses ALDH is enriched for MICs in NSG mice

We then examined whether ALDH-expressing cells were the tumorigenic cells in NSG mice. Although previous hematopoietic studies used ALDH^hiSSC^lo populations, more recent studies in solid tumors used ALDH⁺ cells. We examined both 2% ALDH^hiSSC^lo cells and ALDH⁺ cells (ALDH⁺ cells comprised 0.46–5.8% of melanoma cells) and no difference in either MIC frequency or in time to first palpability in the NSG mouse was detected (Supplementary Tables S2 and S3 online). Accordingly, in subsequent studies, we used the entire ALDH⁺ cell population. ALDH⁺ human melanoma cells were enriched in MICs such that 1 in 4 (95% CI 3–5) ALDH⁺ cells versus 1 in 418 (95% CI 302–579) UNF cells (P=0.0001), or 1 in 601 (95% CI 463–779) ALDH⁻ cells (P=0.0001) was capable of tumorigenesis (Figure 3a). Median time to first palpability was significantly shorter for tumors derived from ALDH⁺ cells versus tumors from equal numbers of ALDH⁻ or UNF cells (Figure 3b and Supplementary Figure S4 online). These studies demonstrate that ALDH⁺ cells were enriched in tumorigenic cells compared with ALDH⁻ or UNF cells and that tumors from ALDH⁺ cells had shorter time to first palpability than tumors from UNF or ALDH⁻ cells.

An increased MIC frequency might be found if all cells were equally tumorigenic, but the proliferation of the cells in one tumor versus another was more rapid. To address this possibility, we studied tumors from Figure 3 and assessed the frequency of ALDH⁺ cells by FACS, MIC frequency, mitotic index, Ki-67 expression, and the weeks to first palpability (an in vivo reflection of doubling time) (Figure 4). In these cases, the tumor with the highest proportion of ALDH⁺ cells in the tumor by FACS had the highest MIC frequency. There was no significant difference in mitotic index between the tumors, or in Ki-67 staining. For UNF cells, the time to first palpability was longest for the tumor with the lowest percentage of ALDH⁺ cells. However, when ALDH⁺ cells only were injected, no statistical difference was detected in the median weeks to first palpability. Thus, at least for the tumors examined, no correlation was found between the mitotic index and MIC frequency.

To determine whether ALDH⁺ melanoma cells were capable of ongoing self-renewal, whereas ALDH⁻ cells were not, serial transplantation in vitro was performed. Xenografts that developed from ALDH⁺ or ALDH^hiSSC^lo melanoma cells contained similar proportions of ALDH⁺ cells to the tumors that developed from UNF cell injections (Figure 3c), whereas tumors that developed from ALDH⁻ cells contained small numbers of ALDH⁺ cells. Xenografts (from Figure 3) were transplanted in vitro to examine self-renewal ability (20,000 cells per well, in 24-well plates). Cells from xenografts that developed from ALDH⁺, ALDH^hiSSC^lo, and UNF cells grew well in culture, whereas the xenografts that developed from ALDH⁻ cells did not grow or grew poorly (Figure 3d and Supplementary Figure S5 online). Thus, xenografts from ALDH⁺ melanoma cells displayed superior self-renewal to xenografts from ALDH⁻ cells.

FACS sorting for ALDH⁻ cell populations with Aldefluor

ALDH⁺ cells were present in the xenografts that developed from ALDH⁻ cells, consistent with either, ALDH⁺ cells present in the ALDH⁻ cells selected by FACS, or ALDH⁻ cells producing or becoming ALDH⁺ cells (see Figure 3c). To determine if ALDH⁺ cells persist in FACS-sorted ALDH⁻ populations, we performed double and triple sorts of FACS-sorted ALDH⁻ cells. When ALDH⁻ cells were double-sorted 0.2–1% of cells that were in the ALDH⁻ gate on the first sort were in the ALDH⁺ gate on the second sort, and after triple sorting, 0–0.4% of previously ALDH⁻ cells were found in the ALDH⁺ gate (n=4). FACS sorting produces highly purified populations of cells, yet is clearly unable to eliminate all ALDH⁺ cells.

Immunostaining for ALDH

We then performed immunoperoxidase studies to detect differences in ALDH expression in fresh human melanomas (Supplementary Figure S6 online). Eleven of 13 primary melanomas and 11 of 14 metastatic lesions had 0–10% ALDH-positive cells, whereas 2 of 13 primaries and 3 of 14 metastastic lesions had 50–100% positive cells. In tumors with small numbers of ALDH⁺ cells (0–10%), some ALDH⁺ cells had melanoma cell morphology (large aberrant nucleus) and a nested location, whereas other cells were infiltrating macrophages, making tumors with small numbers of ALDH-positive cells difficult to assess. In three paired primary/metastasis samples, we found that each of the primaries and each of the metastases had 0–10% positive cells, and thus there was no evidence for expansion of the ALDH⁺ population in metastases over corresponding primary lesions.

Melanoma cell preparation

Metastatic melanomas were obtained intraoperatively from patients at the UCSF Helen Diller Cancer Center, with written, informed patient consent, adherence to the declaration of Helsinki principle guidelines and Committee on Human Research approval. As in most human melanoma studies (Monzani et al., 2007; Quintana et al., 2008; Schatton et al., 2008), metastatic melanomas were used, as obtaining fresh primary melanomas is difficult due to the requirement of the entire lesion for pathological evaluation. Extraneous tissue was trimmed and the tumor was minced and placed in 1000 collagen digestion Uml⁻¹ of Collagenase, 1300Uml⁻¹ of Hyaluronidase, and 200 Kunitz Uml⁻¹ DNase I (Sigma, St Louis, MO) in Hank’s solution (Cambrex, Walkersville, MD) with 10% fetal bovine serum (Invitrogen, Carlsbad, CA). After 90–120 minutes at 37 °C, the cell suspension was filtered (100 μm strainer, Becton Dickinson, Mountain View, CA). Cell counts and viability were determined using a Neubauer hemacytometer (Hausser Scientific, Horsham, PA). Dead cells were excluded using 0.4% Trypan Blue (Sigma), and lymphocytes by size and morphology.

Cell culture

Patient melanoma cells and the cell line WM-266-4 (from a human metastatic melanoma in lymph node) were cultured in Eagle’s minimum essential medium (ATCC, Manassas, VA), 10% fetal bovine serum, and 1% antibiotic/antimycotic (Invitrogen).

Subcutaneous injection of melanoma cells

The NOD/SCID mouse and the NSG mouse (Jackson Laboratory, Bar Harbor, ME) models were utilized. Melanoma cells were injected subcutaneously on the dorsal surface. Cell viability was ~90%. Tumors were followed for up to 24 (NOD/SCID) or 36 (NSG) weeks. After killing, injection sites were excised, and histological examination determined the presence of tumor. Tumors were excised for serial transplantations at 1cm diameter.

FACS

Melanoma cells were incubated with 1 μm BODIPY aminoacetaldehyde (Aldefluor), a fluorescent substrate to ALDH, according to the manufacturer’s protocol. Cells were also incubated with phycoerythrin-conjugated anti-human CD133/1(AC133) antibody (Miltenyi, Bergisch Gladbach, Germany), according to the manufacturer’s protocol. Allophycocyanin-conjugated anti-human CD45 antibody was used to exclude bone marrow-derived cells. Sorting was conducted on a FACSAria (Becton Dickinson) utilizing FACSDiva software and a purity prioritization algorithm. FlowJo software (Treestar, Ashland, OR) was used for data analysis. CD133⁺ melanoma cells were separated by an AutoMACS Separator (Miltenyi Biotec, Auburn, CA). The CD133⁺ population was confirmed by FACS.

Immunostaining

Paraffin-embedded sections were deparaffinized, rehydrated, and antigen enhanced with Reaction Buffer (Ventana Medical Systems, Oro Valley, AZ). ALDH1 antibody (ab52492, Abcam, Cambridge, MA) was used according to the manufacturer’s instructions. Humanspecific mAb anti-CD45 and anti-CD31 antibodies (Cell Marque, Rocklin, CA) and mAb anti Ki-67 antibody (Ventana Medical Systems) were used.

Statistical analysis

Injection sites were scored as positive or negative for tumor, based on gross observation and subsequent histological confirmation. The ratio of positive results/total injections for each dose of cells injected was determined. Using this data, limiting dilution analysis of the frequency of MICs for each cell population was performed using L-Calc (Stemsoft, Vancouver, Canada) statistical software. The Spearman correlation coefficient was used to assess the association between variables. A Stratified Cox regression model was used to assess differences in time to first palpability.