Early passage human xenograft tumors generated from cells never expanded in vitro appear to closely resemble those in patients and thus may serve as an excellent model by which to study cancer. We recently demonstrated that the subpopulation of human colorectal tumor cells with an ESA⁺CD44⁺ phenotype is uniquely responsible for generating heterogeneous adenocarcinomas in a xenograft setting (Figure 1A & B)[30]. Human ESA⁺CD44⁺ cells from xenogeneic colorectal tumors can be further subdivided based on CD166 expression, resulting in enrichment for tumorigenic (TG) cells among the CD166⁺ population. In contrast, all remaining cancer cells of human origin appear unable to initiate tumor growth and are referred to as non-tumorigenic (NTG). Having identified colorectal cancer stem cells (CoCSC) capable of fueling xenogeneic tumor growth, we sought to address a lingering question in the field regarding whether CSC are preferentially resistant to chemotherapeutic drugs.

Cyclophosphamide (CPA) is an alkylating agent whose metabolic byproduct, phosphoramide mustard, crosslinks DNA and induces apoptosis in rapidly dividing cells [22]. CPA is a commonly used chemotherapeutic drug in the treatment of various types of cancer, such as soft tissue sarcomas, breast cancer and non-Hodgkins lymphoma, but is not generally used to treat colorectal cancer due to a prevailing resistance on the part of these tumors. To investigate this resistance to therapy, we explored whether CoCSC are enriched in residual tumors following CPA administration in vivo. To address this subject of interest, tumors were initiated with highly purified CoCSC from multiple xenogeneic tumor lines (UM-C4 & UM-C6) and upon reaching ∼400 mm³, mice were randomized to receive either vehicle or 38 mg/kg CPA, twice weekly. Within 15 days of randomization and administration, tumor growth was noticeably retarded in the CPA-treated animals (Figures 1C and Figure S1A). Following euthanization, tumors were removed for histological and phenotypic analysis. Although the percentage of proliferating tumor cells (i.e. Ki-67⁺) did not appear altered, there was noticeably more cell death (i.e. TUNEL⁺ cells; Figure 1B) and ESA⁺CD44⁺ cells were more frequent in tumors from CPA-treated mice (Figure 1D). The trend towards more frequent ESA⁺CD44⁺ cells in CPA-treated tumors generally held true independent of tumor volume and was more pronounced at higher CPA doses (e.g. 38 mg/kg versus 25 mg/kg; data not shown). Enrichment of ESA⁺CD44⁺CD166⁺ cells was even more striking and particularly evident when CD166 expression was analyzed on residual human ESA⁺CD44⁺ cells (Figures 1E, S1B & C). Similar results were observed with both xenogeneic tumor lines investigated.

To determine whether the phenotypic increase in CoCSC following CPA administration correlated with an authentic rise in TG cell frequency, we serially transplanted 4 sets of mice with bulk UM-C4 tumor cells using a limiting dilution approach and scored them as positive or negative for tumor growth after three months. Mice with palpable tumors at 90 days were kept alive to determine whether tumor growth would continue beyond 200 mm³, as rare palpable masses can occur, but which contain murine stromal elements and no detectable human cells upon analysis at the termination of the study (data not shown). Based on Poisson distribution statistics, TG cells were >2.2-fold more frequent in CPA-treated tumors than in mice administered vehicle (P = 0.0016; Table 1 & Figure 2A). Importantly, the cellular phenotype of these serially transplanted secondary tumors were identical to the parental tumor cells used to initiate the study (Figure 2B), demonstrating that intervening treatment and serial transplantation of cells in limiting dilution did not alter either the tumorigenic potential of CoCSC nor their capacity to generate heterogeneous tumors containing predominantly non-tumorigenic (i.e. CD44⁻) cells.

Since ALDH1A1 gene expression is elevated in TG versus NTG cells, ALDH1 enzymatic activity is disproportionably high in cells with the CoCSC phenotype, and this activity may mediate resistance to CPA, we next asked whether the frequency of ESA⁺CD44⁺ALDH⁺ cells was elevated in UM-C4, UM-C6 and OMP-C8 tumors from mice receiving CPA versus those being administered vehicle. Like the observed phenotypic increase in CD166⁺ cells (Figures 1E, S1B & S1C), the ALDH⁺ subpopulation of ESA⁺CD44⁺ cells was consistently higher in CPA- versus vehicle-treated control mice (67.6±6.3% versus 56.8±6.8%, respectively; n = 5, P<0.012 in paired t-test) (Figure 4A).

A number of gene products have been deemed critical to normal colon development and are widely expressed in colorectal tumors [36]–[39]. Following the isolation of TG versus NTG populations from both vehicle- and CPA-treated UM-C4 tumors with >99% purity, expression of numerous genes with suggested ties to colorectal cancer was determined by Taqman™ qRT-PCR. Those differentially expressed in TG versus NTG cells included not only ALDH1A1, as described above (Figure 3C), but also c-Myb (MYB) and c-Myc (MYC; Figure 4B). In contrast, genes associated with differentiation and a mesenchymal phenotype [40], [41], such as Vimentin (VIM), were more highly expressed in NTG cells.

Consistent with the elevated frequency of residual tumor cells expressing high levels of tumorigenicity-associated markers, such as CD166, relative expression of CoCSC-associated versus NTG-associated transcripts, such as ALDH1A1 and VIM, respectively, was further disparate in CPA- versus vehicle-treated control UM-C4 tumors (Figure 4B). Consistent with the hypothesis that ALDH1 may have a role in mediating resistance to CPA, ALDH1A1 gene expression was further elevated in residual cells with the CoCSC phenotype, but ALDH3A1 and ALDH5A1 were not (Figures 4B & S3). In addition, genes previously identified as highly expressed in the epithelial stem/progenitor cell compartment at the base of colon crypts and in neoplastic tissue (e.g. MYB & MYC) were more highly expressed in residual tumor CoCSC than the NTG population of cells that represent the bulk of the tumor, but which are generated by CoCSC and appear more susceptible to CPA-induced cytotoxicity. Similar expression patterns were also observed in other xenogeneic tumor lines following CPA-treatment, including UM-C6 (Figure S3).

Finally, gene expression differences in CPA- versus vehicle-treated control tumors were further correlated with protein expression by immunofluorescence and IHC. Intracellular levels of ALDH1 and nuclear c-Myc, for example, were more prevalent in CPA-treated tumors (Figure 4C). Interestingly, ALDH1 staining in CPA-treated tumors appeared to localize to the apical surface, where it may be hypothesized to intercept CPA metabolites as they enter the cell. In contrast, ALDH1 protein levels were lower and more diffuse in control tumors. Similarly, c-Myc was noticeably more concentrated in the nucleus of colorectal tumor cells of CPA-treated mice, consistent with the increased frequency of TG cells and elevated MYC expression within this resilient population.

To determine whether CoCSC also exhibit resistance to other important chemotherapeutic agents, we examined the impact of Irinotecan (a.k.a. SN-38) on CoCSC frequency. Experiments were initiated whereupon tumor bearing mice received weekly injections of 15 mg/kg Irinotecan. After a slight delay of approximately 1 week, tumor growth was halted by Irinotecan (Figure 5A). Upon tumor harvest two weeks after the initiation of chemotherapy, the percentage of remaining CoCSC was assessed by flow cytometry. In Irinotecan-treated tumors, the percentage of bulk tumor cells with the CoCSC phenotype was increased 61% versus control mice administered vehicle (Figure 5B). Next, the inherent tumorigenicity of ESA⁺CD44⁺CD166⁺ cells from vehicle- or Irinotecan-treated mice was tested in a serial transplant setting versus those cells that did not phenotypically classify as CoCSC. Similar take rates were observed with CoCSC from control- or Irinotecan-treated tumors in these transplants, and cells phenotypically classified as “Other” did not initiate tumors (data not shown). The increased frequency of CoCSC phenotype cells in residual tumors from mice administered Irinotecan and the inability of cells with other phenotypes to transplant disease, together suggest that CoCSC are also preferentially resistant to Irinotecan.

Seeing that ESA⁺CD44⁺ tumor cells have high ALDH activity, the frequency of ESA⁺CD44⁺ALDH⁺ cells is increased in tumors from mice treated with CPA, and ALDH1A1 gene expression is elevated in CPA-resistant CoCSC, we therefore sought to determine whether resistance to CPA was mediated by ALDH1 enzyme activity. Because the ALDH1 specific inhibitor DEAB is highly unstable in vivo [42], in vitro culture conditions that support TG colorectal tumor cell expansion were established. Notably, the cellular phenotype of cells best able to establish colonies in vitro was ESA⁺CD44⁺CD166⁺ (also that of CoCSC), whereas FACS purified CD44⁻ cells were unable to do so (data now shown). To validate that this culture system supports maintenance of TG cells, we sorted ESA⁺CD44⁺CD166⁺ CoCSC in limiting dilution into plates containing Mitomycin C-treated 3T3 or MEF feeder cells as support, depending on the tumor line used: UM-C4 or UM-C6, respectively. Colony formation frequency of CoCSC under these conditions was roughly similar to the TG cell frequency observed upon transplantation (∼1∶86±33; n = 7 for UM-C4 and 1∶52±5; n = 3 for UM-C6). Colonies were cultured for 14 days without passaging prior to reimplantation into mice. Not only did cells maintain ESA, CD44 and CD166 expression during in vitro culture (Figures 6A & B), but the ability to generate heterogeneous tumors resembling parental tumors was conserved as well (Figure 6C). To validate that tumor-initiating cells in tumors originating from single in vitro colonies both resembled parental tumors and were derived from a single cell, UM-C4 cells were transduced with a lentivirus carrying GFP-Luciferase, tumors were generated, and single colonies obtained from ESA⁺CD44⁺CD166⁺ cells were transplanted into mice. Upon harvest of these tumors derived from a single colony, either cells with the CoCSC phenotype or all other tumor cells with differing phenotypes were isolated by FACS, and tumorigenicity and clonality studies were performed. An input of 2,000 FACS purified cells resulted in a 100% take frequency with CoCSC phenotype cells, whereas human cells with all other phenotypes generated only 1 tumor (of 10 mice transplanted); likely resulting from rare contaminating CoCSC (Figure 6D). Of note, tumor heterogeneity can result from a single cell, as lentiviral transduction studies show that both ESA⁺CD44⁻ (i.e. NTG) and ESA⁺CD44⁺CD166⁺ (i.e. CoCSC) cells have identical lentiviral insertion sites (Figure 6D inset), but only CoCSC are able to propagate tumors. Phenotypic and histological analysis of these serially transplanted tumors, which originated from single in vitro colonies, show that the ability to generate adenocarcinomas similar to parental UM-C4 tumors is maintained during short-term in vitro culture in these conditions (Figures 6E & F).

To test whether CPA resistance is mediated by ALDH1 enzyme activity, UM-C4 or UM-C6 colorectal tumor cell colonies established in vitro over the course of 3–4 days were then exposed to the bioactive CPA-metabolite, 4-HC, for 4 hours without or including various concentrations of the ALDH1-specific inhibitor, DEAB. In the absence of DEAB, the EC₅₀ of 4-HC was ∼60 µg/mL (Figure 7A). Inhibition of ALDH1 activity resulted in increased susceptibility to 4-HC mediated cell death, as the EC₅₀ was shifted to 43 µg/mL and 6 µg/mL in the presence of 30 µM and 100 µM DEAB, respectively. Like DEAB, all-trans retinoic acid (ATRA) has been demonstrated to reduce ALDH enzyme activity; however, it does so by reducing ALDH1 and ALDH3 protein levels [43]. Like DEAB, pre-treatment with 1 µM ATRA sensitized colorectal tumor colonies to 4-HC (n = 3; P<0.015) (Figure 7B). Surprisingly, the combination of ATRA pre-treatment and concurrent DEAB exposure synergized to facilitate increased chemosensitivity (n = 3; P = 0.003). Preferential ALDH1A1 gene expression in CoCSC and sensitization of colorectal tumor cells to 4-HC by inhibiting ALDH1, together strongly suggest that ALDH1 enzyme activity mediates CPA resistance.

The above observation raised the question as to whether high ALDH1 enzyme activity also confers resistance to Irinotecan. We next established colonies of colorectal tumor cells and exposed them to serial dilutions of Irinotecan for one week in the presence or absence of 75 µM DEAB; a concentration known to potently inhibit ALDH1 activity. In contrast to 4-HC, the unaltered cytotoxicity profile of Irinotecan despite the presence of DEAB (Figure 7C) suggests that inherent resistance of CoCSC to Irinotecan occurs by a mechanism other than ALDH1 enzyme activity. Of note, neither DEAB nor an equal volume of its vehicle (1∶200 dilution of 95% Ethanol) significantly reduced proliferation or survival during extended culture (7 days), suggesting that ALDH1 activity is only critical in a chemotherapeutic setting.

The above in vitro studies strongly suggest that the ALDH1A1 gene product, ALDH1, mediates resistance to CPA. To determine the importance of ALDH1 to CoCSC tumorigenicity and resistance to CPA, we transduced UM-C6 tumor cells with shRNA targeting ALDH1A1. UM-C6 tumors were chosen because of their high ALDH1A1 expression and relatively low expression of other ALDH family members (e.g. ALDH3A1 & ALDH5A1; see Figure 4C). Cells successfully transduced with either an ALDH1A1-targeted shRNA or Luciferase-targeted shRNA control vector were isolated by FACS and implanted subcutaneously to generate tumors. Taqman™ qRT-PCR studies of UM-C6 tumor cells verified that ALDH1A1 gene expression was reduced 88% versus Luciferase-targeted shRNA transduced control cells, whereas neither ALDH3A1 nor ALDH5A1 gene expression were altered (Figure 8A). After tumors reached roughly 175 mm³, tumors were randomized within each group and then either treated with vehicle or 38 mg/kg of CPA, twice weekly. By ten days post-randomization, tumors in vehicle-treated mice from both groups had almost doubled in size. Of note, the growth of ALDH1A1-targeted shRNA transduced cells did not differ significantly from controls (not shown). Consistent with observations described above, control Luciferase-targeted shRNA transduced tumors undergoing a regimen of CPA were retarded in their growth versus vehicle-treated controls (Figure 8B; open circles). In agreement with results observed in vitro with the ALDH1 inhibitor, DEAB, shRNA targeting of ALDH1A1 gene expression appeared to sensitize tumors to CPA in vivo, as tumor growth essentially stopped, whereas ALDH1A1-targeted shRNA cells treated with vehicle continued to grow (Figure 8B; black triangles). Compared to vehicle-treated, ALDH1A1-targeted shRNA containing tumors, the difference in tumor volumes between CPA and vehicle-treated groups significantly differed by day 10 (n≥5, P = 0.0061).

Upon termination of the study, tumors were removed and the cellular phenotype was assessed to determine whether CoCSC frequency differed in tumors where ALDH1A1 gene expression was being suppressed by shRNA. Whereas control Luciferase-targeted shRNA tumors were significantly enriched for ESA⁺CD44⁺CD166⁺ cells in CPA- versus vehicle-treated mice (n≥5, *P = 0.0002), both CoCSC and NTG cells from ALDH1A1-targeted shRNA containing tumor cells appeared equally sensitive to CPA, as the frequency of CoCSC was no different (Figure 8C). Of note, shRNA-mediated knockdown of ALDH1A1 in vivo, significantly reduced the frequency of CoCSC in residual tumors following CPA therapy (n≥6, **P = 0.045).

Human colorectal tumor lines used in this study were obtained and passaged in mice as previously described [30]. Briefly, all tumors were initiated by subcutaneous implantation of either bulk tumor cells from frozen stocks or FACS purified cell populations into 6–8 week old NOD/SCID mice (Jackson Laboratories). Mice were anesthetized with Isoflurane or a single IP injection of 75–100 mg/kg Ketamine and 5–10 mg/kg Xylazine. Of the 3 xenograft lines used in this study, two (UM-C4 & UM-C6) originated at the University of Michigan and one (OMP-C8) originated at OncoMed Pharmaceuticals Inc. All experiments were carried out under approved institutional IACUC guidelines and protocols.

Tumor tissue was minced into tiny fragments (∼2 mm³), followed by enzymatic digestion with 300 u/mL Collagenase, 100u/mL Hyaluronidase, 0.5 mg/mL Dispase and 100 u/mL DNAseI (all obtained from Stem Cell Technologies; Vancouver, BC) for 1 hour at 37°C/5% CO₂ with intermittent pipetting to disperse cells. Cells were then filtered sequentially through 70 µm and 40 µm screens, followed by a wash with excess FACS Buffer (1× Hanks Buffered Saline Solution [HBSS], 2% Heat-inactivated Fetal Calf Serum [FCS] and 25 mM HEPES [pH 7.4]). Red blood cells were lysed during a brief exposure to Ammonium Chloride and washed again with excess FACS buffer.

All analyses and cell isolations were performed using freshly dispersed cell suspensions. Antibody staining was performed in FACS buffer for 30 minutes at 4°C at a density of 1×10⁷ cells/mL. Antibodies used in this study include: anti-mouse H-2K^d (SF1-1.1; BD Pharmingen), anti-mouse CD45 (30-F11; BioLegend), anti-human ESA (HEA-125; Miltenyi Biotec), anti-mouse/human CD44 (IM7; eBioscience), anti-human CD49f (GoH3; BD Pharmingen) and anti-human CD166-PE (105902; R&D Systems). The Aldefluor™ reagent was purchased from StemCell Technologies and used per manufacturer instructions. In all experiments, cells staining positively for murine lineage markers (mLin⁺; H-2K^d and murine CD45) were excluded during flow cytometry using Cy5.5PE-labelled antibodies. Dead cells were excluded using the viability dye DAPI and cell doublets and clumps were excluded using doublet discrimination gating. Cellular phenotype and viability of FACS purified cells was confirmed by serial flow cytometric analysis prior to injection for tumorigenicity studies. Purity was typically >99%.

Following gentle centrifugation at 900 rpm×5 min, cells were resuspended in 50 µL of FACS buffer per mouse and mixed 1∶1 with Matrigel (BD Biosciences), followed by subcutaneous injection into the lower abdominal region while mice were under general anesthesia as described above. Only one tumor was initiated per mouse. Health was monitored daily and tumor growth was measured weekly using a digital caliper for up to 4 months. Animals were euthanized when tumors exceeded 1500 mm³ or the 120 day timepoint was reached. Statistical analysis of tumor growth includes only those mice with palpable tumors (Mean±SEM).

Female mice were implanted with mLin⁻ESA⁺CD44⁺ cells (400 per mouse) while under anesthesia as described above. Once palpable, tumors were calipered twice weekly, and their length and width were used to calculate tumor volume based on the following formula: V = (length×[width])²/2. When tumors reached ∼400 mm³, mice were randomized into either the control group, which received vehicle (sterile water), or the therapeutic group, which received CPA administered intraperitoneally at the dose of 38 mg/Kg twice a week. Irinotecan was diluted in PBS and similarly administered once weekly at a dose of 15 mg/kg.

Immunohistochemical studies were performed on formalin-fixed paraffin embedded tissues using monoclonal antibodies raised against Ki-67 (Vector Laboratories, catalog number VP-RM04) and c-Myc (DAKO, Catalog number M3570). Briefly, 4 µm-thick paraffin sections were deparaffinized and hydrated. Antigen retrieval was performed in 0.1M Tris-Cl, pH 9.0, using the Decloaking chamber from Biocare. Sections were then incubated with 3% hydrogen peroxide to block endogenous peroxidase activity and incubated with 1∶50 dilution of primary antibody for 1 hr at room temperature, and then detected using ImmPRESS anti-mouse or anti-rabbit Ig (peroxidase) followed by Vector® NovaRED™ substrate for visualization (Vector Laboratories). Hematoxylin was used for counter staining. TUNEL staining was performed using the In Situ Cell Death Detection™ kit (Roche). For Immunofluorescence staining of ALDH1 (Abcam), frozen sections were fixed with −20°C cold methanol for 10 min, followed by a blocking step. Sections were incubated with ALDH1 at 1 ug/ml for 1 hr at room temperature and visualized with anti-mouse antibody conjugated with Alexa Fluor 488. Anti-hapten IgG was used as negative control and mounted with Prolong Gold DAPI containing antifade (Invitrogen).

Following tissue disaggregation, as described above, cell suspensions were depleted of murine lineage cells using magnetic beads and plated in 96-well Primaria (BD Bioscience) plates at a density of ∼20,000 cells per well in serum free Medium-D (3∶1 low glucose DMEM:F-12 Media, B27 supplement, ITS-X, Pen/Strep [all from Invitrogen] and 0.5 µg/mL hydrocortisone [Stem Cell Technologies]), supplemented with 20 ng/mL bFGF and EGF, 5 u/mL Heparin and 1×10⁶ u/mL LIF. Plates were gently spun at 500 rpm for 5 min at room temperature following plating to promote attachment. When cultured in vitro for more than 7 days, media was changed weekly.

For cytotoxicity studies, cells were cultured for 3–4 days at 37°C/5% CO₂/5% O₂, non-attached cells were removed, and medium was replaced with that containing 4-hydroxycyclophosphamide (4-HC; obtained from Dr. OM Colvin; Duke University Medical Center), diethylaminobenzaldehyde (DEAB; Sigma), Irinotecan or their respective vehicle. In studies using 4-HC, cells were exposed for only 4 hours, after which the medium was removed, cells were washed twice with PBS, and fresh Medium-D was added. In experiments involving ATRA, cells were pre-incubated overnight prior to exposure to DEAB and/or 4-HC. Cell viability was then assessed using CellTitre-Glo (Promega) either 20 hr post-treatment for 4-HC studies, or 1 week after addition of Irinotecan.

To assess clonality using lentiviral insertion site analysis, or introduce shRNA targeting ALDH1A1 or Luciferase, UM-C4 and UM-C6 tumor cells were transduced in vitro over a 3–4 day period. Lentiviral vectors used for these studies contained either an eGFP-Luciferase fusion gene (pLOM65) for clonality studies, or ALDH1A1 (pLOM205) or a control Luciferase (pLOM145) targeted shRNA embedded in the eGFP mRNA; each of which is driven by the CMV promoter. Cells were isolated by FACS based on GFP fluorescence and transplanted into mice. For clonality studies, pLOM65-transduced cells were deposited into 96-well plates in limiting dilution and single colonies were injected into mice 14 days later. Lentiviral insertion site analysis was performed as described elsewhere [49] using the NlaIII restriction endonuclease.