Tumor-initiating cells are rare in many human tumors

The cancer stem cell (CSC) hypothesis proposes that clonogenic growth and self-renewal potential are restricted to a subset of cells within tumors. Central to the hypothesis are tumor-initiating cells (TICs), which are often defined by their ability to form tumors following xenotransplantation into immunodeficient NOD/SCID mice and appear to be relatively rare in most human cancers (Clarke et al., 2006). However, recent data indicate that the frequency of TICs can increase dramatically when NOD/SCID mice lacking the interleukin-2 receptor gamma chain (NSG), which are more immunodeficient than NOD/SCID mice, are used as xenograft recipients or when syngeneic mouse models of cancer are studied (Kelly et al., 2007; Quintana et al., 2008). These reports raise the possibility that the true frequency of TICs has been greatly underestimated in most human tumors primarily due to xenotransplantation barriers and the degree to which immune recognition is impaired in recipient mice. We compared the growth of human pancreatic, non-small cell lung, and head and neck squamous cell carcinomas in NOD/SCID and NSG mice. TIC frequency was up to 10-fold higher in NSG mice for some tumors, but remained relatively low (0.0028% – 0.04%) in all cases. Moreover, tumor formation by pancreatic cancer aldehyde dehydrogenase-positive (ALDH⁺) and CD44⁺CD24⁺ cells, previously identified as TICs (Li et al., 2007; Rasheed et al., 2010), did not differ between NOD/SCID or NSG mice. Our findings demonstrate TICs are rare in human pancreatic adenocarcinoma, lung squamous cell carcinoma, lung adenocarcinoma, and head and neck squamous cell carcinoma, despite using highly permissive xenotransplantation conditions.

Specific modifications to the xenotransplantation protocol can dramatically increase the frequency of human melanoma TICs from approximately 0.0001% to 25% (Quintana et al., 2008; Schatton et al., 2008). These alterations include the use of the more immunodeficient NSG strain, injection of tumor cells in Matrigel, and increasing the length of observation for tumor formation (>20 weeks). We optimized our xenotransplantation model to assess the TIC frequency for four malignances and compared subcutaneous tumor formation in NSG and NOD/SCID mice by performing limiting dilution assays of tumor cells suspended in Matrigel. Animals were then monitored for tumor growth for at least 20 weeks following injection to assess tumor formation.

We investigated the frequency of TICs in pancreatic adenocarcinoma, non-small cell lung adenocarcinoma, squamous cell lung carcinoma, and head and neck squamous cell carcinoma using NSG mice. For each type of cancer, tumors from at least 3 different patients were analyzed. We found that the TIC frequency ranged from 1/2,500 – 1/18,000 in pancreatic adenocarcinoma, 1/16,000 – 1/30,000 in non-small cell lung adenocarcinoma, 1/4,200 – 1/16,000 in squamous cell lung carcinoma, and 1/4,300 – 1/36,000 in head and neck squamous cell carcinoma (Table 1). Although TIC frequencies were highly variable even within a specific diagnosis, they were relatively low (< 0.04%) in all of the tumors we tested.

We also compared the frequency of TICs in NOD/SCID and NSG mice. Similar to findings for melanoma (Quintana et al., 2008), we found a trend towards decreased latency in NSG mice in all tumor types (data not shown). However, TIC frequency did not differ significantly for pancreatic adenocarcinoma or head and neck squamous cell carcinoma cells when measured in either strain of mice (Table 1). By contrast, in all three cases of squamous cell lung carcinoma, TICs were more frequent when measured in NSG mice compared with NOD/SCID mice, ranging from 1/4,600 – 1/16,000 and 1/42,000 – 1/125,000 cells, respectively (Table 1). Similarly, the TIC frequency in 1 out of 3 cases of non-small cell lung adenocarcinoma was higher when assayed in NSG mice (Table 1). In the lung adenocarcinoma in which a difference was detected, no tumors formed at the highest dose of cells (100,000) injected into NOD/SCID mice, whereas the TIC frequency measured in NSG mice was 1/16,000 (Table 1). In the two other cases of non-small cell lung adenocarcinoma, the TIC frequencies in both mouse strains were not significantly different (Table 1). Therefore, depending on the specific cancer or among individual tumors within a single diagnosis, TIC may be detected up to 10-fold more frequently in NSG than NOD/SCID mice, but even then, remain quite rare.

Phenotypically distinct cells enriched in tumorigenic capacity have been identified in an increasing number of human malignancies. However, in melanoma, when more permissive experimental conditions were used, specific phenotypes failed to enrich for tumor-forming capacity (Quintana et al., 2008). In pancreatic cancer, both CD44⁺CD24⁺ and ALDH⁺ cells have been identified as TICs (Li et al., 2007; Rasheed et al., 2010). We isolated CD44⁺CD24⁺ cells (0.2 – 1.6% of total cells; Figure S1) and ALDH⁺ (2.3 – 3.5% of total cells; Figure S1) and compared tumor formation by each of these cell populations in both models. The TIC frequency was 1/300 and 1/330 (P=0.471) for ALDH⁺ cells and 1/240 and 1/180 (P=0.435) for CD44⁺CD24⁺ cells in NSG and NOD/SCID mice, respectively (Table 2). Thus, CD44⁺CD24⁺ or ALDH⁺ cells were equally tumorigenic in NOD/SCID and NSG mice, similar to findings using bulk pancreatic cancer cells. We also examined the histology of tumors formed by both strains of mice and found that they were all similar to the original surgical specimen (Figure S2). These data demonstrate that the degree of immunodeficiency of the recipient mouse strain does not alter the frequency or phenotype of pancreatic cancer cells with enriched tumorigenic potential.

We also examined whether the proportion of TICs changed as tumors were serially propagated as xenografts. In non-small cell lung adenocarcinoma, TIC frequency increased from 1/16,000–1/30,000 in primary specimens to >1/10 – 1/320 in extensively passaged tumors (6 - 10 generations; Table S1). These findings are similar to those recently reported in melanoma (Boiko et al., 2010), and also demonstrate our ability to detect and quantify highly tumorigenic cells under our xenograft conditions. Moreover, the use of subcutaneous implantation did not appear to negatively impact TIC frequency as orthotopic intratracheal or intravenous injections did not result in an increase in TIC frequency (Table S2).

Our data demonstrate that TICs are relatively infrequent in pancreatic adenocarcinoma, lung adenocarcinoma, squamous cell lung carcinoma, and head and neck squamous cell carcinoma. These findings are in agreement with studies of several other human malignancies (Al-Hajj et al., 2003; Boiko et al., 2010; Dalerba et al., 2007; Fang et al., 2005; Lapidot et al., 1994; Li et al., 2007; Matsui et al., 2004; Prince et al., 2007; Rasheed et al., 2010; Schatton et al., 2008; Singh et al., 2004; Suva et al., 2009; Yang et al., 2008). In contrast, Quintana et al. recently reported that approximately 25% of unselected melanoma cells are tumorigenic under more permissive conditions (cells co-injected with Matrigel in NSG mice) (Quintana et al., 2008). The variability in these results may not be entirely due to the use of xenotransplantation assays to measure TIC frequency as similar findings have been observed in mouse cancer models using syngeneic transplants (Bruce and Van Der Gaag, 1963; Kelly et al., 2007). An obvious explanation for the differences in TIC frequency and detection is the specific malignancy studied. Clinically, melanoma sometimes responds to immune-based therapeutics, such as interferon-alpha or interleukin-2 (Smith et al., 2008), whereas these approaches are generally ineffective for most malignancies, including pancreatic cancer (Bellone et al., 2006). Therefore, residual immune cells, most likely NK cells, in NOD/SCID mice decrease the apparent TIC frequency in melanoma but not in other cancers. Additional features, such as tumor stage or grade, may also affect the frequency of TICs in a specific malignancy (Shackleton et al., 2009), and recent findings have demonstrated that metastatic melanoma lesions are relatively enriched for TICs compared to primary tumors (Boiko et al., 2010). The tumors in our studies were all obtained from patients with localized disease and it remains to be seen whether TIC frequency remain low in metastatic pancreatic, lung and head and neck carcinomas.

Our findings also suggest that the use of primary tumors or xenografts may influence TIC frequency. We found that TICs were relatively rare in primary non-small cell lung adenocarcinomas, but significantly increased if tumors were heavily passaged as xenografts. Similarly, a recent report in melanoma (Boiko et al., 2010) demonstrated increased TIC frequency when melanomas were passaged as xenografts, and these data may in part explain the findings of the Quintana et al. study (Quintana et al., 2008). We also found that the distinct cellular phenotypes (ALDH⁺ and CD44⁺CD24⁺) that enrich for tumorigenic cells in NOD/SCID mice are preserved in NSG mice. These data suggest that a more permissive xenotransplantation model does not alter the functional characteristics of specific populations enriched for TICs in pancreatic cancer. However, the impact of xenotransplantation into NSG rather than NOD/SCID mice on the tumor-initiating capacity of specific cell populations in other malignancies remains to be seen. The data presented here and previous findings in melanoma emphasize the need to study TICs within the context of specific malignancies and disease states.