Heterogeneity in Colorectal Cancer Stem Cells

It has long been recognized that malignant cells within a single tumor display significant heterogeneity in morphology, proliferative activity, and function (1). The cancer stem cell (CSC) concept provides a convincing explanation for the mechanisms underlying this heterogeneity. CSCs, also called tumor-initiating cells, are defined by their capacity to self-renew and generate diverse cells that comprise the tumor (2). They are thought to initiate and continually sustain tumor growth.

The first evidence for the existence of CSCs came from studies on acute myelogenous leukemia in the 1990s, in which a rare subset of leukemic cells with a CD34⁺/CD38⁻ phenotype was shown to induce leukemia in immunocompromised mice (3, 4). A decade following the identification of leukemic stem cells, the discovery of a CSC subset in breast cancers (5) expanded the concept to solid tumors and thereafter CSCs were identified in multiple solid tumors, including colorectal cancer (6–8). Because CSCs share phenotypic and molecular characteristics with their corresponding normal tissue-resident stem cells, and intestinal stem cells (ISC) play a crucial role in maintenance of homeostatic epithelial renewals in the normal colonic tissues (9), colorectal cancer provides a compelling model of a hierarchically organized solid tumor, with CSCs at the apex. Furthermore, data from ISCs should offer clues to understanding CSCs in colorectal cancer.

In this review, we summarize the past and current evidence related to heterogeneity within colorectal CSC populations and our understanding of the potential mechanisms underlying this heterogeneity.

Since the initial publication on brain tumors (10), CD133 (also known as prominin-1) has been established as a marker for CSC populations in a variety of cancers. Human colorectal CSCs were first identified by CD133 expression (7, 8). CD133⁺ cells were consistently capable of generating tumors that resembled the original patient tumor when serially transplanted in immunocompromised mice, while their CD133⁻ counterparts failed to give rise to xenografts. To date, other markers have been reported to identify CSCs in human colorectal cancers (Table 1), including CD44 (6), CD44v6 (11), and aldehyde dehydrogenase 1 (ALDH1) (12).

Flow-cytometric analysis has revealed overlaps between populations of CD133⁺ colorectal cancer cells and populations expressing CD44, CD29, CD24, and CD166, all of which have been described as enriched for colorectal CSCs (13). This finding suggests that CD133 expression is the most broadly distributed marker for colorectal CSCs. However, as has been reported in human glioblastomas (14–16), colorectal cancer cells with CSC properties do not necessarily express CD133. It has been shown that both CD133⁺ and CD133⁻ subsets from hepatic metastases of colorectal tumors are capable of reconstituting tumors when subcutaneously injected into immunocompromised mice; moreover, tumors derived from CD133⁻ cells grew at a more rapid rate (17). Further analysis revealed that CD133⁻ cells frequently expressed CD44, another CSC marker (6), in colonospheres derived from CD133⁻ cells (17). The following finding is also indicative of the existence of CSCs phenotypically distinct from CD133⁺ cells in colorectal tumors: CD44⁺/EpCAM^high cells with CSC properties were found even in colorectal tumors in which CD133⁺ cells were not contained (6).

Complicating matters further, it has been shown that CD133 and CD44 can be nonmutually exclusive markers, as a partial overlap between the two cell subsets in colorectal cancers has been repeatedly reported (6, 13, 18). CD133⁺/CD44⁺ cells induced tumors in immunocompromised mice under conditions in which the same number of CD133⁺/CD44⁻ cells failed to engraft, indicating that CD44 provides further enrichment of CSCs in the CD133⁺ subset (18).

Distant metastasis is the primary cause of lethality in patients with colorectal cancer. Previous studies have suggested that only certain subsets of CSCs are capable of distant metastasis (Table 1). Experiments showing differences in metastatic capacity between CSC subsets are the most convincing demonstrations of functional heterogeneity.

Hermann and colleagues were the first to identify a specific subset of CSCs responsible for metastasis in human pancreatic cancer cell lines, based on the cell-surface expression of CD133 and CXCR4, a receptor for the chemokine CXCL12 (19). When orthotopically injected into nude mice, both CD133⁺/CXCR4⁺ and CD133⁺/CXCR4⁻ cells could produce tumors at the injection site, but only CD133⁺/CXCR4⁺ cells produced liver metastases (19). A subsequent study showed that the colorectal cancer cell line HCT116 also contains CD133⁺/CXCR4⁺ cells, which have a significantly higher metastatic capacity in nude mice than CD133⁺/CXCR4⁻ cells (20).

Pang and colleagues demonstrated that a subpopulation of colorectal CSCs expressing CD26 has both tumor-initiating and metastatic capacities (21). Orthotopic implantation into mice of CD133⁺/CD26⁺ cells isolated from the primary tumor of a colorectal cancer patient with hepatic metastasis produced liver metastases following tumor formation in the cecal wall, while their CD26⁻ counterparts induced tumor growth only at the site of injection (21). They also showed that circulating CD133⁺/CD26⁺/CD44⁺ cells could be detected in portal blood after cecal-wall injection and that intraportal injection of CD133⁺/CD26⁺/CD44⁺ cells, but not CD133⁺/CD26⁻/CD44⁺ cells, led to liver metastasis (21). In vitro evaluations revealed that CD26 knockdown by small interfering RNA (siRNA) reduced the migratory and invasive capacities of the CD26⁺ cells, with a downregulation of epithelial–mesenchymal transition (EMT) markers (21). Consistent with these findings, clinical studies have reported that CD26 expression in colorectal cancer is correlated with poor prognoses (22–24). Interestingly, CD26⁺/CD326⁻ circulating cancer cells have been proposed as prognostic markers for the recurrence of colorectal cancer (24).

In a recent study, Todaro and colleagues phenotypically identified colorectal CSCs with metastatic capacity based on their expression of the variant form 6 of CD44 (CD44v6; ref. 11). CD44v6⁺ cells were able to induce tumor growth in gut, lung, and liver after orthotopic injection into mice, while their CD44v6⁻ counterparts were not metastatic (11). Interestingly, while there was a substantial overlap between CD44v6⁺ and CD26⁺ populations, CD44v6⁺/CD26⁻ cells showed considerable metastatic potential in the orthotopic model (11), suggesting phenotypic heterogeneity even within metastatic CSCs.

Interestingly, a recent study suggested that preferential metastasis to particular target organs could be attributable in part to phenotypic/functional diversity within metastatic CSCs. Gao and colleagues reported that only colorectal CSCs expressing CD110, a specific receptor for thrombopoietin, were able to colonize the liver after orthotopic implantation in immunocompromised mice, while CSCs expressing CUB-domain-containing protein 1 (CDCP1) produced lung metastases (25). They showed that knockdown of either CD110 or CDCP1 by siRNA reduced liver or lung metastasis, respectively, but neither had a discernible effect on primary tumor growth. In addition, they showed that CD110 and CDCP1 functioned in extravasation into the liver parenchyma and adhesion on the pulmonary endothelium, respectively (25). It is notable that both CD110⁺ and CDCP1⁺ CSCs were exclusively contained within the CD133⁺ population with no overlap between these subpopulations (25), indicating that distinct metastatic CSC subsets can be contained within CD133⁺ CSC populations.

Recent studies demonstrate colorectal CSCs to be dynamic rather than static populations that are continuously altered by multiple factors including genetic and epigenetic alterations, interactions between tumor cells, microenvironmental factors, hormone action, and cancer therapy. Any or all of these factors could contribute to producing heterogeneity in colorectal CSC populations. This section summarizes past and current findings related to the dynamics of colorectal CSCs, particularly the signaling pathways and intra- and extratumoral factors involved in their regulation. While ISCs are not the primary focus of this section, we have provided a brief description of them, focusing on ISC findings that offer clues to understanding CSC dynamics in colorectal cancer.

CSCs share phenotypic and molecular characteristics with their normal counterparts, tissue-resident, adult stem cells. Notably, the colorectal CSC markers identified to date are also expressed by normal ISCs (12, 26). Conversely, the human homolog of the best-documented marker for murine ISCs, leucine-rich repeat-containing G-protein–coupled receptor 5 (LGR5), could also serve as a marker for CSCs in human colorectal cancers (27). A recent study showed that LGR5-expressing cells possess CSC properties in xenografted organoids derived from human colorectal cancer (28). CSCs do not necessarily originate from the transformation of normal stem cells, but in the case of colorectal cancer, the similarity could be attributable, at least in part, to the origin of CSCs. Colorectal cancer develops and progresses by the sequential accumulation of genetic alterations, in which activation of the Wnt/β-catenin signaling pathway, via an APC or CTNNB1 (β-catenin) mutation, marks the first step in tumor formation. Given that intestinal epithelial cells are consistently renewed for a shorter period than that required to accumulate causative genetic alterations, it is reasonable to hypothesize that CSCs arise from long-lived ISCs. The discovery of reliable murine ISC markers, including Lgr5 (9) and B lymphoma Mo-MLV insertion region 1 (Bmi1; ref. 29), enables the testing of this hypothesis. Actually, it has been demonstrated that specific activation of Wnt/β-catenin signaling in ISCs expressing Lgr5, Bmi1, or Cd133 results in adenoma formation in mice (29–31).

These findings do not necessarily preclude the possibility that intestinal epithelial cells other than ISCs can be the founders of intestinal tumors under specific conditions. In fact, activation of Wnt and Notch signaling in basic helix–loop–helix family member a15 (Bhlha15)-positive progenitor cells resulted in the rapid development of intestinal tumors in mice (32). Unlike observations from ISCs (29–31), Wnt activation alone in Bhlha15⁺ cells was insufficient to drive colonic carcinogenesis (32), suggesting that ISCs are at higher risk for tumorigenic transformation. Similarly, activation of Wnt signaling in doublecortin-like kinase 1 protein (Dckl1)-positive tuft cells led to colorectal tumor development in mice under inflammatory conditions, but not under normal conditions (33).

ISCs are one of the most intensively studied subjects in stem cell biology. While previous works based on label-retaining cells (34) hypothesized the existence of a single quiescent population of stem cells residing in a specific location of the intestinal crypt, it is currently accepted that both quiescent and active ISCs coexist in distinct niches (35). Both populations possess the capacity to self-renew and give rise to all differentiated intestinal epithelial cell types, despite having entirely different proliferative activities. In the small intestine of mice, Lgr5 is the best-known marker for active ISCs (9), while quiescent ISCs have multiple markers including Bmi1 (29), homeodomain-only protein (Hopx; ref. 36), telomerase reverse transcriptase (Tert; ref. 37), and leucine-rich repeats and immunoglobulin-like domains protein 1 (Lrig1; ref. 38). Active and quiescent ISCs represent functionally distinct ISC populations: actively cycling Lgr5⁺ ISCs contribute to homeostatic epithelial renewal, while slow-cycling Bmi1⁺ ISCs are considered to be a reserve stem cell pool, based on their active contribution to regeneration after radiation damage (39). Consistently, it has been reported that Bmi1⁺ ISCs give rise to Lgr5⁺ ISCs in the small intestine of mice after the selective ablation of Lgr5⁺ cells (40). Conversely, Takeda and colleagues showed that Lgr5⁺ cells can give rise to Hopx⁺ cells in mice and in organoid cultures (36). These results indicate that active and quiescent ISCs can interconvert and/or replenish each other. In the colon of mice, Lgr5⁺ cells also reside at the crypt base and serve as active ISCs (9), while Lrig1 marks predominately quiescent stem cells (38). Asfaha and colleagues identified Lgr5⁻ cells with ISC properties within a Krt19⁺ population distributed above the crypt base in the colon of mice (41). Genetic lineage-tracing experiments revealed that Krt19⁺/Lgr5⁻ colonic cells generated entire colonic crypts, including Lgr5⁺ cells at the colonic crypt base (41).

The differentiation of intestinal epithelial cells normally proceeds in a unidirectional manner, from ISCs to the terminally differentiated cells via lineage-committed progenitor cells, but recent studies have demonstrated the dedifferentiation capacity of the intestinal epithelial cells. In the small intestine of mice, Alkaline phosphatase intestinal (AlPi)-expressing enterocyte progenitors can dedifferentiate into Lgr5⁺ ISCs upon the depletion of Lgr5⁺ cells (42); similarly, Delta-like 1 (Dll1)^high secretory progenitor cells regenerate stem cell compartments that contain Lgr5⁺ cells following irradiation damage (43). In the colon, atonal homolog-1 (Atoh1)⁺ or Bhlha15⁺ secretory progenitor cells significantly contribute to colonic crypt regeneration after mucosal damage (32, 44–46). These findings indicate the dedifferentiation capacity of the intestinal epithelium; however, this capacity seems limited to fated progenitors rather than to terminally differentiated cells.

Recent studies have demonstrated the reversion of differentiated non-CSC populations to CSCs in novel colorectal cancer models using genetically engineered organoids (28, 47). By xenotransplantation of organoids derived from human colorectal cancer cells, Shimokawa and colleagues demonstrated that KRT20⁺ differentiated tumor cells revert to LGR5⁺ CSCs and contribute to tumor growth after selective ablation of LGR5⁺ CSCs (28). Likewise, the recovery of functional Lgr5⁺ CSCs was also observed in transplanted mouse colon cancer organoids, in which Lgr5⁻ cells rapidly replenished Lgr5⁺ CSC populations after the ablation of Lgr5⁺ cells (47).

Consistent with the importance of Wnt/β-catenin signaling in both intestinal stemness and colon carcinogenesis, Wnt signaling plays a central role in the regulation of colorectal CSCs. Although other signaling pathways have been implicated in the control of colorectal CSCs, here we focus on Wnt/β-catenin signaling. Vermeulen and colleagues demonstrated that colorectal cancer cells show variable levels of Wnt activation and that only those with the highest levels possess CSC properties (48). A study from our group also supports the notion that the Wnt activity levels define colorectal CSCs. We investigated the dose-dependent effect of Wnt activation in the mouse colonic epithelium by regulating the expression levels of S33Y mutant β-catenin (49). Higher levels of mutant β-catenin expression induced the amplification of Lgr5⁺ cells in colonic crypts with de novo crypt formation, whereas lower levels of expression only enhanced cell proliferation (49). Consistent with these findings, Ordonez-Moran and colleagues showed that HOXA5 induction counteracted CSC traits, preventing tumor growth and metastasis by inhibiting Wnt signaling activity in colorectal cancers (50). Colorectal CSCs is regulated through the interaction between Wnt and other signaling pathways, as is the normal ISCs. For example, bone morphogenetic protein 4 (BMP4) promotes differentiation and apoptosis by antagonizing Wnt signaling in colorectal CSCs (51). In a recent study, Whissell and colleagues identified the transcription factor GATA-binding factor 6 (GATA6) as a key regulator of Wnt and BMP signaling in colorectal cancers (52). GATA6 enabled CSC self-renewal through the repression of BMP gene expression, competing with β-catenin/Tcf4 to bind to a regulatory region of the BMP4 locus (52).

Microenvironmental factors are important in the regulation of colorectal CSCs as well as of their normal counterparts, ISCs (53). Tumor stroma undergoes dramatic changes during the process of tumor progression and therefore could produce more complex effects than it does in normal homeostatic conditions. Hepatocyte growth factor (HGF) secreted from myofibroblasts induced CSC properties in Wnt^Low colorectal cancer cells by enhancing Wnt signaling activity (48). When cocultured with colorectal cancer cell lines, mesenchymal stem cells increased the number of cancer cells with tumor-initiating capacity by producing prostaglandin E2 and cytokines that activated Wnt/β-catenin signaling (54). It is possible that CSCs acquire a metastatic capacity by the actions of cytokines secreted from cancer-associated fibroblasts (CAF) during tumor progression. As described above, colorectal CSCs with metastatic capacity are phenotypically identified by CD44v6 expression (11). Cytokines secreted from CAFs, including HGF, osteopontin, and stromal-derived factor 1α, enhanced CD44v6 expression by activating Wnt signaling, transforming nonmetastatic progenitors into metastatic CSCs (11). In addition to stromal cells, Paneth cells and cKit⁺ secretory cells constitute niches for Lgr5⁺ ISCs in the murine small intestine and colon, respectively, where they provide essential signals for stem cell maintenance, including the Notch ligand Dll4 (55, 56). As Notch signaling is elevated in colorectal CSCs (57) and antibody blockade of DLL4 reduces CSC frequency in colorectal cancers (57), the CSC state might also be regulated through interactions between tumor cells in colorectal cancers.

The regulation of Wnt and BMP4 signaling pathways by activated thyroid hormones has been reported in colorectal CSCs (58). Type 3 deiodinase 3 (D3), a chief thyroid hormone T3–inactivating enzyme, has been shown to be expressed at high levels in CD133⁺ and Wnt^high cell populations (58). T3 treatment induced CSC differentiation, decreasing the tumorigenic potential in CSCs, accompanied by the upregulation of BMP4 and attenuation of Wnt signaling (58). These findings suggest that colorectal cancers may be regulated not only by local factors within or surrounding a lesion, but also by hormonal action.

The CSC concept has attracted a great deal of interest for its potential clinical implications. Colorectal cancer remains the fourth leading cause of cancer-related deaths worldwide. Although its occurrence is declining in developed countries, its incidence is still rapidly rising in many developing countries (59).

Multiple reports have shown that colorectal CSCs are more resistant to chemotherapy (60, 61) and therefore likely play an essential role in recurrence following conventional anticancer treatments. Enrichment of colorectal CSCs in xenografts after chemotherapy clearly showed the limitation of conventional therapies for colorectal cancers (28, 60). Therefore, CSCs represent an attractive target for more effective therapies, and several potential CSC-targeted drugs or strategies have been actually proposed (62). However, considering the heterogeneity and dynamism of colorectal CSC populations, CSCs no longer represent a fixed target population, making it necessary to establish improved strategies, designed to counteract to their dynamic nature.

Recent studies have indicated that the selective ablation of colorectal CSCs is not sufficient to produce complete tumor regression (28, 47). In allografts of mouse colorectal cancer organoids, continuous ablation of Lgr5⁺ CSCs restricted tumor growth but did not produce tumor regression, because proliferative Lgr5⁻ cells maintained the tumors even with depletion of Lgr5⁺ CSCs (47). Furthermore, Lgr5⁻ cells can rapidly replenish the Lgr5⁺ population to reinitiate tumor growth after cessation of Lgr5⁺ cell ablation (47). A similar phenomenon was observed in xenografted human colorectal cancer organoids (28). Upon ablation of LGR5⁺ CSCs, tumor size was reduced (28). However, compensatory proliferation of KRT20⁺ non-CSC cells reduced the effectiveness of the ablation and the tumor eventually regrew in parallel with the reappearance of LGR5⁺ cells (28).

Although CSCs were initially considered to be cell populations with well-defined phenotypic and molecular characteristics, CSCs have more recently been shown to be phenotypically/functionally heterogeneous and highly dynamic populations. These characteristics present a serious problem in establishing therapeutic strategies targeting CSCs. Considering that the stemness of colorectal cancer cells is a dynamic state that is constantly altered by multiple extrinsic factors, in addition to intrinsic cellular factors (genetic and epigenetic alterations), a better understanding of the tumor environment should lead to improved strategies for the eradication of colorectal CSC by regulating their CSC state.

No potential conflicts of interest were disclosed.