Introduction

Colon cancer still represents an ideal model to study the molecular and cellular mechanisms that underlie tumour onset and progression towards malignancy [1], the so-called adenoma-carcinoma sequence [2]. Overall, loss-of-function mutations in the APC (adenomatous polyposis coli) tumour suppressor gene or oncogenic mutations in β-catenin (CTNNB1) lead to the constitutive activation of Wnt/β-catenin signalling and represents the most common rate limiting (adenoma-forming) events among colon cancer patients. Adenoma growth and progression is often accompanied by alterations in KRAS or BRAF, followed by loss of the TP53 and SMAD4 tumour suppressors thought to underlie the malignant transformation into locally invasive adenocarcinoma [1]. However, this well established genetic evolution model does not take into account other essential characteristics of human colon cancers, namely their cellular heterogeneity (different cell lineages are often present within the primary mass) and the putative role played by a subpopulation of tumour cells, the cancer stem cells (CSCs), in driving tumour growth and determining local invasion into surrounding tissues and distant metastasis [3]. In fact, although the above genetic model would predict that every tumour cell within a colon cancer allegedly initiated by an APC or β-catenin mutation should invariably be earmarked by the hallmark of constitutive Wnt activation, namely nuclear β-catenin accumulation, this is only observed in a minority of cells usually located at the invasive front of the primary lesion [4] from where they detach and invade the surrounding stroma [5], [6]. This “β-catenin paradox” nicely illustrates how intra-tumour heterogeneity and possibly tumour stemness ensue at the very initial stages of the adenoma-carcinoma sequence and lead to different Wnt signalling levels among different tumour cells lineages sharing the same (APC) mutations [7]. It also indicates that loss of APC function (or oncogenic β-catenin activation) is presumably necessary for the onset of the initial dysplastic lesion but insufficient to fully activate Wnt signal transduction and promote malignant transformation in the absence of additional environmental and (epi)genetic factors.

Previously, by employing in vivo mutagenesis [8], [9] and gene targeting in the mouse [10], [11], it was shown that loss of Apc function results in adenoma formation in the upper GI tract. However, these mouse adenomas fail to progress to malignancy and do not spontaneously accumulate additional genetic hits at the endogenous Kras and Tp53 genes [12]. Notably, whereas oncogenic KRAS activation on its own is unable to initiate intestinal tumorigenesis if not with very late onset and only upon somatic hits at the Tp53 gene [13], compound Apc^1638N/+/KRAS^V12G mice are characterized by a 10-fold increase in tumour multiplicity and by accelerated tumour progression when compared with Apc^1638N/+ littermates, with the vast majority of the tumour lesion being represented by adenocarcinomas [14]. Further analyses revealed that Apc and KRAS mutations are synergistic in promoting β-catenin nuclear translocation, thus enhancing canonical Wnt signal transduction [14]. The latter is likely to result from the ability of activated KRAS, through downstream and yet unknown kinases, to induce β-catenin tyrosine phosphorylation thus leading to a substantial increase of its cytoplasmatic pool and its subsequent translocation to the nucleus where it acts as a transcriptional activator of several Wnt downstream target genes. Accordingly, intestinal tumours from Apc^1638N/+/KRAS^V12G mice show a significant increase in cells with nuclear accumulation of β-catenin when compared with Apc^1638N/+ animals [14].

In recent years, CSCs have been successfully purified from human colon cancers by employing different cell surface markers such as CD133 [15], [16], EpCAM, CD44 and CD166 [17], and EphB2 [18]. However, although the above cell surface antigens have been instrumental for the identification of tumour cell subpopulations with enriched tumour-initiating properties when transplanted into immune-incompetent recipient mice, our understanding of the mechanisms underlying intestinal cancer stemness and of the role played by CSCs in progression towards malignancy is still largely incomplete. Previously, Vermeulen et al. showed that high Wnt activity earmarks CSCs within suspension spheres derived from colon tumors [19]. From this perspective, a number of additional issues need to be addressed: is nuclear β-catenin accumulation a functional marker for intestinal CSCs in vivo? Are tumor cells with stem-like properties already present in early, benign lesions such as adenomas? Here, we took advantage of the Apc^1638N/+ and Apc^1638N/+/KRAS^V12G mouse models for intestinal tumorigenesis to prospectively identify subpopulations of tumour stem-like cells and characterize them with regard to their multipotency, self-renewal, genome-wide expression profile, and Wnt/β-catenin signalling activity.

Results

Tumour-initiating Cells are Present in Apc^1638N/+/KRAS^V12G but are Very Rare in Apc^1638N/+ Intestinal Tumours

The Apc^1638N/+ mouse model develops an average of 4–5 benign upper GI tumours (adenomas) in the C57Bl/6J genetic background [10], [11]. These lesions only rarely (and with late onset) develop into adenocarcinomas as also shown by the lack of spontaneous somatic mutations occurring at the Kras and Tp53 genes [12]. To assess the presence of tumour-initiating cells in Apc^1638N/+ adenomas, i.e. cells capable of recapitulating the primary lesion when transplanted into a recipient animal, intestinal tumours were collected from Apc^1638N/+ animals and dissociated both mechanically and enzymatically into single cell suspensions and depleted from endothelial and hematopoietic cells (Lin⁺) by fluorescence-activated cell sorting (FACS). Next, different multiplicities of the resulting Lin⁻ population of bulk tumour cells were transplanted subcutaneously into NOD-SCID animals. As shown in Table 1, no tumour growth was observed even upon injection of as many as 0.5–1.0*10⁶ Lin^- cells and 6 months after transplantation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Subcutaneous transplantation of bulk intestinal tumor cells in immune-incompetent mice indicates the presence of CSCs in Apc^1638N/+/KRAS^V12G but not in Apc^1638N/+ tumours.

https://doi.org/10.1371/journal.pone.0073872.t001

Next, we repeated the transplantation assay with bulk Lin⁻ tumour cells from Apc^1638N/+/KRAS^V12G intestinal tumours. As previously reported, the majority of the intestinal tumours found in these compound animals in the same inbred C57B6/J genetic background are locally invasive adenocarcinomas [14]. In sharp contrast with the Lin^- cells from Apc^1638N/+ adenomas, tumour growth was observed in 23 out of 33 injections with 10⁵ Apc^1638N/+/KRAS^V12G Lin^- cells and, though at lower incidence (3 out of 48 transplantations), even with as low as 1500 cells (Table 1). By limiting dilution analysis (L-Calc™) the frequency of tumour-initiating cells in the Lin⁻ population from Apc^1638N/+/KRAS^V12G intestinal tumours was estimated as 1 in 72838 (95% CI of 109467 to 48465). As for the Apc^1638N/+ adenomas, tumour-initiating cells are likely to be present, if at all, at considerably lower frequencies (<10⁻⁶). Hence, the presence of tumour-initiating cells, as defined by transplantation assays, appears to be limited to the tumours from Apc^1638N/+/KRAS^V12G mice when compared with those from the Apc^1638N/+ mouse model.

The Lin-CD24^hiCD29⁺ Subpopulation from Apc^1638N/+/KRAS^V12G Intestinal Tumours Encompass Tumour-initiating and Self-renewing CSCs

In order to prospectively enrich and eventually isolate tumour-initiating cells from the bulk Lin^- population of Apc^1638N/+/KRAS^V12G tumours, we first tested a panel of previously established (cancer) stem cell markers including CD24, CD29 (β1 integrin), CD44, CD97, and L1CAM by FACSorting and subsequent transplantation in NOD-SCID mice. In contrast to CD44, L1CAM, and CD97 (Table S1), transplantation of Lin^-CD24⁺CD29⁺ cells revealed a slight yet significant enrichment in tumour-initiating cells (estimated frequency of 1 in 56463, with CI of 399211 to 7986) (Table 2). Given that the Lin^-CD24⁺CD29⁺ population represents a relatively large proportion of the bulk cells (∼80%; data not shown), we then further defined three additional FACS gates based on the relative expression of the CD24 cell surface antigen (CD24^hi, CD24^med, and CD24^low) (Figure 1a). Out of 30 primary Apc^1638N/+/KRAS^V12G tumours analysed by FACS, the average size (expressed in percentage of the bulk Lin⁻ fraction) of each CD24/CD29 sorted subpopulation was determined: CD24⁻CD29⁻, 3.4% (SD 2.6); CD24⁻CD29⁺, 7.8% (SD 4.7); CD24⁺CD29⁻, 4.4% (SD 3.4); CD24^loCD29⁺ (P1), 7.9% (SD 2.5); CD24^medCD29⁺ (P2), 52.9% (SD 7.4); CD24^hiCD29⁺ (P3), 10.0% (SD 4.7). Please note that these percentages do not add to 100% simply because of the deliberate exclusion of cells located in the separations between sorting gates (see Figure 1a).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. FACS analysis of cell suspensions from Apc^1638N/+/KRAS^V12G tumours.

a. Large panel: dot plot representative of the staining pattern obtained by staining with anti-CD24 APC-conjugated and anti-CD29 PE-conjugated antibodies. Lineage positive cells (Lin⁺) were excluded (gated out) by staining with biotinylated antibodies against lineage markers and Streptavidin-PerCPCy5.5. P1 (Lin⁻CD24^lowCD29⁺), P2 (Lin⁻CD24^medCD29⁺), and P3 (Lin⁻CD24^hiCD29⁺) populations are indicated in the plot. Small panels: dot plots representative of cells stained with isotypic control antibodies (left), compensation control stained only with anti CD24-APC antibodies (middle), compensation control stained only with anti CD29-PE antibodies (right). b. FACS analysis of the CD24/CD29 pattern of tumours obtained by serial transplantation of P3 cells suspensions from Apc^1638N/+/KRAS^V12G intestinal tumours. Left: primary transplantation. Right: secondary transplantation. c. Immunohistochemistry analysis of tumors obtained by 3 rounds of serial transplantation of P3 cells suspensions from Apc^1638N/+/KRAS^V12G intestinal tumours.

https://doi.org/10.1371/journal.pone.0073872.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Subcutaneous transplantation of sorted cells from Apc^1638N/+/KRAS^V12G tumours in immune-incompetent mice indicates the presence of tumor-initiating cells in the CD24^hiCD29⁺ subpopulation.

https://doi.org/10.1371/journal.pone.0073872.t002

The frequency of tumour-initiating cells within the P1, P2 and P3 subpopulations was determined by transplanting 1500 cells of each into NOD-SCID mice. Notably, whereas at this multiplicity CD24^medCD29⁺ and CD24^loCD29⁺ cells consistently failed to form tumours (only 1 growth out of 56 transplantations), the Lin^-CD24^hiCD29⁺ subpopulation was found to encompass a substantial enrichment in tumorigenic cells (13/28; Table 2). Hence, although in this case we could not perform limiting dilution analysis by L-Calc™ (since a fixed multiplicity of cells was employed) the Lin^-CD24^hiCD29⁺ tumour subpopulation appears to be characterized by a significant relative enrichment of approx. 20–25 fold when compared with total Lin^- bulk cells (1 CSC out of ∼3000 tumour cells vs. 1 in 72838).

The definition of cancer stem cells cannot be exclusively based on their ability to form tumours when transplanted at low multiplicity in immune-incompetent mice. Equally important characteristics of CSCs are their unique ability to self-renew and differentiate to fuel and recapitulate the heterogeneous composition of the primary tumour they are derived from. In order to determine whether the tumour-initiating cells encompassed by the Lin^-CD24^hiCD29⁺ population are also capable of self-renewal and differentiation, we performed serial transplantations experiments. First, 1500 Lin^-CD24^hiCD29⁺ cells were isolated from Apc^1638N/+/KRAS^V12G primary intestinal tumours and transplanted in NOD-SCID recipient mice. As anticipated from our previous results, this initial assay gave rise to subcutaneous tumours within 8 to 10 weeks. The resulting tumours were then excised from the recipient NOD-SCID animals and employed for both FACS and histological analysis. As for FACS, tumours were dissociated into single cell suspensions and analysed, sorted and transplanted according to their CD24 and CD29 expression levels. Secondary tumours originated from Lin^-CD24^hiCD29⁺ cells fully recapitulated the CD24/CD29 FACS expression profile of the primary lesions (Figure 1b and Figure S2). Likewise, upon transplantation of 1500 cells from each of the Lin⁻CD24CD29 populations obtained from the secondary tumours, only the Lin⁻CD24^hiCD29⁺ cells were capable of forming tertiary tumours. Accordingly, the FACS profile of the tertiary tumours recapitulates that of the primary lesions (Figure 1b). Notably, immunohistochemistry (IHC) and enzymatic staining revealed a progressive increase in the relative presence of intestinal differentiation lineages, namely Goblet (Periodic Acid Schiff; PAS), Paneth (lysozyme), and entero-endocrine (synaptophysin) cells in the transplanted intestinal tumours when compared with the primary Apc^1638N/+/KRAS^V12G lesions (Figure 1c). However, FACS analysis of the serially transplanted tumours showed that the relative size of the individual CD24/CD29 subpopulations did not significantly changed (Figure S3).

Thus, the Lin^-CD24^hiCD29⁺ subpopulation of tumour cells from Apc^1638N/+/KRAS^V12G adenocarcinomas encompass bona fide CSCs with tumour-initiating, self-renewing and differentiation capacities.

Lin-CD24^hiCD29⁺ Cells from Apc^1638N/+/KRAS^V12G Intestinal Tumours Show Increased Intracellular β-catenin Accumulation

We previously proposed that the minority of colon cancer cells featuring nuclear β-catenin accumulation and non-randomly distributed along the invasive front, represent CSCs [7]. Notably, both Apc^1638N/+ adenomas and Apc^1638N/+/KRAS^V12G carcinomas share the same “β-catenin paradox” observed in human colon cancers in that, upon IHC analysis, only a minority of tumour cells show nuclear β-catenin accumulation notwithstanding that the majority, if not all, share the two-hits at the Apc locus [12], [14] (Figure 2a). To assess whether the CSCs enriched in the Lin⁻CD24^hiCD29⁺ tumour subpopulation are characterized by an increased level of intracellular β-catenin, we analysed protein expression in the different FACSorted tumour cell subpopulations by two independent assays, namely immuno-staining and western blot analysis. Immuno-staining showed that the majority of Lin^-CD24^hiCD29⁺ intestinal tumour cells are characterized by intracellular accumulation of β-catenin when compared with other sorted populations and the bulk (Lin⁻) tumour cells (Figure 2b). This result was also confirmed in a more quantitative fashion by western analysis performed with antibodies specific for the signalling-competent fraction (i.e. dephosphorylated at residues Ser37 and Thr41) of the β-catenin protein (Figure 2c and Figure S4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. β-catenin expression analysis in Apc^1638N/+ and Apc^1638N/+/KRAS^V12G intestinal tumours.

Immuno-histochemistry (a.,b.) and western blot (c.) analysis of β-catenin in primary Apc^1638N/+ intestinal adenomas (a.) and in FACSorted tumour populations from Apc^1638N/+/KRAS^V12G intestinal tumours (b. and c.). The bars in c. represents the quantification of the bands obtained with an anti-active β-catenin Ab (anti-ABC; clone 8E7, #05–665, Millipore) by scanning and analyzing the western blot with the Odyssey scanner and after normalization with β-actin.

https://doi.org/10.1371/journal.pone.0073872.g002

Overall, these data confirm that the Lin^-CD24^hiCD29⁺ subpopulation from Apc^1638N/+/KRAS^V12G tumours, here shown to be enriched in CSCs, encompasses a significantly higher level of intracellular and signalling-competent β-catenin when compared to bulk Lin⁻ and other tumour cell subpopulations. The increased Wnt signalling activity in CSCs was later confirmed by expression profiling and Taqman qPCR analysis of Wnt downstream target genes (e.g. Lgr5, Axin2, T, and Lef1; see here below).

The Expression Signature of CSCs from Apc^1638N/+/KRA^SV12G Intestinal Tumours is Distinct from that of Differentiated and Bulk Tumour Cells and Encompasses Both Stem and Paneth Cell Markers

To identify molecular differences between stem-like and more differentiated (bulk) tumour cells from Apc^1638N/+ and Apc^1638N/+/KRAS^V12G intestinal tumours, we isolated total RNA from 10⁴ Lin^-CD24^hiCD29⁺ (P3), Lin^-CD24^medCD29⁺/Lin⁻CD24^loCD29⁺ (P1+P2, merged gate) and Lin^- (bulk) tumour cells from 5 individual mice of each genotype (Apc^1638N/+ and Apc^1638N/+/KRAS^V12G). Total RNA samples were then employed to hybridize oligonucleotide microarrays (Affymetrix Mouse Genome 430A 2.0 Array) according to conventional protocols.

First, the different tumour cell populations isolated from mice with different genotypes (Apc^1638N/+ and Apc^1638N/+/KRAS^V12G) were compared by ANOVA (3 ways) with a FDR (false discovery rate) set at 0.05 to select for genes with ≥2-fold differential expression. Notably, the P3 (CSCs from both genotypes) vs. Lin⁻ (bulk; both genotypes) and P3 vs. P1+P2 comparisons resulted in significant differences and in the definition of two lists of differentially expressed probe sets (n = 1062 [851 non-redundant, annotated genes] and 746 [602 non-redundant, annotated genes], respectively; Tables S3 and S4). In this fashion, we identified a list of 587 differentially expressed probe sets from the intersection of the Lin- vs. P3 and P1+P2 vs. P3. The identified probe sets correspond to 482 genes (non-redundant) (Table S5).

The expression profiles obtained from the CSC-enriched subpopulations (Lin⁻CD24^hiCD29⁺) of both genotypes (Apc^1638N/+ and Apc^1638N/+/KRAS^V12G) differ from those of more differentiated (Lin⁻CD24^medCD29⁺/CD24^loCD29⁺) and of the Lin⁻ bulk subpopulations as clearly visible in the hierarchical clustering (HCA) and principal component (PCA) analysis (Figure 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Expression profiling analysis of tumor cell subpopulations from Apc^1638N/+ and Apc^1638N/+/KRAS^V12G intestinal tumours.

(a.) Hierarchical clustering and (b.) Principal Components Analysis (PCA) (both implemented in Partek) of Lin⁻CD24^hiCD29⁺ (P3), Lin⁻CD24^medCD29⁺/Lin⁻CD24^loCD29⁺ (P1+P2, merged gate) and Lin⁻ (bulk) tumor cells from 5 individual mice of each genotype (Apc^1638N/+ and Apc^1638N/+/KRAS^V12G). For better visualization individual colours were used for each group and in b. ellipsoids were drawn around the three tumour populations.

https://doi.org/10.1371/journal.pone.0073872.g003

Notably, the Apc- and Apc/KRAS-mutant genotypes are not resolved by HCA and PCA, possibly due to the relatively limited number of Apc^1638N/+ and Apc^1638N/+/KRAS^V12G tumour samples (n = 5 for each group) employed for the comparative expression profiling analysis, insufficient to highlight the allegedly more subtle differences between benign and malignant CSCs.

Next, from the above lists of differentially expressed probes between the P3 and other tumour cell populations we validated the expression of a total of 35 genes by quantitative real-time PCR (Table S2). The selection of validated genes includes, apart from the top up- and down-regulated genes, also additional members of the canonical Wnt signal transduction pathway, and genes known to play relevant roles in cancer. Overall, the vast majority of the selected genes (33/35) were validated for their differential expression by qPCR (Figure S1).

Gene ontology analysis [20] of the intersection signature revealed a rather broad spectrum of cellular functions, structures and processes among the up-regulated genes including extracellular matrix, cell adhesion, organ development and morphogenesis (Table S6). In particular, analysis of the genes differentially expressed in the CD24^hiCD29⁺ cells from Apc^1638N/+ and Apc^1638N/+/KRAS^V12G tumours in comparison with bulk and more differentiated tumour cells, revealed several biological processes likely to play functional roles in cancer stemness. First, as also shown by the intracellular accumulation of active β-catenin, several targets and members of the Wnt signalling cascade are differentially expressed in the P3 signature including Lgr5, Mmp2 and Mmp7, Dkk2, Pla2g2a, Prox1, Sox17, T (brachyury), Wif1, and Fzd5. The presence of Lgr5 among the upregulated genes is of interest as it indicates that this well-known marker of normal cycling stem cells in the mouse intestine [21] might also represent a useful CSC marker in mouse intestinal tumours as recently demonstrated by lineage tracing [22]. Also, the transcription factor Prox1, a direct and dose-dependent target of the Wnt/β-catenin signalling pathway, was upregulated in the P3 population. Prox1 was previously shown to promote dysplasia in colonic adenomas and colorectal cancer progression [23]. However, we could not find any significant differences between Prox1 expression levels between Apc^1638N/+ and Apc^1638N/+/KRAS^V12G tumour cells both in the original microarray data and upon qPCR validation (Figure S1). This observation reflects a more general lack of significant differences between the P3 populations of Apc^1638N/+ and Apc^1638N/+/KRAS^V12G tumours, as also shown by hierarchical clustering and PCA analysis (Figure 3a and b, respectively). In our previous study [14], expression profiling of bulk tumours from Apc^1638N/+ and Apc^1638N/+/KRAS^V12G mice also did not resolve the two genotypes. In fact, only the relative number of tumour cells with nuclear β-catenin could significantly distinguish between Apc^1638N/+ from Apc^1638N/+/KRAS^V12G intestinal tumours [14].

Apart from Wnt, additional signalling pathways are represented by the differentially regulated genes as shown by the differential expression of Bmp7 and Bmper (Bmp signaling), Fgfbp1, Fgfrl1, and Etv5 (fibroblast growth factor receptors, binding proteins and transcription factors), and Igf1, Igfbp1, Igfbp5, and Igfbp7 (insulin-like growth factors and binding proteins).

Overall, these results show that CSCs from Apc- and Apc/KRAS-mutant tumours have distinct expression profiles from other tumour cell populations and are characterized by increased Wnt signalling activity, in agreement with their enhanced levels of intracellular β-catenin, together with other signalling pathays (Bmp, Igf), and by the expression of Paneth cell-specific genes.

Discussion

Mutations in the APC tumour suppressor gene represent the main initiating and rate-limiting event in the adenoma-carcinoma sequence leading to colon cancer in man [1]. Loss of APC function leads to the constitutive activation of the canonical Wnt/β-catenin signalling pathway known to play crucial roles in the regulation of self-renewal and differentiation in a broad spectrum of tissue-specific stem cell niches including the intestinal crypt and, accordingly, in the onset of many cancer types [24]. Constitutive Wnt signalling activation in the intestinal epithelium triggers adenoma formation and represents a necessary, though insufficient, step for malignant transformation. Somatic mutations in KRAS, BRAF, TP53, and SMAD4 usually underlie the further progression of the benign tumour into locally invasive adenocarcinoma and metastasis at distant organ sites [1]. Mutations in the endogenous mouse Apc gene also lead to intestinal polyp formation, though mainly located in the upper GI tract. Notably, mouse Apc-driven intestinal adenomas do not spontaneously accumulate Kras or Tp53 somatic mutations and, accordingly, very rarely progress to adenocarcinomas [12].

In our laboratory, we have generated the Apc^1638N/+ mouse model encoding for a hypomorphic Apc mutation resulting in few (5–6) upper GI adenomas, longer survival, and an increased chance of spontaneous malignant transformation though only in animals older than 1 year [10], [11], [12]. In contrast, compound heterozygous Apc^1638N/+/KRAS^V12G mice are characterized by increased tumour multiplicity (∼10 fold) and accelerated malignant progression with the majority of lesions being locally invasive adenocarcinomas [14]. Notably, oncogenic activation of KRAS/Kras on its own is insufficient to initiate intestinal tumorigenesis [25], if not with late onset and upon somatic inactivation of the Tp53 gene [13]. Hence, the dramatic phenotypic differences brought about by oncogenic activation of the KRAS gene in compound Apc^1638N/+/KRAS^V12G mice results from its synergistic action in promoting canonical Wnt signalling, as shown by the increase in TOP-Flash reporter activity and in the number of tumour cells earmarked by nuclear β-catenin accumulation [14].

The observed relative increase in the number of Apc^1638N/+/KRAS^V12G tumour cells earmarked by nuclear β-catenin is of importance in view of the so-called “β-catenin paradox” [7]: notwithstanding the fact that loss of APC function is common to all tumour cells and is predicted to result in the intracellular and nuclear accumulation of β-catenin, this is only observed in a minority of cancer cells undergoing an epithelial-to-mesenchymal transition (EMT) and non-randomly distributed at the invasive front of the tumour mass [4], [5] (see also Fig. 2a). This observation led to the hypothesis according to which nuclear β-catenin earmarks stem-like intestinal tumour cells with higher Wnt signalling activity capable of detaching from the primary mass and efficiently disseminate and home in distal, vital organs [6], [19], [26].

Here, we attempted the prospective purification of cancer stem cells (CSCs) from Apc^1638N/+/KRAS^V12G intestinal tumours. Notably, the presence of tumour-initiating cells, could not be demonstrated in Apc^1638N/+ tumours. According to the operational CSC definition, namely their capacity to form tumours upon limiting dilution transplantation into recipient mice, bona fide CSCs are either absent or extremely rare (>10⁻⁶) in the intestinal lesions characteristic of Apc^1638N/+ mice. In contrast, the malignant transformation and accelerated adenoma-carcinoma sequence brought about by the gut-specific expression of oncogenic KRAS result in the establishment of a subpopulation of tumour-initiating cells (estimated as 1 in approx. 7×10⁴ bulk tumour cells). The observed lack of grafting potential by cells derived from Apc^1638N/+ tumours is not surprising in view of previous studies showing that both established adenoma cell lines and primary human adenomas cannot form tumours even when injected at very high multiplicities (∼10⁷) [27], [28], [29], [30]. This is also in agreement with the recent study by C. Blanpain and colleagues showing that benign papillomas of the skin form secondary tumours at very low frequencies exclusively when transplanted together with stroma cells [31]. Of note, although the tumours here utilized for the transplantation and expression profiling analyses were not analysed by histopathology to maximize the number of cells available for these assays, it is safe to assume that the employed Apc^1638N/+ tumours were adenomas, whereas those from Apc^1638N/+/KRAS^V12G animals were adenocarcinomas. This assumption is based on our own 10-yrs long experience with the histo-pathological analysis of the intestinal tumours found in these inbred models, in agreement with the original descriptions [10], [11], [12], [14].

Overall, these results call for a revision of the current operational definition of CSCs, especially when dealing with benign lesions where the alleged subpopulation of stem-like tumour cells are unlikely to form tumours upon transplantation in immune-deficient animals. Alternatively, it is also plausible to think that benign tumours do not encompass a bona fide subpopulation of CSCs and that the true cancer stem cell properties are acquired only upon malignant transformation.

CSCs enrichment (approx. 20–25 fold) was observed in the CD24^hiCD29⁺ subpopulation (here also referred to as P3) of tumour cells from Apc^1638N/+/KRAS^V12G mice. Notably, P3 cells are characterized by increased intracellular β-catenin accumulation thus indicating enhanced Wnt signalling activity and validating previously proposed models [6], [7]. These results are in apparent contradiction with a recent publication showing that oncogenic Kras activation does not result in the expansion of the intestinal stem cell pool [25]. However, these observations were made in an otherwise wild type genetic background (e.g. Apc-proficient) rather than in the context of constitutive Wnt activation resulting from loss of Apc gene function. As previously shown, Apc and KRAS mutations appear to be synergistic in enhancing Wnt signalling levels [14] and as such result in the expansion of the CSC compartment (earmarked by β-catenin intracellular accumulation) as shown here.

The CSCs concept [3] postulates that tumour masses are organized in a hierarchical fashion with a subpopulation of stem-like tumour cells endowed with self-renewal and differentiation capacity which underlie tumour maintenance and growth. Subpopulations of cancer cells enriched in CSCs have been isolated from the majority of tumour types, including colon cancer [15], [16], [17], [18]. However, apart from their capacity to recapitulate tumorigenesis when transplanted at limiting dilutions in immune-deficient mice, CSCs need to be further characterized by additional assays to test their self-renewal and differentiation capacity. As shown here, the CSCs enriched in the P3 subpopulation of Apc^1638N/+/KRAS^V12G intestinal tumours are also characterized by self-renewal and differentiation: the relative size of the CSC subpopulation remains constant in serially transplanted tumours and the resulting malignant growths encompass several differentiated cell types (enterocytes, Goblet and Paneth cells) similar to the primary lesions. Notably, an apparent increase in the relative presence of intestinal differentiation lineages, namely Goblet, Paneth, and entero-endocrine cells, was noted among the serially transplanted intestinal tumours when compared with the primary Apc^1638N/+/KRAS^V12G lesions. However, this was not reflected by a corresponding increase in the size of the CD29/CD24 tumor subpopulations as observed by FACS (Figure S3). Also, as these transplantation assays were subcutaneous and not orthotopic, it is likely that fluctuations in the incidence of specific lineages results from the influence of the host microenvironment on the CSCs’ self-renewal and differentiation rates.

Expression profiling analysis of CD24^hiCD29⁺ cells from Apc^1638N/+ and Apc^1638N/+/KRAS^V12G tumours in comparison with bulk and more differentiated tumour cells revealed several biological processes likely to play functional roles in cancer stemness. Gene ontology (GO) analysis [20] of the P3 signature revealed a broad spectrum of biological and molecular functions (Table S6). Of note, protein (mainly growth factors) binding activity and extracellular matrix (ECM) structural constituents seems to characterize the expression profile of the P3 subpopulation of tumour cells. Whereas the former (growth factor binding) may represent the main mechanism underlying the activity of specific signalling pathways in CSCs, ECM components are likely to play a rate/limiting supportive role in the regulation of essential stem cell properties cells (eg, self-renewal, multipotency, proliferation, and differentiation). Stem cell niches are known to be regulated by direct and paracrine interactions with supporting cells and the extracellular matrix. Likewise, cancer-associated fibroblasts and other (e.g. inflammatory) cellular components of the tumour microenvironment modulate the CSC niche not only by releasing specific growth factors and cytokines but also through the synthesis of modified ECM components rarely found in normal tissues [32]. Nevertheless, one should not forget that the P3 subpopulation is enriched in stem-like tumour cells and by no means represents purified CSCs. Hence, bioinformatic analysis of the corresponding gene profiles might be confounded by the presence of different cellular components which do not make integral part of the CSC subpopulation and should be interpreted with caution. Recently, a signature that predicts disease relapse in colon cancer in man was developed based on mouse intestinal stem cells (ISC) expressing high levels of the Lgr5 and EphB2 genes [18]. Notably, comparison between the P3 and EphB2^hi and Lgr5^hi signatures revealed no major overlap (7 and 13 genes respectively, among which Lgr5 but not EphB2).

Of note, several Paneth cell specific genes including defensins (Defa5, Defa 21/22, Defa24), secretory phospholipase A2 (Pla2g2a), Fzd5 and matrix metallopeptidase 7 (Mmp7) are differentially expressed in the P3 population [33]. On one hand this is expected as the P3 (Lin⁻CD24^hiCD29⁺) FACS gate is likely to overlap with the CD24^hiSSC^hi sorting gate previously employed to enrich for Paneth cells from normal small intestinal epithelial cells [34], [35]. Nevertheless, this observation is of potential interest in view of the recently elucidated dual role of Paneth cells both as niche cells (supporting Lgr5⁺ stem cells) [34], [35], and as quiescent stem cells [35], [36]. Given the concomitant Lgr5 upregulation in P3 cells, it is plausible that tumour-specific Paneth cells constitute the niche for Lgr5⁺ CSCs in Apc^1638N/+/KRAS^V12G adenocarcinomas [37]. Alternatively, quiescent (Paneth cell precursor) tumour stem cells may provide a source of actively cycling Lgr5⁺ CSCs whose function is then further supported by more mature Paneth cells.

Notably, the Apc- and Apc/KRAS-mutant genotypes could not be resolved by HCA and PCA which does not allow us to pinpoint qualitative differences between benign (Apc^1638N/+) and more malignant (Apc^1638N/+/KRAS^V12G) CSCs. This is possibly due to the relatively limited number of Apc^1638N/+ vs. Apc^1638N/+/KRAS^V12G tumour samples (n = 5 for each group) employed for the comparative expression profiling analysis, insufficient to highlight the allegedly more subtle differences between the two CSCs genotypes. However, both in the HCA and PCA a trend is apparent where the P1+2 and P3 tumour cell subpopulations derived from Apc^1638N/+ and Apc^1638N/+/KRAS^V12G tumours appear to cluster away from each other. In the future, the use of more tumour samples and possibly the profiling of microdissected tumour cells earmarked by nuclear β-catenin accumulation from the two genotypes will shed more light on the more qualitative differences between benign vs. malignant neoplastic stem cells.

Overall, our results show that cancer stemness in Apc-driven intestinal tumorigenesis is underlined by an increased Wnt/β-catenin signalling activity that is apparently decreased in more differentiated tumour cells. As previously [7] and here reported (Figure 2), in both human (colon) and mouse (upper GI tract) APC-driven intestinal cancers cells with intracellular β-catenin accumulation are clustered around the invasive front, thus suggesting that factors secreted from the stromal microenvironment are likely to play a role in the enhancement of Wnt signalling activity and the maintenance of stem-like cancer cells. Also, the differential upregulation in the P3 subpopulation of markers of both cycling stem cells (Lgr5) and Paneth cells (defensins, matrix metalloproteases, and phospholipases) suggest that a niche similar to what described for the normal intestinal crypt of the mouse is present within adenocarcinomas. Further purification and (epi)genetic profiling of the individual components of the CSC niche are necessary to elucidate the cellular and molecular mechanisms underlying cancer stemness in the digestive tract.

Materials and Methods

Supporting Information