Cdx2 determines the fate of postnatal intestinal endoderm

Cdx2 is expressed in intestinal endoderm posterior to the stomach throughout gestation and postnatally throughout life; it is never expressed in gastric tissues (Beck et al., 1995; Silberg et al., 2000). Previously, we disabled Cdx2 by homologous recombination (Chawengsaksophak et al., 1997). Homozygous Cdx2^–/– embryos failed to implant because trophoblast did not differentiate (Strumpf et al., 2005) but heterozygotes were born alive. These had polyps in the paracaecal region of the intestine that consisted of normally organised stomach mucosa arranged in an anatomically sequential manner from the stratified squamous epithelium characteristic of forestomach through to pyloric antral mucosa (Beck et al., 1999). The mesodermally expressed stomach-specific gene Barx1 was found in the mesenchyme immediately deep to the developing polyps and Sox2 was expressed in the overlying endoderm (Stringer et al., 2008). It was postulated that local sporadic haploinsufficiency of Cdx2 caused an anterior homeotic shift in which undifferentiated intestinal endoderm defaulted to a rostral gastric phenotype (forestomach); the remaining gastric tissue types (cardia, corpus and pylorus) were subsequently intercalated between forestomach and surrounding colon, in order to maintain the orderly succession of tissue types that are established during normal stomach development. Chimaeric studies using Cdx2^–/– cells in wild-type hosts confirmed these observations (Beck et al., 2003). Gao et al. (Gao et al., 2009) overcame the embryonic lethality of Cdx2 ablation by targeting its knock out specifically to embryonic endoderm early in development, thus inactivating gene expression throughout the intestine. They confirm that the intestinal lining reverted to a forestomach phenotype.

We have now conditionally knocked out the Cdx2 gene in the adult mouse intestine and have monitored the phenotype over 62 weeks to determine whether it has retained the ability to give rise to histologically and organotypically normal stomach mucosa, and also to investigate whether differentiation of the intestinal cell types that normally arise from the stem cells is at all possible in the absence of Cdx2 expression. Neither of these matters have previously been investigated. We describe a novel phenotype that differs fundamentally from that following knock out at conception or early in development. While the adult intestinal stem cells have the ability to express gastric-type genes, they are fixed into maintaining an intestinal architecture and have lost the positional information cues that determine morphogenic organisation of stomach mucosa. Consequently, a variety of gastric genes are expressed in an intestinal setting throughout the small intestine; in the paracaecal region, these heterotopias occasionally exhibit a tubulo-glandular picture that is suggestive of pyloric antral glands but do not express the gastric marker Sox2 immunocytochemically. In this respect they differ fundamentally from normal antral mucosa. Cdx2-negative crypts give rise to subepithelial cystic vesicles that are devoid of Paneth cells. Similar thin-walled Cdx2-null empty cysts are produced from stem cell-targeted Cdx2 inactivation in cultured crypts instead of the intestinal organoids produced from control crypts. The results of our in vivo and in vitro studies indicate that, in the absence of Cdx2 expression, the stem cells lose the ability to differentiate into any of the types that constitute normal intestinal epithelium. They express a variety of gastric-type genes but are unable to form normal gastric mucosa.

The Cdx2 conditional knock-out targeting vector was generated using a cloning strategy resulting in loxP sites flanking exon 2. This vector was electroporated into ES cells (129/Ola), positive clones identified and chimaeric mouse lines generated from which a backcross line was achieved, producing Cdx2^+/flox mice as described by T. Young (PhD Thesis, University of Utrecht, 2009).

Experiments were performed under UK Home Office licence authority. Cdx2^flox/flox mice on a mixed background were crossed to C57/BL6 AhCreER^T+/– (Kemp et al., 2004) transgenic mice generating Cdx2^flox/flox//AhCreER^T+/– mice and relevant controls. The Cdx2^flox/flox line was also crossed to the Lgr5-EGFP-ires-CreERT2 (Lgr5CreER) line (Barker et al., 2007) to generate Cdx2^flox/flox//Lgr5CreER^T+/–. Additionally, Cdx2^flox/flox//AhCreER^T+/– mice were crossed to C57/BL6 Cdx1^–/– mice generating Cdx1^–/–//Cdx2^flox/flox//AhCreER^T+/– mice and controls. To generate Cdx2^+/ko mice, the Cdx2^+/flox mice were crossed to the CMV-Cre line causing the floxed allele to be knocked-out in the germline. The Cdx2^+/– line was also crossed to Cdx2^+/flox mice and resultant Cdx2^–/flox mice were bred to the Sox2Cre line, generating Cdx2^–/ko//Sox2Cre embryos.

AhCreER^T, Lgr5CreER and Cdx1^– alleles were detected as described previously (Subramanian et al., 1995; Kemp et al., 2004; Barker et al., 2007). The Cdx2^flox allele was detected using primers shown in Fig. 1 and listed in supplementary material Table S2. PCR conditions were 35 cycles of 94°C for 30 seconds, 60°C for 30 seconds and 72°C for 30 seconds.

Cdx2^flox/flox//AhCreER^T+/– mice and controls were given a combined intraperitoneal injection of 1.6 mg tamoxifen (Sigma-Aldrich) and 1.6 mg β-naphthoflavone in corn oil, once daily for 4 days. Lower levels of recombination were achieved in one animal with a single intraperitoneal of the same concentration. The final injection was considered as day 0 after Cdx2 knockout. Cdx2^flox/flox//Lgr5CreER^T+/– mice and controls were given a single intraperitoneal injection of 2 mg tamoxifen in sunflower oil. Carrier oil only was administered to Cdx2^flox/flox controls to detect any spontaneous recombination of the Cdx2^flox allele. Mice were 6 weeks old when injected.

Intestines were divided into six sections and the majority fixed in methacarn as ‘swiss roll’ preparations. Sections between swiss rolls were formalin fixed and small regions were also transferred to RNAlater (Ambion) (Fig. 3F). Paraffin wax-embedded sections (5 μm) were used for histology.

Staining was performed using primary antibodies against Cdx2 (Beck et al., 1995), Cdx1 (Bonhomme et al., 2003), Sox2 (Neuromics; 1:250), villin (Santa Cruz; 1:750), Shh (Santa Cruz; 1:50), pepsinogen II/C (Abcam; 1:1500), Ki67 (Neomarkers; 1:200), Sox9 (Millipore; 1:1500), Myc (Abcam; 1:50), β-catenin (Sigma; 1:2000 upon formalin fixed tissue), lysozyme (Dako; 1:500 upon formalin fixed tissue) and H⁺/K⁺ ATPase (MBL; 1:1). Sections were developed using biotinylated secondary antibodies (Abcam) and a DAB peroxidise substrate kit (Vector Labs). Tissue was methacarn fixed unless otherwise stated. Sections apart from those stained for Cdx2, were counterstained with Haematoxylin. Cdx2 stained sections were counterstained with Alcian Blue.

Mice were given an intraperitoneal injection (1 ml/100 g body weight) of BrdU labelling reagent (5-bromo-2′-deoxy-uridine labelling and detection kit II, Roche) and culled 1 or 48 hours later. Labelling kit (Roche) for BrdU detection was used on formalin fixed 5 μm paraffin wax-embedded sections.

Tissues were fixed in 4% formaldehyde/2.5% glutaraldehyde in 0.1 M Sörensens phosphate buffer (pH 7.2) followed by buffered 1% osmium tetroxide. Thin sections (80 nm) were counterstained with 2% aqueous uranyl acetate followed by Reynolds’ lead citrate, and examined in a JEOL 1220 electron microscope.

RNA was extracted and DNAse treated using the RNeasy and RNase-free Dnase kits (Qiagen). For RT-PCR, 0.2 μg RNA was reverse transcribed and amplified with the primers listed in supplementary material Table S2. RNA-seq libraries were prepared from independent samples of three Cdx2^ko/ko//AhCreER^+/– (AH) and three control mice following Illumina’s protocol and sequenced on the Illumina Genome Analyzer II. After data normalization and quantification, genes with |log2 fold-change|>2 and adjusted P-value<5% were retained.

Small intestine was isolated from 9-month-old Cdx2^-/flox//Lgr5Cre^+/– mice. Crypts were isolated (Sato et al., 2009), and cultured in matrigel in medium containing Egf, noggin and RSpondin1 (ENR), which have previously been shown to allow long-term growth of intestinal crypts (Sato et al., 2009). After organoids had grown (see Fig. 6B), 100 nM 4 OH tamoxifen was added to the culture for 15 hours. Induced and non-induced (control) organoids were washed, dissociated and plated in culture medium in matrigel. Isolated cells from the non-induced organoids gave rise to branching organoids after 3 days that could be passaged indefinitely. From the dissociated 4OH tamoxifen-induced organoids, empty cysts appeared (30%). Five cystic and five normal-looking organoids were picked manually, pooled together and genotyped using primers P2 and P3 (Fig. 1; supplementary material Table S2). In three experiments from tamoxifen-treated cultures, Cdx2-null cysts and their control Cdx2-positive organoids were fixed at day 7, sectioned and stained for lysozyme (Sato et al., 2011).

We used Cre-LoxP technology to investigate Cdx2 loss of function in adult intestinal epithelium (Fig. 1). Tamoxifen/β-naphthoflavone given to Cdx2^flox/flox//AhCreER^+/– adult mice activated Cre recombinase. This generated Cdx2^ko/ko//AhCreER^+/– (AH) experimental mice in which Cdx2 knock out was targeted to all cell types lining the intestinal epithelium, except Paneth cells (Kemp et al., 2004). We also specifically targeted the Lgr5-expressing stem cells located in the intestinal crypt base by crossing the Lgr5-EGFP-ires-CreERT2 (Lgr5CreER) mouse line described by Barker et al. (Barker et al., 2007; Barker et al., 2009) to mice carrying our floxed Cdx2 allele. Drug-injected Cdx2^+/+//Cre-expressing mice were controls and Cdx2^flox/flox//Cre-expressing mice injected with corn oil alone indicated absence of spontaneous recombination of the Cdx2^flox allele. Animals were killed at various times to a maximum of 62 weeks after injection. Experimental animals at all stages remained healthy. Tissue was prepared as shown in Fig. 3F for histological and RNA analysis at various stages. We extracted RNA 4 and 62 weeks after drug injection for analysis by RT-PCR. We also prepared RNA from the jejunal region 1 week after drug administration for analysis of the total transcriptome. Finally, we investigated the ability of isolated crypt cells to develop into intestinal organoids in vitro following Cdx2 knock out. We tested the efficacy of the conditional knock out by two methods. First, we crossed Cdx2^flox/– mice with a line that expresses Cre under a Sox2 promoter (Sox2Cre) (Hayashi et al., 2002). The Cre was expressed exclusively in the inner cell mass, thus allowing development beyond the implantation block. The Cdx2^ko/–//Sox2Cre embryos were therefore Cdx2 null and proved to be identical to those we obtained previously by tetraploid ‘rescue’ of Cdx2^–/– embryos (Chawengsaksophak et al., 2004) (Fig. 1A). Second, we obtained germline Cdx2^+/ko mice from the floxed construct by crossing Cdx2^flox/+ mice with a CMV-Cre strain expressing Cre under a CMV promoter (Schwenk et al., 1995). Intestines of these mice, at the adult stage, contained organotypically normal gastric polyps in the paracaecal region that were identical to those we described when Cdx2 was inactivated by straightforward homologous recombination (Beck et al., 1999; Beck et al., 2003) (Fig. 1B).

One week after knock out, staining the Cdx2^ko/ko//AhCreER^+/– series (AH) with Cdx2 antibody indicated 75% knock out in the proximal intestine, maintaining a high level of ∼70% in the caecum (Fig. 2A) but falling to very low levels in the large intestine. Cdx2-negative cells were present over the whole extent of the majority of the villi but villous architecture was unchanged (n=5/5). The Cdx2^ko/ko//Lgr5CreER^+/– (LGR) series showed a 35% level of Cdx2 knock out; Cdx2-negative cells were largely confined to uniformly negative crypts without significant migration up the villi, even though a week had passed since their creation (n=3/3) (Fig. 2A). An additional specimen in the AH series given a lower dose of tamoxifen/β-naphthoflavone produced knockout levels similar to those seen in the LGR series; again, Cdx2-negative cells were demonstrable over the whole extent of relevant villi, thus excluding the possibility that the difference between the two groups resulted from a dose effect (Fig. 2A). It therefore appears that Cdx2-negative cells are at a selective disadvantage with respect to providing cells that migrate onto villi.

Pepsinogen C was expressed immunocytochemically in strips along histologically normal duodenal villi in cells in which Cdx2 had been inactivated (Fig. 2B), H⁺/K⁺ ATPase was seen in scattered Cdx2-negative crypts. Claudin 18 (Hewitt et al., 2006) (another stomach specific gene) was demonstrated on villi in cells that had lost staining for alkaline phosphatase (Fig. 2B). Genome wide analysis of the jejunal transcriptome in the AH series revealed upregulation of numerous genes principally or exclusively expressed in the stomach (supplementary material Table S1). These include Tff1 and Tff2 and their binding proteins, gastrokine 1, gastrokine 2 and gastrokine 3 (members of the gasdermin gene family), progastricsin (pepsinogen C), gastric intrinsic factor, the two subunits of H⁺/K⁺ ATPase, aquaporin 3, aquaporin 5, and claudin 18. Significantly, there was no evidence of Barx1 expression nor of any genes specifically associated with forestomach epithelium.

Four weeks after injection in the AH series, the mucosal surface of the gut became abnormally viscid and histological examination indicated a bias towards the production of secretory cells (n=4/4). AB/PAS staining at this and all subsequent stages showed excess of magenta mucous-secreting cells in the upper part of the crypts and towards the bases of the villi (Fig. 3B). Villi continued to be significantly clothed by Cdx2-negative cells but were now misshapen. We performed 48-hour BrdU-pulse studies, which showed that the majority of Cdx2-negative cells remaining on the villi were BrdU negative, indicating that they originated from crypts before labelling and suggesting failure of replacement from below due to the previously mentioned lack of ability of Cdx2-negative cells to replenish the villous surface. Confirming this, crypts containing Cdx2-negative cells had retained label while this was cleared from controls after 48 hours (Fig. 3A).

Paneth cells, stained for lysozyme, appeared abnormal and frequently moribund, an observation confirmed by electron microscopy (Fig. 3C). Cdx2-negative crypts, containing few Ki67-positive cells, were becoming sealed off into cystic vesicles. Ki67 counts in these crypts indicated a severe degree of hypoproliferation when compared with normal crypts in control animals (Fig. 3D). However, Ki67 staining showed an increase in the untargeted crypts of the AH experimental mice compared with controls, indicating that untargeted Cdx2-positive crypts were beginning a process of repopulating the surface epithelium (Fig. 3D). This was reflected in an increase in untargeted crypt length in experimental mice compared with controls (Fig. 3B).

In the LGR series, cells clothing the villi remained largely Cdx2 positive. A lower level of excess mucous secretion compared with the AH series suggests that the bias to secretory cell production in the AH series might include a direct response of the transit amplifying or post mitotic cells to Cdx2 inactivation. Cdx2-negative cells issuing from crypts remained deep in the mucosa, again indicating their relative inability to repopulate the villous surface (n=3/3) (Fig. 3E; LGR data). Pepsinogen C was demonstrable in cells of Cdx2-negative crypts adjacent to and between the Paneth cells of both series (Fig. 3E).

RT-PCR in all regions of the gut of the AH series demonstrated strong expression of pepsinogen C, H⁺/K⁺ ATPase and Muc6 with complete absence from controls. Barx1 was undetectable in all regions but strongly expressed in stomach controls (Fig. 3G).

Eight weeks after Cdx2 knock out, Cdx2-negative cells on the villous surfaces in the AH series had been almost entirely replaced by Cdx2-positive cells issuing from purified untargeted crypts (n=3/3). However, Cdx2-negative cells originating in targeted crypts persisted in great numbers as cystic vesicles within the mucosa (Fig. 4A,B). These were frequently filled with mucous detritus and when stained with AB/PAS, contained PAS-positive, Alcian Blue-negative material, in contrast to the Alcian Blue-positive mucous secretion of adjacent normal intestinal epithelium. The simple epithelium lining the vesicles stained positively for pepsinogen C, claudin 18 and occasionally for H⁺/K⁺ ATPase (Fig. 4A,B) and did not express villin or alkaline phosphatase. Cells expressing gastric genes thus appear to lose at least some intestine-specific markers. Sonic hedgehog is a ligand that is restricted in its gastrointestinal expression to the stomach endoderm (van den Brink et al., 2002); in both LGR and AH series, it was demonstrable in the secretions of Cdx2-negative intestinal crypts and in some Cdx2-negative cystic vesicles (Fig. 4A). Occasional Ki67 staining in vesicle cells indicated a retained capacity for proliferation. In the LGR series, GFP acts as a reporter for Lgr5 expression; GFP staining was sometimes also seen in cells lining the cystic vesicles, demonstrating the incorporation of cells bearing the Lgr5 stem cell marker into them (Fig. 4C).

Sox9 is a transcription factor that is regulated by the Wnt pathway. In the normal intestine, it is expressed in the lower region of the crypts. In our experimental material, Sox9 staining was seen in Cdx2-negative crypts as well as in the cystic vesicles, particularly in their basal regions (supplementary material Fig. S1). Staining for Myc produced similar results (supplementary material Fig. S1). β-Catenin, although expressed, as expected, in the nuclei of the Paneth cells of the crypts, is not nuclear in the cystic vesicles (Fig. 4D) and is therefore likely to be associated with low Wnt expression and a low proliferating ability. This is supported by the low Ki67 expression findings reported above. However, continued expression of Sox9 and Myc in the cysts indicates that replication potential is not entirely lost; taken together, these findings do not demonstrate changes in the Wnt pathway consequent upon the loss of Cdx2. Electron microscopy showed that cysts had a healthy, homogeneous lining of cells connected by junctional complexes. These exhibited sparse microvilli and contained electron-dense profiles compatible with mucous secretion (Fig. 4E). There was no histological differentiation into recognisable intestinal or gastric phenotypes. Scattered occasionally in the cystic lining were pale cells with few organelles (Fig. 4E).

The LGR phenotype was now similar to that in the AH line (n=3/3). This is because the knocked out cells that produced excessive mucous on the villi and in the transit amplifying region of the AH series had been replaced by Cdx2-positive cells and, in both series, the persisting Cdx2-negative stem cells were forming subsurface cystic vesicles (Fig. 4). A prominent feature of the crypts was the continuing number of dying Paneth cells.

In isolated areas of distal ileum, the progeny of Cdx2-negative crypts had replaced the villi. The mucosa had lost its villous morphology (Fig. 4F), contained pepsinogen C-secreting cells and presented a villin-negative branched tubulo-glandular appearance secreting Alcian Blue-negative, PAS-positive mucous reminiscent of a pyloric phenotype (Ichinose et al., 1991). Significantly, and in contrast to normal antral mucosa, however, these regions did not express immunocytochemically detectable amounts of Sox2, a gene normally associated with gastric mucosal morphogenesis (Que et al., 2007).

Beyond 19 weeks (up to a maximum of 62 weeks) after Cdx2 ablation, the majority of the surface epithelium continued to be Cdx2 positive but Cdx2-negative cells persisted as closed subsurface cystic vesicles (n=5/5). Stomach-specific genes continued to be expressed in both Cdx2-negative crypts and cysts. Sox 9 and Myc staining produced similar results to those described at 8 weeks after Cdx2 knock out. Isolated Cdx2-negative regions of the terminal ileum had lost villous morphology and were organised into antral-like branched tubular glands secreting PAS-positive, Alcian Blue-negative mucous but these regions were again Sox2 negative, unlike normal antral mucosa. In the caecum of three animals, a similar transformation resulted in development of polyps containing large irregular vesicles (Fig. 5). Pepsinogen-secreting cells were present in these regions (Fig. 5A, part c). However, in contrast to the organotypically normal regions of stomach mucosa following constitutive Cdx2 ablation, the endodermal cells of the polyps that developed following conditional knock out were Sox2 negative by immunocytochemical staining (Fig. 5B; shown schematically in Fig. 6F). Staining for β-catenin in these regions again showed membranous localisation and no evidence of nuclear staining (Fig. 5A, part d). RT-PCR of the paracaecal region-bearing polyps was also performed and, as at previous stages, failed to demonstrate the presence of Barx1 transcripts (Fig. 5C).

The LGR series phenotype was similar to that of the AH series but large polyps were not present in the caecum (n=3/3), most likely reflecting the lower level of knock out in these animals.

Both AH and LGR series at this stage contained a marked excess of Paneth cells demonstrable by both lysozyme and AB/PAS staining (Fig. 5D). It involved crypts containing Cdx2-positive cells and constituted compensation for Paneth cell death due to Cdx2 inactivation. On electron microscopy, the Paneth cells were healthy and contained an abnormally abundant number of normal looking secretory granules (Fig. 5E).

Cdx2-negative cells gradually downregulated levels of Cdx1, while neighbouring Cdx2-positive cells remained Cdx1 positive. Thus, Cdx1 expression was still strong in Cdx2-negative cells 1 week after Cdx2 knock out but, beginning proximally and gradually extending distally, was downregulated at later stages (Fig. 6A-D). We conclude that Cdx2 is required to maintain Cdx1 expression in intestinal epithelium. We also investigated the possibility that Cdx1 might perform a compensatory role by crossing to a Cdx1^–/– line, thus generating Cdx1^–/–//Cdx2^ko/ko//AhCreER^+/– (AH1//2) mice upon administration of drug (n=3). A qualitatively similar but greatly accelerated phenotype to that seen in the AH series resulted; PAS-positive cystic vesicle profiles appeared 1 week after knock out in the AH1//2 series, whereas this took 4 weeks in the AH series (Fig. 6E). Pepsinogen C expression was apparent in Cdx2-negative crypts and occasionally in runs along the villi. These results agree with those of Verzi et al. (Verzi et al., 2011) who describe an exaggerated phenotype and rapid death within 2 days of treatment in animals lacking both Cdx1 and Cdx2, in contrast to a more prolonged survival in mice lacking Cdx2 only.

We inactivated Cdx2 in intestinal crypts cultured in vitro. Intestinal crypts were isolated from a Cdx2^flox/–//Lgr5CreER^+/– mouse and grown in matrigel (Fig. 7B). 4-Hydroxy tamoxifen was added to the culture overnight. Induced and control organoids were then washed and dissociated cells were grown under conditions used for growing intestinal crypts in vitro (Sato et al., 2009). Control cultures produced rapidly growing and budding intestinal organoids previously described (Sato et al., 2009), but Cre-induced cultures consisted of a mixture of empty cysts (30%) and normal-looking intestinal organoids (Fig. 7D,E). The empty, thin-walled cysts devoid of Paneth cells appeared from day 3 of culture, persisted through the 7-day culture period (Fig. 7G-N) but did not survive passage. From tamoxifen-induced cultures, cystic and normal-looking organoids were picked manually at day 5 and genotyped, confirming that the empty cysts were Cdx2 negative, whereas the phenotypically normal intestinal organoids contained the wild-type Cdx2 allele (Fig. 7F,G). These observations were confirmed in three different experiments. We sectioned and stained cysts and normal organoids for the Paneth cell marker lysozyme. The Cdx2-null cysts did not express lysozyme, whereas the control organoids did (Fig. 7O,P). These in vitro observations support the conclusion that Cdx2-negative stem cells are unable to form mature intestinal endoderm and to produce the definitive intestinal stem cell niche that has previously been shown to depend upon the presence of Paneth cells (Sato et al., 2011).

We indicate a fundamental difference between effects of constitutive Cdx2 loss in intestinal endoderm and conditional knock out in the adult. When Cdx2 is inactivated prior to inception of the gut, intestinal endodermal differentiation reverts to a ‘default’ condition of stratified squamous epithelium characteristic of forestomach/oesophagus (Gao et al., 2009) followed by intercalation of remaining stomach regions between forestomach and surrounding normal intestine (Beck et al., 2003). By contrast, Cdx2 knockout in the adult never resulted in the development of forestomach or gastric units characteristic of corpus epithelium even after 12 months. Instead, gastric genes were expressed in Cdx2-negative crypts and submucous cysts, while the intestine largely retained its normal architecture. In small parts of the paracaecal region, a tubulo-glandular phenotype developed lacking expression of Cdx2 but, significantly, differing from normal antral mucosa by not expressing Sox2. The second important conclusion arising from the present study is that intestinal stem cells cannot differentiate into enterocytes, goblet cells or Paneth cells in the absence of Cdx2. The excess of dying Paneth cells seen 4 weeks after drug injection reflects senescence of those Paneth cells that were already present at the time of injection and the inability of Cdx2-null stem cells to replace them. At later stages, the abundant Paneth cells (with overabundant secretory granules per cell) seen in untargeted crypts reflect an attempt to re-establish and maintain intestinal homeostasis. The eventual similarity between the AH and LGR series indicates that changes in the potential of stem cells at the crypt base are responsible for the long-term phenotype produced by ablation of Cdx2 expression.

Cdx2-negative crypts in vivo seal off into closed subepithelial cystic vesicles expressing a variety of gastric genes in a cell-autonomous manner; these are long lived and appear to retain the capacity for very limited proliferation indicated by continued expression of Ki67, Sox9 and Myc. The normal distribution of β-catenin staining suggests that, according to this criterion, they are not neoplastic. In vitro, Cdx2-inactivated cultured intestinal crypts develop into similar empty cystic vesicles in contrast to long term organoids derived from wild-type counterparts; they do not contain lysozyme-positive cells, confirming the in vivo conclusion that Cdx2-negative intestinal stem cells cannot generate Paneth cells, thereby disrupting the intestinal stem cell niche. Cdx2-null cultured cysts do not persist long enough to express intestinal or stomach markers.

We suggest that, following Cdx2 inactivation experiments, crypts undergo purification (Lopez-Garcia et al., 2010; Snippert et al., 2010). Those that had retained Cdx2-positive stem cells became hypertrophic (as evidenced by proliferative cell counts) and some (perhaps all) subsequently became entirely Cdx2 positive, serving to replace the Cdx2-negative surface epithelium. This, together with the lack of acute digestive symptoms indicates the homeostatic capacity of the intestine.

Verzi et al. (Verzi et al., 2011) conditionally inactivated Cdx2 in mice using tamoxifen-responsive Cre under a villin promoter; their animals died within 3 weeks of treatment, making it impossible to trace the long-term fate of disabled cells. In our experiments sufficient normal (untargeted) enterocytes remained on the villi to maintain intestinal function until the intestine regained a clothing of Cdx2-positive cells from purified untargeted Cdx2 positive crypts, allowing us to follow the longer term fate of Cdx2-negative crypt areas. Verzi et al. (Verzi et al., 2011) found that Cdx2 ablation produced different effects, depending upon the state of differentiation of the maturing intestinal cell. This may explain some differences between the mucous secretion seen in the AH versus the LGR series 4 weeks after injection. In the AH series, both differentiated cells on the villi and partially differentiated transit amplifying cells are targeted for Cdx2 knock out, whereas in the LGR series only proliferating stem cells are targeted and these remain deep in the mucosa.

In normal gut development prior to histodifferentiation, the stomach region is characterised by expression of Sox2 in endoderm and Barx1 in underlying mesoderm (Kim et al., 2005), while the nascent intestinal endoderm expresses Cdx2 and does not express mesodermal Barx1. Our previous observations, confirmed here, indicated that constitutive deletion of Cdx2 resulted in the formation of histologically normal stomach mucosa expressing Sox2 in regions of Cdx2 deficiency (Fig. 1, Fig. 5B, part ii). We now demonstrate that following conditional Cdx2 ablation in adult mice, the intestinal endodermal cells fail to differentiate into intestinal cell types and begin to express gastric genes in a cell-autonomous manner. However, they are not organised into normal stomach mucosa and fail to express Sox2 immunocytochemically. In Fig. 6F, we illustrate a mechanistic basis for this change in stem cell potential based upon the inability of adult intestinal mesoderm to express the homeobox transcription factor Barx1. This gene has been shown by Kim et al. (Kim et al., 2005; Kim et al., 2007) to be required for normal spatially ordered stomach mucosal development. Our previous studies (Stringer et al., 2008) demonstrated that, during embryonic development, Barx1 is active in the intestinal mesoderm underlying gastric heterotopias in constitutively Cdx2-deficient areas of the midgut. Crucially, our present experiments show that Barx1 is not switched on in the mesoderm of the adult intestine in which Cdx2 is conditionally ablated. We suggest that the maturing intestinal mesoderm loses its earlier ability to express Barx1 and possibly other mesodermal signalling molecules, and is no longer able to impart the positional information to the adjacent endoderm that is necessary for organisation into morphologically normal stomach mucosa. The overall histology of the conditionally Cdx2-deficient adult intestine remains essentially intestinal in type, thus allowing the regeneration of intestinal villi from untargeted crypts that would not be possible in the presence of mesodermal Barx1. Our results following staining for Sox9, β-catenin and Myc suggest that changes in the Wnt pathway are not centrally involved.

The intestine exhibits a changing phenotype along its length and Cdx2 ablation therefore occurs within an altering context resulting in a varying phenotype proximodistally. Thus, throughout the small intestine, Cdx2 loss results in expression of gastric genes and development of subepithelial cystic vesicles but in the paracaecal area, scattered tubulo-glandular regions develop. The partial redundancy demonstrated between Cdx1 and Cdx2, and the dependence of Cdx1 expression upon Cdx2 may contribute to the region-specific phenotypes seen, as Cdx1 expression is highest in the distal gut whereas Cdx2 peaks in the paracaecal region. It is likely that the ‘rescue’ action of Cdx1 will be weakest in the paracaecal region where Cdx2 levels are normally high and so Cdx1 compensation is difficult. This is consistent with partial overlap of Cdx gene function that has been demonstrated elsewhere (van den Akker et al., 2002).

In summary, we have shown that conditional knock out of Cdx2 in adult mice and in cultured crypts prevents differentiation of stem cells into the various normal intestinal lineages. In vivo, Cdx2-negative crypt areas retain limited plasticity, enabling them to express gastric genes in a cell-autonomous manner but they are unable to default to morphologically normal forestomach and corpus, associated with the fact that the underlying mesenchyme is no longer capable of imparting appropriate positional information. Intestinal function is maintained by compensatory growth of unaffected crypts, which we suggest is permissible only on a Barx1-negative mesoderm. In effect, we indicate that the adult intestinal mesoderm becomes permanently programmed to provide support for an intestinal-type phenotype. Our observations provide new insights concerning the developmental potential and plasticity of the stem cell niche in the adult mammalian intestine.

We thank J. Edwards, E. Martin, H. Begthel, J. Korving, K. Beck-Sander and M. van den Born for help with histology; N. Allcock for electron microscopy; and Biomedical services, University of Leicester for technical expertise. T. Young contributed to the generation of the Cdx2 conditional knockout strains.