Colonic Hamartoma Development by Anomalous Duplication in Cdx2Knockout Mice¹

The Drosophila caudal gene, the first member of the caudal-type homeobox gene family to be described, is expressed in a gradient pattern highest at the posterior pole of the embryo (1). Whereas the caudal mutants lose posterior structures (2), the ectopic expression of caudal leads to a disruption of head segmentation and development (3). Several other caudal-like genes have been isolated in mammals. These include Cdx1, Cdx2, and Cdx4 in the mouse, Cdx3 in the hamster (4), and CDX1 (5) and CDX2/3 (6) in the human. Expression of Cdx2 in the mouse embryo begins at 3.5 dpc³ and is confined to the trophectoderm but absent from the inner cell mass (7). Subsequently, it is expressed in the extraembryonic ectoderm adjacent to the epiblast, sparing the more superficially placed polar and mural trophoblastic cells. Its expression continues in the fetal membranes, such as the chorion and allantoic bud, and in the spongiotrophoblasts at later stages. After 8.5 dpc, Cdx2 is expressed principally in the posterior gut, tail bud, and caudal neural tube (7).

The CDX2 protein has been characterized as a key transcriptional factor for intestinal development. It regulates some intestine-specific genes such as sucrase-isomaltase and carbonic anhydrase-1, which are indispensable for differentiation and homeostasis of the intestine (8). Cdx2 has been reported to play a role in the extracellular matrix-mediated intestinal cell differentiation (9). A gene knockout analysis showed that the homozygous null Cdx2 mutants die between 3.5 and 5.5 dpc, whereas the heterozygous mutant mice are fertile, with skeletal malformations, and form multiple adenomatous polyps in the colon (10). Although several reports suggest that expression of Cdx2 is down-regulated in human colon cancer cells (5, 11), it remains to be determined whether the down-regulation is responsible for colorectal carcinogenesis.

To investigate the expression and function of Cdx2, we have constructed ES cell lines in which the Cdx2 gene was inactivated by homologous recombination, with a lacZ cassette placed under the control of the Cdx2 promoter-enhancer. We have established Cdx2 mutant mice from these ES cells and analyzed Cdx2 expression and the precise mechanism of polyp formation.

We used a panel of DNA samples from interspecific backcross mice that has been characterized for over 1100 genetic markers throughout the genome (12). The informative restriction fragment length variants for Cdx2 were defined using MspI; the C3H allele showed a band of 1.8 kb, whereas that of Mus spretus was 2.3 kb.

A recombinant λ clone containing the mouse Cdx2 was isolated from a 129/Sv genomic library using a Cdx2 genomic DNA fragment probe [nucleotide positions 14–1153 according to James et al. (13)]. A 9.6-kb BglII-NotI fragment, containing a part of exon 1, and another downstream fragment of about 900 bp (NotI-KpnI) were subcloned in pUC19, respectively. The Cdx2 ATG translation initiation codon in exon 1 was deleted and replaced with a cassette containing the lacZ reporter and PGKneobpA (Ref. 14; Fig. 1 A).

Fifty μg of the targeting vector were linearized at the unique Sse8387I site and electroporated into ES cell line RW4 (Genome Systems, St. Louis, MO) as described (15). Of 200 G418-resistant colonies, 3 independent homologous recombinant candidates were identified by PCR and confirmed by Southern hybridization. The PCR primers were: Cdx2-F2, complementary to the neo gene (5′-CCG GTG GAT GTG GAA TGT GTG-3′); and Cdx2-R2, located 75 bp downstream from the 3′ KpnI site in Cdx2 (5′-CCC CTT CAG TAG ACC CTA AT-3′; Fig. 1). Amplifications were performed for 35 cycles (denaturation, 1 min at 94°C; hybridization, 1 min at 53°C; and elongation, 2 min at 72°C) and the homologous recombinants showed a 973-bp band (Fig. 1,C). Upon Southern analysis, their BamHI-digested genomic DNA showed two bands, 3.5 and 4.0 kb long, corresponding to the wild-type and knockout alleles, respectively (Fig. 1,B). The probe used was a 630-bp PstI-XbaI fragment containing the neo^r or a 1246-bp fragment downstream of the 3′ KpnI site (Fig. 1 A).

Two germ-line chimeras were obtained (Fig. 1,D). The wild-type allele was identified by PCR using two primers: Cdx2-end (5′-TTT GTC AGT CCT CCG CAG TA-3′); and Cdx2-R2, as described above, except that annealing was at 55°C (Fig. 1 C). Although most data presented here were collected in mice with the (129/Sv × C57BL/6)F₁ background, we also analyzed the mice backcrossed to C57BL/6 (N₃-N₄) and found the same phenotypes.

The day of the vaginal plug was considered as 0.5 dpc. To detect the β-galactosidase activity, the samples were incubated at 30°C for 30 min to overnight, depending on the developmental stage, in a staining solution (5 mm potassium ferricyanide, 5 mm potassium ferrocyanide, and 1 mg/ml X-gal in PBS).

Essentially, the same procedures were used as described previously (13), except that alkaline phosphatase-coupled anti-digoxigenin antibody (Boehringer Mannheim, Mannheim, Germany) was used. Whole-mount embryos were hybridized and sectioned.

Embryos were flushed from uteri at 3.5 dpc, placed in gelatin-coated wells in HEPES-buffered DMEM supplemented with 20% fetal bovine serum, and incubated at 37°C in 5% CO₂ in air. The polyp cells were cultured as described (16) with a minor modification.

Total RNA from the cultured cells were reverse-transcribed to cDNA with random hexamer primers, and each transcript was detected by PCR using oligonucleotide primers: Cdx2 mRNA, 5′-CGC CAC CAT GTA CGT GAG CT-3′ and 5′-GTC ACT GGG TGA CAG TGG AG-3′ (13); lacZ mRNA, 5′-GCG TTA CCC AAC TTA ATC G-3′ and 5′-TGT GAG CGA GTA ACA ACC-3′; and Ck8 mRNA, 5′-ACC ACC AGC GGC TAC TCA GGA GGA C-3′ and 5′-CTT GGA CAC GAC ATC AGA AGA CTC G-3′ (17). Thirty-five cycles were performed as described above, except that hybridization was at 53°C or 63°C.

Incorporations of BrdUrd were for 4 h for adults and 1 h for embryos, as described (18).

It has been reported that mouse Cdx2 maps on chromosome 5 in an interspecific backcross mapping using a panel of (C57BL/6J × M. spretus)F₁ × M. spretus mice (19). We independently mapped Cdx2 using different markers in another panel of 114 backcrossed mice from (C3H/HeJ-gld × M. spretus)F₁ × C3H/HeJ-gld (12). The Cdx2 gene was found tightly linked to the myocyte nuclear factor (Mnf; Ref. 20) and platelet-derived growth factor α (Pdgfa; Ref. 20) genes on distal chromosome 5 and not very far from the β-glucuronidase (Gus; Ref. 21) or erythropoetin (Epo; Ref. 22) gene. The linkage distances and the gene order were: Gus - 1.8 ± 1.2 cM - Epo - 1.8 ± 1.2 cM - Cdx2/Mnf/Pdgfa.⁴ These results are consistent with, and complementary to, those reported recently (19). No known mouse mutations were found in the vicinity of the Cdx2 locus that were likely to be Cdx2 mutant alleles.

One of the Cdx2 alleles was inactivated by homologous recombination in ES cells so that the translation initiation site was disrupted by a bacterial β-galactosidase gene (lacZ; Fig. 1; see “Materials and Methods”). Their germ-line chimeras were generated, and the resultant heterozygotes were viable and fertile (see below), whereas no homozygous pups were obtained from the heterozygous intercrosses. Although all embryos genotyped from 7.5 to 13.5 dpc were either wild type or heterozygous at the ratio of ∼1:2 (actual numbers, 17:38), the Mendelian ratio was obtained when the morulae and blastocysts were cultured in vitro for 7 days and genotyped (actual numbers, 4:8:4). However, the homozygous cells of the inner cell mass failed to proliferate, although the trophectoderm cells showed normal outgrowths (data not shown). As expected, an RT-PCR analysis of the embryonic cells thus cultured revealed that the homozygous mutants lacked the Cdx2 mRNA, whereas the wild-type and heterozygous embryos expressed the gene (Fig. 1 D). These results suggest that the homozygous embryos failed to develop after implantation.

It has been reported that Cdx2 heterozygous mice show an anterior homeotic shift of vertebrae and corresponding malformations of the ribs and develop multiple intestinal adenomatous polyps, particularly in the proximal colon (10). We also confirmed the skeletal anomaly as well as their runted growth (data not shown). In addition, examination of the gut of the adult Cdx2 heterozygotes revealed abnormal structures in the cecum and the proximal colon. Whereas the mucosa of the wild-type cecum and colon consisted of crypts without villi (Fig. 2,B), the heterozygous mutants contained numerous villiform structures in the cecum and proximal colon (Fig. 2, A and C). The densities of these structures were lower than that of the villi in the ileum, and they were often aligned in rows (Fig. 2,C). They were found abundantly in the body of the cecum, with decreasing densities and heights toward the apex (data not shown). Histologically, these villiform structures resembled the villi in the ileum but were extended from crypts of the colonic type, with their epithelium containing enterocytes and a few goblet cells (Fig. 2 D). Fine blood vessels were evident in the villiform structures, and the lacteals appeared to exist. Immunohistochemistry with antibodies against the small intestine-specific peptide transporter and sucrase-isomaltase, respectively, showed expression of these proteins in the structures (data not shown). Accordingly, they can be explained as the result of an incomplete anteriorization of the cecal and colonic epithelia caused by the haploinsufficiency of the Cdx2 gene expression (see below).

In addition to the villiform structures above, some polyps were found in the proximal colon. In adult mice, they had a sessile or pedunculate configuration, ranging from one-half to 8 mm in diameter. At 3 months of age, up to 10 polyps were found per mouse. Most of them were found on the mesenteric side near arterioles (Fig. 3,A). Histologically, the polyp mucosa was often divided into compartments by arborizing bands of smooth muscles and sometimes appeared to line within smooth muscle (Fig. 3,C). These are characteristics observed in hamartomatous polyps such as in PJS (23). However, there were some additional cytological differences as well. In the normal proximal colon, goblet cells are scattered throughout the colonic crypt. This is also the case for the colonic hamartomas of the PJS type (23). In contrast, polyps in the Cdx2 heterozygotes were of two cell types; one looked like colonocytes, and the other, pseudostratified epithelial cells (Fig. 3, G and B, respectively). But they were devoid of goblet cells or Paneth cells (compare with Fig. 3,H). Upon extensive examinations of the serial sections, these polyps in the adult mice did not contain any dysplastic changes such as prominent budding, crypt branching or resultant crowding/expansion of the epithelium, or cytological atypia as hyperchromatism or nuclear stratification. To determine the nature of the polyp cells further, we labeled nascent DNA with BrdUrd and stained the sections with a specific antibody. Whereas parts of the polyp epithelium incorporated some BrdUrd, as shown in Fig. 3,E, the center of the hamartoma incorporated BrdUrd relatively little (Fig. 3 F). These results are consistent with a benign course of the polyps. Although all heterozygotes carried polyps, the life spans of the heterozygotes were the same as their wild-type littermates, at least up to 1.5 years of age. The polyps in such mice did not show any signs of malignant changes (data not shown).

To monitor the Cdx2 expression in the heterozygotes, we stained their intestinal tissues with X-gal for the lacZ reporter activity. In the ileum, the villi were stained more intensely than the crypts, although the Paneth cells at the base were not stained (Fig. 4,A). In contrast, the hamartoma tissues were not stained with X-gal, despite the fact that the villiform structures nearby were stained (Fig. 4,B). To characterize this phenomenon further, we determined the expression of the Cdx2 and lacZ mRNAs by RT-PCR analyses. Because it was difficult to separate the polyp epithelial cells from the stromal cells, we cultured the polyp tissue in vitro and established a polyp epithelium-derived cell population, which aggregated as colonies (Fig. 4,C). Neither transcript of the lacZ nor Cdx2 gene was detected (Fig. 4,D), while the cells expressed the cytokeratin 8 (Krt2–8) mRNA, verifying their epithelial origin (Fig. 4,D). Thus, we conclude that both Cdx2 alleles were inactivated in the hamartomas. To investigate whether the tumor cells lost the wild type Cdx2 allele (i.e., LOH), a PCR genotyping was then carried out for each colony. In all colonies that showed biallelic inactivation of Cdx2, both the wild-type and knockout Cdx2 gene segments were detected (Fig. 4,E). These results ruled out the loss of the wild-type Cdx2 allele in the hamartoma. Accordingly, Cdx2 LOH does not appear to be the mechanism of its inactivation in the polyps developed in the Cdx2 heterozygous knockout mice (see “Discussion”). Additional in situ hybridization studies showed the lack of endogenous Cdx2 mRNA as well (see below and Fig. 5).

To investigate the polyp formation mechanism, we made a chronological analysis of the differentiating embryonic gut by serial sectioning. At 10.5 dpc, lacZ was expressed in a gradient pattern along the gastrointestinal tract (Fig. 5,A). A strong expression of lacZ was observed in the simple epithelium of the colorectal region (e in Fig. 5,B), whereas the proximal region including the stomach did not express lacZ. At 11.5 dpc, an “outpocketing” of the gut epithelium became visible on the opposite side from the cecal bud, i.e., on the mesenteric side (p in Fig. 5, C and D). Although the outpockets still expressed lacZ, an in situ hybridization analysis showed that the endogenous Cdx2 mRNA had already been lost (Fig. 5,E). By 12.0 dpc, the outpocketing epithelium lost the lacZ expression, except for its base connected to the gut epithelium, and formed a “duplicated structure” (p in Fig. 5, F–I), the epithelial cells of which proliferated faster than the normal gut epithelium, as demonstrated by more incorporation of BrdUrd (Fig. 5,J). By 15.5 dpc, the duplicated structure reached a diameter comparable with that of the proximal colon (Fig. 5,K), and it was still connected to the proximal colon (Fig. 5,L). The columnar epithelium of the duplicated structure was histologically indistinguishable from that of the proximal colon at this stage, except that it ceased to express both lacZ and the remaining Cdx2 allele. At around 16 dpc, the columnar epithelium of the colon starts to form the crypt structure by a rapid growth (24). Such normal development did not take place in the duplicated structure. At 17.5 dpc, its epithelium still remained lacZ negative and undifferentiated, and the duplicated structure was gradually outgrown by the normal colonic epithelium, although it was still connected to the colonic lumen (Fig. 5 M).

After birth, a dramatic change occurred in the epithelium lining the duplicated structure. A part of it differentiated into a pseudostratified epithelium with some keratinization, especially in the area exposed to the colonic lumen (s in Fig. 5,N), whereas the normal colonic epithelium showed fully developed colonic crypts. The relative size of the duplicated structure became even smaller because the normal colonic mucosa outgrew the duplicated structure. Mesenchymal walls were formed around the duplicated epithelium at 1 week of age (w in Fig. 5,O), and its morphology showed an appearance of a polyp at 2 weeks of age, with a smooth muscle layer developing in the mesenchymal walls (Fig. 5, P and Q). Accordingly, the original duplicated structure was contained by the submucosal mesenchyme, which superficially resembled an invasive neoplasm. This containment, however, was not complete, but an opening to the colonic lumen remained. Even in adults of 3 months of age, some hamartomas were found still connected to the colonic lumen.

Compared with an earlier report of Cdx2 knockout mice (10), we found some distinctly different gastrointestinal phenotypes, i.e., the heterozygotes developed cecal and colonic villiform structures and colonic hamartomas. It is likely that the villiform structures resulted from a homeotic anterior transformation of the gut mucosa due to the Cdx2 haploinsufficiency, because Cdx2 transcripts are abundant in the normal cecum and proximal colon (13), and expression of Cdx2 monitored by lacZ continued in the gut from the embryonic to adult life. On the other hand, the hamartomas appeared to be caused by the biallelic Cdx2 inactivation, because their epithelial cells did not express either lacZ or the remaining Cdx2 allele (Figs. 4 and 5).

In the Apc^Δ716 knockout mice, the crypt proliferative zone cells form outpockets after birth, beginning about 3 weeks of age (15, 18). In the Cdx2 knockout mutant, however, the outpocket formation took place around 11.5 dpc, when the gut epithelium differentiation was suppressed by the biallelic inactivation of Cdx2, which probably kept the cells proliferating faster than the surrounding epithelium. Forced expression of Cdx2, on the other hand, results in differentiation and slower growth rates in rat intestinal epithelial cell line IEC-6 (25).

The outpocketed epithelium developed thereafter into a partially duplicated gut structure (Fig. 5 I). However, this duplication is different from those often seen in teratological anomalies of the human gastrointestinal tracts. Such duplications are suggested to form by errors of recanalization after the stage of complete occlusion of the intestinal lumen (26) or by errors of separation between the notochord and embryonic endoderm (27). In the Cdx2 hamartomas, on the other hand, the initially duplicated gut epithelium did not go through any occluded stages, nor did it continue the normal development after 16 dpc; it was outgrown by the normal gut and contained as a hamartoma after birth. Hamartomas are nonneoplastic nodules composed of an abnormal overgrowth of indigenous cell types with evidence of an underlying developmental etiology (28). Most human intestinal hamartomas are found along the mesentery (23). Interestingly, the outpockets extruded into the mesenchyme along the mesenteric arteries. In the adult Cdx2 heterozygotes, the hamartomas were also found along the mesentery of the proximal colon.

It is worth noting that transgenic mice expressing Hox3.1 (HoxC8) from the Hox1.4 (HoxA4) cis-regulatory elements develop hamartomatous nodules in the stomach in addition to homeotic transformations of the skeletal system (29). Recently, several genes have been identified, the mutations of which are responsible for inherited hamartoma polyposis syndromes such as Cowden syndrome, Bannayan-Ruvalcaba-Riley syndrome, and PJS (30). Although Cowden syndrome and Bannayan-Ruvalcaba-Riley syndrome appear to be caused by mutations in PTEN/MMAC1/TEP1, a gene encoding phosphatidylinositol (3, 4, 5) triphosphate₃ phosphatase (31), PJS is caused by germ-line mutations in a protein serine-threonine kinase gene LKB1/STK11 (32). Thus far, there have been no hereditary hamartoma syndromes reported that carry mutations in the Hox or Cdx2 gene. It is interesting that Hox3.1 (HoxC8) is expressed in the mesenchyme of the stomach, whereas Cdx2 is expressed in the epithelium. The histological characteristics of the hamartomas in the Cdx2 heterozygotes are most similar to those of PJS, and PJS polyps can have features of diverticula, suggesting a common mode of formation. However, there were distinct differences as well, such as the lack of goblet cells and the presence of the pseudostratified epithelium that showed keratinization. Because the latter was seen only after birth, it is possible that the homeostatic changes upon milk digestion may be responsible for the change. Although some case studies report the squamous metaplasia in colonic adenomas (33, 34), no experimental analyses have been made, and the biological significance of the metaplasia in the Cdx2 hamartomas remains to be investigated.

Because the description of the colonic polyps in the earlier paper (10) is incomplete, it is difficult to judge whether they had similar hamartomas to ours. They did not give the long-term changes (i.e., prognosis) of their colonic polyps either. At least, the hamartomas of our Cdx2 knockout mice showed a benign course, and not a single tumor showed any malignant changes, even after a year and a half. It is interesting that CDX2/3 expression is down-regulated in some human colon cancer tissues; 10 of 12 colon carcinoma tissues examined had reduced levels of CDX2/3, as well as those of CDX1 (5). Such down-regulation was also observed in some cell lines obtained from colorectal carcinomas as well. However, it remains to be investigated further whether such Cdx2 down-regulation is responsible for malignancy in combination with other gene mutation(s). The possibility may be tested by crossing the Cdx2 mice with other tumorigenic mutant mice. In conclusion, our results suggest that the hamartoma development in the Cdx2 heterozygous mice is caused by the biallelic gene inactivation, whereas the cecal and colonic villiform structures are formed due to the haploinsufficiency of the homeobox gene Cdx2.

We thank P. Soriano for the PGKneo cassette, K. Miyamoto for antitransporter antibodies, R. G. Russell and E. Fearon for histopathological evaluations, and M. Capecchi for discussions.