Combined Mutation of Apc, Kras, and Tgfbr2 Effectively Drives Metastasis of Intestinal Cancer Free

Colorectal cancer is a leading cause of cancer-related death worldwide (1, 2), and the 5-year survival rate drops significantly to about 14% for patients with metastasis (3). It is therefore extremely important to clarify the biological mechanisms of malignant progression to identify novel therapeutic target pathways. The gradual accumulation of genetic alterations in driver genes is known to cause development and malignant progression of colorectal cancer and is an established concept of multistep tumorigenesis (4, 5). Recent genome-wide analyses have confirmed the presence of frequently mutated driver genes in human colorectal cancer (6, 7). Using organoid culture systems, it has been shown that the introduction of genetic alterations in driver genes, APC, KRAS, SMAD4, TP53, and PIK3A, in intestinal epithelial cells induces tumorigenesis (8–10).

These driver mutations are classified as cancer signaling pathways (11), and the possible mechanisms of each mutation in tumorigenesis have been well studied. For example, loss of APC results in Wnt signaling activation, leading to the acquisition of stemness (12). KRAS plays a major role in tumorigenesis through the activation of RAF–MAPK and PI3K pathways (13). In contrast, TGFβ signaling promotes differentiation of epithelial cells, thus playing a tumor suppressor role in colorectal cancer (14). Furthermore, FBXW7 is a component of the ubiquitin ligase complex that degrades proto-oncogene products, thereby functioning as a tumor suppressor (15), and the disruption of Fbxw7 promotes intestinal tumorigenesis (16, 17). Recent results indicate that mutant p53 alters the gene expression globally by gain-of-function mechanism, which promotes tumorigenesis (18, 19).

Moreover, the effects of simultaneous mutations in combination on colorectal cancer have also been genetically studied. It has been shown that the suppression of TGFβ pathway or expression of mutant p53 in Apc^Δ716 mice induces the submucosal invasion of intestinal tumors (20–22). Moreover, the Kras mutation in addition to Apc and Trp53 mutation causes invasion and metastasis (23, 24). However, despite these findings from genetic studies, the big picture regarding how specific combinations of driver mutations promote each step of malignant progression in the primary tumors and metastatic foci remains unclear.

In this study, we generated mouse models carrying the Apc^Δ716 mutation together with conditional mutant alleles of Kras^G12D, Trp53^R270H, Tgfbr2^−/−, or Fbxw7^−/− in various combinations and examined their intestinal tumor phenotypes. Using these mouse models and allografts of tumor-derived organoids, we found the specific combinations of the driver mutations that were responsible for submucosal invasion, epithelial–mesenchymal transition (EMT)-like morphology, intravasation, and metastasis. Particularly, the combinations including Apc^Δ716 Kras^G12D Tgfbr2^−/− mutations caused efficient liver metastasis. Moreover, RNA sequencing (RNA-Seq) analysis corroborated the distinct expression profiles for the specific mutational combinations that are associated with distinct malignant phenotypes in both mice and human. Accordingly, targeting the pathways that regulate specific progression processes will be effective for the prevention of colorectal cancer metastasis.

Apc^Δ716, Fbxw7^flox/flox, and villin-CreER mice were described previously (25–27). Tgfbr2^flox/flox, Trp53^LSL·R270H, and Kras^LSL•G12D mice were obtained from the Mouse Repository (NCI-Frederick, Frederick, MD; refs. 28–30). Primer sequences for genotyping are previously reported except for Trp53 forward primer, 5′-CCTGCCAGCTCCGAAAGATT-3′. The genetic background of all strains used in this study is C57BL/6. All mice were treated with tamoxifen intraperitoneally at 4 mg/mouse once a week from 8 weeks of age for 4 weeks. For the survival curve analysis, mice were observed until 270 days of age, and the mice were euthanized when they showed a moribund phenotype. The total numbers of polyps in both the small intestine and colon were scored at 13–16 weeks of age (n = 3–5 for each genotype). NOD/Shi-scid Il2rg^−/− mice (NSG mice) and C57BL/6 mice were purchased (CIEA). All animal experiments were performed with the protocol approved by the Committee on Animal Experimentation of Kanazawa University.

The primary intestinal tumors and liver metastasized tumors were fixed in 4% paraformaldehyde, paraffin-embedded, and sectioned at 4-μm thickness. The sections were stained with hematoxylin and eosin (H&E) or Masson's trichrome stain. Antibodies against E-cadherin (R&D Systems), αSMA (Sigma), F4/80 (Serotec), CD3ϵ (Santa Cruz Biotechnology), Ki67 (Life Technology), Lyve-1 (Acris Antibodies GmbH), vWF (DakoCytomation, Denmark), Snail2 (Bioss), CD4, CD8, and CD45R (BD Pharmingen) were used as the primary antibody. Staining signals were visualized using the Vectastain Elite Kit (Vector Laboratories). For fluorescent IHC, Alexa Fluor 594- or Alexa Fluor 488–conjugated antibodies (Molecular Probes) were used as the secondary antibody. The histologic classification of the tumors was performed according to the Japanese Classification of Colorectal Carcinoma (31).

The mean Ki67-labeling indices were calculated as the number of Ki67-positive cells per total number of tumor cells by counting 5 independent microscopic fields (×200) for three independent tumors per mouse (n = 3) or 3–8 independent foci of liver metastasized tumors (NSG mice n = 3 for each, C57BL/6 mice n = 1–5).

The immunostaining-positive areas for F4/80, αSMA and CD3ϵ in the microscopic fields were measured using the Hybrid cell count software program of All-In-One microscope (Keyence). For the primary tumors, the ratio of the αSMA immunostaining-positive area was calculated from 3–5 polyps per mouse (n = 3). For the liver metastasized tumors, the ratios of αSMA, F4/80, and CD3ϵ immunostaining-positive areas were calculated from 3 to 12 foci per mouse (NSG mice n = 3, C57BL/6 mice n = 1–5).

The total numbers of invading polyps in submucosa and total polyps were counted on the H&E-staining sections of whole intestine and the invasion efficiency was calculated (n = 4–5). Solitary tumor cells were detected by IHC for E-cadherin, and the numbers of solitary cells were counted on the sections (n = 4–18) while the mean numbers were calculated per 1-mm² area using the Hybrid cell count software program (Keyence).

The organoid cultures were prepared from small intestinal tumors, as described previously (22). Authentication of the organoid cells was done by genotyping for the respective combined mutations. Cryopreserved stocks were established after testing, and experiments were performed within 5–10 passages following thaw of frozen stocks. Mycoplasma testing was done by indirect immunofluorescence test and nested PCR. The cultured organoids were mechanically dissociated, and 3 × 10⁵ organoid cells were injected with Matrigel into the spleen of NSG mice (n = 3–5) or C57BL/6 mice (n = 4–18). At four weeks after transplantation, the liver and lung were examined histologically. The multiplicity of metastatic foci in the liver was scored by the measurement of the metastasized tumor areas on H&E sections and the calculated percentages per total liver (NSG, n = 3–5; C57BL/6, n = 5–12).

Total RNA was extracted from organoid cells using an RNeasy plus Micro kit (Qiagen). RNA-Seq Libraries were prepared using a SureSelect Strand Specific RNA Reagent Kit (Agilent Technologies) according to the manufacturer's protocol. Thirty-six bp of single-end sequencing was performed using an Illumina HiSeq3000 (Illumina). Obtained reads were aligned to UCSC mm10 using tophat2. For the expression estimation, we counted the number of reads, which uniquely mapped to the exon of RefSeq transcripts, and calculated the reads per kilo-base of exon model per million mapped reads (RPKM) as the expression value. The sequencing data were deposited in the DNA Data Bank of Japan (DDBJ, accession #DRA005647). Clean reads that average quality scores for all libraries were more than Q30 were aligned to the mouse reference (Ensembl 85) using TopHat2 (32).

Gene expression quantification was performed using Cufflinks (33). Differential expression analysis between samples with replicates was performed using Cuffdiff with the cut-off set at P < 0.01 and ≥1.5-fold change. Hierarchical clustering for selected genes were analyzed with MeV (http://mev.tm4.org) using Euclidean distance and complete linkage method. Functional annotation for module members were performed by DAVID (34), and relevant gene ontology (GO) terms were selected with a cutoff of P < 0.01.

We used publicly available, level 3 data of The Cancer Genome Atlas (TCGA) in the current study. Clinical information and mRNA expression data obtained by RNA-seq of the TCGA samples were downloaded from the USCS Cancer Browser (https://genome-cancer.ucsc.edu). The patients' clinical data and molecular subtype status of the tumors were referred from the TCGA database (6). Among colorectal cancer samples, 343 tumor samples that included RNAseq data and metastasis information (M0, n = 285; M1, n = 58) were analyzed for the current study.

The RPKM values from RNA-seq of intestinal tumor organoids were subjected to quantile normalization and 5,000 genes with the highest variances were input to principal component analysis (PCA). PCA was performed using the prcomp function implemented in the R software (https://www.r-project.org).

The data were analyzed using an unpaired t test and are presented as the means ± SD. A value of P < 0.05 was considered as statistically significant. The significance of the differences of RNA expressions according to same or more than two groups was calculated by the Wilcoxon rank-sum test. For RNA-seq data, statistical analyses and data presentations were performed in R language 3.1.3 (http://www.r-project.org)

We crossed Apc^Δ716 (A), Kras^+/LSL-G12D (K), Tgfbr2^flox/flox (T), Trp53^+/LSL-R270H (P), Fbxw7^flox/flox (F), and villin-CreER mice to generate compound mutant mice carrying driver mutations in different combinations (Fig. 1A and B). These are frequently mutated genes in human colorectal cancer (6, 7). Of note, missense-type mutations of p53 are found in about 50% of human colorectal cancer (6), and we recently showed that mutant p53^R270H induces the submucosal invasion of intestinal tumors (22). Accordingly, we used Trp53^R270H mice in this study to examine the gain-of-function mechanism of mutant p53. In all mice, intestinal tumors were initiated by somatic loss of wild-type Apc, that causes Wnt signaling activation (25). The treatment of compound mice with tamoxifen induced genetic alterations in conditional alleles in an intestinal epithelia-specific manner, and we confirmed the tamoxifen-induced recombination in the intestinal tumor cells (Supplementary Fig. S1).

Although none of compound mice developed spontaneous metastasis (see below), the mean life spans were decreased depending on the increased numbers of driver mutations (Fig. 1C), suggesting that the accumulation of driver mutations affects systemic conditions through secreted factors.

We examined the number and size of intestinal polyps in A, AK, AKP, AKT, ATP, AKTP, and AKTPF mice. Intestinal tumors of other genotype mice with the Fbxw7 mutation were not polypotic and thus uncountable (see below). Notably, the polyp number increased significantly in mice that carried the AK combination, that is, AK, AKP, AKT, AKTP, and AKTPF, in comparison to that in A and ATP mice, indicating that Kras activation increases the multiplicity of tumors (Fig. 1D and E). Interestingly, the ratio of small polyps (<1 mm) dramatically increased in mice carrying AK combination (Fig. 1F), and most small adenomas were located in the upper part of the mucosa without connecting to the crypt bottom (Supplementary Fig. S2). Therefore, these small adenomas in Kras^G12D mice may be eliminated by shedding to the lumen, which may explain the increase in the ratio of small polyps. We previously showed that Apc^Δ716 mouse tumor cells cannot survive without generation of COX-2–expressing microenvironment (35, 36). Thus, it is possible that Kras activation contribute to the survival of tumor cells through the generation of such microenvironment.

Histologic types of all genotype mouse tumors are shown in Supplementary Table S1. Notably, mice carrying AP or AT mutations, that is, AKP, ATP, AKT, ATPF, AKTF, AKTP, and AKTPF mice, developed adenocarcinomas with invasion to submucosa or deeper, while invasive tumors were not found in A, AK, and AKF mice (Fig. 2A and B). These results are consistent with the previous findings that TGFβ signaling suppression or mutant p53^R270H expression induces submucosal invasion intestinal tumors (20–22).

Notably, the ratio of stromal volume was significantly increased in the invaded submucosal area with the increased number of αSMA-expressing myofibroblasts in comparison with those in mucosal area regardless of mutation types (Fig. 2A and C). Increased collagen fiber deposition detected by Masson's trichrome staining was consistently found in the submucosal invasion area (Supplementary Fig. S3A). These results indicate that desmoplastic reaction is induced in the submucosa, but not in the mucosa. It has been shown that such fibrotic reactive stroma plays a role in malignant progression through the secretion of various factors including TGFβ and Wnt ligands (37). Tumor cells may therefore acquire malignant characteristics in the submucosa by exposure to desmoplastic microenvironment.

Interestingly, the proliferation rates of tumor cells were significantly lower in the submucosal area than in mucosal area, possibly due to the effect of desmoplastic stroma (Fig. 2D; Supplementary Fig. S3B). We noted no significant difference in the proliferation rates in the mucosa of all genotypes compared with that in simple Apc^Δ716 mouse adenomas, indicating that any of the driver mutations does not accelerate the proliferation in primary tumors.

EMT is a hallmark of cancer malignancy (38). We found tumor cell clusters that had lost their glandular architecture and solitary tumor cells in the submucosa, which showed the nuclear accumulation of the EMT marker Snail2 (Fig. 3A; Supplementary Fig. S3C). Of note, these invasive tumor cells still expressed E-cadherin, so we classified them as “EMT-like.” Importantly, the number of solitary cells in the submucosa was markedly high in AKP, AKT, AKTP, AKTF, and AKTPF mouse tumors that carried AKP or AKT combination but rarely found in ATP and ATPF mice (Fig. 3A and B, right). These results indicate that Kras activation together with mutant p53^R270H expression or TGFβ suppression can induce EMT-like morphology when tumor cells are exposed to a desmoplastic microenvironment in submucosa. In contrast, solitary tumor cells were found in the mucosal area of AKF, AKTF, and AKTPF mice that carried AKF combination (Fig. 3A and B, left), suggesting that the combination of Kras and Fbxw7 mutations induces EMT-like morphology without the support of a desmoplastic microenvironment, although AKF combination is not sufficient for invasion.

Importantly, intravasation into the Lyve-1–expressing lymph vessels was found in AKT, AKTP, and AKTPF mouse tumors but rarely found in other genotype mice (Fig. 3C). Furthermore, intravasation to the blood vessels was also found in the AKTPF mice. The AKT combination may therefore induce intravasation with higher efficiency than the AKP or AKF combination.

It has been shown that FBXW7 plays a role in the differentiation of intestinal epithelia (16, 17), suggesting that Fbxw7 mutation may alter the tumor morphology. Interestingly, the histologic type of AKF, AKTF, and AKTPF genotype mouse tumors that included the AKF combination mutations was villous-type, while other genotype mouse tumors were glandular-type (Supplementary Fig. S4A). Because it was technically difficult to assess the tumor multiplicity of AKF common mice, the tumor distribution was measured by the length ratio on histology sections (Supplementary Fig. S4B). Notably, the tumor distribution was significantly increased in the AKF and AKTF mice compared with other genotype mice that did not carry the AKF combination (Supplementary Fig. S4B and S4C). These results indicate that the combination of Kras activation and Fbxw7 disruption accelerates tumor growth in addition to inducing EMT-like morphology, and these properties are independent of the invasion ability.

We detected no spontaneous metastasis in any genotype mice examined, indicating that the accumulation of these driver mutations is not sufficient for metastasis. We therefore established organoids from the tumors of A, AK, AP, AT, ATP, ATPF, AKP, AKT, AKTF, AKTP, and AKTPF mice and transplanted them to the spleens of immunodeficient NSG mice as well as immunocompetent C57BL/6 mice to examine their ability of metastasis to the liver (Fig. 4A). At 4 weeks after spleen injection to NSG mice, tumor organoids with ≥3 mutations except for ATP (i.e., AKP, AKT, ATPF, AKTF, AKTP, and AKTPF) developed metastatic tumors with 100% incidence. Among them, the multiplicity of metastasis was significantly higher in the AKT common organoids, that is, AKT, AKTF, AKTP, and AKTPF, than in those of AKP or ATPF (Fig. 4B). The organoids carrying AKT combination efficiently formed metastatic tumors also in C57BL/6 mice, while the incidences of metastasis of AKP and ATPF in C57BL/6 mice were 10% and 0%, respectively (Fig. 4A, C, and D). These results indicate that Wnt activation, Kras activation, and TGFβ suppression by AKT mutations are the core combination for efficient liver metastasis. Notably, additional mutations in Trp53 and Fbxw7 to AKT, that is, AKTP and AKTPF, induced the development of lung metastasis after the spleen injection of organoids, which was not found in other genotypes (Fig. 4A). Furthermore, AKTPF organoids showed the highest mean multiplicity of liver metastasis among the AKT common organoids.

Histologically, metastasized tumors were diagnosed as moderately to poorly differentiated adenocarcinomas with desmoplastic reaction in the stroma consisting of αSMA-positive myofibroblasts (Fig. 4E and F). The desmoplastic reaction was induced in all invaded tumors, suggesting that the AKP or AKT combination is sufficient to induce fibrotic responses in metastatic lesions. In contrast to the primary tumors, the mean proliferation rate of tumor cells increased gradually from AKP to AKT, AKTF, AKTP, and AKTPF (Fig. 4G; Supplementary Fig. S5). Of note, macrophages, CD4⁺ and CD8⁺ T cells, and B cells were infiltrated into the metastasized tumors at the similar levels, regardless of the tumor genotypes, whereas T-cell infiltration was rarely found in the primary tumors of the same genotype mice (Fig. 4H and I; Supplementary Fig. S5). Accordingly, the interaction with microenvironment in the metastasized tumors contributes to the induction of the host immune response to tumor cells, although the underlying mechanism remains to be investigated.

To examine whether genotype-related malignant phenotypes are associated with distinct expression profiles, we next performed RNA-seq of the tumor organoids for the ATP, AKP, AKT, AKTP, and AKTPF genotypes that showed submucosal invasion in the primary tumors. Notably, a principal component analysis (PCA) indicated that expression profiles of tumors carrying the Kras^G12D mutation (AK common, i.e., AKP, AKT, AKTP, and AKTPF) were divergent from that of ATP (Fig. 5A, left), which reflected the fact that Kras activation is required for metastasis. Furthermore, the expression profiles of AKT common (AKT, AKTP, and AKTPF) organoids were distinct from that of AKP (Fig. 5A, center), which reflected the high metastatic ability of AKT common organoids compared with AKP. We also found the evolutional expression changes from AKT to AKTPF via AKTP (Fig. 5A, right). These results suggest that expression changes induced by AKT mutations cause efficient metastasis, and additional mutations in Trp53 and Fbxw7 accelerate malignancy, which is consistent with the results of phenotype analyses (Fig. 5B).

We next extracted 363 differentially expressed genes (DEG) in AKT common genotype tumors by selection of overlapping genes in AKT, AKTP, and AKTPF versus A, respectively (Fig. 5C; Supplementary Table S2). GO term analysis using AKT common DEGs indicated significantly enriched pathways relating to angiogenesis, signal transduction, hypoxia, and epithelial morphogenesis, suggesting the role of these pathways in AKT combination-induced metastasis (Fig. 5D). Hierarchical clustering analysis indicated that expression pattern of ATP was divergent from that of AKT common genotypes (Fig. 5E). Although expression of AKP was distinct from AKT common, we found 175 overlapping DEGs in AKT common and AKP but not with ATP (Fig. 5E–G). These results are consistent with those found by PCA (Fig. 5B). Notably, the expression profile of AK tumors was similar to that of AKT common. We found that organoids from AK tumors but not simple Apc^Δ716 tumors formed submucosal tumors with desmoplastic reactions when they were injected directly into colonic submucosa (Supplementary Fig. S6). Taken together, these results suggest that the AK combination significantly contributes to the AKT-induced malignant phenotype.

We thus classified 363 AKT common DEGs to four clusters of AKT (CL1, CL2, CL3, and CL5 in Fig. 5F; Supplementary Table S2), and performed hierarchical clustering analysis. As expected, 73 genes classified to CL5 showed expression patterns specific to the AKT common genotypes, which were not overlapped with AKP and ATP (Fig. 6A and B). Among 73 AKT-common genes, we selected 60 genes, 21 upregulated and 39 downregulated, which were annotated in the TCGA colorectal cancer database, and extracted expression data of these 60 genes. Importantly, the mean expressions of upregulated AKT-common genes were significantly higher in human colorectal cancer with metastasis (M1) than those without metastasis (M0), suggesting that AKT-common signature is important also for human colorectal cancer metastasis (Fig. 6C). Moreover, we found that expression levels of R3HDML and SYT7 were significantly upregulated in M1 than M0 colorectal cancer tissues, while those of PSME1 and TMEM150C were downregulated, which were consistent with those found in the AKT-common profile (Fig. 6B and C). Of note that SYT7, a calcium sensor for synapse exocytosis, plays a role in cell migration (39, 40). Accordingly, it is possible that upregulated genes in AKT-common profile will be effective therapeutic targets for colorectal cancer metastasis.

To understand how the specific combinations of driver mutations promote each step for colorectal cancer malignant progression, we performed comprehensive phenotype characterization of the primary intestinal tumors and liver metastasis in mouse models that carried various driver mutations in combination (Fig. 7). Among the driver genes examined in this study, the activation mutation of Kras is essential for EMT-like morphology and metastasis. This is consistent with the previous reports that liver and lung metastasis of human colorectal cancer are more likely to develop in patients whose tumors have KRAS mutations (41, 42). In addition, a recent genetic study indicated that the continuous expression of activated Kras is required for the maintenance of invasive and metastatic disease (23). In this study, we found that the Kras^G12D mutation induced EMT-like morphology of Apc^Δ716 mouse tumors when combined with mutations in either of Fbxw7^−/−, Tgfbr2^−/−, or Trp53^R270H. Furthermore, Kras activation promoted intravasation and efficient liver metastasis when combined with Apc^Δ716 and Tgfbr2^−/− mutations (AKT common; Fig. 7). In contrast, the combination of Apc^Δ716 and Kras^G12D mutations was not sufficient for invasion, indicating that malignant progression by Kras mutation requires other driver mutations in combination.

We previously showed that either TGFβ signaling suppression or mutant p53^R270H expression induces submucosal invasion of Apc^Δ716 mouse tumors (20–22). Consistently, compound mice that carried AT or AP mutations in their genotypes consistently developed invasive adenocarcinoma. Unexpectedly, however, simultaneous mutations of Tgfbr2^−/− and Trp53^R270H did not induce further intravasation or metastasis, and the expression profile of ATP organoids was divergent from that of metastatic AKP and AKT organoids. These results confirm the requirement of a Kras mutation in addition to either TGFβ suppression (AKT) or p53 mutation (AKP) for metastasis of colorectal cancer (Fig. 7).

On comparing phenotypes between AKP and AKT tumors, AKT showed markedly more advanced malignancy, such as lymph vessel invasion and significantly higher incidence and multiplicity of metastasis in the liver. Interestingly, the expression profiles of AKTP and AKTPF more closely resemble that of AKT than AKP; therefore, we would like to propose that Wnt activation, Kras activation, and TGFβ suppression (i.e., AKT) are core-combination pathways for efficient metastasis, and the selected AKT common DEGs play a role in this process (Fig. 6B). Importantly, the mean expression level of AKT-common upregulated genes was significantly increased in human colorectal cancer with metastasis (M1), suggesting that AKT-common signature is important also for metastasis of human colorectal cancer. Moreover, we found that Syt7 is upregulated both in AKT-common genes and human colorectal cancer with metastasis. Synaptotagmin (Syt) is a well-established calcium sensor for calcium-induced exocytosis of synaptic vesicles (39). Notably, it has been shown that Syt7-mediated vesicle fusion regulates the chemotaxis of leukocytes (40). Accordingly, it is possible that Syt7 plays a tumor-promoting role by accelerating the migration of colorectal cancer cells. Accordingly, activated pathways by AKT-common mutations may play a role in malignant progression of human colorectal cancer.

In contrast, it has been shown that the TGFβ pathway is activated in the invasive tumor cells of Apc Trp53^Null Kras^G12D mice (23) and that TGFβ signaling in the stroma facilitates tumorigenesis through the promotion of fibrosis and immune evasion (43). Accordingly, it is possible that TGFβ signaling in the stroma plays a role in the malignant progression of AKT tumor cells that have lost their TGFβ signaling.

In this study, we also found that a desmoplastic reaction is induced in submucosal tumors and liver metastasis, while such fibrotic reactions were rarely found in noninvading mucosal tumors of the same mice. Notably, the histologic features of tumors were distinctively more malignant in the invasive and metastatic lesions, suggesting that desmoplastic stroma contributes to the induction of malignant phenotypes. It has been reported that fibrosis with the deposition of type I collagen promotes metastasis by inducing the proliferation of dormant cells via integrin signaling (37, 44, 45), and collagen crosslinking in the stroma results in the promotion of invasion and metastasis (46). Therefore, it is possible that EMT-like morphology and metastasis of AKT tumors are induced by cooperation of driver mutations and fibrotic microenvironment.

The disruption of Fbxw7 increased the number of intestinal tumors in Apc^Min mice (16, 17); however, its role in malignant progression remains unclear. In this study, we showed that the combination of the Fbxw7^−/− and Kras^G12D mutations (AKF) induced EMT-like morphology in mucosal tumors without the support of fibrotic microenvironment (Fig. 7). It has been shown that Fbxw7 plays a role in stem cell regulation, and its deletion accelerates the self-renewal activity of stem cells (47). Furthermore, there is a direct link between EMT and the gain of epithelial stem cell properties (48). It is therefore possible that Fbxw7 disruption leads to EMT-like morphology through the acquisition of stem cell properties, which may contribute to the increased multiplicity of metastasis of AKTPF tumors compared with AKTP.

In this study, we did not find spontaneous metastasis to distant organs in any genotype mice, even if tumor cells can invade into vessels and metastasize to liver from spleen (Fig. 7). It is therefore possible that additional genetic or epigenetic alteration(s) are required to overcome physical stresses in the blood stream and escape from immune attack, which are required to be survived as circulating tumor cells (Fig. 7; ref. 37).

Microsatellite instability (MSI)-high colorectal cancer is associated with an increased level of immune cell infiltration (49), therefore, MSI-high colorectal cancer responds to immune checkpoint blockade (50). However, the association with tumor-infiltrating lymphocytes was also evident in microsatellite-stable colorectal cancer tumors (7), indicating that non-MSI colorectal cancer also can induce an immune reaction. In this study, we found that both CD4⁺ and CD8⁺ T cells infiltrated into metastasized tumors of C57BL/6 mice at the similar level among all genotypes, although T-cell infiltration was rarely detected even in the invasive area of the primary tumors. These results suggest a mechanism of the microenvironment in the metastasized tumors to induce immune responses against non-MSI colorectal cancer cells, although further studies will be needed. This mechanism is particularly important for understanding tumor immunity, which will help develop therapeutic strategies for preventing colorectal cancer metastasis in the future.

In conclusion, we constructed mouse models with multiple combinations of driver mutations and performed the comprehensive characterization of the primary and metastatic tumors. Suppression of the TGFβ pathway or expression of mutant p53 induces submucosal invasion, and additional Kras mutation activation is required for further malignant progression including EMT-like morphology, intravasation, and metastasis. In contrast, the combination of Fbxw7 inhibition and Kras activation induces EMT-like morphology, possibly through increased stemness. Finally, Wnt activation, Kras activation, and TGFβ suppression represent important combination for efficient metastasis, which is corroborated by the RNA-seq analysis. Accordingly, pathways activated by the AKT-common DEGs may be preventive or therapeutic targets against colorectal cancer metastasis.

No potential conflicts of interest were disclosed.

Conception and design: M. Oshima

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Sakai, H. Oshima, S. Fujii, A. Ochiai, Y. Suzuki

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Nakayama, Y. Kouyama, A. Niida, K. Mimori, C.P. Hong, C.-Y. Ock, S.-J. Kim

Writing, review, and/or revision of the manuscript: S. Fujii, C.-Y. Ock, M. Oshima

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.I. Nakayama

Study supervision: K.I. Nakayama, M. Oshima

This work was supported by grants from AMED-CREST, AMED, Japan Agency for Medical Research and Development, Japan and Grants-in-Aid for Scientific Research (A) (15H02362), (C) (26430110), Innovative Areas (17H05616) and (16H06279) from the Ministry of Education, Culture, Sports, Science and Technology of Japan. We thank Yoshie Jomen, Ayako Tsuda, and Manami Watanabe for their technical assistance.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.