FBXW7/hCDC4 Is a General Tumor Suppressor in Human Cancer

The ubiquitin-proteasome system regulates important cellular processes including differentiation and development, apoptosis, protein trafficking, the immune and inflammatory responses, cell cycle progression, and cellular division (1). The latter processes are primarily regulated by two ubiquitin ligases known as the anaphase-promoting complex (APC) and SCF. SCF ubiquitin ligases are composed of Cul1, Rbx1 (also called Roc1 or Hrt1), and Skp1 bound to a member of the F-box protein family, which provide substrate specificity. Approximately 65 F-box proteins exist in humans, each characterized by an ∼40-amino-acid F-box motif that associates with the SCF complex through Skp1 (1).

In S. cerevisiae, the F-box protein Cdc4 plays a critical role in cell cycle control by mediating the ubiquitin-dependent proteolysis of DNA replication protein Cdc6 and cyclin-dependent kinase (Cdk) inhibitors Far1 and Sic1 (2, 3). Similarly, the orthologue of Cdc4 in humans, designated Fbxw7 (also known as hCdc4, Fbw7, or SEL-10), mediates the ubiquitin-dependent proteolysis of several key regulatory proteins involved in cell division and cell fate determination, including cyclin E1, c-Myc, c-Jun, Notch, and Aurora-A (2). Furthermore, FBXW7 deficiency in mice has been shown to result in early embryonic lethality (day ∼10.5) with embryos exhibiting abnormalities in hematopoietic and vascular development (2).

Several lines of evidence suggest that Fbxw7 is a putative tumor suppressor in human tumorigenesis, including (a) the well-defined oncogenic potential of its putative substrates (described above); (b) localization of FBXW7 to chromosome 4q31.3, which is deleted in ∼30% of human cancers; (c) the finding that allelic loss of FBXW7 cooperates with p53 in tumorigenesis in mice (4); and (d) demonstration that targeted disruption of FBXW7 in cultured cells leads to an increase in genetic instability (5), a hallmark of human cancers.

In this study, we carry out an extensive genetic analysis of FBXW7 in primary human tumors of diverse tissue origin to determine its role as a putative tumor suppressor. We find that FBXW7 is mutated in a variety of human tumor types with an overall mutation frequency of ∼6%. Furthermore, we show that expression of an Fbxw7 mutant corresponding to one of the major mutational hotspots in primary tumors interferes with wild-type Fbxw7 function, suggesting a potential dominant negative mechanism of Fbxw7 inactivation. Our results show that FBXW7 is a general tumor suppressor in human tumorigenesis and provide insight into how Fbxw7 function is inactivated during tumorigenesis.

Tumor specimens. Primary tumor specimens (534 in total) were obtained from (a) Department of Obstetrics and Gynecology, Innsbruck Medical University, Innsbruck, Austria (breast and ovarian tumor specimens); (b) Sidney Kimmel Cancer Center, San Diego, CA (lung and prostate); (c) Department of Pathology and Gastroenterology, Uppsala University Hospital, Uppsala, Sweden (bile duct and liver); (d) Department of Oncology/Pathology, Karolinska Hospital, Stockholm, Sweden (hematologic malignancies and melanomas); and (e) Research Center for Gastrointestinal and Liver Disease, Talghani Hospital, Tehran, Iran (stomach, bladder, colon, and esophagus). This project was approved by the research ethics committees at the Karolinska Institute, Uppsala University Hospital, and Shahid Beheshti University of Medical Sciences. A total of 1,556 primary tumor specimens (our results pooled with previously published data) are included in our final analysis.

FBXW7 mutation screening. Screening for FBXW7 mutations was done as previously described (6).

Methylation analysis. Methylation status of mutational hotspot 1435-6 was determined by bisulfite sequencing as described (7). PCR amplifications were done with primers 5′-TGTGGAATGTAGAGATTGGAGAATGTATA-3′ and 5′-AAAAAATCCCAACCATAACAAAATTT-3′.

Antibodies. Antibodies used in this study include anti-Flag (Sigma), anti-Skp1 (Lab Vision), anti–cyclin E1 (HE12, Santa Cruz Biotechnology), anti–phospho-T380 cyclin E1 (Santa Cruz Biotechnology), anti-Myc (9E10, Santa Cruz Biotechnology), anti–glutathione S-transferase (GST; Santa Cruz Biotechnology), anti-actin (Santa Cruz Biotechnology), anti-p53 (DO-1, Santa Cruz Biotechnology), and anti-p21 (Ab-1, Calbiochem). Precipitation experiments were done using anti-Flag agarose (Sigma) or Glutathione-Sepharose (GE Healthcare). Immunohistochemical analysis of p21 and p53 was done as described (8). Analysis of ploidy was determined by Feulgen staining of histopathologic sections and image cytometry.

Plasmids and transfection. The full-length α-Fbxw7 cDNA was cloned into pFlag-CMV2 (Sigma) and GST expression vector pEBG. Expression plasmids for all FBXW7 mutants were generated using the GeneTailor Site-Directed Mutagenesis System (Invitrogen). Expression plasmid for 3× Myc-cyclin E1 was generated by cloning the full-length cyclin E1 cDNA into pcDNA3.1 (Invitrogen). Transfections were done using Lipofectamine 2000 (Invitrogen).

Molecular modeling. The molecular structure of α-Fbxw7 was created using the SWISS-MODEL program and viewed using the DeepView-SwissPDB-Viewer.

FBXW7 is mutated in a variety of primary human tumor types. To determine the role of FBXW7 as a tumor suppressor in human tumorigenesis, we genetically analyzed the complete coding region, including sequences unique to the three alternatively spliced Fbxw7 isoforms (α, β, and γ), in a variety of human tumor types. The results of this analysis, combined with previously reported data (9–17) for FBXW7 in primary tumors (1,556 specimens in total), are presented in Table 1. FBXW7 mutations were detected in tumors of diverse tissue origin, including those of the blood, breast, bile duct, colon, endometrium, stomach, lung, bone, ovary, pancreas, and prostate. The highest frequencies of mutations were observed in tumors of the bile duct (cholangiocarcinomas, 35%), blood (T-cell acute lymphocytic leukemia, 31%), colon (9%), endometrium (9%), and stomach (6%). Overall, the frequency of FBXW7 mutations in all primary human cancers analyzed was ∼6% (87 of 1,556). DNA sequencing of matching negative biopsies from patients with stomach cancers showed that the FBXW7 mutations were exclusively in the tumor specimens, indicating that the mutations were somatic in nature.

Spectra of FBXW7 mutations reveal mutational “hotspots” and isoform-specific mutations. Of the mutations detected in FBXW7, the vast majority (94%) were single nucleotide changes, with the remainder (6%) being either small nucleotide deletions or insertions. Seventy-four percent of mutations introduced a single amino acid substitution (missense) in the Fbxw7 protein, whereas 26% induced a premature termination of translation (nonsense; Table 1). FBXW7 mutations were found to occur throughout the coding region, including the F-box motif, WD40 repeats, and two of three unique 5′ exons that determine the Fbxw7 isoforms (Fig. 1). Interestingly, isoform-specific mutations were found to represent ∼6% of FBXW7 mutations. Approximately 43% of all FBXW7 mutations occurred at two mutational hotspots positioned at codons Arg⁴⁶⁵ (nucleotides 1,393–1,394; 29%) and Arg⁴⁷⁹ (nucleotides 1,435–1,436; 14%). Additional minor mutational hotspots were detected at nucleotides 670 (Arg²²⁴; 3%), 832 (Arg²⁷⁸; 4%), 1,177 (Arg³⁹³; 3%), and 1,745 (Ser⁵⁸²; 4%). Molecular modeling revealed that codons Arg⁴⁶⁵, Arg⁴⁷⁹, and Ser⁵⁸² are localized to the surface or lumen of the β-propeller structure of Fbxw7; Arg³⁹³ is at the rear of the β-propeller structure; and Arg²⁷⁸ is situated close to the isoform-specific NH₂ terminus of Fbxw7.

Mutation hotspots in FBXW7 correspond to methylated CG dinucleotides with signatures of deaminations of 5′-methylcytosine. Investigation of the mutational hotspots of FBXW7 revealed that all seven are centered on CG dinucleotides, potential sites of DNA methylation in humans. Interestingly, the prominent mutation type that occurs at each mutational hotspot was either a C→T or G→A transition within the CG dinuclotide, consistent with the spontaneous deamination of 5′-methylcytosine→thymine in DNA (18). Because 5′-methylcytosine has been shown to act as an endogenous mutagen (18), we analyzed the methylation status of DNA isolated from several primary tumor types to determine whether the cytosine corresponding to the mutation hotspot at nucleotides 1,435–1,436 (Arg⁴⁷⁹) was methylated in vivo. Indeed, bisulfite-based sequence analysis revealed that the cytosine was methylated in 18 of 18 primary tumors examined, including those of the ovary, breast, and endometrium (data not shown).

Functional analysis of Fbxw7 mutants reveals defects in localization and substrate binding. Substrate recognition by Fbxw7 is mediated through the interaction of the β-propeller surface formed by the WD40 repeats with the phosphodegron of the substrate protein (2, 3). To examine the consequence on substrate binding of the hotspot mutations, we expressed α-Fbxw7^Arg465 and α-Fbxw7^Arg479 in HEK 293T cells and assessed their ability to bind substrate cyclin E1. Whereas wild-type α-Fbxw7 readily bound cyclin E1 in vivo, mutants α-Fbxw7^Arg465 and α-Fbxw7^Arg479 did not (Fig. 2), despite their proper localization as assessed by immunofluorescence (data not shown). Interestingly, many FBXW7 mutations fall outside of the WD40 repeats and the effect of these on Fbxw7 function is unclear. To analyze this in more detail, we functionally tested mutant FBXW7^+p16 identified in a prostate tumor specimen that contains a proline residue inserted at amino acid 16 of the α-isoform (Table 1). Expression of α-Fbxw7^+p16 in HEK 293T cells showed that it was incapable of binding cyclin E1 substrate in vivo, although it could readily associate with the SCF core components (Fig. 2). Because residue 16 is in close proximity to the proposed nuclear localization signal of the α-isoform (2), we next tested whether α-Fbxw7^+p16 is properly localized in cells. Immunofluorescence analysis showed that whereas wild-type α-Fbxw7 is almost exclusively localized to the nucleus in HEK 293T cells, α-Fbxw7^+p16 is restricted to the cytoplasm (Fig. 2). These results show that the insertion of proline at residue 16 abolished the nuclear localization signal, preventing interaction with cyclin E1 substrate in the nucleus.

FBXW7 mutants corresponding to a mutational hotspot in primary tumors act dominant negatively to inactivate Fbxw7 function. Recently, it was shown that isoforms of Fbxw7 can form homodimers and heterodimers in vivo (19), raising the possibility that mutants of Fbxw7 could interfere with wild-type Fbxw7 function in a dominant negative fashion. To explore this possibility, we cotransfected HCT116^FBXW7−/− cells that have both FBXW7 alleles inactivated by targeted disruption (5) with plasmids that express Myc-cyclin E1, Cdk2, α-Fbxw7, and increasing amounts of α-Fbxw7^R465C. The results of this experiment showed that coexpression of α-Fbxw7^R465C with wild-type α-Fbxw7 resulted in a marked accumulation of cyclin E1 compared with expression of wild-type α-Fbxw7 alone (Fig. 3A). Similar results were observed for expression of α-Fbxw7^R479Q, corresponding to the other major mutational hotspot in primary tumors, in HCT116^FBXW7−/− cells and for both mutants in HEK 293T cells that contain wild-type FBXW7 alleles (data not shown). We next carried out cycloheximide-chase experiments to determine whether the increase in cyclin E1 level observed for α-Fbxw7^R465C expression was due to increased protein stability. As expected, expression of wild-type α-Fbxw7 resulted in a decrease in cyclin E1 stability (half-life of ∼1.2 h) compared with expression of α-Fbxw7^R465C or vector alone (half-life >3 h). In contrast, coexpression of α-Fbxw7 and α-Fbxw7^R465C resulted in an increase in cyclin E1 half-life (>3 h), comparable to expression of α-Fbxw7^R465C or vector alone (Fig. 3).

Our results show that FBXW7 is a general tumor suppressor in human tumorigenesis. FBXW7 mutations were detected in primary human tumors of the blood, breast, bile duct, bone, colon, endometrium, stomach, lung, ovary, prostate, and pancreas, with an overall mutation frequency of ∼6%. Mutations were most frequent in tumors of the bile duct (35%), blood (T-cell acute lymphocytic leukemia, 31%), colon (9%), endometrium (9%), pancreas (9%), and stomach (6%), suggesting a potential tissue specificity of FBXW7 inactivation in human tumorigenesis.

FBXW7 mutations were found localized throughout the coding region of the gene including the isoform-specific 5′-exons, nuclear localization signal, F-box motif, and WD40 repeats. These mutations are predicted to adversely affect isoform-specific functions, subunit dimerization, Pin-1 association, localization, SCF assembly, and substrate recognition, highlighting the plethora of ways Fbxw7 can be functionally inactivated in tumors. Interestingly, ∼6% of FBXW7 mutations are isoform specific, suggesting that inactivation of an individual isoform could play a causative role in human tumorigenesis. This finding could reflect the potential of a mutant Fbxw7 isoform to act dominant negatively (see below) or that individual Fbxw7 isoforms may target a unique set of substrates for ubiquitination. In support of the latter hypothesis, Fbxw7 isoforms are differentially localized in mammalian cells (α-nuclear, β-cytoplasmic, and γ-nucleolar; ref. 20). Furthermore, ubiquitination of c-Myc has been shown to be mediated by γ-Fbxw7, which selectively binds substrate in the nucleolus (2). Additionally, ubiquitination of cyclin E1 is mediated through a cooperative effort of α- and γ-Fbxw7, in which α-Fbxw7 binds cyclin E1 and stimulates prolyl-isomerization by Pin-1, which then promotes efficient ubiquitylation by SCF^γ-Fbxw7 (20). Indeed, one of the α-isoform–specific mutations detected in tumors (D124Y) was shown to be defective in supporting Pin-1–mediated isomerization of cyclin E1 but was not defective in ubiquitylation function (20). It is currently not understood whether other SCF^Fbxw7 substrates are ubiquitinated via an analogous cooperative mechanism of Fbxw7 isoforms.

Our analysis identified seven mutational hotspots in FBXW7, representing >60% of the total mutations. Two major hotspots located at nucleotides 1,393–1,394 (Arg⁴⁶⁵) and 1,435–1,436 (Arg⁴⁷⁹) account for ∼43% of all mutations. The amino acid residues corresponding to six of seven mutational hotspots in primary human tumors are evolutionarily conserved in flies, worms, and yeast (2), showing their potential importance in Fbxw7 function. Interestingly, all seven FBXW7 mutational hotspots involve CG dinucleotides, where the resultant mutation is primarily a C→T or G→A transition. Consistent with 5′-methylcytosine deamination as a mechanism for FBXW7 mutation, the cytosine nucleotide at hotspot position 1,435–1,436 (Arg⁴⁷⁹) was found to be methylated in all primary tumors analyzed. These data stress the potential importance of endogenous methylation in mutational inactivation of Fbxw7 in tumors, as has previously been postulated for the tumor suppressor p53 (18).

It is currently not understood what role, if any, other genetic and epigenetic mechanisms play in FBXW7 inactivation in human tumors. In many primary tumors and derived cell lines, FBXW7 mutations frequently occur without a concomitant loss or mutation of the remaining allele, suggesting that FBXW7 may not follow the classic “two-hit” model of tumor suppressor gene inactivation. It is possible that reduced expression of the wild-type allele might be sufficient to abrogate tumor suppressor activity or that FBXW7 mutations could act dominant negatively. The former possibility is supported by a study that showed that loss of a single FBXW7 allele can accelerate tumor development in p53^+/− mice (4), suggesting FBXW7 is a p53-dependent haploinsufficient tumor suppressor. However, we analyzed FBXW7 and p53 mutations in gastric cancers and failed to detect cooperation between these proteins (Supplementary Table S1). It remains to be determined whether FBXW7 expression is reduced or silenced in tumors through epigenetic mechanisms such as promoter methylation.

Our data show that Fbxw7 mutants corresponding to the major mutational hotspots in primary tumors (Arg⁴⁶⁵ and Arg⁴⁷⁹) can act dominant negatively to abrogate wild-type Fbxw7 function. Coexpression of wild-type α-Fbxw7 and α-Fbxw7^R465C was found to increase the steady-state level of cyclin E1 as well as its half-life, compared with expression of wild-type α-Fbxw7 alone. The different isoforms of Fbxw7 interact in vivo through association of their D-box motifs located in the NH₂-terminal region of the protein, and this interaction enhances SCF^Fbxw7-associated ubiquitylation activity (19). If Fbxw7 predominantly functions as a homodimer or heterodimer in vivo, then expression of a single mutant allele may be sufficient to functionally inactivate its tumor suppressor function. This could explain why most primary tumors contain only a single mutant FBXW7 and retain a wild-type allele. Further study is needed to determine whether other Fbxw7 mutations also act dominant negatively and whether expression at physiologic levels is sufficient to exert the dominant negative effects on wild-type Fbxw7.

Grant support: University of California Tobacco-Related Disease Research Program, Swedish Cancer Society, the Swedish Research Council, the Swedish Institute, the Children Cancer Foundation, the Cancer Society of Stockholm, and Karolinska Institute Foundations.

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.

We thank Dr. Katayoun Aghajani, Mohsen Chiani, and Dr. Ali Mansouri for technical support in tumor specimen preparation. We also thank Bert Vogelstein for providing HCT116^FBXW7−/− cells for this study.