A subset of Wnt-addicted cancers are sensitive to targeted therapies that block Wnt secretion or receptor engagement. RNF43 loss-of-function (LOF) mutations that increase cell surface Wnt receptor abundance cause sensitivity to Wnt inhibitors. However, it is not clear which of the clinically identified RNF43 mutations affect its function in vivo. We assayed 119 missense and 45 truncating RNF43 mutations found in human cancers using a combination of cell-based reporter assays, genome editing, flow cytometry, and immunofluorescence microscopy. Five common germline variants of RNF43 exhibited wild-type activity. Cancer-associated missense mutations in the RING ubiquitin ligase domain and a subset of mutations in the extracellular domain hyperactivate Wnt/β-catenin signaling through formation of inactive dimers with endogenous RNF43 or ZNRF3. RNF43 C-terminal truncation mutants, including the common G659fs mutant are LOF specifically when endogenous mutations are examined, unlike their behavior in transient transfection assays. Patient-derived xenografts and cell lines with C-terminal truncations showed increased cell surface Frizzled and Wnt/β-catenin signaling and were responsive to porcupine (PORCN) inhibition in vivo, providing clear evidence of RNF43 impairment. Our study provides potential guidelines for patient assignment, as virtually all RNF43 nonsense and frameshift mutations, including those in the C-terminal domain and a large number of patient-associated missense mutations in the RING domain and N-terminal region compromise its activity, and therefore predict response to upstream Wnt inhibitors in cancers without microsatellite instability. This study expands the landscape of actionable RNF43 mutations, extending the benefit of these therapies to additional patients.
Systematic examination of patient-derived RNF43 mutations identifies rules to guide patient selection, including that truncation or point mutations in well-defined functional domains sensitize cancers to PORCN inhibitors.
Wnts are secreted short-range signaling molecules that play key roles during embryonic development and adult homeostasis. As Wnt proteins are synthesized, they are post-translationally palmitoleated by the O-acyltransferase porcupine (PORCN), a modification that is required for both Wnt secretion and subsequent binding to receptors, including Frizzled (FZD) on the receiving cell surface (1–3). Wnt ligand–receptor interaction causes the accumulation of β-catenin in the cytoplasm and its subsequent translocation to the nucleus where it interacts with transcription factors such as TCF/LEF to regulate target gene expression.
Mutations that activate the Wnt/β-catenin signaling pathway are found in many cancers. The most common mutations in Wnt signaling affect downstream regulators of β-catenin degradation such as the adenomatous polyposis coli (APC) protein. However, there is a distinct class of cancers that are driven by mutations in genes, including RNF43, ZNRF3, and RSPO3 (4–9). These cancers have hypersensitivity to Wnt ligands and multiple approaches to treat these Wnt-addicted cancers, including PORCN inhibitors, anti-FZD antibodies and anti–R-spondin (RSPO) antibodies have shown efficacy in preclinical studies (10–14).
RNF43 and its paralog ZNRF3 are integral membrane E3 ubiquitin ligases that ubiquitylate cytoplasmic sites on FZD, driving its lysosomal degradation and hence negatively regulating its abundance at the cell surface (6, 15). The activity of RNF43 and ZNRF3 at the cell surface is in turn tightly regulated by RSPO ligands and their membrane co-receptors LGR4/5/6 (16, 17). The heterotrimeric complex of LGR–RSPO–RNF43/ZNRF3 inhibits ubiquitylation of FZD, thus increasing FZD cell surface abundance and cellular sensitivity to Wnts. This is clinically relevant because loss-of-function (LOF) mutations in RNF43 and gain-of-function mutations in RSPO2 and RSPO3 that increase FZD abundance are found in multiple cancer types and cause Wnt addiction (5, 18–20).
One of the goals of precision medicine is to identify actionable mutations. Actionable RSPO chromosome translocations are identifiable by PCR-based assays (9). However, missense and indel mutations in the RNF43 gene identified by targeted or NextGen sequencing present a challenge. These mutations occur along the entire RNF43 coding sequence in multiple cancer types, including colorectal, pancreatic, ovarian, biliary tract and gastric (4, 21–26) and with an elevated frequency at two G•C tracks in cancers with microsatellite instability (MSI; refs. 4, 20). Although selected mutations have been studied, a broad survey of how these mutations affect the activity of RNF43 is needed (6, 21, 24). As PORCN inhibitors advance in clinical trials, delineating which RNF43 mutations drive Wnt addiction and which are simply passengers is essential in selecting patients with cancer who might benefit from this treatment.
In this study, we profiled 164 RNF43 mutations identified in human cancers. We find that missense mutations in well-defined functional domains, as well as frameshift and nonsense mutations in the N-terminal half of RNF43 are largely LOF. A subset of these are hyperactivating, likely due to their ability to form inactive multimers with nonmutant protein. Multiple truncating mutations in the RNF43 C-terminus, including the recurrent G659Vfs, show LOF in vivo despite variable results in vitro and may therefore be actionable in patients with cancer without MSI. This study of the mutational landscape of RNF43 facilitates the selection of patients who may benefit from upstream Wnt pathway inhibitors.
Materials and Methods
Cell lines, plasmids, and antibodies
HEK293 and Panc 08.13 cells were obtained from the ATCC and were authenticated by testing for known RNF43 mutations. Mycoplasma test was performed regularly using MycoAlert Mycoplasma detection kit (Lonza). HEK293 cells were cultured in DMEM medium supplemented with 10% FBS, l-glutamate, pen/strep, and sodium pyruvate. Panc 08.13 cells were cultured in RPMI-1640 medium supplemented with 15% FBS, l-glutamate, pen/strep, sodium pyruvate, and insulin. The cells were passaged every 3 to 4 days and used for experiments before passage number 20.
The set of HA–FZD constructs were the generous gift of Jeffrey Rubin. They contain the signal peptide of human FZD5, followed by 2XHA tags and then the corresponding mature FZD sequences. FZD1 and FZD2 are rat, FZD5 is human, and the remainders are mouse sequences. Monoclonal antibodies clone F2 and clone F2.A were obtained from S. Sidhu and S. Angers, University of Toronto (27). OMP-18R5 was used as previously described (11, 28).
Generation of RNF43 mutant constructs
A C-terminal Myc-FLAG (also known as DDK)-tagged RNF43 ORF expression construct was purchased from Origene (Origene, Cat# RC214013) and verified by Sanger sequencing. This vector was used to introduce all patient-derived mutations using site-directed mutagenesis following the manufacturer's protocol (Stratagene). For some experiments, the FLAG epitope was deleted by introducing a stop codon after the sequence encoding Myc. For coimmunoprecipitation assays, we also used a 3xFLAG-internal tagged RNF43 that was generated by 2-step PCR, placing 3xFLAG after the signal peptide. All constructs including the patient-derived mutants were verified by Sanger sequencing.
Wnt/β-catenin reporter assay
One day before transfection, HEK293 cells were seeded into 24-well plates coated with poly-L-lysine. Cells were transfected using Lipofectamine 2000 (Thermo Fisher) mixed with DNA in a ratio of 3:1 using the manufacturer's protocol. For each well 100-ng TOPFlash, 50-ng mCherry, 10-ng PGK-mWNT3A, and 5-ng (unless otherwise indicated) RNF43 plasmid mix were used. Forty-eight hours after transfection, cells were lysed with 100 μL Reporter Lysis Buffer (Promega) supplemented with protease inhibitor cocktail (Roche) for 20 minutes at 4°C with shaking. The cell lysates were transferred to a 96-well black plate and then an equal volume of Luciferase assay reagent (Promega) was added. Luciferase activity was measured on a Tecan Infinite 2000 plate reader. mCherry values were used for normalization.
For reporter assays performed using WNT3A-conditioned medium, HEK293T cells in 24-well plates were transfected with 66-ng Wnt/β-catenin reporter, 66-ng RNF43 expression, and 8-ng CMV-Renilla expression plasmid. L-WNT3A or L-Control–conditioned medium (diluted 10-fold in normal growth media) was added after 24 hours. Cells were lysed in passive lysis buffer (Promega) 72 hours after transfection. The Firefly and Renilla luciferase activity was measured with the Dual-Luciferase Reporter Assay System (Promega) using a LumiStar Optima luminescence counter (BMG LabTech). All luciferase reporter assays were carried out in triplicate. Firefly luciferase value was normalized to Renilla luciferase to calculate relative luciferase units.
IHC staining for surface FZD
Formalin-fixed and paraffin-embedded (FFPE) sections of the xenograft tumors were dewaxed following a standard protocol. The section slides were boiled in Tris-EDTA (pH 9.0) buffer for 20 minutes for antigen retrieval. A human pan-FZD antibody clone F2 (27, 29) was used at 1:1,000 dilution with anti-human IgG secondary antibody conjugated with horseradish peroxidase.
Mouse xenograft studies
All mouse studies were approved by the Institutional Animal Care and Use Committee. The patient-derived xenograft (PDX; PAXF1861 and PAXF1869) studies were performed by Oncotest. FFPE with PDXs PA3127 and PA1457 were purchased from Crown Bioscience Inc. Mouse xenograft models were established by subcutaneously injecting HPAF-II or AsPC-1 cells mixed with Matrigel into NSG mice or implanting patient-derived solid tissue fragments in NMRI nude mice. ETC-159 formulated in 50% PEG400 (vol/vol) in water was administered by oral gavage at a dosing volume of 10 μL/g bodyweight. Tumors were measured by calipers and volumes calculated using the formula (length/2 × width⁁2). All mice were sacrificed 8 hours after the last dosing. At sacrifice, tumors were resected, weighed, and snap-frozen in liquid nitrogen or fixed in 10% neutral-buffered formalin.
Analysis of the RNF43 mutational landscape
We systematically examined the distribution of RNF43 mutations identified in multiple human cancer types using cBioPortal (30, 31). Truncating (nonsense and frameshift) mutations were enriched in the first half of the protein, up until the end of the ubiquitin ligase RING domain (Fig. 1A; Supplementary Fig. S1A; P = 1.45 × 10−8, binomial test). Missense mutations appeared more evenly distributed (Fig. 1B; Supplementary Fig. S1B). We therefore investigated whether they were enriched or depleted in specific functional domains. RNF43 has an extracellular protease-associated (PA) domain that interacts with RSPOs (32) and FZD (24), a single-pass transmembrane (TM) domain, and an intracellular RING domain with E3 ubiquitin ligase function. Both the PA and RING domains of RNF43 had significantly more missense mutations than expected by chance (Fig. 1C, P < 0.05), suggesting that there is positive selection acting on these domains in cancer. There was no evidence of positive selection for missense mutations in the C-terminal domain (CTD) of RNF43 (Fig. 1C). A permutation test confirmed that several amino acids in RNF43 had a significantly higher frequency of truncation or missense mutations than expected by chance alone, including three we identified as common germline variants (Fig. 1A and B). Furthermore, we observed that only 16% (N = 79) of the RNF43 mutations were shared between different cancer types, the most common being G659Vfs*41 (Fig. 1D), indicating that each cancer type has unique RNF43 mutations. Given the diverse nature of the mutational landscape, we asked if there were general principles that determined whether an RNF43 mutation was a LOF.
Systematic study of RNF43 missense mutations reveals 30% to be LOF or hyperactivating
It can be difficult to determine a priori if a specific missense mutation is a passenger or a driver. To get a better estimate of the actionable RNF43 mutations, we systematically analyzed 119 patient-derived RNF43 missense mutants, including five common RNF43 germline variants, as these are likely to be found in many cancers. The common germline variants were identified using the genome aggregation database gnomAD and a cutoff of 10% allele frequency, with I47V and L418M being the most common with an allele frequency of approximately 0.4 and R117H, P231L, and R343H with frequencies of approximately 0.15–0.18 (33).
The remaining 114 disease-associated missense mutants were selected on the basis of the following criteria. First, we chose to evaluate virtually all missense mutations from pancreatic cancers found in cBioPortal, because RNF43 LOF mutations are well established to render those cancers sensitive to Wnt inhibition (5, 19). In addition, we also examined a subset of RNF43 mutations present in colorectal cancers with an emphasis on those found in microsatellite stable (MSS) cancers (as assessed by mutational burden) as well as mutations that are present at high frequencies in other cancer types. We focused on MSS tumors because MSI tumors have multiple additional oncogenic driver mutations and are responsive to alternative therapies such as checkpoint inhibitors (34, 35). Finally, we evaluated four RNF43 mutations found in commercially available pancreatic ductal adenocarcinoma PDX models.
We used transient expression of RNF43 in a cell-based reporter assay to study the effect of individual RNF43 mutations. In this assay, wild-type RNF43 reduces Wnt-activated signaling by reducing the abundance of FZD and other Wnt receptors on the cell surface. Benign RNF43 variants should retain activity and reduce Wnt/β-catenin signaling similar to wild-type. To minimize overexpression artifacts and ensure the sensitivity of the assay, we first titrated the quantity of RNF43 expression plasmid to ensure that the amount used is not near saturation (Supplementary Fig. S2A). Both common variants and disease-associated alterations were introduced into an RNF43 expression vector by site-directed mutagenesis and their ability to regulate Wnt-induced β-catenin reporter activity (TOPFlash) was measured. To assess the robustness of the assay, we compared the activity of selected mutants on signaling stimulated by WNT3A or WNT7B in HEK293 cells, and by WNT3A in a pancreatic cancer cell line, Panc 08.13, both cell lines with wild-type RNF43 and ZNRF3 (Fig. 2A; Supplementary Fig. S2B and S2C). Similar results were obtained in all assays, indicating that the results obtained are robust and generalizable. Importantly, the five common RNF43 germline variants were as active as wild-type RNF43 (Fig. 2A), suggesting that they do not contribute to cancer risk and are not actionable. Subsequent screening of the 84 RNF43 mutants was performed using HEK293 cells with transfected WNT3A. Independently, a set of 41 patient-associated RNF43 mutants were tested for their ability to modify the response of cells to WNT3A-conditioned media (21), with 11 mutants overlapping with the first screening. The activity of the RNF43 mutants is reported as the percentage of inhibition of Wnt signaling, where wild-type RNF43 produces 80%–90% inhibition. The mutants that inhibited signaling less than 20% were defined as LOF and those that actually increased signaling greater than 20% above baseline were termed hyperactivating or dominant negative (Fig. 2B and C).
In the region preceding or in the PA domain (amino acid 87–186), 33% of the missense mutations were either LOF or hyperactivating (Fig. 2B and C; Supplementary Tables S1 and S2). Almost all of the missense mutations in the RING domain (aa 272–313) were hyperactivating. There was also a small cluster of missense mutations in the C-terminus around aa 470–490 that had compromised activity, suggesting the presence of an additional functional domain that may be related to Casein Kinase 1 binding (36). Taken together, around 30% of the tested missense mutations in RNF43 are either LOF or hyperactivating (Supplementary Tables S1 and S2).
To specifically test whether the RNF43 hyperactivating mutants are actionable, that is, if they are sensitive to Wnt inhibition, we compared the activity of selected mutants using the Wnt/β-catenin reporter assay in the presence of the PORCN inhibitor ETC-159 (19). As expected, PORCN inhibition blocked baseline WNT3A-induced signaling (Fig. 2D). Importantly, PORCN inhibition also reduced the enhanced signaling from all tested hyperactivating mutants (V162M, G166V, I186T, R286W, and D300G) by >95%. Thus, Wnt secretion is required for the enhanced signaling from RNF43-hyperactivating mutants. Inhibition of Wnt signaling using upstream inhibitors such as anti-FZD antibodies or PORCN inhibitors could, therefore, be a useful treatment in cancers with either RNF43 LOF or hyperactivating mutations.
RNF43 regulates the cell-surface abundance of multiple FZD proteins
An established molecular function of RNF43 is to ubiquitylate FZD, leading to its endocytosis and lysosomal degradation (6). There are 10 FZD genes in the mammalian genome, and whether RNF43 can act on all of them is not known. Flow cytometry allows for assessment of FZD abundance at the plasma membrane. The FZDs can be subdivided into four groups based on their sequence similarity (Fig. 3A). We first transiently expressed each of the hemagglutinin-epitope (HA)–tagged FZD proteins in HEK293 cells. FZD 2, 4, 5, 7, and 10, representing three of the four groups, were well expressed and readily detectable in immunoblots with an anti-HA antibody (Fig. 3B). To test whether RNF43 could regulate the cell surface abundance of these FZDs, they were co-expressed with or without RNF43 in Panc 08.13 cells that have wild-type RNF43 and ZNRF3 and relatively low endogenous FZD levels. As analyzed by flow cytometry (Fig. 3C,–G), co-expression of wild-type RNF43 reduced the cell surface abundance of each of the five FZDs tested. As an alternative approach to predict whether FZD 1, 3, 6, 8, and 9, which we were unable to assess due to poor expression levels, are also RNF43 targets, we performed multiple sequence alignment of all the 10 FZD family members. RNF43 regulates FZD cell surface accumulation by ubiquitination of specific lysine residues (6). The lysines present in both the intracellular loops and the CTD were largely conserved in all ten FZDs (Supplementary Fig. S3). This suggests that all FZD proteins may potentially be regulated by RNF43.
The activity of RNF43 mutants correlates with their ability to regulate FZD abundance, endocytosis, and degradation
We used several approaches to test whether the activity of RNF43 mutants in regulating the Wnt/β-catenin reporter correlated with their ability to regulate endogenous cell surface FZD abundance. We assessed multiple mutants, selecting representative hyperactivating, WT and LOF mutants in the N-terminal region. First, cell surface FZD levels were measured by flow cytometry using a pan-FZD antibody OMP-18R5 that recognizes five of the 10 FZD family members (FZD 1, 2, 5, 7, and 8; ref. 11). HEK293 cells have relatively high baseline levels of cell surface FZD (Fig. 4A, blue line). Consistent with the ability of RNF43 to reduce Wnt/β-catenin reporter activity, expression of WT RNF43 significantly reduced the abundance of FZD on the cell surface (Fig. 4A, red line). RNF43 PA domain mutant I186T and RING domain mutant R286W that were hyperactivating in reporter assays (Fig. 2E) increased the abundance of surface FZD (Fig. 4A) whereas S94I, P118T, and A146G mutants that had wild-type activity in Wnt/β-catenin reporter assays behaved like wild-type RNF43 in their ability to reduce FZD abundance (Fig. 4B). We also analyzed the LOF RNF43 mutants and, as expected, these mutants did not change the baseline level of FZD in HEK293 cells (Fig. 4C). Similar results were obtained with several mutants tested using a different pan-FZD antibody in Panc 08.13 cells (Supplementary Fig. S4A).
Flow cytometry allows the assessment of membrane-bound FZD but not its intracellular fate. To further test whether patient-derived RNF43 mutants are impaired in inducing FZD internalization, we used a SNAP–FZD5 construct (6). As reported previously, when cells expressing SNAP–FZD5 are treated with a membrane-impermeable SNAP Alexa549 reagent, the labeled FZD5 receptor is visualized at the cell surface (Fig. 4D). Co-expression of RNF43 leads to rapid internalization of surface-labeled FZD5 into endocytic vesicles (Fig. 4E). Supporting the flow cytometry and the Wnt/β-catenin reporter assays, co-expression of RNF43-hyperactivating mutants V162M, G166V, I186T or R286W, did not lead to FZD5 endocytosis but instead enhanced its membrane localization and abundance (Fig. 4F,–I).
As reported by Tsukiyama and colleagues (24), mutations (I48T, L82S, R127P) in RNF43 can lead to its mislocalization. We confirmed and extended their observations, that WT RNF43 localizes to cytoplasmic vesicles with a granular-staining pattern, whereas extracellular domain mutants G166V, I186T, L82S, and A169T had a reticular pattern and largely colocalized with the endoplasmic reticulum (ER) marker calnexin (Supplementary Fig. S4B). The altered localization may compromise the ability of these mutants to interact with FZD. Consistent with this, we observed that the hyperactivating mutants in the PA domain, but not the silent mutant S94I, had decreased ability to coimmunoprecipitate with FZD (Supplementary Fig. S4C). Mapping these hyperactivating mutants on the structure of the RNF43 extracellular domain (PDB: 4KNG; ref. 32) demonstrates that they cluster on or near the surface of a discrete region that we speculate could be involved in interaction with the extracellular domain of FZD and/or other membrane proteins (Supplementary Fig. S4D).
FZD proteins are glycosylated during their maturation process in the ER and Golgi. The failure of inactive RNF43 mutants to induce FZD internalization was also reflected in the relative abundance of the mature (glycosylated) and the immature (non-glycosylated) forms of FZD5 (6, 37). Treatment with the glycosylation inhibitor tunicamycin collapsed the mature as well as the immature bands of FZD5 into a single band with faster mobility (Fig. 4J, lane 3). Co-expression of FZD5 with wild-type RNF43 or RNF43 with silent mutations S94I, P118T or A146G led to the disappearance of the upper mature band of FZD5, indicating clearance of the mature form from the plasma membrane. In contrast, the amount of the mature FZD5 was enhanced in the presence of dominant negative RNF43 mutants (Fig. 4J, lanes 8–12). The disappearance of the mature form of FZD5 in the cells expressing wild-type RNF43 or RNF43 with silent mutations S94I was rescued by treatment with bafilomycin, which blocks lysosomal degradation (Supplementary Fig. S4E). Taken together, these results are consistent with the model that RNF43 induces FZD endocytosis and subsequent lysosomal degradation, and the inactive RNF43 mutants fail to interact with and promote FZD internalization.
The dominant negative function of RNF43 is linked to dimer formation
We explored potential mechanisms for the dominant negative, hyperactivating nature of some RNF43 mutants. Similar to our observation, three disease-associated hyperactivating RNF43 mutations have been reported previously, but the underlying mechanisms are still unclear (24). We hypothesized that RNF43 may normally form a homodimer, or a heterodimer with ZNRF3, and that a mutant RNF43 in the complex might inhibit the function of the wild-type endogenous protein. RNF43 dimers have not been reported previously, but its paralog ZNRF3 has been observed to form dimers in the crystal structure (38). Importantly, homodimerization is required for the activity of multiple RING domain proteins (39, 40). To test whether RNF43 can form higher order complexes, we co-expressed RNF43 with two different tags, FLAG–RNF43 and RNF43-Myc. Co-immunoprecipitation experiments indeed revealed strong association of these two differently tagged RNF43 proteins, supporting the hypothesis of RNF43 homodimerization (Fig. 4K). Interestingly, after removing the intracellular region of RNF43, including the RING domain, the protein could still interact with itself (Supplementary Fig. S4F). Furthermore, both wild-type and mutant RNF43 could interact with ZNRF3 (Fig. 4L). This is consistent with a model where RNF43 normally dimerizes with itself or forms heterodimers with ZNRF3. Dominant negative mutants may then dimerize with wild-type protein forming an inactive complex, hence reducing the clearance of FZDs from the cell surface (Figs. 2 and 4; see model Fig. 4M).
RNF43 truncation and frameshift mutations, including G659fs, are LOF
We examined the activity of truncation and frameshift mutations in the N-terminal half of RNF43 protein that includes the known functional domains, including the hotspot G•C track around R117. Nearly all of the N-terminal truncation mutants were either LOF or hyperactivating (Fig. 5A). The RNF43 with frameshift V287Gfs*7 in the RING domain also significantly increased surface FZD abundance, consistent with its hyperactivating activity in the Wnt/β-catenin reporter assay (Fig. 5B).
We next tested the activity of 12 C-terminal–truncating mutations. In the RNF43 Wnt/β-catenin reporter assay, a number of these mutants had partial function, whereas others appeared to be wild-type (Supplementary Fig. S5A and S5B). R117fs and G659fs are the most frequent mutations in RNF43 in multiple cancer types as the G•C stretches near amino acids R117 and G659 are targets for indels. Indeed, the frequency of the G659fs truncation mutant was shown to be significantly greater than predicted by chance, even in MSI cancers, suggesting that they confer a growth or proliferation advantage to tumors harboring this mutation (4). In contrast, two recent publications using overexpression assays concluded that indels around G659 do not compromise RNF43 activity (21, 41).
To more closely examine whether G659fs and other C-terminal truncations are LOF, we first tested the effect of G659Vfs*41 expression on FZD cell surface abundance. Compared with the wild-type RNF43, the G659Vfs*41 mutant was modestly compromised in its ability to inhibit FZD abundance on the cell surface (Fig. 5C) showing partial function.
We tested whether the C-terminal truncations and frameshifts destabilized RNF43 protein. Because antibodies detecting endogenous RNF43 were not available, we introduced several patient-associated RNF43-truncating mutants into the FLAG–RNF43 construct. Indeed, we observed that although most of these constructs transcribed comparable amounts of mRNA (Supplementary Fig. S5C), the mutant proteins, including the recurrent G659Vfs*41, were expressed at lower levels than WT RNF43 (Fig. 5D). The low protein abundance might be due to enhanced proteasomal degradation of prematurely truncated and frameshifted proteins. In fact, treating the transfected cells with the proteasome inhibitor bortezomib markedly increased the protein levels of the mutant RNF43 proteins (Fig. 5E). Thus, RNF43 proteins that are prematurely truncated and/or frame-shifted may be less stable. Although this decreased protein abundance of C-terminal mutants should compromise their activity, we were unable to detect the LOF when overexpressing these mutants in the Wnt/β-catenin reporter assay (Supplementary Fig. S5A—S5B).
Editing the endogenous RNF43 locus reveals C-terminal truncations are LOF
To eliminate the role of exogenous expression of RNF43, we used CRISPR-Cas9 to perform genome editing of the endogenous RNF43 gene in Panc 08.13 cells that have no underlying mutation in the Wnt/β-catenin pathway. CRISPR-Cas9–mediated nonhomologous end joining leads to indels, resulting in frameshift mutations. The human RNF43 gene has 10 exons. Exons 7 and 8 encode the functional RING domain, whereas exon 9, the longest, encodes for almost half of the protein, including most of the C-terminus. As illustrated in Fig. 5F, we used multiple single-guide RNAs (sgRNA) to target specific regions of RNF43. We designed an sgRNA targeting the translation start site in exon 2 to serve as a positive control. To target the RING domain, we designed two sgRNAs targeting exon 7, sgE7_1 and sgE7_2, recognizing sequences upstream of and in the RING domain, respectively. Four sgRNAs were designed to target RNF43 Exon 9 to test the impact of truncating mutations at the C-terminus of the RNF43.
Panc 08.13 cells were transfected with a plasmid expressing both Cas9 and an sgRNA, and then selected with puromycin. The genome-editing efficiency of the sgRNAs in the cell pools was analyzed using the Surveyor nuclease assay (Supplementary Fig. S5D and S5E; Supplementary Table S3). All the specific sgRNAs produced the expected bands following nuclease S treatment (arrows), indicating genome editing at the target locus. sgE9_4 sgRNA showed the same banding pattern as the nontargeting control sgRNA and turned out to be a fortuitous additional negative control, as Panc 08.13 cells were unexpectedly homozygous for a P686R polymorphism (allele frequency 0.099), altering the PAM sequence to which sgE9_4 was targeted (Supplementary Fig. S5F).
We then examined the consequences of targeted editing of the genomic locus of RNF43. Cell surface FZD levels (Fig. 5G) were significantly increased in all the successfully edited cell pools. As a functional consequence of these increased FZD levels, the CRISPR-edited pools had enhanced basal Wnt/β-catenin reporter activity that was increased more than 2-fold in response to both single and synergistic Wnt ligands (Fig. 5H; Supplementary Fig. S5G; ref. 42). This genome-editing result stands in marked contrast with the results from transient overexpression using cDNA. We note that RNF43 has been reported to undergo nonsense-mediated decay (NMD) in HCT116 cells (43). However, consistent with what has been reported in other systems (21), we did not observe a significant reduction in the mRNA levels in any of the successfully edited cell pools (Supplementary Fig. S5H), suggesting that the level of RNF43 in these cells is not regulated by NMD. Taken together, these genome-editing studies indicate that frameshift and truncation mutations in the C-terminal half of endogenous RNF43 compromise its activity, and this is in part mediated by decreased protein stability.
Patient-derived xenografts with RNF43 C-terminal truncation mutations, including P660fs have high FZD abundance and are sensitive to PORCN inhibitor
Having established in Panc 08.13 cells that truncating mutations in endogenous RNF43 locus in the C-terminus are LOF and enhance sensitivity to Wnt ligands, we next examined pancreatic cancers bearing similar mutations. The pancreatic cancer cell lines, AsPC-1 has a distal C-terminal–truncating mutation in RNF43 S720X, whereas Capan-2 has RNF43 R330fs. Both AsPC-1 and Capan-2 cells are Wnt-addicted and sensitive to PORCN inhibitors in vivo or in soft agar assays (19, 29), supporting the conclusion that endogenous C-terminal truncations are LOF.
To further test whether chromosome-based truncating mutations in the RNF43 C-terminal region are indeed LOF, we examined additional pancreatic PDX models. We selected two models with distinct C-terminal frameshift mutations, PAXF 1861 with a heterozygous R371fs and PAXF 1869 with the recurrent P660fs mutation and loss of heterozygosity (Fig. 6A). We first assessed FZD abundance by IHC staining with a pan-FZD monoclonal antibody (Fig. 6B; refs. 21, 32). Xenograft tumors from Wnt-addicted RNF43-mutant cell lines HPAF-II (E174X) and AsPC-1 (S720X) that are sensitive to PORCN inhibitors had prominent FZD staining (19), whereas two negative controls, pancreatic PDXs PA3127 and PA1457 with wild-type RNF43 had minimal FZD staining (Fig. 6B and C). Importantly, PDXs PAXF 1861 and PAXF 1869 with RNF43-truncating mutations both had high FZD abundance with clear membrane staining. The observation that PAXF 1861, a PDX with a heterozygous-truncating mutant, had elevated cell surface FZD supports the model that inactive RNF43 mutants can also inactivate wild-type copies (Fig. 4M). These data further support the conclusion that endogenous C-terminal truncation mutations compromise the activity of RNF43, leading to increased FZD levels on the cell surface.
To directly examine whether these mutations are actionable, we tested the sensitivity of PDXs PAXF 1861 and PAXF 1869 to the PORCN inhibitor ETC-159. The growth and final weight of xenograft tumors in mice generated from both PAXF 1861 and PAXF 1869 was significantly reduced by ETC-159 treatment (Fig. 6D and E). Consistent with Wnt/β-catenin inhibition by ETC-159, expression of the Wnt target gene AXIN2 was significantly reduced in these tumors as well (Fig. 6F). Taken together, the data from CRISPR-Cas9–targeted Panc08.13 cells, the cancer cell lines AsPC-1 and Capan-2, and the PDXs PAXF 1861 and PAXF 1869 confirm that in vivo RNF43 C-terminal–truncating mutations are actionable LOF.
RNF43 is frequently mutated in diverse cancers, including colorectal, endometrial, mucinous ovarian, pancreatic, and gastric (4, 20, 23, 44). Wnt inhibition using upstream pathway inhibitors such as anti-FZD–blocking antibodies or PORCN inhibitors is effective in preclinical cancer models with RNF43 LOF mutations or RSPO translocations, and hence these agents have advanced to clinical trials. RNF43 is now included in sequencing panels designed to identify actionable mutations in patients with cancer. However, as clinical mutations in RNF43 span the entire length of the gene it is challenging to predict which are drivers and which are passengers. Here, we analyzed a broad range of patient-derived mutations to determine those that can be used to predict Wnt addiction.
Our study consistently found that all tested nonsense and frameshift mutations, as well as a large percentage of missense RNF43 mutations found in human cancers are either LOF or hyperactivating, leading to an increased cell surface abundance of FZDs and subsequently sensitivity to PORCN inhibition. In the absence of additional downstream Wnt pathway mutations or MSI status, these results can guide decisions as to which patients may benefit from upstream Wnt pathway inhibitors.
One unexpected finding was that a large fraction of missense mutants in the N-terminus of RNF43 acted as dominant negative mutations that hyperactivated Wnt/β-catenin signaling. These mutants failed to induce cell surface FZD receptor internalization and subsequent lysosome-mediated degradation. It was recently reported that the PA domain located in the N-terminal region of RNF43 is dispensable for its suppression of Wnt/β-catenin signaling (45). However, we find that several PA domain mutants compromised RNF43 activity. In addition, missense mutations are significantly enriched in the PA domain (Fig. 1C), suggesting a growth advantage for those tumors. Mechanistically, RNF43 can form higher-order complexes with itself and its paralog ZNRF3. We therefore propose that inclusion of a mutant in the heterodimer can render the wild-type copy inactive. This suggests that cancers with one mutant RNF43 allele, and/or tumors with mutant RNF43 that co-express ZNRF3, may still have enhanced FZD abundance at the cell surface, exhibit higher sensitivity to Wnts, and hence may benefit clinically from upstream Wnt pathway inhibitors.
We find that C-terminal–truncating mutations in RNF43 are also LOF. Importantly, we confirmed the deleterious consequences of C-terminal truncations in pancreatic cancer cell lines and in several pancreatic PDXs. We find that two xenografts with distinct C-terminal mutations, P660fs and R371fs, have increased cell surface FZD and are sensitive to therapeutic doses of PORCN inhibitors. This is consistent with published data that the Capan-2 cells with R330fs and AsPC-1 cell line with S720X are sensitive to PORCN inhibitors in xenograft models (19, 29). Our conclusion is also supported by genome editing of the endogenous RNF43 locus, as well as in vivo functional studies in PDX models bearing C-terminal truncation or frameshift mutants, including the G659fs mutation that is common in cancers due to its G•C track. The data indicate that C-terminal truncations predict response to PORCN inhibitors in the absence of microsatellite instability. Whether RNF43 mutations predict response to PORCN inhibitors in MSI tumors with multiple additional Wnt pathway and other mutations is not known.
Additional pathways to regulate RNF43 function may exist as well. In a recent study, another class of C-terminal truncation mutants falling in a small domain were identified that trap casein kinase 1 at the plasma membrane to regulate β-catenin signaling (36).
Why G659fs and other C-terminal truncations lose function in vivo but not in cell-based reporter assays is not clear. The results from HEK293 cell-based reporter assays were discordant with results from both our genome-editing data in Panc 08.13 cells and from functional studies in xenografts, suggesting that these more complex assays using chromosomal rather than plasmid-based mutations may measure additional relevant properties of the C-terminus of RNF43. Our in vitro assays using an approach similar to others agrees with their observations (21, 41). However, there are limitations of transient transfections, including the highly simplified promoter, the nature of the cell types used, and that the assays are performed in vitro whereas mutant tumors grow in a far more complex environment in vivo. Some differences observed in the CRISPR-Cas9–edited cells may be explainable by cell type, as in Li and colleagues (21), an MSI colorectal cell line KM12 with APC and AXIN1 mutations was used. These mutations activate Wnt signaling downstream of FZD. In contrast, Panc 08.13 cells have no downstream mutations in Wnt signaling components (46). Our in-depth in vivo analysis of various C-terminal–truncating mutations, including those in the most recurrently mutated region around G659, predicts them to be LOF and potentially actionable mutations.
This analysis provides a comprehensive evaluation of RNF43 as a biomarker for predicting Wnt addiction and hence sensitivity to PORCN inhibitors. We find that all truncation or frameshift mutants are LOF. Nearly all missense mutants in the RING domain and several missense mutations in the N-terminal region are also LOF. In the absence of additional downstream Wnt pathway mutations, these mutations predict actionable RNF43 mutants. The finding that simple transfection assays fail to identify a number of in vivo LOF mutants suggests that our results are actually an underestimate of the number of such mutations. Increased membrane FZD abundance in RNF43 mutant cancers has potential to identify these additional actionable tumors. Developing FZD immunohistochemical assays might further assist in selecting patients for treatment with PORCN inhibitors.
D.M. Epstein reports being the founder, President, and CEO of Black Diamond Therapeutics, an oncology precision medicine company developing therapeutics against oncogenic kinases. D.M. Virshup reports consulting fees from Experimental Drug Development Center, ASTAR outside the submitted work, as well as a financial interest in the drug ETC-159 used in this study through Duke-NUS. B. Madan reports grants from The Singapore Ministry of Health's National Medical Research Council during the conduct of the study, as well as a patent for Wnt pathway modulators WO/2014/175832 issued. No disclosures were reported by the other authors.
J. Yu: Investigation, writing-original draft, writing-review and editing. P.A.M. Yusoff: Investigation. D.T.J. Woutersen: Investigation. P. Goh: Investigation. N. Harmston: Formal analysis, investigation. R. Smits: Conceptualization, supervision, funding acquisition. D.M. Epstein: Conceptualization, supervision, funding acquisition. D.M. Virshup: Conceptualization, supervision, funding acquisition, writing-review and editing. B. Madan: Conceptualization, supervision, funding acquisition, writing-original draft, writing-review and editing.
We gratefully acknowledge the assistance of Jamal Aliyev and other past and present members of the Virshup and Epstein laboratories. We thank Dr. Madelon Maurice for the SNAP–FZD5 plasmid and Dr. Jeffery Rubin for the HA–FZD constructs. We thank Drs. Stephane Angers, Sachdev Sidhu, and University of Toronto for the pan-FZD antibodies. We acknowledge the Advanced Bioimaging facility of Singhealth/DukeNUS for assistance with the generation of imaging data. This research is supported in part by the National Research Foundation Singapore and administered by the Singapore Ministry of Health's National Medical Research Council under the STAR Award Program MOH-000155 (to D.M. Virshup). B. Madan acknowledges the support of the Singapore Ministry of Health's National Medical Research Council Open Fund—Independent Research grant NMRC/OFIRG/0055/2017.
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.