A Small-Molecule Antagonist of the β-Catenin/TCF4 Interaction Blocks the Self-Renewal of Cancer Stem Cells and Suppresses Tumorigenesis

Great efforts have been made over the past two decades to develop rational targeted cancer therapies, which are more effective than chemo- and radiotherapies and may cause fewer side effects. Successes involve drugs that act as tyrosine kinase inhibitors; to date, 28 such substances have been approved for clinical use (1). Drugs that interfere with other fundamental signaling systems, such as γ-secretase inhibitors that target Notch signals (2) or inhibitors that target Sonic hedgehog (3), have been developed and are being evaluated in preclinical and clinical trials. Other signaling systems whose deregulation is essential in the initiation and maintenance of cancers—such as Wnt/β-catenin signaling—need to be identified and targeted as means of expanding the current repertoire of antitumor strategies (4).

Wnt/β-catenin signaling is a system that is essential for embryonic development and tissue homeostasis that has been highly conserved throughout evolution (5–7). The key mediator of the pathway is β-catenin, a transcriptional coactivator whose level is tightly controlled by a multiprotein destruction complex in the cytoplasm. This complex is composed of Axin, adenomatous polyposis coli (APC), casein kinase 1α (CK1α), glycogen synthase kinase 3β (GSK3β), and β-transducin repeats containing protein (β-TrCP). The latter phosphorylate β-catenin and induce its ubiquitination and proteosomal degradation. Canonical Wnt ligands, such as Wnt1 or Wnt3a, stimulate the pathway by inhibiting the destruction complex; this results in the stabilization and nuclear translocation of β-catenin. β-Catenin interacts with members of the TCF/LEF family of transcription factors and recruits a number of coactivators including B-cell lymphoma 9 (BCL9) and CREB-binding protein (CBP), ensuring efficient activation of Wnt target genes. A comprehensive list of Wnt target genes can be found at web.stanford.edu/group/nusselab/cgi-bin/wnt/.

The discovery of an association between β-catenin and the tumor suppressor gene APC led to the recognition of the essential role of Wnt/β-catenin signaling in various cancers (8–11). Aberrant Wnt signaling is frequently caused by mutations in APC, β-catenin, and Axin2. A noted exception is melanoma, where activated Wnt/β-catenin signaling appears to be associated with reduced tumor formation (12). Deregulated Wnt signaling is not only important for cancer initiation, but is also crucial in late-stage cancers and in metastasis formation (13, 14). Recent studies have shown that cancer stem cells (CSC) in various tumors exhibit deregulated Wnt/β-catenin signaling (15–17) and require higher Wnt activity than differentiated cancer cells to maintain their self-renewal and tumorigenic properties. This feature may be useful in specifically targeting CSCs, a crucial factor in combatting cancer recurrence (18, 19).

Over the past decade, researchers in academia and industry have developed various strategies to identify small molecules that target components involved in Wnt signaling. Wnt inhibitors were initially identified through ELISA-based high-throughput screening (HTS) of natural compounds that interfere with β-catenin/TCF4 interactions (20). Cell-based HTS has been applied to identify additional small-molecule inhibitors of the pathway, some of which target Wnt secretion (the porcupine inhibitor IWP-2; ref. 21) or the Wnt receptor complex (Dishevelled/Frizzled interaction inhibitors NSC668036 and FJ9; refs. 22, 23). Other compounds have been found that enhance the stability of the β-catenin destruction complex (tankyrase inhibitors XAV939 and JW55; refs. 24, 25) or disrupt the transcription complex (β-catenin/CBP interaction inhibitor ICG-001, β-catenin/TCF interaction inhibitors iCRT3 and CWP232228, and β-catenin/BCL9 interaction inhibitor carnosic acid; refs. 26–29). These studies allowed the further development of small-molecule Wnt inhibitors that entered early clinical trials, including PRI-724, CWP232291, and LGK974 (30).

In human cancer, activating mutations of Wnt signaling components are frequently found at the level of the destruction or transcription complexes. We reasoned that efforts to develop new Wnt inhibitors could therefore be directed to the downstream components of the pathway, in particular to the β-catenin–TCF transcription complexes. This may require interference with protein–protein interactions, which is a challenging field of research (31). A further concern was the overlap of the binding sites of β-catenin with TCFs and E-cadherin, as seen in the structural analyses (32). Substances interfering with E-cadherin interaction could thus have opposing effects than interfering with Wnt signaling. We should however note that because of the much larger β-catenin/E-cadherin interaction surface, the K_d of this interaction can be 1,000-fold lower than with the TCFs (33). β-Catenin/TCF4 interaction inhibitors have been recently described (27, 29), suggesting the great potential of such inhibitors to be developed for clinical use.

Here, we took advantage of the AlphaScreen technology to perform HTS to identify small molecules that interfere with the interaction of β-catenin and TCF4. The compound LF3 was identified as a potent and specific inhibitor of activated Wnt/β-catenin signals in reporter cells, but it did not interfere with the interaction of β-catenin with E-cadherin. LF3 blocked properties related to Wnt addiction in colon cancer cells, including enhanced cell migration and the progression of the cell cycle, and interfered with the self-renewal of CSCs. In a xenograft tumor model derived from colon CSCs, LF3 did not only reduce tumor growth but also strongly induced tumor differentiation. We thus report on the discovery of a small-molecule Wnt inhibitor with promising chemical and biologic properties, with a potential for further development into a lead compound for targeted therapies of Wnt-addicted tumors.

The cell lines HCT116, HCT15, HT29, SW480, SW620, LS174T, SW48, MCF7, HeLa, HEK293, and MDCK were obtained from the ATCC and cultured in DMEM (Invitrogen) supplemented with 10% FBS, penicillin, and streptomycin. Mouse salivary gland CSCs were isolated and cultured in DMEM/F12 (Invitrogen) supplemented with 20% KSR, nonessential minimal amino acids, penicillin, streptomycin, l-glutamine, and β-mercaptoethanol. The cell identity was tested by PCR phenotyping for β-catenin and Bmpr1a (16).

The cDNA encoding residues 134–668 of human β-catenin were inserted into the pGEX-4-T1 vector (GE Healthcare), and the cDNA encoding residues 1–79 of human TCF4 were inserted into the pET-16b vector (Qiagen). A mutation of residue Asp16 of TCF4 was introduced by PCR mutagenesis with the primers: forward 5′-GACCTAGGCGCCAACGCCGAACTGATTTCCTTCA-3′ and reverse 5′-TGAAGGAAATCAGTTCGGCGTTGGCGCCTAGGTC-3′. Recombinant fusion proteins were expressed in E. coli BL21 cells and purified through GSTrap FF or HisTrap FF columns (GE; details in Supplementary Materials and Methods). Biotinylated human TCF4 peptide (aa 8–53) was purchased from Biosyntan.

A library containing 16,671 World Drug Index–derived compounds was provided by ChemBioNet, designed by the drug design and modeling group of the Leibniz-Institut für Molekulare Pharmakologie, Berlin, Germany (34). Analogues of LF3 were purchased from ChemDiv and ChemBridge and dissolved in DMSO to a concentration of 50 mmol/L. In the AlphaScreens, purified GST-hβ-catenin was first incubated with nickel chelate acceptor beads and His-hTCF4 incubated with glutathione donor beads (Perkinelmer) in PBS. They were then combined into 384-well AlphaPlates in the presence of 20 μmol/L test compounds. Titration was used to determine the concentration of purified proteins used in the AlphaScreens, yielding signals in linear ranges. After 1-hour incubation, protein–protein interactions induced luminescence, as measured by an EnVision multilabel plate reader (PerkinElmer).

Twenty-four–well cell culture plates were precoated with 250 μL polyhema (12 mg/mL in 95% ethanol; Sigma) to promote the growth of nonattached spheres. SW480 reporter cells and mouse salivary gland CSCs were trypsinized and seeded as single cells into the sphere culture medium (F12:DMEM 1:1, 1× B-27 supplement, 20 ng/mL EGF, 20 ng/mL FGF, 0.5% methylcellulose) with or without treatment (16, 19). After 10 days, spheres were counted under a phase contrast microscope, and pictures were taken by AF6000 and DFC350FX (Leica).

Unsorted GFP^low and GFP^high SW480 cells (1 × 10⁴) were subcutaneously injected into the back skin of NOD/SCID mice. Tumor growth was monitored over a period of 45 days. For therapy, LF3 was administered i.v. at 50 mg/kg body weight for three rounds over 5 consecutive days, with 2-day breaks.

After desired treatments, total RNA of salivary gland CSCs was extracted according to the standard TRIzol protocol (Invitrogen). RNA was purified with the RNeasy Kit (Qiagen) and prepared with the TotalPrep RNA amplification Kit (Illumina) for the microarray analysis using the MouseRef-8 v2.0 expression beadchip Kit. Microarray data (GEO accession: GSE73732) were analyzed and exported with the software Genome Studio (Illumina). Pearson correlation was calculated with http://www.wessa.net/rwasp_correlation.wasp. Expression clustering was calculated and visualized with software Cluster and TreeView (Stanford University), respectively. Functional clustering of interesting genes was performed with http://david.abcc.ncifcrf.gov/.

The techniques ELISA, coimmunoprecipitation, Western blotting, the establishment of TOP-GFP reporter cell lines, Luciferase reporter assays, immunofluorescence, chromatin immunoprecipitation and gene transcription analyses with qRT-PCR, cell migration, proliferation with BrdUrd, cell death by apoptosis, cell cycle, and cell surface marker analyses are described in Supplementary Materials and Methods.

We performed AlphaScreen-based HTS (Supplementary Fig. S1A; ref. 35) with a library of 16,000 synthetic compounds (see Supplementary Fig. S1C) from the central open access technology platform of ChemBioNet (34), using a GST-tagged armadillo repeat domain of human β-catenin (GST-β-catenin, aa residues 134-668) and a His-tagged N-terminal region of human TCF4 (His-hTCF4, aa 1–79; Fig. 1A; refs. 32, 36, 37; see Materials and Methods). The interaction between recombinant β-catenin and TCF4 was specific: in ELISA (Supplementary Fig. S1B), binding could be inhibited by recombinant TCF4 (His-hTCF4) and an N-terminal peptide of TCF4 (Bio-TCFpeptide, aa 8–53), but not by TCF4 mutated at critical Asp16 [His-TCF4(D16A); refs. 32, 36, 37] or irrelevant proteins (Supplementary Fig. S1C). From the primary AlphaScreens, compounds that inhibited interactions by at least 40% at 20 μmol/L were selected, and false positives (which also appeared in nonrelated screens) were excluded (see the course of the screening procedure in Supplementary Fig. S1D). A total of 132 compounds were re-examined using both AlphaScreens and ELISAs, and finally 5 compounds were identified (termed LF1 to LF5) that inhibited interactions by 50% at concentrations below 10 μmol/L in both assays (Supplementary Fig. S1E and S1F). Due to its excellent chemical properties and expected plasma membrane permeability (predicted by Lipinski's rule; ref. 38), compound LF3 (Fig. 1B, structure on the left) with an IC₅₀ of less than 2 μmol/L in both assays (Fig. 1B, right) was chosen for further studies.

To clarify the specific structural requirements of LF3, analogues were purchased and tested by ELISA. Deletion or replacement of the sulfonamide group (R1) by methyl, fluorine, or nitro eliminated inhibitory activity (Fig. 1C). Deletion of the benzene tail (R2) strongly reduced activity (Fig. 1D). The distance between the benzene ring and core structure was also important; among the analogues with 2 or fewer methyl groups between the benzene and the core, only LF3h exhibited inhibitory activity as potent as that of LF3 in these biochemical experiments (Fig. 1D).

To investigate the effects of LF3 on the interactions of endogenous cellular proteins, protein extracts from HCT116 cells were incubated with LF3 at 3.3 to 60 μmol/L, and β-catenin together with its binding partners were pulled down by coimmunoprecipitation using specific anti–β-catenin antibodies. LF3 interfered with both β-catenin/TCF4 and β-catenin/LEF1 interactions in dose-dependent manners, as shown by Western blotting (Fig. 2A, quantification in B).

To evaluate biologic activity, LF3 and analogues were examined for inhibition of Wnt/β-catenin signaling in stable HEK293 reporter cells that express GFP in a TCF/LEF-dependent fashion (TOP-GFP; SABiosciences). LF3 was the most potent inhibitor of TOP-GFP production with an IC₅₀ of 22.2 ± 4.9 μmol/L, when cells were stimulated with 3 μmol/L of CHIR99021 (39), which is a GSK3β inhibitor and thus a Wnt activator (Fig. 2C and D). LF3h was less potent and was therefore not studied further. When compared with ICG-001 and XAV939, which are Wnt inhibitors that suppress Wnt signaling at the levels of nuclear transcription and β-catenin destruction, respectively (scheme in Supplementary Fig. S2A; refs. 25, 30), LF3 inhibited TOP-GFP signals with kinetics and at levels comparable with ICG-001, but different from XAV939 (Supplementary Fig. S2B). Conventional TOPflash assays in HeLa cells showed that LF3 inhibited luciferase signals with IC₅₀ between 2.4 and 4.0 μmol/L, which were stimulated by recombinant Wnt3a, CHIR99021, or overexpression of an activating β-catenin mutant (with four mutated N-terminal phosphorylation sites), but not by overexpression of a β-catenin/LEF1-HMG fusion protein (40), which has no interface for LF3 binding (Fig. 2E). LF3 did not significantly inhibit other reporters, i.e., Oct-Luciferase stimulated by Oct4, NF-κB-Luciferase by TNFα, or Notch-Luciferase by NICD (SABiosciences; Fig. 2F; Supplementary Fig. S2C). These data indicate that the inhibition of canonical Wnt signaling by LF3 in cells is specific and may indeed occur downstream in the pathway, i.e., at the β-catenin/TCF4 level.

Both TCF4 and E-cadherin interact with β-catenin through common armadillo repeats (32, 41, 42); this was a major concern as we developed our inhibitor, because a loss of function of β-catenin in adherens junctions might increase cell migration and induce metastasis (43). We therefore examined whether LF3 might also disrupt the interaction between β-catenin and E-cadherin. Immunofluorescence staining of MDCK cells revealed that E-cadherin and β-catenin colocalized at cell–cell adhesion sites and showed merged cobblestone patterns (Supplementary Fig. S2D; top; Supplementary Fig. S2H, left). This location was not altered by LF3 treatment, for instance at 60 μmol/L, which however strongly disturbed Wnt/β-catenin signaling at this concentration (Supplementary Fig. S2D, middle; see also above). In contrast, treatment of cells with HGF at 40 U/mL leads to scattering of cells, to the loss of cell–cell adhesions, and to a relocation of E-cadherin and β-catenin to nonoverlapping cytoplasmic sites (Supplementary Fig. S2D, bottom). When β-catenin was immunoprecipitated from protein extracts of HCT116 cells incubated with increasing concentrations of LF3 at 3.3 to 60 μmol/L, the amount of E-cadherin bound to β-catenin remained unchanged (Supplementary Fig. S2E, quantification in F). MDCK cells were also incubated with LF3 and subjected to trypsin treatment for 5 minutes, which largely digested E-cadherin and abrogated E-cadherin–mediated cell–cell adhesion (Supplementary Fig. S2G and Supplementary Fig. S2H; 2nd plot). After removing the trypsin, cells were allowed to reform E-cadherin/β-catenin complexes at cell–cell adhesions: in the presence of LF3, the time course of recovery of E-cadherin/β-catenin–mediated cell–cell adhesion was as in controls (Supplementary Fig. S2H, 3rd–5th plots). Overall, these data show that LF3 inhibits Wnt/β-catenin signaling, but does not interfere with E-cadherin/β-catenin–mediated cell–cell adhesion.

In colon cancer cells that harbor Wnt/β-catenin–activating mutations, the Wnt target genes Axin2 and c-Myc are actively transcribed upon binding of β-catenin/TCF4 to their promoter regions. Sequences near the promoters of Axin2 and c-Myc were enriched by anti–β-catenin chromatin immunoprecipitation from HCT116 cells, and the enrichments were significantly reduced by LF3 at 30 μmol/L (Fig. 3A). As shown by qRT-PCR, LF3 at 10 to 60 μmol/L also inhibited the expression of many Wnt target genes (Bmp4, Axin2, survivin, Bambi, and c-Myc) in a dosage-dependent manner (Fig. 3B); ICG-001 at 25 μmol/L or siRNA treatment against β-catenin had similar effects. The production of Axin2, c-Myc, and cyclinD1 proteins was also suppressed by LF3 at 30 μmol/L in a panel of colon cancer cell lines (Fig. 3C–F; Supplementary Fig. S3A–S3H). Thus, LF3 blocks the expression of a series of Wnt target genes in Wnt-addicted colon cancer cells.

Wnt/β-catenin signaling contributes to high cell motility, high cell proliferation, and cell-cycle progression in colon cancer cells. Boyden chamber motility assays with SW480 colon cancer cells showed that LF3 at 10 and 30 μmol/L strongly reduced cell migration in a dosage-dependent manner (Fig. 4A). LF3 at 30 and 60 μmol/L also reduced proliferation, as measured by BrdUrd incorporation, in the Wnt-dependent colon cancer cell lines HCT116 and HT29, but not in the Wnt-independent breast cancer cell line MCF7 (Fig. 4B). Cleaved caspase 3 and propidium iodide staining indicated that the inhibition of proliferation, for instance in HCT116 cells, was not due to apoptotic or cytotoxic effects of LF3 at 60 μmol/L, which are induced in cells by puromycin at 1 μg/mL (Supplementary Fig. S4A and S4B). Instead, we observed that LF3 at 30 and 60 μmol/L induced cell-cycle arrest in the G₁ phase of HCT116 and HT29 cells in dosage-dependent manners (Fig. 4C), but had only an insignificant influence on the cell cycle of MCF7 cells. These data show that LF3 inhibits proliferation of Wnt-addicted colon cancer cells through induction of cell-cycle arrest.

It has been reported that high Wnt/β-catenin activity characterizes colon CSCs and other CSCs, which are fractions of tumor cells capable of regenerating entire tumors following transplantation (18, 19, 44). A small-molecule Wnt inhibitor CWP232228 has recently been shown to be able to inhibit the proliferation of breast CSCs (29). Stem and CSCs are generally enriched in nonattached sphere cultures, which reflect their self-renewal capacity (45, 46). We asked whether LF3 could specifically target colon and salivary gland CSCs through Wnt/β-catenin inhibition. A reporter cell line that expressed TOP-GFP (see above) was generated from SW480 colon cancer cells. Flow cytometry analysis revealed heterogeneous expression of GFP in this cell line (Fig. 5A). GFP intensity correlated with Wnt/β-catenin activity, as shown by upregulated mRNA and protein levels of Wnt target genes (Bmp4, Lgr5, vimentin, Axin2, Bambi, and Lef1) in the GFP^high2 population, compared with the GFP^low2 population (Fig. 5B and C). Immunofluorescence revealed intense nuclear β-catenin staining in the GFP^high2 population, whereas E-cadherin staining at cell borders was strongly reduced (Fig. 5D, bottom). In GFP^low2 cells, E-cadherin was homogeneously high and β-catenin was low, whereas unsorted cells showed separated subclones of the different cell subpopulations (Fig. 5D, middle and top plots, respectively). The expression of the human stem cell markers CD44 and CD29 also correlated with Wnt/β-catenin activity; the fraction of CD44^highCD29⁺ cells increased from 20% in the GFP^low1 population to 91% in the GFP^high2 population (Supplementary Fig. S5A and S5B). Cells from different subpopulations were then cultured under anchorage-independent conditions in serum-free medium to test their sphere-forming ability in the absence or presence of LF3 at 10 and 30 μmol/L. All four cell populations formed spheres, but GFP^high populations formed significantly more spheres than GFP^low populations (Fig. 5E and F), expressed TOP-GFP, and they were larger (Fig. 5E, 3rd plot). Remarkably, treatment with LF3 reduced the number and size of spheres formed from GFP^high populations, but had no significant impact on GFP^low cells (Fig. 5E and F). In sequential rounds of colony assays, GFP^high cells produced similarly high number of spheres in each passage, but in the presence of LF3, sphere numbers were reduced continually (Supplementary Fig. S5C).

We had found previously that salivary gland squamous cell carcinomas were generated in mice that harbor β-catenin gain-of-function (β-cat^GOF) and Bmp receptor 1a loss-of-function (Bmpr1a^LOF) mutations driven by keratin14-Cre recombinase (Fig. 6A; ref. 16). A CD24^highCD29⁺ cell population (Fig. 6B, encircled in red in the left plot and following purification in the right plot) was characterized as salivary gland CSCs by their potent sphere-forming capacity in culture and their high tumorigenic potential following transplantation (16). Remarkably, the expression of stem cell–associated, chromatin modifier and Wnt target genes that are highly expressed in the salivary gland CSCs were downregulated by LF3 at 30 μmol/L, by ICG-001 at 25 μmol/L, by the recently identified new Wnt inhibitor iCRT3 (27) at 60 μmol/L, and by β-catenin siRNA (Fig. 6C). LF3 also inhibited proliferation and sphere formation of these cells in dosage-dependent manners, as did ICG-001 (Fig. 6D–F). Similarly, treatment with LF3 or ICG-001 induced differentiation of salivary gland CSCs, as indicated by upregulated expression of Amylase1, CA6, and Loricrin, and formation of hollow spheres in three-dimensional matrigel culture (Fig. 6G; Supplementary Fig. S6A and S6B; refs. 16, 17).

We also investigated Wnt/β-catenin target genes using gene expression profiling in salivary gland CSCs treated with LF3 or ICG-001, because these inhibitors affect canonical Wnt signaling through different mechanisms, e.g., by inhibition of β-catenin-TCF/LEF1 and β-catenin-CBP interactions, respectively, in this study (30). Several canonical Wnt target genes, including vimentin, Ephb2, Ash2l, Birc5, and Id2, were downregulated by both LF3 and ICG-001 (Supplementary Fig. S6C). Overall, a majority of 325 genes (72.5%) were downregulated by both LF3 and ICG-001 (Supplementary Fig. S6D), giving a high Pearson correlation coefficient of 0.73 (Supplementary Fig. S6E). Gene clustering identified genes that were similarly or differentially regulated by LF3 and ICG-001 (Supplementary Fig. S6F). Then a functional clustering of genes from the different groups was performed using DAVID bioinformatics resources. Genes that are involved in essential cellular processes, such as cell-cycle progression, macromolecular metabolism, transcription regulation, and others, were similarly regulated by LF3 and ICG-001 (Supplementary Fig. S6F). Two clusters of genes that contribute to metal ion binding and purine nucleotide binding were differentially regulated by the two inhibitors, respectively (Supplementary Table S3).

To examine the effects of LF3 on tumor formation, GFP^high and GFP^low SW480 cells (see Fig. 5A) were injected subcutaneously into the back skin of NOD/SCID mice (1 × 10⁴ cells per mouse). Within 40 days, all mice injected with GFP^high cells developed tumors, whereas only 1 of 5 mice injected with GFP^low cells developed a tumor (Fig. 7A). When mice with GFP^high cells were treated with LF3 at 50 mg/kg through systemic i.v. injection for 3 weeks before tumor formation, tumor growth was significantly reduced (Fig. 7B; Supplementary Fig. S7A). Inhibition was also observed, when treatment was started after tumors were palpable (Supplementary Fig. S7B). No systemic toxicity, e.g., by weight loss, was observed in the treated mice (Supplementary Fig. S7C). Immunofluorescence showed that tumors induced by GFP^high cells exhibited undifferentiated features, including high expression of Lef-1 or vimentin, nuclear location of β-catenin, and loss of E-cadherin, keratin20, and Muc2 (Fig. 7C; Supplementary Fig. S7D, top). Remarkably, LF3-treated tumors became highly differentiated, as shown by expression of high E-cadherin, keratin20, and Muc2, by membrane localization of β-catenin and by downregulation of Lef-1 and vimentin (Fig. 7C; Supplementary Fig. S7D, bottom, quantified in Fig. 7D; Supplementary Fig. S7E). LF3 treatment did not disturb the normal histology of the gut of mice (Supplementary Fig. S7F). These data show that LF3 has potent tumor inhibition and differentiation capacities in xenotransplanted tumors.

Wnt signaling's crucial role in cancer biology has made it an important target to pursue in the development of new therapeutic strategies. So far, however, few rationally designed Wnt inhibitors have entered clinical trials, and of those being tried, some target Wnt ligand secretion or receptor binding (reviewed in ref. 30). In the present study, we focused on inhibiting the interaction between β-catenin and TCF4, with the goal of overcoming deregulations of Wnt signaling caused by mutations of components that lie upstream in the pathway. The small-molecule LF3 was identified in a biochemical screen using purified recombinant interaction partners. By specifically inhibiting the β-catenin/TCF4 interaction, LF3 diminished Wnt-dependent biologic characteristics of colon cancer cells, including high proliferation and high motility. Remarkably, LF3 also inhibited the self-renewal capacity of human colon and mouse salivary gland cancer cells and induced their differentiation.

By comparing gene expression profiles of mouse salivary gland CSCs treated with LF3 or ICG-001, two compounds that target the transcription complexes at different positions, we observed a large overlap in dis-regulated genes, including well-known Wnt targets. This was surprising given the fact that the ICG-001 target CBP is known to interact with numerous transcription factors and can thus regulate many signaling systems (47). LF3 and ICG-001 apparently inhibit deregulated Wnt signaling with a similar overall outcome. Further investigations are needed to understand whether the small sets of genes regulated by either TCF4 or CBP are critical for specific CSCs and tumors. Moreover, we show that the potency of LF3 is comparable with ICG-001 or iCRT3. LF3 at 50 mg/kg significantly reduced the growth of colon CSCs during tumor formation while exhibiting no significant toxicity for mice. These results suggest that LF3 has the potential for further development with the aim of preclinical and clinical trials.

Drug development, especially downstream in Wnt signaling, faces the challenge of finding inhibitors that target specific protein–protein interactions. Yet there is increasing evidence that small molecules that effectively dock onto hotspots of the interaction sites between partners and thus inhibit their interactions can indeed be found (reviewed in ref. 48). Such hotspots have been reported for the β-catenin/TCF4 interaction (32, 37). Efforts have been made to identify such inhibitors using, for instance, the compounds PKF115-584, CGP049090, PNU-74654, and iCRTs (20, 27, 49). In the present study, advanced biochemical screening techniques and the use of in-house synthetic compound libraries allowed us to identify compound LF3. The mechanistic details of the way LF3 disrupts β-catenin/TCF4 interaction remain to be shown; a strong hypothesis is that the negatively charged sulfonamide group of LF3 may compete with Asp16 of TCF4 to bind to the positively charged pocket of β-catenin formed by Lys435. The hydrophobic tail of LF3 might insert itself into a hydrophobic cleft lined by the residues Cys466 and Pro463 of β-catenin to facilitate interaction (32, 37). Solving the structure of LF3 with the partner proteins will help answer these questions and will also permit medicinal chemistry approaches to be applied to improve the activity of the compound.

CSCs are implicated in therapy resistance, metastasis formation, and cancer relapse (50). In various cancers, CSCs are maintained by single signaling systems or by combinations of systems. If these are inhibited, tumors may be forced to differentiate, as we have shown here in transplanted colon cancer cells. By inhibiting Wnt signaling, LF3 blocks the self-renewal of Wnt-high CSCs and induces their differentiation both in vitro and in xeno-transplants. Remarkably, the population of Wnt-low cells in SW480 cells did not respond to LF3 treatment. These cells appear to be supported by other signaling systems and driver mutations. Tumors that exhibit complex cellular heterogeneity may include different CSC populations supported by different signaling systems, and thus a combination of targeted therapies may be required to treat these cases in a personalized way.

No potential conflicts of interest were disclosed.

Conception and design: L. Fang, W.I. Weis, W. Birchmeier

Development of methodology: L. Fang, Q. Zhu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Fang, Q. Zhu, A. Wulf-Goldenberg, J.P. von Kries, W. Birchmeier

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Fang, Q. Zhu, M. Neuenschwander, E. Specker, J.P. von Kries, W. Birchmeier

Writing, review, and/or revision of the manuscript: L. Fang, Q. Zhu, W.I. Weis, W. Birchmeier

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Fang, W. Birchmeier

Study supervision: W. Birchmeier

Other (support in HTS with about 35,000 drug like compounds for inhibitors of the protein interaction of b-Catenin with TCF4): J.P. von Kries

This study was supported by the German Research Foundation (DFG), grant FG806.

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.