Discovery of Novel Inhibitor for WNT/β-Catenin Pathway by Tankyrase 1/2 Structure-Based Virtual Screening

Aberrant activation of the WNT/β-catenin signaling pathway is implicated in various types of cancers. Inhibitors targeting the Wnt signaling pathway are intensively studied in the current cancer research field, the outcomes of which remain to be determined. In this study, we have attempted to discover novel potent WNT/β-catenin pathway inhibitors through tankyrase 1/2 structure-based virtual screening. After screening more than 13.4 million compounds through molecular docking, we experimentally verified one compound, LZZ-02, as the most potent inhibitor out of 11 structurally representative top hits. LiCl-induced HEK293 cells containing TOPFlash reporter showed that LZZ-02 inhibited the transcriptional activity of β-catenin with an IC50 of 10 ± 1.2 μM. Mechanistically, LZZ-02 degrades the expression of β-catenin by stabilizing axin 2, thereby diminishing downstream proteins levels, including c-Myc and cyclin D1. LZZ-02 also inhibits the growth of colonic carcinoma cell harboring constitutively active β-catenin. More importantly, LZZ-02 effectively shrinks tumor xenograft derived from colonic cell lines. Our study successfully identified a novel tankyrase 1/2 inhibitor and shed light on a novel strategy for developing inhibitors targeting the WNT/β-catenin signaling axis.


Introduction
The highly conserved WNT/β-catenin signaling pathway plays pivotal roles in the context of development, differentiation and cellular self-renewal [1][2][3]. Not surprisingly, aberrant activation of WNT/β-catenin signaling has been implicated in various types of cancers [4,5]. Therefore, the identification of a novel inhibitor targeting the WNT/β-catenin pathway for cancer treatment is imperatively needed in the current onco-clinic. However, the limited numbers of identified inhibitors targeting the Wnt/β-catenin signaling pathway restrain the scope by which the proteins of the Wnt/β-catenin signaling pathway could be targeted.
Being important activators of the WNT/β-catenin pathway in colon cancer, tankyrase is a hot topic for drug development and promising clinical data have been reported for several tankyrase-targeting reagents as anti-cancer drugs [6,7]. The abnormal activation of tankyrase stabilized cellular β-catenin through the degradation of axin proteins are the crucial components of the β-catenin destruction complex [8,9]. Tankyrase belong to the family of proteins known as poly (ADP)-ribose polymerases XAV939 revealed that the tankyrase inhibitor interacts with the NAD + binding groove of the catalytic domain [49].
We retrieved crystal structures of TNKS-1 (PDB: 2RF5) and TNKS-2 (PDB: 3KR8). The co-crystallized inhibitor XAV939 occupies the whole nicotinamide binding region of TNKS-2, which was referenced to construct the grids for docking screening (Figure 1). TNKS-1 displayed a similar substrate-binding and overall 3D structure to TNKS-2. Our simulation revealed that it has the same targeting region as TNKS-2 ( Figure 1A). Prior to screening the ZINC database, evaluation of the accuracy of the docking programs, DOCK6.5 and Autodock4.2, was necessary. The re-docking test is an experimental method to evaluate the reproducibility of a complex structure by docking calculation and it has always been used to evaluate the performance of a docking program [50][51][52]. The docking programs were considered to qualify for virtual screening if the root mean square deviation (RMSD) for the ligand between the docked conformation and the crystallographic conformation was < 2 Å [53,54]. Herein, the TNKS-2 protein bound original ligand XAV939 was re-docked into the binding pocket by DOCK6.5 or Autodock4.2, the RMSD of the ligand between obtained the docked pose. We found that the RMSD was only 0.2 Å. Moreover, similar patterns were confirmed with three known TNKS-2 inhibitors when docked into the same area by DOCK6.5 and Autodock4.2, respectively ( Figure 1C). Taken together with the above data, it showed that DOCK6.5 and Autodock4.2 were suitable for our virtual screening.
The flexible docking program could improve the accuracy of screening due to its capacity to more closely mimic the protein in nonrigid residues, although at the price of longer computing time. The co-crystal structures of TNKS-2 with XAV939 revealed that some residues are important for inhibitors to occupy the catalyzing site: the Gly1032 and Ser1068 form hydrogen bonds with XAV939; Tyr1071 also shown a π-π stacking interaction with it [39]. Therefore, we selected residues around XAV939 including Tyr1071, Ser1068, Gly1032, Phe1061, Tyr1050, Tyr1071, Pro1034, Phe1035 and Ile1075 as flexibility residues in our flexible docking screening ( Figure 1D).
Superimposing known inhibitors has been reported to serve as a screening criterion in the virtual screening process to improve accuracy and effectively avoid false positive compounds. We selected three of the experimentally validated TNKS-1/2 inhibitors reported by Huang et al. [23] (XAV939, ABT-888 and LDW643) as the reference compounds in our screening (Table S1). IWR-1-endo and IWR-1-exo were excluded due to the fact that they bound in a site other than the catalytic domain [55].

Discovery of Candidate Compounds by Virtual Screening
Following the above process, we screened the drug-like and natural product databases in ZINC (http://zinc.docking.org/) of more than 13.4 million compounds through three steps in silico ( Figure 2). Prior to each virtual screening, the three reference compounds were used to test whether the dock and score system gave results that were consistent with the biological assay. We only collected the compounds with scores better than LDW643.
Given the volume of compounds in the database, the rapid docking program and simple scoring evaluation for efficiency were the chief goal in the primary screening. Since DOCK6.5 is fast for matching compounds to a specific site on target protein and also utilizes a relatively simple scoring function to evaluating van der waals (VDW) and electrostatic interactions, it was employed for the first round of screening. First, the compounds were docked in tankyrase 2; chemicals with scores below −40 kcal/mol were chosen for docking into tankyrase 1; subsequently, the compounds which scored below −25 kcal/mol were chosen for the second round of docking (Table S2).

Discovery of Candidate Compounds by Virtual Screening.
Following the above process, we screened the drug-like and natural product databases in ZINC (http://zinc.docking.org/) of more than 13.4 million compounds through three steps in silico ( Figure  2). Prior to each virtual screening, the three reference compounds were used to test whether the dock and score system gave results that were consistent with the biological assay. We only collected the compounds with scores better than LDW643. We performed the second round of more accurate screening with Autodock4.2, a program scoring the docking of chemical base on the Lamarckian Genetic Algorithm. The compounds were likewise first docked into tankyrase 2 and collected the chemicals which scored below −10 kcal/mol for docking into tankyrase 1; subsequently, compounds with a score less than −7 kcal/mol were subjected to further screening.
At the final step, a flexible ligand-protein docking program was used for precisely filtering compounds and those which scored below −8.5 kcal/mol were selected. After removing duplicate compounds and excluding incorrect and similar structures, 326 compounds were obtained as the hit compounds (Table S3). We visually inspected the hit-list and purchased 11 structurally representative compounds for further validation (Table 1).  Given the volume of compounds in the database, the rapid docking program and simple scoring evaluation for efficiency were the chief goal in the primary screening. Since DOCK6.5 is fast for matching compounds to a specific site on target protein and also utilizes a relatively simple scoring function to evaluating van der waals (VDW) and electrostatic interactions, it was employed for the first round of screening. First, the compounds were docked in tankyrase 2; chemicals with scores below −40 kcal/mol were chosen for docking into tankyrase 1; subsequently, the compounds which scored below −25 kcal/mol were chosen for the second round of docking (Table S2).
We performed the second round of more accurate screening with Autodock4.2, a program scoring the docking of chemical base on the Lamarckian Genetic Algorithm. The compounds were likewise first docked into tankyrase 2 and collected the chemicals which scored below −10 kcal/mol for docking into tankyrase 1; subsequently, compounds with a score less than −7 kcal/mol were subjected to further screening.
At the final step, a flexible ligand-protein docking program was used for precisely filtering compounds and those which scored below −8.5 kcal/mol were selected. After removing duplicate compounds and excluding incorrect and similar structures, 326 compounds were obtained as the hit

Biological Evaluation of the Purchased Compounds
We used a 9X TCF luciferase reporter gene assay (TOPFlash) to functionally validate the ability of chemicals to inhibit the β-catenin transcription activity in HEK293 cells [23]. We found that LiCl and Wnt3a strongly enhanced the transcription activity of β-catenin, indicating the authenticity of our system to validate the ability of a chemical for its ability to target the WNT/β-catenin pathway. Interestingly, six compounds (LZZ-02, LZZ-03, LZZ-05, LZZ-08, LZZ-09 and LZZ-10) significantly inhibited both Wnt3a-and LiCl-upregulated TOPFlash activity in this screening ( Figure 3A-C). All of the six compounds were well docked into the same sites as shown in Figure S1.  (Table S3). We visually inspected the hit-list and purchased 11 structurally representative compounds for further validation (Table 1).   (Table S3). We visually inspected the hit-list and purchased 11 structurally representative compounds for further validation (Table 1).   We used a 9X TCF luciferase reporter gene assay (TOPFlash) to functionally validate the ability of chemicals to inhibit the β-catenin transcription activity in HEK293 cells [23]. We found that LiCl and Wnt3a strongly enhanced the transcription activity of β-catenin, indicating the authenticity of our system to validate the ability of a chemical for its ability to target the WNT/β-catenin pathway.

Biological Evaluation of the Purchased Compounds.
We used a 9X TCF luciferase reporter gene assay (TOPFlash) to functionally validate the ability of chemicals to inhibit the β-catenin transcription activity in HEK293 cells [23]. We found that LiCl and Wnt3a strongly enhanced the transcription activity of β-catenin, indicating the authenticity of our system to validate the ability of a chemical for its ability to target the WNT/β-catenin pathway. Interestingly, six compounds (LZZ-02, LZZ-03, LZZ-05, LZZ-08, LZZ-09 and LZZ-10) significantly

Biological Evaluation of the Purchased Compounds.
We used a 9X TCF luciferase reporter gene assay (TOPFlash) to functionally validate the ability of chemicals to inhibit the β-catenin transcription activity in HEK293 cells [23]. We found that LiCl and Wnt3a strongly enhanced the transcription activity of β-catenin, indicating the authenticity of our system to validate the ability of a chemical for its ability to target the WNT/β-catenin pathway. Interestingly, six compounds (LZZ-02, LZZ-03, LZZ-05, LZZ-08, LZZ-09 and LZZ-10) significantly inhibited both Wnt3a-and LiCl-upregulated TOPFlash activity in this screening ( Figure 3A-3C). All of the six compounds were well docked into the same sites as shown in Figure S1.

Biological Evaluation of the Purchased Compounds.
We used a 9X TCF luciferase reporter gene assay (TOPFlash) to functionally validate the ability of chemicals to inhibit the β-catenin transcription activity in HEK293 cells [23]. We found that LiCl and Wnt3a strongly enhanced the transcription activity of β-catenin, indicating the authenticity of our system to validate the ability of a chemical for its ability to target the WNT/β-catenin pathway. Interestingly, six compounds (LZZ-02, LZZ-03, LZZ-05, LZZ-08, LZZ-09 and LZZ-10) significantly inhibited both Wnt3a-and LiCl-upregulated TOPFlash activity in this screening ( Figure 3A-3C). All of the six compounds were well docked into the same sites as shown in Figure S1. expression of downstream target genes [56]. To further validate the inhibitory activity of these compounds against the WNT pathway, we checked the protein level of β-catenin in DLD-1 cells treated with the above chemicals respectively. Strikingly, a significant decrease of β-catenin protein level was seen in cells treated with LZZ-02, LZZ-08 and LZZ-10. Of note, LZZ-02 showed the strongest inhibitory effect on β-catenin protein level in DLD1 cells ( Figure 3D). Moreover, assaying the activity of the TOPFlash reporter in transfected cells revealed an IC50 value showed around 10 ± 1.2 μM ( Figure 3E). Therefore, we focused our further experimental efforts on LZZ-02.  DLD-1 cells harbor an APC truncation, rendering β-catenin stabilization and constitutive expression of downstream target genes [56]. To further validate the inhibitory activity of these compounds against the WNT pathway, we checked the protein level of β-catenin in DLD-1 cells treated with the above chemicals respectively. Strikingly, a significant decrease of β-catenin protein level was seen in cells treated with LZZ-02, LZZ-08 and LZZ-10. Of note, LZZ-02 showed the strongest inhibitory effect on β-catenin protein level in DLD1 cells ( Figure 3D). Moreover, assaying the activity of the TOPFlash reporter in transfected cells revealed an IC 50 value showed around 10 ± 1.2 µM ( Figure 3E). Therefore, we focused our further experimental efforts on LZZ-02.

LZZ-02 Suppresses WNT/β-Catenin Signaling by Increasing Axin 2 Protein Level
Tankyrase inhibitors regulate WNT signaling by increasing the level of axin 2, the scaffold protein and main component of the β-catenin destruction complex [23]. We went further to examine whether LZZ-02 was able to suppress the β-catenin induced WNT signaling pathway. The luciferase assay indicated that LZZ-02 restrains β-catenin mediated TOPFlash activity in HEK293 cells ( Figure 4A). Aberrant activation of the WNT/β-catenin pathway promotes target proto-oncogenes c-Myc and Cyclin D1 transcription, thereby increasing their protein level [19,20]. We found that LZZ-02 decreased the protein level of β-catenin, c-Myc and CyclinD1 in a dose-dependent manner ( Figure 4B). In addition, TOPFlash assays suggested that LZZ-02 inhibits constitutive WNT activity in colon cancer cells DLD1 ( Figure 4C). LZZ-02 is also able to increase axin 2 protein level and strongly decreased total β-catenin level in DLD1 cells. Meanwhile, the expression level of β-catenin was also reduced by LZZ-02 in SW480 cells ( Figure 4D). Collectively, these data suggested LZZ-02 suppressed WNT/β-catenin signaling by increasing axin 2 protein level.

LZZ-02 Inhibits the Growth of Colon Cancer Cells in Vitro.
The WNT/β-catenin signaling cascade plays an important role in cell proliferation [57]. XAV939 has been shown to reduce the proliferation of human colorectal cancer cells [29,58]. Herein, we asked

LZZ-02 Inhibits the Growth of Colon Cancer Cells In Vitro
The WNT/β-catenin signaling cascade plays an important role in cell proliferation [57]. XAV939 has been shown to reduce the proliferation of human colorectal cancer cells [29,58]. Herein, we asked whether LZZ-02 could inhibit the growth of colorectal cancer cell-harboring, aberrantly-active tankyrases. Interestingly, CCK8 assay revealed that LZZ-02 significantly inhibited the proliferation of DLD1 and SW480. Moreover, LZZ-02 has a similar inhibition efficiency to XAV939 (Figure 5A,B). Meanwhile, a dose-dependent decrease of cell numbers was also noted in DLD-1 and SW480 in the presence of LZZ-02 ( Figure S2). More importantly, LZZ-02 also decreased the numbers of the colony forming of DLD1 and SW480 ( Figure 5C,D). Considering that APC mutation and the aberrant activation of the WNT/β-catenin are common in colorectal cancer, LZZ-02 deserved further characterization.
Molecules 2020, 25, x FOR PEER REVIEW 10 of 18 whether LZZ-02 could inhibit the growth of colorectal cancer cell-harboring, aberrantly-active tankyrases. Interestingly, CCK8 assay revealed that LZZ-02 significantly inhibited the proliferation of DLD1 and SW480. Moreover, LZZ-02 has a similar inhibition efficiency to XAV939 (Figure 5A &  B). Meanwhile, a dose-dependent decrease of cell numbers was also noted in DLD-1 and SW480 in the presence of LZZ-02 ( Figure S2). More importantly, LZZ-02 also decreased the numbers of the colony forming of DLD1 and SW480 ( Figure 5C & D). Considering that APC mutation and the aberrant activation of the WNT/β-catenin are common in colorectal cancer, LZZ-02 deserved further characterization.

LZZ-02 Inhibits the Growth of Subcutaneous DLD1 Xenografts.
To validate the antitumor activity of LZZ-02 in vivo, we inoculated DLD1 cells subcutaneously in flanks of nude mice. When tumors reached a volume of around 80 mm 3 , mice were randomized into two groups for treatment through oral gavage with 30 mg/kg of LZZ-02 or vehicle for 18 days. Interestingly, treatment with LZZ-02 caused significant tumor shrinkage ( Figure 6B). At day 18 post treatment, the average weight of tumor in LZZ-02-treated group was around 0.26 g, in stark contrast to an average of around 1 g in the control group ( Figure 6D & E). We also found that LZZ-02 is well

LZZ-02 Inhibits the Growth of Subcutaneous DLD1 Xenografts
To validate the antitumor activity of LZZ-02 in vivo, we inoculated DLD1 cells subcutaneously in flanks of nude mice. When tumors reached a volume of around 80 mm 3 , mice were randomized into two groups for treatment through oral gavage with 30 mg/kg of LZZ-02 or vehicle for 18 days. Interestingly, treatment with LZZ-02 caused significant tumor shrinkage ( Figure 6B). At day 18 post treatment, the average weight of tumor in LZZ-02-treated group was around 0.26 g, in stark contrast to an average of around 1 g in the control group ( Figure 6D,E). We also found that LZZ-02 is well tolerated by mice as judged by their constant body-weight ( Figure 6A,C), indicative of favorable in vivo toxicity profile of LZZ-02.

Docking Study by Molecular Modeling of Interactions Between LZZ-02/XAV939 and TNKS-1/2.
We went further to check the detail of binding mode of LZZ-02 with TNKS-1/2. We found that LZZ-02 nested in the NAD + pockets of TNKS-1/2 in a similar way to that of XAV939 ( Figure 7A-7D). Simulation analysis revealed that XAV939 anchored and formed four H-bonds in the binding site of TNKS-2: The pyrimidine nitrogen and hydroxyl and the sulfur atom in the thiopyrano ring were

Docking Study by Molecular Modeling of Interactions Between LZZ-02/XAV939 and TNKS-1/2
We went further to check the detail of binding mode of LZZ-02 with TNKS-1/2. We found that LZZ-02 nested in the NAD + pockets of TNKS-1/2 in a similar way to that of XAV939 ( Figure 7A-D). Simulation analysis revealed that XAV939 anchored and formed four H-bonds in the binding site of TNKS-2: The pyrimidine nitrogen and hydroxyl and the sulfur atom in the thiopyrano ring were within hydrogen bonding distance of the Gly1032, Ser1068 and Phe1061 backbone, respectively ( Figure 7B), consistent with a previous study [39]. Interestingly, LZZ-02 forms three H-bonds with TNKS-2, except the Gly1032 side chain stack with carbonyl and amino. Of note, nitro of LZZ-02 formed interactions with Ile1075 backbone near the opening of the binding crevice ( Figure 7D). Similarly, LZZ-02 also formed a hydrogen bond with a Glu1291 side chain on the verge of the binding site of TNKS-1; its nitro group interacted with His1184, Gly1185 and Ser1221 backbones deep in the of pocket ( Figure 7C). In contrast, XAV939 only formed interactions with Phe1214 and Glu1291 inside of TNKS-1 ( Figure 7A). Compared with XAV939, LZZ-02 formed more additional interaction with amino acid near the edge of the binding crevice in TNKS-1/2, which could stabilize their binding position and thereby enhance the inhibitory effects, as suggested in an earlier report to optimize tankyrase inhibitors [39].
Molecules 2020, 25, x FOR PEER REVIEW 12 of 18 within hydrogen bonding distance of the Gly1032, Ser1068 and Phe1061 backbone, respectively ( Figure 7B), consistent with a previous study [39]. Interestingly, LZZ-02 forms three H-bonds with TNKS-2, except the Gly1032 side chain stack with carbonyl and amino. Of note, nitro of LZZ-02 formed interactions with Ile1075 backbone near the opening of the binding crevice ( Figure 7D). Similarly, LZZ-02 also formed a hydrogen bond with a Glu1291 side chain on the verge of the binding site of TNKS-1; its nitro group interacted with His1184, Gly1185 and Ser1221 backbones deep in the of pocket ( Figure 7C). In contrast, XAV939 only formed interactions with Phe1214 and Glu1291 inside of TNKS-1 ( Figure 7A). Compared with XAV939, LZZ-02 formed more additional interaction with amino acid near the edge of the binding crevice in TNKS-1/2, which could stabilize their binding position and thereby enhance the inhibitory effects, as suggested in an earlier report to optimize tankyrase inhibitors [39].

Receptor Preparation
The 3D protein structure for docking was based on the X-ray crystal structure of the PARP domain of tankyrase. Free Tankyrase 1 and tankyrase 2 bound to its complex inhibitor, XAV939, 2-[4-(trifluoromethyl) phenyl]-7,8-dihydro-5H-thiopyrano [4,3-d] pyrimidin-4-ol], were obtained from the Protein Data Bank (PDB entry 2RF5 [38], 3KR8 [39], respectively). Prior to docking with DOCK6.5, the protein was fixed by deleting waters molecules, adding hydrogens and adding the Gasteiger charge. Before docking with Autodock4.2, the MGL Tools was used to treat proteins: adding the Gasteiger charge, adding the polar hydrogens, removing the water and writing in the pdbqt format.

Receptor Preparation
The 3D protein structure for docking was based on the X-ray crystal structure of the PARP domain of tankyrase. Free Tankyrase 1 and tankyrase 2 bound to its complex inhibitor, XAV939, 2-[4-(trifluoromethyl) phenyl]-7,8-dihydro-5H-thiopyrano [4,3-d] pyrimidin-4-ol], were obtained from the Protein Data Bank (PDB entry 2RF5 [38], 3KR8 [39], respectively). Prior to docking with DOCK6.5, the protein was fixed by deleting waters molecules, adding hydrogens and adding the Gasteiger charge. Before docking with Autodock4.2, the MGL Tools was used to treat proteins: adding the Gasteiger charge, adding the polar hydrogens, removing the water and writing in the pdbqt format.

Ligand Preparation
The XAV939, ABT-888 and LDW6433D structures of these compounds were constructed using the Sketch Molecule module in the SYBYL software (Version 6.9, Tripos Associates, St. Louis, MO, USA). Energy minimization was performed by the Powell gradient algorithm with the Tripos force field [58] and the Gasteiger-Huckel charge [59]. More than 13.4 million structurally diverse compounds were downloaded from all-purchasable subsets of the publicly accessible ZINC database, which contains two databases: the Drug-Like Database of 13.3 million molecules and the Nature Products Database of 89.4 thousand molecules [60,61]. These compounds are selected because they are commercially available and they were filtered by applying Lipinski's rule of five. All compounds were downloaded in MOL2 format, then added hydrogens and charge for virtual screening with DOCK 6.5 directly. For docking with Autodock, compounds were papered with prepare_ligand4.py script in Autodock tools.

Virtual Screening
The molecular docking program DOCK 6.5 [62] was utilized to perform the first round virtual screening owing to its fast calculating speed, followed by the rigid dock program and the flexible docking using Autodock 4.2 software, respectively. The job was performed on a Dawning A620-F cluster of 16 processors-each 2.6G AMD Opteron at the Gansu Computational Center.

Transfection and Reporter Assay
HEK293 cells were transfected by the standard calcium phosphate precipitation method, and DLD1 cells were transfected with lipofectamine 2000 according to manufacturer's instruction. An empty amount of vector plasmid was served as control in each transfection. To normalize for transfection efficiency, 0.1 µg Renilla luciferase reporter plasmid was added to each transfection. Luciferase assays were performed using a dual-specific luciferase assay kit.

Cell Proliferation Assay
Cells were seeded in 96-well plates at 1 × 10 3 per well. Cell proliferation was evaluated using the Cell Counting Kit-8 according to the manufacturer's instructions. Briefly, 10 µL of the CCK-8 solution were added to culture medium and incubated for 2 h, and the absorbance at a 450 nm wave length was determined.

Colony Formation Assay
Cells were seeded in 6-well plates at 200 or 1000 per well and maintained in a medium containing 0.5% FBS. Sixteen hours later, LZZ-02 were added at the indicated concentrations. The medium was replenished every two days until colony formation was observed. The cells were then washed twice with PBS, fixed with cold methanol, and stained with 0.5% crystal violet.

Immunoblot Analysis
The cells were washed twice with ice-cold PBS and then lysed with ice-cold RIPA lysis buffer. Lysates were kept on ice for 30 min and then centrifuged at 13,000 rpm for 5 min at 4 • C. The supernatant was subjected to 10% SDS/PAGE and followed by immunoblot analysis with the indicated antibodies.

In vivo Xenograft Model
All mice were housed in a pathogen-free environment in Jinan University. All experimental protocols were approved by the Institutional Committee for Animal Care and Use at Jinan University. All animal work was performed in strict accordance with the approved protocol. DLD1 cells (2.5 × 10 6 ) suspended in a 100 µL mixture of equal volumes of PBS and Matrigel were implanted subcutaneously into the flank of 6-week-old female BALB/c nude mice. When the tumors had reached a volume of about 80 mm 3 , the 12 mice were then randomly divided into two groups. Animals received the compound by oral gavage, whereas the control group received vehicle solution. LZZ-02 were orally gavaged at 30 mg/kg once daily for 18 consecutive days. Tumor volumes and the body weight of animals were measured every three days. Tumor volume (tumor volume (cm 3 ) = D × d 2 /2, where D is the longest and d is the shortest diameter, respectively) were monitored once every three days after 18 days up to the end of the experiment.

Statistical Analysis
Statistics were performed using GraphPad Prism 7.04, and the student t-test was used to compare differences between the two experimental groups. The data are presented as means ± SD and p < 0.05 was considered statistically significant.

Conclusions
In conclusion, tankyrase 1/2 structure-based virtual screening was performed successfully to identify inhibitors of the WNT/β-catenin pathway from the ZINC Database. In total, 11 compounds were selected to test their in vitro inhibitory activities against the WNT/β-catenin pathway. Among them, LZZ-02 showed the most significant inhibitory potency against LiCl-, Wnt3a-or β-catenin-induced TOPFlash activity in HEK293 cells. Meanwhile, LZZ-02 showed remarkable inhibition of TOPFlash activity in DLD-1 cells, which express constitutively activated β-catenin. Furthermore, LZZ-02 stabilized axin 2 protein level and restrain β-catenin, c-Myc and Cyclin D1 expression in DLD1 and SW480 cells. LZZ-02 also inhibits the colonic cancer cells' viability and colony formation. In the whole process of treatment, the mice were active, LZZ-02 showed antitumor function in vivo and the bodyweight of mice remains constant after treated. LZZ-02 is therefore worthy of further characterization in clinic. Equally important, LZZ-02 could serve as a lead candidate for developing highly potent tankyrase inhibitors, but the precise mechanism requires more investigation. : Table S1. Known TNKS-1/2 inhibitors for virtual screening method validations. Table  S2. The scoring results of training set compounds in each round screening. Table S3. The list of hit compounds by virtual screening (kcal/mol). Figure S1. Binding conformation of the selected 6 compounds. XAV939 as the positive control show similar conformation. Figure S2. LZZ-02 reduces proliferation in dose-dependent of human colorectal cancer cell. The DLD-1 (A) and SW480 (B) cell line were treated different concentration of LZZ-02.