Selectively Targeting STAT3 Using a Small Molecule Inhibitor is a Potential Therapeutic Strategy for Pancreatic Cancer

Signal transducer and activator of transcription 3 (STAT3), a latent cytoplasmic transcription factor, belongs to the STAT family proteins. After stimulating by cytokines, growth factors, and other signals, STAT3 is constitutively activated and supports tumor cell survival, growth, migration, drug resistance, and immune evasion. The activated STAT3 signaling pathway is closely related to disease progression and poor prognosis, and thus STAT3 represents a promsing therapeutic target for the cancer treatment.

In canonical activation, STAT3 is recruited to cytokine receptor intracellular domains via its Src homology 2 (SH2) domain, followed phosphorylation by JAK at Tyr705, and translocated into the nucleus to start transcription. For its essential role in homodimerization and the consequent transcription, the SH2 domain becomes a pivotal target for inhibiting STAT3 activities. Despite two decades of effort, direct targeting of the SH2 domain still faces numerous difficulties. First, STAT3 activities are regulated by protein–protein or protein–DNA interactions that involve a diffuse and large interface of protein, which are difficult to target by small molecules (1). Second, STAT family proteins share high sequence homology in their SH2 domain, making it hard to specifically target STAT3 over other STAT members (2). Hence, despite persistent research in the last 20 years, targeting STAT3 by a small-molecule inhibitor remains challenging.

The poor 5-year survival rate of pancreatic cancer is largely due to inefficient diagnosis and unsatisfactory treatment options (3, 4). Thus, it is urgent to develop effective targeted therapeutics for pancreatic cancer. STAT3 is constitutively activated in pancreatic cancer for the following mechanisms.The activation of STAT3 signaling promotes cancer cell survival and inhibits apoptosis (5, 6). Besides, STAT3 signaling enhances the expression of MMP7 and therefore promotes pancreatic cancer growth and metastasis (7). STAT3 also directly regulates the microenvironment of pancreatic cancer to promote tumor progression (8–10), and genetic deletion of STAT3 in pancreatic epithelial cells reduces tumorigenesis in Kras^G12D (KC) mouse models (7, 10). In contrast, the STAT3 pathway is inactive in the normal pancreas and is dispensable for pancreatic development and hemostasis (11). Consequently, STAT3 has become a potential target for the treatment of pancreatic cancer.

In this paper, we described the discovery and mechanism of WB436B, a potent and highly selective STAT3 inhibitor. WB436B exhibited selective cytotoxicity to pancreatic cancer cells with high levels of p-STAT3^Tyr705, and induced their apoptosis. WB436B was well tolerated, and suppressed pancreatic cancer growth and metastasis in vivo. To sum up, this work not only provided a validated drug discovery pocket within the STAT3 SH2 domain, but also evaluated WB436B as a lead compound for pancreatic cancer treatment.

WB436B synthesis was described in supporting information.

Human pancreatic cancer cell lines including PANC-1, MIAPaCa-2, Capan-2, CFPAC-1, SW1990, AsPC-1, BxPC-3, the human normal pancreatic ductal cell line (hTERT-HPNE), human atrial fibroblast (HAF), human epidermal keratinocyte line (HaCaT), and human embryonic kidney cell line (HEK293T) were obtained from Shanghai Cell Bank, Chinese Academy of Sciences and ATCC (Manassas, VA). Cell lines were tested for Mycoplasma prior to the beginning of the studies (Mycoplasma Detection Kit, TransGen Biotech). Identification of cells was confirmed by short tandem repeat analysis, and the authenticated cell lines purchased from the Shanghai Cell Bank, Chinese Academy of Sciences 6 months prior to the studies were expanded and stocks were frozen. A cell stock was thawed and used for less than 20 passages in this study. The Aqueous One Solution Cell Proliferation Kit was purchased from Promega CellTiter-Glo (G3581). Detailed information about antibodies are listed in Supplementary Table S1.

PANC-1, HEK293T, HaCaT, HAF, and MIAPaCa-2 were cultured in DMEM; Capan-2, BxPC-3, AsPC-1 were cultured in RPMI1640 medium; SW1990 was cultured in Leibovitz's L-15 medium; CFPAC-1 was cultured in Iscove's modified Dulbecco's medium; and hTERT-HPNE was cultured in 75% DMEM without glucose medium and 25% Medium M3 Base medium. All media contained 10% to 15% FBS and 1% penicillin–streptomycin. For MIAPaCa-2, an additional 2.5% horse serum was added. All cell lines were maintained at 37°C in the presence of 5% CO₂.

Docking was performed by Schrödinger Glide software (New York, NY). The crystal structure of STAT3-SH2 domain (PDB code 6QHD; ref. 12) was prepared by Glide Grid to detect druggable pockets using Fpocket 2.0 server (13). The druggability score was calculated to assess the possibility of a pocket to accommodate drug-like molecules. The score more than 0.5 (the threshold) means the pocket might be druggable. On the basis of the druggable pocket, we screened Chembridge Library with 1.3 million compounds, and the pan-assay interference compounds (PAINS) were removed before virtual screening. The molecules were prepared and tautomerized at pH 7.0 by LigPrep module of the Schrödinger software suite. Compounds were screened using the standard precision docking module in Glide. The OPLS3 force field (14) was used to parameterize both ligands and protein. Glide G-score was used to rank the results list and the 1,000 top-scoring compounds from each virtual screen were subjected to visual inspection. To allow for diversity of molecular structures, binding modes, and drug-like properties, 11 compounds were selected and purchased for the bioassay with a purity of >95%.

The induced-fit docking (IFD) protocol was performed on compound WB436B using the induced-fit tool in maestro 11.5 from Schrödinger. IFD employed the use of glide and prime for ligand docking and protein refinement, respectively. The active site was centered on the Tyr705 phosphorylated site near the STAT3^SH2 domain (PDB code 6QHD). Induced-fit protein–ligand complexes were generated using prime and further subjected to side chain and backbone refinement.

The molecular dynamics (MD) simulation was performed as previously described (15). In brief, MD simulation was performed on WB436B-bound STAT3 from the IFD. MD simulation was carried out using GROMACS 5.1.4 (16) with CHARMM36 parameters (17). The force-field parameters of WB436B in the simulation were generated on the basis of the CHARMM General Force Field (CGenFF; ref. 18). Energy minimizations were first performed to relieve unfavorable contacts, followed by equilibration steps of 11 ns in total to equilibrate the solvent with restraints (isotropic force constant κ = 1 × 10³ kJ/mol/nm²) on ligand and Cα atoms in STAT3. For the equilibrated system, at least two replicates of all-atom MD simulations were generated with different initial velocities. The temperature of each system was maintained at 300 K using the v-rescale method (19) with a coupling time of 0.1 ps. The pressure was kept at 1 bar using the Parrinello-Rahman (20) with τ_p = 2.0 ps and a compressibility of 4.5 × 10⁻⁵ bar⁻¹. LINCS (21) constraints were applied to the hydrogen-involving covalent bonds of water molecules and other molecules, respectively, and the time step was set to 2 fs. Electrostatic interactions were calculated with the Particle-Mesh Ewald algorithm (22) with a real-space cutoff of 1.2 nm. Van der Waals interactions used a 1.0-nm cutoff. Analysis of the trajectories was performed using Gromacs analysis tools. PyMOL (Version 1.3, Schrödinger, LLC) was used to visualize the structural models and generate figures.

The RosettaLigand application (23, 24) was applied to dock WB436B to STAT3. The RosettaLigand protocol was described in our previous work (25). The best binding mode obtained from MD simulation was adopted as the initial docking model. The top 1,000 models were sorted by total score, and the binding energy between WB436B and the STAT3^SH2 domain was calculated. In addition, in silico alanine scans were conducted by individually changing the residue to alanine without otherwise changing the conformation of the protein or ligands in Rosetta. To explore the distribution of binding interactions between WB436B and STAT3, the average energies of the top 10 models with the lowest binding energies (interface score) were calculated.

pET28A-STAT3^SH2 and pET28A-STAT3^127–722 were generated by cloning the STAT3 SH2 domain (amino acid residues 586–685) and STAT3 (amino acid residues 127–722) into the pET28A vector, respectively. All the mutants of the STAT3-SH2 were produced by site-directed mutagenesis. Each insert was cloned into the pET28A vector. In addition, STAT1 (127–716), STAT2 (138–702), STAT4 (133–705), STAT5B (136–703), and STAT6 (113–658) were also cloned into the pET28A vector. Recombinant human fusion proteins were expressed in E. coli BL21 (DE3). Briefly, BL21 (DE3) was transformed with indicated plasmids. For STAT3 plasmids, when the OD value reached about 0.6 to 1.0, the E. coli was transferred to 22°C and 0.5 mmol/L Isopropyl β-D-1-thiogalactopyranoside (IPTG) was added. When the OD value reached about 0.8 to 1.0, 0.5 mmol/L IPTG was added to the Luria-Bertani (LB) medium of the BL21 (DE3) transformed with the pET28A-STAT1^127–716, pET28A-STAT2^138–702, and pET28A-STAT6^113–658 plasmids, and the condition was changed to 22°C for 12–16 h. For BL21 (DE3) transformed with the pET28A-STAT4^133–705 and pET28A-STAT5B^136–703 plasmids, when the OD value reached about 0.5–0.6, 0.2 mmol/L IPTG was added to the LB medium, and the condition was changed to 30°C for 2.5 to 3 hours. The soluble protein was obtained by sonication and centrifugation. The protein was incubated with Ni NTA Beads (Qiagen), and then eluted with a concentration gradient of Imidazole.

Microscale thermophoresis (MST) was conducted as follows. Binding affinities of WB436B, Stattic and C188–9 against purified STAT3^127–722 or WB436B against STAT3-SH2 WT and mutants were measured by a Monolith NT.115 (Nanotemper Technologies). The proteins were fluorescently labeled according to the manufacturer's procedure and kept in the MST buffer (50 mmol/L HEPES, pH 7.0, 500 mmol/L NaCl, 0.01% NP40, 50 mmol/L l-arginine) at a concentration of 200 nmol/L. Next, the RED fluorescent dye was added, mixed, and incubated for 30 minutes at 25°C in the dark. For each assay, the labeled protein was mixed with the same volume of the unlabeled compound at 16 different serially diluted concentrations at room temperature. The samples were then loaded into premium capillaries (NanoTemper Technologies) and measured at 25°C by using 20% to 40% LED power and medium MST power. Each assay was repeated two or three times. Data analyses were performed using Monolith. Affinity analysis used v.2.2.4 software. All figures were made by GraphPad Prism 7.0.

Surface plasmon resonance (SPR) experiments were performed with a Biacore 8K instrument (Cytiva) with CM7 sensor chip (Cytiva). To test WB436B binding of STAT families’ protein, including STAT1, STAT2, STAT3, STAT4, STAT5B, and STAT6 protein, serially diluted concentrations of WB436B were injected into the flow system. Experiments were conducted using PBS and the analyte was injected at the flow rate of 30 μL/min. The association time was 90 s and the dissociation time was 90 s. WB436B was dissolved in PBS containing 5% dimethyl sulfoxide and a solvent correction assay was performed to adjust the results. STAT protein was immobilized on the sensor chip (CM7) using the amine-coupling method according to standard protocols. WB436B at various concentrations was injected into the flow system. The K_D values were calculated with the kinetics and affinity analysis option of Biacore evaluation software.

Cell viability was measured by CellTiter-Glo (Promega, G3581) assay following the manufacturer's instructions. Briefly, pancreatic cancer cells or normal cells were seeded in 96-well plates at appropriate densities, allowing attachment overnight, followed by 72-hour WB436B exposure. Aqueous One solution (MTS) was then added. The OD values at 490 nm were acquired.

Tumor cells were seeded into 6-well plates, allowing attachment overnight. Different concentrations of WB436B were added for 1-week incubation. Then colonies were fixed by 4% paraformaldehyde and stained with 0.2% crystal violet. Images were photographed using a digital camera, and colonies were quantified by manual counting.

Pancreatic cancer cells were lysed in RIPA buffer supplemented with 1 mmol/L phenylmethylsulfonyl fluoride, a proteinase inhibitor cocktail and a phosphatase inhibitor cocktail (Sigma). Lysates were separated by 8% to 12% SDS-PAGE and transferred to nitrocellulose. The blots were probed with specific antibodies followed by a secondary antibody, and then membranes were examined by the LI-COR Odyssey infrared imaging system (LI-COR Biotechnology, Lincoln, NE). The Western blot results were quantitated by ImageJ following the manufacturer's instructions, and these results are shown in Supplementary Files, named “Quantification results of Western blots”.

Total RNA was extracted using TRIzol (Takara, Japan) according to the manufacturer's instructions. A total of 1,000 ng RNA was used for cDNA synthesis using a cDNA Reverse Transcription Kit (Takara, Japan). Real-time PCR was performed in triplicates using gene-specific primers on the QuantStudio3 Real-Time PCR System machine (Applied Biosystems). The mRNA expression levels were normalized to β-actin and the gene-specific primers are listed in Supplementary Table S2.

Pancreatic cancer cells were seeded on gelatin-coated glass coverslips. After 20 hours of WB436B treatment in basic medium, IL6 (20 ng/mL) was added for 30 minutes. Cells were washed with cold PBS, fixed with 4% paraformaldehyde, and treated with 0.2% Triton-X 100. After blocking in 0.5% BSA, cells were incubated with primary antibody overnight at 4°C before further incubation with secondary antibody at 37°C for 1 hour in the dark. Then DAPI was added for 5 minutes in the dark. Images were recorded by microscopy (Leica).

The target sequence for silencing STAT3 is listed in Supplementary Table S3. shRNAs were inserted into pLKO.1 vector respectively and were co-transfected into HEK293T cells with the packing plasmids pMD2.G and pxPAX2, using Lipofectamine 2000 (Thermo Fisher Scientific). After transfection for 72 hours, the harvested lentiviruses were infected into pancreatic cancer cell lines and the puromycin was added to screen to obtain stable STAT3 knockdown cells. Finally, the knockdown efficiency was detected by Western blots.

Pancreatic cancer cells were cross-linked with 3 mmol/L ethylene glycol bis (succinimidyl succinate; EGS; 30 minutes) and 1% formaldehyde (10 minutes) at room temperature. Cross-linking was stopped by 125 mmol/L glycine. Cells were collected by PBS and lysed in 0.1% SDS, 1% Triton X-100, 10 mmol/L Tris-HCl, pH 7.5, 1 mmol/L EDTA, pH 8.0, 0.1% NaDoc, 0.3 mol/L NaCl, and protease inhibitors. Next, after ultrasonic treatment, the DNA fragments of the lysates were interrupted to 500 bp. Immunoprecipitation with Protein G-plus Agarose beads prebound with antibody (anti-STAT3) was performed for 15 hours at 4°C overnight. The following day, cross-linking was reversed and DNA fragments were purified for analysis by real-time PCR. The primer pairs were shown in Supplementary Table S4.

Six × 10⁵ cells were seeded on a 6-well plate, treated with or without 100 nmol/L WB436B for 24 hours and harvested for RNA extraction. For the STAT3 knockdown group, the siRNA target STAT3 was transfected with BxPC3. After 72-hour transfection, the RNA was extracted and knockdown efficiency was examined by Western blots. Strand-specific libraries were constructed using the TruSeq RNA sample preparation kit (Illumina, San Diego, CA), and sequencing was carried out using the Illumina Novaseq 6000 instrument. The raw data were handled by Skewer and data quality was checked by FastQC v0.11.2. The read length was 2×150 bp. Clean reads were aligned to the Human genome hg38 using STAR and StringTie. The expression of the transcript was calculated by FPKM (Fragments Per Kilobase of exon model per Million mapped reads) using Perl. Differentially expressed transcripts were determined using the MA plot–based method with Random Sampling (MARS) model. R statistical programming was used for further analysis. The upregulated or downregulated genes upon WB436B treatment were subjected to gene set enrichment analysis (GSEA). Gene sets with a P value < 0.05 and with an FDR < 0.25 were considered statistically significant.

BALB/c-nude, male, 6- to 8-week-old mice were obtained from the Animal Center of East China Normal University. All animal experimental protocols were approved by the Animal Investigation Committee of the Institute of Biomedical Sciences, East China Normal University. The PANC-1 xenograft tumor models were developed by injecting 5 × 10⁶ PANC-1 cells in suspension into the right flank of a BALB/c-nude mouse while cells were suspended in PBS with 50% matrigel. Treatment began after the tumor nodules grew to 100 mm³. Tumor-bearing BALB/c-nude mice were randomly assigned to four groups and treated by intraperitoneal injection of compound or drug. The tumor volume and mouse body weight were measured twice a week. The tumor volume was calculated using the following equation: tumor volume (V) = length × width × width × 0.52. At the end of the experiment, the mice were euthanized by CO₂. The tumors were removed and prepared for Western blot and IHC analyses.

Animal experimental protocols were approved by the Animal Investigation Committee of the Institute of Biomedical Sciences, East China Normal University. NPI mice (male, 6- to 8-week-old) were obtained from the Beijing IDMO. The patient-derived xenograft (PDX) studies were performed by Beijing IDMO as a customized service. Primary PDX was established from patient with pancreatic cancer maintained in NPI mice. One generation was transplanted, and the second generation was used for experiments. When the xenografts reached about 1,000 mm³, the mice were sacrificed, and the tumors were washed in PBS and sectioned into a tissue block approximately 2 to 3 mm in diameter. The tumors were passaged and injected subcutaneously into the flanks of NPI mice. Treatment began after the tumor nodules grew to 200 mm³; mice were randomly assigned to three groups and intraperitoneally injected with WB436B (2.5 and 5 mg/kg/d), with an equivalent volume of vehicle serving as the control group. Tumor volume was measured using a digital caliper and calculated with the formula (V) = length × width × width × 0.52. Tumor size was monitored per 5 days until the endpoint. After 3 weeks of drug treatment, the mice were euthanized by CO₂, and the tumors were collected. Hematoxylin and eosin (H&E) and IHC were performed on paraffin-embedded xenograft tumor specimens.

As previously described (26), 4- to 6-week-old male C57BL/6 mice were injected with 5×10⁵ murine pancreatic cancer cells (PAN02-luciferase) embedded in Matrigel (BD Biosciences) in the tail of the pancreas. On the next day (day 0), the mice were subjected to bioluminescent imaging and divided into four groups according to the luciferase luminescence value. Mice were treated with intraperitoneal injection of indicated compound or drug. The bioluminescence of cancer cells in lungs was monitored every week by an IVIS Imaging System (Xenogen Corporation, Alameda, CA). When the mice showed near-death indicators such as loss of mobility and body temperature drop, they were euthanized immediately in consideration of animal ethics. Another independent animal experiment was performed to determine whether WB436B prolongs the survival of mice.

Four- to 6-week-old male C57BL/6 mice were injected with 5×10⁵ murine pancreatic cancer cells (PAN02-luciferase) embedded in Matrigel (BD Biosciences) in the spleen. On the next day (day 0), the mice were divided into four groups according to the luciferase luminescence value of the mice. All mice in drug treatment groups were intraperitoneally injected with WB436B (2.5 and 5 mg/kg/d) or C188–9 (20 mg/kg/d), with an equivalent volume of vehicle serving as the control group. When the mice showed near-death indicators such as loss of mobility and body temperature drop, they were euthanized immediately in consideration of animal ethics. Another independent animal experiment was performed to determine whether WB436B prolongs the survival of mice.

Tissue samples were dewaxed and then washed with alcohol gradient, followed by treatment with 3% H₂O₂ in methanol and then blocked. Then sections were incubated with primary antibodies (p-STAT3^Tyr705 and Ki67) overnight at 4°C. The following day, sections were incubated with secondary antibodies for 1 hour and the signals were amplified and visualized by avidin–biotin complex system and DAB substrate. Images were acquired by Leica photomicroscope. The IHC images were quantified using the Aperio ImageScope.

Experiments were carried out with three or more replicates. Statistical analyses were done by Student t test. P values < 0.05 were considered significant. The differences between control and experimental groups were determined by one-way ANOVA. Since treatment and time course were investigated, two-way ANOVA followed by post hoc test was also applied. Data were expressed as means and 95% confidence intervals and P < 0.05 was considered significant. All analyses were performed using Microsoft Excel 2010 and GraphPad Prism 7 software.

The authors declare that the data supporting this study are available within the paper and its Supplementary Data file. All other data are available from the authors upon reasonable request.

To specifically identify STAT3 inhibitors, structure-based virtual screening and computer-aided drug design were used. A cavity under the crystal structure of human STAT3-SH2 domain (12) and nearby the Tyr705 site may be a drug binding pocket as it has high druggability scores (Supplementary Fig. S1A). Then, based on this druggable pocket, about 1.3 million compounds from Chembridge Library were screened (Supplementary Fig. S1B). The top-scoring compounds were selected for the in vitro experimental analysis using a STAT3 direct binding assay. The 8910106 was the potential hit compound with the most potent binding affinities to STAT3 (Supplementary Fig. S1C). After several rounds of lead optimization, we found that WB436B (Fig. 1A) was the most selective and potent STAT3 inhibitor among them. In MST assays, WB436B directly bound STAT3-SH2 and STAT3^127–722 with K_D of 94.3 ± 22.1 nmol/L and 129.0 ± 28.8 nmol/L, respectively (Fig. 1B). The SPR results also proved that WB436B directly bound to STAT3 (Fig. 1C). In addition, WB436B showed no appreciable binding to other STAT members including STAT1, STAT2, STAT4, STAT5B, and STAT6 (K_D > 10 μmol/L; Fig. 1D; Supplementary Fig. S2; Supplementary Table S5). Besides, WB436B had a weak inhibition of a panel of human kinases in vitro (Supplementary Table S6).

Subsequently, we investigated whether WB436B achieved high cellular selectivity to STAT3. Engagement of IFNα to IFN receptor complex leads to activation of STAT1 and STAT3 by tyrosine phosphorylation. Consistent with the direct binding and kinase inhibition results in vitro, WB436B specifically inhibited IFNα induced p-STAT3^Tyr705, but had no significant inhibition against STAT1 phosphorylation (Fig. 1E). Subsequently, we confirmed the effect of WB436B on the phosphorylation of JAK kinases and other pathway proteins, as results showed, WB436B selectively blocked p-STAT3^Tyr705 at lower than 100 nmol/L in PANC-1 cells, while it showed minimal impact on JAK1, STAT1, STAT5, AKT, and ERK1/2 phosphorylation (Fig. 1F). To our surprise, WB345 (Supplementary Fig. S3A), an inactive analog of WB436B, had a weak binding affinity to STAT3 in vitro and also less sensitive to pancreatic ductal adenocarcinoma (PDAC) cells (Supplementary Fig. S3B and S3C). These results further proved that WB436B was a selective STAT3 inhibitor.

Next, we compared the inhibitory capacity of WB436B with the representative STAT3 inhibitors including a JAK inhibitor (AZD1480), a STAT3 and cancer stemness inhibitor (BBI-608), and STAT3-SH2 domain inhibitors (Stattic, BP-1–102, C188–9). WB436B was more efficient in suppressing STAT3 phosphorylation and its regulatory target gene expression (e.g., c-Myc, Cyclin D1) than the above list of STAT3 inhibitors (Fig. 1G). Importantly, the WB436B-mediated suppression of STAT3 target gene expression was in line with the impact of shRNA-mediated genetic depletion of STAT3 (Fig 1G and H). Furthermore, the growth inhibitory activity of WB436B was potent by 10- to 1,000-fold than the listed STAT3 inhibitors (Fig. 1I). Then, we detected the binding affinities of STAT3 to WB436B, or Stattic, which was previously reported to directly bind to the STAT3-SH2 domain. The results showed that WB436B displayed a greater binding affinity to STAT3 than Stattic (Fig. 1J).

In summary, we found and validated a potent and highly selective STAT3 inhibitor, WB436B, which specifically targeted the STAT3-SH2 domain over other STAT members or upstream kinases by in vitro binding experiments, kinase inhibition assays, and cellular assessments. Compared with the reported STAT3 inhibitors, the higher affinity between WB436B to STAT3 contributed to its higher potency in suppression of cancer cell growth, STAT3 phosphorylation, and STAT3 target gene expression.

To provide a structural basis for the selective and high binding affinity between STAT3 and WB436B, an induced-fit molecular docking simulation was applied, and the binding model was optimized by MD simulation, which revealed that the predominant binding mode of WB436B was within the STAT3-SH2 domain (Fig. 2A). Residues Ser611, Gln635, Ser636, Glu638, and Tyr657 directly interacted with WB436B by hydrophilic interaction and other surrounding residues Val637 and Ile659 likely contributed hydrophobic interactions with the ligand. We next examined the ligand–STAT3 interactions in this binding mode using mutagenesis experiments and RosettaLigand (24). RosettaLigand computations predicted that these residue mutants decreased WB436B binding to differing extents (Fig. 2B). Next, we mutated the predicted amino acid residues for further testing by direct binding assay. In MST assays, WB436B lost its binding affinities to some of the indicated site-directed STAT3 mutations (K_D > 5 μmol/L), and the experiments concluded that Ser611, Glu612, Ser636, Glu638, Tyr657, and Ile659 were the most critical amino acid residues for the STAT3–WB436B interaction (Fig. 2C and D).

To verify whether the amino acid residues critical for WB436B mediated cytotoxicity to pancreatic cancer cells, we reintroduced wild-type STAT3 or the indicated SH2 mutants in stable STAT3 knockdown cells. In WT-STAT3 re-expression cells, WB436B anti-proliferation capacity was restored, and this effect was specific because it did not occur upon expression of the indicated SH2-mutant or vector plasmids (Fig. 2E–H). Then, we explored whether these amino acid residues were required for STAT3 transcription; a STAT3-reporter luciferase assay was performed. The well-characterized STAT3 gain-of-function mutations (27), including Y640F, D661V, N647I, and E616K were used for system validation. In luciferase experiments, all of the gain-of-function mutations directly activated luciferase reporter activity with or without IL6 stimulation. However, some mutants including S611A, E612A, E638A, Y657A, and I659A failed to respond to IL6 stimulation of STAT3 transcription (Supplementary Fig. S4), demonstrating that above amino acid residues were critical for STAT3 transcription. In summary, we concluded that the binding between WB436B and the STAT3-SH2 domain within the critical pocket resulted in suppression of STAT3 target gene transcription, thereby inhibiting tumor growth and metastasis in vitro and in vivo.

To gain an in-depth understanding of whether the growth inhibitory activities of WB436B were dependent on targeting STAT3, we assessed the expression of STAT3, its phosphorylation (p-STAT3^Tyr705 and p-STAT3^Ser727) in seven PDAC cell lines and three normal cells. All tested cell lines, except for PDAC cells AsPC-1 and three normal cell lines HPNE, HaCaT, and HAF, harbored higher levels of STAT3, p-STAT3^Tyr705, and p-STAT3^Ser727 (Fig. 3A). Next, we evaluated the growth inhibitory activity of WB436B in a panel of pancreatic cancer cell lines and normal cell lines. As shown in Fig. 3B, all of the high p-STAT3^Tyr705 cell lines were highly sensitive to WB436B (IC₅₀ < 0.1 μmol/L), while normal cell lines were insensitive (IC₅₀ > 10 μmol/L) and the low p-STAT3^Tyr705 pancreatic cancer cell line AsPC-1 showed intermediate sensitivity. Furthermore, WB436B inhibited the colony formation of pancreatic cancer cells with high p-STAT3^Tyr705, but had no obvious inhibitory effect on AsPC-1 cells (Fig. 3C and D). These findings established that WB436B achieved high selectivity in PDAC cells. Meanwhile, WB436B significantly induced several high p-STAT3^Tyr705 PDAC cells apoptosis (Supplementary Fig. S5A and S5B). As STAT3 is frequently activated in pancreatic cancer, it is conceivable that these high p-STAT3^Tyr705 cell lines may rely heavily on STAT3 signaling for survival. Therefore, WB436B inactivation of STAT3 activity made the compound a potent and selective growth inhibitor of high p-STAT3^Tyr705 cells.

In an attempt to investigate the biological effects of STAT3 in pancreatic cancer cell lines, we knocked down endogenous STAT3 in several pancreatic cell lines (Supplementary Fig. S6A). Knockdown of STAT3 by shRNA markedly inhibited the cell growth and colony formation of high p-STAT3^Tyr705 cell lines, including Capan-2, PANC-1, and BxPC-3 cells, while having no obvious effect on AsPC-1 (Fig. 3E; Supplementary Fig. S6B and S6C). To assess whether STAT3 was the primary target of WB436B, different shSTAT3 or negative control (shNC) pancreatic cancer cells were treated with WB436B and the corresponding cell viabilities and colony formation were detected. In cells with STAT3 loss or knockdown, sensitivity to WB436B decreased, while remaining sensitive to shNC cells (Fig. 3F–I; Supplementary Fig. S7A and S7B). While, in low p-STAT3^Tyr705 cells (AsPC-1), knockdown STAT3 did not cause a significant reduction in cell growth and colony formation, and AsPC-1 was also relatively less sensitive to WB436B treatment (Fig. 3J; Supplementary Fig. S7C). Consistent with the stable shRNA data, doxycycline (DOX)-induced STAT3 depletion led to marked inhibition of colony formation in PDAC cells (Supplementary Fig. S8A and S8B), and WB436B was also relatively insensitive to the DOX-induced STAT3 knockdown cells (Supplementary Fig. S8C). In summary, STAT3 was critical for pancreatic cancer cells survival, and WB436B achieved strong cytotoxicity to STAT3-dependent PDAC cells.

To determine whether WB436B suppressed STAT3 activity in pancreatic cancer cells, PANC-1 and Capan-2 were treated with WB436B, and immunoblot experiments were performed. WB436B selectively suppressed p-STAT3^Tyr705 but not total STAT3 or p-STAT3^Ser727 (Fig. 4A). IL6 stimulated the expression of p-STAT3^Tyr705 in PDAC, which was suppressed by WB436B at concentrations lower than 300 nmol/L (Fig. 4B). Immunofluorescence analysis demonstrated that WB436B decreased IL6-mediated STAT3 nuclear translocation in PDAC cells (Fig. 4C and D). In addition, WB436B suppressed STAT3 target gene expression, including c-Myc, Cyclin D1, Bcl-2, and VEGF (Fig. 4E and F). In brief, WB436B selectively inhibited p-STAT3^Tyr705 and STAT3 target genes expression in pancreatic cancer cells.

To futher explore the on-target effects of WB436B towards STAT3, we next examined the effect of WB436B on the whole-transcriptome analysis with total RNA sequencing in BxPC3 cells. We first compared the whole-transcriptome gene expression patterns in the absence or presence of WB436B to those induced by acute knockdown of STAT3. As results shown, the effects of WB436B treatment in gene expression profile in BxPC3 cells were consistent with STAT3 silencing (Fig. 4G; Supplementary Fig. S8D). GSEA with the STAT3 targets gene set GSE21670 (2, 28) affirmed a highly significant enrichment (Fig. 4H), suggesting that STAT3 inhibition by WB436B leads to downregulation of STAT3 transcriptional targets in BxPC3. Besides, previous studies demonstrated that c-Myc is a downstream target of the canonical JAK–STAT3 pathway (29). In our results, Myc-Targets pathways were significantly enriched in the downregulated genes (Fig. 4I). To further clarify how WB436B inhibited STAT3 downstream gene expression, we first detected the distribution of STAT3 after WB436B treatment. The nuclear and cytoplasmic extraction experiments demonstrated that WB436B decreased IL6-mediated STAT3 nuclear translocation in PDAC cells (Supplementary Fig. S9A and S9B). Chromatin immunoprecipitation assays further revealed that WB436B disturbed interaction between STAT3 and its target genes (Supplementary Fig. S9C and S9D). Taken together, WB436B selectively bound STAT3-SH2 domain to inhibit STAT3 nuclear translocation, and then deceased the binding between STAT3 with its target genes in pancreatic cancer cells.

Based on the potent effects of WB436B in vitro, we next evaluated its efficacy in vivo in the PANC-1 subcutaneous xenograft model. Administration of 2.5 or 5 mg/kg/d WB436B significantly suppressed tumor growth (Fig. 5A). After 20 days of treatment, tumors were weighed (Fig. 5B), and WB436B showed more potency in inhibiting pancreatic cancer growth than STAT3 phase I inhibitor C188–9 (Fig. 5A and B). Tumor tissues were analyzed by IHC, and these results proved that WB436B suppressed p-STAT3^Tyr705 and Ki67 expression in vivo (Fig. 5C and D), and Western blot results of tumor tissues also showed that WB436B efficently inhibited STAT3 phosphorylation and downstream gene expression in vivo (Fig. 5E). WB436B was well tolerated as demonstrated via the insignificant mouse body weight loss and lack of obvious toxicity in major organs visualized by H&E staining (Supplementary Fig. S10A and S10B). In conclusion, WB436B potently suppressed pancreatic tumor growth in vivo.

PDX animal models have been increasingly used in translational research since their development because they maintain more similarities to the parental tumors (30). We next explored whether WB436B suppressed tumor growth in pancreatic PDX models. WB436B potently inhibited PDX tumor growth (Fig. 5F–H). IHC revealed that WB436B inhibited p-STAT3^Tyr705 and Ki-67 expression (Fig. 5I and J). Besides, insignificant mouse body weight loss and lack of obvious toxicity in major organs visualized by H&E staining further proved that WB436B was with good safety profile in vivo (Supplementary Fig. S10C and S10D). Taken together, these results showed that WB436B specifically targeted STAT3, and had promising capability in suppressing tumor growth in preclinical pancreatic tumor models.

Owing to the potent antitumor growth effects in vivo, and evidence suggesting that STAT3 was involved in key steps of pancreatic cancer metastasis (31), we next investigated whether WB436B suppressed pancreatic cancer metastasis in vivo. WB436B was well tolerated (Supplementary Fig. S11A and S11B) and significantly suppressed PDAC tumor growth and metastasis in the orthotopic model and liver metastatic model of pancreatic cancer (Fig. 6A–D; Supplementary Fig. S12A and S12B). Further analysis of the pancreatic tumors showed that the metastatic foci in control and C188–9–treated mice became large but those in the WB436B-treated groups remained relatively small in the PDAC orthotopic model (Fig. 6A) and liver metastastic model (Fig. 6B). In PDAC orthotopic mouse model, the mice treated with 2.5 and 5 mg/kg/d WB436B had tumor bioluminescence decreases of 84% and 88%, respectively, versus control mice, while C188–9–treated mice only decreased by 33% (Fig. 6C). In liver metastastic mouse model, intraperitoneal injection of WB436B caused significant inhibition of pancreatic cancer liver metastasis (Fig. 6D). Moreover, under the 5 mg/kg/d of WB436B administration, we hardly found any metastatic tumor nodes in the liver (Fig. 6E; Supplementary Fig. S12B). In addition, administration of 2.5 and 5 mg/kg/d of WB436B significantly prolonged mouse survival in the PDAC orthotopic model (Fig. 6F) and liver metastastic model (Fig. 6G). In contrast, C188–9 treatment had little effect on tumor-bearing mouse survival (Fig. 6F and G). To sum up, WB436B potently abrogated tumor metastasis in the orthotopic and liver metastatic models of pancreatic cancer and significantly prolonged mouse survival.

To investigate the mechanisms of antitumor activities of WB436B in vivo, we first evaluated whether WB436B induced apoptosis in the orthotopic transplantation tumor model. As IHC staining results shown, WB436B induced expression of the marker of apoptosis (cleaved caspase 3), indicating WB436B significantly caused the PDAC tumors apotosis in vivo (Supplementary Fig. S13A and S13B). We also found administration of WB436B resulted in reduced tumor vasculature, with a marked decrease of the angiogenesis marker CD31 in tumors from WB436B-treated mice versus control mice (Supplementary Fig. S13A and S13B). The microenvironment of PDAC, which promotes tumorigenesis, disease development, and metastasis, consists of fibroblasts, immune cells, pancreatic stellate cells (PSC), adipocytes, and extracellular matrix (32). The persistent STAT3 activation supports a protumorigenic microenvironment in PDAC by increasing the levels of immunosuppressive cells (33, 34), such as tumor-associated macrophages (TAM). Thus, we detected the numbers of the F4/80+ macrophages in pancreatic orthotopic tumor model upon WB436B treatment, and administration of WB436B significantly decreased the number of TAM (Supplementary Fig. S13A and S13B). In brief, WB436B decreased PDAC tumors angiogenesis, inhibited TAM infiltration, and induced apoptosis in vivo.

Persistent activation of STAT3 is prevalent in a variety of human cancers. By promoting tumor survival, antiapoptotic signals, metastasis, and inhibiting antitumor immune responses, the activation of STAT3 signaling is closely related to poor prognosis of several human tumors. Thus, STAT3 is established as a promising therapeutic target for cancer treatment. The available strategies directly targeting STAT3 include the STAT3 DNA binding domain decoy (35), antisense oligonucleotides (36), and small molecules that bind to STAT3. Owing to its important role in regulating STAT3 activation, the STAT3-SH2 domain is turning into a feasible target for antagonizing STAT3. OPB31121 (37) and OPB51602 (38) have completed early clinical trials (Phase I/II) in multiple human tumors (39, 40). Recent reports also demonstrated that OPB31121 inhibited STAT3 activity by downregulating its upstream signaling factors JAK2 and GP130 (41). However, these two drugs were terminated for their poor pharmacokinetic properties and potential toxicities (42). C188–9 has been determined to be an effective small-molecule STAT3 inhibitor, which directly binds to STAT3 SH2 domains and significantly blocks the two steps of STAT3 activation: recruitment to activated receptors and homodimerization. Preclinical studies have shown that C188–9 markedly inhibits the proliferation of various tumor cells in vitro and in vivo, including hepatocellular carcinoma, breast cancer, acute myeloid leukemia, etc. (43–45). C188–9 is currently being studied in the several clinical trials (Clinical trial information: NCT03195699, NCT05384119, and NCT05440708). Besides, C188–9 has shown target engagement, no toxicity, and evidence of clinical benefit in a Phase I study in patients with solid tumors (NCT03195699). Two clinical trials were announced in 2022: Phase I evaluation of C188–9 in patients with metastatic hormone receptor–positive (ER⁺) and human epithelial receptor 2–negative (HER2⁻) breast cancer and locally advanced or metastatic unresectable hepatocellular carcinoma are currently underway. Preclinical results indicated that the inhibitory activity and target selectivity of C188–9 still has areas with potential of improvement (46). Thus, at present, highly selective STAT3 inhibitors to treat cancer are still urgently needed.

Here, we described a small molecule, WB436B, directly bound to the STAT3-SH2 domain at particular amino acid residues including Ser611, Glu612, Ser636, Gln635, Tyr657, and Ile659. Further results demonstrated that STAT3 and the above amino acid residues were critical for the antitumor activity of WB436B. In vitro direct binding and kinase assays, together with the cellular STAT3 phosphorylation inhibition experiments, confirmed that WB436B selectively targeted STAT3 over other STAT family proteins and upstream kinases. Notably, the binding affinities (K_D), cellular STAT3 phosphorylation inhibition, and the growth inhibitory activity (IC₅₀) of WB436B were 10- to 100-fold more potent than the STAT3 inhibitors tested in this study.

For pancreatic cancer, the liver is the most common metastatic site and disease with liver metastases currently is incurable (47, 48). Previous reports have demonstrated that blocking STAT3 with ectopic expression of dominant-negative STAT3 markedly suppressed pancreatic tumor liver metastasis (49). Indeed, our results demonstrated that WB436B significantly blocked pancreatic cancer liver metastasis and prolonged tumor-bearing mouse survival. Meanwhile, the mechanism of WB436B inhibiting pancreatic cancer liver metastasis still needs further investigation. Recent studies indicated that the STAT3 signaling pathway was activated in pre-metastatic sites, leading to the formation of a pre-metastatic niche, thereby facilitating tumor metastasis (31). Thus, further experiments are needed to clarify whether WB436B suppressed pancreatic pro-metastatic niche formation and liver metastasis by interfering with the IL6–STAT3 pathway.

Tumor microenvironment in pancreatic cancer is highly relevant to tumor growth, invasion, and metastasis. The PSC cells were the most abundant component of the pancreatic cancer stroma, which secreted leukemia inhibitory factor to activate STAT3 in cancer cells, and STAT3 was the key signaling factor for the cross-talk between PSCs and tumor cells (50). Blocking STAT3 resulted in tumor microenvironment remodeling, stromal formation, and immune cell infiltration (51). In this study, WB436B significantly inhibited tumor growth and metastasis in immunocompetent and immune-deficient mice, indicating that WB436B might enhance its antitumor effects through regulating the microenvironment of pancreatic cancer. Therefore, it is necessary to explore whether WB436B disrupts the interaction between tumor cells and stromal cells or immune cells in the pancreatic tumor microenvironment.

Taken together, this study presents WB436B as a highly potent and selective STAT3 inhibitor that markedly inhibited pancreatic cancer growth and metastasis in vivo and in vitro. Furthermore, WB436B caused complete PDX tumor regression and significantly prolonged the survival of mice in preclinical pancreatic cancer models. Our results laid a solid foundation supporting STAT3 as a promising drug target for pancreatic cancer, and evaluated WB436B as a potential candidate for pancreatic cancer treatment.

H. Chen reports a patent for CN108558848A pending. W. Zhou reports a patent for CN108558848A pending. Y. Chen reports a patent for CN108558848A pending. Z. Yi reports a patent for CN108558848A pending. No disclosures were reported by the other authors.

H. Chen: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. W. Zhou: Investigation, writing–original draft. A. Bian: Formal analysis, investigation, writing–original draft. Q. Zhang: Software, formal analysis, validation, investigation, writing–original draft. Y. Miao: Validation, investigation, writing–original draft, writing–review and editing. X. Yin: Validation, investigation, writing–original draft. J. Ye: Data curation, formal analysis, investigation. S. Xu: Formal analysis, investigation. C. Ti: Formal analysis, investigation. Z. Sun: Formal analysis, investigation. J. Zheng: Writing–original draft. Y. Chen: Supervision, validation, writing–original draft, project administration, writing–review and editing. M. Liu: Supervision, writing–original draft, project administration, writing–review and editing. Z. Yi: Supervision, writing–original draft, project administration, writing–review and editing.

We thank Dr. Stefan Siwko (Texas A&M University, USA) for revising the paper. We thank the staff members of the Large-scale Protein Preparation System at the National Facility for Protein Science in Shanghai (NFPS), Zhangjiang Lab, China for providing technical support and assistance in data collection and analysis.

This work was supported by the grants from National Key R&D Program of China (2018YFA0507001 to M. Liu), National Natural Science Foundation of China (81830083 to M. Liu; 82073310 and 81773204 to Z. Yi; 81973160 to Y. Chen; 81903431 to W. Zhou; 81872418 to Z. Sun), The Science and Technology Commission of Shanghai Municipality (20JC1417900 to Z. Yi; 21S11902000 to Z. Sun; 21S11902100 to J. Zheng; 21S11907800 to Y. Chen), ECNU Construction Fund of Innovation and Entrepreneurship Laboratory (44400-20201-532300/021 to Z. Yi), and China Postdoctoral Science Foundation (2021M691031 to H. Chen).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.