Synthesis and biological evaluation of celastrol derivatives as potent antitumor agents with STAT3 inhibition

Abstract Using STAT3 inhibitors as a potential strategy in cancer therapy have attracted much attention. Recently, celastrol has been reported that it could directly bind to and suppress the activity of STAT3 in the cardiac dysfunction model. To explore more effective STAT3 inhibiting anti-tumour drug candidates, we synthesised a series of celastrol derivatives and biologically evaluated them with several human cancer cell lines. The western blotting analysis showed that compound 4 m, the most active derivative, could suppress the STAT3’s phosphorylation as well as its downstream genes. SPR analysis, molecular docking and dynamics simulations’ results indicated that the 4m could bind with STAT3 protein more tightly than celastrol. Then we found that the 4m could block cell-cycle and induce apoptosis on HCT-116 cells. Furthermore, the anti-tumour effect of 4m was verified on colorectal cancer organoid. This is the first research that discovered effective STAT3 inhibitors as potent anti-tumour agents from celastrol derivatives.


Introduction
Celastrol (CEL, also named tripterine), is a natural friedelane pentacyclic triterpenoid isolated from Tripterygium wilfordii 1 . It exhibits multiple pharmacological potentials, including anti-cancer, antiinflammatory, and anti-obesity effects 2,3 . In 2007, celastrol was voted as one of the five most promising natural products for turning traditional medicines into modern drugs in Cell 4 . Scientists discovered new mechanisms of celastrol's tumour-suppressing effects in recent decades by identifying various new pathways and key targets 5,6 . For instance, celastrol was found to suppress growth and induce apoptosis of human hepatocellular carcinoma through the STAT3/JAK2 signalling pathway 7 . Other than that, structural modifications of celastrol also made significant progress, and various celastrol derivatives have been synthesised with biological activity evaluation [8][9][10] .
STAT3 is a cytoplasmic transcription factor that modulates many genes' transcription activity to regulate critical biological functions, including cell proliferation, differentiation, cell survival, angiogenesis, and immune response 11 . Abnormal STAT3 activation (such as phosphorylation) plays a crucial role in tumour cell proliferation 12 . Identifying specific and potent STAT3 inhibitors as a candidate substance in cancer therapy has attracted much attention [13][14][15][16] . Recent research showed that celastrol could inhibit Ang II-induced cardiac dysfunction via directly binding to STAT3 instead of the upstream mediator 17 . Thus, STAT3 is speculated to be one of the main targets of celastrol for suppressing tumours. Furthermore, it is theoretically feasible to screen novel small molecule inhibitors of STAT3 from celastrol derivatives.
Organoids are microscopic self-organising, three-dimensional (3 D) structures grown from stem cells in vitro 18 . These amazing 3D constructs represent a promising, physiologically similar model for human cancers, which laid a foundation for various experimental applications in cancer research 19 . Every organoid culture derived from each patient showed incredible resemblance to the original tumours, making it the most clinically relevant and representational model for novel medicine development research 20 . In comparison, traditional pre-clinical models such as cell lines and animal models do not truly represent the original human tumour, making clinical experiments have a greater chance of failure and poor therapeutic performance 21 . So patient-derived organoids are widely considered to serve as a better platform for anti-tumour candidate drug screening and efficacy evaluation in recent years.

Cytotoxic assay in vitro
The cytotoxic activities of compounds were evaluated against A549, HCT-116 (cultured in RPMI 1640), HepG2 (cultured in DMEM-high glucose) by CCK-8 method. For this assay, 100 lL (5 Â 10 3 /mL) cells per well were seeded in 96-well plates and allowed to incubate for 24 h. Then, the compounds with different concentrations (0.02, 0.1, 0.5, 2.5, 12.5 lM) were added. After 24 h of incubation, CCK-8 solution (10%) was added, and the plates were incubated again for another 4 h at 37 C. The OD values at 450 nm was immediately read by a microplate reader (Tecan Infinite M200 Pro). Subsequently, the IC 50 values were calculated by Graphpad Prism 5. Three independent experiments were performed. Data are presented as the mean ± SD (n ¼ 3). By using the same method, 4m's toxicity to normal colon cell was detected in NCM-460 cells.

In silico ADMET analysis
The ADMET properties of the compounds were predicted using Qikprop module in Schr€ odinger 2009 platform (Schr€ odinger, LLC, New York, NY) with the default settings.
Western blot analysis HCT-116 cells were seeded (1 Â 10 7 /mL, 2 ml/orifice) in 6-well plates and incubated for 24 h. The cells were serum-starved. Then celastrol or 4m with specific concentrations were added for 6h incubation. After 6 h, cells were stimulated by IL-6 (25 mg/mL). The cells were harvested after 30 min and then the proteins were extracted with lysis buffer and quantified by BCA method. Each sample was separated on 10% SDS-PAGE and transferred to PVDF membrane (Millipore). The membranes were blocking and sequentially incubated with primary and secondary antibodies diluted in 5% bovine serum albumin (BioSharp). Subsequently, the bands were detected in a gel imaging analysis system ((Bio-Rad. ChemiDoc XRS).

SPR analysis
Purified rhSTAT3 protein (optimum pH ¼ 5.5) was immobilised on sensor chip CM5 (GE Healthcare) and carried out at 25 C on Biacore T200 instruments (GE Healthcare). Approximately 14937 RU of STAT3 was amino coupled to a CM5 Chip (according to the manufacturer's protocol), and another cell being left blank for reference subtraction. HBS-EP buffer was used as running buffer. The test compounds (10 mM, DMSO) were diluted to 3.91 lM, 7.81 lM, 15.63 lM, 31.25 lM, 62.5 lM, 125 lM and 250 lM using PBS buffer with 5% DMSO. The operating conditions: contact time: 80 s, flow rate: 30 lL/min, dissociation time: 180 s. The ratio of the association and dissociation rate constants was determined as theaffinity (equilibrium constants, K D ). The K D (equilibrium constant) values were calculated as k d (dissociation rate constant)/k a (associated rate constant) for each interaction and determined by globally fitting with Biacore T200 evaluation software 2.0 (GE Healthcare).

Molecular docking
The crystal structure of the STAT3 homodimer (PDB entry: 1BG1) was downloaded from the Protein Data Bank (PDB). The molecular docking studies were performed using the flexible docking protocol in Discovery Studio 2019. Small molecules 4m and celastrol were prepared with full minimisation. Protein STAT3 was processed with clean protein and the active site was defined based on the key residues of the pTyr705 pocket (LYS-591, ARG-609, SER-611, GLU-612, SER-613) 22 and set the radius of the sphere as 12.0 Å . We also selected the residues LYS-591, ARG-609, SER-611, GLU-612, SER-613 for flexible processing. For other parameters, we kept them as default settings in flexible docking.

Molecular dynamics simulations
Molecular dynamics (MD) simulations were performed to study the stability of molecular docking between 4m and CEL with STAT3 in nanosecond scales for the duration of 100 ns. NAMD 2 (NAnoscale Molecular Dynamics) software was used with default parameters and procedures. The NPT ensemble with a temperature of 300 K and a pressure of 1 bar was applied to all the runs. The simulation length was 100 ns with relaxation time one ps for the ligand with STAT3. The docked complex system was solvated by a pre-equilibrated simple point charge in an orthorhombic box with a box wall distance of 10 Å. The system was neutralised by adding a salt concentration of 0.15 M salt (NaCl). The generated trajectories were then aligned and analysed by calculating their root mean square deviations (RMSD).
Cell-cycle and apoptosis assay HCT-116 cells (log phases) were seeded in 6-well plates, and preincubated at 37 C for 24 h. Then cells were further incubated with DMSO (blank control) or 4m at different concentrations (1, 2.5, 5 lM) for 48 h. After the respective period of incubation, cells were harvested softly and washed twice with cold PBS, and collected by centrifugation. For apoptosis analysis, after centrifugation, the cells were resuspended with 1 Â binding buffer solution and partial shipments to 100 lL. Then PI and Annexin V-FITC were added and the mixture was incubated at room temperature for 15 min in the darkness. Finally, the mixture was diluted with 1 Â binding buffer and analysed by flow cytometry (Agilent Technologies, NovoCyte 3000) at 488 nm. NovoExpress software was used to analyse and generate the double dispersion point diagram. For cell-cycle analysis, the cells were fixed in 70% ethanol at 4 C overnight.
After washing with PBS, cells were suspended in PBS containing 50 mg/ml PI and 100 mg/mL RNase A, incubated at 37 C for 30 min, and protected from the light. Then the fluorescence intensity was measured at 488 nm by flow cytometry.

Activity verification on colorectal cancer organoids
Human specimens Malignant and primary colorectal tissues were collected through Nanjing Hospital of Chinese Medicine affiliated to Nanjing University of Chinese Medicine from patients undergoing surgical procedure at Colon and Rectal Surgical department. All experiments were reviewed and approved by the Ethics Board of Nanjing Hospital of Chinese Medicine and performed under protocols. Written informed consent form for research was acquired from donors prior to sampling procedure. Samples were acquired from adult patients who were treatment-naive. Pathological analysis of all samples was performed at department of pathology, Nanjing Hospital of Chinese Medicine.

Chemistry
The synthetic routes of celastrol derivatives are outlined in Schemes 1-4. The structures of the synthesised compounds were elucidated by 1 H NMR, 13 C NMR, and HR-MS. All spectral data were in accordance with the assumed structures.
Firstly, celastrol was esterified with the halides on the carboxyl to obtain the four C-20 ester derivatives 1a-1d (Schemes 1-2). Secondly, on the basis that the carboxyl group has been protected, the C-2 carbonyl group of 1a was reduced by NaBH 4 , and then directly esterified with cinnamic acid or other carboxy-containing compounds to obtain 2a-2g. After that, their allyl group was de-protected by Pd[P(C 6 H 5 ) 3 ] 4 respectively to obtain a series of novel A-ring double-substituted modified derivatives 3a-3g (Schemes 1). In addition, 2h was obtained through the methylation of 1a with methyl iodide (Schemes 1). 3h and 3i were another two A-ring disubstituted derivatives produced by the reaction of celastrol with corresponding anhydride after C-2 carbonyl group reduction (Schemes 3). Finally, fourteen C-20 amide derivatives 4a-4n were synthesised refer to the literature method 23 (Schemes 4). Among them, 2a-2h, 3a-3h, 4d, 4f, 4h-4n were new synthesised compounds.

Cytotoxic assay in vitro
According to the in vitro results (Table 1), the C-20 ester derivatives 1a-1d showed more or less cytotoxicity compared with celastrol, except that 1d have no effect on A549 under the test concentration. Almost all the C-20 amide derivatives except for 4l showed equivalent or improved activities to at least one of the three cell lines. In particular, the antiproliferation effect of 4a, 4d, 4g, 4j, and 4m was higher than celastrol in all the screened cell lines.We speculated that the C-29 carboxyl group contributed finitely to the celastrol's anti-tumour activity, consistent with previous studies' conclusions 23,24 . Thus, the structural modification of E-ring for derivatives with enhanced activity worth further investigation. In addition, the cytotoxic assay indicated that the most active compound 4m showed lower toxicity to normal colon cells NCM-460 (IC 50 ¼1.76 ± 0.42 lM) when compared with colon cancer cells HCT-116 (IC 50 ¼ 0.61 ± 0.07 lM).
No matter the carboxyl group is protected or not, the A-ring disubstituted derivatives 2a-2g and 3a-3g had no antiproliferation effect below the maximum test concentration (12.5 lM), indicating the quinone methyl structure of celastrol is vital for activity. Significantly, celastrol was considered to affect protein function by forming covalent Michael adducts via binding to the electrophilic sites on quinone methide rings (A/B rings) of celastrol and nucleophilic thiol groups of cysteine residues [25][26][27] . It suggested us that preserving the unique quinone methyl structure of celastrol should be prioritised during structural modification. However, another two disubstituted compounds (3h and 3i, with smaller substituent groups) retain similar activity as celastrol, and the previous study 28 reported similar results, which contradicts our conclusion. Therefore, the structure-activity relationship of A-ring modification needs to be further analysed.

In silico ADMET analysis of physicochemical and pharmacokinetic parameters
To evaluate active compounds' potential for novel drug development, pharmacokinetic properties such as absorption, distribution, metabolism, excretion, and toxicity (ADMET) profiling of celastrol derivatives were determined using ADMET Predictor software. The obtained results are presented in Tables S1 in Supplementary Information. For the most active compounds 4m, the lipophilicity was acceptable with the predicted octanol/water partition coefficient (logPo/w ¼ 6.493). The predicted apparent Caco-2 cell permeability (PCaco) of 4m in nm/s was 570.622, which means the ability to cross gut-blood barrier for non-active transport was great. Meanwhile, compound 4m was expected to easily get across the blood-brain barrier, since the predicted brain/blood partition coefficient (logBB) was À0.855 and the predicted apparent MDCK cell permeability (PMDCK) in nm/sec was 628.767. Moreover, compound 4m was predicted to be metabolised smoothly while it may undergo four kinds of metabolic reactions in vivo. In addition, the predicted skin permeability (logKp) of 4m was À2.491 and the predicted human oral absorption was greater than 80%. All above indicated that 4m has good physicochemical properties and drug development potential.  Compound 4m inhibited the phosphorylation of STAT3 at site pTyr705 Since celastrol had been confirmed as a STAT3 targeted inhibitor 17 , it is highly plausible that its derivatives with similar structures could function as novel STAT3 inhibitors. To further investigate whether the anti-tumour effects of 4m were related to inhibition of STAT3 phosphorylation, we evaluated the level of p-STAT3 stimulated by IL-6 in HCT-116. As displayed in Figure 1, 4m could effectively decrease the level of p-STAT3 at site pTyr705 in a concentration dependent manner, and the function of 4m was stronger than CEL at 1.0 lM. However, the total level of STAT3 was not changed with 4m treatment. These results indicated that 4m could effectively inhibit the phosphorylation of STAT3 and this may be the main mechanism by which 4m exerts its antitumour effect.

Compound 4m inhibited the expression of the downstream gene of STAT3
To further study the effects of 4m on the STAT3 pathway, we examined the expressions of STAT3-targeted genes (Survivin and Mcl-1). As shown in Figure 2, compound 4m down-regulated the level of Survivin and Mcl-1. These results preliminarily indicated that 4m could inhibit the activity of STAT3.
Compound 4m had no effects on the typical upstream kinases JAK2 As a signalling pathway closely related to the occurrence and development of tumours, the phosphorylation of STAT3 was usually mediated by p-JAK2, which is one of the typical STAT3 upstream tyrosine kinases 29 . To evaluate whether the inhibitory effect of 4m on p-STAT3 is regulated by upstream JAK2, corresponding western blot was carried out. It was evident in Figure 3   that the levels of p-JAK2 and total proteins were not affected by 4m, illustrating that 4m might inhibit the phosphorylation of STAT3 by directly binding to STAT3 protein.
SPR analysis of CEL and 4m to recombinant human STAT3 protein (rhSTAT3) Surface plasmon resonance (SPR) analysis detected the direct interaction between celastrol and STAT3 17 . In order to test the hypothesis that 4m inhibits the phosphorylation of STAT3 through targeting STAT3 directly, we used SPR analysis to evaluate the interaction at the molecular level and celastrol (CEL) as a positive control. We observed that both CEL and 4m could interact with the rhSTAT3 protein and the response unit (RU) values were proportional to compound concentrations within the selected ranges (Figure 4). According to the fitting calculation results, compound 4m showed a slightly stronger binding affinity with a lower K D value (the equilibrium dissociation constant) of     45.33 mM, when compared with CEL (K D ¼ 60.38 mM). In addition, the representative A-ring disubstituted derivative 3g was also tested with SPR analysis (See Figure S1 in Supplementary  Information). Compound 3g showed no creditable binding to STAT3 protein. And this might be the reason that 3g and other A-ring disubstituted derivatives appear no effect in cell experiments. In general, the SPR experiment provides dependable evidences for the conception of screening novel STAT3 inhibitors from celastrol derivatives.
Molecular docking between 4m, CEL, and STAT3 SH2 domain As mentioned above, 4m can bind to rhSTAT3 protein and inhibit the phosphorylation of STAT3 at site pTyr705. To further investigate the binding sites of 4m, molecular docking was performed. As illustrated in Figure 5(A), multiple strong hydrogen bonds were formed between 4m and the residues LYS-591, ARG-609, GLU-612, SER-613, ILE-634, ARG-595 of STAT3 protein. At the same time, 4m had hydrophobic interactions with the residues TRP-564, ILE-589, PRO-639, VAL-637, SER-636, GLN-635, GLN-633, LYS-591. Compared with 4m, the hydrogen bond and hydrophobic interaction between CEL and STAT3 protein are much less ( Figure 5(B)), suggesting that 4m binds to the SH2 domain more tightly than CEL, thereby inhibiting the phosphorylation of STAT3. These results were consistent with SPR analysis. Furthermore, it can be seen in Figure 5(C,D), 2-thiopheneethylamine moiety of 4m was just inserted into the hole composed of residues SER-611, GLU-612, ILE-589, LYS-591 to enhance the affinity, indicating the size of this part just matches the size of the hole, which points out the direction for the future structural modification of CEL.

Molecular dynamics simulations of the docking modes
Based on docking results, 100 ns MD simulations were carried out to estimate the dynamic stability of the docking modes of 4m and CEL ( Figure 6). Average RMSD of STAT3-CEL system increased from 1.3 to 2.8 Å around 15 ns and then increased to 3.5 Å at around 20 ns to 100 ns slowly and undulating. The average RMSD of STAT3-4m increased from 1.3 to 2.3 Å around 10 ns and then increased to 2.8 Å at around 10 ns to 45 ns and then reaches equilibrium from 45 ns and converges to 2.8 Å, at 100 ns. The average RMSD plot of STAT3-4m complex simulation has shown less deviation. It is consistently stabilised around 2.8 Å and signifies less structural deviation than the STAT3-CEL complex, supporting structural stability.

Effects of 4m on cell-cycle distribution and apoptosis
It has been reported that celastrol induces cell-cycle arrest and apoptosis in different cancer cell lines 30,31 . And there is more evidence showing that STAT3 activity has critical role in various carcinogenic processes such as preventing apoptosis and regulating cell-cycle 32 . Therefore, compound 4m was selected for the characterisation of its effect on cell-cycle distribution and apoptosis on HCT-116. As observed in Figure 7(A), compound 4m could induce apoptosis of HCT-116 cells in a concentration dependent manner compared with the control conditions (DMSO). Furthermore, compound 4m also had a retardation effect on the cell-cycle distribution of HCT À116 cells compared with the control conditions. Moreover, low concentration of 4m mainly arrested cell cycle at G2/M phase while 4m arrested cell cycle in both S and G2/M phase at high concentration (Figure 7(B)). Hence, these results suggest that growth inhibition of the HCT-116 cells treated by 4m may be related to induction of cell-cycle arrest and apoptosis.

Effects of 4m on colorectal cancer organoids
Four colorectal cancer organoid cultures (CCO-1, CCO-2, CCO-3, and CCO-4) from humans were used to further investigate the activity of representative derivative 4m and assess its potential in clinical application. The inhibitory effect of 4m on human colorectal cancer organoids (CCOs) was evaluated. The commonly used drug oxaliplatin (L-OHP) in clinical was chosen as a control. Their cytotoxicity was also tested on colorectal normal organoid (CNO). As the results, 4m showed much better activity than the positive drug L-OHP on all the colorectal cancer organoids (Figure 8). While L-OHP gradually emerges drug resistance in clinical application 33 , celastrol derivative 4m would offer alternative as drug candidate. However, 4m also had certain toxicity to CNO, which need to be taken into account before clinical application. As an available strategy, nano based drug delivery systems are functional to achieve optimal efficacy and reduce toxicity 34 which can offer more opportunities for the translational development of celastrol and its derivatives.

Conclusions
In this study, we reported the design and synthesis of a series of celastrol derivatives including 2,3-disubstitution in A-ring, esterification and amidation of C-20 carboxy group in E-ring. Some of the C-20 amides derivatives (4a, 4d, 4g, 4j, and 4m) showed improved anticancer activity against all the tested cancer cell lines compared with celastrol. The SAR was discussed and the preliminary results showed the possible negative effect of A-ring modifications of celastrol analogs on anticancer activity. Among all the synthesised derivatives, compound 4m was the most potent, highly active on all tested tumour cell lines. Mechanism studies showed that 4m exerts a significant inhibitory effect on STAT3 phosphorylation as well as the downstream genes without affecting the activated JAK2 and showed stronger binding affinity with STAT3 protein than celastrol. And the docking and molecular dynamics studies revealed that 4m could bind to the SH2 domain of STAT3 protein. Furthermore, the flow cytometry test showed that the potent anti-tumour activity of compound 4m on HCT-116 cells might be mediated by inducing cell-cycle arrest and apoptosis. Finally, we further verified the activity of 4m on colorectal cancer organoids. These results indicated the potential of celastrol derivatives to be promising STAT3 inhibitors in the cancer treatment.