Introduction

The RAS-RAF-MEK-ERK signaling pathway is the most extensively characterized cascade of three mitogen-activated protein kinase (MAPK) pathways in the human body, which transduces the signals from extracellular spaces to intracellular locations and plays a prominent role in cell proliferation, differentiation and survival1,2,3. Approximately one-third of human cancers possess mutations in this pathway3, among which the mutation of B-RafV600E has been identified in ∼34% of malignant melanoma4, ∼58% of papillary thyroid cancer (PTC)5, 4.7% to 10% of colorectal cancer6,7, 12.5% of Grade 1 serous ovarian carcinoma8, ∼94% of papillary craniopharyngioma9 and a wide variety of other cancers10. Compared with other isoforms of the RAF kinase family, A-Raf and C-Raf (also known as Raf-1), the regulation of B-Raf activation requires less molecular events, resulting in increased kinase activity, and its activity is more frequently induced by a single point mutation11. The B-RafV600E mutation has ∼500-fold greater activity compared with wild type in vitro12. Persistent B-Raf activation stimulates cancer cell proliferation and protects cells from apoptosis, which makes it an attractive anticancer drug target. B-RafV600E inhibitors have been heavily investigated by both academia and the pharmaceutical industry. Two small molecule drugs, vemurafenib (PLX4032, brand name: Zelboraf) and dabrafenib (GSK2118436, brand name: Tafinlar), have been approved by the FDA for the treatment of multiple malignant cancers with the B-RafV600E mutation; thus, B-RafV600E is a novel anticancer drug target13,14,15,16. There are some other B-RafV600E inhibitors in clinical trials or at various stages of drug development3. However, new B-RafV600E inhibitors are still urgently required due to unaffordable cost and emerging resistance to present inhibitors16.

All protein kinases contain a segment connecting the amino- and carboxy-terminal kinase catalytic domain, known as the hinge region (Supplementary Figure S1), which could form hydrogen bonding interactions with the adenine ring of ATP, playing a central role in ATP-binding and substrate phosphorylation17,18. Accordingly, hindering ATP from binding to the hinge region should be a straightforward and effective approach to inhibit kinase activity. This approach has been confirmed by its use in numerous successful kinase inhibitors, eg, PD166326 (ABL1 inhibitor), axitinib (ABL1 and VEGFR2 inhibitor), lapatinib (EGFR and ERBB2 inhibitor), and sorafenib (a dual RAF-KDR inhibitor)18,19,20. As a member of the kinase family, apart from the hinge region, the B-Raf catalytic area also includes a glycine-rich loop (G-loop), αC helix, and activation loop (A-loop) (Supplementary Figure S1)12. To our knowledge, the vast majority of B-RafV600E inhibitors bind to the hinge region in kinase catalytic areas and overlap with the ATP-binding region to some extent. One or more critical hydrogen bonds are found between the hinge region of B-RafV600E and the inhibitors, which make a significant contribution to the binding affinity3,15. In accordance with the crystal structures of B-RafV600E and the drugs, both the 7-azaindole moiety of vemurafenib (PDB code: 3OG713, Supplementary Figure S1B) and the 2-amino pyrimidine moiety of dabrafenib (PDB code: 4XV215, Supplementary Figure S1D) form two critical hydrogen bonds with the hinge region, which greatly supports the central role of the hinge-binding area. In our previous work, combining pharmacophore-based virtual screening and scaffold hopping, a series of compounds were identified as B-RafV600E inhibitors. This study showed that forming strong hydrogen bonding with the hinge region results in enhanced IC50 values for the series of inhibitors12.

Fragment-based drug discovery (FBDD) has been widely used in the pharmaceutical industry during recent years, typically as a mainstream alternative to high-throughput screening. With a high hit rate and satisfactory ligand efficiency (LE) of the hits, numerous inhibitors have advanced to clinical trials or have been approved using the FBDD approach21. Vemurafenib is a remarkable representative of successful FBDD cases, which implies the applicability and high-efficiency of FBDD targeting B-RafV600E. The 7-azaindole moiety of vemurafenib was derived from high-concentration screening and was shown to form hydrogen bonding with the hinge area by crystallographic analysis. Another successful FBDD drug is the first approved PPI inhibitor Venetoclax (brand name: Venclexta), which was approved in April 2016 for the treatment of patients with chronic lymphocytic leukemia (CLL). Venetoclax was the first FDA-approved treatment that targets the B-cell lymphoma 2 (BCL-2) protein22. In research, docking-based fragment screening and fragment reassembly, ie, fragment growing, fragment linking and fragment merging, were also applied and exhibited high efficiency in lead discovery and optimization23. In addition to virtual screening programs, computational approaches have been developed to facilitate fragment reassembly. For instance, the AutoT&T2 software suite developed by Wang RX at the State Key Laboratory of Bioorganic and Natural Products Chemistry achieves automatic tailoring and transplanting with the hits from screening, which establishes a complete pipeline of FBDD and provides insight into de novo drug design24. Accordingly, it is clear that appropriate application of FBDD could accelerate the drug discovery process.

In this context, we sought to identify a novel molecular fragment that can bind to the hinge region of B-RafV600E with high affinity and then performed further optimization using the FBDD strategy, as described in Figure 1.

Figure 1
figure 1

Schematic representation of the B-RafV600E inhibitor discovery process with FBDD.

PowerPoint slide

Full size image

Materials and methods

Fragment preparation, molecular docking and assembly

Molecular fragments were derived from the small molecular drugs listed in the top 200 pharmaceutical products by US retail sales in 2011. In consideration of the hinge-binding areas of vemurafenib and dabrafenib, we filtered the fragments generated by Pipeline Pilot 7.5 with the component named Generate Fragments using the following criteria: molecular weight ranges from 50 to 300 and number of heavy atoms ranges from 5 to 1625. Molecular fragments were prepared using LigPrep with all possible protonation states generated at pH 7.0±3.0 by Epik26,27,28. Then, Glide was utilized to perform molecular docking in its SP mode with the post-docking minimization including 10 000 poses per ligand, and the remaining parameters were set to default. The X-ray structure of the B-RafV600E binding by vemurafenib (PDB code: 3OG7) was retrieved from the PDB as the docking structure in this study. To predict the binding modes of the new compounds, molecular docking was performed using Glide in its SP mode in a standard procedure29,30,31. The docked conformations of the molecules with the lowest energy were selected for further studies.

Chemistry

All starting materials and solvents were purchased from commercial suppliers and used without further purification unless otherwise noted. The chemical synthesis of all the designed compounds is fully described in the Experimental Section of the Supplementary Information. The 1H and 13C spectra were obtained on Bruker Avance III (Karlsruhe, Germany) with 300, 400, 500 and 600 NMR spectrometers operating at 300 MHz, 400 MHz or 600 MHz for 1H NMR and 100 MHz or 125 MHz for 13C NMR, respectively. The deuterated solvents, such as CDCl3 and DMSO-d6, were used with the internal standard of tetramethylsilane (TMS). Chemical shifts are provided in δ values of ppm. The abbreviation s indicates singlet, d indicates doublet, t indicates triplet, and m indicates multiplet. Coupling constants (J) were measured in hertz (Hz). The LRMS and HRMS were recorded on a ThermoFinnigan LCQ Deca (San Jose, CA, USA) and a Micromass Q-TOF Ultima (in ESI mode, Manchester, UK) spectrometer, respectively.

Biological evaluation

Cell viability assessment

The human melanoma cell line A375 (B-RafV600E) and human colorectal cancer cell line HCT116 (B-RafWT) were used in the preliminary evaluation of the anti-tumor activity of the compounds using the MTT assay32. The cells were dissociated into single suspension with H-DMEM culture medium containing 10% fetal bovine serum and then seeded into 96-well plates with 2500 (A375) or 3500 (HCT116) cells per well. After 24 h of cultivation, compounds were treated at the initial concentration of 10 μmol/L, and an equal amount of DMSO was used as the vehicle control. Cells were further incubated for 72 h, and then MTT solution was added to the well to a final concentration of 0.5 mg/mL. The plates were incubated at 37 °C for 4 h. The medium was discarded, followed by addition of 100 μL DMSO. The plates were shaken for 10 min and absorbance was detected at 492 nm with a multi-well spectrophotometer.

The sulforhodamine B (SRB) staining assay was performed to detect the activity of compounds on the proliferation of HT-29, COLO205, LOVO, HCT-15, HCT-8, HCT116, SW1116, SW620 and SW480 human colorectal cancer cells. In brief, cells were seeded in 96-well plates at 3000 to 4000 cells/well, followed by treatment of serially diluted compounds for 72 h. The medium was replaced by 10% trichloroacetic acid, and then cells were stained with SRB. After cells being washed with 1% acetic acid, remaining SRB was dissolved in 100 μL of buffer containing 10 mmol/L Tris-base, and the OD value was measured at 560 nm with a multi-well spectrophotometer. The inhibitory rate of cell proliferation was calculated using the formula (ODcontrol–ODtreatment)/ODcontrol×100%. The average IC50 values were from at least three independent tests.

In vitro B-Raf kinase assay

Given that B-Raf catalyzes the phosphorylation of MEK1/2, an ELISA-based MEK1/2 phosphorylation assay was developed and widely used in the evaluation of the kinase activity of B-Raf or the inhibitory activity of B-Raf inhibitors. To determine the IC50 values of compounds in this study at the molecular level, we performed the ELISA-based MEK1/2 phosphorylation assay as previously described33. A sigmoidal dose response curve was generated with duplicate or triplicate measurements at each inhibitor concentration using B-RafV600E or B-RafWT protein. Accordingly, IC50 values were generated. In addition, Western blot assays were also performed to estimate the phosphorylation level of MEK1/2 using a standard procedure.

In vitro ADME profile and solubility

Human liver microsome assays were performed to evaluate in vitro metabolic stability. The concentrations of the parent compound in reaction systems were determined by LC-MS/MS to estimate the stability (the detailed experimental procedures and data analyses are included in the Supplementary Information). Solubility was measured in different buffer solutions using the classical shake test method. Permeability determination was performed using bidirectional permeability assays. In addition, metabolic evaluation with cytochrome P450 was also performed to assess the metabolic stability of the compound.

Results and discussion

Fragment generation and evaluation

Based on the structures of the top 200 drugs, 283 fragments were generated. Taking into account the different protonation states, 429 fragment structures were prepared for docking. All of the fragment structures were then docked against B-RafV600E with one pose output for each structure (Supplementary Table S1). The top 10 fragments with the highest score (Figure 2) all formed hydrogen bonding with the hinge region, except for fragments f3, f6 and f7. In particular, the fragment of pemetrexed (7-deazaguanine) f1 with the highest docking score of −7.920 caught our attention given its 5 hydrogen bond acceptor/donors. We re-docked the fragment to B-RafV600E and output 3 poses to search for more binding modes. All 3 binding poses were located around the hinge region. Of note, in addition to the first pose with a docking score value of −7.920, the second pose, with a docking score of −7.603 and ligand efficiency (the docking score divided by the number of heavy (non-hydrogen) atoms) of −0.691, was predicted to superimpose well with the hinge-binding fragment of vemurafenib (Figure 3A). In addition, we also docked the hinge-binding fragment of vemurafenib, and the second conformation could reproduce the crystal structure of vemurafenib (Supplementary Figure S2). Impressively, the corresponding docking score (−7.742) is slightly lower than f1 (−7.603), but the ligand efficiency (−0.484) is significantly increased compared with the latter (−0.691). Therefore, f1 should be an ideal fragment to replace the 7-azaindole moiety of vemurafenib.

Figure 2
figure 2

Structure of the top 10 molecular fragments ranked by docking score.

PowerPoint slide

Full size image
Figure 3
figure 3

Molecular docking and fragment reassembly identified 1a as a potential B-RafV600E inhibitor. Overlay of (A) f1 (cyan), (B) hinge-binding fragment of vemurafenib (pink) and (D) 1a (green) on vemurafenib (gray) with B-RafV600E (PDB entry 3OG7). The protein is shown in cartoon form and the ligands are in stick form. (C) Molecular tailoring and fragment reassembly.

PowerPoint slide

Full size image

Hit compounds derived from fragment reassembly

Using f1 as an ideal hinge-binding fragment, we sought to replace the hinge-binding area of vemurafenib with f1. To ensure the validity of the aforementioned fragment reassembly strategy, we performed molecular docking using the remaining vemurafenib moieties with the same parameters as the aforementioned fragment docking. As expected, the binding mode similar to that of vemurafenib was ranked third in the docking results with a docking score of −7.368 (Figure 3B). In this context, we found that no fragment apart from f3 could bind to B-RafV600E at the same position with a higher docking score than −7.368. For synthesis, we replaced the 7-azaindole moiety of vemurafenib and obtained compound 1a (Figure 3C). The docking study revealed that 1a is a potent inhibitor of B-RafV600E given its similar docking score and binding mode as vemurafenib (Table 1 and Figure 3D). Indeed, primary biological evaluation indicated that the IC50 of 1a was 0.80 μmol/L against the A375 cell line, whereas that of vemurafenib is 0.56 μmol/L. Given the availability of the reagents in our laboratory, compounds 1b-1n were designed and synthesized for exploring the structure-activity relationship (Table 1). Among the new compounds, 1m was predicted to have similar B-RafV600E inhibitory potential as 1a.

Table 1 Structures, docking scores and biological activities of compounds.
Full size table

Chemistry

The synthetic route of deazapurine derivatives studied in this study is outlined in Figure 4. 2-Bromo-1,1-dimethoxyethane was hydrolyzed under concentrated hydrochloride to obtain bromoacetaldehyde, which was then heated to condense 2,6-diaminopyrimidin-4(3H)-one (2) to give compound 3 under sodium acetate. Compound 3 was acetylized under acetic anhydride to obtain 5a, which was then hydrolyzed to give 6a under ammonium hydroxide without further purification. Compound 3 was methylated to obtain 4 under dimethyl sulfate, which was acetylized to achieve 6b via 5b (Figure 4A). 2,6-Difluorobenzoic acid (7) was nitrified to obtain 8 under the mixture of nitric acid and sulfuric acid, and 8 reacted with methanol under a catalytic amount of concentrated sulfuric acid to form 9, which was reduced by iron powder in the solvent of acetic acid to obtain 10. Then, 10 reacted with two equivalents of sulfonyl chloride to afford 11, which was hydrolyzed under the aqueous solution of NaOH to achieve 12. The derivatives of benzoic acid (8, 12a and 12b) were chloridized to afford the corresponding acyl chlorides (13a-c) (Figure 4B). Compounds 1a-d, 1m and 1n were synthesized from the fragments obtained from procedures A and B (6 and 13a-c) via the Friedel-Crafts reaction in the presence of AlCl3 (Figure 4C). Compound 14 was hydrolyzed by NaOH aqueous solution to achieve 15, which was then protected by Boc to give 16. Compounds 17a-c were obtained by the reaction of 16 with CH3I, 4-nitrobenzyl bromide or allyl bromide, which was deprotected to give the important intermediates 18a-c (Figure 4D). Compounds 1e-l were synthesized from the fragments (15 or 18a-c and 13a-c) obtained from the procedures B and D via the Friedel-Crafts reaction in the presence of AlCl3 (Figure 4E).

Figure 4
figure 4

Synthesis of deazapurine derivatives. Reagents and conditions: (a) Bromoacetaldehyde dimethyl acetal, HCl, sodium acetate trihydrate, 80 °C, over night; (b) Acetic anhydride, acetic acid, 130 °C, 2 h; (c) 0.1 mol/L NaOH, dimethyl sulfate; (d) Acetic anhydride, acetic acid, 130 °C, 2 h; (e) Ammonium hydroxide, methanol; (f) 65%–68% HNO3, H2SO4; (g) Conc sulfuric acid, methanol; (h) Acetic acid, ethanol, Fe powder, 110 °C, 0.5 h; (i) Triethylamine, 3-chloropropylsulfonyl chloride, 3.5 h; (j) NaOH (aq) THF, reflux, 2 h; (k1) SOCl2, toluene, reflux, 3 h; (k2) Oxalyl chloride, cat. DMF, DCM; (l) AlCl3, MeNO2, 60 °C, 80–90 °C or 100–105 °C, over night; (m) Acetic acid, ethanol, Fe powder, 110 °C, 0.5 h; (n) NaOH (aq) 80 °C, over night; (o) Triethylamine, DMAP, Boc2O, THF, RT, over night; (p1) KF, MeI, acetonitrile; (p2) Anhydrous DME and DMF, NaH, LiBr, 4-Nitrobenzyl bromide or allyl bromide; (q) Conc HCl, RT, over night; (r) AlCl3, MeNO2 or nitrobenzene, 60 °C, 80–90 °C or 100–105 °C, over night; (s) AlCl3, MeNO2, 60 °C, over night; (t) Acetic acid, ethanol, Fe powder, 110 °C, 0.5 h.

PowerPoint slide

Full size image

Structure-activity relationship study

Vemurafenib forms 3 hydrogen bonds with residues D594, F595 and G596 by the sulfonamide moiety (Figure 5A). If 1a interacts with B-RafV600E in a similar manner, replacing the sulfonamide moiety with amino (1b) should lead to loss of the inhibitory activity. The nitro substituted derivative 1c exhibited weak inhibitory activity, but the selectivity against B-RafWT disappeared, which is consistent with the notion that the deprotonated sulfonamide favors an interaction with the B-RafV600E mutation other than the wild type34. Using 2,6-difluorobenzenesulfonamido, the fragment of dabrafenib substituted at the 1'-position, 1d, also exhibited weak inhibitory activity on A375 cells (Table 1).

Figure 5
figure 5

The binding mode of B-RafV600E inhibitors (docked with Glide in SP mode, PDB entry 3OG7). (A) Crystal structure of vemurafenib (orange) binding to B-RafV600E. (B) Predicted binding mode of 1a (green) to B-RafV600E. (C) Predicted binding mode of 1m (magenta) to B-RafV600E. (D) Superimposition view of all the three compounds. The protein is show in cartoon form with critical residues and the inhibitors in stick form.

PowerPoint slide

Full size image

Compared with vemurafenib, 1a was predicted by a docking study to form one more hydrogen bond with C532 on the hinge region via the amino of 7-deazaguanine (Figure 5B). To explore how the hydrogen bonding contributes to the binding affinity, we replaced the amino groups of 1a, 1b, 1c and 1d with a chlorine atom, resulting in 1e, 1f, 1g and 1h, respectively. As expected, the activities of chlorine-substituted compounds became weaker compared with the amino-substituted compounds except for 1g, whose inhibitory activity was extremely inconsistent with the docking score (Table 1). In view of the structure of 1g, we hypothesized that the activity of 1g in cells was not achieved by targeting B-RafV600E given that its selectivity between A375 and HCT116 cell lines was lost.

On account of the similar binding mode of 1a and vemurafenib, increased substitution at the 3-position was proposed as rational (Figure 5D). From the organic synthetic perspective, methyl, allyl and 4-nitrobenzyl were added to 1e, leading to 1i, 1j and 1k, respectively. The activities of the three compounds were all improved compared with 1e. Moreover, 1i had the best performance with respect to both activity and selectivity. The methyl substituted derivative of 1h, 1l, also had a higher inhibition rate, implying that methyl substitution is applicable. Then, we synthesized two additional methyl-substituted derivatives, 1m (Figure 5C) and 1n. Compound 1n exhibited increased activity against the A375 cell line compared with 1d, whereas 1m had the same IC50 value as 1a (0.80 μmol/L). At the molecular level, the IC50 values of 1a and 1m on B-RafV600E were 0.50 μmol/L and 0.05 μmol/L, respectively, whereas that of vemurafenib was 0.13 μmol/L. In addition, the selectivity of 1m (9 times) for B-RafV600E was increased compared with vemurafenib (5 times) against B-RafWT (Table 1). Accordingly, we obtained a potential B-RafV600E inhibitor with comparable activity and better selectivity than vemurafenib.

Assessment of 1m against multiple colorectal cancer cell lines

For the high proportion (4.7%–10%) of colorectal cancer patients bearing the B-RafV600E mutation, we used multiple colorectal cancer cell lines to evaluate the inhibitory activity and selectivity of 1m (Table 2). The IC50 values of 1m were at the micromolar level in B-RafV600E mutant HT-29 and COLO205 cells, and these values were comparable to vemurafenib. However, the IC50 values of 1m were over 10 μmol/L in LOVO, HCT-15, HCT-8, SW1116, HCT116, SW620 and SW480 cells, which harbor wild type B-Raf. The IC50 values of vemurafenib in LOVO, HCT-15 and HCT-8 cells were 8.88 μmol/L, 7.75 μmol/L and 5.84 μmol/L, respectively, indicating that 1m exhibits superior selectivity against the V600E mutated cells compared with vemurafenib. A molecular docking study demonstrated that 1m has a similar binding mode as vemurafenib. Furthermore, the sulfonamide moiety of vemurafenib is retained in 1m, suggesting that the moiety preferred binding to the V600E mutant34, accounting for the selectivity of 1m. In addition, the 7-deazaguanine moiety of 1m forms more interactions with B-RafV600E than vemurafenib, which might be another reason why 1m has lower activity against wild type B-Raf.

Table 2 The selectivity of 1m for B-RafV600E over B-RafWT in multiple colorectal cancer cell lines.
Full size table

The phosphorylation level of MEK1/2, which are the downstream kinases of B-Raf, was determined in HT-29 cells to validate B-RafV600E inhibition by 1m at the cellular level. After incubation with serially diluted 1m or vemurafenib for 2 h, the phosphorylation level of MEK1/2 was measured by Western blot. Phosphorylated MEK1/2 significantly decreased with increasing concentrations of 1m (Figure 6), confirming its target of B-RafV600E in HT29 cells. Similar observations were obtained with vemurafenib.

Figure 6
figure 6

The phosphorylation level of MEK1/2 in HT-29 cells was detected with Western blot.

PowerPoint slide

Full size image

In vitro ADME properties

Finally, we determined the solubility, permeability and metabolic stability of 1m. The human liver microsome assay indicated that the remaining rate (proportion of compounds that are not decomposed) of 1m decreased slightly from 30 min to 2 h, implying that 1m was stable in human liver microsomes (Supplementary Table S2 and Figure 7). Moreover, the metabolic stability evaluation concerning cytochrome P450 indicated that the metabolic bioavailability of 1m was greater than 90% and varied slightly with different species. Furthermore, 1m also exhibited good solubility, whereas the minimal soluble concentration of 1m in buffers A and B was 50 μmol/L and that in buffer C was 20 μmol/L. Regarding permeability, given that 1m might be a substrate of P-gp (efflux ratio was 67.4), the in vivo absorption and bioavailability of 1m should be higher than the in vitro predictive value (mean Fabs was 35%) for the saturation of the efflux transporter (Table 3). Accordingly, 1m should have good ADME properties, including metabolism and bioavailability.

Figure 7
figure 7

TMetabolic stability of 1m in human liver microsomes. Midazolam was a positive control and negative control was ultrapure-water rather than NADPH.

PowerPoint slide

Full size image
Table 3 Solubility, permeability and metabolic stability of 1m.
Full size table

Conclusion

In view of the structure and interaction analyses of approved drugs and inhibitors in development, we utilized molecular tailoring, molecular docking and fragment reassembly strategies to design and synthesize a series of deazapurine derivatives, among which compound 1m was identified as a potential B-RafV600E inhibitor with comparable activity and superior selectivity compared with vemurafenib. Moreover, 1m exhibits significant activity against the proliferation of HT-29 colorectal cancer cells harboring B-RafV600E with an IC50 value of 0.55 μmol/L, which is lower than that of vemurafenib (0.69 μmol/L), while sparing B-RafWT cells (IC50>10 μmol/L). An obvious decrease in the phosphorylation of MEK1/2 was observed upon treatment with 1m, confirming that 1m inhibited B-RafV600E in cells. In addition, 1m also exhibits good solubility, bioavailability and metabolic stability. These results of 1m lay a solid foundation for further development. In addition, this study provides a novel strategy for designing inhibitors targeting B-RafV600E.

Author contribution

Wei-liang ZHU, Bo LI, He-yao WANG, Ling-hua MENG, and Jian DING conceived and designed the research; Gui-min WANG, Xiang WANG, Jian-ming ZHU, Bin-bin GUO, and Zhuo YANG performed the research; Gui-min WANG, Xiang WANG, Jian-ming ZHU, Bin-bin GUO, and Zhi-jian XU analyzed the data; Gui-min WANG, Wei-liang ZHU, Jian-ming ZHU, Ling-hua MENG, and He-yao WANG wrote the paper.