K-Ras^G12D has an allosteric small molecule binding site

Abstract

KRAS is the most commonly mutated oncogene in human cancer, with particularly high mutation frequencies in pancreatic cancers, colorectal cancers and lung cancers¹. The high prevalence of KRAS mutations and its essential role in many cancers makes it a potentially attractive drug target; however, it has been difficult to create small molecule inhibitors of mutant K-Ras proteins. Here, we identified a small molecule binding site on K-Ras^G12D using computational analyses of the protein structure, and then used a combination of computational and biochemical approaches to discover small molecules that bind to this pocket, which we have termed the P110 site, due to its adjacency to proline-110. We determined that one compound, named K-Ras Allosteric Ligand KAL-21404358, bound to K-Ras^G12D, as measured by microscale thermophoresis (MST), thermal shift assay (TSA), and nuclear magnetic resonance (NMR) spectroscopy. This compound impaired the K-Ras^G12D interaction with B-Raf, and disrupted the RAF-MEK-ERK and the PI3K-AKT signaling pathway. We synthesized additional compounds, based on the KAL-21404358 scaffold with more potent binding and greater aqueous solubility. In summary, these findings suggest that the P110 site is a promising pocket for binding of small molecule allosteric inhibitors of K-Ras^G12D.

INTRODUCTION

Ras proteins belong to the small GTPase family and are involved in transmitting growth, survival, and proliferation signals within cells. As a GTPase, Ras cycles between a GTP-bound state and a GDP-bound inactive state, the transition of which is regulated by guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs)². Two regions of Ras proteins, Switch I (residues 30–40) and Switch II (residues 60–76), undergo substantial conformational changes and form effector-protein-interaction surfaces upon GTP binding³. In the GTP-bound active state, Ras interacts with effector proteins and activates downstream cellular signal transduction pathways, including the RAF-MEK-ERK, the PI3K-AKT-mTOR and the RalGDS pathways⁴. Oncogenic mutants of Ras are locked in an active signaling state: the constitutive activation of Ras downstream signaling results in sustained proliferation, metabolic reprogramming, inhibition of apoptosis, and other hallmarks of cancer⁵.

There are three common RAS genes in humans—KRAS, HRAS, and NRAS. The frequencies and distribution of RAS gene mutations are not uniform among these three family members⁶. KRAS is the most frequently mutated gene, and is altered in 86% of RAS-altered cancers. G12, G13, and Q61 are three hotspot point mutations found around the RAS GTP-binding site. Among these mutations, G12D mutations are predominant in pancreatic ductal adenocarcinoma, and colon and rectal carcinomas⁷. The high prevalence of KRAS mutations in cancers suggests it may be a potentially valuable drug target. However, there are still no effective inhibitors directly targeting K-Ras mutant proteins that are suitable for clinical use.

K-Ras is considered a challenging drug target for two main reasons. First, there does not seem to be a deep, hydrophobic pocket on the surface of K-Ras suitable for potent and selective small molecule binding; the only notable binding pocket on K-Ras is the nucleotide-binding pocket, which binds GTP/GDP with picomolar affinity, making it an impractical target site for small molecule drugs⁸. Second, K-Ras, like roughly 85% of other human proteins, exerts its biological effects via protein-protein interactions, which are often difficult to disrupt with small molecules, due to their large surface areas and the diffuse nature of the interactions between them⁹. Despite these difficulties, direct K-Ras inhibitors have been explored using several strategies—(1) targeting G12C-specific K-Ras mutants with covalent, cysteine-reactive electrophilic inhibitors^10–12, (2) blocking K-Ras-effector interactions by developing small-molecule and peptides inhibitors^13–15, (3) interrupting nucleotide exchange, including the K-Ras-GEF interaction and modification of the GTP-binding site^16–18, and (4) targeting potential allosteric regulatory sites^{19, 20}.

Here, we described a strategy to target oncogenic K-Ras by combining computational methods and biochemical assays. We discovered an allosteric binding site, the P110 site, near the C-terminus of K-Ras^G12D. The P110 site involves residues Arg97, Asp105, Ser106, Glu107, Asp108, Val109, Pro110, Met111, Tyr 137, Gly138, Ile139, Glu162, Lys165, and His166. Using virtual screening, we discovered a P110-site-binding compound, termed KAL-21404358. We used biochemical assays to validate the binding of KAL-21404358 to the P110 site. A combination of MST, TSA, line broadening NMR, and HSQC NMR demonstrated binding of KAL-21404358 to the P110 site of K-Ras^G12D with a K_D of 100 μM, and allosteric effects on switch I and switch II. KAL-21404358 was further found to disrupt the K-Ras^G12D-B-Raf interaction using a NanoBiT split luciferase assay, and to impair the Raf-MEK-ERK and the PI3K-AKT signaling pathways. We designed analogs to define the structure-activity relationship around the KAL scaffold. These findings suggest that the P110 site is an allosteric regulatory site for targeting oncogenic K-Ras^G12D. Moreover, this structure-based approach provides a strategy to discover small-molecule inhibitors for otherwise challenging drug targets.

MATERIALS AND EXPERIMENTAL DETAILS

Software and computational methods

MD simulations, MxMD simulations, molecular docking, and modeling were performed using Maestro (Schrödinger Suite), Molecular Operating Environment (MOE) and PyMOL. Chemical structures were drawn using ChemDraw Professional 16.0. Statistical analyses were produced using Prism 7.0 (GraphPad Software). Libraries of commercially available compounds were compiled from the inventories of Asinex, Enamine, Chembridge, ChemDiv, IBS, Life, Maybridge and TimTec. A fragment subset of ~3.5 millions compounds was selected and screened.

Molecular cloning

The KRAS^G12D plasmid was previously described¹⁴. Binding-deficient mutants of KRAS^G12D plasmid were generated using a QuikChange XL site-directed mutagenesis kit. Primers were designed using the Agilent QuikChange Primer Design application, and purchased from Integrated DNA Technologies. KRAS^G12DR97G forward primer 5’ GAA GAT ATT CAC CAT TAT GGA GAA CAA ATT AAA AGA GTT AAG G 3’ KRAS^G12DR97G reverse primer 5’ CTT AAC TCT TTT AAT TTG TTC TCC ATA ATG GTG AAT ATC TTC 3’, KRAS^G12DE107A forward primer 5’ GAG TTA AGG ACT CTG CAG ATG TAC CTA TGG TCC 3’ KRAS^G12DE107A reverse primer 5’ GGA CCA TAG GTA CAT CTG CAG AGT CCT TAA CTC 3’, KRAS^G12DD108A forward primer 5’ TAA GGA CTC TGA AGCT GT ACC TAT GGT CC 3’ KRAS^G12DD108A reverse primer 5’ ACC ATA GGT AC AGC T TCA GAG TCC TTA ACT C 3’ and KRAS^G12DP110D forward primer 5’ AGA TGT AGA TAT GGT CCT AG 3’, KRAS^G12DP110D reverse primer 5’ AGG ACC ATA TCT ACA TCT TC 3’. DNA sequencing was performed to confirm the amino acid sequence of the construct (GeneWiz).

Protein expression and purification

The KRAS^G12D construct was expressed in Escherichia coli BL21-Gold (DE3) cells (Stratagene). An isolated colony was transferred to 8 mL LB media with 100 μg/mL ampicillin and the inoculated culture was incubated with shaking (225 rpm) at 37 °C for 4.5 h. The starter culture was added to 1 L fresh LB with 100 μg/mL ampicillin. The culture was incubated with shaking at 37 °C and 225 rpm until the OD₆₀₀ reached 0.6. The temperature was then reduced to 15°C. Cells were incubated with 500 μM Isopropyl β-D-1-thiogalactopyranoside (IPTG) with shaking at 15°C and 225 rpm overnight. The next day, the bacteria were harvested by centrifugation at 4,000 × g for 20 min at 4°C and the pellet obtained was ready for purification or stored at −20°C.

The pellet was resuspended in 25 mL chilled lysis buffer (10 mM Tris pH 7.5, 500 mM NaCl, 5 mM MgCl₂, 5 mM imidazole, 2 mM TCEP, and Roche protease inhibitor cocktail). The bacteria were lysed by sonication on ice for 6 min and the lysate was centrifuged at 15,000 rpm for 45 min at 4°C to remove cell debris. The clarified lysate was incubated with Ni Sepharose 6 Fast Flow beads (GE Life Sciences) on a rotator at 4°C for at least 1 h. The beads were washed with wash buffer (10 mM Tris pH 7.5, 500 mM NaCl, 20 mM imidazole, 5 mM MgCl₂ and 2 mM TCEP) to remove non-specific binding. The protein was eluted with 10 mM Tris pH 7.5, 500 mM NaCl, 250 mM imidazole, 5 mM MgCl₂ and 2 mM TCEP. The protein was further purified using gel filtration Superdex 200 column in FPLC buffer containing 25 mM Tris pH 7.5, 100 mM NaCl, 5 mM MgCl₂, and 2 mM TCEP. The fractions containing K-Ras^G12D were pooled together and analyzed by SDS-PAGE. Protein concentration was determined using absorbance at 280 nm with extinction coefficient of 11,920 M⁻¹ cm⁻¹.

Nucleotide exchange

Nucleotides in endogenous recombinant K-Ras^G12D were exchanged with GDP or GppNHp using an EDTA-loading procedure. K-Ras^G12D protein (70 μM final) was incubated with 70-fold molar excess of EDTA (5 mM final), and 70-fold molar excess of new nucleotide (5 mM final) for 2 h at 30°C. After incubation, the sample was put on ice for two min and then MgCl₂ was added (65 mM final) to stop the reaction. To remove excess unbound nucleotide, the sample was added to a NAP-5 column (GE Life Sciences) equilibrated with FPLC buffer and eluted with FPLC buffer at 100 μL per fraction. Eluted fractions were evaluated using the NanoDrop method to determine protein concentration.

Microscale thermophoresis (MST)

100 μL of 200 nM K-Ras^G12D was combined with 100 μL of 100 nM RED-tris-NTA dye in PBS buffer with 3 mM DTT and 0.05% Tween-20 (PBSTD buffer). The protein/dye mixture was incubated at r.t. for 30 min, followed by 10 min centrifugation at 4°C and 15,000×g. The compounds were arrayed across a 16-point dilution series in PBSTD buffer and mixed 1:1 with labeled protein solution in 20 μL. Reaction mixture were loaded into standard treated capillaries and analyzed by Monolith NT.115 (Nanotemper Technologies) at 60% LED power and 40% MST power with a laser-on time of 5 s. The K_D was calculated by taking the average of triplicate F_norm measurements at each concentration and fitting the data to a sigmoidal four parameter fitting function in Prism (GraphPad Software). R-Ras, R-Ras2, and Rap1A were purchased from ProSpecBio. K-Ras WT was purchased from Cell Biolabs. H-Ras WT was acquired from Enzo Life Sciences.

Thermal shift assay (TSA)

A fluorescent thermal shift assay was used to validate the binding and confirm the success of the nucleotide exchange procedure of all GTPases used in the study. The assay was carried out in triplicate in Fast 96-well optical plates containing 5 μM protein and 5X SYPRO Orange dye (Invitrogen) in 20 μL total volume/well. Samples were heated at 3°C/min from 25°C to 95°C and protein unfolding was observed by monitoring the fluorescence of SYPRO Orange dye at 470 nm excitation and 623 nm emission using ViiA7 real-time PCR machine (Applied Biosystems). K-Ras^G12D with DMSO was used on the same plate as a reference for the shift in melting temperature (T_m) with compounds. Each GTPase with endogenous nucleotide was also used on the same plate as a reference for the shift in melting temperature (T_m) with the new nucleotide. All experiments were performed in triplicate. Data were analyzed using Protein Thermal Shift Software (Applied Biosystems) to determine the Tm of each well.

NMR line broadening

For NMR line broadening studies, the samples were buffer exchanged into Milli-Q water using a Millipore spin column (13,000 × g for five min at 4 °C, repeated five times, each time adding fresh Milli-Q water and discarding the flow-through). 10% D₂O was added to the protein/compound mixtures: protein only, 1:1, 1:3 and compound only. NMR line broadening experiments were performed on Bruker Avance III 500 Ascend (500 MHz) spectrometers (Columbia University) at 298 K. Compound peak assignment was completed using MestReNova version 10.0.2.

NMR HSQC

Uniformly ¹⁵N-labeled K-Ras^G12D protein without N-terminal His₆ tag was prepared. The KRAS^G12D construct was expressed in Escherichia coli BL21-Gold (DE3) cells (Stratagene) growing at 37°C in M9 minimal medium supplemented with 100 μg/mL ampicillin, 2 mM MgSO₄, 100 mM CaCl₂, 1X trace metals, 1X RPMI 1640 vitamin stock (Sigma-Aldrich, cat. R7256), 10 μg/mL biotin, 10 μg/mL thiamine hydrochloride, and 3 g/L ¹⁵NH₄Cl as the sole nitrogen source. The remaining steps were identical to the K-Ras^G12D expression and purification described above. Thrombin was then added at 5 U/mg protein to cleave the N-terminal His₆ tag. The reaction was allowed to proceed overnight at 4°C. The next day, protein solution was passed over Ni-Sepharose 6 Fast Flow beads (GE Life Sciences) and flowthrough containing the ¹⁵N-labeled K-Ras^G12D protein without histidine tag was concentrated and flash frozen. Purity was checked by SDS-PAGE gel.

¹H-¹⁵N HSQC experiments were performed on Bruker Avance III 500 Ascend (500 MHz) spectrometers (Columbia University) at 298 K. Uniformly ¹⁵N-labeled K-Ras^G12D was dissolved at 100 μM to 150 μM in NMR Buffer (50 mM HEPES pH 7.4, 50 mM NaCl, 2 mM MgCl₂, 2 mM TCEP, and 10% D₂O). The ¹H carrier frequency was positioned at the water resonance. The ¹⁵N carrier frequency was positioned at 115 ppm. Suppression of the water signal was accomplished using the WATERGATE sequence. Heteronuclear decoupling was accomplished using the GARP decoupling scheme. Assignments of K-Ras^G12D were previously published¹⁴. All data were processed and analyzed using TopSpin 3.1 (Bruker) and Sparky (Developed by T. D. Goddard and D. G. Kneller, UCSF).

NanoBiT split luciferase assay

HEK293T cells from ATCC (cat. CRL-1573) were seeded 16 h prior to use in 10% FBS in DMEM. Plasmids (KRAS-SmBiT/BRAF-LgBiT or SmBiT/LgBiT positive controls) were transfected into HEK293T cells and incubated 48 h to 72 h. After transfection was complete, compounds with indicated concentration were added and treated for 1 h at 37°C and 5% CO₂. Plates were then read with a Tecan Infinite M200 for luminescence every 6 min for 3 h at 37°C. Data were analyzed using Prism 7.0 (GraphPad Software).

Cell-based K-Ras^G12D-Raf RBD pulldown

LS513 cells from ATCC (cat. CRL-2134) were seeded 16 h prior to use in 10% FBS in RPMI-1640. The medium was then aspirated and replaced with serum-free medium containing KAL-21404358 and cells were incubated for 24 h. The medium was removed, washed with cold PBS, lysed and spun down at 13,000 rpm at 4°C to remove unlysed cells and debris. The lysate was incubated with Raf-1 RBD agarose beads (EMD Millipore) for 2 h with rotation at 4°C. The solution was then spun down at 1500 × g and the supernatant removed. The beads were washed twice with PBS, resuspended in 4X SDS, and then analyzed by the western blotting procedure detailed below.

Western blots

LS513 cells were seeded in RPMI-1640 and 10% FBS with 1% penicillin and streptomycin (PS) 16 h prior to use. The medium was then aspirated and compounds added as solutions in serum free medium (RPMI-1640 with 1% PS) at indicated concentration. Following treatment, the medium was aspirated from each dish and cells washed twice with PBS. Cells were lysed with 70 μl lysis buffer (RIPA buffer from ThermoFisher, cat. 89900, 1 mM EDTA, 1 mM PMSF, 1X Halt^™ protease inhibitor cocktail from ThermoFisher, cat. 78430 and 1X Halt^™ phosphatase inhibitor cocktail from ThermoFisher, cat. 78426). Unlysed cells and debris were pelleted for 15 min at 16,000 × g at 4°C. Samples were separated using SDS-PAGE and transferred to a polyvinylidene difluoride membrane. Transfer was performed using the iBlot2 system (Invitrogen). Membranes were treated with Li-COR Odyssey blocking buffer for at least 1 h at r.t., then incubated with primary antibody (1:1000) in a 1:1 solution of PBS-T (PBS with 0.1% Tween 20) and Li-COR odyssey blocking buffer overnight at 4°C. Following three 5 min washes in PBS-T, the membrane was incubated with secondary antibodies (1:3000) in a 1:1 solution of PBS-T and Li-COR Odyssey blocking buffer for 1 h at r.t. Following three 5 min washes in PBS-T, the membrane was scanned using the Li-COR Odyssey Imaging System. Antibodies for pErk1/2, Erk1/2, pAkt ser473, Akt, pan-Ras (Cell Signaling), and Raf-1 (Santa Cruz) were detected using a goat anti-rabbit or goat anti-mouse IgG antibody conjugated to an IRdye at 800CW and 680CW conjugated, respectively (Li-COR Biosciences).

RESULTS

Discovery of the P110 site and KAL-21404358

Given the challenges of directly targeting the nucleotide binding site and effector-interaction surface, we hypothesized there might be allosteric pockets regulating the on and off cycle of K-Ras. To identify such sites, we first performed computational analysis of the K-Ras^G12D crystal structure (PDB ID: 4DSN) using the SiteMap (Schrödinger Suites) prediction tool^{21, 22}. The P110 site had the highest score for a potential small molecule binding site (r_sitemap_SiteScore = 0.92) and identified a binding site including residues Arg97, Asp105, Ser106, Glu107, Asp108, Val109, Pro110, Met111, Tyr 137, Gly138, ILE139, Glu162, Lys165, and His166 (Figure 1A, highlighted in red).

Figure 1.

P110 Site and KAL-21404358 Binding Pose

A. K-Ras^G12D (PDB: 4DSN) model with the P110 site highlighted in red. The P110 site residues are Arg97, Asp105, Ser106, Glu107, Asp108, Val109, Pro110, Met111, Tyr137, Gly138, Ile139, Glu162, Lys165, and His166.

B. KAL-21404358 docking pose in the P110 site, and its structure, docking score, chemical formula, mass, and molecular weight.

C. Detailed view of KAL-21404358 binding in P110 site. Four potential hydrogen bonds are labeled with green lines.

D. KAL-21404358 superimposed with four high-energy water molecules shown in red with ΔG > 2, in purple with ΔG > 1.

To explore the potential robustness of the P110 site across different protein conformations, we conducted similar analyses of other crystal structures of K-Ras (PDB ID: 4EPR, 4OBE) and other Ras isoforms H-Ras (PDB ID: 4L9W) and N-Ras (PDB ID: 3CON). We found that the P110 site still appeared across these structures, but the SiteMap scores were not as high as in structure 4DSN (Figure S1A). This indicates that the P110 site is more apparent in GTP-bound K-Ras^G12D than GDP-bound K-Ras^G12D. Computational analysis of the P110 site on 4DSN (orange) and 4EPR (blue) showed different poses of residues Arg97, Asp105, Ser106, Glu107, Asp108 and Lys165 (Figure S1B). These conformational changes make the P110 site on GTP-bound K-Ras^G12D open and larger than the ones on GDP-bound K-Ras^G12D. Comparison of P110 sites of 4DSN (orange) with 4L9W (green) showed different poses of residues Asp105, Ser106, Glu107, Asp108, Pro110 and Met111, making the P110 site open and larger in K-Ras (Figure S1C). These observations suggest that the P110 site is specific to K-Ras and is not as apparent in H-Ras or N-Ras, at least in the x-ray structures currently available.

Next, we performed a molecular dynamics (MD) simulation to mimic different conformations of K-Ras^G12D other than the ones found in the crystal structures. 20 clusters were generated from a 200 ns MD simulation. The P110 site appeared consistently in these simulations, of which one cluster (#6) showed the best SiteMap score of 1.06 (Figure S1D). This suggests that this pocket can become even more accessible during motion of the K-Ras^G12D protein.

Mixed Solvent Molecular Dynamics (MxMD) simulations were then run with structure 4DSN using acetonitrile, isopropanol, pyrimidine, acetone, imidazole, and N-methylacetamide as organic probes. These organic probes can affect the conformation of K-Ras^G12D and reveal which sites can be accessible to small molecules of various chemotypes²³. The P110 site was found to contain all of these solvents as a hot spot, indicating its potential as binding pocket for small organic molecules (Figure S1E).

We evaluated whether we could identify fragments or lead-like compounds predicted to bind with reasonable affinity to the P110 site. We tested 3.5 million compounds using the Glide docking algorithm (Schrödinger Suites), which generates a score in which the more negative the score, the higher the predicted affinity²⁴. 77 fragments with scores <−6.5 were obtained for further validation. Four rounds of biochemical screening using microscale thermophoresis (MST), thermal shift assay (TSA), nuclear magnetic resonance (NMR) line broadening, and NMR heteronuclear single quantum coherence (HSQC) spectroscopy were used to select promising compounds from among these 77 candidates. KAL-21404358 was the most favorable compound, as it showed positive binding results in all four tests (Figure 1, 2 & 3).

Figure 2.

Validation of KAL-21404358 Binding to K-Ras^G12D

A. MST assay of KAL-21404358 with GppNHp-bound K-Ras^G12D (K_D = 88 ± 1 μM)

B. MST assay of KAL-21404358 with GDP-bound K-Ras^G12D (K_D = 146 ± 2 μM) and indicating selectivity towards GppNHp-bound versus GDP-bound K-Ras^G12D.

C. KAL-21404358 increases the melting temperature of K-Ras^G12D in a thermal shift assay.

D. KAL-21404358 lacks binding to P110 site mutants and has differential selectivity to K-Ras^G12D compared to other RAS family member proteins.

Figure 3.

NMR Validation of KAL-21404358 Binding to K-Ras^G12D

A. NMR line broadening experiment of KAL-21404358 with an increased concentration of K-Ras^G12D (1:0, 1:0.3, 1:1 and 0:1 ratios). Peaks of hydrogens of quinolinol and piperazinyl group (highlighted in red) were broadened, indicating KAL-21404358’s binding to K-Ras^G12D.

B. An NMR HSQC experiment showed conformational changes in switch I and switch II of K-Ras^G12D, which could be explained by P110 site’s allosteric effect. In the upper graph, blue represents GDP-bound K-Ras^G12D and red represents GDP-bound K-Ras^G12D with KAL-21404358 at 1:7 concentration ratios. In the lower graph, blue represents GppNHp-bound K-Ras^G12D only, red represents GppNHp-bound K-Ras^G12D with KAL-21404358 at 1:7 concentration ratios and green represents GppNHp-bound K-Ras^G12D with KAL-21404358 at 1:14 concentration ratios. Residues changed upon binding with KAL-21404358 are labeled.

KAL-21404358 was predicted to interact well with the P110 site, with a Glide docking score of −7.37 (Figure 1B). A closer view of KAL-21404358 in the P110 site showed four potential hydrogen bonds (between the -NH in the hydroxyquinoline and the carboxyl group of Asp108, between the -OH in the hydroxyethyl group and the carboxyl group of Glu107, and two between the -OH in the hydroxyethyl group and the amine group of Arg97), as well as strong polar interactions (between the bridge of KAL-21404358 and the amide group of Glu107, between the piperazinyl group and the amide group of Gly138) (Figure 1C).

WaterMap (Schrödinger Suites) was then used to estimate thermodynamic property changes resulting from water molecule displacement by fragments binding in the P110 site^{25, 26}. Four high-energy water molecules were predicted to be located in the P110 pocket superimposed upon KAL-21404358, which suggested that displacement of these water molecules would improve the binding affinity of KAL-21404358 even further (Figure 1D).

Validation of KAL-21404358 binding with K-Ras^G12D

MST and TSA were used as first-line screening methods. The K_D of KAL-21404358 for GppNHp-bound K-Ras^G12D was 88 μM, and was 146 μM with GDP-bound K-Ras^G12D assessed by MST (Figure 2A and B). This suggests that KAL-21404358 has a slightly higher binding affinity to the GppNHp-bound form of K-Ras^G12D. TSA experiments confirmed a 2.1°C melting temperature shift, indicating binding of KAL-21404358 stabilizes GDP-bound K-Ras^G12D to thermal denaturation, possibly further inhibiting activation of GDP-K-Ras^G12D (Figure 2C). The binding of KAL-21404358 to GppNHp-bound K-Ras^G12D did not cause a temperature shift (Figure S2A).

To test whether KAL-21404358 bound specifically to the P110 site, we constructed four mutants predicted to be deficient for binding to KAL-21404358—R97G, E107A, D108A and P110D. No binding between KAL-21404358 and these four mutants was detected using MST, supporting the hypothesis that this compound binds in the P110 site (Figure 2D, the original MST curves are shown in Figure S2B). We also examined the specificity of KAL-21404358 for K-Ras^G12D over K-Ras^WT, H-Ras^WT, Rap1a, R-Ras and R-Ras2. Each protein was tested for its ability to bind to KAL-21404358 using MST. Much weaker binding of KAL-21404358 was detected towards these other proteins, suggesting that KAL-21404358 has selectivity for K-Ras^G12D (Figure 2D, the original MST curves are shown in Figure S2C).

Validation of the KAL-21404358 binding site and binding mode

To further elucidate how KAL-21404358 binds to K-Ras^G12D, we used NMR line broadening. The disappearance of hydrogens 1, 5, 6, 14, 15, 17 and 18 in the ¹H NMR spectrum of the compound indicated binding to K-Ras^G12D, which is likely due to these being the interacting atoms on KAL-21404358 (Figure 3A). Analysis of the structure of KAL-21404358 suggested that it bound to K-Ras^G12D with the quinolinol and piperazinyl group, but not the neopentyl group, which was consistent with the computational prediction.

HSQC NMR was then used to identify the residues on K-Ras^G12D that change upon KAL-21404358 binding to test for possible allosteric effects. Conformational changes in the switch I and II regions (Asp33, Thr35, Ser39, Leu56, Gly60, and Gly75) were observed in both GppNHp-bound and GDP-bound K-Ras^G12D (Figure 3B) upon KAL-21404358 binding, suggesting an allosteric effect on the K-Ras^G12D conformation. We observed KAL-21404358’s differential selectivity towards GppNHp-bound K-Ras^G12D. However, KAL-21404358 only interacted with side chains of residues, as no backbone shifts were observed in the HSQC NMR experiments near the P110 residues.

Inhibition of the K-Ras^G12D-B-Raf interaction

Based on KAL-21404358’s binding mode, we tested whether this compound could disrupt the interaction between K-Ras^G12D and B-Raf using a NanoBiT split luciferase assay²⁷. K-Ras^G12D was fused to SmBiT, and B-Raf was fused to LgBiT. The Ras-Raf interaction inhibitor 3144 was used as a positive control for disruption of the interaction¹⁴. KAL-21404358 and its analog KAL-YZ0965 exhibited lower luminescence compared to DMSO-treated control samples, suggesting that these compounds disrupted the K-Ras^G12D–B-Raf interaction (Figure 4A). KAL-21404358 has higher effectiveness at lower concentrations, whereas KAL-YZ0965’s effectiveness was low. The compounds were simultaneously tested in cells with SmBiT/LgBiT to rule out non-specific inhibition (Figure 4B). A K-Ras^G12D-Raf-1-RBD pulldown assay was also conducted to validate disruption of this interaction. Less K-Ras^G12D was bound to Raf-1-RBD beads in the presence of KAL-21404358, which supported the hypothesis that this compound disrupts this interaction in cells (Figure 4C).

Figure 4.

KAL-21404358 Inhibits the K-Ras^G12D-B-Raf Interaction and K-Ras^G12D-dependent signaling

A. A NanoBiT split luciferase assay showed that KAL-21404358 and its analog KAL-YZ0965 disrupted the K-Ras^G12D-B-Raf interaction. 3144 was used as positive control. K-Ras^G12D was fused to SmBiT and B-Raf was fused to LgBiT, and constructs were transfected in HEK293T cells. Luminescent signals were detected when K-Ras^G12D bound to B-Raf. Ordinary one-way ANOVA test were conducted in Prism 7. **** indicates p<0.0001, *** indicates p<0.001 and ns indicates p>0.05.

B. KAL-21404358, KAL-YZ0965 and 3144 were screened against SmBiT-LgBiT to rule out non-specific inhibition and toxicity.

C. LS513 cells (with K-Ras^G12D) were treated with KAL-21404358 as indicated and the amount of Raf-1-RBD bound K-Ras^G12D proteins were determined.

D. The effects of KAL-21404358 on abundance of phosphorylated Akt and total Akt (left) and phosphorylated Erk and total Erk (right) were determined at the indicated concentrations in LS513 cells (with K-Ras^G12D mutations). LS513 cells from ATCC (cat. CRL-2134) were seeded 16 h prior to use in 10% FBS in RPMI-1640. The medium was then aspirated and replaced with serum-free medium containing 21404358 and cells were incubated for 24 h. Relative intensities of phosphorylated forms versus total forms were quantified and labeled.

Next, the effect of KAL-21404358 on downstream Raf-Mek-Erk and PI3K-Akt-mTor pathways was investigated. Less cellular phosphorylated Akt and phosphorylated Erk were detected after treatment of cells with KAL-21404358 (Figure 4D). However, the inhibitory effect of KAL-21404358 in the pulldown and western assays was not strong, consistent with a low binding affinity. Further optimization of this compound or other compounds that bind in this site is needed to enable efficient allosteric inhibition of K-Ras^G12D in cells.

Modification of the KAL-21404358 scaffold

To optimize the properties of KAL-21404358, we designed and synthesized a series of close structural analogs. We divided KAL-21404358 into four functional groups: the hydroxyquinoline (blue), the bridge (green), the amine group (black) and the neopentyl group (red) (Table 1). First, we hypothesized that the neopentyl group might be replaced to improve properties of the compounds, because (1) it did not interact with the P110 pocket in the computational models, and (2) this group is hydrophobic, decreasing aqueous solubility of the compound. We thus identified and tested seven KAL-21404358 analogs (red shade) which had replacements of the neopentyl group only. The binding affinities of these compounds, as assessed by MST, were not improved, consistent with a lack of interaction at this site. These replacements did not generate new interactions between compounds and receptor, likely because this group is facing solvent. We next focused our attention on the hydroxyquinoline moiety, which fits well into the P110 pocket. Our hypothesis was that adding functional groups or changing the hydroxyquinoline moiety might form new hydrogen bonds between KAL-21404358 analogs and K-Ras^G12D, thereby improving binding. Six KAL-21404358 analogs (blue shade) were identified and tested, among which analog KAL-11067146 (in bold) showed more potent binding in the MST assay (Table 1). However, this analog did not affect the K-Ras^G12D conformation in an NMR HSQC experiment, suggesting a loss of allosteric inhibitory activity. Computational docking did not show the presence of a hydrogen bond formed by the addition of the fluorine (Figure 5A), consistent with the NMR result.

Table 1.

KAL-21404358 Structure-Activity Relationship Analysis and Optimization

Four functional groups are labeled: hydroxyquinoline in blue, bridge in green, amine group in black and neopentyl group in red. KAL-21404358 analog names, structures, and K_D measured by MST and NMR HSQC results are shown. More active compounds are in bold.

Figure 5.

Computational docking poses of three representative analogs of KAL-21404358.

A. KAL-11067146 docking pose in P110 site. KAL-11067146 has a similar docking pose as KAL-21404358. Addition of fluorine molecule did not improve KAL-11067146 binding to the receptor. Potential hydrogen bonds are labeled with green lines.

B. KAL-PHB6003 has a reverse docking pose compared to KAL-21404358. Potential hydrogen bonds are labeled with green lines.

C. KAL-YZ0965 has a similar docking pose as KAL-21404358. Addition of a carbonyl group to the bridge helps form an additional hydrogen bond to improve KAL-YZ0965 binding. Potential hydrogen bonds are labeled with green lines.

Next, we added a carbonyl group to the bridge region to increase hydrogen bond interactions with K-Ras^G12D and also to increase the hydrophilicity of KAL-21404358. Considering the difficulty of synthesizing the hydroxyethyl piperazinyl moiety, we decided to replace it with other cyclic amines. Three analogs were synthesized and one was commercially available, among which KAL-PHB6002, KAL-PHB6003 and KAL-2241124388 exhibited more potent binding in the MST assay. Computational docking showed a reverse pose for KAL-PHB6003 in the P110 site as an example of this series of analogs (Figure 5B). Two potential hydrogen bonds are likely to be formed between the oxygen molecule of the hydroxyquinoline and the amine group of Lys165, as well as between amine group of the bridge and carboxyl group of Glu162. However, HSQC NMR experiments again showed a lack of allosteric inhibitory effect of KAL-PHB6002 and KAL-PHB6003. The NanoBiT split luciferase assay showed a trend towards a decreased luminescence signal in the presence of KAL-2241124388. However, this inhibitory effect was not statistically significant in a one-way ANOVA test (Figure S3).

We further synthesized 15 analogs with carbonyl bridges and cyclic amines (see Supplementary Methods). Three of these compounds (KAL-YZ0965, KAL-YZ0968 and KAL-YZ0970, grey shade and bold) exhibited higher binding affinity by MST and chemical shifts in NMR HSQC experiments (Table 1). Computational docking showed that KAL-YZ0965 fit well into the P110 site, as an example of this series of analogs. The addition of amide group as the bridge could form two potential hydrogen bonds with K-Ras^G12D, stabilizing the binding of KAL-YZ0965 (Figure 5C). KAL-YZ0965, KAL-YZ0968, and KAL-YZ0970 together with KAL-55883121 (similar binding affinity with KAL-21404358 in MST assay) were thus selected for testing in the NanoBiT K-Ras^G12D-B-Raf interaction assay.

We detected trends towards decreased luminescence signal in the presence of these four analogs, but not in control cells (Figure 4 & Figure S3). A one-way ANOVA test showed that only the inhibitory effect of KAL-55883121 was statistically significant, suggesting this compound may also be capable of interrupting K-Ras^G12D-B-Raf interaction. Moreover, KAL-YZ0965, KAL-YZ0968 and KAL-YZ0970 have improved aqueous solubility, which provides opportunities for future studies with these compounds (Figure 4 & Figure S3).

DISCUSSION

K-Ras^G12D has been considered a challenging target over the past 30 years. Here, we explored a strategy for discovering small-molecule inhibitors that directly bind to this oncogenic K-Ras mutant. This strategy started with computational design, leading to discovery of a potential binding pocket. The P110 site is in the allosteric lobe, which is opposite to the functional P-loop (residues 10–17), switch I (residues 30–40) and switch II (residues 60–75) regions, which constitute the active site for GTP hydrolysis and interaction sites for effector proteins, including Raf, PI3K, RalGDS and GAP (Figure 6A)²⁸. Despite the distance of the P110 site to effector domains, residues in switch I and switch II undergo conformational changes upon binding of KAL-21404358, causing disruption of K-Ras^G12D signaling activity (Figure 6B).

Figure 6.

The location of the P110 site in relation to effector domains

A. The P110 site (orange) is in the allosteric lobe, opposite to the functional P-loop (blue), switch I (red) and switch II (green) domains which constitute the active site for GTP hydrolysis and effector proteins binding.

B. Residues in P110 site, P-loop, Switch I, Switch II and allosterically affected with binding of KAL-21404358 are listed and compared. Key residues which undergo conformational changes upon binding are in bold.

We used a tiered set of computational and biochemical assays to evaluate whether compounds can bind to this site, and if so, what effect they have on the K-Ras^G12D protein. The first-line screening consisted of computational ligand docking to select compounds predicted to bind in this site. The second screening filter involved the use of MST and TSA to validate binding to K-Ras^G12D experimentally. The third tier was NMR line broadening and NMR HSQC binding assay to further discover the binding mode and location, and to assess allosteric effects of binding. The fourth tier was a Ras functional assay involving direct interactions with Raf and two well-established cellular signaling pathways.

KAL-21404358 was found to be the best candidate from this set of assays. KAL-21404358 was further validated to specifically bind to K-Ras^G12D in the P110 site. Although the binding affinity was moderate, KAL-21404358 was able to disrupt the K-Ras-B-Raf interaction, and Akt and Erk signaling pathways at high concentration. GTP-bound K-Ras^G12D exists in two distinct conformations, state 1 and state 2, where state 1 has a lower binding affinity to effectors^29–31. Based on our experimental results, we hypothesize that KAL-21404358 binds to GTP-bound K-Ras^G12D state 1, and thus shifts the protein equilibrium from state 2 towards state 1 (Figure 7). KAL-21404358 can also bind to GDP-bound (inactive) conformation with a two-fold lower affinity. We hypothesize that KAL-21404358 stabilizes and traps K-Ras^G12D in this inactive GDP-bound state by reducing its probability to nucleotide exchange, thus decreasing the amount of GTP-bound K-Ras^G12D level (Figure 7). Thus, the P110 pocket, according to this model, is slightly more pronounced in state 1 of the GTP-bound protein, but is also present in the GDP-bound protein.

Figure 7.

KAL-21404358 binding scheme

The orange circle represents inactive GTP-bound K-Ras^G12D (state 1). The green circle represents active GTP-bound K-Ras^G12D (state 2). The red circle represents inactive GDP-bound K-Ras^G12D. The size of the circles represents the relative amounts of different K-Ras^G12D states. In the absence of KAL-21404358, K-Ras^G12D favors the active state 2 conformation. When KAL-21404358 binds to GTP-bound K-Ras^G12D, it stabilizes state 1, disrupting the binding of effectors. When KAL-21404358 binds to GDP-bound K-Ras^G12D, it traps the protein in this inactive state, thereby reducing the amount of active-state protein.

To optimize this P110-binding scaffold, a high-resolution structure is likely needed. We attempted to obtain such a co-crystal structure without success, likely due to the low binding affinity and low solubility of KAL-21404358. Nonetheless, we did synthesize a series of analogs to define the structure-activity relationship, and found that addition of a carbonyl group to the bridge enhances binding affinity by forming new hydrogen bonds and increasing hydrophilicity of KAL-21404358. The binding affinities of those analogs were improved as shown by the MST assay, but their ability to disrupt K-Ras-B-Raf interaction was not enhanced. This remains to be studied further, especially by using structural biology approaches.

In summary, these findings suggest that the P110 site is a potential allosteric regulatory site for targeting oncogenic K-Ras proteins. KAL-21404358 is the first small-molecule discovered to bind this site and disrupt K-Ras^G12D function. Moreover, this approach provides a strategy to discover small-molecule inhibitors for RAS family and other challenging target proteins.