Protein preparation for assay development

Several preparations of purified KRAS were created to compare wildtype vs mutant and prenylated vs non-prenylated proteins. The prenylated KRAS proteins were expressed in insect cells using a baculovirus system and produced as a mixture of farnesylated and geranyl-geranylated forms in a roughly 40:60 ratio (based on mass spectrometry analysis). The purity of the various KRAS preparations was ≥ 95%. The preparation of BRAF was also produced in insect cells and was partially purified to ~ 50%. This preparation could be activated by KRAS (characteristics described below) and was co-purified with endogenous insect cell 14-3-3, which is a known RAF co-factor [34, 35]. Any further purification resulted in loss of ability to be activated by KRAS. Finally, in order to avoid MEK inhibitors from interfering with the coupled assay read-out, we used a preparation of MEK^K97R, an inactive form of that protein.

Biochemical assay measures inhibition of KRAS

At the time of assay development for the coupled assay, no small molecule inhibitors of RAS had been disclosed so we relied on recapitulating the known biology for assay validation. As a first step, we examined phospholipid dependencies as shown in Fig 1B. This experiment confirms observations reported by others that KRAS activation of BRAF requires a negatively charged lipid environment, while a neutral lipid such as phosphatidylcholine was unable to support KRAS activation of BRAF [32]; for further assay development we chose phosphatidylserine (PS). Fig 1C shows that KRAS activation of wildtype BRAF is dependent on the presence of protein prenylation, PS, and the presence of a GTP analogue (GTPγS, which hydrolyzes at a lower rate than GTP). Low level basal activity of the wildtype BRAF was increased 50-fold or more with the addition of prenylated, GTP-loaded KRAS along with PS, which was prepared as a solution of liposomes and added to assay buffer. When using KRAS loaded with a non-hydrolyzable GDP analogue (GDPβS) or when omitting PS or prenylation, only the basal level activity of BRAF was observed. Comparing wildtype (WT) KRAS with KRAS^G12V showed that both proteins are able to activate BRAF, with the mutant achieving a higher level of activation. All further coupled assay experiments utilize GTPγS-loaded, prenylated KRAS^G12V in a PS environment.

As a control to assess inhibition of the KRAS activation of BRAF, we added the RAS-Binding-Domain (RBD) of CRAF and showed that this blocked the activation with an IC₅₀ of 0.013 μM (Fig 1D). RAF inhibitors are also expected to inhibit in this coupled assay (due to inhibition of the kinase activity of BRAF) so we designed a counter-screen that matches the coupled assay but is lacking KRAS. As is shown in Fig 1C, the wildtype BRAF preparation that can be activated by KRAS is not able to phosphorylate MEK without KRAS so we used BRAF^V600E to drive the phosphorylation of MEK in the counter-screen. We tested a diverse set of RAF inhibitors from historical programs in both the coupled assay and the counter-screen and showed that the coupled assay IC₅₀s correlated well with the IC₅₀s in the BRAF^V600E counter-screen across five orders of magnitude (Fig 1E). The results from the control experiments with RBD and with RAF inhibitors provide reassurance that the BRAF preparation used in the coupled assay behaves as expected.

Validated hits inhibit KRAS function

We identified KRAS-specific inhibitors following the workflow in Fig 2, which includes the KRAS^G12V-dependent coupled assay with the BRAF^V600E counter-screen, hit-validation by protein-observed NMR, determination of cellular potency, and biophysical assessments. With respect to the KRAS-dependent coupled assay, 8800 diverse fragments (average molecular weight 251) were screened at 250 μM and 8800 diverse small molecules (average molecular weight 349) were screened at 50 μM. We defined primary hits as those compounds with ≥ 80% inhibition in the coupled assay and ≤ 20% inhibition recorded from historical BRAF^V600E screening efforts. The hit-rate after this selection step was 1.9%. In parallel, a fragment library of 500 diverse fragments was screened by NMR against KRAS^G12V. From this NMR-based fragment screen, we identified indoles as a privileged KRAS binding scaffold. After hit-list triaging and hit assessment from the coupled assay screen, we selected 17 compounds (3 containing the indole substructure) for hit confirmation in dose-response format for the coupled assay and counter-screen as well as hit validation by NMR.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Workflow to identify and validate chemical matter that inhibits KRAS activation of BRAF through direct interaction with KRAS.

Green boxes utilize full-length prenylated KRAS in a lipid environment, yellow boxes utilize truncated KRAS in aqueous buffer and the grey box is the BRAF^V600E counter-screen.

https://doi.org/10.1371/journal.pone.0174706.g002

The first validation of these hits was focused on confirming a direct interaction with KRAS^G12V for which we used a protein-observed NMR Heteronuclear Single-Quantum Correlation (HSQC) assay. This assay was performed using non-prenylated GDP-KRAS^G12V since we found a lack of important resonances in KRAS bound to a GTP analogue due to dynamic conformational exchange. Two indole-containing fragments scored positive in the HSQC assay and also confirmed in the dose-response assays, including compound 1 (Fig 3) which recorded a coupled assay IC₅₀ of 87 μM and an IC₅₀ in the counter-screen of >250 μM (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Chemical structures of compounds 1–6.

https://doi.org/10.1371/journal.pone.0174706.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of biochemical, cellular, and biophysical measures of potency in μM.

https://doi.org/10.1371/journal.pone.0174706.t001

We explored the SAR around compound 1 by substructure and similarity searches from the existing compound archive as well as focused medicinal chemistry efforts, resulting in several compounds with improved potency in the coupled assay and similar binding behavior to the KRAS^G12V protein by NMR. Subsequent experiments focused on two active analogues from this effort (compounds 2 and 3, Fig 3), as well as three control compounds. Structures are shown in Fig 3 for all compounds and their IC₅₀ values from the coupled assay and counter-screen are shown in Table 1. The active analogues 2 and 3 had IC₅₀ values in the coupled assay of 0.8 μM and 1.2 μM, respectively, with a good window of selectivity in the RAF-based counter-screen, where minimal activity was observed. Compound 4 has negligible biochemical activity in the coupled assay, yet similar physical chemical properties compared to compounds 2 and 3. Compounds 5 and 6 are RAS ligands identified by others [18, 19]. Compound 5 exhibits a 7 μM IC₅₀ in the coupled assay and a small selectivity window in the counter-screen; compound 6 has negligible activity in both the coupled assay and the counter-screen.

Cellular effects of the compounds were assessed by Western blot analysis and inhibition of cell proliferation. Fig 4A shows the dose-dependent effects of the compounds on pAKT and pERK, representing two different pathways downstream of KRAS^G12D in SW1990, a human pancreatic cancer cell line. Inhibition of cell proliferation was recorded as an EC₅₀ value for the compounds in SW1990 cells (Table 1). The active analogs 2 and 3 showed single digit μM potency in these cellular assays, with dose-dependent decreases in phosphorylation of both of the pathway markers, while total protein levels were not affected, consistent with on-target inhibition of KRAS^G12D. The negative control compound 4 and the initial fragment hit compound 1 were inactive at the concentrations tested, in line with their activities in the coupled assay. As published previously, compound 5 showed double-digit μM potency in cells and compound 6 was inactive in cells [18, 19]. All compounds tested show good correlation between their IC₅₀ in the coupled assay measuring inhibition of the KRAS switching function and their cellular EC₅₀. As a critical control for non-specific inhibition of RAS, we tested the compounds in a melanoma cell line driven by BRAF^V600E (A375) and pERK levels were not affected by the compounds (Fig 4B). The pAKT levels in this cell line show an unexpected pattern of modulation for the active analogs 2 and 3, which is consistent with on-target RAS activity since the PI3K-AKT pathway is still expected to be regulated by RAS in these cells.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Cellular activity of ligands on signaling in KRAS and BRAF mutant tumor cells.

Dose-response data for all compounds from Fig 3 on SW1990 cells (A) and A375 cells (B), assessing impact on pAKT and pERK, together with their total protein controls and a GAPDH control.

https://doi.org/10.1371/journal.pone.0174706.g004

NMR studies with truncated KRAS protein

In order to assess mechanism of inhibition and establish the binding site for the new inhibitors, we employed NMR studies using compound 2 and non-prenylated KRAS^G12V in the absence of PS. It has been previously reported that GTP RAS has two conformational states that can be detected by ³¹P-NMR [36, 37]. These states are seen in splitting the γ-PO₄ resonance peak into two peaks. Each peak has been associated with either an inactive conformation or an active conformation by recording spectra from inactive HRAS (T35S) or by trapping HRAS or KRAS in its active form with RBD and comparing those spectra with the two-state spectrum from GTP RAS. We have repeated these experiments using non-prenylated KRAS^G12V loaded with the non-hydrolyzable GTP analogue GMPPNP and compound 2 (Fig 5A). The relationship of the ³¹P γ-PO₄ peaks for KRAS^G12V and KRAS^G12V plus RBD are as described previously [36]. We find that compound 2 traps KRAS in a conformation lacking the γ-³¹P resonance of the “active” RBD binding form, whereas the γ-³¹P resonance corresponding to the “inactive” form persists and is slightly shifted further downfield, comparable to the observation of the inactive HRAS mutant.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. NMR experiments probe mechanism and binding site.

(A) ³¹P-NMR spectrum of non-prenylated GMPPNP-loaded KRAS^G12V shows two environments for the γ-phosphate (top) with a shift to a single inactive population with 1 mM compound 2 (middle) and a single active population with 0.5 mM of the CRAF-RBD (bottom); α and β phosphate have single environments and the KRAS^G12V is present at 5 mg/ml. (B) ¹⁵N,¹H-HSQC spectrum for non-prenylated GDP-KRAS^G12V in absence (black) or presence (blue) of 400 μM compound 2 (protein concentration is 0.5 mg/ml). The residues with major chemical shift changes are indicated with circles. (C) Residues engaged in nOe’s between non-prenylated GDP-KRAS^G12V and compound 2 shown in blue stick model with labels using the crystal structure for compound 6 (PDB-ID 4EPY) as reference; compound 6 in orange, GDP and RAS protein in green, switch-I in magenta and switch-II in yellow.

https://doi.org/10.1371/journal.pone.0174706.g005

Next, we assessed the KRAS binding site for these compounds. The HSQC NMR shifts for compound 2 observed in non-prenylated KRAS^G12V are shown in Fig 5B. Many of the residues showing significant shifts belong to a conformationally flexible region around the switch-II loop, known to be involved in effector activation. Further NMR experiments resulted in assignment of multiple intermolecular nOe’s between indole protons of compound 2 and V7, L56, T58, and M72 of non-prenylated KRAS^G12V, confirming a direct and specific interaction. The binding site indicated by these residues has since been disclosed by others using similar truncated protein preparations and different ligands [18–20]. Fig 5C maps the residues involved in the intermolecular nOe’s between compound 2 and KRAS onto the public domain crystal structure of one of these published co-crystal structures, namely KRAS and compound 6 (PDB-ID 4EPY). This mapping shows that the indole from compound 2 binds at the same site as the indole from compound 6, which is also the same site as the indole from compound 5 (as observed in PDB-ID 4DST).

An HSQC titration experiment for compound 2 with non-prenylated KRAS^G12V showed an estimated K_d of 690 μM (Table 1), well above the IC₅₀ from the coupled assay (0.8 μM). Since ligands in this series show cellular effects in the low micromolar range (Fig 4A and Table 1), there is a significant decrease in potency when comparing the affinity in the biochemical and cellular assays with the affinity determined in the NMR experiment. A similar disconnection between cellular activity and binding affinity was reported for compound 5 [19]. We hypothesize that this difference is due to the use of full-length prenylated KRAS in a native or native-like lipid environment in the cellular and biochemical assays, whereas the NMR assessments used truncated non-prenylated KRAS in aqueous buffer lacking PS.

Biolayer interferometry studies with prenylated KRAS protein

To further evaluate if the biochemical and cellular potencies of the compounds are consistent with binding to KRAS, and to test the hypothesis that the affinity of binding to KRAS is dependent on the presence of the C-terminal prenyl group and the presence of lipids, we explored the binding of compounds directly to prenylated KRAS using ForteBio’s biolayer interferometry (BLI) platform [38]. In this method, KRAS is attached to a sensor surface via streptavidin:biotin interactions, and the binding of compounds to the target is measured directly. The K_ds for the compounds determined for prenylated KRAS^G12V loaded with GTPγS in the presence of PS (prepared as a solution of liposomes) correlate well with their biochemical IC₅₀s and the cellular EC₅₀s (Table 1), and the rank-order of the compounds is consistent across the assays. Of specific note is that control compound 4 was inactive in the biochemical and cellular assays and did not bind in BLI studies at 10 μM. Compounds 5 and 6 also showed K_ds in the BLI assay that correlate with their biochemical IC₅₀ and cellular EC₅₀ values (Table 1). These results support the hypothesis that the binding of these inhibitors to KRAS results in an effect in the coupled assay and in the cellular assays.

To understand the impact of prenylation and the presence of PS, we performed a series of control experiments with BLI. First, a positive control was selected to ensure run-to-run reproducibility so that results could be compared across studies. Compound 3 was selected because it showed good behavior up to 10 μM, whereas compound 2 showed minor non-specific binding at 10 μM (K_d determinations described earlier were not impacted). In the absence of PS or for KRAS constructs lacking the C-terminus (and thus the prenyl group), binding for compound 3 was significantly lower. Specifically, minimal binding was observed in a KRAS construct 1–180, which lacks the C-terminus (and therefore the prenyl group), at 10 μM of compound 3. For full-length KRAS, but in the absence of PS, no binding was observed up to 3.3 μM of compound 3, and some binding was observed at 10 μM. The differences in responses for these controls and full-length KRAS in the presence of PS are shown in Fig 6A. We also observed that none of the compounds bind non-specifically to biosensors lacking KRAS. These observations show that the correlation between affinity in the BLI binding assay and potency in the coupled biochemical assay is dependent on full-length prenylated KRAS and PS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. BLI experiments assess dependence of K_d on prenylation, PS, and HVR.

(A) Response data at various concentrations of compound 3 with different protein preparations (with or without prenylation, with or without PS). Error bars represent assay method variability (three standard deviations for buffer). (B) Dose-response data for compounds 3 and 4 with prenylated HVR (light blue & yellow) and with full-length prenylated KRAS^G12V (dark blue & orange); all in PS. Error bars represent assay method variability (three standard deviations for buffer).

https://doi.org/10.1371/journal.pone.0174706.g006

Because we compared KRAS constructs with and without the prenylated C-terminus, we also examined the binding of our inhibitors to an 18-residue KRAS C-terminal prenylated peptide. This peptide is immobilized to the biosensor through biotin and has a farnesyl group attached to the cysteine residue through the thioether link: Biotin-GGG GGG GGE KMS KDG KKK KKK SKT KC (-OCH3)-Farnesyl. Surprisingly, we observed that compound 3 binds to the prenylated peptide in a concentration-dependent manner in BLI experiments with PS (Fig 6B). The response profiles for compound 3 binding to the peptide were more complex than those for compound 3 binding to KRAS protein, showing possible non-specific interactions with the peptide. In this control experiment, binding signals for the compound do not fully dissociate to background levels at 3.3–10 μM, and some secondary binding is observed in the association phase. Consistent with KRAS binding studies using the full-length protein, control compound 4 did not bind to the prenylated peptide (Fig 6B).

Protein preparation

The KRAS constructs His-Tev KRAS^G12V(1–180) and His-Tev Avi-Tag (Gly)8 KRAS^G12V(1–180) were cloned in the pET22b E. coli plasmid (Novagen) with standard molecular biology techniques and the genes were driven by the T7 promoter. The BirA gene was cloned in a pACYC184 derived E. coli plasmid and is driven by the tac promoter. The plasmid with His-Tev KRAS^G12V(1–180) was transformed into Rosetta2 DE3 E. coli cells (Novagen) and grown in a fermenter at 37°C to an OD600 = 8-10/ml in YEP medium using carbenicillin and chloramphenicol for selection, induced with 0.81mM IPTG and grown to OD600 = 30 at 18°C. The His-Tev Avi-Tag (Gly)8 KRAS^G12V(1–180) and the BirA plasmids were co-transformed into BL21 (DE3) E. coli cells (Invitrogen) and grown in flasks in LB medium at 37°C using carbenicillin and chloramphenicol for selection. Induced at 1–2 OD/ml with IPTG to 0.5 mM and grown at 18°C to OD 4/ml. GST-CRAF RBD (aa51-131) construct was cloned in the E. coli pGEX2T plasmid (GE Healthcare) and is driven by the tac promoter. This plasmid was transformed into BL21 E. coli cells (Invitrogen) and grown in flasks in LB medium at 37°C using carbenicillin for selection. Induced at 1–2 OD/ml with IPTG to 0.5 mM and grown at 37°C for 4 hrs.

KRAS full length, BRAF full length and BRAF^V600E full length genes were cloned in the baculovirus expression vector pBlueBac4.5 (Thermo Fisher Scientific) and were driven by the polyhedrin promoter. These plasmids were subsequently used to generate baculovirus by the co-transfection/plaque purification method [51]. Viruses generated from the transfected insect cells were amplified using a standard low MOI infection method. Avi-tag-(Gly)8 –KRAS^G12V (2–188) was cloned into the pFastBac1 vector under the polyhedrin promoter. MEK1^K97R-cAvi tag and BirA were cloned pFastBac Dual vector, driven by the polyhedrin and the p10 promoters respectively. The pFastBac1 plasmids containing Avi-tag-(Gly)8 –KRAS^G12V (2–188) and the pFastBac Dual vector with Avi-MEK^K97R and BirA were used to generate bacmid DNA as instructed in the Bac-to-Bac manual (Invitrogen). Baculoviruses expressing the recombinant proteins were generated using the Bac-To-Bac Expression System method (Invitrogen) following manufacturer’s protocol. The viruses generated from the transfected insect Sf9 cells were amplified using a standard low MOI infection method. The pFastBac1 plasmids containing Avi-tag-(Gly)8 –KRAS^G12V (2–188) and the pFastBac Dual vector with Avi-MEK^K97R and BirA were used to generate bacmid DNA as instructed in the Bac-to-Bac manual (Invitrogen). Baculoviruses expressing the recombinant proteins were generated using the Bac-To-Bac Expression System method (Invitrogen) following manufacturer’s protocol. The viruses generated from the transfected insect Sf9 cells were amplified using a standard low MOI infection method. To generate recombinant proteins for all the insect cell expressed proteins, suspension cultures of insect cells (Sf9, Sf21 or Tn5) were seeded at a density of 1.5x10⁶ cells/ml and infected with virus at an MOI of 10 or with a 3% volume. The infected insect cells were cultured for 48 hours at 27°C shaking at 120 rpm using 2L glass Erlenmeyer flasks and serum free media. Cells were harvested two days post-infection. To generate biotinylated Avi-tagged protein, a suspension culture of insect cells (Tn5) were seeded at a density of 1.5x10⁶ cells/ml and co-infected with virus at an MOI of 5:5 (recombinant baculovirus and BirA-expressing virus, both at passage 2). D-biotin was added to 50uM final concentration.

Cells were collected by centrifugation and resuspended in ~5x volumes of lysis buffers. E. coli cells were lysed by sonication or Microfluidizer (Microfluidics) and insect cells were lysed by Dounce homogenization. The proteins expressed in E. coli or baculovirus/insect cells were purified by a combination of affinity and size-exclusion chromatography.

More detailed descriptions of cloning, expression and purification are provided in S1 Protocols.

Compound synthesis and characterization

Detailed descriptions of synthetic methods and compound characterization are provided in S2 Protocols.

Phospholipid preparation for biochemical and biophysical assays

Phosphatidylserine (PS) liposomes were prepared by suspending the lipid (porcine brain; Avanti Polar Lipids #840032P) at 5 mg/ml (13 mM) in 50 mM Tris buffer, pH 7.5. The suspension was extruded 11 times through a 50 nm pore polycarbonate membrane using an Avestin LiposoFast-Basic Membrane Extruder and the solution was stored at 4°C. This method is similar to previously described methods for the preparation of phosphatidylcholine-based liposomes containing PS [52] with the exception that PS was the only lipid solubilized and it was dissolved directly in 50mM Tris pH 7.5 prior to extrusion. The resulting liposomes were characterized in triplicate by dynamic light scattering on a Wyatt DynaPro reader and were 120nm (polydispersity 16nm). The expected sizes for 100 and 300nm Nanosphere^™ size standards (ThermoFisher Scientific 3100A and 3300A, respectively) were observed. Upon addition of KRAS, the liposomes were 129nm (polydispersity 19nm) in the presence of assay buffer.

Assay for BRAF activation by KRAS

Reactions were run in 50 mM Tris, pH 7.0, 5 mM MgCl₂, 0.01% BSA (Sigma #4503), 0.01% TritonX-100, 100 mM NaCl and 1 mM DTT. BRAF activation was measured as the amount of phosphorylation of inactive MEK protein. The AlphaScreen technology (PerkinElmer #6760671) with an anti-phospho-MEK antibody (Cell Signaling #9121) was used to detect MEK phosphorylation. Recombinant G12V-KRAS (2 nM final), either prenylated or nonprenylated, and either pre-loaded with GDPβS or GTPγS, was combined with PS (final 2 μg/ml), biotinylated MEK (final 20 nM) and ATP (final 10 μM) at 2x for 30 minutes prior to the addition of 2x full length wildtype BRAF fraction. The wildtype BRAF preparation was not quantified but used at a dilution that minimized basal activity. Reactions were stopped with the addition of EDTA plus Tween, final 20 mM and 0.003%, respectively. A mixture of AlphaScreen beads, composed of streptavidin donor and protein A acceptor beads, and anti-phospho-MEK antibody in 50 mM Tris, pH 7.5 and 0.01% Tween 20 was added for a final 20 μg/ml AlphaScreen beads and 1:2000 dilution of antibody. After an overnight incubation, the samples were read for emission at 570 nm after excitation of the donor beads at 680 nm on an Envision 2101 instrument (PerkinElmer). An initial survey of phospholipids was done with recombinant wildtype KRAS (final 100 nM), prenylated and pre- loaded with GPTγS, biotinylated MEK (final 40 nM), ATP (final 50 uM) and full length wildtype BRAF fraction. The liposomes phosphatidylinositol (Avanti Polar Lipids #840042P), L-α-phosphatidylcholine (Avanti Polar Lipids #840051P) and L-α-phosphatidic Acid (Avanti Polar Lipids #840101P) were prepared as described for PS liposomes above. All liposomes were assessed at 100 μg/ml.

For library screening and IC₅₀ determinations (including the control experiment with RBD), the 2x mixture of RAS, lipid, MEK and ATP was combined with compound in DMSO at 2x for 30 minutes before the addition of 2x RAF to start the reactions. The final DMSO concentration was 2.5%. This assay was reproducible from run to run. The geomean of the IC₅₀ ± the standard deviation for a known RAF inhibitor, compound C from [33], was 0.002±0.003 μM (N = 53). The geomean of the IC₅₀ ± the standard deviation for compound 2 was 0.8±0.7 μM (N = 18). After establishing reproducibility with these compounds, all other compounds were measured with N≥2.

Counter assay to eliminate BRAF inhibitors

An assay consisting of full length BRAF^V600E was used to flag inhibitors acting directly on BRAF, taking advantage of the constitutive activity of the BRAF mutant. The BRAF^V600E stock was diluted to give a reasonable signal upon phosphorylation of inactive MEK protein. The counter assay buffer, other components and order of addition matched the RAS-dependent assay except for the presence of KRAS. This assay was reproducible from run to run. The geomean of the IC₅₀ ± the standard deviation for a known RAF inhibitor, compound C from [33], was 0.0007±0.0007 μM (N = 49). After establishing reproducibility with this compound, all other compounds were measured with N≥2.

NMR experiments

All NMR experiments except for the ³¹P experiments were performed at 296K on a Bruker AV600 or AV800 NMR spectrometer, operating at 600 MHz or 800 MHz proton frequency, respectively, and equipped with cryoprobes. The buffer contained 20mM Tris HCl, pH8.0, 50mM NaCl, 1mM MgCl₂, 2mM DTT. HSQC experiments were run with ¹⁵N-labeled and GDP-loaded K-RAS^G12V (1–169) at a concentration of 0.5 mg/ml, and ¹³C,¹⁵N-edited NOESY spectra were recorded with ¹³C,¹⁵N-labeled and GDP-loaded K-RAS^G12V (1–169) at a concentration of 7 mg/ml. The same samples were used to record triple-resonance experiments for resonance assignment. The NMR fragment screen was carried out using GDP-loaded KRAS^G12V (1–180). 500 fragments from a fragment library [53] were screened by T1ρ ligand observation experiments, followed by protein-observed HSQC experiments. ³¹P NMR spectra were acquired at 278°K on a Bruker AV400 spectrometer operating at 161.6 MHz ³¹P frequency, equipped with a BBO probe. Unlabeled GMPPNP-loaded K-RAS^G12V (1–169) was used at a concentration of 5 mg/ml alone or with 1 mM compound 2 or 0.5 mM of the CRAF-RBD.

NMR K_d values were determined by titrating ¹⁵N-labeled GDP-loaded K-RAS^G12V (1–169) protein with increasing amounts of compound 2 and following the resulting chemical shift perturbations. The weighted change in chemical shift of two peaks was calculated using a ¹⁵N scaling factor of 0.1 and the K_d calculated by fitting the data with an equation describing the Langmuir binding isotherm.

For binding site determination, 0.4 mM samples of KRAS^G12V (1–169) with intrinsically bound GDP were prepared in presence of 1 mM compound 2. As reported earlier, backbone and methyl-group assignments were obtained with automated projection spectroscopy [54] and complete side-chain assignments with an H(C^aliC^aro)H-TOCSY [55]. For obtaining inter-molecular NOEs, double-half filter experiments were used [56, 57]. Additionally, a high-resolution spectrum was recorded on a specifically ILVM ¹³C-methyl labeled sample [58], in order to resolve ambiguities in NOE assignments.

Cell lines and reagents

SW1990 and A375 cell lines were obtained from the Broad-Novartis Cancer Cell Line Encyclopedia collection and cultured in DMEM or RPMI supplemented with 10% FBS. Anti-p-ERK1/2 T202/Y204, -ERK1/2, -p-AKT S473, and -AKT antibodies were purchased from Cell Signaling. Anti-GAPDH was purchased from Millipore. MSD Phospho/Total ERK1/2 Whole Cell Lysate Kit (K15107D) was purchased from Meso Scale Discovery. CellTiter-Glo^® Luminescent Cell Viability Assay was purchased from Promega.

Western blot analysis

Cells were seeded in 6-well plates and treated with compounds or DMSO at the indicated concentrations for 3hrs. Total cell lysates were prepared in either M-PER (Pierce) or RIPA lysis buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.5% NP-40, 0.5% sodium deoxycholate) supplemented with 1x protease inhibitor cocktail (Sigma) and 1x phosphatase inhibitor cocktails (Sigma). Lysates were normalized for protein concentration, resolved by SDS-PAGE, transferred onto nitrocellulose membranes and probed with the indicated antibodies.

Proliferation assay

Cells were seeded in 384-well plates and incubated with a 1:3 serial dilution of compound starting from 30 μM at 37°C with 5% CO₂ for 5 days. Cell viability was measured using the CellTiter-Glo^® Luminescent Cell Viability Assay kit (Promega) and read on an Envision (Perkin Elmer) plate reader. Percent growth inhibition was calculated by normalizing treatment to DMSO control. EC₅₀ values, where cell growth is inhibited by 50%, were calculated from dose response curves generated in GraphPad PRISM 6.07 using a non-linear regression 4-parameter curve fitting model. For every compound, each concentration was run in triplicate and all points were used for fitting a single curve. Compound 2 was tested in three independent assays and the geomean of the EC₅₀ ± the standard deviation was 2.7±1.4 μM.

Determination of binding constants with biolayer interferometry

Determination of K_ds was performed on the Octet RED platform, according to the manufacturer's recommendations in their application note on small molecule analysis. The prenylated KRAS peptide containing the farnesyl group attached to the terminal cysteine (biotin-GGGGGGGGEKMSKDGKKKKKKSKTC(farnesyl)-OCH₃) was obtained from AnaSpec (Fremont, California). Proteins were obtained as described above. The assay buffer was 25mM Tris containing 190mM glycine, 5mM MgCl₂, 0.01% Triton X-100, 1mM DTT and 1.3uM PS (liposome preparation as described above). The protein was attached to SuperStreptavidin biosensors (ForteBio 18–5057) via Avi-tagged-KRAS G12V:GTPγS with C-terminal prenylgroup by adding 2μL of the 5mg/mL solution of PS to 2mL of assay buffer, followed by the KRAS protein to a final concentration of 40nM. After incubation for 1h at RT, the protein was loaded into a 96-well plate (Greiner 655209), and loaded a BLI response of 3-4nm. These sensors were subsequently blocked with a 100 μg/mL solution of biocytin, and a 6-point, 3X dilution series of compound was analyzed in the presence of 2.5% DMSO at 28°C. A reference sensor was run to remove systematic drift, and a second set of SuperStreptavidin biosensors blocked with biocytin was examined to remove systematic optical interferences, and to check for non-specific binding of compounds to the biosensor. K_ds were determined using the ForteBio analysis software tool, which determines K_d from the steady-state plot, or from the determination of the kinetic constants k_on and k_off. This assay was reproducible from run to run; the K_d for compound 3 obtained from the steady-state model with multiple preparations of PS liposomes was 2.6±0.8 μM (geomean ± standard deviation; N = 10). After establishing reproducibility with compound 3, all other compounds were measured in singleton.