KRAS-specific inhibition using a DARPin binding to a site in the allosteric lobe

Inhibiting the RAS oncogenic protein has largely been through targeting the switch regions that interact with signalling effector proteins. Here, we report designed ankyrin repeat proteins (DARPins) macromolecules that specifically inhibit the KRAS isoform by binding to an allosteric site encompassing the region around KRAS-specific residue histidine 95 at the helix α3/loop 7/helix α4 interface. We show that these DARPins specifically inhibit KRAS/effector interactions and the dependent downstream signalling pathways in cancer cells. Binding by the DARPins at that region influences KRAS/effector interactions in different ways, including KRAS nucleotide exchange and inhibiting KRAS dimerization at the plasma membrane. These results highlight the importance of targeting the α3/loop 7/α4 interface, a previously untargeted site in RAS, for specifically inhibiting KRAS function.

M embers of the RAS family of oncogenic proteins are frequently mutated in human cancers. In particular, KRAS mutations are the most prominent ones, representing around 86% of all RAS mutations 1 . KRAS mutants are major drivers of cancers, such as colorectal, lung or pancreatic cancers 1 . Isolation of selective KRAS inhibitors that block its function is therefore an important goal 2 . Nonetheless, selectively targeting KRAS is challenging, as RAS isoforms are highly similar in primary sequence with 82-90% amino acid sequence identity 3 .
Specifically targeting directly mutant KRAS has been achieved with small molecules covalently binding the G12C mutant KRAS [19][20][21] . This approach targets the G12C mutation that represents around 12% of KRAS mutations in cancers (Cosmic database v86, https://cosmic-blog.sanger.ac.uk/), and is only present in a subset of cancers, such as non-small cell lung cancers 22 . Therefore, alternative strategies are needed to inhibit the most frequent mutations of KRAS accounting for 88% of KRAS mutant cancers.
We report here the characterisation of two potent DARPins that selectively bind KRAS on a site of the allosteric lobe, encompassing histidine residue 95. The DARPin binding inhibits KRAS nucleotide exchange and KRAS dimerisation, thus impairing mutant KRAS-effector interactions and the downstream signalling pathways. These findings reveal a unique strategy to selectively inhibit KRAS.

Results
Isolation of anti-KRAS-specific DARPins. We performed a phage display selection of a diverse DARPin library 8 , followed by immunoassays with KRAS G12V to isolate hits. We have identified two DARPins (designated K13 and K19) that bound to KRAS G12V . Biochemical analysis of the DARPins show K13 and K19 interact with KRAS independently of the nucleotide-bound state of the GTPase, and have K d s around 30 and 10 nM, respectively ( Supplementary Fig. 1a). The nucleotide and protein sequences of DARPins K13 and K19 are shown in Supplementary  Fig. 1b, c and highlight a conserved amino acid sequence in the repeat regions with only six amino acids difference.
The X-ray structure data of K13 and K19 in complex with KRAS G12V show these DARPins bind to the allosteric lobe of KRAS, at the interface between helix α3/loop 7/helix α4 (Fig. 1a, b; Supplementary Table 1). The crystal structures show that when DARPins K13 or K19 bind to KRAS, a structural change appears in the KRAS molecule on the effector lobe, especially on the switch 1 and 2 when compared with two unbound KRAS G12V -GDP structures ( Supplementary Fig. 2a, b). However, the exact conformation of the switch 1 loop in the K13-and K19-bound states differ somewhat. This difference is most likely due to their different crystal-packing environments ( Supplementary Fig. 2c, d). NMR chemical shift perturbation HSQC and hydrogen deuterium exchange with mass spectrometry (HDX-MS) data support the observed binding interface in solution of K19 in the allosteric lobe  [4][5][6]. After K19 binding to KRAS, a small but significant increase in the dynamic mobility of the switch 2 loop is shown by the increase in de-protection observed by HDX-MS , and some small perturbations of the effector lobe HSQC resonances are observed in a few residues in the switch 2 region (Fig. 2a-c). Our data suggest that the conformational change observed by X-ray crystallography on the switch regions is most likely due to the crystal-packing effect, because the switch regions are flexible by nature.  (Fig. 3a, b). In addition, K13 and K19 interact equally with the dominant-negative KRAS S17N , the constitutively activated mutant KRAS G12D and wild-type KRAS, unlike K27, which does not contact the KRAS S17N mutant (Fig. 3a, b). Indeed, K13 and K19 bind KRAS away from the switch regions, and are not affected by the S17N mutation while K27 binding is expected to be affected as the flexibility of the switch regions is modified on that mutant 24 . These selectivity findings were further supported using a co-immunoprecipitation assay with cells co-expressing 3xFLAG-tagged wild-type RAS and GFP-tagged DARPin. KRAS, NRAS and HRAS proteins were immunoprecipitated on beads, and the presence of DARPin-GFP bound to RAS proteins was assessed. The DARPin K27 was captured by all three RAS isoforms, but K13 and K19 were only captured by KRAS, while the negative control DARPin E3.5 8 was not co-immunoprecipitated by any RAS proteins (Fig. 3c).
K13 and K19 bind KRAS by interacting with the H95 of KRAS. We evaluated which residue(s) are involved in this KRAS selectivity of K13 and K19. Analysis of the amino acids engaged in the interaction between the DARPins and KRAS, showed that K13 and K19 make extensive interactions with the histidine 95 and the glutamic acid 107, which are residues only present in KRAS ( Supplementary Fig. 7a) 15 N TROSY spectra for apo KRAS G12V at 100 μM (black) overlaid with K19-bound KRAS G12V at 89 μM (green). DARPin concentrations were in slight excess. An expanded region is shown on the right. b Amino acids undergoing chemical shifts upon DARPin K19 binding on KRAS G12V -GDP are shown in blue, residues not experiencing shifts are shown in turquoise in the KRAS G12V sequence (unassigned residues are not highlighted). The P-loop is highlighted in yellow, the switch 1 in dark yellow and the switch 2 in purple. c Ribbon representation of (b) with amino acids undergoing chemical shifts in blue and residues not experiencing shifts in turquoise in the KRAS G12V -GDP structure (PDB 4TQ9) after DARPin K19 binding (unassigned residues are in grey). L7: loop 7, α3: helix 3, α4: helix 4 HIS95 with leucine or glutamine (found at residue 95 in NRAS and HRAS, respectively) and substituting GLU107 with aspartic acid (found at residue 107 in both NRAS and HRAS). While the E107D mutation did not affect K13, K19 or K27 DARPins binding with KRAS ( Fig. 3d), H95Q and H95L mutations drastically decrease K13 and K19 binding on KRAS WT but do not affect K27 binding as shown by BRET (Fig. 3d) or coimmunoprecipitation ( Supplementary Fig. 7b). Supplementary Table 2 summarises the effects of RAS mutations on the binding of the DARPins. Finally, we assessed the residues of the DARPins important for binding to KRAS by mutating two tryptophans found in the first repeat of both K13 and K19 to glycine residues   Fig. 9a, b). These results suggest that K13 and K19 are selective KRAS inhibitors. K13 and K19 DARPins selectively inhibit the PI3K/AKT pathway in the KRAS mutant HCT116 cell line, but not in the other cancer cell lines, while K27 inhibits this pathway in HCT116 and MCF-7 cells (Fig. 5a, b). These data are consistent with the finding that the PI3K/AKT pathway is only partly dependent on RAS, and can be activated independently of RAS 27,28 . These results demonstrate that K13 and K19 selectively inhibit endogenous mutant KRAS function without affecting RAS WT cancer cells.
K13 and K19 inhibit KRAS function by at least two mechanisms. While K13 and K19 bind to the allosteric lobe of KRAS and interfere with effector interactions, it suggested new mechanism (s) by which DARPins inhibit KRAS function. To this end, we tested these DARPins in biochemical assays, where K13 and K19 were compared with the known RAS nucleotide exchange inhibitor K27 and known RAS/CRAF RBD interaction inhibitor K55 8 . In vitro analyses show that DARPins K13 and K19 inhibit the biochemically coupled KRAS assay based on the inhibition of nucleotide exchange or KRAS-CRAF RBD interaction (Fig. 6a) as does K27 but, unlike K55, K13 and K19 do not prevent KRAS/ CRAF RBD interaction in vitro ( Fig. 6b) even though they do in the cell-based BRET assay (Fig. 4a). Furthermore, these DARPins lock down the GDP-and GTP-bound KRAS, thus preventing the nucleotide exchange by SOS (Fig. 6c). Nonetheless, the binding of the DARPins to KRAS is nucleotide independent ( Fig. 3a; Supplementary Fig. 1a), and the binding occurs on a surface remote from the switch regions ( Fig. 1). However, SOS not only contacts RAS near the switch 1 and 2 regions but also makes additional interactions with the residues 95-105 of the helix α3 29 . These contacts would cause a clash between K13 and SOScat for KRAS binding (Fig. 6d) and indicate why K13 and K19 inhibit the nucleotide exchange of KRAS.
In addition, we investigated whether the DARPins could interfere with KRAS dimerisation, as it has been previously described with a different allosteric inhibitor, the monobody NS1 12 . Indeed, the α3-α4 and α4-α5 regions are defined as the potential dimer interfaces of KRAS 14-17 and K13 and K19 interact with the α3-α4 site of KRAS ( Supplementary Fig. 7a).  Consequently, the binding of K13/K19 with KRAS is expected to perturb its dimerisation. We tested this hypothesis by BRET experiments involving the dimerisation of mutant K-, N-and H-RAS. K13 and K19 essentially inhibit KRAS G12D dimerisation, while NRAS Q61H and HRAS G12V dimers formation is less affected by the DARPins (Fig. 7a-c). It has been previously reported that the CAAX motif of RAS is mandatory for its dimerisation 18 . Therefore, we investigated RAS dimerisation using KRAS G12V deleted of its last 22 amino acids (a KRAS G12V 166 mutant missing the whole hypervariable region, HVR, including the CAAX motif) that should remove its ability to form RAS dimers. First, we showed that KRAS G12V 166 was soluble and expressed in the cytoplasm of cells (Supplementary Fig. 10a) and was not able to form dimers in cells by BRET donor saturation assays ( Supplementary Fig. 10b). We then performed a BRET competition assay with K13 and K19 against KRAS G12V FL or KRAS G12V 166 interaction with CRAF RBD showing that removal of the HVR from KRAS G12V only reduced K13/K19 inhibition of KRAS G12V /CRAF RBD interaction ( Supplementary Fig. 10c). These results show that RAS dimerisation in cells is not the only mechanism of inhibition of KRAS by the DARPins, as these retain the ability to inhibit KRAS G12V 166 (Fig. 5). In a previous study, an antisense oligonucleotide (AZD4785) was described that selectively targets KRAS mRNA 30 , but both wild-type and mutant mRNAs. AZD4785 downregulates both WT and mutant KRAS protein, but only inhibits the downstream MAPK and PI3K signalling pathways of mutant KRAS expressing cells and not of KRAS WT expressing cancer cells 30 . Therefore, our data also suggest that inhibition of both WT and mutant KRAS do not affect the RAS downstream signalling pathways of RAS WT cancer cells. This effect could be explained by compensatory and/or redundancy mechanisms of the other RAS isoforms 31 .
Although the crystal structures of K13 and K19 bound to KRAS show substantial conformational change on both switch regions, the NMR HSQC and HDX-MS only show a small conformational change on the switch 2 region of KRAS upon the DARPins binding ( Fig. 2 and Supplementary Figs. 3-6). In addition, K13/K19 do not prevent the interaction KRAS G12V / CRAF RBD in vitro (Fig. 6b), indicating that the conformational change of the switch 2 is not the main driver of KRAS function inhibition in cells. The mechanism by which K13 and K19 inhibit KRAS function is complex. KRAS dimerisation occurs through the α3-α4 and/or α4-α5 interfaces that are dimerisation interfaces of KRAS [14][15][16][17] . However, only the α4-α5 has been functionally tested in cells [11][12][13] . We now show that targeting α3-α4 interface also leads to KRAS function inhibition. The mechanism underlying this inhibition could rely on inhibition of RAF activation as described with a monobody 12 . The DARPin binding may also promote a certain degree of membrane occlusion of the effector-binding region of KRAS and impede the binding of effectors on the switch regions 32 . Nevertheless, because the GEF SOS contacts the α3 region of RAS 29 (as well as switch 1 and 2), K13 and K19 presumably also impede KRAS nucleotide exchange by engaging the α3-α4 site. Therefore, it adds another level of complexity in the interpretation of the mechanisms responsible for KRAS inhibition. We conclude that K13/K19 can affect KRAS function by at least two different mechanisms, including nucleotide exchange and RAS dimerisation inhibition.
None of the mechanisms above would explain why DARPins K13 and K19 disrupt the binding of K27 to KRAS WT in a BRET competition assay in cells ( Supplementary Fig. 11a). This interaction should not rely on RAS dimerisation, since K27 interacts predominantly on the switch 1 of KRAS-GDP and no steric clash appears visible between K27 and K13 or K19 ( Supplementary  Fig. 11b, c). Therefore, the mechanism of inhibition of KRAS by K13 and K19 seems more complex than only the prevention of nucleotide exchange and KRAS dimerisation. While K27 only binds KRAS-GDP, K13/K19 also interact and lock down GTPbound KRAS ( Fig. 6c; Supplementary Fig. 1a). Hence, K13/K19 could inhibit the GTPase-activating proteins (GAP) binding on KRAS. The superimposition of the structures of two GAP proteins (p120 RAS GAP, PDB 1WQ1 and NF1, PDB 1NF1) with the structures of KRAS G12V -GDP/K13 ( Supplementary Fig. 12a, b) and KRAS G12V -GDP/K19 ( Supplementary Fig. 12c, d) shows that the amino terminal end of these DARPins would overlap with the bound GAP proteins, suggesting that K13/K19 may be GAP inhibitors. Even though a putative GAP inhibitory mechanism could explain the inhibition of KRAS WT /K27 interaction by K13/ K19, it is not the mechanism that prevails in either mutant KRAS or KRAS WT cancer cells. Indeed, RAS signalling is decreased in HCT116 cells (bearing both mutant and WT KRAS alleles), and is not increased in MCF-7 cells (homozygous for KRAS WT ) when K13/K19 are expressed (Fig. 5). Nevertheless, HT1080 cells (bearing NRAS Q61K and KRAS WT alleles) expressing K19 present an increased level of pERK and pMEK signals (Fig. 5). This result could be assigned to the inhibition of the GAP binding on KRAS WT by K19. However, this effect is not observed in T24 cells (bearing HRAS G12V and KRAS WT alleles), suggesting this increase of signalling might be cell line dependent. Therefore, the GAP inhibitory effect on RAS downstream signalling could be balanced by the dimerisation inhibition of KRAS WT and/or by compensatory mechanisms involving for instance the two other RAS isoforms 31 and explaining why RAS signalling is not always affected in KRAS WT cancer cells, such as MCF-7.
The therapeutic use of these inhibitors with an intracellular protein such as KRAS is currently limited, but this work has identified an allosteric site on KRAS and the different inhibitory mechanisms involved in targeting this interface. Hence, the α3/loop 7/α4 region could be targeted for inhibition via alternative approaches. In future studies, the DARPins K13 and K19 could be used as surrogates to select small molecules targeting the α3/loop Statistical analyses in a-c were performed using a one-way ANOVA followed by Dunnett's post-hoc tests (**P < 0.01, ***P < 0.001, ****P < 0.0001). Each experiment was performed four times (a-c). Error bars are mean ± SD of biological repeats. a-c Source data are provided as a Source Data file 7/α4 allosteric interface, especially around the KRAS selective residue H95. The surrogate method has been successfully used previously to isolate RAS-binding compounds using an anti-RAS intracellular single-domain antibody fragment 7 . Furthermore, an in silico study showed a potential small molecule-binding site on that interface 33 . This would be a promising way to isolate KRAS selective small inhibitors for the treatment of KRAS mutant cancers.

Methods
Isolation of DARPins K13 and K19. DARPins were isolated from a phage display library 34 by panning selections on biotinylated KRAS G12V (1-166) pre-bound to streptavidin magnetic beads, as previously described 8 . DARPin K19 was isolated from a first round phage display output that was further selected against KRAS G12V (1-166) by ribosome display and error prone mutagenesis, using methods described by Groves et al. 35 To convert a DARPin phage display library into a format compatible with ribosome display, primer DARPin-RD1 (5′-AGACCACA ACGGTTTCCCTCTAGAAATAATTTTGTTTAACTTTAAGAAGGAGATATAT CCATGGCCGATCTGGGAAA-3′) was used in place of primer SDCAT-DP47 35 .
For characterisation, DARPins were sub-cloned to the pET16b vector and expressed cytoplasmically in BL21 (DE3) Escherichia coli (New England Biolabs). A list of the primers used in this study is shown in Supplementary Table 3. Following lysis in BugBuster plus Benzonase (EMD Millipore), DARPins were purified to homogeneity using nickel-chelate chromatography, followed by size exclusion chromatography to provide a monomeric protein in PBS (pH 6.5).
KRAS expression and purification. The human KRAS gene sequence (residues 1-166, Isoform 2B, P01116-2) containing the G12V substitution was cloned into a pET28b vector with an N-terminal His6 tag. KRAS G12V was expressed in BL21 (DE3) Escherichia coli (New England Biolabs) and cell pellets resuspended in 50 mM HEPES (pH 7.4), 100 mM NaCl, 20 mM Imidazole, 2 mM TCEP and 2 mM MgCl 2 . Following sonication, the sample was centrifuged at 20,000 rpm and proteins purified from the supernatant by Ni-NTA chromatography on a 5 -mL HisTrap HP column (GE Life Sciences, cat # 17-5248-02). Following elution in 50 mM HEPES (pH 7.4), 500 mM NaCl, 400 mM Imidazole, 2 mM TCEP and 2 mM MgCl 2 , samples were further purified by size exclusion chromatography on a Superdex 75 column. Final buffer composition was 50 mM HEPES (pH 7.4), 100 mM NaCl, 2 mM MgSO 4 . KRAS G12V was biotinylated via an Avi tag using BirA enzyme and then exchanged with either GDP or GTPγS, catalysed by SOS (Son of Sevenless). The GTPγS in the KRAS GTPγS had a tendency to be hydrolysed to GDP on storage, so the levels were determined by electrospray mass spectroscopy.
X-ray structure. DARPins K13 and K19 were mixed with KRAS G12V -GDP in equimolar ratio with a final concentration of 16 mg.mL −1 and incubated on ice for 30 min prior to crystallisation screening at 277 and 293 K. Hits were observed in several conditions. Both K13 and K19 complexes gave bipyramidal crystals in 0.5-1.0 M lithium sulphate, 0.5-1.0 M ammonium sulphate, 100 mM tri-sodium citrate pH 5.5. A cryoprotectant solution of 2 M lithium sulphate could be used for crystals of both complexes.
The diffraction data were collected at 100 K temperature at the Diamond Light Source, UK on beamlines I04-1 and I03 at 0.920 and 0.976 Å wavelength, respectively. XDS, pointless and scala were used to process the data. Model building and subsequent refinement were carried out with Coot and Refmac. There are no Ramachandran outliers in either structures and over 98% are in the most favoured region. Coordinates and structure factors were deposited in the PDB with accession codes 6H46 and 6H47. Data collection and refinement statistics are summarised in Supplementary Table 1. NMR spectroscopy. Protein was expressed in BL21-Gold cells grown in M9 media supplemented with 5 g.L −1 Celtone N powder and containing 50 μg.mL −1 kanamycin and 15 μg.mL −1 tetracycline. Purification was performed as described above. Protein was dialysed overnight into NMR buffer (50 mM HEPES pH 7.4, 50 mM NaCl, 2 mM MgCl 2 , 2 mM TCEP, 0.1 mM EDTA, 0.02% NaN 3 ) 36 . All NMR spectra were collected at 298 K on a Bruker 800 MHz, Avance III spectrometer, equipped with a 5 -mm TCI Cryoprobe with z-axis gradients using standard experiments and parameters from Bruker library. Uniformly 15-N-labelled KRAS G12V was used at a concentration of 100 μM with concentrations of added non-labelled DARPins as indicated. NMR samples were prepared in HEPES-based buffer at pH 7.4 supplemented with 5% D 2 O. All spectra were processed in Topspin 3.1. Binding was monitored by 1H-15N 2D TROSY (transverse relaxation optimised spectroscopy) spectra of the protein acquired with (F2 × F1) 2048 × 160 complex pairs (in Echo-Antiecho mode), 12019 × 2757 Hz sweep width and 85.2 × 29.0 ms acquisition times. Assignments were mapped from deposited assignments of KRAS WT BMRB ID: 18529 and HRAS G12V BMRB ID: 25730.
HDX mass spectrometry. Protein samples for HDX-MS were prepared in 20 mM HEPES-Na pH 7.4, 100 mM NaCl, 5 mM MgCl 2 , 1 mM TCEP at final protein concentrations of 10 μM GDP-loaded KRAS G12V and 20 μM DARPin, and kept at 4°C, unless otherwise stated. The deuterium exchange reactions were performed using a LEAP PAL RTC system (LEAP Technologies) and all conditions were performed in triplicate. Peptides were identified that covered 98.8% of the protein sequence; considering the rapid back-exchange of the two N-terminal residues, this reduces to 91.7%.
Peptide identification was performed in Protein Lynx Global Server (Waters) using MS E data collected for the triplicate undeuterated control samples. The resultant peptide lists were imported into DynamX (Waters) where peptides were filtered: minimum intensity of 5000, minimum of 0.3 products per amino acid, a maximum MH + error of 5 ppm, and found in all of the undeuterated data sets. The automatic peptide assignment in DynamX was performed using the standard parameters, but all peptides were manually checked for the charge-state assignment, overlapping peptides and retention time. The data were not corrected for back exchange, so are relative rather than absolute deuterium uptake values. The first two residues of each peptide were excluded from analyses due to rapid back exchange. Significant differences were determined by performing a Student's t test using the pooled standard deviation or peptide standard deviation, whichever was the greater. The p-values were adjusted for false discovery using the Benjamini-Hochberg procedure. Differences with p < 0.05 were considered significant.
Peptide-level differences in deuterium uptake are calculated according to the Eq. (1): pep i state : Difference in deuterium uptake for peptide i in state of interest uptake i state : Deuterium uptake for peptide i in state of interest uptake i apo : Deuterium uptake for peptide i apo state The mean residue-level differences were calculated using the Eq. (2): res j : mean difference for residue j N: number of peptides containing residue j pep i : deuterium uptake difference for peptide i that contains residue j amide i : number of exchangeable residues within peptide i KRAS biochemical coupled assay and KRAS/CRAF RBD assay. The KRAS biochemical coupled assay was performed as previously described 8 . Briefly, biotinylated KRAS G12V was pre-incubated with streptavidin-Europium chelate to form a complex, and in a separate reaction, CRAF RBD-GST (glutathione-S-transferase) was pre-incubated with anti-GST-XL665 to form a second complex. Test samples were incubated with the streptavidin:KRAS complex for 15 min before addition of GTPγS and SOS (to initiate nucleotide exchange) and the CRAF RBD:anti-GST complex. Final concentrations added were 2 nM biotinylated KRAS; 37.5 ng.mL −1 streptavidin-Europium; 12 nM CRAF RBD-GST; 2 mg.mL −1 anti-RAF GST XL665; 4 mM GTPγS and 2 mM SOS. In addition, the buffer for the CRAF RBD: anti-GST complex contained 0.1 mg.mL −1 BSA and 0.1 M potassium fluoride. After 1 h incubation, fluorescent resonance energy transfer (FRET) was measured on an Envision plate reader at emission wavelengths 620 and 665 nm. The ratio of these values was fitted using non-linear regression in the application Prism (GraphPad Software). An assay to measure only the inhibition of KRAS/CRAF RBD was like the coupled assay above, with the exception that test samples were added after the completion of the nucleotide exchange step.
MANT nucleotide exchange and release assays. Exchange of GDP with fluorescent MANT-dGDP or fluorescent MANT-GTP on KRAS was studied in an assay previously described 8 . DARPins were tested for inhibition of nucleotide exchange by incubating at 10 μM with 500 nM KRAS G12V , 1 mM SOS and 500 nM MANT-dGDP or MANT-GTP (Biolog). Exchange was measured over time by the increase in fluorescence of MANT-dGDP or MANT-GTP upon binding to KRAS and detected using an Envision plate reader (PerkinElmer) using 340 nm/450 nm excitation/ emission filters. The data were captured at times from zero to 60 min. For nucleotide release assays, DARPins were added after equilibration of MANT nucleotides. medium complemented with 4% FBS. Cells were incubated for an additional 20-24 h at 37°C before the BRET assay reading. A step-by-step protocol is described elsewhere 37 .
BRET2 measurements. BRET2 signal was determined immediately after addition of coelenterazine 400a substrate (10 μM final) to cells (Cayman Chemicals), using an Envision instrument (2103 Multilabel Reader, PerkinElmer) with the BRET2 Dual Emission optical module (515-30 and 410-80 nm; PerkinElmer). The total GFP 2 fluorescence was detected with excitation and emission peaks set at 405 and 515 nm, respectively. The total mCherry fluorescence was detected with excitation and emission peaks set at 530 and 615 nm, respectively. The total RLuc8 luminescence was measured with the luminescence 400-700 nm-wavelength filter.
The BRET signal or BRET ratio corresponds to the light emitted by the GFP 2 acceptor constructs (515-30 nm) upon addition of coelenterazine 400a divided by the light emitted by the RLuc8 donor constructs (410-80 nm). The background signal is subtracted from that BRET ratio using the donor-only negative control, where only the RLuc8 plasmid is transfected into the cells. The normalised BRET ratio is the BRET ratio normalised to a negative control (DARPin control) during a competition assay. The total GFP 2 , mCherry and RLuc8 signals were used to control the protein expression level from each plasmid.
Immunoprecipitation assay. HEK293T cells were transfected 48 h in duplicate (two wells per condition) with pEF-3xFLAG-RAS and pEF-DARPins-GFP 2 plasmids. Cells were washed once with PBS and lysed in the immunoprecipitation buffer (150 mM NaCl, 50 mM Tris-HCl pH 7.4, 10 mM MgCl 2 , 10% glycerol and 0.5% Triton) supplemented with protease inhibitors (Sigma) and phosphatase inhibitors (Thermo-Fisher) for 20 min. Lysates were centrifuged for 15 min, and the supernatant incubated with protein G magnetic beads (Life Technologies) and anti-FLAG antibody (Sigma). The complexes were incubated 4 h at 4°C on a wheel. Beads were washed five times with the IP buffer, the bound proteins were eluted with 1× loading buffer and resolved on 12.5% SDS-PAGE.
Western blot analysis. Cells were washed once with PBS and lysed in SDS-Tris buffer (1% SDS, 10 mM Tris-HCl pH 7.4) supplemented with protease inhibitors (Sigma) and phosphatase inhibitors (Thermo-Fisher). Cell lysates were sonicated with a Branson Sonifier, and the protein concentrations determined by using the Pierce BCA protein assay kit (Thermo-Fisher). Equal amounts of protein were resolved on 10 or 15% SDS-PAGE and subsequently transferred onto a PVDF membrane (GE). The membrane was blocked either with 10% non-fat milk (Sigma) or 10% BSA (Sigma) in TBS-0.1% Tween20 and incubated overnight with primary antibody at 4°C. After washing, the membrane was incubated with HRPconjugated secondary antibody for 1 h at room temperature (RT, 25°C). The membrane was washed with TBS-0.1% Tween and developed using Pierce ECL Western Blotting Substrate (Thermo-Fisher) and CL-XPosure films (Thermo- Confocal microscopy. HEK293T cells were seeded on coverslips and transfected 24 h with pEF-GFP 2 -KRAS G12V 166 construct. Coverslips were washed in PBS, fixed 10 min in 4% paraformaldehyde and washed twice in PBS. Then the coverslips were mounted with DAPI Fluoromount-G overnight. Slides were analysed using a Zeiss 880 Inverted Confocal Microscope with a ×63 objective. Confocal images were analysed with ImageJ software. Quantification and statistical analysis. The quantifications were performed using ImageJ or Prism 7.0c (GraphPad Software), BRET titration curves and statistical analysis were performed using Prism 7.0c (GraphPad Software). The data are typically presented as mean ± SD or SEM as specified in the figure legends. Statistical analyses were performed with a one-way ANOVA followed by Dunnett's post hoc tests, unless otherwise indicated in the figure legends. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Structure files and coordinates have been deposited to PDB under these accession numbers: 6H46 and 6H47. All relevant data are within the paper and its Supplementary Information file and the Source data file. The source data underlying Figs. 3c, 4a-d, 5a, b, 7a-c and Supplementary Figs. 7b, 9a, b, 10c and 11a