RAS-inhibiting biologics identify and probe druggable pockets including an SII-α3 allosteric site

RAS mutations are the most common oncogenic drivers across human cancers, but there remains a paucity of clinically-validated pharmacological inhibitors of RAS, as druggable pockets have proven difficult to identify. We have identified two RAS-binding Affimer proteins, K3 and K6, that inhibit nucleotide exchange and downstream signalling pathways with distinct isoform and mutant profiles. Affimer K6 is the first biologic to bind in the SI/SII pocket, whilst Affimer K3 is the first non-covalent inhibitor of the SII region, revealing a novel RAS conformer with a large, druggable SII/α3 pocket. This work demonstrates the potential of using biologics with small interface surfaces to select novel druggable conformations in conjunction with pharmacophore identification for hard-to-drug proteins.


INTRODUCTION
The RAS family of small GTPases consists of four members, KRAS4A, KRAS4B, HRAS and NRAS, which act as bi-directional molecular switches that cycle between an inactive GDP-bound form and an active GTP-bound form 1 . Mutations in RAS are the most common oncogenic drivers, with KRAS being the most frequently affected member especially in pancreatic, lung and colon cancer 1 . This makes RAS a strong therapeutic target, but despite having been identified as a drug target for over 30 years, only recently have compounds been developed that show promise in preclinical trials 2 . This paucity of agents has been driven by the lack of clearly druggable pockets on the surface of RAS. However, recent work has identified two pockets that may be amenable for drug binding [3][4][5][6][7][8][9][10][11] . The first of these, the SI/II-pocket, exists between the Switch I and Switch II regions of RAS in an area involved in the binding of the nucleotide exchange factor, Son of Sevenless (SOS). Several groups have independently developed compounds that bind this pocket with varying affinities and efficacies, predominately in the micromolar range [5][6][7][8] , except for BI-2852 which has nanomolar binding affinity and efficacy 11 . The second, the SII-pocket, is located under the Switch II loop and was identified using a tethered warhead approach relying on the reactive nature of the cysteine in the G12C mutant 9,10 . This pocket is not fully formed in published KRAS structures in the absence of the inhibitor; however, a groove is evident in some structures and it has been identified as a potential allosteric site computationally 3,4 . Development of tethered compounds targeting this pocket has led to the only series of RAS inhibitors currently in clinical trials 12,13 . This compound series is limited to cancers harbouring G12C mutations and it would be interesting to determine whether this cryptic groove can be exploited in other RAS isoforms and mutants.
Although to date some of these have been used to assist the identification of small molecules 5,7,21 , none has directly probed druggable pockets on RAS, as the majority of these biologics tend to bind over large protein interfaces 14,15,17,20 that are difficult to mimic with small molecules 22 . As some biologics form smaller interfaces 19,23,24 there emerges the tantalising prospect that biologics could be used as tools to identify druggable pockets and novel conformers, and could also have the potential to act as pharmacophore templates for in silico-informed drug discovery. Here, we explore the possibility of using Affimer proteins, an established biologic with a small probe surface formed by two variable regions 24 known to bind at protein interaction 'hotspots' 23,25,26 , to identify and probe RAS for druggable pockets and conformers that might be amenable to small molecule inhibition. Such a direct approach utilising small probe surfaces has not previously been used with biologics and could revolutionise drug discovery, exemplifying a novel pipeline for small molecule design that has the potential to unlock the vast number of currently 'undruggable' proteins.
Here, we demonstrate the use of Affimer proteins to identify and directly probe two druggable pockets on wild-type KRAS associated with inhibition of nucleotide exchange and effector molecule binding. The Affimer that binds to the SI/SII pocket actually mimics the current small molecule inhibitors that target the pocket, providing a proof-of-principle for using Affimer-target interfaces as pharmacophore templates.
The Affimer that binds the SII region selects a novel conformer of the pocket present in wild-type KRAS. The Switch II region adopts a more open position, demonstrating that selecting and targeting this site via non-covalent binding is possible. Our work demonstrates two important concepts in the use of biologics: first, biologics can act as pharmacophore templates for development of small molecule inhibitors; second, they can be used to select for, and stabilise, conformations that are only present as a small fraction of the conformations of the target protein in solution, particularly those that may not be present in extant crystal structures. This approach is likely to be applicable to other important therapeutic targets, and exemplifies a novel pipeline for drug discovery.

Identification and biochemical characterization of anti-RAS Affimers.
Seven unique Affimer proteins that bind wild-type KRAS in both the inactive GDPbound form and the active form, bound to the non-hydrolysable GTP analogue -GppNHp, were isolated by phage display 27 (Supplementary Table 1). To identify inhibitors of RAS, these Affimer proteins were screened for their ability to inhibit SOS1-mediated nucleotide exchange, the primary process in RAS activation. Three of the Affimers, K3, K6 and K37, showed clear inhibition of this process (Fig. 1a), the remaining four Affimers that bound to KRAS showed partial (Affimers K19 and K68) or no inhibition (Affimers K69 and K91) of nucleotide exchange. The latter four Affimers are not discussed further. Affimer K3 displayed the greatest inhibition of nucleotide exchange on wild-type KRAS with an IC 50 of 144 ± 94 nM, with Affimers K6 and K37 also displaying strong inhibition with IC 50s of 592 ± 271 nM and 697 ± 158 nM, respectively (Supplementary Table 2). Next, the abilities of the inhibitory RAS-binding Affimers, K3, K6 and K37 to disrupt the interaction of RAS with its effector protein RAF were determined by KRAS:RAF immunoprecipitation experiments ( Fig. 1b and c)

Affimer proteins bind to intracellular RAS and inhibit downstream signalling.
We then examined whether the Affimer proteins retained their ability to interact with and inhibit RAS in human cells by using transiently transfected HEK293 cells and His-tagged Affimer proteins. Affimers K3, K6, and K37 all showed the ability to pull down endogenous RAS, while the control Affimer showed no such activity (Fig. 2a Fig. 2b and c).
To further study the impacts of RAS-binding Affimer proteins on ERK phosphorylation, we developed an immunofluorescence assay to allow the phosphorylation levels of the endogenous ERK to be examined. HEK293 cells were transiently transfected with tGFP-tagged Affimer-expressing constructs, stimulated with EGF, fixed and stained with an anti-phospho-ERK (pERK) antibody and analysed for alterations to nuclear pERK levels. In accordance with results from the immunoprecipitation experiments, expression of all three Affimer proteins resulted in significant reductions in pERK levels (One-way ANOVA with Dunnett's post-hoc test p<0.001 for all three Affimers), with K3 having a less pronounced effect (ca. 50% reduction for K3, compared with K6, 90% and K37, 85% reductions) ( Fig. 2d and   in the Q61R background showed a significant lack of inhibition of pERK nuclear intensity (p<0.05 Two-way ANOVA with Tukey's post-hoc test) (Fig 3d)

K6 binds the SI/II hydrophobic pocket on KRAS
The crystal structure of Affimer K6 in complex with GDP-bound wild-type KRAS was determined at 1.9 Å resolution, revealing that Affimer K6 binds to a shallow hydrophobic pocket on KRAS between the switch regions ( Fig. 4a and Supplementary Table 3). The Affimer K6 binding sites overlaps that of SOS1 providing structural evidence that K6 acts as a SOS1 competitive inhibitor ( Supplementary Fig. 1a). The binding site further overlaps with that of the RAS- W43 of Affimer K6, and V7/L56 from KRAS form a hydrophobic cluster strengthened by Affimer residues T41 and Q45 forming hydrogen bonds with KRAS switch region residues D38/S39 and Y71, respectively ( Fig. 4b -top right panel). The importance of these amino acid residues for K6 function was confirmed by mutational analysis.
Individual replacement of P42, W43, F44 or Q45 by alanine reduced Affimermediated inhibition of nucleotide exchange (Fig. 4d). These data also revealed the importance of residues F40, N47 and R73 for the inhibitory function of Affimer K6.
Indeed, complete removal of the second variable region abolished the inhibitory ability of K6 (Fig. 4d ∆VR2). This effect is most likely a result of F40, N47 and R73, and indeed Q45 forming intra-Affimer hydrogen bonds that stabilise the tripeptide, P42, W43, F44 (Fig. 4b -bottom panel). These data suggest that the functional Affimer motif responsible for binding and inhibition of KRAS is small, in agreement with the total interacting surface interface estimated by PISA analysis 29 of 478.3 Å 2 for the K6:KRAS complex, a substantially smaller area than most common proteinprotein interaction surfaces. The combination of a functional motif and small interaction interface provides a good basis from which to consider the development of small molecule inhibitors.
Affimer K6 binds the SI/SII pocket that has been previously documented [3][4][5][6][7][8]11 , and alanine scanning has suggested a functional pharmacophore from K6 is responsible for the observed inhibition. This PWFQxN peptide motif is also present in Affimer K37 and it is thus likely that K37 interacts in a similar manner to K6, but this remains to be confirmed. This binding pocket had previously been identified, and a number of small molecules exist to target it, including DCAI 6 , compound 13 8 , Abd-7 7 and BI-2852 11 . We measured the affinity of K6 for both GDP-and GppNHp-bound KRAS by SPR to test whether this was comparable to these small molecules. Affimer K6 bound both forms of KRAS with low nanomolar affinities and showed a significant preference for GDP-bound KRAS (K D =1.36 ± 0.87 nM for GDP and K D = 7.88 ± 1.09 nM for GppNHp, Student T-test p=0.0095). Thus Affimer K6 has a 10-fold higher affinity for KRAS than Abd-7 7 , the strongest-binding small molecule with a K D of 51 nM (c.f 750 nM for BI-2852 11 and 1.1 mM for DCAI 6 and 340 µM for compound 13 8 ).
We also inspected the SI/SII-binding small molecules for structural similarities to the K6 pharmacophore (Fig. 4e). All of the small molecules have an aromatic ring that inserts into the pocket and which is reproduced by the side chain of W43 of Affimer K6. However, only K6 and BI-2852 appear to interact across the whole of the pocket surface. This suggests that additional points of interaction may underlie efficacy, as BI-2852 is the most potent of the compounds to date [6][7][8]11 and Affimer K6 shows a similar degree of potency to BI-2852, both showing IC 50 values in the nanomolar range for inhibition of nucleotide exchange (IC 50 = 592 ± 271 nM for Affimer K6 and IC 50 = 490 nM for BI-2852 11 ). E37 acts as the propagating residue responsible for a hydrogen bonding/salt bridge network, terminating at the RAF-RBD residue R100 30 . It is possible that the positioning of KRAS residue E37 whilst interacting with S65 and R68 is now unable to interact with RAF-RBD and thus may explain the ability of K3 to inhibit the KRAS:RAF interaction as seen in the immunoprecipitation experiments. In addition, the conformational shift of SII upon K3 binding has orientated the M67 side chain of α3 helix towards the KRAS:RAF interface that could give rise to steric clashes with RAF-RBD residues such as R67 (Supplementary Fig 1d). Furthermore, as SII acts as a main anchor point for SOS1 binding, 31 the significantly reduced flexibility of this site may, together with occlusion of the Cdc25 domain of SOS1 forming a steric clash ( Supplementary Fig 1c), underlie the K3-mediated inhibition of SOS1-mediated nucleotide exchange. The intra-molecular hydrogen bonds between R68/E37, and Thus, Affimer K3 has identified a conformer of wild-type KRAS that generates a druggable pocket, with an estimated interface area of 790.6 Å (PISA (EMBL-EBI) 29 analysis), buried between the SII region and the α3 helix, not previously observed in wild-type KRAS. A similar pocket has previously been documented in the KRAS G12C mutant, together with a small molecule series that inhibits RAS function via binding at this pocket 9,10,12 (Fig. 5g). Of this series, the most recently published AMG510 compound has reached clinical development 12 . The compound is covalently tethered to the C12 residue and explores the same SII/α3 helix pocket.

K3 locks KRAS in a novel conformation to reveal a SII/α3 pocket
We hypothesise that AMG510 induces a similar mode of inhibition seen by K3, whereby the switch regions are stabilised by the ligand. However, it is clear that although the AMG510 compound and Affimer K3 explore the same cryptic groove, the conformations of SII lead to distinct pocket conformations with widely different electrostatics. Affimer K3 stabilises a more open conformation compared to the closed conformation seen with AMG510 (Fig. 5g). Further to this, binding of K3 to KRAS decreases flexibility by inducing hydrogen bond interactions between SI/SII, and SII/α3 helix, that are not present in the KRAS G12C :AMG510 structure (PDB code: 6OIM). The shape, size and physiochemical composition of the pocket identified by Affimer K3 suggests a potential druggable site 34,35 . The K3 data shows that we have isolated a non-covalent KRAS binder and have identified a druggable pocket and pharmacophore combination through which to inhibit KRAS preferentially over other RAS isoforms.

Discussion
We have isolated RAS-binding Affimer reagents that inhibit RAS both in vitro and in cells. The Affimer proteins generated show nanomolar affinities for KRAS together with IC 50 values in the nanomolar range for inhibition of SOS1-mediated nucleotide exchange. Furthermore, they are functional intracellularly demonstrating inhibition of the MAPK pathway as assessed by ERK phosphorylation levels. Structural analyses showed that the Affimer proteins interact with RAS within druggable pockets, notably identifying a novel pocket between the Switch II region and the α3 helix, with a non-covalent binder. Thus, we have exemplified a new site for the development of compounds to inhibit KRAS, together with a pharmacophore as a starting point for this approach.
The biochemical and cellular profiles of the Affimer proteins used in this study are comparable with biologics that have previously been identified that also inhibit RAS, again with nanomolar affinities and IC 50 values 14,15,[17][18][19][20]36 . However, the majority of these do not distinguish between RAS variants, and/or mutants, and structural analyses reveal that these pan-RAS inhibitors are binding in the Switch I/II region, except for the NS1 monobody that binds the α4-β6-α5 dimerization domain 18 , and the DARPins K13 and K19 14 ( Supplementary Fig. 2) as discussed below. The binding positions of the scFV, iDab6 and the DARPins K27 and K55 all span the SI/SII pocket 15,20 , which is the location of Affimer K6 binding; however, none of these other biologics have been shown to protrude into the pocket. Indeed, structural analysis of the K6:KRAS complex showed that a tripeptide motif, P42, W43 and F44, inserts into this SI/SII pocket in a manner that mimics the binding of known small molecules targeting this pocket. The aromatic indole ring of W43 extends into the pocket, a motif seen with all the compounds targeting this site. The interactions of the compounds are then diverse compared to Affimer K6 (Fig. 4e). It would be interesting to determine if compounds based on the K6 pharmacophore were more potent than the current compounds, as K6 binds with a higher affinity than any published reagent and shows comparable inhibition [5][6][7][8]11 .
Thus, Affimer K6 demonstrates that the use of biologics with small interaction interfaces can not only bind and inhibit difficult-to-drug proteins, but also identify and probe druggable pockets on such proteins, potentially acting as templates for small molecules. Importantly, the K3 Affimer shows inhibition of RAS, but also demonstrates a preference for KRAS over the HRAS and NRAS variants. To our knowledge, the only other biologics to express such RAS variant specificity are DARPins K13 and K19 14 . This preferential behaviour is underpinned by the involvement of the H95 residue unique to KRAS; mutation of this residue abolished binding of both Affimer K3 and the DARPins K13 and K19 14 . This ablation was more complete with mutation to glutamine for the DARPins, but was still significant with Affimer K3. These differences may in part be due to the distinct binding locations of the DARPins K13 and K19 on the allosteric lobe side of H95, whereas K3 binds on the effector lobe side and locks KRAS in a conformation where a pocket is revealed 14 . The residues involved in this pocket, specifically Q61, underlie the mutational preferences of Affimer K3 for wild-type/G12 mutants vs. Q61; this selectivity is not seen with DARPins K13 and K19 14 .
The pocket revealed by Affimer K3 binding is a previously unseen conformer of the SII pocket 3,4,9,10,12,13 , which we have termed the SII/α3 pocket, and it coincides with a cryptic groove identified computationally 4,12 . The SII pocket has previously been targeted by the covalently-tethered KRAS G12C inhibitors, the ARS series, and the most recent iterations that are in clinical trials 9,10,12,13 demonstrating the clinical importance of this pocket. Whilst this compound series has yielded the most clinically promising RAS-inhibitors to date, its dependence on covalent tethering to C12 restricts its utility to KRAS G12C mutant cancers only. Affimer K3 is, to our knowledge, the only RAS inhibitor to bind to a SII-derived pocket non-covalently. K3 demonstrates a similar degree of in vitro potency to AMG510 with IC 50 values for nucleotide exchange of 0.15 µM and 0.09 µM, respectively 12 , suggesting that targeting this region is a good approach for allosteric inhibition of RAS.
As Gentile et al 37 noted, for binding to the SII pocket, a substituted phenolic ring is required for insertion within the subpocket formed by V9, R68, D69 and M72, and to form hydrogen bonds with R68 and D69 residues. Affimer K3 fulfils these criteria with the aromatic ring of W44 extending into this subpocket and its surrounding residues, S40 and D42 forming the necessary hydrogen bonds, thus the SII/α3 pocket shares a key subpocket with the SII pocket. Indeed, K3 also forms hydrogen bonds with H95 and Q99 in common with AMG510, albeit with different orientations of H95.
Nevertheless, the positioning of the α2 helix of SII is significantly different with K3 inducing a conformation where the α2 helix distal to the α3 helix, and AMG510 a more closed conformation where the α2 helix is semi-distal to the α3 helix, but the loop region of SII is held across the pocket due to hydrogen bonding between AMG510 and KRAS E63. Whilst ARS-1620, the most potent ARS compound, leaves the helices proximal to one another as seen in WT-KRAS GDP (PDB: 4OBE), ( Fig.5g) 10,12 . These differences suggest that there may be an extended pocket area for small molecules based on the K3 pharmacophore, as identified by mutational analysis that is targetable to achieve the first non-covalent small molecule inhibitors of KRAS via the SII/α3 pocket.
Our work presented here demonstrates a novel concept of using biologics that bind with a relatively small interface as precursors for the development of small molecule inhibitors for difficult-to-drug proteins. The Affimer proteins identified in this study inhibited RAS, by binding to shallow pockets previously identified, or pockets derived from those previously identified, with comparable affinities and in vitro efficacies to the best small molecules available that target these pockets. This highlights the ability of Affimer proteins to select novel conformers of target proteins to reveal druggable regions on protein surfaces concurrent with pharmacophore identification.
Indeed, it will be interesting to use the pharmacophore motifs identified in this study as templates for novel series of RAS-binding small molecules and for potential hit-tolead optimisation using an Affimer-target NanoBRET system, as has been previously achieved with RAS-binding biologics 5,7 . The approach utilised in this study is likely to be applicable to other important therapeutic targets, and provides a novel pipeline for drug discovery.

RAS Interaction Assays
The interaction of KRAS with RAF-RBD or Affimer K3 was assessed by immunoprecipitation using a KingFisher Flex (ThermoFisher Scientific). Glutathione magnetic agarose beads were blocked overnight at 4ºC with 2x blocking buffer

Construction of K6ΔVR2 mutant
To generate Affimer K6ΔVR2 mutant, the 9 residues of the K6 VR2 were replaced with AAE. Affimer K6 VR1 and control Affimer VR2 (AAE) were amplified and subjected to splice overlap extension (SOE) PCR. The spliced product was subcloned into pET11a and Affimer K6ΔVR2 produced as previously described 24 .

Statistical Analysis
Data were analyzed in Prism v8.1.0 (GraphPad Software). Normality was tested using Shapiro Wilk test. Data presented are mean ± SEM unless otherwise stated.

Data Availability Statement
The Affimer constructs generated during the current study are available under a standard MTA from the University of Leeds via the corresponding author (DCT). The X-Ray crystal structures generated during and analysed during the current study are available in the PDB repository (https://www.rcsb.org).   (ARS1620 is shown in yellow and AMG510 is shown in magenta). Data is mean ± SEM, n=3 independent experiments, One-way ANOVA with Dunnett's post hoc test