Structure-based development of new RAS-effector inhibitors from a combination of active and inactive RAS-binding compounds

Significance The RAS family of oncogenic proteins is important as therapy targets because of the frequency of activating mutations in almost all major cancers. An important approach is development of small molecules with drug-like properties that can inhibit RAS-effector protein interactions inside cells. We present a strategy for identification of such compounds, and their development as RAS-effector interaction inhibitors, utilizing a structure-based design approach and cell-based assays. By combining moieties from two distinct sets of RAS-binding molecules, we generated cross-over compounds that showed improved efficacy in vitro and in cell-based assays.

The RAS gene family is frequently mutated in human cancers, and the quest for compounds that bind to mutant RAS remains a major goal, as it also does for inhibitors of protein-protein interactions. We have refined crystallization conditions for KRAS 169 Q61H -yielding crystals suitable for soaking with compounds and exploited this to assess new RAS-binding compounds selected by screening a protein-protein interaction-focused compound library using surface plasmon resonance. Two compounds, referred to as PPIN-1 and PPIN-2, with related structures from 30 initial RAS binders showed binding to a pocket where compounds had been previously developed, including RAS effector protein-protein interaction inhibitors selected using an intracellular antibody fragment (called Abd compounds). Unlike the Abd series of RAS binders, PPIN-1 and PPIN-2 compounds were not competed by the inhibitory anti-RAS intracellular antibody fragment and did not show any RAS-effector inhibition properties. By fusing the common, anchoring part from the two new compounds with the inhibitory substituents of the Abd series, we have created a set of compounds that inhibit RAS-effector interactions with increased potency. These fused compounds add to the growing catalog of RAS protein-protein inhibitors and show that building a chemical series by crossing over two chemical series is a strategy to create RAS-binding small molecules. cancer | RAS | drugs | antibody | intracellular antibody T he oncogenic family of RAS genes is of significant interest in the fight against cancer because of the frequency of activating mutations (1). Their presence in almost all major cancers makes them a highly valued therapeutic target, in particular the KRAS gene, since it has been identified as one of the most frequently mutated oncogenes (2,3). RAS proteins are linked to the plasma membrane by COOH-terminal prenylation mediated by farnesyl transferases (4). All family members function by signal transduction to the nucleus of cells via interaction with effectors (such as RAF, RALGDS, and PI3K) that catalyze phosphorylation of downstream proteins (5). When KRAS is bound to GDP, the protein is in the inactive state and becomes activated by nucleotide exchange from GDP to GTP. Normally, the activation/deactivation cycle is catalyzed by guanine nucleotide exchange factors and GTPase-activating proteins (GAPs) (6,7). Mutant RAS proteins remain in the active state and hydrolyze GTP at a much slower rate than wild-type (WT) RAS (8). Mutations reduce GAP activity leading to constitutive activation of RAS effector pathways (2), constantly generating a signaling cascade that activates cell functions such as division, survival, and invasion (9).
Despite its great potential as a cancer target, KRAS has proved to be very difficult to inhibit in a therapeutic setting. KRAS signaling works via protein-protein interactions (PPI) that can be very difficult to disrupt (10). In addition, the nucleotides that regulate KRAS function (GTP and GDP) bind to the protein with picomolar affinity, making them problematic to displace (11). Attempts at targeting RAS function using farnesyl transferase inhibitors also proved to be ineffective, failing to demonstrate antitumor activity in KRAS-driven cancers (12). As an alternative to compounds, various macromolecules [called macrodrugs (13)] have been developed that can bind to RAS and prevent PPI with the RAS effectors, such as has been shown with intracellular antibody fragments (14,15). The possible clinical use of these macrodrugs has not been implemented thus far due to difficulties in their delivery into cells, although methods are becoming available that may solve this problem (16).
Although there are a large number of mutant RAS protein isoforms, their structural conformation is highly conserved (17) because of the invariant N-terminal domain up to amino acid 166. The interest in inhibition of RAS proteins by small molecules has increased again recently (18), and several compounds have been described that bind to RAS (19)(20)(21)(22)(23)(24)(25)(26)(27). Recently, we have defined a chemical series based on an intracellular antibodybinding domain (28) that interact with a hydrophobic pocket (designated pocket I, SI Appendix, Fig. S1A), previously identified in silico (29) and confirmed as the binding site for 4,6-dichloro-2methyl-3-aminoethyl-indole (DCAI) near the switch I region of KRAS (23).

Significance
The RAS family of oncogenic proteins is important as therapy targets because of the frequency of activating mutations in almost all major cancers. An important approach is development of small molecules with drug-like properties that can inhibit RAS-effector protein interactions inside cells. We present a strategy for identification of such compounds, and their development as RAS-effector interaction inhibitors, utilizing a structure-based design approach and cell-based assays. By combining moieties from two distinct sets of RAS-binding molecules, we generated cross-over compounds that showed improved efficacy in vitro and in cell-based assays.
A critical step in drug development programs for progressing small molecules is the use of X-ray crystallography with compounds after crystal soaking or cocrystallization to identify where such molecules bind to the target protein. We have optimized KRAS 169 Q61H crystallization and applied crystal soaking to assess a set of RASbinding compounds selected from an initial diverse PPI-net l compound library (kindly provided by Andrew Wilson, University of Leeds, Leeds, UK), of which two bind in pocket I. However, unlike our previous Abd compounds, their binding was not impaired by binding of an inhibitory anti-RAS intracellular antibody fragment nor did they interfere with RAS-effector interactions. Comparison of the structures of these two PPI-net RAS-binding compounds and the lead compound Abd-7 allowed us to synthesize chimeric cross-over compounds that bind to RAS with improved potency and inhibit RAS-effector interactions whereas the PPI-net did not.

Results
Crystallography Conditions for KRAS 169 Q61H GppNHp, Suitable for Crystal Soaking. Initially, we wanted to establish crystallization conditions to obtain a crystal form that would allow free movement of RAS-binding compounds through the lattice for crystal soaking. We reproduced the KRAS 169 Q61H crystal structure found in the database (PDB ID code 3GFT) but with optimized crystallization conditions, using sparse matrix crystallization screening with a protein spanning residues 1-169. This produced diffraction quality crystals that showed similar packing to PBD ID code 3GFT with six chains in the asymmetric unit ( Fig. 1A and SI Appendix, Table S1). All six crystallographically independent chains (identified as chains A-F) have the same fold. Noncrystallographic symmetry averaging of the electron density maps allowed the assignment of all of the polypeptide backbone for switch I (in all six chains). Switch II is more flexible in this crystal form and a complete model could only be built in three or four copies of RAS per asymmetric unit. Further evidence for the flexibility of switch II is that chain A adopted a different conformation in this region owing to interactions with a symmetryrelated molecule (Fig. 1B). The six chains are less sterically hindered than in other published RAS crystal structures, such as KRAS 188 G13D (21) PDB ID code 4DST. The solvent channels are also much larger, facilitating compound diffusion; consequently, bound compounds have more freedom to be accommodated within KRAS 169 Q61H crystals, and the structures are more likely to represent the interaction in solution. In addition, six independent protein chains are available for binding of compounds within the asymmetric unit, making this crystal form particularly suitable for compound-soaking experiments.
As a comparison with the KRAS 169 Q61H crystals, we also determined crystal structures of WT KRAS 188 , mutant KRAS 188 G12V (SI Appendix, Fig. S2 and Table S2), and KRAS 188 G12D (all isoform 4B) (SI Appendix, Fig. S3 and Table S2) with a GTP analog (GppNHp), using crystallization conditions similar to those described previously for full-length KRAS 188 G12D (23). When the crystal packings of KRAS 169 Q61H and KRAS 188 G12D mutant proteins are compared, the switch regions are more solvent-accessible and less sterically hindered for the Q61H crystal. A further comparison was carried out between the switch regions of the two fulllength KRAS structures (G12V and G12D) with our new KRAS 169 Q61H structure (amino acids 1-169, Fig. 1 C and D). Both KRAS 169 Q61H and KRAS 188 G12V lack stabilization in the switch II region, and this instability could be attributed to the lack of a Mg ion binding to the switch II. All of the comparisons and observations between different crystal forms and mutants led us to conclude that the KRAS 169 Q61H crystals were the best option for crystal-soaking experiments.

G12V
. We previously identified a compound series that binds to RAS in pocket I using a highaffinity anti-RAS intracellular antibody fragment in competition surface plasmon resonance (SPR) (28). In the present paper, we used direct screening of KRAS with a compound library using SPR. The library used was triaged for possible PPI inhibitors and comprised 1,534 compounds (the PP1-net screening collection). To identify specific KRAS binders, the library was simultaneously negatively screened against two control proteins, namely the LIMonly protein 2 (LMO2) and a fusion protein consisting of LMO2 bound by an intracellular antibody VH fragment (LMO2-VH fusion) (30). Responses were referenced by subtracting those measured against the control protein LMO2 from the responses measured against KRAS (R ref Fig. S4A) of which 7 showed RAS specificity. Four of these compounds were available in sufficient quantities (with identities confirmed by mass spectroscopy) to allow waterLOGSY NMR to be carried out showing that PPIN-1 and PPIN-2 (the chemical structures are shown in Fig. 2 A and B) have good interaction properties with KRAS 166 G12V -GppNHp (SI Appendix, Fig. S4 A-C). Furthermore, the two PPIN compounds still bind to KRAS in waterLOGSY experiments in the presence of the RASinhibitory single-chain variable region antibody fragment (scFv) as a competitor. Thus, in this orthogonal assay, PPIN-1 and -2 were confirmed to bind to KRAS 166 G12V , but neither compound was prevented from binding to KRAS by the anti-RAS scFv intracellular antibody fragment as shown using Bioluminescence Resonance Energy Transfer (BRET)-based RAS biosensors (31) A B to assess intracellular RAS-protein interactions. As predicted from the waterLOGSY data using the intracellular antibody fragment, the PPIN compounds did not disrupt the interaction of KRAS 166 G12D with the anti-RAS iDAb (VHY6) with a dematured version of the iDAb (VHY6dm) or with full-length CRAF (CRAF FL ) (SI Appendix, Fig. S4D). This discrepancy between in vitro affinity and in cell potency could be attributed to targets with high conformational variability (like RAS switch regions) interacting with allosteric binders (32,33).
Crystal Soaking of KRAS 169 Q61H with PPIN-1 and -2. We determined the RAS-binding sites of PPIN-1 and PPIN-2 by soaking KRAS 169 Q61H crystals. Crystals of KRAS soaked with PPIN-1 diffracted to 1.63 Å (Fig. 2 C and E and SI Appendix, Table S3). Good electron density was observed for the ligand in one chain of the asymmetric unit (chain A, in which the switch regions are stabilized due to additional interactions on the opposite face with residues Arg102 and Lys101 from chain B in a neighboring asymmetric unit) and located adjacent to the C-terminal end of the switch I and switch II regions in pocket I. PPIN-1 primarily contacts KRAS 169 Q61H with the biphenyl head group via van der Waals interactions. No hydrogen bonds are formed with the protein. Crystals soaked with PPIN-2 diffracted to 1.7 Å. For this compound, clear electron density corresponding to the methoxybiphenyl anchor group was observed in four of the six chains ( Fig.  2 D and F). Weaker electron density was observed around the linker and tail groups, suggesting that the rest of the compound remains flexible when bound to the protein. It was noteworthy that both PPIN compounds have the same biphenyl anchoring group but different tail functional groups, suggesting that the biphenyltype groups of PPIN-1 and -2 are key in targeting these compounds to the pocket I-binding site.

Design and Characterization of RAS-Binding Cross-Over Compounds.
Our data show that PPIN-1 and PPIN-2 bind to KRAS at the same pocket I as several previously identified compounds (23), but they do not disrupt RAS function. To understand the lack of RAS inhibition of the PPIN compounds, we selected one of our RAS-binding intracellular antibody-derived compounds (Abd-7) able to interfere with RAS PPI in cells (28). We used the computational chemistry suite FORGE (https://www.cresset-group. com/forge/; ref. 34) that employs a ligand comparison method to align and score molecules independently using their shape and electrostatic properties, aiding the understanding of structureactivity relationships. FORGE was used to compare the structures of PPIN-1 and -2 with Abd-7 by performing an alignment based on the surfaces of the compounds only and detected similarities in the lower half of the molecules (Fig. 3 A and B). The aromatic ring of the benzodioxane moiety in Abd-7 aligned with the terminal biphenyl aryl ring in both PPIN compounds, and the pyridine ring of Abd-7 aligned with the middle aromatic ring of both PPIN compounds. This suggested that the PPIN biphenyl system might act as the anchor to the RAS protein. The alignment with the functional groups on the upper half of the molecules was poor. We used these analyses to design of cross-over compounds linking the common diphenyl-anchoring moiety of PPIN-1 and PPIN-2 to the Abd-7 aniline fragment. Three cross-over compounds (Fig. 3C: Ch-1, Ch-2, and Ch-3) were synthesized and their binding geometries were determined by X-ray crystallography using KRAS 169 Q61H crystal soaking (Fig. 3 D and E and SI Appendix, Table S4). All three compounds showed very similar binding modes to Abd-7 with van der Waals contacts to K5, L6, V7, S39, Y40, R41, D54, I55, L56, G70, Y71, T74, and G75. Ch-1 was found in four of the six KRAS 169 Q61H chains (A, B, C, and F); Ch-2 was again found in three of the six chains (A, B, and C) and Ch-3 was found in four of the six chains (A, B, C, and D). Thus, by combining the anchor constituent of the PPIN compounds with the aniline fragment of Abd-7, we generated a compound series showing good electron density for the entire molecule when bound to KRAS pocket 1.
The RAS-Binding Cross-Over Compounds Are PPI Inhibitors in Cells.
The purpose of the compounds was to generate inhibitors of RAS PPI. The ability of the compounds to interfere with RAS-associated PPI was analyzed with our BRET-based RAS biosensor toolbox (31). The interaction of full-length KRAS G12D and either the anti-RAS iDAb VHY6, a dematured version of this iDAb (VHY6dm), or the natural RAS partner CRAF FL was assessed as a dose-response with the three compounds (respectively, Fig. 4 A-C). While the interaction of the high-affinity WT iDAb with RAS was minimally affected, even at the highest dose of compound (i.e., 20 μM), both the lower-affinity-dematured iDAb and CRAF binding to RAS were progressively impaired, starting at the lowest dose of 5 μM. No alteration in the BRET signal for the PPI of a nonrelevant protein pair (LMO2-VH576dm) was observed (Fig. 4D), confirming that the dose-response effects of the Ch compounds was not due to loss of cell viability.
These BRET data show that the three cross-over compounds can enter cells and reach their target protein (RAS) in the cytoplasmic environment. The BRET assay relies on cotransfected donor and acceptor expression plasmids, and we confirmed this inhibitory capability by testing the effect of the compounds on biomarker phosphorylation in DLD-1 colorectal cancer cells. The cells were incubated with our previously described antibodyderived compounds Abd-2 and Abd-7, the original PPIN-2 compound, or the three cross-over compounds Ch-1, -2, and -3. Phosphorylation of AKT (downstream of RAS-PI3K signaling) or phosphorylation of ERK (downstream of RAS-RAF signaling) was determined following EGF stimulation (Fig. 5 A-C). The PPIN-2 (and control Abd-2) had no effect on the levels of phospho-AKT or phospho-ERK even at 20 μM (addressed using Western blotting, Fig. 5 A-C), but we found that the three cross-over compounds caused loss of phospho-AKT or phospho-ERK, as did the previously described compound Abd-7. The most potent compound appears to be Ch-3, which invokes an almost complete reduction of phospho-AKT at a concentration of 10 μM while not affecting AKT protein levels (Fig. 5C).
The biomarker Western blotting assay was carried out 2.5 h after addition of the compounds, at which time no loss of viability was observed. The survival of DLD-1 cells was determined over 48-and 72-h periods using a dose-response (0-20 μM) ( Fig. 5D; data shown at 72 h), allowing a calculation of IC 50 for each compound (SI Appendix, Table S5). The previously characterized low-affinity Abd-2 compound does not affect DLD-1 viability over the range of concentrations nor does the PPIN-2 compound. Conversely, the intracellular antibody-derived compound Abd-7 causes loss of viability with IC 50 of 10.8 μM at 48 h and 8.2 μM at 72 h. The potency of the two of the cross-over compounds (Ch-1 and Ch-3) is improved, relative to Abd-7, as these show IC 50 at 72 h of 5.3 and 4.5 μM, respectively (SI Appendix, Table S5). This increased efficacy of Ch-3 in the challenge of DLD-1 viability matches the most efficacious compound in the signaling biomarker assay.

Discussion
The RAS family of genes is among the most frequently mutated in human cancer (e.g., up to 96% in pancreatic cancer) and therefore an important target for drug development. Targeting the RAS-effector PPI is one possible route to RAS inhibitors. Screening chemical compound libraries per se does not guarantee selection of compounds that will act as PPI inhibitors unless there is a method to guide the screen to functionally specific locations. Strategies are needed that will allow protein-binding compounds Statistical analyses were performed using a one-way ANOVA followed by Dunnett's posttests (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). Error bars correspond to mean values ± SD of biological repeats. RLuc8-KRAS, NRAS, and HRAS all comprised full-length RAS components.
to be selected and improved in their properties related to the functional mechanism to be disrupted. In particular, X-ray crystallography of RAS proteins is a key method to determine the location and geometry of bound compounds. We optimized production of KRAS 169 Q61H crystals for soaking of compounds for this purpose. It should be noted that these conditions, in principle, could be used to produce crystals of other RAS mutants suitable for compound-soaking experiments. Surface analysis of KRAS (using the CASTp server: sts.bioe.uic.edu/castp/) confirmed three pockets with internal volumes greater than 80 Å 3 (SI Appendix, Fig. S1A: pockets I, III, and IV) and a less deep, more like a shallow groove, pocket (SI Appendix, Fig. S1A: pocket II). Pocket I was previously identified in silico (29) and was also the site where DCAI was first selected (23). In addition, this pocket is close to the switch region and could act as a point of inhibition for PPI. Furthermore, we have focused on the identification and optimization of compounds inhibiting RAS-effector interactions and not other RAS modulation effects, such as DCAI has shown. Structural comparison of the six chains in the asymmetric unit show that compound binding has a minimal effect on the conformation of the switch regions in RAS protein. Any observed differences are no more than twice the coordinate error (35) and therefore cannot be considered significant. Comparison of the temperature factors between RAS molecules with compound bound and 3GFT (where no compound is present) show no evidence that compound binding results in significant reduction in flexibility of switch II relative to the rest of the molecule, suggesting that any stabilization of the switch regions by compound binding is, at most, slight. Therefore, we conclude that any inhibitory effect observed is due to a disruption of RAS-effector PPI rather than any other modulation of the RAS protein.
We have screened a chemical library that yielded two RASbinding compounds (the PPINs) and have shown, by crystallography, that they bind to pocket I near to the effector binding sites. However, when these were tested in an orthogonal NMR waterLOGSY binding assay, their binding to KRAS 166 G12V was not impaired by the presence of the anti-RAS intracellular antibody fragment, nor did they interfere with PPI using a cellbased BRET assay (SI Appendix, Fig. S4). Crystallography and medicinal chemistry was undertaken to improve the initial PPIN hits using structure-based design combining the crystal information and molecular fragments from two different chemical series, namely PPIN and the Abd series (28). The resulting crossover compounds illustrate that this strategy guided the conversion of the PPIN RAS-binding compounds to RAS PPI inhibitors in a series (compounds designated Ch-1-3). These compounds have a low molecular weight suitable for further medicinal chemistry to improve drug-like properties and with better ligand efficiencies than their progenitors (SI Appendix, Table S5). Furthermore, the compounds have also shown better cell viability results than their progenitors. Their low molecular weight makes them a better starting point for the development of RAS inhibitors based on this promising chemical series.
We have previously shown that the Abd chemical series affected RAF, RAL, and PI3K interactions with RAS (28) and, due to the similarities in binding mode and orientation of the Ch series with the Abd series, we expect the Ch compounds to have a similar range of profiles with other KRAS mutants and also with NRAS and HRAS isoforms. This was confirmed using various KRAS G12 mutants and the NRAS Q61H and HRAS G12V mutants with four effector molecules in BRET assays.
This approach shows that compounds binding in pocket I are not necessarily able to inhibit RAS PPI, but synthetically linking components of two classes of RAS-binding compounds can generate new active molecules that inhibit PPI. It should also be noted that the presence of pocket I in nonmutated forms of RAS as well as in mutant RAS (SI Appendix, Fig. S1 B-G) presents a technical challenge for the development of anti-RAS drugs since these will bind to the pocket in WT-activated RAS as well as in mutant RAS. Development of methodologies for specific drug delivery could avoid drug interference with nonmutated RAS. Approaches such as Antibody-Drug Conjugates (ADC) offer one route to avoid drugs entering normal cells (reviewed in ref. 36) by targeting antigens expressed on tumors. While this, in turn, has difficulties, since few surface antigens are tumor-specific, surfaceome studies of tumors (37, 38) can find possible markers or pairs of markers that may be useful for monoor bispecific ADCs. An alternative approach, based on structureactivity relationships, could be the development of compounds anchoring at pocket I and moving toward the nucleotide-binding region of RAS or perhaps linking the "unselective" but potent compounds identified in pocket I to those binding in pocket II.
In conclusion, our approach demonstrates the importance in drug development of combining assays for PPI with the identification of compounds that bind at important locations but not necessarily with PPI properties. Thus, compounds binding with good potency to the target but not showing any effect on PPI could be utilized in combination chemistry to create new chemical series. A combination of high-resolution crystallography from different chemical series with biophysical competition assays, such as using high-affinity antibody fragments, is thus a powerful way to identify hit compounds of interest in analogous settings. It is also useful in the development of new chemical series, even when initial compounds are inactive, and should allow directed medicinal chemistry for drug development.