KRAS Binders Hidden in Nature

Abstract Natural products have proven to be a rich source of molecular architectures for drugs. Here, an integrated approach to natural product screening is proposed, which uncovered eight new natural product scaffolds for KRAS—the most frequently mutated oncogenic driver in human cancers, which has remained thus far undrugged. The approach combines aspects of virtual screening, fragment‐based screening, structure‐activity relationships (SAR) by NMR, and structure‐based drug discovery to overcome the limitations in traditional natural product approaches. By using our approach, a new “snugness of fit” scoring function and the first crystal‐soaking system of the active form of KRASG12D, the protein–ligand X‐ray structures of a tricyclic indolopyrrole fungal alkaloid and an indoloisoquinolinone have been successfully elucidated. The natural product KRAS hits discovered provide fruitful ground for the optimization of highly potent natural‐product‐based inhibitors of the active form of oncogenic RAS. This integrated approach for screening natural products also holds promise for other “undruggable” targets.

Natural products have proven to be ar ich source of molecular architectures which bind to drug targets. [1] In contrastt os ynthetic small molecules, these secondary metabolites, which have evolvedt op rotect host organisms,t end to feature more stereogenic centers, contain rigid cores and multiple ring systems with low degrees of conformational entropy( compounds 1-6,F igure 1). [2] However,t he future role of natural productsi n drug discovery is constrained due to the limitations in isolation of pure natural products for screening, hit validation and hit characterization. [3] KRAS is the most frequently mutated oncogene and belongs to the protein family of small GTPasest hat functiona sb inary moleculars witchesi nvolved in cell signaling. [4] Although KRAS could serve as an excellent target for many cancers, no therapeutic agent directly targeting RAS has been clinicallya pproved. The main reason for this is the lack of druggable pockets on the surface of RAS. In recent years, the existence of two pockets on the surfaceo fK RAS has been discovered, whichc ould potentially be amenable to small-molecule drug discovery. [5] The shallow,p olar pocket situated be-  ( 1), the tetrahydroisoquinoline derived pentacycle morphine (2), the pentacyclic quinolone camptothecin(3), the tetracyclic indoleergonovine (4), the heptacyclic indoline strychnine (5) tween the switch Ia nd switch II region of KRAS (SI/II-pocket) is of particulari nterest because it is present in the active, oncogenic form of KRAS. However,m olecular scaffolds capable of binding to the activef orm of KRAS remains carce and are of weak affinity. [6] In this paper,w ed escribe an integrated approach to screening natural products in which virtuals creening of natural products has been combined with elements of fragment-based screening, [7] the use of SAR by NMR, [8] and structure-based drug discovery [9] to uncover natural-productbased KRAS binders.
Our previousa ttempts to identify starting points through high-throughput screening of 1.6 million compounds, including ad iversity-oriented synthesis library (150 000 compounds), failed to deliver any inhibitors of GTP-KRAS. This demonstrates that the chemical space covered by our historical compound collection was not sufficiently complementary to the pockets on KRAS in the affinity range tested by biochemical and cellular assays.N atural products, however,c ould represent an area of chemical space containing molecular architectures with sufficient complementarity to KRAS.
Considering that many naturalp roduct structures are not ideal for virtuals creening due to their large size and high degrees of substitution, ac hemoinformatics based PipelinePilot workflow, analogous to that used by other groups, [10] deconstructed all structures in the CRC Dictionary of NaturalP roducts (DNP) [11] (180 000 entries at the time of application). This produced al ibrary of 226 000 original and deconstructed natural products amenable to virtual screening, referred to subsequently as the virtual DNP (vDNP). At the time of the virtual screening campaign,o nly one X-ray co-crystal structure of a ligand boundt ot he active form of KRAS had been reported. [5b] Due to ambiguity in the electron density for the proposed binding mode in this structure, we decided insteadt ou se the published GDP-KRAS structure (PDB code 4EPV) with indole 7 bound as published by the Fesik group, [5a] as as urrogate for GTP-KRAS.
The vDNP was docked by using Glide into the SI/II-pocket of GDP-KRAS.T he compounds with the best 500 molecular-weight-normalized dockings cores [12] (Glide score/MW) were visually inspected, and compounds that mimicked the known indole occupation of the SI/II-pocket ( Figure 3A)a nd formed H-bonds with D54 were selected. Compoundsd eemed chemically unstable were removed. 3-Substituted indoles were also  not pursued further because they did not provide sufficient structural noveltya sastarting point versus 7 (Figure 2).
From the virtual screen of the vDNP,n atural product hits isolated from fungi, sponges, plants and bacteria were identified ( Figure 2). b-Carbolines and g-carbolines, which are relatively ubiquitous in nature,s ubstituted at the 1-, 2-and 3-positions were found. b-Carboline 8 (Figure2), similart oi ndole 7,f illed the lipophilic SI/II-pocket flanked by L56 and K5 and formed two H-bonds to D54:o ne throught he indole NH and as econd through the glycerols ubstituent ( Figure 3B). The tricyclici ndolopyrrole alkaloid 9a (Figure 2) first isolated from the fungus Fusariumi ncarnatum mimics the indole binding mode of 7 and 8,w hile formingasecond H-bond to D54 through the pyrrole nitrogen (Figure3C). The marine sponges Agelas longissima and Phakellia flabellate provided two KRAS virtualh its derived from bromopyrrole alkaloids, namely,t he debrominated analogue 10 b of longamide (10 a) ( Figure 2) and the tetracyclic guanidines caffold Phakellin 11 b,w hichc o-occurs with its dibromo analogue 11 a ((+ +)-dibromophakellin;F igure2). Compound 10 b forms one H-bond to D54 throughahydroxyl group (Figure3d), whereas 11 b forms two H-bonds to D54 throught he guanidine group (Figure 3e). Twov irtualh its were also uncovered from plant secondary metabolites, namely,d iffusarotenoid 12 a and vasicine 13,at ricyclic quinazoline ( Figure 2). The cis-fused tetrahydrochromenochromene ring system of the de-esterifiedd iffusarotenoid 12 b (Figure 2) complementedt he shape of the SI/II-pocket while formingt wo Hbonds to D54:one through the phenolic oxygen and the backbone carbonyl, and one between the 6-hydroxyl and the side chain (Figure3f). Compound 13 also occupies the SI/II-pocket and forms an H-bond with D54 through the 3-hydroxy group (Figure 3g). Deconstruction of the methyle ther of quinocarcinol 14 a, isolated from the bacteria Streptomyces melanovinaceus, led to the virtualh it 14 b (Figure 2) for which the iminoazepinoisoquinoline skeleton occupiest he SI/II-pocket and forms two H-bonds to the side chain of D54 with the phenolic and alcoholicoxygens (Figure 3h).
To validate the virtuals creen, 9b,aclose analogue of the tricyclic indolopyrrole alkaloid 9a wast ested for binding to GCP-KRAS by using NMR.C ompound 9b demonstrated cross peak shifts in the 2D 1 H/ 15 NHSQC NMR spectrao fGCP-KRAS G12D (Figure 4a), confirming that these virtualh its do indeed bind to the active form of KRAS. The dissociation constant (K D )f or 9b, as measured by NMR, was in the millimolar range with no saturation obtainedw itht he concentrations tested (up to 1.5 mm, Figure 4b). To be ablet op erform high throughput crystallization of very weakly binding fragments (K D > 1mm)w ith KRAS, such as those reported here, we developedarobust crystalsoakings ystem for active KRAS G12D .B yu sing GDP-KRAS G12D ,w e exchanged the nucleotide with the non-hydrolysable GTP analogue GMP-PCP (GCP) to yield an ovel crystal form amenable to high-throughputc rystallization, which mimics the beta sheet dimeric form of active KRASG12D [13] (Figure 4c). With this soakings ystem,w ew ere able to determine the co-crystal structureo f9b with GCP-KRAS at ar esolution of 1.2 (Figure 4c). The observed binding mode confirmed the docking prediction (Figure 4d), namely binding to the SI/II-pocket and the formation of two H-bonds to D54.
The docking pose of rotenoid 12 b displayed striking complementarity to the shape of the SI/II-pocket ( Figure S2 A, Supporting Information);h owever,s emiempiricalq uantum mechanicsc alculations (AM1, PM3, MNDO) and crystal structure analysiso fs tructurally related compounds in the Cambridge StructuralD atabase suggest that rotenoids exist in af lat conformation.T he complementary "kinked" docking pose of 12 b in the SI/II-pocket of KRAS encouraged us to generate as ub-library of 1383 rotenoid-like "kinked" molecules with contiguous fused rings from our corporate database for as econd virtual screening campaign.
To ensure that the shape complementarity between the rigid three-dimensional topology of our "kinked" library and the KRAS protein were assessed accurately,w ed eveloped a "snugness-of-fit" scoring function. The shapes of ap rotein and al igand were calculated as envelopes around their respective atoms and were represented as 3D grids. Envelope overlaps that occurred at small atomic radii corresponded to those areas where the protein and ligand were close, that is, "snug", whereas overlaps that only occurred at larger radii represented fewer snug areas. Addition of all envelope overlap grids (at different radii)p rovided aq uantitative measuref or the "snugness of fit" of the ligand at each point in space. The grids can be also used for visualizing the snugness of fit of ad ocked or crystallized compound using iso-contour surfaces or colorcoded surfaces ( Figure S2, Supporting Information). The "kinked" libraryw as virtually screened in an analogous procedure to that used for the vDNP,w ith the addition of the "snugnesso ff it" scoring functionf or pose post-processing ( Figure 5a). After visual inspection, 7c ompounds were selected for NMR testinga nd the racemic compound 15 showed NMR cross peak shifts. Both enantiomers 15 R and 15 S docked with conserved indole binding modes( Figure S3 A,B, Supporting Information);h ence,t he racemate 15 was separated into its two enantiomers 15 R (2R,3S)a nd 15 S (2S,3R). 15 R displayed a K D of 1mm ( Figure 5B,C), as determined by the 2D 1 H/ 15 NHSQC NMR spectra to GCP-KRAS G12D ,w hereas 15 S bound more weakly with ad issociation constant in the millimolar range (FigureS3C and Figure S3 D). Enantiomer 15 R was successfully soakedi nto the crystallography system ( Figure 5D) with the crystal structure confirming the docking pose (Figure 5E). 15 R binds to the SI/II-pocket of active KRAS and forms two chelating H-bonds to D54 with an NÀOd istance of 2.9 and 3.0 ,r espectively,a nd aw eaker third H-bond to the K9 side chain throught he carbonyl group with an OÀNd istance of 3.4 .
In conclusion, from the two virtual screensc onducted, virtual hits stemming from seven classes of natural products from fungi, sponges, plants, bacteria and one natural-product-like class from our corporate compound collection were identified. Of these, we confirmed biophysical binding to the active form of KRAS G12D for the tricyclic indolopyrrole alkaloid 9b and the indoloisoquinolinones 15 R and 15 S.I mportantly,w ee lucidated the binding mode by determining the protein-ligand complex structures of 9b and 15 R with the first X-ray crystal-soaking system for the active form of KRAS G12D .T he wealth of natural product KRAS hits, discovered by using this integrated approach of natural product screening, should act as inspiration for the design of more potent and selectiveR AS inhibitors. This approach also promises to be applicable to other "undruggable" targets beyondK RAS,f or which classical screening methods have proven to be unsuccessful.