Genomic Discovery and Structure-Activity Exploration of a Novel Family of Enzyme-Activated Covalent Cyclin-Dependent Kinase Inhibitors

Abstract

Fungi have historically been the source of numerous important medicinal compounds, but full exploitation of their genetic potential for drug development has been hampered in traditional discovery paradigms. Here we describe a radically different approach, top-down drug discovery (TD³), starting with a massive digital search through a database of over 100,000 fully genomicized fungi to identify loci encoding molecules with a predetermined human target. We exemplify TD³ by the selection of cyclin-dependent kinases (CDKs) as targets and the discovery of two molecules, 1 and 2, which inhibit therapeutically important human CDKs. 1 and 2 exhibit a remarkable mechanism, forming a site-selective covalent bond to the CDK active site Lys. We explored the structure-activity relationship via semi- and total synthesis, generating an analog, 43, with improved kinase selectivity, bioavailability, and efficacy. This work highlights the power of TD³ to identify mechanistically and structurally novel molecules for the development of new medicines.

Figure 1

Gene content and taxonomic distribution of a representative ros BGC. (A) The ros BGC has 11 genes, including a polyketide synthase core (rosJ) and a putative ETaG homologous to human CDKs (rosG). (B) Shown are 4 of the 70 distinct alleles of the ros BGC discovered in diverse taxonomic classes (indicated on the left) within LifeBase. Though the order and orientation of genes differ, the consensus genes are maintained, including the ETaG (red), consistent with functional essentiality.

Figure 2

Phylogenetic comparison of the candidate ros BGC ETaG protein (red) from P. flanaganii with yeast and human CDK protein isoforms. Note that the ETaG (red) is a second copy of and is most closely related to the corresponding housekeeping Pho85 (blue).

Figure 3

Residue conservation analysis identified three putative resistance mutations in the active site of the CDK ETaG (Table 1), suggesting that the product of the ros BGC is an active site, ATP-competitive inhibitor of CDKs. Mutations are mapped onto the structure of the yeast Pho85-Pho80 CDK-Cyclin complex (PDB ID 2PMI).

Figure 4

Heatmap of the sum of LC-MS/MS response from targeted detection of 1 and 2 (green) and the average expression of ros cluster genes (purple), as determined by quantitative RNAseq, in fermentations of LifeBase strains containing the ros BGC. Plotted values are the maximum observed in fermentations on 7 different media conditions at days 7 and 14. The heatmap is organized by multilocus phylogenetic analysis of 100 fungal housekeeping proteins.

Scheme 1. Biosynthetic Pathway of 1 and 2

(A) Biosynthetic pathway of 1 and 2, showing nonenzymatic degradation pathway of 2 to 9 by hydrolysis and spontaneous decarboxylation at C11a to form 10 (gray arrows). (B) Heatmap of LC-MS/MS response for biosynthetic intermediates of 2 accumulated in single-gene KO strains of Penicillium sp.

Figure 5

Kinase activity assays utilizing the CDK2/cyclin E kinase complex to characterize covalent binding interactions with 2. (A) Enzyme inhibition by 2 exhibited time-dependence of the IC₅₀, here plotted in relative units (RU). (B) CDK2/cyclin E activity as a function of preincubation time was fit to determine the k_obs rate constants which are plotted as a function of 2 concentration. Mean k_obs reported ± SEM. These data were fit to determine K_I (the apparent inhibitor affinity), k_inact (the maximal inactivation rate), and the ratio k_inact/K_I (potency).

Figure 6

(A) Co-crystal structure of CDK2/cyclin E in complex with 2 (PDB ID 9BJB). (B) Active site of CDK2 with bound 2, showing H-bonding interactions from C3-OH to the kinase hinge; electron density map of 2 overlaid in mesh (2fo-fc contoured at 1 sigma) demonstrates covalent attachment to Lys33. (C) Binding of amide analog XC208 51 (PDB ID 9BJC) without covalent interaction, demonstrating a prereaction complex. (D) 2D map of binding site interactions between CDK2 and 2, showing covalent (solid line) and H-bonding (dotted lines) positions. (E) Structure of CDK2/cyclin A with ATP bound showing similar interactions to 2 and 51 at the hinge and catalytic Lys33 (PDB ID 4EOQ).

Figure 7

(A) % Inhibition of pRbT821 dose response of 2 in OVCAR-3 cells. Potency of 2 with transient 4 h exposure was similar to constant treatment. Mean IC₅₀ reported ± SEM (B) % growth inhibition dose response measured by Cell Titer-Glo in OVCAR-3 cells. Potency of 2 with transient 4h exposure was similar to constant treatment. Mean IC₅₀ reported ± SEM.

Scheme 2

Scheme 3

Scheme 4

Scheme 5

Figure 8

(A) Cocrystal structure of CDK2/cyclin E with 2. (B) Cocrystal structure of CDK2/cyclin E with XC219 43 (PDB ID 9BJD) showing the same binding pose as 2 with rotation of Val64 to accommodate the larger R₂ = Cl substituent.

Figure 9

Kinome binding profile of 2 and analogs using the Kinobead assay. (A) Profile of GEM 2 (2 μM) compared to (B) chlorinated analog 43 (2 μM) with improved CDK2 selectivity; (C) acetylated analog 47 (2 μM) with alternate selectivity in the CAMK group; (D) analog 38 (50 μM) with alternate specificity in the CK1 group. Kinases with engagement >50% are labeled, kinase classes that show differentiation between analogs of 2 are highlighted.

Figure 10

Antitumor growth activity in a CCNE amplified (CN = 12) gastric PDX model CRT00292. NOG mice were inoculated subcutaneously with tumor fragments and randomized (n = 7/group) when tumors reached an average of 150–200 mm³. Animals were dosed daily with XC219 43 (30 and 100 mpk × 14d, i.p.). In the highest dose group 3 animals were moribund or withdrawn from treatment. One-way ANOVA, Dunnett’s multiple comparison test was performed reporting statistical significance of 100 mpk group, p value = 0.0079.