Discovery of potent small-molecule inhibitors of lipoprotein(a) formation

Abstract

Lipoprotein(a) (Lp(a)), an independent, causal cardiovascular risk factor, is a lipoprotein particle that is formed by the interaction of a low-density lipoprotein (LDL) particle and apolipoprotein(a) (apo(a))^1,2. Apo(a) first binds to lysine residues of apolipoprotein B-100 (apoB-100) on LDL through the Kringle IV (K_IV) 7 and 8 domains, before a disulfide bond forms between apo(a) and apoB-100 to create Lp(a) (refs. ^3-7). Here we show that the first step of Lp(a) formation can be inhibited through small-molecule interactions with apo(a) K_IV7-8. We identify compounds that bind to apo(a) K_IV7-8, and, through chemical optimization and further application of multivalency, we create compounds with subnanomolar potency that inhibit the formation of Lp(a). Oral doses of prototype compounds and a potent, multivalent disruptor, LY3473329 (muvalaplin), reduced the levels of Lp(a) in transgenic mice and in cynomolgus monkeys. Although multivalent molecules bind to the Kringle domains of rat plasminogen and reduce plasmin activity, species-selective differences in plasminogen sequences suggest that inhibitor molecules will reduce the levels of Lp(a), but not those of plasminogen, in humans. These data support the clinical development of LY3473329-which is already in phase 2 studies-as a potent and specific orally administered agent for reducing the levels of Lp(a).

Fig. 1. Identification of a small-molecule apo(a) binder and inhibitor of Lp(a) formation.

a, Chemical structure of LSN3353871. b, ITC of LSN3353871 interacting with the apo(a) domains K_IV8 (orange, one experiment), K_IV7–8 (blue, two experiments) and K_IV5–8 (violet, two experiments). DP, power differential; ΔH, enthalpy change. c, Average data for the inhibition of in vitro Lp(a) assembly by LSN3353871 (43 independent experiments; compound concentration 0.03–10 µM). d, Left, crystal structure of the K_IV8–LSN3353871 complex (Protein Data Bank (PDB) ID: 8TCE), with the amino acids involved in ligand binding labelled. Ligand hydrogen bonds to D54, E56, Y62 and R69 are shown as yellow rods. The weak hydrogen bond between the pyrrolidine ring and W70 is shown as a dotted line. Right, the ligand is also shown in the same orientation as a space-filling model in a ribbon structure of the whole Kringle domain with disulfide bonds labelled. e, Percentage change in the steady-state levels of Lp(a) in LPA × ApoB100 transgenic mice after five days of oral dosing. Data are from one experiment with n = 5 per group except n = 4 for the 10 mg kg⁻¹ group, and are shown as mean ± s.e.m. with individuals plotted as circles. Two-sided P values were calculated using one-way ANOVA with Dunnett’s comparison to vehicle. f, Percentage change from baseline in the steady-state levels of Lp(a) in female cynomolgus monkeys during 14 days of oral dosing (n = 8 per group except n = 9 for the vehicle group) in a single experiment. Intersecting lines between box plots indicate median values, box limits indicate the 25th and 75th percentile and whiskers indicate the smallest and largest value. Two-sided P values are from a repeated measures ANOVA with Bonferroni comparisons to vehicle at each time point. The null hypothesis was rejected at P < 0.05. BID, twice daily; QD, once daily; NS, not significant. Source Data

Fig. 2. Mechanistic interactions with apo(a) K_IV domains and discovery of multivalency.

a, Chemical structure of the dimeric molecule LSN3441732. b, Chemical structure of the optimized monovalent molecule LSN3374443. c, Average inhibition of in vitro Lp(a) assembly by LSN3353871 (triangles; 43 independent experiments), LSN3374443 (circles; 3 independent experiments) and LSN3441732 (squares; 8 independent experiments). d, Representative data from direct radioligand binding of ³H-LSN3374443 (circles; three experiments) and ³H-LSN3441732 (squares; one experiment) to full-length recombinant 16K apo(a). e, Representative data from ITC of LSN3374443 (circles; two experiments) and LSN3441732 (squares; three experiments) with apo(a) K_IV5–8. f, Plot of direct binding of ³H-LSN3441732 to K_IV7 (black; two experiments), K_IV8 (orange; two experiments), K_IV7–8 (blue; two experiments) and K_IV5–8 (violet; one experiment); data are representative.

Fig. 3. Multivalency increases the potency and efficacy of Lp(a) reduction.

a, Percentage change in the steady-state levels of Lp(a) in LPA × ApoB100 transgenic mice after five days of BID oral dosing with LSN3441732. Data are from one experiment with n = 5 per group, shown as mean ± s.e.m. with individuals plotted as circles. Two-sided P values were calculated using one-way ANOVA with Dunnett’s comparison to vehicle. b, Percentage change from baseline in the steady-state levels of Lp(a) in female cynomolgus monkeys during 14 days of BID oral dosing (n = 8 per group except n = 9 for the vehicle group) with LSN3441732 in a single experiment. Box plots as in Fig. 1f. Two-sided P values are from a repeated measures ANOVA with Bonferroni comparisons to vehicle at each time point. The null hypothesis was rejected at P < 0.05. c, Chemical structure of LY3473329. d, Crystal structure of LY3473329, shown as a CPK representation, bound to three separate K_IV8 domains (PDB ID: 8V8Z). e, Average inhibition of in vitro Lp(a) assembly by LY3473329 (nine independent experiments; compound concentration 0.03–3 nM). f, Percentage change in the steady-state levels of Lp(a) in LPA × ApoB100 transgenic mice after five days of BID oral dosing with LY3473329. Data are from one experiment with n = 5 per group, shown as mean ± s.e.m. with individuals plotted as circles. Two-sided P values were calculated using one-way ANOVA with Dunnett’s comparison to vehicle. g, Percentage change from baseline in the steady-state levels of Lp(a) in male cynomolgus monkeys during 14 days of QD oral dosing (n = 6 per group) with LY3473329 in a single experiment. Box plots as in Fig. 1f. Two-sided P values are from a repeated measures ANOVA with Bonferroni comparisons to vehicle at each time point. The null hypothesis was rejected at P < 0.05. Source Data

Fig. 4. Selectivity against plasminogen.

a,b, Change in plasmin activity (blue) and plasminogen concentration (black) in rat plasma after four days of oral dosing with LSN3441732 (a) or LY3473329 (b). Data are from one experiment with n = 5 per group and individuals are plotted as circles. Two-sided P values are from an ANOVA with Dunnett’s comparisons to vehicle. The null hypothesis was rejected at P < 0.05. c, Primary sequence alignment of human apo(a) K_IV domains, rat and human plasminogen Kringle domains and the presence of the consensus binding sequence (DxE) for the binding of the small-molecule inhibitor. d, Representative saturation binding curves of ³H-LSN3441732 binding to rat (squares; seven experiments) or human (circles; four experiments) plasminogen. e, ITC of LSN3441732 binding to rat (left) or human (right) plasminogen. Data are representative of two independent experiments. f, Schematic model of human closed plasminogen, on the basis of the crystal structure of human type II plasminogen (PDB ID: 4DUR). Crucial interactions for maintaining the closed conformation are K-4 and K-5 with Lys and Arg from the plasminogen–apple–nematode (PAN) domain, and K-2 with Lys from the trypsin domain. Our hypothesis is that multivalent molecule interactions with the K-2 and K-3 subdomains of rat plasminogen promote a plasminogen open (active) conformation. Box plots show the median (centre line), 25th and 75th percentile (box limits) and the smallest and largest value (whiskers). LBS, lysine-binding site. Source Data

Extended Data Fig. 1. Schematic of apolipoprotein(a) domains.

a, The type IV Kringle domains of apo(a) are shown, ending with a single type V Kringle domain and C-terminal peptidase domain. b, Numbering of domains based on the Uniprot (P08519-1) and GenBank RefSeq (NP_005568.2) amino acid residue numbers are provided in the table, with the Kringle domains defined by their first and last cysteine residues that form a disulfide bond.

Extended Data Fig. 2. Co-crystal structures of apo(a) K_IV domains with small-molecule inhibitors.

a, Apo(a) K_IV8 in complex with LSN3353871 (PDB ID: 8TCE). Amino acids involved in ligand binding are labelled and the ligand hydrogen bonds to D54, E56, Y62 and R69 are shown as yellow rods. b, Apo(a) K_IV7 in complex with LSN3441732 (PDB ID: 8V9B). This bivalent compound binds two Kringle domains simultaneously while conserving all the interactions observed in monomeric binding. c, Apo(a) K_IV8 in complex with LY3473329 (PDB ID: 8V8Z), also conserving the same interactions.

Extended Data Fig. 3. Human and rat in vitro clot dissolution assay.

a, Percentage inhibition of clot dissolution mediated by LSN3353871 in human platelet-poor plasma (top) or rat citrate plasma (bottom) (compound concentration 0.03–100 µM) as compared to vehicle. b, Percentage inhibition of clot dissolution mediated by LSN3441732 in human platelet-poor plasma (top) or rat citrate plasma (bottom) (compound concentration 0.03–100 µM) as compared to vehicle. c, Percentage inhibition of clot dissolution mediated by LY3473329 in human platelet-poor plasma (compound concentration 0.03–100 µM) as compared to vehicle. d, Percentage inhibition of clot dissolution mediated by plasminogen inhibitor, tranexamic acid (compound concentration 0.1–1,000 µM).

Extended Data Fig. 4. Rat hepatic Plg mRNA after oral doses of LSN3441732.

Change in Plg mRNA in male rat liver after four days of oral dosing with LSN3441732. Data are from one experiment with n = 5 per group (circles) with mean ± s.e.m. shown. Plg mRNA was measured by RT-qPCR from liver of male Sprague Dawley rats dosed with varying oral doses of LSN3441732 for 4 days. Relative quantification was calculated using the comparative Ct method and Gapdh mRNA served as the normalizing control. Source Data

Extended Data Fig. 5. Analysis of Kringle domains in LPA and potential off-target proteins.

a, Multiple sequence alignment of Kringle domains from LPA, plasminogen (PLG)/angiostatin, plasminogen activator urokinase (PLAU), plasminogen activator tissue type (PLAT, tPA), and prothrombin (F2). Residue numbering is based on the type IV Kringle domains in LPA from the first to the last cysteine, which form a disulfide bond. Residues interacting with compounds published in this study are indicated with black triangles, and the key DxE residues are indicated with red triangles. Disulfides form between Cys1-Cys78, Cys22-Cys61, Cys50-Cys73. b, Detailed comparison of LPA Kringle domains with off-targets containing Kringle domains, highlighting variations that are likely to reduce binding affinity. The only off-target Kringle domain that conserves the DxE motif seen in LPA K_IV5/6/7/8 is that of the human plasminogen Kringle II domain. Yellow squares indicate potentially disruptive variations, and orange squares indicate the most disruptive variations. Asp54 is the only residue strictly conserved across all these human Kringle domains. c, LY3473329 is shown binding to one of three copies of LPA K_IV8; interacting residues are labelled. The protein ribbon is coloured blue to red from the N to the C terminus. Ser67 is also shown, because it is this residue in LPA K_IV9 (Cys67) that forms a disulfide bond with ApoB (in the absence of inhibitor).