Tc-43/Lipoprotein-a-Binder-Designs

Lipoprotein(a) Binder Designs — apo(a) KIV-8 / 8TCE

Why this target matters. Lp(a) is the most common genetic cardiovascular risk factor in the world, affecting roughly a fifth of the population, and the only major one with no approved therapy.

100 de novo small-molecule designs docked into the lysine-binding site of kringle IV type 8 (KIV-8) of apolipoprotein(a), using the 1.07 Å co-crystal 8TCE — Lp(a) KIV-8 in complex with LY3353871 — as the receptor.

This is a score-filtered set: every design here passed a docking-score cut, so the score range is deliberately narrow and is not a random sample of the generative run. Please read "How to interpret the docking scores" below before using vina_score as a ranking signal — a control experiment included in this card shows it is not one at this particular site.

---

Why lipoprotein(a) matters

Lp(a) is the most common genetic cardiovascular risk factor in the world, and the only major one with no approved therapy.

Lipoprotein(a) is an LDL-like particle in which apolipoprotein(a) — apo(a) — is joined by a single disulfide bond to the apoB100 of an LDL particle. Apo(a) is a plasminogen paralogue built from tandem kringle domains: ten types of kringle IV, one kringle V, and an inactive protease domain. The KIV type 2 domain is present in a variable number of tandem repeats, from about 2 to more than 40 copies, and this copy number is the main determinant of an individual's Lp(a) concentration.

Scale of the problem. Roughly 20–25% of the global population — on the order of 1.4 billion people — carry an Lp(a) above the 50 mg/dL (about 125 nmol/L) threshold at which risk rises measurably. Levels are set almost entirely at the LPA locus: they are 80–90% heritable, essentially fixed from early childhood, and largely unresponsive to diet, exercise, or statins. Statins in fact raise Lp(a) modestly. Median levels differ several-fold by ancestry, being highest in people of African descent and lowest in several East Asian populations, which has contributed to systematic under-recognition of the risk in some groups.

The evidence that it is causal, not merely correlated. Elevated Lp(a) is associated with myocardial infarction, ischaemic stroke, peripheral artery disease, heart failure, and — distinctively — calcific aortic valve stenosis, for which it is the strongest known genetic risk factor. Crucially, the case rests on more than epidemiology: Mendelian randomisation studies using LPA variants and KIV-2 copy number show a dose-dependent, lifelong relationship between genetically determined Lp(a) and cardiovascular events. Because genotype is randomised at conception and precedes disease, this design is largely immune to the confounding and reverse causation that weaken observational studies. Lp(a) is therefore considered causal.

Why it is thought to be pathogenic. Three mechanisms are usually invoked, and they are not mutually exclusive: the particle delivers cholesterol to the arterial wall like LDL; it is the major carrier of oxidised phospholipids in plasma, which are pro-inflammatory and pro-calcific; and apo(a)'s kringle domains give it structural homology to plasminogen, allowing it to interfere with fibrinolysis and promote thrombosis. The kringle lysine-binding sites targeted by this dataset sit at the centre of that third mechanism, and are also the interaction surface used during particle assembly.

The therapeutic gap. Diet, exercise, statins, and ezetimibe do not meaningfully lower Lp(a). PCSK9 inhibitors reduce it 15–25%, well short of what Mendelian randomisation suggests is needed — estimates for a clinically meaningful effect run to an absolute reduction on the order of 50–100 mg/dL. Lipoprotein apheresis works but is invasive, scarce, and expensive. As of this writing no Lp(a)-specific therapy is approved anywhere.

What is in development. Five agents have reached late-stage clinical testing, four of them injectable nucleic-acid drugs that suppress hepatic apo(a) synthesis — pelacarsen (antisense; the Lp(a)HORIZON outcomes trial, the first ever for an Lp(a)-lowering agent), olpasiran, lepodisiran, and zerlasiran (siRNA) — plus muvalaplin, an oral small molecule that blocks assembly of the particle by disrupting the initial apo(a)–apoB100 interaction. Muvalaplin's mechanism is the one relevant to this dataset. Additional oral small molecules have since entered development, including HRS-5346 (Hengrui/Merck) and YS2302018 (CSPC/AstraZeneca), which signals real commercial conviction in an oral approach.

Why an oral small molecule is worth pursuing. Lp(a) is an asymptomatic, lifelong, genetically fixed risk factor affecting a fifth of humanity. That is a population in which a pill is a fundamentally different proposition from an injection — particularly for primary prevention, where the treatable population is largest and adherence to parenteral therapy is hardest. The two clinical oral compounds are also unusually large and polar for oral drugs (MW ~710–740, cLogP ≈ −2.8, 18–20 rotatable bonds), which leaves obvious room for chemotypes with more conventional properties.

---

Target and site

The receptor is 8TCE chain A. Apo(a) kringle domains carry a lysine-binding site (LBS): a shallow, solvent-exposed surface groove evolved to bind a lysine side chain, and the same site through which apo(a) engages fibrin and lysine analogues such as ε-aminocaproic acid and tranexamic acid.

The site is a bipolar charge clamp with an aromatic floor:

Element	Residues (chain A)	Recognises
Anionic centre	Asp54, Glu56	the lysine ε-ammonium
Cationic centre	Arg35, Arg69	the lysine α-carboxylate
Aromatic cage	Trp60, Tyr62, Trp70	boxes the ammonium (cation-π)

The co-crystallised ligand of 8TCE, LY3353871 (HET code HWF, (2S)-3-phenyl-2-[(3R)-pyrrolidin-3-yl]propanoic acid), is a zwitterionic lysine mimetic and corresponds to a single arm of muvalaplin. Measured from the crystal pose, its amine nitrogen sits 2.48 Å from Glu56 OE1 and 2.71 Å from Asp54 OD2, while its carboxylate is 2.88 Å from Arg69 NE and 2.40 Å from Tyr62 OH. The internal N⁺···C(carboxyl) span is 4.85 Å; the protein clamp measures 8.08 Å between the Asp54/Glu56 and Arg35/Arg69 centres.

Its phenyl group, by contrast, is almost entirely solvent-exposed (2.3 protein neighbours within 5 Å, versus 16.8 for the ammonium). It is not a binding element — it is the attachment vector, the position that becomes the benzyl linker in multivalent compounds such as muvalaplin.

Contents

File	Description
designs.csv	Index of all 100 designs: ID, SMILES, docking score, physicochemical properties, pharmacophore flags, per-design contact residues
ligands.sdf	All 100 docked poses as a single 3D SDF, with bond orders and formal charges taken from the source SMILES and coordinates from the pose
structures/	Per-design ligand-only PDB (<design_id>.pdb), 100 files, in the receptor frame
receptor_8TCE.pdb	The single shared receptor used for every docking run

The receptor is byte-identical across all 100 runs, so it is stored once. Ligand poses are in the receptor's coordinate frame — load receptor_8TCE.pdb alongside any ligand file or SDF record with no superposition needed.

Receptor detail. Chains A and B, two crystallographic copies of KIV-8. Docking was performed in chain A, from which the co-crystallised HWF was deleted. HWF is retained in chain B and can be used directly as a reference pose after superposing B onto A (Cα RMSD 2.04 Å over 87 residues). One structural water is retained. All 100 designs contact chain A only.

designs.csv columns

design_id, smiles, vina_score (kcal/mol), mw, clogp, tpsa, qed, hbd, hba, rotatable_bonds, heavy_atoms, rings, aromatic_rings, formal_charge, ligand_efficiency (−score / heavy atom), murcko_scaffold, has_carboxylate_or_tetrazole, has_basic_amine, lysine_mimetic_zwitterion, pose_in_sdf, n_contacts, contact_residues (chain:residue, within 4 Å of the pose, ;-separated).

Rows are sorted by docking score, best first.

Set composition

Designs	100 (97 unique structures; 3 molecules contribute two poses)
Murcko scaffolds	84
Docking score	−10.10 to −8.40 kcal/mol
MW	371–437 (median 411)
cLogP	−0.57 to 3.96 (median 2.59)
TPSA	58–153 (median 93)
QED	0.26–0.73 (median 0.55)
Heavy atoms	28–31
Rotatable bonds	2–5

All designs are Rule-of-5 compliant, which distinguishes them from the two clinical oral compounds in this mechanism class.

Contact frequency (within 4 Å): Glu56, Tyr62, Arg69, Trp70 at 100%; Trp60 99%; Arg35 75%; Asp54 62%; Arg34 58%; Trp32 23%; His33 12%. Every design engages the lysine-binding site.

Pharmacophore composition. Of the 100: 21 carry a carboxylate or tetrazole, 13 a basic amine, and 4 carry both (lysine_mimetic_zwitterion = 1). Since every molecule with experimental evidence of binding a kringle LBS is a zwitterion — L-lysine, ε-aminocaproic acid, tranexamic acid, and LY3353871 itself — those 4 are the designs most likely to engage the site by its native recognition chemistry. They are flagged in designs.csv so they can be selected directly.

How to interpret the docking scores

The vina_score column records the value each design was filtered on. It should not be read as predicted affinity at this site. A control experiment run under an identical protocol — same receptor, box, and settings, changing only the ligand:

Ligand	Vina (kcal/mol)	Ligand efficiency
Designs in this set	−10.10 to −8.40	0.27–0.33
LY3353871 (HWF) — the crystallographic ligand	−4.81	0.301
Tranexamic acid	−4.24	0.385
Naphthalene (neutral decoy)	−4.19	0.419
ε-aminocaproic acid	−4.05	0.450
L-lysine	−3.92	0.392

The co-crystallised ligand of this very structure scores worse than every design, and a naphthalene decoy with no functional groups outscores L-lysine, the site's natural ligand. Ligand efficiency inverts the ranking entirely.

The cause is structural, not a bug: AutoDock Vina's function is gauss1 + gauss2 + repulsion + hydrophobic + H-bond, with no electrostatic term. The KIV-8 LBS is a bipolar salt-bridge clamp, so its dominant interactions are close to invisible to the scoring function, while per-atom hydrophobic contact is rewarded. At a shallow, solvent-exposed site, larger aromatic ligands therefore accumulate favourable score without engaging the recognition chemistry. Across the parent generative run, the mean score for acid-bearing molecules was −8.33 versus −8.19 for fully neutral ones — a 0.14 kcal/mol spread.

Pose prediction is unaffected: re-docking HWF reproduces its crystal pose to 1.93 Å RMSD, and all 100 designs place into the correct groove. Geometry is reliable here; ranking is not.

Recommended use. Treat this set as a pose and chemotype library, not a ranked hit list. If you need a ranking, rescore with an electrostatics-aware method (AD4, MM-GBSA on minimised complexes, or free-energy methods). If you are selecting compounds to pursue, start from lysine_mimetic_zwitterion = 1 and from ligand_efficiency rather than from raw score.

Design guidance for follow-up work

Derived from the crystal pose, for anyone extending this set:

1. A protonatable primary or secondary amine (pKa > 9) — not an aniline, amide, or aromatic nitrogen. It must be cationic at pH 7.4 to make the Asp54/Glu56 bidentate contact and occupy the Trp60/Trp70 cation-π box.
2. A carboxylate or bioisostere (tetrazole, acylsulfonamide) for the Arg69 salt bridge and the Tyr62 hydrogen bond. Neutral esters and amides will not substitute.
3. An N⁺···C(carboxyl) span of ≈ 4.85 Å, pre-organised. Reaching the distance in a low-energy conformer is necessary; reaching it without a large conformational search is what buys affinity. Head groups that satisfy this well include 2-(pyrrolidin-3-yl)acetic acid, 2-(piperidin-4-yl)acetic acid, 3-aminocyclopentane-1-carboxylic acid, and tranexamic acid. Para-substituted benzoic acids overshoot to ~6.5 Å and cannot reach the clamp.
4. Keep the exit vector free. The solvent-exposed position corresponding to LY3353871's phenyl is where a linker attaches. Monovalent affinity at a kringle LBS is intrinsically modest — muvalaplin is trivalent, and its sub-nanomolar potency is avidity across multiple KIV domains rather than per-site affinity. Designs intended for eventual multivalent assembly should preserve that vector.

Provenance and limitations

Designs were produced by the TechnetiumTx GA-II generative pipeline and docked with AutoDock Vina into 8TCE chain A. These are computational designs. None has been synthesised or assayed, and no experimental binding, activity, selectivity, or safety data exist for any molecule in this set. Docking scores are not affinities; see above for a direct demonstration of their limits at this site. Nothing here is a therapeutic recommendation.

Citation

If you use this dataset, please cite the source structure alongside it:

Structure: PDB 8TCE — Lp(a) kringle IV domain 8 in complex with LY3353871. Reported in: Discovery of potent small-molecule inhibitors of lipoprotein(a) formation, Nature (2024). doi:10.1038/s41586-024-07387-z

License

CC BY 4.0.