Computational study ranks Andrographis diterpenoids at the PCSK9–LDLR interface

Researchers have carried out a staged computational analysis of six diterpenoids from Andrographis paniculata, the medicinal herb long studied for possible lipid-modulating effects, to identify which of the compounds best complements the protein-protein interface between PCSK9 and the low-density lipoprotein receptor (LDLR). The work, published in Discover Chemistry, was designed explicitly as a structure-based prioritization exercise rather than a demonstration of inhibition, and its authors are careful to state that no binding free-energy or biochemical measurements were obtained.

PCSK9 is a validated target in lipid management because reducing circulating or functional PCSK9 increases the abundance of LDL receptors on hepatocyte surfaces and accelerates plasma LDL clearance. Elevated low-density lipoprotein cholesterol is a major modifiable contributor to atherosclerotic cardiovascular disease, and human genetic evidence reinforces the importance of this axis: gain-of-function variants in PCSK9 raise LDL cholesterol, whereas loss-of-function variants lower both LDL cholesterol and cardiovascular risk. At the hepatocyte surface, PCSK9 binds the epidermal growth factor-like repeat A of LDLR, and the 3BPS crystal structure shows this engagement occurring across an extended surface, with acidic endosomal conditions strengthening the interaction and favoring lysosomal targeting rather than receptor recycling.

That binding geometry is what makes the problem computationally demanding. The PCSK9-LDLR interface is broad, solvent-exposed, and lacks a canonical deep pocket, so docking scores on such shallow surfaces can reward nonspecific accommodation without predicting productive competition. The study therefore treated surface topology, pose reproducibility, and dynamic persistence as structural metrics independent of nominal docking affinity, reasoning that structural complementarity is more informative than a docking score alone at this kind of interface.

The team, led by Muhammad Nazrul Hakim with colleagues including Armania Nurdin, Mohd Sofian Omar Fauzee, and Zuraini Ahmad, evaluated a predefined six-compound panel of A. paniculata labdane diterpenoids through four sequential levels: ligand and receptor quality control, triplicate interface-focused docking, integration of interaction, interface, and ADMET predictions, and short-timescale molecular dynamics. Docking used AutoDock Vina 1.1.2 with three independent seeds per ligand, and the median Mode 1 score was used for affinity ranking. The prospective interface set of 19 PCSK9 contact residues was defined before final ligand-contact counts were interpreted, based on the 3BPS structure and established studies of PCSK9-EGF-A recognition.

The docking results revealed how narrow the margins were. Median affinities spanned only 0.7 kcal/mol, from -5.6 to -6.3 kcal/mol. Isoandrographolide, labeled L05, produced the most favorable median score at -6.3 kcal/mol, followed by L02 at -6.2 and L01 at -6.1. Andrograpanin, L06, scored -6.0 kcal/mol, less favorable than L05 by just 0.3 kcal/mol. The authors explicitly declined to treat this small separation as decisive, noting that such differences fall below the precision generally required for mechanistic conclusions from empirical docking functions.

Interface contact mapping, however, produced a strikingly different ranking. Every ligand except L06 overlapped only one of the predefined interface residues. L06, by contrast, contacted eight: SER153, PRO155, ASP238, ALA239, ILE369, PHE379, VAL380, and SER381, spanning proximal, central, and distal segments of the LDLR-recognition surface. L06 also showed the strongest cross-seed convergence, with pairwise heavy-atom RMSDs of just 0.018 to 0.036 angstroms among seed-specific poses, indicating that independent docking searches repeatedly recovered essentially the same surface pose. L05, despite its top score, overlapped only ILE154. The docking stage therefore deliberately generated two distinct candidates: L05 as the affinity comparator and L06 as the interface-directed lead.

A notable methodological episode concerned the receptor model itself. The initial workflow built on the experimental crystal chains was rejected after quality control identified unresolved coordinate gaps that generated artificial long-range connectivity and contradictory RMSD and RMSF behavior. The team rebuilt a comparative model of PCSK9 residues 153-446 using SWISS-MODEL, achieving 100 percent sequence identity and coverage to the template region, with GMQE 0.93 and QMEANDisCo 0.90 plus or minus 0.05. Internal missing segments at residues 173-176 and 212-219 were reconstructed, and contacts involving those reconstructed residues were treated as model-sensitive and excluded from primary mechanistic interpretation. Importantly, the residues that differentiated L06 from L05 lie outside the reconstructed segments. The rebuilt apo model completed 1 nanosecond of production dynamics with low backbone RMSD of 0.1141 nanometers, low C-alpha RMSF, and preserved compactness.

The matched 1-nanosecond simulations of the L05 and L06 complexes then preserved the interface distinction. Both ligands remained in close contact with PCSK9 in all 101 sampled frames, with similar atom-pair contact counts and nearly identical minimum distances, and neither caused evidence of global fold disruption. But L06 showed lower protein-aligned ligand RMSD than L05, 0.0961 versus 0.1355 nanometers, and a higher proportion of frames containing at least one protein-ligand hydrogen bond, 58.4 percent versus 47.5 percent. Residue-level occupancy was also sharply different: L05 remained dominated by ILE154 at 85.1 percent of frames with no persistent distal-segment contact, whereas L06 maintained PRO155, ALA239, and ILE369 in 100 percent of frames, SER153 in 99.0 percent, PHE379 in 98.0 percent, ASP238 in 96.0 percent, ILE154 in 94.1 percent, and SER381 in 64.4 percent, bridging the 153-155, 238-239, and 369-381 regions during the sampled interval.

The authors attach firm caveats to these numbers. Because the trajectories were 1 nanosecond, unreplicated, and not designed to establish convergence, the occupancy and hydrogen-bond frequencies are descriptive of the sampled trajectories only, and longer independent replicate simulations extending into the tens of nanoseconds would be needed to test whether the differences persist. Docking was performed on a largely rigid receptor, ADMET predictions are model-dependent, and no biochemical, cellular, pharmacokinetic, or in vivo validation was available. The findings therefore represent structure-based prioritization rather than evidence of PCSK9-LDLR inhibition.

The ADMET profiling added a practical counterweight. All six compounds met Lipinski criteria without violations and were predicted to have high gastrointestinal absorption, but the cross-platform picture was not uniformly favorable. L02, despite the second-best docking score, showed the lowest predicted intestinal absorption and the only hERG II and hepatotoxicity alerts, supporting its de-prioritization. L06 combined predicted 96.632 percent intestinal absorption with no pkCSM AMES, hERG, or hepatotoxicity alerts, yet also showed the highest lipophilicity, the poorest predicted solubility, low predicted fraction unbound, possible CYP1A2 inhibition, and a ProTox class 2 prediction. L05 had more moderate lipophilicity but was predicted as a P-glycoprotein substrate with several toxicity endpoint alerts. The authors conclude that andrograpanin is the higher-priority interface-directed candidate for direct experimental testing within this computational workflow, while isoandrographolide remains a valuable affinity comparator, and they propose biophysical competition assays, surface plasmon resonance or microscale thermophoresis, hepatocyte LDLR-recycling measurements, and LDL uptake studies as suitable next steps. They also note that L06 did not contact the canonical hot spot Arg194, so it should not be described as reproducing the complete EGF-A recognition network. Preclinical evidence supports lipid-modulating activity of A. paniculata extracts and related compounds, but direct PCSK9 modulation has not been established for the molecules studied here; the present results supply a testable structural hypothesis, not a therapeutic claim.

Subject of Research: Structure-based computational prioritization of Andrographis paniculata diterpenoids as candidate ligands at the PCSK9–LDLR protein-protein interface

Article Title: Structural prioritization of Andrographis paniculata diterpenoids at the interface between PCSK9 and LDLR

Article References: Hakim, M. N., Nurdin, A., Fauzee, M. S. O., & Ahmad, Z. (2026). Structural prioritization of Andrographis paniculata diterpenoids at the interface between PCSK9 and LDLR. Discover Chemistry, 3(1), Article 592. https://doi.org/10.1007/s44371-026-01049-x

Image Credits: AI Generated

DOI: 10.1007/s44371-026-01049-x

Keywords: PCSK9, LDLR, Andrographis paniculata, andrograpanin, isoandrographolide, molecular docking, molecular dynamics, diterpenoids, ADMET, protein-protein interface, drug discovery, cholesterol metabolism