The motivation behind the work lies in a quiet revolution in the drug pipeline. Under the Biopharmaceutical Classification System, and its more modern cousin the Developability Classification System, orally administered drugs are sorted into four classes according to how soluble they are in gastrointestinal fluids and how easily they permeate intestinal membranes. In earlier eras, most new medicines fell into classes I and III, compounds with no real solubility problems. Today, the majority of new drug candidates are poorly soluble molecules belonging to classes II and IV, often the products of combinatorial chemistry and high-throughput screening that favour potent, lipophilic targets. For these compounds, a dissolution profile alone is a poor predictor of what happens in a patient, because the rate-limiting step may not be how fast the tablet breaks apart but how fast the dissolved molecules squeeze through the lipid barriers of the intestinal wall.

Commercial dissolution-permeation systems do exist, and they have become valuable tools for formulation screening. Yet the Latvian team, led by Andrejs Sitovs and Valentyn Mohylyuk, identified three persistent shortcomings. The acceptor compartments of commercially available systems that include compendial dissolution testing hold no more than 50 millilitres, a volume that quickly saturates when a poorly soluble drug finally dissolves and needs somewhere to go. None of the existing platforms is compatible with ex vivo membranes, such as excised segments of animal or human intestine, which would allow active and facilitated transport mechanisms to be studied. And the dissolution step in most devices is not the standardised, pharmacopoeial kind that regulators around the world already recognise. The researchers’ answer was to bolt a CHM1 Ussing chamber onto a compendial USP2 dissolution apparatus, connecting the two with silicone tubing and a six-channel peristaltic pump in a closed circulation loop.

The engineering of the setup is elegantly straightforward. Two one-litre USP2 vessels ran side by side at 37 degrees Celsius with paddles spinning at 50 revolutions per minute. One vessel held the donor medium, a phosphate buffer at pH 6.8 that mimics the small intestinal lumen, where the tablet dissolved. The other held the acceptor medium, a phosphate buffer at pH 7.4, representing the blood side of the intestinal barrier. Medium was pumped from the donor vessel, roughly one centimetre below its surface, through the Ussing chamber, where it contacted a membrane mounted over a 1.13 square centimetre aperture, and then dripped passively back into the donor vessel from a height of about one centimetre. The acceptor vessel circulated through the opposite side of the chamber. Each tube ran one metre in length, and adhesive tape over the dissolution apparatus apertures kept evaporative losses to between 0.3 and 1.3 percent of the total volume over the course of an experiment.

The membrane at the heart of the chamber was an artificial lipid barrier modelled on the parallel artificial membrane permeability assay, better known in the field as PAMPA. The researchers spread 18.4 microlitres of soy L-α-phosphatidylcholine dissolved in dodecane, a mixture abbreviated LiDo, across one of two supporting filter membranes: a hydrophobic polyvinylidene fluoride membrane with 0.45 micrometre pores and a thickness of 125 micrometres, or a polycarbonate membrane with 0.20 micrometre pores and a thickness of only 10 micrometres. They tested LiDo concentrations of 5 and 10 percent by weight, deliberately lower than the 20 percent used in conventional PAMPA, because previous work had shown that increasing lecithin concentration in dodecane tends to slow naproxen permeation. Scanning electron microscopy confirmed clear differences in pore size and structure between the two filter materials, differences that would later prove consequential.

Naproxen itself was chosen with care. The drug, administered here as Nalgesin 550 milligram tablets containing naproxen sodium equivalent to 500 milligrams of naproxen, is a BCS class II compound: poorly soluble in acidic fluids but highly soluble at near-neutral pH, and transported across intestinal membranes purely by passive diffusion. That last property matters, because an artificial lipid membrane can only model passive transport. At the pH values of the small intestine, between 6.8 and 7.4, more than 99 percent of naproxen exists in its ionised, water-soluble form, according to the Henderson-Hasselbalch relationship. The researchers confirmed that proper sink conditions were maintained, with sink indices of 7.2 in the donor medium and 2.9 in the acceptor medium, meaning the buffers could hold far more dissolved drug than the experiments actually delivered. The tablets released more than half their payload within ten minutes and essentially the entire dose within thirty.

The permeability numbers that emerged were striking, and humbling. Across the 0.45 micrometre PVDF membrane, the apparent permeability coefficient of naproxen was 0.91 plus or minus 0.35 times ten to the minus eight centimetres per second with 5 percent LiDo, and 0.75 plus or minus 0.23 times ten to the minus eight with 10 percent LiDo, a difference that statistical analysis found not significant. The thinner polycarbonate membrane, however, roughly doubled the measured permeability, yielding 1.99 plus or minus 0.57 times ten to the minus eight centimetres per second at 5 percent LiDo. A control experiment in which donor and acceptor media were identical, both at pH 6.8, produced a similar value of 0.95 times ten to the minus eight, suggesting the pH gradient was not the dominant factor. Every one of these figures was approximately one hundred times lower than permeability values reported in the literature for naproxen, whether from artificial membrane-Ussing models, which report 19.55 times ten to the minus six centimetres per second, or from PAMPA experiments, which report around 4.2 times ten to the minus six.

The authors attribute this hundredfold gap primarily to the composition of their acceptor medium. Other studies routinely add chemical scavengers that mimic serum proteins, binding permeated drug and pulling the equilibrium forward, thereby inflating the apparent flux. The Latvian system used plain phosphate buffer with no such additives, so the acceptor side offered no extra thermodynamic incentive beyond the pH gradient. Membrane architecture likely played a supporting role: although the PVDF pores are on average 2.25 times larger than those of the polycarbonate filter, they appear structurally more complex under the microscope, which may affect how the lipid solution coats and distributes across the surface. The polycarbonate membrane is also twelve times thinner, shortening the diffusional path. Notably, electron microscopy showed the PVDF membrane used here matched the one in the commercial MicroFLUX system, meaning the discrepancy with published values cannot be blamed on a defective filter.

What the study delivers, then, is not a perfected predictive model but a validated platform with unusual flexibility. Because the permeation module is a standard Ussing chamber, the same rig can accept ex vivo tissue mounts, opening the door to dissolution-permeation experiments that capture active and facilitated transport, something no artificial membrane can do. The one-litre vessels provide large and adjustable acceptor volumes that poorly soluble compounds demand, and the dissolution step remains fully compendial, which matters for regulatory conversations. The researchers frame the potential payoff in broad terms: better bioavailability for oral formulations, fewer failed clinical trials, lower therapeutic doses, less unabsorbed drug excreted into the environment, and reduced financial risk across the development pipeline.

The work, funded through the European Union Recovery and Resilience Facility under a Latvian scientist grant for innovative biopharmaceutical evaluation of hard-to-develop oral medicines, is explicitly a feasibility study, and the authors are candid about its limits. The artificial membrane setup simulates passive transport only, and the permeability coefficients it generates are not yet calibrated against in vivo data. But the direction of travel is clear. If the methodology is adequately improved, as the team puts it, the humble dissolution paddle and the century-old Ussing chamber, working in tandem, could give formulation scientists a far more honest preview of how a new medicine will behave inside the human body, long before a single volunteer swallows the first tablet.

Subject of Research: A compendial dissolution-permeation system combining a USP2 dissolution apparatus with an Ussing chamber for biopharmaceutical evaluation of poorly soluble drugs

Article Title: Dissolution-permeation approach for biopharmaceutical evaluation: a feasibility study using naproxen

Article References: Sitovs, A., Gurkina, K., Petersone, L., & Mohylyuk, V. (2026). Dissolution-permeation approach for biopharmaceutical evaluation: a feasibility study using naproxen. Journal of Pharmacy & Pharmaceutical Sciences, 29, Article 16570. https://doi.org/10.3389/jpps.2026.16570

Image Credits: AI Generated

DOI: 10.3389/jpps.2026.16570

Keywords: dissolution-permeation, Ussing chamber, naproxen, BCS class II, biopharmaceutics, PAMPA, drug permeability, oral drug absorption, pharmaceutical analysis, LiDo membrane, sink conditions, in vitro testing

The research team designed ten new naproxen derivatives, labeled ND1 through ND10, by replacing the drug’s terminal carboxylic acid with an amide linkage attached to a variety of aromatic, heterocyclic, and aliphatic fragments. The naphthalene methoxy backbone that the cyclooxygenase enzymes recognize was left untouched, preserving the pharmacophore responsible for anti-inflammatory action. The logic is rooted in well-established medicinal chemistry: the acidic group anchors classical NSAIDs inside the COX channel through a salt bridge with the conserved Arg120 residue, but that same ionizable acid undergoes ion trapping in the acidic gastric lumen, directly irritates the epithelial barrier, and inhibits mitochondrial oxidative phosphorylation in stomach lining cells. Swapping the acid for a neutral amide could, in principle, keep the anchoring interactions while eliminating the gastric toxicity, and the amide’s terminal fragment could even reach into the less conserved side pocket that distinguishes COX-2 from COX-1, a strategy long used to convert non-selective NSAIDs into COX-2-preferring inhibitors.

To test this idea without synthesizing a single molecule, the researchers deployed a multi-layered computational pipeline. Density functional theory calculations at the B3LYP/6-31+G(d,p) level probed the electronic character of each derivative, yielding frontier molecular orbital energies, chemical hardness, softness, electronegativity, and electrophilicity. The results were striking: derivatives ND1 and ND4 showed the smallest HOMO-LUMO energy gaps, around 0.30 and 0.31 electron volts, compared with 0.87 electron volts for naproxen itself, indicating far greater electronic softness and reactivity. Soft molecules polarize more readily, which favors effective dispersion forces and electronic redistribution during non-covalent binding within the hydrophobic COX-2 pocket. ND2, meanwhile, posted the highest dipole moment at over 30 Debye, suggesting the strongest potential for polar interactions with the receptor. Molecular electrostatic potential mapping reinforced the picture, revealing that the high-polarization derivatives, including ND1, ND4, ND5, ND6, and ND9, retained sharply localized electron-rich and electron-poor regions capable of directional hydrogen bonding, effectively replacing the electrostatic role of the discarded carboxylic acid.

Before any binding predictions could be trusted, the docking protocol itself had to be validated. The researchers re-docked the co-crystallized ligands into the active sites of COX-1 and COX-2, using crystal structures 3KK6 and 3NT1 from the Protein Data Bank, and obtained root-mean-square deviations of 1.48 and 0.936 angstroms respectively, well within accepted thresholds. With the protocol confirmed, all ten derivatives were docked against both enzymes using AutoDock Vina and AutoDock4, alongside reference drugs including diclofenac, aspirin, mefenamic acid, and indomethacin. Every single derivative outperformed the parent drug. Naproxen scored roughly minus 7.5 kilocalories per mole against COX-2, while the derivatives ranged from minus 8.3 to minus 10.1 kilocalories per mole under Vina scoring. ND10, bearing an aminomethyl benzoic acid terminus, achieved the most favorable AutoDock4 score of minus 11.49 kilocalories per mole, driven largely by superior van der Waals contacts, while ND5’s bulky cycloheptylethyl group delivered a minus 10.43 kilocalories per mole result.

The docking poses revealed how the amide modification preserves potency through a clever molecular sleight of hand. Although converting the carboxylate into an amide removes the classical ionic salt bridge with Arg120, the amide carbonyl steps in as an alternative hydrogen-bond partner for the Arg120 guanidinium group in complexes ND3, ND5, and ND10. This neutral interaction, combined with deep penetration of the derivatives into the hydrophobic core near GLY526 and ALA527, provides a rational basis for COX-2 selectivity. Several complexes, including ND2, ND4, ND5, ND6, ND7, ND9, and ND10, reproduced the essential binding pattern of naproxen through pi-pi stacking with aromatic residues such as PHE381, TYR385, and TRP387, while ND4, ND6, and ND10 added hydrogen bonds and pi-sigma interactions with TYR385. By replacing the ionic anchor essential for COX-1 recognition with amide-based hydrogen bonding and expanded aromatic stacking, the derivatives exploit the larger active-site volume of COX-2 while potentially reducing the COX-1-mediated gastrointestinal harm of the parent acid.

Pharmacokinetic profiling through SwissADME and ProTox 3.0 added a crucial safety dimension. All ten derivatives fell safely below the 500 grams per mole Lipinski ceiling, satisfied hydrogen-bond donor and acceptor limits, and showed topological polar surface areas between 38 and 76 square angstroms, comfortably within the threshold for passive oral absorption. Predicted gastrointestinal absorption was high across the series, and all compounds were predicted non-cardiotoxic. The toxicity results were particularly encouraging: naproxen showed an LD50 of 248 milligrams per kilogram in toxicity class 3, whereas ND1 and ND8 exhibited LD50 values of 4000 and 5000 milligrams per kilogram in the far safer class 5. ND1, ND8, and ND10 were additionally predicted to be non-neurotoxic, and nearly all compounds showed inactive nephrotoxicity predictions, in contrast to active nephrotoxicity for naproxen itself. Caveats remain, however, as several analogues were predicted to inhibit multiple cytochrome P450 isoforms, flagging a potential risk of drug-drug interactions that will require experimental confirmation.

The three most promising candidates, ND2, ND4, and ND10, were then subjected to 100-nanosecond molecular dynamics simulations in GROMACS using the CHARMM36 force field, solvated in explicit water with physiological ion concentrations. The trajectories told a dramatic story. Naproxen repeatedly dissociated and rebounded at three distinct binding-site regions over the simulation, maintaining only intermittent hydrogen bonding of zero to two bonds and showing the highest conformational fluctuation of any system studied. ND4, by contrast, locked into its binding pose almost immediately and maintained a steady four to six hydrogen bonds, occasionally reaching seven, throughout the entire 100 nanoseconds. Its average root-mean-square deviation of 0.215 nanometers was the lowest of the series, and its solvent-accessible surface area of 255.90 square nanometers indicated the most compact, tightly bound complex. ND10 required roughly 50 nanoseconds to settle but then held its final pose with remarkable rigidity, achieving the steadiest plateau in the RMSD analysis.

Free-energy calculations using the MM/GBSA method over 501 trajectory snapshots quantified these differences with impressive precision. Naproxen’s binding free energy averaged minus 13 kilocalories per mole, while ND2 reached minus 16.5 and ND10 minus 25.4 kilocalories per mole. ND4 was in a class of its own at minus 46 kilocalories per mole, powered by an enormous van der Waals contribution of nearly minus 49 kilocalories per mole and an electrostatic term of minus 27 kilocalories per mole, reflecting its extended hydrogen-bond network with TYR324, ARG89, and ARG120. Per-residue energy decomposition confirmed that Arg120 contributed favorably to ND4 binding, demonstrating that the amide linkage had successfully recreated the anchoring interaction that the carboxylic acid once provided. Conformational analysis added a final layer of insight: 75 percent of the ND4 trajectory occupied a single conformational family, compared with only 43 percent for naproxen, whose energy landscape was fragmented across many shallow, poorly defined basins.

The authors are careful to frame these findings as exploratory rather than conclusive. Computational docking, dynamics, and ADMET predictions cannot fully capture the complexity of living systems, and predicted binding affinities do not always translate to in vitro or in vivo efficacy and safety. Experimental verification of the amide modification’s impact on cyclooxygenase selectivity, anti-inflammatory activity, metabolic stability, and gastrointestinal protection remains essential. Still, the study stands as a compelling demonstration of modern computer-aided drug design, showing how density functional theory, electrostatic mapping, validated docking, long-timescale simulation, and ensemble free-energy calculations can converge to identify ND4 and ND10 as lead candidates worthy of synthesis and biological testing. If laboratory work confirms what the simulations suggest, the humble painkiller in millions of medicine cabinets could one day be replaced by a safer, more potent descendant of itself.

Subject of Research: Computational design of naproxen amide derivatives as improved anti-inflammatory COX inhibitors

Article Title: Structure-based computational design of naproxen analogues with improved anti-inflammatory activity by DFT analysis, ADMET assessment, molecular docking, and molecular dynamics simulations

Article References: Abdullah, A., & Akhtar, M. (2026). Structure-based computational design of naproxen analogues with improved anti-inflammatory activity by DFT analysis, ADMET assessment, molecular docking, and molecular dynamics simulations. Results in Chemistry, 31, Article 103933. https://doi.org/10.1016/j.rechem.2026.103933

Image Credits: AI Generated

DOI: 10.1016/j.rechem.2026.103933

Keywords: naproxen, NSAIDs, COX-2 inhibition, computer-aided drug design, molecular docking, molecular dynamics simulation, DFT, ADMET, MM/GBSA, amide derivatives, gastrointestinal toxicity, drug discovery