The review, conducted by researchers at Western New England University, Princeton University, Jahrom University of Medical Sciences, and Kurdistan University of Medical Sciences, followed PRISMA 2020 guidelines and screened 2,044 records from PubMed and Europe PMC covering literature from January 2010 through December 2025. After a two-stage automated screening process verified independently by two researchers, 123 studies met the inclusion criteria, each one combining an experimental microphysiological system with a computational model such as physiologically based pharmacokinetic modeling, quantitative systems pharmacology, in vitro-to-in vivo extrapolation, or machine learning. What emerged from the synthesis is a portrait of a field in explosive adolescence: publication output has been compounding at roughly 35% per year, and nearly 38% of all qualifying studies appeared in 2024 and 2025 alone.

To understand why this integration matters, it helps to grasp what each half brings to the table. Microphysiological systems are multicellular culture platforms engineered to recreate the physiological conditions of specific organs, complete with three-dimensional tissue architecture, mechanical forces such as stretch and fluid perfusion, and the cell types that make an organ function, including parenchymal cells, endothelial barriers, and immune cells. The field’s founding moment came in 2010, when Donald Ingber and colleagues at Harvard’s Wyss Institute built a lung-on-a-chip that reproduced the alveolar-capillary interface and demonstrated that cyclic breathing motions amplified inflammatory and toxic responses, something no static petri dish could ever capture. In the years since, chips have been built for virtually every major organ, and sophisticated multi-organ platforms now link human heart, liver, bone, and skin niches through recirculating vascular flow, generating pharmacokinetic profiles that begin to resemble those of an actual body.

But even the most elegant chip produces data that must be translated into predictions about intact humans, and that is where the computational half comes in. Physiologically based pharmacokinetic models represent the body as a network of compartments connected by blood flow, mechanistically predicting absorption, distribution, metabolism, and excretion from a drug’s physicochemical properties. Quantitative systems pharmacology models go further, folding disease biology and drug mechanism into mathematical frameworks that can forecast therapeutic outcomes. The digital twin concept binds these two worlds together through bidirectional information flow: the virtual model is initialized and calibrated with measurements from the physical system, and its predictions, in turn, guide which experiments to run next. In principle, this creates a continuously learning loop in which a chip’s human-relevant biology feeds a model that can simulate thousands of virtual drug trials before a single animal is touched.

The systematic review reveals exactly where this synergy has matured and where it remains embryonic. The liver dominates the landscape, appearing in 30.9% of studies, followed by vasculature at 19.5%, the gut at 18.7%, and immune components at 17.1%. This distribution is no accident: hepatic metabolism is the single most important determinant of drug clearance, and drug-induced liver injury remains one of the costliest reasons for late-stage clinical failures. Crucially, a gap analysis showed that only two organ-model pairings qualify as genuinely established, defined as more than five studies each: liver with physiologically based pharmacokinetic modeling, and liver with pharmacokinetic modeling. Countless combinations, particularly involving the kidney, lung, brain, skin, and pancreas, remain unexplored. Kidney, brain, and tumor models each appeared in just over 10% of studies, while lung, heart, and pancreas hovered near 5%, and skin registered a mere 1.6%.

The technical infrastructure underpinning the field is equally revealing. Among commercial platforms, Emulate led with 55.8% of citations, followed by Mimetas at 23.3% and CN Bio at 19.2%, yet custom and academic-built platforms still accounted for 74.2% of systems used, a sign that innovation remains distributed and that standardization is incomplete. On the software side, MATLAB was the most frequently reported computational environment, appearing in 20% of full-text analyzed studies, ahead of dedicated pharmacokinetic suites such as Simcyp, GastroPlus, and PK-Sim. Microfluidic designs dominated platform architecture at 76.7%, with polydimethylsiloxane as the fabrication material of choice in roughly half of studies, and 48.3% of platforms described explicit multi-organ configurations. Notably, adoption of multi-organ systems accelerated sharply after 2021, reflecting a growing recognition that organ-organ interactions, such as enterohepatic recirculation between gut and liver, are essential for accurate whole-body pharmacokinetic prediction.

Perhaps the most encouraging finding concerns validation, the make-or-break criterion for regulatory acceptance. A striking 95.8% of studies discussed comparisons with clinical or human data, and 77.5% reported quantitative accuracy metrics such as average fold error, the same standard regulators apply to physiologically based pharmacokinetic model qualification. The benchmark here is the Emulate Liver-Chip, which in a blinded assessment of 27 compounds recommended by the industry’s Innovation and Quality consortium achieved 87% sensitivity and 100% specificity for predicting drug-induced liver injury, dramatically outperforming the 50 to 60% concordance typical of animal toxicity studies. Animal models fail so often because species-specific differences in drug metabolism, immune responses, and disease mechanisms make rodent and non-rodent data unreliable guides to human outcomes, and methodological weaknesses like inadequate randomization and publication bias compound the problem.

The regulatory ground has shifted beneath all of this in ways that would have seemed unthinkable a few years ago. The FDA Modernization Act 2.0, signed in December 2022, eliminated the federal mandate for animal testing that had stood since 1938, explicitly defining cell assays, organ chips, microphysiological systems, and computer modeling as acceptable nonclinical tests. In September 2024, the first organ-on-a-chip, a Liver-Chip designed to predict drug-induced liver injury, was accepted into the FDA’s ISTAND pilot qualification program, and in April 2025 the agency announced a roadmap for phasing out animal testing in preclinical safety studies, beginning with monoclonal antibodies. Economic modeling based on the Liver-Chip’s validated performance suggests that routine adoption could save the pharmaceutical industry more than $3 billion annually by catching hepatotoxic drugs earlier, rescuing safe compounds wrongly flagged by insensitive assays, and accelerating go/no-go decisions.

Yet the review also punctures some hype with unsparing honesty. Despite the digital twin branding, most studies sit at the lower rungs of a maturity spectrum: the majority perform one-directional parameter transfer or one-time model calibration rather than the closed-loop, continuously updated coupling that defines a true twin. A manual review of the twelve machine-learning studies found that only one actually trained a predictive model on chip-derived data, a neural network predicting oxygen transport from vascularized chip morphology, while most used machine learning for image analysis or discussed it without applying it. And on the economics, the picture is nearly blank: while 92.5% of studies mentioned efficiency gains and 73.3% invoked animal reduction, only 10.8% cited dollar figures and not a single study performed a formal cost-effectiveness or cost-benefit analysis. The authors argue that liver-focused applications are now mature enough to serve as a template, and that the priority frontiers are kidney, lung, and brain integration, multi-organ systems paired with multi-compartment models, standardized parameter extraction, uncertainty quantification, shared benchmark datasets, and the health-economic evidence that will convince pharmaceutical decision-makers to abandon a century of dependence on animal testing.

Subject of Research: Integration of microphysiological systems with digital twin computational modeling for pharmaceutical development

Article Title: Integration of microphysiological systems with computational and digital twin modeling for pharmaceutical development: a systematic review

Article References: Soheili, M., Gilzad Kohan, B., Rastegar, L., Moradi, Y., & Gilzad Kohan, H. (2026). Integration of microphysiological systems with computational and digital twin modeling for pharmaceutical development: a systematic review. Journal of Pharmacy & Pharmaceutical Sciences, 29, Article 16454. https://doi.org/10.3389/jpps.2026.16454

Image Credits: AI Generated

DOI: 10.3389/jpps.2026.16454

Keywords: microphysiological systems, organ-on-chip, digital twin, PBPK modeling, drug development, pharmacokinetics, drug-induced liver injury, FDA Modernization Act 2.0, machine learning, IVIVE, preclinical testing, systematic review

The review’s central argument is that multi-epitope vaccines are chemical products as much as immunological constructs. Once short antigenic peptide segments are stitched together with linkers, adjuvant sequences and carrier domains, the resulting molecule must survive formulation, storage, administration and biological transport. Peptide constructs frequently display unfavorable charge distribution, low aqueous solubility, excessive hydrophobicity, non-specific aggregation or susceptibility to proteases. Linkers intended to improve antigen processing can instead create new junctions, bury epitopes or introduce instability. A sequence that ranks highly in an MHC-binding predictor may therefore fail long before it reaches any immunological endpoint, and the authors insist that these chemical liabilities are part of the vaccine hypothesis itself, not a downstream inconvenience to be fixed later.

Nanocarriers enter the picture as chemically tunable systems that can partially rescue these liabilities. Polymeric nanoparticles built from poly(lactic-co-glycolic acid), chitosan or alginate offer biodegradable matrices and charge-based interactions, though they bring acidic degradation microenvironments, burst release and batch variability. Lipid-based systems, including the lipid nanoparticles that made messenger RNA vaccines possible, can encapsulate peptides and adjuvants while controlling membrane composition and PEG density, but cationic or ionizable formulations may raise reactogenicity and complement activation. Virus-like particles provide multivalent, repetitive antigen display, yet inserted epitopes can disrupt capsid assembly or reduce yield. Extracellular vesicles, a biologically derived platform, can display membrane-associated antigens and carry peptide or mRNA cargo, but they suffer from source-cell heterogeneity, variable cargo copy number and difficult purification. The review stresses that no carrier class is universally superior; performance emerges from the specific peptide-carrier-method combination.

Three primary studies cited in the review illustrate the prediction-to-formulation gap with unusual clarity. In the MS2 virus-like particle system, insertion of HIV-derived peptides into the wild-type coat protein disrupted capsid assembly entirely, and manufacturability was restored only through a single-chain dimer scaffold. In hepatitis B core virus-like particles carrying Toxoplasma gondii epitopes, four epitope combinations assembled and were immunogenic, but only one construct, HBcΔH82, improved both mouse survival and brain cyst burden, showing that display geometry rather than immunogenicity scores controlled protection. Conversely, the bioinformatically selected peptide rOmp22, a candidate against Acinetobacter baumannii, required chitosan-coated PLGA encapsulation to achieve roughly 55 percent encapsulation, sustained release and stronger protection than the free peptide. Together these cases span assembly failure, context-dependent translation and formulation rescue at distinct points of the same pipeline.

On the computational side, the authors assess the current generation of epitope predictors with measured skepticism. Tools such as NetMHCpan-4.1 and NetMHCIIpan-4.0 use neural-network ensembles trained jointly on binding-affinity and eluted-ligand data, offering broad allele coverage but weaker support for rare alleles and no direct model of T-cell receptor recognition. MHCflurry 2.0 separates affinity, antigen processing and presentation, improving mechanistic attribution for class I predictions, while BepiPred-2.0 provides scalable but conformationally limited B-cell epitope screening. Protein language models and structure-prediction systems such as AlphaFold add transferable sequence and structural representations, yet their embeddings are not calibrated vaccine endpoints and can propagate biases from training data skewed toward well-studied pathogens and common HLA alleles. Train-test leakage from homologous sequences can further inflate reported performance metrics.

The review proposes that epitope prediction be treated strictly as an upstream prioritization step, filtered through a two-stage developability gate. Candidates should first be ranked immunologically, then screened against intrinsic sequence descriptors including molecular weight, net charge, isoelectric point, hydrophobicity, aromaticity, aggregation propensity and protease-sensitive motifs. Only candidates within a predefined developability domain should proceed to context-dependent testing in the intended formulation. The authors argue that this prevents a high epitope score from overriding insolubility, aggregation or interface incompatibility, and they map each peptide liability to orthogonal analytical confirmation and explicit go or no-go decisions.

Formulation-oriented machine learning shows genuine but bounded promise. Wang and colleagues trained a LightGBM model on 325 mRNA-lipid nanoparticle formulations to predict immunoglobulin G titers and experimentally confirmed the resulting formulation ranking. Sato and colleagues used XGBoost to identify ethanol fraction, buffer pH and total flow rate as determinants of particle size and encapsulation efficiency, then applied Bayesian optimization to target lipid nanoparticles of approximately 80 and 200 nanometers. Kumar and Ardekani curated 6,454 lipid nanoparticle formulations from 21 studies and found descriptor-based ensemble models, particularly balanced random forest and extra trees classifiers, most accurate for predicting activity and cell viability. These successes, the review notes, demonstrate prediction of bounded quality attributes within a single platform, not proof that one model can optimize peptide nanovaccines across carrier classes.

Controlled release and biodistribution modeling represent the frontier where computational ambition most clearly exceeds current evidence. Classical kinetic models such as Higuchi, Korsmeyer-Peppas and Weibull remain useful but rest on assumptions that real nanocarriers strain through simultaneous diffusion, swelling, erosion, peptide-matrix binding and burst release. Physics-informed neural networks, which embed governing differential equations into neural-network training, could in principle connect sparse release data with mechanistic transport assumptions and estimate parameters that are difficult to measure directly. The authors treat this cautiously: if the governing physics is oversimplified or parameters are not identifiable, a physics-informed network can create the appearance of rigor without improving prediction. Similarly, physiologically based pharmacokinetic models borrowed from soluble small molecules cannot capture nanovaccine disposition, which involves injection-site depots, lymphatic transport, time-varying protein coronas and separate fates for carrier, intact peptide, degradation fragments and adjuvant.

The review’s conclusions are operational rather than celebratory. Future progress, the authors argue, depends less on isolated prediction accuracy than on standardized datasets, external validation, uncertainty quantification, reproducible experiments and chemically interpretable models. They propose a three-tier agenda: harmonized experimental protocols reported in FAIR-compliant formats; multiscale model integration linking release, degradation, lymphatic transport and immune timing through modular mass balances; and systematic sharing of negative data, including failed syntheses, aggregation, low loading and non-protective immune responses, which define platform boundaries and counteract publication bias. Their minimum end-to-end reporting framework requires that candidates advance only after passing predefined gates for peptide developability, interface compatibility, process robustness, intact release, pathway-specific antigen presentation and separate disposition of carrier, peptide and adjuvant, with uncertainty propagated at every stage.

The broader message for the field is a disciplined reframing of what artificial intelligence can and cannot yet do. AI has genuinely accelerated the nomination of candidate epitopes, and machine learning has delivered experimentally validated optimizations for lipid nanoparticle formulations. But the translation of predicted epitopes into chemically viable delivery systems remains the bottleneck, and it will not be solved by better binding predictors alone. The highest-impact work, Garcia-Atutxa and Villanueva-Flores conclude, will be the studies that connect immunological prediction with formulation chemistry, release kinetics, biodistribution and validation in a way transparent enough to be tested and reproduced. Until peptide solubility, aggregation, antigen orientation, corona-mediated masking and intact-antigen release are jointly represented in externally validated models, the chemical design of multi-epitope vaccine nanocarriers will remain a problem where chemistry, not computation, sets the pace.

Subject of Research: AI-supported chemical design of multi-epitope vaccine nanocarriers

Article Title: Artificial intelligence can support the chemical design of multiepitope vaccine nanocarriers

Article References: Garcia-Atutxa, I., & Villanueva-Flores, F. (2026). Artificial intelligence can support the chemical design of multiepitope vaccine nanocarriers. Discover Chemistry, 3(1), Article 469. https://doi.org/10.1007/s44371-026-00929-6

Image Credits: AI Generated

DOI: 10.1007/s44371-026-00929-6

Keywords: artificial intelligence, multi-epitope vaccines, nanocarriers, epitope prediction, peptide vaccines, lipid nanoparticles, controlled release, physics-informed neural networks, PBPK modeling, extracellular vesicles, machine learning, vaccine formulation

The review lays out five interconnected processes that govern the fate of mRNA-lipid nanoparticles in the body: route-dependent transport and initial tissue distribution, interactions with blood proteins that form the so-called protein corona, receptor-mediated cellular uptake and organ tropism, endosomal escape and intracellular trafficking, and finally metabolic degradation and immune-mediated clearance. These pathways act sequentially, meaning that a change at any stage ripples through the entire pharmacokinetic profile. Crucially, the authors emphasize that the localization of LNP-associated lipids, intact mRNA, and encoded protein follows distinct spatial and temporal patterns, so detecting one component cannot be assumed to demonstrate the presence or activity of another. Apparent contradictions between studies using radiolabeling, molecular quantification, imaging, or protein-based readouts often reflect differences in the biological endpoint measured rather than genuinely conflicting results.

Administration route emerges as one of the most powerful determinants of biodistribution. Intramuscular and subcutaneous injection favor retention at the injection site followed by drainage to regional lymph nodes, with peak tissue concentrations typically occurring within two to eight hours; subcutaneous delivery produces similar patterns but with slower kinetics and prolonged persistence. Intravenous administration bypasses local lymphatics entirely, producing immediate systemic exposure with predominant accumulation in the liver. Respiratory delivery via intranasal, intratracheal, or nebulized routes deposits nanoparticles throughout the airway epithelium according to aerodynamic diameter, bypassing first-pass hepatic clearance and enabling mucosal immunity characterized by secretory IgA and tissue-resident memory T cells. The review notes, however, that intranasal delivery is not straightforward, as animal studies suggest some formulations may be reactogenic.

Perhaps the most striking examples of route-driven tropism come from less conventional approaches. Intraperitoneal administration of ionizable LNPs has been shown to deliver mRNA to pancreatic beta cells through an indirect mechanism in which peritoneal macrophages internalize the particles before transferring functional mRNA to pancreatic islets, largely avoiding the hepatic dominance seen after intravenous injection. More recently, researchers identified a pancreas-selective delivery mechanism based on organ capsule filtration, in which arginine-histidine-modified LNPs undergo protein-induced size enlargement after systemic administration, allowing preferential accumulation within the pancreatic capsule. This platform achieved efficient delivery of Cas9 mRNA and therapeutic cytokine mRNAs in rodents and non-human primates, demonstrating that particle size, protein association, organ anatomy, and administration route act synergistically to determine organ selectivity rather than lipid chemistry alone.

Intrathecal injection, which introduces nanoparticles directly into cerebrospinal fluid, bypasses the blood-brain barrier and enables widespread distribution across the brain and spinal cord. Biodegradable brain-targeting LNPs have achieved reporter expression in the cortex, hippocampus, and cerebellum, transfecting both neurons and astrocytes, while related work delivered CRISPR genome-editing components throughout the central nervous system with limited peripheral exposure. Notably, formulation chemistry remained a critical determinant even within this route, with ionizable lipid structure influencing both cellular uptake and regional distribution. Subretinal injection likewise places formulations directly adjacent to the retinal pigment epithelium and photoreceptors, enabling robust local transfection for inherited retinal disorders while minimizing systemic exposure. Even the choice of limb for sequential vaccine doses matters: murine studies showed enhanced early germinal center B-cell responses and higher-affinity antibodies following ipsilateral boosting, although these differences diminished over time, with comparable long-term protection regardless of injection side.

Once nanoparticles reach biological fluids, they rapidly adsorb plasma proteins to form a protein corona that replaces the synthetic surface as the interface recognized by cells. Apolipoprotein E is the best-characterized component, promoting uptake through low-density lipoprotein receptor family members and explaining the intrinsic hepatic tropism of many formulations. But quantitative proteomics reveals a far more complex picture: the corona is a dynamic mixture of high-density lipoproteins, albumin, vitronectin, complement proteins, immunoglobulins, fibrinogen, prothrombin, C-reactive protein, and alpha-2-macroglobulin, whose composition depends on lipid chemistry, administration route, and host physiology. HDL has been associated with enhanced uptake and transfection, albumin prolongs circulation by improving colloidal stability, and complement proteins and immunoglobulins accelerate clearance by the mononuclear phagocyte system, potentially contributing to complement activation-related pseudoallergy and anti-PEG immune responses during repeated dosing. Formulations with nearly identical physical properties can therefore exhibit markedly different tissue tropism because organ selectivity is determined by the integrated corona composition rather than any single protein.

Even successful cellular uptake does not guarantee productive delivery. The review identifies endosomal escape as the principal intracellular bottleneck, with only a small fraction of internalized nanoparticles releasing intact mRNA into the cytoplasm before lysosomal degradation. Ionizable lipids facilitate escape by becoming protonated within acidifying endosomes, destabilizing the endosomal membrane, but recent in vivo LysoTag and lysosomal barcoding studies have provided the first quantitative measurements of escape kinetics, showing that relatively small improvements in escape efficiency produce disproportionately large increases in protein expression. This finding shifts the focus of formulation optimization from maximizing tissue uptake toward improving intracellular trafficking and cytosolic release. Once released, mRNA undergoes ribosomal translation before rapid degradation by endogenous ribonucleases, while the encoded protein frequently persists considerably longer, creating a temporal dissociation between nanoparticle localization, mRNA persistence, and biological activity that explains why different analytical methods yield different pharmacokinetic profiles for the same formulation.

The human evidence, though less comprehensive, broadly supports the preclinical picture. After intramuscular vaccination, vaccine-derived mRNA and encoded spike protein can be transiently detected in plasma within the first few hours, consistent with systemic dispersion of a small fraction of the dose. Lymph node biopsies confirm local expression in germinal-center B cells and dendritic cells, correlating with strong antibody and T-cell responses. One study using RT-qPCR detected vaccine mRNA in lymph nodes, liver, spleen, and myocardium from recently vaccinated individuals, with persistence up to 30 days in some axillary lymph nodes, but the authors of that work concluded that observed cardiac changes likely reflected pre-existing conditions rather than direct vaccine effects. Quantitative imaging in non-human primates has found no evidence of significant mRNA or lipid presence in brain tissue, and examination of gonadal tissues reveals only trace, fragmented RNA species without functional translation. The placenta, meanwhile, appears to be selectively targetable: optimized ionizable LNPs efficiently transfected placental trophoblasts with minimal fetal transfer in pregnant mice and ex vivo human placental perfusion systems, opening possibilities for treating pre-eclampsia and fetal growth restriction.

Clearance kinetics are now better characterized as well. In most models, mRNA persists for less than 48 hours before enzymatic breakdown, while lipid components are metabolized more slowly through hepatic and biliary routes; PEG-lipids and cholesterol derivatives may remain for several days, particularly within macrophages, but are gradually cleared without significant histopathological changes. Mass spectrometry studies of a biodegradable ionizable lipid showed complete hepatobiliary and renal clearance by 168 hours despite rapid systemic distribution of lipid-derived radioactivity. Repeated exposure to PEG-containing formulations may induce anti-PEG antibodies and complement activation, leading to accelerated blood clearance and altered biodistribution after subsequent administrations, a phenomenon driving the development of PEG-free alternatives. Cross-species comparisons reveal that rodents tend to overestimate hepatic accumulation relative to humans, while non-human primates better replicate human plasma kinetics but differ in immune activation thresholds, prompting growing interest in organ-on-chip systems, organoids, and humanized models to improve translational accuracy.

Looking forward, the review argues that rational, data-driven engineering is replacing empirical formulation. Selective organ targeting (SORT) lipids redirect expression from liver to spleen, lung, or bone marrow in a charge-dependent manner; biodegradable ionizable lipids with ester or ketal linkages reduce long-term retention while maintaining delivery efficiency; and barcode-based screening now enables high-throughput evaluation of hundreds of formulations within a single animal. Artificial intelligence and machine learning are accelerating lipid design, with graph neural networks predicting blood-brain barrier permeability and generative models proposing novel ionizable structures, though the authors caution that most models are trained on small, heterogeneous datasets and require rigorous external validation. Physiologically based pharmacokinetic and quantitative systems pharmacology models are increasingly integrating these diverse data streams to predict tissue exposure and protein expression across species. The overarching message is clear: biodistribution cannot be defined by a single analytical signal, and progress toward clinically predictable mRNA therapeutics will depend on determining not only where each formulation distributes, but which cells receive intact cargo, whether functional expression occurs, and how long its effects persist.

Subject of Research: Biodistribution and pharmacokinetics of mRNA lipid nanoparticle delivery systems

Article Title: Mechanisms and determinants of mRNA Lipid Nanoparticle biodistribution: From pharmacokinetics to clinical translation

Article References: Alhareth, Z., & López‐Camacho, C. (2026). Mechanisms and determinants of mRNA Lipid Nanoparticle biodistribution: From pharmacokinetics to clinical translation. Bioengineering & Translational Medicine, Article e70183. https://doi.org/10.1002/btm2.70183

Image Credits: AI Generated

DOI: 10.1002/btm2.70183

Keywords: mRNA therapeutics, lipid nanoparticles, biodistribution, protein corona, endosomal escape, pharmacokinetics, drug delivery, SORT lipids, intrathecal delivery, placental transfer, PBPK modeling, clinical translation

The drug at the center of the study, roxithromycin, is a macrolide antibiotic related to erythromycin, used against respiratory tract and other bacterial infections. Like many macrolides, it is given orally and its absorption into the bloodstream depends on a chain of events that includes dissolution of the tablet in the gastrointestinal tract, permeation across the intestinal wall, and distribution into tissues. Any change in the formulation, the manufacturing process, or even the excipients used to bind and coat the tablet can in principle alter how quickly and completely the drug reaches the circulation. That is why regulators require bioequivalence testing whenever a generic version or a modified formulation is brought to market. The standard metrics are the area under the concentration-time curve, which reflects total drug exposure, the maximum plasma concentration, which reflects peak exposure, and the time at which that maximum occurs. If the ratios of these parameters between test and reference products fall within a ninety percent confidence interval of eighty to one hundred twenty-five percent, the products are considered bioequivalent.

Conducting such trials in humans is expensive, slow, and ethically burdensome, since volunteers are exposed to a drug without any personal therapeutic benefit. This is where physiologically based pharmacokinetic modeling, or PBPK, enters the picture. Unlike simpler empirical models that treat the body as a set of mathematical compartments with fitted parameters, a PBPK model attempts to represent the actual anatomy and physiology of the body: organs and tissues are connected by blood flows, and drug movement between them is governed by physicochemical properties such as solubility, permeability, lipophilicity, and protein binding, together with physiological parameters like organ volumes, blood flow rates, and enzyme abundances. Because the underlying structure reflects real biology, a well-constructed PBPK model can be adapted to different populations, doses, and even formulations simply by swapping in the relevant inputs, making it a powerful tool for predicting drug behavior in scenarios that have never been tested clinically.

The research team, led by Danfeng Yu, Qin Li, and colleagues including corresponding authors Jiyu Chen and Yan He, assembled roxithromycin’s physicochemical, in vitro, and in vivo parameters from established databases and the published literature, and constructed an adult PBPK model using the open-source software platform PK-Sim. The initial model was developed and calibrated against clinical pharmacokinetic data from a 150 milligram dose of the reference formulation. The critical test of any model, however, is not how well it fits the data used to build it, but how well it predicts data it has never seen. To address this, the researchers performed a comparative evaluation using clinical data from the test formulation, checking whether the model could reproduce concentration-time profiles for a product whose behavior had not informed the model’s construction.

To quantify the agreement between prediction and observation, the team applied the mean fold error approach, a standard metric in pharmacokinetic modeling that measures the ratio of predicted to observed values. For the three key parameters, the area under the curve from zero to the last measurable time point, the maximum concentration, and the time to maximum concentration, more than ninety percent of the mean fold error ratios fell within the range of 0.5 to 2. This means that for the overwhelming majority of comparisons, the model’s predictions were within a factor of two of the measured values, and in most cases considerably closer. With the model thus validated, the researchers imported formulation-specific in vitro dissolution data for both products, allowing the simulations to distinguish between the two tablets based on how rapidly each released its drug payload in the digestive tract.

The virtual bioequivalence trial itself was conducted in a simulated cohort of ninety-nine East Asian subjects, a population chosen to match the demographic context in which the formulations are intended to be used. Virtual subjects in PBPK platforms are not identical clones; the software samples from realistic distributions of physiological parameters such as body weight, gastric emptying time, and intestinal transit, so that the simulated population exhibits the same variability that a real trial population would. Each virtual subject received both formulations, and the resulting pharmacokinetic profiles were analyzed exactly as they would be in a conventional bioequivalence study. The outcome was unambiguous: the ninety percent confidence intervals for the geometric mean ratios of the primary pharmacokinetic parameters all fell within the regulatory acceptance range of eighty to one hundred twenty-five percent, indicating that the two formulations would be judged bioequivalent.

The implications of this result extend well beyond a single antibiotic. Bioequivalence studies are among the most numerous clinical trials conducted worldwide, required for every generic approval and for many post-approval manufacturing changes. If validated PBPK models can reliably substitute for some of these trials, the savings in cost, time, and human exposure could be substantial. Regulators, including the United States Food and Drug Administration and the European Medicines Agency, have already issued guidance on when modeling and simulation can support or replace certain clinical studies, and virtual bioequivalence is an active area of regulatory science. The present study adds to a growing body of evidence that, for drugs with well-characterized pharmacokinetics and formulations whose dissolution behavior can be measured reliably in vitro, a computer can indeed stand in for the clinic.

The study also illustrates the methodological discipline that makes such virtual trials credible. The researchers did not simply assume the model was correct; they built it from one dataset, tested it against an independent dataset from a different formulation, and reported a formal quantitative measure of predictive performance before trusting it for the bioequivalence question. The work was registered retrospectively at ClinicalTrials.gov under number NCT06798051, the underlying clinical study was approved by the Medical Ethics Committee of the Affiliated Hospital of Guizhou Medical University, and all participants in the source trials provided informed consent. The modeling was funded by the Science and Technology Planning Project of Guizhou Province and related institutional programs, and the authors declared no competing interests.

Certain caveats remain. A virtual trial is only as good as its underlying model, and PBPK predictions for drugs with complex metabolism, food effects, or narrow therapeutic windows demand even more extensive validation than was possible here. The ninety-nine subject cohort, while adequate for this comparison, is a simulation of a specific population rather than a demonstration across all patient groups. Nevertheless, the study offers a concrete template for how virtual bioequivalence can be conducted and documented: assemble the drug’s parameters, build and validate the model against real data, import formulation-specific dissolution profiles, and simulate a demographically realistic cohort under the same statistical criteria that govern human studies. As more models of this kind accumulate and are independently verified, the day when a generic drug can win approval on the strength of a computer trial, with human studies reserved for cases the models cannot yet handle, moves measurably closer.

Subject of Research: Virtual bioequivalence assessment of two roxithromycin formulations using a physiologically based pharmacokinetic model

Article Title: Virtual bioequivalence assessment for two roxithromycin formulations using a physiologically based pharmacokinetic model

Article References: Yu, D., Li, Q., Zeng, L., Liao, M., Lu, X., Chen, J., & He, Y. (2026). Virtual bioequivalence assessment for two roxithromycin formulations using a physiologically based pharmacokinetic model. BMC Pharmacology and Toxicology. https://doi.org/10.1186/s40360-026-01244-7

Image Credits: AI Generated

DOI: 10.1186/s40360-026-01244-7

Keywords: roxithromycin, PBPK modeling, virtual bioequivalence, PK-Sim, pharmacokinetics, generic drugs, drug dissolution, macrolide antibiotics, clinical trial simulation, regulatory science, East Asian population, BMC Pharmacology and Toxicology

The technical foundation of the dispute lies in two biomarkers. Ethylene is a simple gaseous molecule that humans both produce internally and inhale from the environment, including from combustion sources and ripening fruit. Inside the body, a fraction of inhaled ethylene is converted by cytochrome P450 enzymes into ethylene oxide, a more reactive epoxide that can bind to hemoglobin and DNA. The specific biomarker used to track this chemistry is N-(2-hydroxyethyl)valine, abbreviated HEV, an adduct formed when ethylene oxide reacts with the N-terminal valine residue of hemoglobin proteins. Because red blood cells circulate for roughly four months, HEV provides an integrated record of ethylene oxide exposure over time. Lin and colleagues compiled a database of breath ethylene measurements and, using physiologically based pharmacokinetic modeling, sought to estimate how much endogenous ethylene oxide production contributes to total HEV burdens in the general population.

What Kirman and Bus highlight in their correspondence is a fundamental analytical problem embedded in that database. Breath ethylene concentrations reported across the literature are extraordinarily variable, and the variability appears to track with the measurement technology rather than with genuine differences between study populations. Studies employing laser-based photoacoustic spectrometry, a technique that detects gas absorption of laser light, reported mean ethylene concentrations of roughly 61 parts per billion with a standard deviation of 130 ppb, an enormous spread. By contrast, studies relying on gas chromatography reported a mean of approximately 20 ppb with a standard deviation of 16 ppb, far tighter and consistently lower. Even within the subset of studies that Lin and colleagues designated as high confidence, mean ethylene levels hovered around just 0.5 ppb, orders of magnitude below the values coming from the laser-based literature.

This is not a new concern, and Kirman and Bus invoke a striking historical precedent. More than two decades ago, Berkelmans and colleagues, working with online laser photoacoustic detection, observed that endogenous ethylene production rates derived from laser-based studies were significantly lower than published values based on gas chromatography, an inconsistency they attributed to fundamental differences between the two analytical approaches. When Berkelmans’s team attempted to fit a physiologically based pharmacokinetic model to their laser-derived data, they found they had to modify measured physiological and biochemical parameters to make the model work, indicating that the laser measurements were inconsistent with the well-established understanding of ethylene toxicokinetics built on gas chromatography. Kirman and Bus argue that this methodological fault line has not been resolved, and that Lin and colleagues’ database largely inherits the problem rather than correcting it.

The PBPK modeling itself is the second pillar of the critique. Physiologically based pharmacokinetic models are mathematical descriptions of how a chemical moves through the body, incorporating blood flow, tissue partitioning, metabolic rates, and ventilation. The model used by Lin and colleagues traces back to the foundational work of Filser and Klein, who developed a toxicokinetic model for inhaled ethylene and ethylene oxide across mouse, rat, and human species, and to earlier work by Csanády and colleagues, who modeled the formation of 2-hydroxyethyl adducts with hemoglobin and DNA from both exogenous and endogenous sources. Running the model with the breath ethylene data, Lin’s team concluded that endogenous production pathways account for less than 20 percent of total HEV in nonsmokers, leaving more than 80 percent unexplained. Kirman and Bus point out that this large unexplained fraction is itself a signal that something in the exposure reconstruction may be missing, whether it be analytical bias in breath measurements, unrecognized internal sources, or contributions from pathways not captured by the model.

Exogenous exposure, meaning ethylene and ethylene oxide inhaled from outside sources such as ambient and indoor air, is the other candidate explanation for the unaccounted adduct burden. Here, Kirman and Bus note that available air monitoring data, including large-scale studies such as Health Canada’s Windsor Exposure Assessment Study, indicate that environmental contributions to total ethylene oxide body burden are expected to be small for the general population. This conclusion aligns with their own prior publications, including a 2021 comprehensive review characterizing total ethylene oxide exposure from endogenous and exogenous pathways and a 2025 assessment of background exposures in the United States that questioned theoretical health risks for populations living near industrial sources. The implication is that the apparent gap between modeled endogenous production and measured HEV cannot simply be closed by invoking ambient air pollution, contrary to some popular narratives about ethylene oxide risk.

The broader context makes this technical disagreement consequential beyond the laboratory. Ethylene oxide is used industrially to sterilize roughly half of all medical devices in the United States, and the Environmental Protection Agency’s recent regulatory actions targeting sterilization facilities have relied on risk estimates derived from inhalation unit risk values. If a substantial fraction of the population’s hemoglobin adduct burden derives from endogenous biology rather than industrial emissions, then the margin between background internal exposure and levels associated with elevated cancer risk is narrower than many assume, and any risk assessment that fails to account for endogenous background risks mischaracterizing the true incremental danger of industrial emissions. Kirman and Bus have argued elsewhere that recognizing endogenous ethylene oxide formation is essential for a coherent risk management framework, since regulatory limits that ignore the body’s own production of the chemical cannot meaningfully protect public health in proportion to actual incremental exposure.

They also point to emerging molecular dosimetry work that helps interpret HEV data. Recent studies by Liu and colleagues on hemoglobin adduct formation in mice exposed to ethylene oxide, and Lin’s own companion work integrating PBPK modeling with tobacco biomarkers to interpret HEV levels in the U.S. population, demonstrate that smoking contributes substantially to measured adduct burdens, since tobacco smoke contains both ethylene and ethylene oxide directly. In nonsmokers, however, the origin of the remaining adduct burden remains contested. Classical biomonitoring studies from the 1990s and 2000s, including molecular dosimetry work by Walker, Fennell, Upton, and Swenberg in rodents exposed to ethylene oxide, established the dose-response framework that still underpins modern interpretation, and those studies consistently emphasized the importance of distinguishing background from incremental exposure when translating adduct measurements into cancer risk.

Neither the correspondence nor the original analysis settles the question of how much endogenous metabolism contributes to human ethylene oxide body burdens, but the exchange underscores a lesson that resonates across exposure science: the quality of a risk assessment is bounded by the quality of the exposure data fed into it. When two analytical technologies applied to the same biological matrix differ by more than an order of magnitude, as laser-based photoacoustic spectrometry and gas chromatography evidently do for breath ethylene, downstream models, no matter how sophisticated, inherit that uncertainty. Kirman and Bus contend that progress requires reconciling the methodological discrepancy, validating breath ethylene measurements against gas-chromatographic benchmarks, and ensuring that PBPK models are fit to data consistent with established toxicokinetics. Until that reconciliation occurs, estimates of how much ethylene oxide in the average person’s blood comes from industry versus biology will remain contested, and the regulatory debate over one of the world’s most widely used sterilants will continue to be fought on contested ground.

Subject of Research: Human exposure to ethylene and endogenous ethylene oxide assessed through breath biomarkers and pharmacokinetic modeling

Article Title: Regarding ethylene exposure and endogenous ethylene oxide levels in humans (Lin et al., 2025)

Article References: Regarding ethylene exposure and endogenous ethylene oxide levels in humans (Lin et al., 2025). (n.d.). https://doi.org/10.1038/s41370-026-00967-3

Image Credits: AI Generated

DOI: 10.1038/s41370-026-00967-3

Keywords: ethylene, ethylene oxide, hemoglobin adducts, PBPK modeling, breath analysis, biomonitoring, exposure science, risk assessment, photoacoustic spectrometry, gas chromatography, environmental epidemiology, Regarding