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 study enrolled 23 patients with muscle-invasive bladder cancer, stages T2 to T3b, who were undergoing radical radiotherapy delivered over four weeks alongside weekly gemcitabine chemotherapy. Blood samples were collected before the first radiation fraction and then every week until treatment ended. The researchers also retrieved each patient’s diagnostic biopsy and measured the activity of a validated 24-gene hypoxia signature, generating a hypoxia score for every tumour. Patients were then split into high and low hypoxia groups based on the median score. This design allowed the team to ask a deceptively simple question: do the proteins circulating in the blood of patients with oxygen-starved tumours look different from those in patients with well-oxygenated tumours, and do they change as treatment progresses?

Answering that question required overcoming a long-standing technical obstacle. Plasma is dominated by a small number of extremely abundant proteins, such as albumin and immunoglobulins, which mask the low-abundance molecules that often carry the most informative biological signals. Previous attempts to find circulating hypoxia markers relied on targeted antibody tests of single proteins, and several promising candidates, including osteopontin in head and neck cancer, ultimately failed prospective validation. The Manchester team took a different approach, using engineered liposomal nanoparticles roughly 105 nanometres in diameter as molecular fishing hooks. When incubated with plasma, these particles attract a shell of bound proteins, known as a protein corona, which preferentially captures lower-abundance molecules. The corona proteins were then purified, digested and analysed by high-resolution liquid chromatography tandem mass spectrometry.

The results were striking. Across all time points, the analysis identified 816 plasma proteins, of which 115 showed significant differences between patients with high and low tumour hypoxia scores, defined as a fold change greater than 1.5 and a p-value below 0.05. The number of differentially abundant proteins fluctuated across the treatment timeline: 24 before radiotherapy, 46 after one week, 40 after two weeks, 14 after three weeks and 37 at the end of treatment. Remarkably, roughly 45 percent of the differentially abundant proteins were immunoglobulins or complement cascade proteins, pointing to a previously underappreciated connection between tumour hypoxia and the humoral immune system, the antibody-producing arm of immunity.

From the full set of hypoxia-associated proteins, seven candidates stood out because they correlated significantly with tumour hypoxia scores at two or more time points and maintained a consistent direction of correlation throughout treatment. Five of them, IGKV3, IGLV2-18, VL_4 and VL_7 alongside the extracellular matrix protein FN1, were consistently elevated in patients with hypoxic tumours, while IGLV2-14 and CAMP were consistently reduced. Principal component analysis and K-means clustering showed that this seven-protein signature cleanly separated high-hypoxia from low-hypoxia patients, and, crucially, that the separation held steady at every time point during radiotherapy. That temporal stability is exactly what a clinically useful monitoring biomarker requires, since a signal that drifts unpredictably during treatment cannot guide real-time decisions.

Because only two of the seven proteins, FN1 and CAMP, had matching RNA expression data in public cohorts, the team distilled the signature into a two-gene version and validated it retrospectively in three independent bladder cancer cohorts. In the TCGA-BLCA cohort of 404 patients treated with cystectomy, high hypoxia scores predicted poorer ten-year overall survival with a hazard ratio of 1.54. In the BC2001 radiotherapy cohort of 313 patients, the hazard ratio was 1.42. A meta-analysis combining TCGA-BLCA, BC2001 and the BCON trial confirmed the association with poor prognosis across 150 patients, with a hazard ratio of 1.48, independently of age, sex, stage and treatment received. The signature also remained prognostic in patients treated with and without concurrent chemotherapy, suggesting its signal is not an artefact of a particular drug regimen.

The most clinically provocative finding came from the BCON phase III trial, which tested whether breathing carbogen, a mixture of oxygen and carbon dioxide, together with nicotinamide, a drug that improves blood flow, could sensitise hypoxic tumours to radiation. That trial previously showed a 17 percent improvement in five-year overall survival with hypoxia modification. When the new two-gene signature was applied to BCON samples, patients with high hypoxia scores had poorer survival in the radiotherapy-only arm but lost that disadvantage when treated with carbogen and nicotinamide, with a hazard ratio of 0.60 favouring hypoxia modification. In other words, the blood-based signature did not merely forecast prognosis; it predicted who would actually benefit from targeting hypoxia, mirroring the performance of the original 24-gene tumour biopsy signature but requiring only a blood draw.

Beyond the biomarker itself, the study revealed a dynamic biological story. Weighted correlation analysis grouped the 115 hypoxia-associated proteins into five co-expression clusters, each peaking at a successive point in the treatment timeline: before radiotherapy, and at weeks one through four. Every cluster was enriched for B cell response and immunoglobulin production pathways, while the pre-treatment cluster was uniquely enriched for extracellular matrix remodelling, and later clusters for coagulation, fibrin formation and metabolic processes. Comparing expression patterns with fold-change dynamics suggested that radiotherapy primarily drove the timing of cluster activation, whereas tumour hypoxia shaped the baseline abundance and magnitude of the response. The researchers propose that tumour re-oxygenation during radiotherapy, which generates reactive oxygen species and activates HIF-1 signalling, may explain the declining protein levels seen in later weeks of treatment.

The immunological implications extend well beyond bladder cancer. Prior laboratory work has shown that hypoxia-inducible factors directly regulate B cell maturation and immunoglobulin class switching, and mouse studies have demonstrated that germinal centre hypoxia can impair humoral anti-tumour responses. This study provides the first clinical, plasma-level evidence that such immune modulation occurs in patients, with hypoxic tumours showing a shifted immunoglobulin profile that evolves over the course of radiotherapy. The authors caution that the discovery cohort was small, included only two female patients, and lacked direct protein-level outcome validation, so prospective validation at the protein level is the essential next step. Even so, the work establishes nanoparticle-enabled proteomics as a powerful discovery platform for circulating cancer biomarkers and suggests a future in which a routine blood test could tell oncologists, week by week, whether a patient’s tumour is suffocating, whether radiation is working, and whether hypoxia-targeting drugs should be added to the plan.

Subject of Research: A plasma protein signature for detecting tumour hypoxia and predicting benefit from hypoxia-modifying radiotherapy in bladder cancer

Article Title: A plasma derived hypoxia signature predicts benefit from hypoxia-modifying radiotherapy and reveals dynamic humoral immune modulation in bladder cancer

Article References: Guerrero Quiles, C., Abumanhal-Masarweh, H., G. Abalos, J., Lodhi, T., Sanchez-Martinez, D., Reeves, K., James, N. D., Hall, E., Huddart, R. A., Porta, N., Hoskin, P., Biolatti, L. V., Hadjidemetriou, M., West, C. M., & Choudhury, A. (2026). A plasma derived hypoxia signature predicts benefit from hypoxia-modifying radiotherapy and reveals dynamic humoral immune modulation in bladder cancer. British Journal of Cancer. https://doi.org/10.1038/s41416-026-03603-x

Image Credits: AI Generated

DOI: 10.1038/s41416-026-03603-x

Keywords: bladder cancer, tumour hypoxia, radiotherapy, biomarker, proteomics, nanoparticles, protein corona, mass spectrometry, humoral immunity, BCON trial, carbogen and nicotinamide, liquid biopsy

The new particles, described in Bioengineering & Translational Medicine, were created by researchers at the University of Michigan who wanted to solve a persistent problem in corticosteroid therapy. Dexamethasone has shown real promise in recent clinical trials for ARDS, improving outcomes where older steroids failed. Yet those benefits typically require high doses over prolonged regimens, and they appear limited to certain patient subgroups, with some patients actually faring worse after treatment. Free dexamethasone circulating through the bloodstream also triggers well-documented systemic effects, including neutrophilia, an abnormal rise in blood neutrophil counts, and lymphopenia, the depletion of lymphocytes. The Michigan team reasoned that if the drug could be packaged so that it preferentially reached neutrophils themselves, the therapeutic effect would be concentrated at the source of tissue damage while the rest of the body would be spared.

The fabrication strategy is elegantly simple. The researchers added an iron sulfate solution dropwise into a stirred solution of dexamethasone phosphate, triggering a nucleation-and-growth process in which amorphous iron-phosphate particles form with dexamethasone incorporated throughout the particle matrix. The resulting nanoparticles, dubbed Dex NP, measured approximately 75 nanometers in diameter with a zeta potential between negative 10 and negative 20 millivolts. Scanning electron microscopy confirmed uniform particle morphology, and energy-dispersive X-ray spectroscopy mapping showed that iron and dexamethasone phosphate were evenly distributed across each particle rather than segregated into layers. There was just one problem: when these first-generation particles were incubated with human neutrophils in whole blood, the cells ignored them completely. No internalization occurred, which meant the drug delivery concept would fail before it even left the bench.

The breakthrough came from reconsidering the particle surface. Nanoparticle behavior in blood is governed largely by the protein corona, the layer of plasma proteins that adsorbs onto the particle within seconds of exposure. Albumin, the most abundant plasma protein, is a known dysopsonin, meaning its presence on a particle surface actively discourages phagocytes from engulfing it. The team hypothesized that the ionic iron within the particle matrix could be exploited to change the surface chemistry. Because iron oxide can form at temperatures as low as 200 degrees Celsius in open air, the researchers heated their particles to 200, 215, and 230 degrees and monitored how the protein corona changed. Only temperatures above 215 degrees altered protein adsorption, so 215 degrees became the standard oxidation step, transforming Dex NP into DexOx NP.

Characterization confirmed the transformation worked exactly as intended. X-ray photoelectron spectroscopy of the top 10 nanometers of the particle surface revealed a shift from ferrous to ferric iron, consistent with iron oxide formation, while Fourier transform infrared spectroscopy showed that the characteristic peaks of dexamethasone phosphate survived the heating intact. This mattered because dexamethasone begins to thermally degrade only above 200 degrees Celsius, with degradation onset at 256 degrees under controlled heating rates, and the brief oxidation window used here was measured in minutes rather than the hour-long exposures that have degraded the drug in prior polymer studies. Release assays in phosphate-buffered saline at physiological temperature showed that dexamethasone emerging from the particles remained as therapeutically active as the free drug, confirming the payload was not compromised during manufacturing.

The biological consequences of oxidation were striking. When FITC-labeled particles were incubated in whole human blood and mouse blood, flow cytometry showed a distinct fluorescence shift in the neutrophil population for DexOx NP but not for the unoxidized particles, and confocal microscopy visualized the oxidized particles sitting inside neutrophils. Protein corona analysis explained why: total adsorbed protein dropped after oxidation, with marked reductions in the albumin and transferrin bands at roughly 60 and 80 kilodaltons, proteins known to adsorb poorly to iron oxide. Meanwhile, a slight increase in bands between 150 and 200 kilodaltons suggested immunoglobulins, which activate complement and drive phagocytosis, had taken their place. Less albumin blocking the surface and more immunoglobulin flagging the particle combined to make DexOx NP irresistible to neutrophils.

Safety testing came next, and the particles passed cleanly. DexOx NP caused no hemolysis of red blood cells, did not activate platelets, and left unactivated neutrophils untouched, indicating the particles could be infused systemically without triggering the very inflammation they were designed to treat. In activated neutrophils, however, the therapeutic effect was dramatic. When lipopolysaccharide was used to simulate bacterial activation, DexOx NP preserved L-selectin, an adhesion molecule shed during neutrophil activation, reducing shedding by 68 percent. This outperformed both free dexamethasone phosphate and the poorly internalized Dex NP, neither of which changed L-selectin expression at all. The effect depended on the glucocorticoid receptor, since mifepristone, a receptor antagonist, blunted the benefit, and control experiments with plain iron oxide particles confirmed the activity came from the dexamethasone rather than the metal.

The particles also tamed NETosis, the explosive process by which activated neutrophils expel DNA and intracellular contents, which drives tissue damage in ARDS. In neutrophils stimulated with phorbol 12-myristate 13-acetate, DexOx NP reduced total NET formation by 21 percent over five hours, while free dexamethasone phosphate, unoxidized particles, and cargo-free polystyrene particles had no effect. The mechanism likely involves inhibition of NADPH oxidase, the enzyme complex that initiates NET formation and whose p47phox subunit is a known corticosteroid target. Because internalized particles release dexamethasone directly into the cytosol where the glucocorticoid receptor resides, the team suggests the intracellular delivery route accelerates and amplifies the drug’s action compared with diffusion of free steroid from the extracellular fluid.

The decisive test came in a mouse model of acute lung injury, where lipopolysaccharide was instilled into the airways and treatments were injected into the tail vein one hour later. Both DexOx NP and free dexamethasone significantly reduced immune cell infiltration into the lungs, cutting total bronchoalveolar lavage cells by 37 and 34 percent respectively and neutrophil counts by 41 and 39 percent. Both treatments lowered the inflammatory cytokines IL-6 and TNF-alpha, and DexOx NP significantly reduced KC, a chemokine that recruits neutrophils to inflamed tissue. Cargo-free polystyrene particles, which neutrophils also engulf, did nothing, proving the benefit came from the drug rather than from particle diversion alone. Crucially, the side-effect profiles diverged sharply: free dexamethasone raised blood neutrophil counts by 57 percent and increased the neutrophil-to-lymphocyte ratio by 64 percent, classic signs of systemic steroid exposure, while DexOx NP produced neither effect. Liver enzymes, leukocyte counts, and body weight remained normal in healthy mice given the particles.

The findings point toward a broader strategy for taming acute inflammation without the blunt instrument of systemic steroids. Because the particles are composed of iron and the drug itself, with no exogenous polymer carrier, they avoid the stability, reproducibility, and loading problems that plague conventional formulations such as liposomes and PLGA particles. The researchers, who have filed a patent on composite drug particles, note that the approach could extend beyond dexamethasone to other corticosteroids and inflammatory diseases driven by neutrophil dysregulation. For a condition like ARDS, where clinicians have long struggled to harness steroid power without immunological collateral damage, a nanoparticle that speaks directly to the immune system’s most volatile cells represents a meaningful step toward precision anti-inflammatory medicine.

Subject of Research: Targeted dexamethasone nanoparticles that modulate neutrophil activity to treat acute neutrophilic inflammation and acute lung injury

Article Title: Iron‐dexamethasone nanoparticles mitigate acute neutrophilic inflammation

Article References: Felder, M. L., Guevara, M. V., Kupor, D., & Eniola‐Adefeso, O. (2026). Iron‐dexamethasone nanoparticles mitigate acute neutrophilic inflammation. Bioengineering & Translational Medicine, Article e70173. https://doi.org/10.1002/btm2.70173

Image Credits: AI Generated

DOI: 10.1002/btm2.70173

Keywords: neutrophils, dexamethasone, nanoparticles, acute lung injury, ARDS, drug delivery, inflammation, NETosis, protein corona, iron oxide, corticosteroids, immunomodulation

The stakes are considerable. Drug development remains notoriously slow and costly, with the vast majority of clinical candidates ultimately failing, and nanoparticle therapeutics have struggled more than most to cross the valley between promising preclinical results and approved medicines. Nanoparticles are exquisitely complex objects: their size, surface charge, shape, composition and manufacturing conditions all influence how they behave, and these variables interact in ways that defy intuition. A lipid nanoparticle built for mRNA delivery, for example, contains several distinct lipid species whose ratios must be tuned precisely, while a polymeric particle may depend on molecular weight, block architecture and solvent conditions simultaneously. Traditional design-of-experiments approaches can explore this space only a few factors at a time. Machine learning, by contrast, thrives on high-dimensional parameter spaces, finding patterns across thousands of variables that no human team could hold in mind at once.

The review emphasizes that the choice of algorithm is not arbitrary; it is dictated by how much data exists. In the small-data regime that still characterizes much of nanomedicine, where a typical laboratory might generate hundreds rather than millions of data points, tree-based ensembles such as random forests and gradient boosting methods, along with Gaussian process models, tend to outperform more fashionable architectures. Gaussian processes carry the additional advantage of quantifying their own uncertainty, telling researchers not just what the model predicts but how confident it is, which is invaluable when deciding which experiment to run next. Deep learning, including transformer-based neural networks, only becomes competitive when large, standardized datasets are available, such as those emerging from high-throughput lipid screening campaigns. This mismatch between model ambition and data reality is one of the field’s central tensions, and the authors argue that honest recognition of it should guide method selection more than novelty does.

At the formulation design and optimization stage, machine learning is already delivering concrete wins. Models can predict the physicochemical properties of lipid, polymeric and self-assembling nanoparticles, including particle size, encapsulation efficiency and drug release kinetics, from composition and process variables alone. One highlighted study used machine learning to predict critical liposome quality attributes and then inverted the model to identify the manufacturing parameters needed to hit a target size, easing the transition to microfluidic production. Another combined automation with data-efficient learning to navigate a large oral lipid nanoparticle formulation space from limited experiments. Generative adversarial networks have been applied to predict nanoparticle size in microfluidic synthesis, while Gaussian process models have been used to optimize polymeric particles for encapsulation efficiency and therapeutic efficacy. In each case, the computational model acts as a surrogate for laborious bench work, compressing months of iterative optimization into a fraction of the experiments.

Perhaps the most striking frontier is the discovery of entirely new excipients. Rather than merely optimizing known ingredients, researchers are now using machine learning to screen vast virtual libraries of candidate molecules. One widely cited effort combined machine learning with combinatorial chemistry to accelerate the discovery of ionizable lipids for mRNA delivery, exploring enormous chemical spaces computationally before synthesizing only the most promising candidates. Transformer-based neural networks have been trained to design lipid nanoparticles de novo, and artificial intelligence-guided frameworks have produced ionizable lipids with improved delivery performance through iterative cycles of virtual screening and experimental feedback. Computationally guided high-throughput approaches have even explored a design space of more than two million drug-excipient pairings for self-assembling nanoparticles, demonstrating that scale of exploration which would be unthinkable by hand.

The review then turns to preclinical evaluation, where machine learning is being asked a harder question: not what a particle is, but what it does to cells and tissues. Models can predict cellular uptake, transfection efficiency and cytotoxicity from formulation features, and interpretable algorithms have linked delivery efficiency to tumor genomic mutations, offering a route toward personalized nanomedicine. Massively parallel pooled screening, combined with machine learning, has revealed biological regulators of lipid nanoparticle delivery, such as the lysosomal transporter SLC46A3, connecting particle-cell interactions to multi-omic data. Other models predict the functional composition of the protein corona, the layer of biomolecules that coats every nanoparticle entering a biological fluid and often determines its fate. Physics-informed neural networks, which embed known physical laws into the learning process, are being used to quantify transcytosis and diffusion across in vitro models of the blood-brain barrier, one of the most formidable obstacles in drug delivery.

In vivo prediction represents the ultimate prize. Machine learning models are now integrating nanoparticle characteristics with complex biological datasets, from tumor genomics to routine medical imaging, to forecast biodistribution, tumor accumulation and therapeutic outcomes. Interpretable radiomics models have been shown to predict nanomedicine tumor accumulation from standard clinical scans, while machine-learning-assisted analysis of individual tumor vessels has illuminated how nanoparticles permeate vasculature. Studies of nanoparticle delivery to the brain, to glioblastoma and to tumors more broadly have all benefited from these approaches. The authors caution, however, that a well-known weakness persists: correlations between in vitro and in vivo performance are often weak, meaning that cell culture results remain an imperfect proxy, and models trained on them inherit that limitation. Bridging this gap is among the field’s most urgent challenges.

A particularly transformative development is the marriage of machine learning with autonomous experimentation, the so-called self-driving laboratories. In these closed-loop systems, a model proposes the next best experiment, robotic platforms execute it, analytical instruments feed the results back, and the model updates itself in an unbroken design-make-test-analyse cycle. Platforms powered by deep learning and foundation models have already accelerated lipid nanoparticle development for mRNA delivery and enabled the autonomous discovery of new ionizable lipid designs. Molecular dynamics simulations and physics-informed frameworks add mechanistic grounding, ensuring that models do not merely interpolate between data points but respect the underlying physics of self-assembly, diffusion and molecular interaction. Together these tools promise workflows in which the boundary between prediction and validation becomes increasingly porous.

None of this will happen automatically, and the review is refreshingly candid about the obstacles. Nanomedicine datasets are frequently small, inconsistently reported and locked away in individual laboratories, which starves algorithms of the fuel they need. The authors call for community-wide adoption of FAIR data principles, making data findable, accessible, interoperable and reusable, alongside standardized minimum-information reporting frameworks, open-source code sharing and large-scale repositories built through public-private partnerships. They also stress the importance of interpretability, since regulators and clinicians will need to understand why a model recommends a particular formulation before trusting it with patients. If those cultural and infrastructural shifts take hold, the authors argue, machine learning could finally deliver on nanomedicine’s long-deferred promise, shortening development pipelines, improving clinical translation and bringing precisely engineered nanoparticle therapies to patients faster than ever before.

Subject of Research: The application of artificial intelligence and machine learning to the design, optimization and preclinical evaluation of nanoparticle drug delivery systems

Article Title: Artificial intelligence and machine learning in nanoparticle drug delivery systems

Article References: Panagiotakopoulou, M., Goren, A., Reker, D., Schroeder, A., Allen, C., & Heller, D. A. (2026). Artificial intelligence and machine learning in nanoparticle drug delivery systems. Nature Reviews Bioengineering. https://doi.org/10.1038/s44222-026-00495-7

Image Credits: AI Generated

DOI: 10.1038/s44222-026-00495-7

Keywords: artificial intelligence, machine learning, nanoparticles, drug delivery, lipid nanoparticles, nanomedicine, mRNA delivery, formulation optimization, self-driving laboratories, protein corona, biodistribution prediction, FAIR data principles