To appreciate why the in vivo approach has generated such intense interest, it helps to understand precisely where the bottlenecks of conventional CAR-T manufacturing lie. The standard workflow begins with leukapheresis, in which a patient’s blood is circulated through an apheresis machine to harvest lymphocytes over several hours. Those cells must then be shipped to a centralized manufacturing facility, activated ex vivo with stimulation reagents, genetically modified to express the chimeric antigen receptor, expanded to therapeutic numbers over one to two weeks, subjected to quality control testing, cryopreserved or shipped fresh, and finally reinfused into the patient, typically after a lymphodepleting chemotherapy regimen designed to make space for the engineered cells. Each step introduces delays, costs, and opportunities for product failure. Patients with aggressive malignancies may progress or die while waiting, and manufacturing failures—caused by poor T cell yield, inadequate expansion, or contamination—affect a meaningful fraction of production runs. The review emphasizes that this model is fundamentally constrained by manufacturing capacity and healthcare resources, which is why only a small proportion of eligible patients worldwide actually receive approved CAR-T therapies despite their demonstrated efficacy.

The conceptual appeal of in vivo CAR-T is that it collapses this multi-week pipeline into a single administration event. Instead of removing T cells from the body, the delivery vehicle itself carries the genetic instructions encoding the chimeric antigen receptor to T cells circulating in blood, resident in lymphoid tissues, or infiltrating tumors. If the payload reaches enough of the right T cells, the patient’s own immune system becomes the manufacturing plant, producing CAR-expressing cytotoxic cells on site. The authors note that this strategy drastically reduces both time and cost, moving from weeks of ex vivo manipulation to practically immediate availability. Beyond logistics, there is a biological argument as well: T cells that are engineered while remaining in their native physiological environment may retain a less differentiated state than cells subjected to culture expansion. In particular, the review highlights the memory stem cell phenotype, a subset of T cells characterized by self-renewal capacity and the ability to generate durable effector populations. Ex vivo culture tends to push cells toward terminal effector differentiation, which correlates with shorter persistence after reinfusion. In vivo engineering, by contrast, could theoretically preserve this naive-like, long-lived compartment, potentially endowing the resulting CAR-T cells with superior persistence and anti-tumor durability.

Delivery technology sits at the heart of whether this vision can be realized, and the review organizes the current landscape into viral and non-viral platforms. On the viral side, targeted lentiviral vectors use engineered envelope proteins to bind specific receptors on T cells, redirecting tropism away from the broad cell types that wild-type lentiviruses can transduce. Lentiviruses offer two practical advantages: large genetic cargo capacity, which accommodates not just the CAR cassette but additional regulatory elements or multi-cistronic payloads, and stable genomic integration, which supports persistent receptor expression across cell divisions. The trade-off is that integration raises long-standing questions about insertional mutagenesis, the risk that the vector lands within or near a gene and disrupts its function. Engineered adeno-associated virus vectors take a different position on this spectrum. AAV genomes persist largely as episomes—circularized DNA molecules that remain separate from the host chromosomes—which substantially reduces insertional mutagenesis risk, though at the cost of more limited packaging capacity and, in some contexts, pre-existing immunity to common AAV serotypes in the human population.

Non-viral platforms are dominated by lipid nanoparticles, the same delivery class that proved itself in mRNA vaccines. LNPs encapsulate mRNA or other nucleic acid cargos within a lipid shell that can be modified on its surface with targeting ligands or tuned through optimized lipid composition to preferentially engage T cells. Because mRNA does not enter the nucleus and is degraded over days, expression is transient by design. The review points out that this transience avoids genotoxicity concerns entirely, and that LNP-mRNA approaches have already shown remarkable efficacy in non-tumor disease models, including cardiac fibrosis and systemic lupus erythematosus—findings that illustrate how a short pulse of CAR expression can be sufficient when the therapeutic goal is resetting a pathological immune process rather than eliminating a large tumor burden. Alongside LNPs, the review catalogs polymeric nanoparticles and virus-like particles as additional non-viral options, each offering different balances of loading capacity, safety profile, and manufacturing scalability.

The translational picture, while still early, is moving quickly. The authors report that both LNP-mRNA and targeted lentiviral systems have achieved efficient T cell transduction and deep target cell clearance in non-human primate models, an important step beyond rodent studies given the closer correspondence of primate immune systems and T cell biology to humans. In the clinic, early-phase trials for relapsed or refractory multiple myeloma and for autoimmune diseases have shown promising safety and preliminary efficacy. The inclusion of autoimmune indications is notable: B cell depletion, the same mechanism that underlies CAR-T therapy in certain leukemias and lymphomas, is being repurposed to reset the autoreactive immune repertoire in conditions such as lupus, where even transient CAR expression may produce durable remissions. This convergence of oncology and immunology applications reflects the broader thesis of the review—that in vivo CAR-T is not merely a cheaper version of existing therapy but a platform whose properties, particularly controllable expression duration, may suit different disease categories in different ways.

That controllability cuts both ways, and the review is candid about the central challenges. Persistence is the first: mRNA-based expression fades within days, which may be insufficient for complete tumor eradication in bulky or aggressive disease. The proposed solutions include circular RNA, whose covalently closed structure resists exonuclease degradation and can extend protein production, and self-amplifying RNA, which encodes its own replication machinery to prolong and amplify expression from a small initial dose. The second challenge is specificity. Off-target transduction—delivery vehicles reaching cells other than the intended T cell population—could create CAR-expressing cells that attack healthy tissue, while on-target effects such as cytokine release syndrome remain a concern even with correctly targeted cells. The review identifies logic-gated dual-recognition strategies, in which full T cell activation requires the simultaneous detection of two distinct antigens, and T cell-specific promoters that restrict receptor expression to the desired lineage, as critical engineering tools for narrowing these risks.

Immunogenicity presents a third obstacle, particularly acute for repeat dosing. Viral vectors and even lipid components can provoke immune responses that neutralize subsequent administrations, effectively converting a potentially titratable, redosable medicine into a one-shot therapy. Stealth vector design—engineering surfaces that evade immune surveillance—and transient immunomodulation to blunt anti-vector responses during the dosing window are the frontier solutions the authors highlight. Finally, there is the problem of the starting material itself. Unlike ex vivo manufacturing, where T cells can be selected, expanded, and quality-controlled before infusion, in vivo engineering must work with whatever T cells the patient possesses. In advanced cancer, patient T cells are often exhausted, functionally impaired by chronic antigen stimulation and an immunosuppressive tumor microenvironment. The review suggests cytokine pre-treatment to rejuvenate the T cell compartment and co-delivery of anti-exhaustion genes—genetic payloads that block inhibitory pathways or restore effector function—as strategies to make in vivo engineering effective even in immunologically compromised patients.

Looking ahead, the authors frame the field’s trajectory around several converging technologies. Artificial intelligence-assisted molecular design is expected to accelerate the optimization of delivery vehicles, targeting ligands, and payload architectures that would otherwise require exhaustive empirical screening. More precise T cell subset targeting aims to direct payloads not just to T cells generically but to specific subsets—memory stem cells, for instance—whose engineered progeny would have the most favorable persistence and safety profiles. Multi-target combination therapy addresses antigen escape, the well-documented phenomenon in which tumors downregulate a single target antigen and evade single-specificity CAR-T cells. Vector engineering tailored to the tumor microenvironment seeks to make delivered constructs responsive to local cues, such as hypoxia or inflammatory signals, so that CAR expression or activity is concentrated where it is needed. Taken together, the review positions in vivo CAR-T as a potential transition point for the entire cellular therapy field: a shift from individually manufactured living drugs toward scalable, off-the-shelf genetic medicines that could be stocked in clinics and administered on demand, dramatically expanding access to the patients whom current CAR-T manufacturing models have largely left waiting.

Subject of Research: In vivo CAR-T cell engineering and delivery strategies

Article Title: In vivo CAR-T therapy: From customized living drugs to universal genetic medicines

Article References: In vivo CAR-T therapy: From customized living drugs to universal genetic medicines. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: CAR-T therapy, in vivo T cell engineering, viral vectors, lipid nanoparticles, mRNA delivery, cancer immunotherapy, autoimmune disease, lentiviral vectors, adeno-associated virus, cell therapy, gene delivery, Immunity & Inflammation

The scale of the improvement is striking. After intramuscular injection, the new formulation reduced liver signal by roughly eighty-fold compared with the parent lipid recipe, and after intravenous injection the reduction was about seventy-two-fold. Crucially, these gains did not come at the expense of local expression: mRNA delivered into muscle still produced robust amounts of the encoded protein. For a field in which off-target expression has been a persistent safety and efficacy concern, the result offers a concrete demonstration that delivery precision can be engineered directly into the nanoparticle itself.

The achievement was not the product of a single lucky substitution. The researchers applied a statistical framework known as design of experiments, or DOE, to interrogate how each component of the lipid mixture, and each ratio between components, shaped delivery outcomes. Rather than adjusting one ingredient at a time, DOE allows scientists to explore combinations of changes simultaneously and to identify the formulation space where desirable properties converge. This systematic refinement retained SM102, the ionizable lipid that anchors the original formulation, but replaced cholesterol with beta-sitosterol, a plant-derived sterol, and recalibrated the proportions of a helper phospholipid and a polyethylene glycol, or PEG, lipid that stabilizes the particle surface.

Each of these changes carries mechanistic significance. Cholesterol has long been a standard structural component of lipid nanoparticles, but plant sterols such as beta-sitosterol differ subtly in molecular shape and behavior, and swapping them can alter how particles interact with proteins in the blood and with cell membranes. The PEG lipid, meanwhile, governs particle size, circulation half-life, and the immune recognition that can follow repeated dosing. By tuning these variables in a statistically guided fashion, the team arrived at a composition whose behavior in the body diverged meaningfully from the original SM102 recipe.

Follow-up experiments illuminated why the new formulation spares the liver. Nanoparticles built with the SM102M recipe differed in physical shape from their predecessors, and they escaped intracellular compartments within liver cells less efficiently, a property that suppresses endosomal release and therefore protein production in hepatic tissue. The particles also persisted in the body for a shorter time and bound less strongly to apolipoprotein E, or ApoE, a blood protein that acts as a molecular escort, shepherding lipid particles toward hepatocytes. Weakening this ApoE interaction effectively removes one of the liver’s principal homing signals. Paradoxically, despite this reduced liver tropism, SM102M delivered mRNA more efficiently into cells in culture, suggesting that the same structural changes that deflect the particles from hepatic uptake can enhance their performance once they reach a target cell population.

The durability of the effect across repeated doses may prove as important as the initial result. Because mRNA therapeutics provide only temporary instructions, many envisioned applications, from chronic muscle disorders to recurring vaccines, will require multiple administrations over time. In the study, SM102M maintained strong local expression and low liver signal across repeated injections, whereas the original SM102 formulation lost much of its local potency with successive doses. The redesigned particles also provoked less anti-PEG immunoglobulin G, an antibody response directed against the PEG coating that can accelerate particle clearance and raise the risk of infusion reactions. Blunting this immune activation could extend the practical dosing window for mRNA therapies that depend on PEG-stabilized nanoparticles.

The benefits extended beyond muscle. When the researchers injected the particles into a brain ventricle or into the vitreous cavity of the eye, SM102M produced protein expression concentrated within the intended anatomical region, with less unwanted liver expression than the reference formulation achieved. Local delivery to the central nervous system and to ocular tissues represents some of the most promising frontiers for mRNA medicine, from neurologic gene supplementation to retinal therapies, and both settings demand that expression remain confined to delicate, hard-to-reach structures. A formulation that keeps expression local could reduce the systemic exposure that complicates these applications.

The study’s methodological approach deserves attention as much as its headline numbers. Design of experiments has deep roots in industrial chemistry and process engineering, but its application to lipid nanoparticle optimization illustrates how modern drug delivery research increasingly resembles rational formulation engineering rather than trial-and-error screening. By mapping how multiple formulation variables interact, the technique can reveal compositional neighborhoods where liver accumulation, circulation time, cellular uptake, and immune reactivity are jointly optimized. The SM102M result suggests that the lipid nanoparticle design space contains substantial untapped room for tissue-selective tuning, and that systematic exploration of that space can yield formulations with properties quite different from the clinically established recipes.

Cautious interpretation remains warranted. All of the findings reported so far come from mouse experiments, and the biodistribution and immunology of lipid nanoparticles in humans can differ in important ways. The authors themselves note that further studies using therapeutic mRNA and established disease models will be needed to demonstrate genuine treatment benefits and longer-term safety before any translation toward human use. Liver accumulation, after all, is not merely a nuisance; it reflects the broader challenge of steering nanoscale carriers through the body’s filtration and clearance systems, and the mechanisms uncovered here, from ApoE binding to endosomal escape, will need to be validated across species and disease contexts.

Nevertheless, the work adds a compelling data point to a rapidly maturing field. Lipid nanoparticles already underpin approved vaccines and advancing gene-editing programs, and the ability to retune their tissue selectivity without abandoning proven core lipids could accelerate a new generation of locally targeted mRNA therapies. If the principles behind SM102M hold up in larger animal models and, eventually, in clinical settings, the lipid recipe that made mRNA medicine famous may be joined by a family of recipes designed not just to deliver, but to deliver precisely where the body needs it, and nowhere else.

Subject of Research: Redesigning lipid nanoparticles for tissue-selective local mRNA delivery with reduced liver accumulation

Article Title: A new lipid recipe helps mRNA delivery stay on target

Article References: A new lipid recipe helps mRNA delivery stay on target. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: lipid nanoparticles, mRNA delivery, SM102M, design of experiments, liver accumulation, beta-sitosterol, ApoE, anti-PEG antibodies, local injection, brain delivery, ocular delivery, repeated dosing

The core challenge the team faced is one that has long frustrated the field of nonviral cell engineering. Lipid nanoparticles, the same delivery technology behind the leading mRNA COVID-19 vaccines, are remarkably efficient at transfecting liver cells after intravenous injection, but they are far less adept at reaching immune cells circulating in the blood. NK cells in particular are notoriously difficult to transfect with conventional LNPs, and their engineering has typically relied on viral vectors or electroporation performed in the laboratory, both of which add expense, delay and complexity. To overcome this barrier, the MIT group engineered a library of targeted nanoparticles decorated with antibody fragments designed to bind receptors found on the surface of NK cells.

The key innovation lies in the use of engineered biparatopic ligands, antibody-derived binding molecules that engage two distinct epitopes on the same receptor. The researchers conjugated these single-chain variable fragments to the surface of ionizable lipid nanoparticles, creating what they call NK cell-targeted LNPs, or NtLNPs. Among the most effective were particles bearing a biparatopic ligand directed against NKG2D, an activating receptor abundantly expressed on NK cells. Molecular dynamics simulations, run alongside the experimental screening, helped the team understand how the biparatopic design engages more residues at the receptor interface than a single-epitope binder, likely stabilizing the interaction long enough for the particle to be internalized. The multivalent engagement appears to be critical: by gripping the receptor at two sites simultaneously, the ligand exploits the additivity of binding energies to achieve an avidity effect that monovalent fragments cannot match.

The performance figures reported in the paper are striking. Lead NtLNP formulations achieved transfection of up to roughly 80 percent of NK cells in mice, with minimal off-target delivery to other cell types. That level of selectivity and efficiency, achieved through a simple intravenous injection of synthetic particles, approaches what is typically only seen with optimized viral transduction in a laboratory dish. The nanoparticles encapsulate mRNA encoding the CAR construct, and once the particles are taken up by NK cells, the mRNA is translated in the cytoplasm, producing the chimeric antigen receptor protein that traffics to the cell surface. Because mRNA delivery is transient, the approach avoids the permanent genomic integration associated with viral vectors, a feature that may translate into a more controllable and potentially safer safety profile.

To test whether the engineered cells could actually function as therapies, the researchers first turned to a CD19-targeted CAR, the same antigen target used in approved CAR T cell therapies for B cell malignancies. After systemic administration of NtLNPs carrying CD19 CAR mRNA, the generated in vivo CAR NK cells completely depleted circulating B cells in healthy mice, demonstrating that the reprogrammed cells were not merely expressing the receptor but were actively patrolling and killing target cells throughout the vasculature. This experiment established proof of concept that a single injection of synthetic nanoparticles could generate a functional population of CAR NK cells distributed across the animal.

Perhaps the most clinically provocative result came from an autoimmune disease model. The team engineered NK cells in vivo with a CAR targeting CD3, a pan-T cell marker, and administered the particles to mice with autoimmune hepatitis, a T cell-driven inflammatory liver disease. The CD3-targeted CAR NK cells eliminated nearly all autoreactive T cells in the model, dramatically ameliorating the disease. This finding highlights a potential structural advantage of NK cells over CAR T cells for treating T cell-mediated disorders. CAR T cells directed against T cell antigens face the problem of fratricide, in which the engineered T cells kill one another because they carry the target antigen themselves, and they risk wiping out the entire T cell compartment, including protective immunity. NK cells, by contrast, do not express CD3, so they can be safely directed against T cells without destroying themselves, and their more transient lifespan may limit the duration of immunosuppression.

The platform also proved adaptable to solid tumor immunotherapy through an entirely different mechanism. Rather than arming NK cells with a tumor-targeting CAR, the researchers used NtLNPs to deliver mRNA encoding a secreted or membrane-tethered blocker of the inhibitory checkpoint NKG2A. NKG2A is a receptor that, upon engaging its ligand Qa-1b on tumor cells, dampens NK cell cytotoxicity and contributes to immune evasion in the tumor microenvironment. By targeting the checkpoint in vivo, the NtLNP treatment enhanced antitumor efficacy by relieving Qa-1b-mediated immunosuppression, effectively unleashing the NK cells’ natural killing capacity against tumors that had learned to hide from it. This dual capability, both CAR programming and checkpoint modulation within a single nanoparticle platform, underscores the modularity that the authors emphasize as the system’s defining feature.

Safety data accompanying the study address some of the most obvious concerns about repeated systemic dosing of targeted nanoparticles. The researchers profiled serum cytokines at acute time points, monitored body weight longitudinally, and performed blood chemistry analyses of liver and kidney function at both short and long-term intervals after repeated intravenous administration. Histological examination of major organs, including heart, liver, spleen, lung and kidney, showed no overt toxicity, and the team also measured anti-drug antibody titers against the conjugated antibody fragments, evaluating whether repeated dosing generated neutralizing serum activity that could blunt the platform’s effectiveness. These analyses, while conducted in mice, provide an early indication that the approach may be compatible with the repeat dosing regimens that transient mRNA-based therapies would likely require.

Finally, the team extended the work toward clinical relevance by optimizing the NtLNPs for in vivo engineering of human NK cells. In humanized mouse models bearing human hematologic malignancies, the optimized particles generated human CAR NK cells in vivo that drove regression of two distinct xenograft models, including a Raji lymphoma model and a T cell leukemia model. Cross-species structural analyses of the NKG2D receptor interface, comparing human and murine sequences, helped rationalize why the targeting ligands retained activity across species, with conserved residues forming hydrogen bonds, salt bridges and cation-pi interactions at the binding interface. The authors also validated human T cell depletion in the xenograft setting, confirming that the CD3-directed approach behaves as expected in a human immune context.

The broader significance of the study lies in what it suggests about the future architecture of cell therapy. Ex vivo CAR cell manufacturing requires apheresis, viral transduction, cell expansion and quality control, costs that can run to hundreds of thousands of dollars per patient and confine treatment to a handful of academic and commercial centers. In vivo engineering, by contrast, would in principle allow a physician to administer an off-the-shelf injection that converts a patient’s own immune cells into therapeutics at the bedside. The MIT team’s demonstration that NK cells, long considered among the hardest immune cells to transfect, can be programmed at high efficiency with ligand-directed lipid nanoparticles opens a path toward that vision. Considerable work remains before human trials, including dose optimization, biodistribution refinement and the inevitable surprises of human biology, but the study establishes NtLNPs as a credible and modular foundation for the next generation of immunotherapies, one in which the manufacturing plant is the patient’s own bloodstream.

Subject of Research: In vivo engineering of CAR natural killer cells using NK cell-targeted lipid nanoparticles for cancer and autoimmune disease immunotherapy

Article Title: In vivo engineering of CAR natural killer cells for immunotherapy using ligand-directed lipid nanoparticles

Article References: Chen, Y., Monkovic, J., Mughal, H., Chen, B., Thadhani, E., McMullen, P., Self, A., Chen, Y., Hu, Y., Detmer, S., Chen, M., Lapteva, A., Kamat, R., Vernon, A., Langer, R., & Anderson, D. G. (2026). In vivo engineering of CAR natural killer cells for immunotherapy using ligand-directed lipid nanoparticles. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03310-1

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03310-1

Keywords: lipid nanoparticles, natural killer cells, CAR cell therapy, in vivo cell engineering, mRNA delivery, cancer immunotherapy, autoimmune hepatitis, NKG2A checkpoint, biparatopic ligands, nanomedicine, gene therapy, Nature Biotechnology

Duchenne muscular dystrophy is an X-linked disorder caused by mutations in the DMD gene, the largest gene in the human genome, spanning 79 exons. These mutations abolish production of dystrophin, a structural protein that acts as a molecular shock absorber, linking the internal cytoskeleton of muscle fibers to the surrounding extracellular matrix. Without it, muscle cells degenerate under the mechanical stress of contraction, leading to progressive wasting, loss of ambulation, cardiomyopathy and premature death. The gene’s sheer size and its array of alternative promoters give rise to multiple dystrophin isoforms of different lengths: the full-length Dp427m isoform dominates in skeletal and cardiac muscle, while shorter versions such as Dp260 in the retina, Dp140 in the central nervous system and kidney, Dp116 in Schwann cells, and the ubiquitous Dp71 and Dp40 serve specialized roles in non-muscle tissues. Where a patient’s mutation falls on the gene determines which isoforms are lost, and thus which tissues bear the brunt of the disease.

Existing preclinical models fall short of capturing this complexity. The classic mdx mouse carries a nonsense mutation in the murine Dmd gene, but mouse and human DMD differ substantially in sequence, exon–intron organization and splicing regulation, and the mouse phenotype is comparatively mild. Humanized mice, in which specific murine exons have been replaced with human sequences, allow editing tools designed against human targets to be tested in vivo — but each line models only a single mutation. Because Duchenne lacks mutational hot spots and instead displays extreme heterogeneity across the gene, building a bespoke mouse for every clinically relevant variant is impractical, particularly for rare mutations. The Korean team, led by Hyuk-Jin Cha and Hyukjin Lee, reasoned that a renewable, patient-derived tissue that carries the patient’s own mutation might sidestep this limitation entirely.

Their solution exploits a structure usually regarded as a laboratory nuisance. When human pluripotent stem cells are injected into immunodeficient mice, they form teratomas containing disorganized derivatives of all three germ layers, including skeletal, cardiac and smooth muscle. Teratomas derived from Duchenne patient iPS cells recapitulate the absence of dystrophin, effectively recreating a patch of dystrophic human tissue inside a mouse. The researchers drew an analogy to patient-derived tumor xenografts, a mainstay of oncology drug development, and adapted the same framework to neuromuscular disease. From a bank of nine patient-derived iPS lines carrying missense, nonsense, duplication and deletion mutations, they focused on a line harboring a C-to-T transition in exon 6 that creates a premature stop codon — a mutation directly correctable by an adenine base editor, which chemically converts A·T base pairs to G·C without cutting the DNA double helix.

First, the team built an isogenic pair. Using electroporation, they delivered the adenine base editor into the mutant iPS cells and isolated edited clones that retained full pluripotency, confirmed by marker gene expression, alkaline phosphatase staining and teratoma formation. When differentiated in a dish over 16 days, the edited cells matured into spontaneously contracting cardiomyocyte-like cells, and immunoblotting revealed that the full-length Dp427m protein, absent in the mutant line, had been restored. A stepwise myogenic protocol — guiding cells through paraxial mesoderm, myogenic progenitor and myoblast stages over roughly 90 days — produced multinucleated, striated myotubes that likewise expressed robust Dp427m. These experiments established that correcting the single exon 6 mutation was sufficient to rescue the muscle-specific isoform in both cardiac and skeletal muscle lineages in vitro.

The ex vivo test followed. The researchers generated day-12 myogenic progenitors from the edited iPS cells, confirmed their expression of the progenitor markers PAX7 and MyoD by flow cytometry and immunofluorescence, and injected 100,000 of these cells into dystrophic teratomas grown from the uncorrected mutant cells. Four weeks later, immunoblotting showed that the engrafted progenitors had restored several shorter dystrophin isoforms — bands at approximately 117, 71 and 55 kilodaltons, presumably corresponding to Dp116, Dp71 and Dp40 — within the teratoma environment. Notably, however, neither the edited teratomas nor the progenitor-engrafted ones displayed the full-length Dp427m band seen in mature mouse muscle, suggesting that the teratoma’s developmental context did not yet support the maturation level required for the full-length protein.

The in vivo arm of the study tackled a delivery problem that has dogged the base-editing field. Adenine base editors such as ABE8e span roughly 5.4 kilobases, exceeding the approximately 4.7-kilobase packaging limit of adeno-associated virus vectors — and AAV immunogenicity has already produced severe, sometimes fatal adverse events in Duchenne gene therapy trials. Messenger RNA encapsulated in lipid nanoparticles, the platform proven by COVID-19 vaccines, accommodates large transcripts readily and avoids viral immunogenicity. The team first attempted a split-editor strategy, encoding the editor in two fragments joined by intein sequences that self-splice after translation, but found that the split mRNAs produced essentially no protein, likely because their engineered 5′ untranslated regions impaired ribosomal scanning. Full-length constructs with naturally functional leader sequences, by contrast, expressed robustly. Formulated by microfluidic mixing into uniform spherical nanoparticles of roughly 60 to 76 nanometers with low polydispersity, the SpRY mRNA–LNP conjugates outperformed even plasmid delivery when tested in patient iPS cells.

Injected directly into teratomas in mice, the editor showed a clear dose response: a single 10-microgram dose yielded about 1 percent editing, while escalating to 80 micrograms achieved up to roughly 15 percent in primary teratomas. Because teratomas vary unpredictably in size, composition and maturation, the researchers then devised a secondary transplantation scheme — sectioning one primary teratoma into four pieces and regrowing them in additional mice — to produce more uniform tissue. In these sequentially matured teratomas, editing efficiencies approached 100 percent at the highest dose, and, strikingly, two of four high-dose samples showed robust restoration of full-length Dp427m by long-exposure immunoblotting. Targeted sequencing of six predicted off-target exonic sites revealed no detectable A-to-G conversion above background, even at the highest dose, underscoring the specificity of the approach in human tissue.

Why did only some fully edited teratomas produce the full-length protein? Single-cell RNA sequencing of a Dp427m-positive and a Dp427m-negative teratoma provided the answer. Both contained broadly distributed DMD transcripts reflecting baseline activity of shorter isoforms, but only the positive sample harbored an enriched myogenic cluster — one of twenty identified cell populations — marked by elevated expression of skeletal muscle genes such as RYR3, LDB3, FBXO32 and MYOM1, along with cardiac maturation markers including TNNT2 and RYR2. Myogenic maturation, the data showed, is a critical determinant of whether corrected DNA translates into the clinically relevant full-length isoform. The authors caution that teratoma growth is inherently difficult to control and that animal welfare limits restrict teratoma size, so reproducible modeling of Dp427m recovery will require further optimization — potentially through inducible myogenic programming to bias teratomas toward muscle. Still, the platform establishes a proof of concept: a scalable, mutation-inclusive, patient-specific human tissue in which cell therapy and in vivo base editing can be tested side by side, bringing the field a step closer to evaluating personalized gene correction before it is ever infused into a patient.

Subject of Research: A patient-derived teratoma platform for evaluating ex vivo cell therapy and in vivo base editing to restore dystrophin in Duchenne muscular dystrophy

Article Title: Patient-derived teratomas as a humanized platform for dystrophin restoration by ex vivo cell therapy and in vivo base editing

Article References: Park, M. J., Kim, H.-B., Hwang, Y., Kim, Y.-J., Park, J.-C., Lim, H.-K., Park, Y.-G., Moon, S., Lee, H., & Cha, H.-J. (2026). Patient-derived teratomas as a humanized platform for dystrophin restoration by ex vivo cell therapy and in vivo base editing. Experimental & Molecular Medicine. https://doi.org/10.1038/s12276-026-01854-5

Image Credits: AI Generated

DOI: 10.1038/s12276-026-01854-5

Keywords: Duchenne muscular dystrophy, dystrophin, base editing, induced pluripotent stem cells, teratoma, lipid nanoparticles, gene therapy, myogenic progenitors, Dp427m, xenograft model, mRNA delivery, single-cell RNA sequencing

The standard manufacturing workflow for approved CAR T cell therapies involves isolating a patient’s own T cells, introducing a CAR-encoding gene, expanding the cells, and reinfusing them. Viral vectors achieve stable CAR expression but carry well-known liabilities: the risk of insertional mutagenesis, potential immunogenicity, limited cargo capacity, labor-intensive production, and costs that restrict broader clinical access. Electroporation, the clinical standard for mRNA transfection of T cells, requires specialized equipment and compromises membrane integrity, reducing viability and altering cellular physiology in ways that demand additional safety evaluation.

Messenger RNA offers an attractive middle path. Because it does not integrate into the genome and drives only transient CAR expression, it confers a self-limiting safety profile, confining receptor expression to a defined window of days. mRNA-based CAR T cells have already shown efficacy against acute lymphoblastic leukemia, melanoma, and Hodgkin lymphoma comparable to virally transduced cells. The challenge has always been delivery: naked mRNA is inherently unstable and cannot cross the cell membrane on its own, so a carrier is essential.

The research team, led by investigators at Shanghai University and the Shanghai Cell Therapy Group, synthesized a library of eight novel ionizable lipids as structural analogs of previously reported candidates effective in immune cell transfection. Each lipid was formulated into a nanoparticle by microfluidic mixing with a fixed excipient composition: cholesterol for membrane integrity and stability, the phospholipid DOPE to facilitate endosomal escape, and the PEG-lipid DMG-PEG2000 to confer colloidal stability and reduce nonspecific uptake. Holding the excipient ratios constant allowed the researchers to isolate the effect of the ionizable lipid head group during screening.

Dynamic light scattering and RiboGreen assays confirmed that all eight formulations were monodisperse, with Z-average diameters between 79.6 and 97.2 nanometers, polydispersity indices between 0.107 and 0.180, and mRNA encapsulation efficiencies above 95 percent. When screened in the murine dendritic cell line DC2.4 and in activated human peripheral blood mononuclear cells using eGFP mRNA, four of the eight candidates transfected more than 90 percent of DC2.4 cells, and lipid 6 emerged as the clear leader, delivering the highest eGFP positivity rates and mean fluorescence intensity and surpassing the clinically validated benchmark lipid ALC-0315.

In vivo validation using firefly luciferase mRNA in mice showed that lipid 6 elicited significantly higher luciferase expression than the benchmark lipids SM-102 and ALC-0315 at the six-hour time point for both intravenous and intramuscular routes. Intravenous injection produced the strongest systemic signal, whereas intramuscular administration yielded a more prolonged, localized signal that remained detectable at the injection site 72 hours after dosing. The researchers then systematically optimized the formulation using a design-of-experiments approach, screening 26 formulations across two rounds by varying the stoichiometry of the four LNP components. Both the ionizable lipid and DMG-PEG2000 concentrations exerted the strongest influence on transfection outcomes, and the team selected a formulation with reduced PEG-lipid content to minimize potential immunogenicity.

Titrating the nitrogen-to-phosphate ratio revealed that eGFP positivity increased with the ratio and plateaued at a value of 5, while mean fluorescence intensity peaked at a ratio of 10 before declining. Supplementing the transfection with recombinant APOE4 boosted efficiency, which plateaued at 1 microgram per milliliter, and an mRNA dose of 150 nanograms per 500,000 cells proved saturating. Under the finalized protocol, more than 90 percent of primary human T cells, natural killer cells, and dendritic cells expressed eGFP, and tumor-infiltrating lymphocytes isolated from hepatocellular carcinoma tissues exceeded 97 percent positivity. Both activated and unactivated T cells were transfected efficiently, at greater than 95 percent and greater than 80 percent positivity respectively. Storage testing showed the nanoparticles remained remarkably stable at 4 degrees Celsius, maintaining high cell viability and roughly 100 percent eGFP expression for up to 110 days, but performance collapsed within two weeks at 25 or 37 degrees Celsius.

The pivotal demonstration came when the team encapsulated mRNA encoding a MUC1-targeting CAR in lipid 6. Roughly 90 percent of transfected T cells expressed the CAR protein 24 hours after transfection, with CD4-positive and CD8-positive subset ratios unchanged and no significant upregulation of the exhaustion markers PD-1, LAG-3, or TIM-3. In real-time cytotoxicity assays, the engineered cells showed significantly enhanced killing of MUC1-expressing MDA-MB-468 breast cancer cells and SKOV3 ovarian cancer cells at every effector-to-target ratio tested, accompanied by markedly increased secretion of the Th1-type cytokines IL-2, TNF-alpha, and IFN-gamma. Longitudinal tracking showed CAR expression peaked at 24 hours and declined to below 5 percent by day 4, yet the cells retained significantly higher tumor-killing capacity than controls at each of the first three days after transfection, all while maintaining normal proliferation and viability.

Transcriptomic sequencing of lipid 6-transfected T cells reinforced the mechanistic picture. Gene Ontology enrichment pointed to upregulation of gene sets involved in transmembrane transport, membrane dynamics, cell junction organization, and intracellular trafficking, consistent with enhanced endosomal escape and cytosolic mRNA release. Kyoto Encyclopedia of Genes and Genomes pathway analysis showed robust activation of cytokine-cytokine receptor interactions, T cell receptor signaling, chemokine signaling, and Th1/Th2 differentiation pathways, indicating that the delivery process triggered physiological T cell activation without exhaustion or aberrant inflammatory signaling. Notably, unlike liver-tropic lipid nanoparticles that depend on APOE-mediated uptake, lipid 6 appears to operate in immune cells through alternative endocytic pathways, though the precise intracellular trafficking and endosomal escape mechanisms remain to be fully elucidated.

The authors are candid about the study’s limitations. All cellular assessments were confined to in vitro experiments, so long-term antitumor efficacy, biodistribution, and biosafety in tumor-bearing animal models remain unverified. Only a MUC1-targeted CAR was examined, leaving the platform’s generality across other tumor antigens untested, and scalable GMP-level production processes have not yet been established. The transient nature of mRNA expression, while a safety advantage, may also limit therapeutic persistence in solid tumor microenvironments. Even so, the work establishes a rationally designed, nonviral platform that overcomes the insertional mutagenesis risks, cellular damage, and prohibitive costs associated with viral vectors and electroporation, and it points toward a future in which CAR T cells and other engineered immune cell therapies can be manufactured more safely, more scalably, and more affordably.

Subject of Research: Lipid nanoparticle-mediated mRNA delivery for nonviral CAR T cell engineering

Article Title: Lipid-based mRNA chimeric antigen receptor T cells for enhanced in vitro antitumor activity

Article References: Gao, H., Fang, Y., Tian, Z., Guo, C., Wang, P., Liu, T., Sun, Y., Zhang, P., & Qian, Q. (2026). Lipid-based mRNA chimeric antigen receptor T cells for enhanced in vitro antitumor activity. iScience, 29(10), Article 117390. https://doi.org/10.1016/j.isci.2026.117390

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117390

Keywords: CAR T cells, lipid nanoparticles, mRNA delivery, ionizable lipids, immunotherapy, nonviral gene delivery, MUC1, T cells, endosomal escape, cancer therapy, transient expression, ex vivo engineering

At their core, VLPs are self-assembling nanostructures built from viral coat proteins that spontaneously organize into capsids closely resembling native virions in size, symmetry, and surface topology. Because the genetic blueprint of the virus is absent or deliberately removed, these particles cannot replicate or cause infection, which makes them inherently safer than attenuated or inactivated virus preparations. That safety, however, does not come at the cost of immunological potency. The repetitive, multivalent display of epitopes on a nanoscale scaffold is a pattern the immune system evolved to recognize as a hallmark of viral threat, so VLPs are taken up efficiently by antigen-presenting cells and can elicit strong antibody responses, often without the need for aggressive adjuvant formulation. The review emphasizes that this structure-property relationship, the link between capsid architecture and immune outcome, is the central design principle governing everything VLPs do.

The authors classify VLPs along several axes, distinguishing enveloped from non-enveloped particles, simple capsids from multi-layered architectures, and particles derived from bacteriophages, plant viruses, and animal viruses. Each structural class brings distinct advantages. Bacteriophage-derived platforms such as those based on Qβ or MS2 are extraordinarily robust and easy to produce in bacterial expression systems, while plant virus-derived particles offer remarkable stability and low manufacturing cost. Mammalian virus-derived VLPs, including those modeled on hepatitis B core antigen or papillomavirus capsids, more closely mimic the glycosylation and conformational presentation of human pathogens. The choice of scaffold is therefore never neutral: it determines particle size, surface chemistry, cargo capacity, and the nature of the immune response the particle will provoke.

Production is where much of the practical complexity lies. The review systematically compares expression systems, from bacteria and yeast to insect cells, plants, and mammalian cell lines, and it is explicit about the trade-offs. Bacterial systems are fast and cheap but cannot perform the post-translational modifications that some VLPs require; mammalian systems offer authentic glycosylation but at higher cost and lower yield. Glycosylation emerges as a recurring theme, because sugar patterns on capsid proteins can alter both antigenicity and manufacturing consistency. Downstream, purification must control impurities such as host-cell proteins, nucleic acids, and misassembled capsid species, and characterization demands a battery of methods, including electron microscopy, dynamic light scattering, mass spectrometry, and potency assays, to establish the critical quality attributes that regulators will scrutinize.

One of the most consequential sections of the review addresses cargo loading, the set of strategies by which VLPs are transformed from empty shells into delivery vehicles. Cargo can be packaged inside the capsid cavity, covalently or non-covalently attached to the interior or exterior surface, or electrostatically condensed within the particle. Genetic fusion during expression allows peptides and proteins to be displayed on the capsid surface with atomic precision, while chemical conjugation after purification offers flexibility for attaching haptens, drugs, or imaging agents. Disassembly-reassembly protocols exploit the reversible self-assembly of many capsids to trap molecules inside. These engineering options are what allow a single structural platform to serve as a vaccine carrier, a drug delivery vehicle, or a contrast agent, depending on what it carries and how it is decorated.

The clinical track record for prophylactic vaccines provides the field’s strongest precedent. Licensed VLP-based vaccines against hepatitis B virus and human papillomavirus have demonstrated that self-assembling capsids can protect millions of people, and the HPV vaccines in particular showed how multivalent capsid display can generate antibody titers far exceeding those produced by natural infection. The review is careful, however, to separate these established successes from the newer applications that remain largely preclinical or in early translational stages. Therapeutic cancer vaccines built on VLP scaffolds, mRNA delivery systems, CRISPR/Cas genome-editing cargo, and AI-assisted capsid design are all presented as promising but unproven at scale, and the authors resist the temptation to extrapolate from prophylactic success to therapeutic promise without evidence.

That distinction matters because the requirements for therapeutic applications are fundamentally different from those for prophylaxis. A cancer immunotherapy built on a VLP must not only present tumor-associated antigens but also break immune tolerance, often requiring co-delivered adjuvants or checkpoint modulation, and it must navigate an immunosuppressed tumor microenvironment. Gene and RNA delivery demands that particles protect fragile nucleic acid cargo from nucleases, achieve cell-type-specific targeting through surface ligands, and release their payload inside the correct intracellular compartment. For CRISPR/Cas delivery, the cargo is not merely large but enzymatically active, adding constraints on loading efficiency and release kinetics that no licensed VLP product has yet had to satisfy. The review treats each of these modalities as a distinct engineering problem rather than a simple extension of vaccine technology.

Biohybrid and composite design represents the direction the field is heading. The authors describe capsid engineering through rational mutation and directed evolution, surface functionalization with polymers such as PEG to extend circulation time, lipid envelope modification to mimic natural virions, and the incorporation of inorganic or polymeric materials to create composite particles with tailored mechanical and optical properties. These hybrid constructs blur the boundary between a biological nanoparticle and a designed material, which is precisely why the review appears in a journal devoted to composites and hybrid materials. Artificial intelligence is beginning to accelerate this design cycle, with computational models used to predict assembly behavior, optimize epitope presentation, and screen surface modifications, although the review notes that AI-assisted designs remain at an early stage of validation.

Diagnostics and imaging complete the application portfolio. VLPs can serve as high-density antigen displays in serological assays, improving sensitivity by presenting target epitopes in the multivalent arrangement antibodies preferentially recognize. When loaded with contrast agents or fluorophores, they become targeted imaging probes whose capsid provides both stability and molecular specificity. The uniformity of VLP dimensions, typically tens of nanometers, also makes them attractive standards for calibrating nanoscale measurement techniques. In all of these uses, the same core properties recur: self-assembly, monodispersity, programmable surface chemistry, and the absence of infectivity.

The review closes with a sober assessment of what stands between the current state of the art and routine clinical use. Manufacturing complexity, batch-to-batch consistency, glycosylation control, particle stability during storage, and scalability of production are identified as the principal barriers, compounded by a regulatory landscape that has not fully harmonized standards for products that are simultaneously biologicals, drugs, and devices. Potency assays must reliably predict clinical performance, and comparability studies must demonstrate that process changes do not alter critical quality attributes. The authors’ central argument is that the clinical value of VLPs will be determined not by any single breakthrough but by reproducible design, validated characterization, scalable production, and evidence-based implementation. In framing VLPs as biohybrid nanomaterials rather than merely as vaccines, the review offers a unifying conceptual framework for a technology that has already saved lives and may, if its manufacturing and regulatory challenges are met, reshape how vaccines, gene therapies, and precision medicines are built.

Subject of Research: Virus-like particles as biohybrid nanomaterial platforms for vaccines, drug and gene delivery, and diagnostics

Article Title: Virus-like particles as a unifying platform bridging prophylaxis, precision medicine, and advanced therapeutic modalities

Article References: Rasouli, M., Farhadi Khoozani, M., Fallahhosseini, N., Azary, H., ZiaRazzaz, M., Shirovi, S., & Doroudian, M. (2026). Virus-like particles as a unifying platform bridging prophylaxis, precision medicine, and advanced therapeutic modalities. Advanced Composites and Hybrid Materials. https://doi.org/10.1007/s42114-026-02050-9

Image Credits: AI Generated

DOI: 10.1007/s42114-026-02050-9

Keywords: virus-like particles, VLPs, vaccines, drug delivery, gene delivery, CRISPR, mRNA delivery, cancer immunotherapy, nanoparticles, biohybrid materials, biomanufacturing, capsid engineering

The stakes of this delivery problem are easy to appreciate when one considers the enzymes involved. Adenine base editors, first described by Gaudelli and colleagues in 2017, are programmable molecular machines that convert A•T base pairs to G•C in genomic DNA without making a double-strand break. Their precision and safety profile have made them attractive candidates for treating genetic disease, but they are encoded by messenger RNAs that are substantially longer than those used in early LNP applications. Cas9 messenger RNA itself already stretches the formulation envelope, and base editors push it further. When a delivery vehicle is asked to carry a transcript several kilobases long, the physics and chemistry of how lipids and RNA organize into a nanoparticle change in ways that conventional screening pipelines, which typically rely on short reporter RNAs, never register.

The clinical context sharpens the urgency. The pivotal trial by Frangoul and colleagues that produced the first approved CRISPR therapy for sickle cell disease and beta-thalassemia relied on ex vivo editing of harvested cells, sidestepping the delivery challenge entirely. In vivo editing is a different proposition. Gillmore and colleagues showed in a first-in-human trial that lipid nanoparticle-delivered CRISPR-Cas9 can edit the liver to treat transthyretin amyloidosis, and Rothgangl and colleagues demonstrated in macaques that LNP-based adenine base editing of PCSK9 can durably reduce LDL cholesterol. These landmark results prove that LNP-mediated in vivo editing works, but they also reveal how much of the therapeutic potential remains locked behind the efficiency ceiling of current formulations. Every incremental improvement in delivery translates into lower doses, reduced off-target exposure, and access to tissues beyond the liver.

Dong and colleagues approached the problem by asking a deceptively simple question: what happens to a lipid nanoparticle when the RNA inside it is the size of a gene editor’s transcript? The answer, revealed through structural characterization, is that the cargo is not a passive passenger. Large messenger RNAs alter how ionizable lipids and nucleic acids pack together, producing nanoparticles whose internal organization differs from those formed with short RNAs. Because the structure of an LNP governs its behavior in the body, particularly its ability to escape the endosome, the membrane-bound compartment that engulfs incoming particles after cellular uptake, a structurally compromised particle is a poorly functional one. Endosomal escape is widely regarded as the critical bottleneck in RNA delivery; only a small fraction of nanoparticles that enter a cell ever release their payload into the cytoplasm, where translation can begin.

Armed with this structural insight, the team built their screening strategy around the principle that the assay should match the application. Instead of evaluating candidate ionizable lipids with compact reporter RNAs, they screened formulations using a gene-editor-sized RNA, ensuring that the structural and functional properties being measured were those that would actually matter in a therapeutic context. This cargo-matched screening identified a lead ionizable lipid, designated LC-1, that had presumably been overlooked or underappreciated in conventional screens because its advantages only manifest when the RNA cargo is large.

The mechanistic payoff of LC-1 lies in endosomal escape. Structural analysis showed that LC-1-containing nanoparticles form an internal arrangement that enhances the release of large RNA payloads from endosomes, directly attacking the step that limits cytoplasmic delivery. This is a notable example of structure-guided formulation design: rather than treating the ionizable lipid as a tunable parameter optimized through iterative empirical variation, the study connects a specific lipid-dependent structural feature to a specific biological bottleneck, and then exploits that connection to select a superior candidate. The finding suggests that the field’s extensive libraries of ionizable lipids may contain many molecules whose true strengths and weaknesses are invisible unless they are tested with realistic cargo.

The functional consequences were validated across three tissues in mice. LC-1-based nanoparticles enabled efficient genome editing in the liver, the traditional stronghold of LNP delivery, but also in the lung and the brain, tissues that have historically been far more difficult targets for intravenously or systemically administered lipid nanoparticles. Achieving robust editing in the lung and brain hints at therapeutic applications that extend well beyond hepatology, from respiratory genetic diseases to central nervous system disorders, although the authors’ mouse data represent an early step on a long translational road. Tissue tropism in LNP delivery is governed by a complex interplay of particle properties, route of administration, and interactions with serum proteins, and the study’s title, which references both large RNA cargo optimization and tissue targeting, indicates that the formulation work was designed with this broader targeting problem in mind.

For the field, the most consequential message may be methodological. LNP development has historically been accelerated by combinatorial chemistry and high-throughput screening, approaches that generated the lipid libraries behind today’s clinical products. But if the performance of an ionizable lipid depends on the size of the RNA it carries, then screening campaigns built on small reporter RNAs may systematically misrank candidates for the applications that matter most, namely the delivery of long editor transcripts. The Dong study effectively establishes cargo-matched screening as a design principle: the RNA used in the assay should approximate the size and presumably the structural behavior of the therapeutic payload. This recalibration could redirect attention to existing lipids that were dismissed in earlier screens and could change how new libraries are evaluated.

The work also deepens a conceptual shift that has been gathering momentum in drug delivery research. Nanoparticles are increasingly understood not as inert shells but as self-assembled systems whose components, including the cargo itself, co-determine the final architecture and thus the biological performance. A several-kilobase messenger RNA is not merely a payload to be protected; it is a structural element of the particle, influencing lipid packing, internal morphology, and ultimately endosomal escape. Recognizing the cargo as a design variable opens a second axis of optimization alongside lipid chemistry: formulators can now think about matching vehicle to payload the way structural biologists think about matching ligand to binding pocket.

Cautious optimism is warranted. The reported results are preclinical, generated in mice, and the translation of LNP performance from rodents to humans has historically involved surprises in biodistribution, tolerability, and potency. Nevertheless, the study arrives at a moment when in vivo gene editing is transitioning from proof-of-concept to therapeutic platform, with liver-directed CRISPR therapies already tested in patients and base editors advancing through development for indications ranging from hypercholesterolemia to genetic blindness. Delivery efficiency is the variable that determines dose, safety margin, and tissue reach for all of these programs. By showing that cargo size matters, that structure explains why, and that a cargo-matched screen can find lipids like LC-1 that unlock efficient editing in liver, lung, and brain, the researchers have supplied the field with both a practical tool and a conceptual correction. The next generation of gene-editing medicines will likely be formulated not around generic nanoparticles, but around vehicles designed for the specific, oversized molecular machines they must deliver.

Subject of Research: Lipid nanoparticle design for delivery of large gene-editor mRNA cargo

Article Title: Cargo size matters when designing lipid nanoparticles for delivery of large gene editors

Article References: Cargo size matters when designing lipid nanoparticles for delivery of large gene editors. (2026). Nature Biotechnology. https://doi.org/10.1038/s41587-026-03297-9 (Author names are not included in the publisher’s citation export.)

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03297-9

Keywords: lipid nanoparticles, gene editing, base editors, ionizable lipids, endosomal escape, mRNA delivery, LC-1, in vivo genome editing, drug delivery, nucleic acid therapeutics, tissue targeting, CRISPR

The problem the researchers set out to solve is rooted in the physics of how lipid nanoparticles form. These particles assemble spontaneously when ionizable lipids, helper lipids, cholesterol, polyethylene glycol lipids and RNA are mixed in carefully controlled conditions. The ionizable lipid is the star of the show: it is positively charged at acidic pH, which allows it to bind and encapsulate negatively charged RNA, but neutral in the bloodstream, which limits toxicity. Decades of optimization have produced lipids such as ALC-0315, used in the Pfizer-BioNTech vaccine, and LP-01, a benchmark for in vivo editing studies. But those lipids were largely selected using short reporter RNAs, and their efficiency drops measurably when the transcript grows to the size of a full-length Cas9 mRNA, which can exceed four kilobases.

That size penalty matters enormously for genome editing. Unlike small interfering RNAs or compact vaccine transcripts, the mRNAs that encode CRISPR nucleases and base editors are among the largest RNAs anyone would want to deliver. Adenine base editor mRNA, for example, stretches to roughly 5.7 kilobases, and it must be co-delivered with a guide RNA and then translated into a functional protein inside the target cell. Viral vectors such as adeno-associated virus struggle to fit such cargo into a single genome, and preexisting immunity to Cas9 proteins and to viral capsids raises additional safety concerns. Nonviral delivery by lipid nanoparticles avoids those constraints, but only if the particle can actually get its bulky payload into the cytoplasm.

To find lipids suited to large transcripts, the Toronto group, led by Bowen Li with co-first authors Songtao Dong, Fanglin Gong and Tyler Thomson, built a combinatorial library of 384 ionizable lipids and screened them against a 5.7-kilobase reporter mRNA encoding an adenine base editor fused to NanoLuc, a luminescent protein. Crucially, the screen was conducted with the large cargo in hand, so the selection pressure reflected the real-world payload rather than a convenient stand-in. The top performer from that screen, LC-1, outshone both LP-01 and ALC-0315 when the same formulations were challenged with large mRNAs, and the advantage held across multiple cell types in culture.

The in vivo results are the heart of the study. When LC-1 nanoparticles carrying Cas9 mRNA and guide RNAs were injected intravenously into Ai9 reporter mice, which light up fluorescently when Cre-mediated recombination occurs, the team measured knockout of the reporter gene in up to 79 percent of liver cells. More remarkable still, LC-1 achieved 48 percent editing in the brain and 27 percent in the lung, depending on the route of administration, which included intrathecal injection into the spinal fluid and intratracheal delivery to the airways. Those figures represent up to fourfold improvements over the benchmark lipids LP-01 and ALC-0315 under the same conditions. For a field in which extrahepatic delivery has long been the bottleneck, editing nearly half of the relevant cells in brain tissue is a striking result.

The researchers then pushed the platform toward therapeutic targets. LC-1 supported base editing of PCSK9 in the liver, a gene whose disruption durably lowers LDL cholesterol and has already been validated in nonhuman primates as a potential one-time treatment for cardiovascular disease. In the lung, the team edited CFTR carrying the R55X mutation, a nonsense variant relevant to cystic fibrosis, where even partial restoration of functional protein is expected to ameliorate disease severity. And in the brain, delivered via intrathecal injection, LC-1 enabled editing at the Ube3a-ATS locus, the long noncoding RNA that silences the paternal copy of Ube3a in neurons. Reactivating that paternal allele is a leading strategy for treating Angelman syndrome, a severe neurodevelopmental disorder. Across all three routes of administration, LC-1 also produced higher correction rates in the LumA reporter mouse model of adenine base editing than the comparison lipids.

Why does LC-1 succeed where established lipids falter? The mechanistic studies in the paper offer an answer grounded in lipid physical chemistry. Using structural analyses, the team found that LC-1 forms stronger interactions with RNA and, critically, retains an ordered, fusogenic inverted-hexagonal phase as mRNA size increases. The inverted hexagonal phase is a nonbilayer lipid arrangement long associated with membrane fusion: lipids in this geometry promote contact between the nanoparticle and the endosomal membrane, allowing the RNA to escape into the cytoplasm rather than being routed to the lysosome for degradation. LC-1 also preserves pH-responsive membrane disruption as the cargo grows. The benchmark lipids, by contrast, become structurally disordered when loaded with large transcripts, losing the very geometry that enables endosomal escape. In other words, big RNA destabilizes ordinary particles but leaves LC-1’s fusogenic architecture intact.

The authors traced these properties to specific subchemical features of the lipid structure, showing how variations in the lipid’s building blocks influence both endosomal escape and the internal organization of the particle. That structure-function insight is arguably the study’s most consequential contribution, because it converts cargo size from an afterthought into a rational design variable. Screening libraries against a representative large transcript, the work suggests, should become standard practice for anyone developing ionizable lipids intended for genome editing, self-amplifying RNA, or other oversized payloads. It also complements a growing toolkit of approaches for tissue targeting, from selective organ targeting formulations to ligand-decorated particles, by ensuring that once a particle reaches its destination cell, it can actually deliver the goods.

Caveats remain before such particles reach the clinic. The results are in mice, and serum factors are known to create species-specific barriers to lipid nanoparticle delivery, meaning performance in larger animals and humans must be demonstrated. Dosing, biodistribution, immune responses to repeated administration, and long-term safety all require further study. The inventors have filed an invention disclosure through the University of Toronto, and the corresponding author serves as an advisor to biotechnology companies, so commercial development is likely to follow. Still, the demonstration that a single optimized lipid can mediate efficient editing in liver, brain and lung through intravenous, intrathecal and intratracheal routes marks a meaningful advance. If the large-cargo-informed design principle generalizes, it could widen the path toward nonviral, potentially redosable gene-editing medicines for diseases of the heart, the airway and the nervous system.

Subject of Research: Design of large-cargo-optimized ionizable lipids for lipid nanoparticle delivery of genome-editing RNA in vivo

Article Title: Lipid nanoparticles optimized for large RNA cargo and tissue targeting enhance in vivo genome editing

Article References: Dong, S., Gong, F., Thomson, T., Cai, Y., Healy, L., Lu, R. X. Z., Zhou, Z., Xu, Y., Chen, J., Savguira, M., Fu, X., Luozhong, S., Zhou, M., Kirtley, P., Wilder, B. K., & Li, B. (2026). Lipid nanoparticles optimized for large RNA cargo and tissue targeting enhance in vivo genome editing. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03298-8

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03298-8

Keywords: lipid nanoparticles, ionizable lipids, genome editing, CRISPR, Cas9, base editing, mRNA delivery, endosomal escape, PCSK9, CFTR, Angelman syndrome, drug delivery

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

To appreciate why this charge choreography matters, it helps to revisit how standard lipid nanoparticles work. A typical formulation combines an ionizable lipid with helper lipids, cholesterol, and polyethylene glycol-conjugated lipids, all assembled around a nucleic acid payload such as mRNA or siRNA. The ionizable lipid is engineered with an amine group whose protonation state depends on acidity: at the near-neutral pH of blood, roughly 7.4, the lipid is mostly uncharged, which keeps the particle circulations friendly and limits interactions with serum proteins and cell membranes. When the particle is taken up by a cell and lands in an endosome, whose interior becomes progressively more acidic, the amine group acquires a positive charge. This electrostatic switch lets the lipid blend with and destabilize the endosomal membrane, allowing the payload to escape into the cytoplasm where it can be translated or processed.

The problem is that cationic charge is inherently disruptive. Positively charged lipids and particles can bind avidly to negatively charged cell surfaces, destabilize plasma membranes, and trigger innate immune signaling pathways, including inflammasome activation and inflammatory cytokine release. Clinicians see the consequences as infusion reactions, fever, and dose-limiting toxicities that constrain how much therapeutic material can be administered, a particular obstacle for applications that require repeated or systemic dosing. Preclinical studies have long documented that ionizable lipids with lower pKa values and more biodegradable linkages tend to be better tolerated, but the field has continued to search for designs that decouple endosomal escape efficiency from extracellular reactivity.

The S-lipids described in the new work take that decoupling a step further. Instead of merely being less protonated at physiological pH, these lipids are designed to carry a neutral or even negative charge while circulating, so that the nanoparticle surface presents little of the cationic character associated with membrane damage and immune activation. Only when the particle encounters the strongly acidic environment inside endosomes does the chemical group undergo its switch, becoming positively charged at precisely the moment and location where membrane disruption is useful. In effect, the designers have shifted the charge transition from a gradual acid-base titration to a sharper, compartment-specific event, concentrating cationic activity where it helps and eliminating it where it harms.

According to the study, lipid nanoparticles built from these charge-switching lipids deliver nucleic acids with an efficiency comparable to that of conventional ionizable lipid formulations, despite presenting a non-cationic exterior in the extracellular environment. In cell-based assays and animal models, the S-lipid nanoparticles induced substantially reduced inflammatory responses, a difference the authors attribute to the absence of persistent positive surface charge during circulation and initial cell contact. The findings suggest that the inflammation commonly associated with lipid nanoparticle therapy is driven in large part by cationic interactions that occur before the particle ever reaches an endosome, and that these interactions can be engineered away without sacrificing the delivery mechanism that makes the technology valuable.

The implications reach across the growing portfolio of nucleic acid medicine. mRNA vaccines and therapeutics, siRNA gene silencing, CRISPR-based genome editing, and prime editing strategies all depend on lipid nanoparticles or close cousins to ferry their fragile cargo past cell membranes and endosomal traps. Each of these modalities is limited by the maximum tolerated dose of lipid. If charge-switching lipids genuinely lower toxicity while preserving potency, they could raise the ceiling on how much genetic instruction or editing machinery can be delivered in a single treatment, which in turn could improve efficacy against tumors, extend duration of protein expression, or reduce the frequency of dosing for chronic conditions. Safer lipids would also matter for patients with pre-existing inflammatory conditions, for whom even mild cytokine activation can be clinically significant.

The chemistry behind the switch reflects a broader trend in nanomedicine toward stimuli-responsive materials. Researchers have previously explored pH-sensitive polymers, acid-labile linkers, and ionizable heads with tuned pKa values, all in pursuit of the same goal: keeping a delivery vehicle inert in circulation and active inside the cell. The S-lipid approach refines this strategy at the level of the lipid headgroup itself, encoding environmental sensitivity into the molecular structure rather than relying on a detachable protecting group or a formulation additive. Such intrinsic responsiveness can be advantageous because it does not depend on a chemical reaction that competes with biological degradation, and because the behavior of the particle is determined uniformly by every lipid molecule in its shell.

There are, of course, familiar gaps between promising preclinical materials and routine clinical tools. Charge-switching lipids must demonstrate manufacturability at scale, stability during storage, predictable behavior across species, and safety profiles that satisfy regulators over repeated exposures. The acidic microenvironments of tumors and inflamed tissues, which some delivery strategies exploit, could in principle trigger premature charge switching, although the authors’ data focus on the endosomal route that dominates lipid nanoparticle uptake. Biodistribution, complement activation, and long-term accumulation of the new lipid species in liver and other organs will require the same scrutiny applied to every prior generation of ionizable lipids. Still, the conceptual advance, that toxicity can be reduced by sharpening the pH dependence of lipid charge rather than by broadly softening the lipid, gives formulation scientists a new design axis to explore alongside biodegradability, pKa tuning, and structural diversity generated by combinatorial synthesis.

The study lands at a moment when the lipid nanoparticle pipeline is expanding far beyond its vaccine origins. Companies and academic groups are advancing inhaled, implantable, and targeted formulations; conjugating antibodies and ligands to particle surfaces; and exploring delivery to cell types, such as T cells and hematopoietic stem cells, that have resisted lipid-based transfection. Every one of those programs inherits the same core constraint of the cationic switch, and any chemistry that relaxes that constraint without weakening delivery is likely to be adopted quickly. As the field matures from proving that nucleic acid delivery works to optimizing how well and how safely it works, charge-switching S-lipids offer a concrete, mechanistically grounded answer to one of the platform’s oldest criticisms: that the vehicle that saves the payload from destruction should not itself become the source of the patient’s discomfort.

For now, the result stands as an elegant demonstration that a single, well-chosen molecular property, the timing of a charge transition, can reshape the biological personality of an entire delivery platform. If subsequent studies in larger animals and human trials confirm the reduced inflammation reported here, the modest S-lipid may take its place alongside the ionizable lipids that preceded it as a quiet but consequential upgrade to the technology that carried mRNA medicine into the clinic.

The distinction between a gradual titration curve and a sharp switching threshold has practical consequences for how such particles behave in blood. Serum albumin and other abundant proteins carry their own net charges, and conventional ionizable lipid nanoparticles can acquire a protein corona whose composition influences both clearance and immune recognition. A particle surface that presents neutral or anionic character should interact with this corona differently, potentially altering which proteins adsorb and how the particle is routed through the liver and spleen. The authors’ observation that reduced inflammation tracks with the absence of persistent cationic surface charge is consistent with this picture, though the precise corona composition of S-lipid formulations remains an open question for follow-up work.

Endosomal escape itself remains one of the least efficient steps in the delivery process, with estimates suggesting that only a small fraction of internalized nucleic acid ever reaches the cytoplasm. Any headgroup chemistry that preserves membrane-disruptive activity at endosomal pH while silencing it elsewhere therefore addresses the central bottleneck rather than trading one limitation for another. The sharper pH dependence reported for S-lipids suggests that the protonation event can be concentrated within the narrow acidity range of late endosomes, where the payload must be released, rather than beginning in early endosomes or, worse, at the cell surface.

It is also worth noting that inflammation from lipid nanoparticles is not a single mechanism but a collection of overlapping pathways, including complement activation, toll-like receptor signaling, and inflammasome engagement, each with different dose thresholds and kinetics. Reducing cationic contact may dampen several of these at once, but disentangling which pathways are most sensitive to surface charge will help predict which patient populations benefit most. Such mechanistic mapping, together with head-to-head comparisons against clinically validated ionizable lipids, will determine whether the safety margin demonstrated in preclinical models translates into meaningfully higher tolerated doses in humans.

Subject of Research: Development of charge-switching ionizable lipids that reduce the toxicity and inflammatory effects of lipid nanoparticles for nucleic acid delivery.

Article Title: Charge-switching ionizable lipids lower the toxicity of lipid nanoparticles

Article References: Liang, D., Qi, Y., Han, H., Ahmadian, N., Gao, K., Sapasap, K., Zhang, Y., Guo, S., Lawanprasert, A., Pimcharoen, S., Zhao, S., Del Buono, M. T., Xia, H., Enders, Z. O., Burgstone, B. W., Calio, A., Dankar, N., Lu, B., Qi, L. S., … Murthy, N. (2026). Charge-switching ionizable lipids lower the toxicity of lipid nanoparticles. Nature Nanotechnology. https://doi.org/10.1038/s41565-026-02262-6

Image Credits: AI Generated

DOI: 10.1038/s41565-026-02262-6

Keywords: lipid nanoparticles, ionizable lipids, charge-switching, mRNA delivery, nucleic acid delivery, inflammation, nanotechnology, endosomal escape, drug delivery, RNA therapeutics, biocompatibility, nanomedicine