Targeted delivery of mRNA to immune cells for in vivo cell therapy Authors: Danyang Wang, Yumin Li, Jinfeng Deng, Mengxin Li, Lin-Ping Wu Published: 2026 ABSTRACT AbstractMessenger RNA (mRNA) therapeutics have revolutionized biomedicine by enabling direct in vivo programming of immune cells. This strategy bypasses the complex manufacturing and high costs associated with ex vivo cell therapies. However, efficient and specific systemic delivery of mRNA to target immune cell subsets remains a major translational hurdle. This review systematically examines engineering strategies that address this challenge. We first outline the key biological barriers to mRNA delivery, such as serum instability, nonspecific biodistribution, cellular uptake heterogeneity, and inefficient endosomal escape. Next, we comprehensively review advances in lipid nanoparticle (LNP) engineering, including discovering novel lipids, modulating compositions, conjugating targeting ligands, and incorporating stimuli-responsive elements, to enable enhanced tropism toward specific immune cells. Representative applications in oncology, protein replacement, autoimmune disease, and tissue regeneration are highlighted. Finally, we address translational challenges in safety, scalable manufacturing, and regulatory issues. The integration of rational material design, high-throughput screening, artificial intelligence, and interdisciplinary collaboration will be essential to advance next-generation targeted in vivo mRNA cell therapies toward clinical translation. Introduction The rapid deployment and demonstrated efficacy of messenger RNA (mRNA) vaccines during the COVID-19 pandemic validated the safety and scalability of mRNA-based therapeutics and nonviral delivery platforms (Wang et al. 2023). This success has made lipid nanoparticle-encapsulated mRNA a modular tool for guiding cells to produce therapeutic proteins. Currently, this technology is particularly prominent in infectious diseases, while its translational potential is expanding into oncology and beyond. One of the most promising frontiers is immune cell engineering, a class of therapeutics designed to use the host immune system to eradicate pathologies (Irvine et al. 2022; Maalej et al. 2023; Patel et al. 2025). To date, autologous CAR-T cell therapy has been the most clinically advanced form of immune cell engineering (Maalej et al. 2023). However, its broader clinical translation is severely constrained by the conventional ex vivo manufacturing modality. Logistically, the requirement for individualised leukapheresis, external genetic modification, and prolonged expansion in centralised facilities imposes extended manufacturing durations, high costs, and limited scalability (Pinto et al. 2025). Clinically, patients must undergo lymphodepletion before infusion to facilitate CAR-T cell engraftment, which increases infection risk and further weakens these critically ill individuals (Amini et al. 2022). During the manufacturing period, many patients need to receive bridging therapy to control tumour progression and prepare for the subsequent infusion, which adds further complexity and burden (Roddie et al. 2023). Biologically, extensive ex vivo expansion often drives T cells toward terminal differentiation and exhaustion, which compromises their native fitness and impairs their in vivo persistence and antitumor potency (Chen et al. 2025). Moreover, ex vivo CAR-T therapy is associated with significant safety concerns, including cytokine release syndrome (CRS), immune effector cell-associated neurotoxicity syndrome (ICANS), on-target off-tumour effects, and organ impairment resulting from persistent CAR-T cell activity (Chen et al. 2025). These challenges are further amplified in solid tumours, where heterogeneous antigen expression, poor T cell infiltration, and a suppressive tumour microenvironment (TME) limit efficacy (Maalej et al. 2023). To overcome these profound systemic limitations, the field is undergoing a paradigm shift toward in vivo cell engineering. By delivering genetic payloads directly to endogenous immune cells, this streamlined approach bypasses complex ex vivo manipulation, transitioning from ex vivo manufacturing of cellular products to in situ drug factories (Li et al. 2025). Notably, in vivo CAR-T generation eliminates the need for lymphodepletion and reduces risks associated with ex vivo cell expansion and reinfusion. Moreover, this strategy preserves the native, naive-like phenotype of immune cells, potentially eliciting durable therapeutic responses (Figure 1). While T cells serve as the foundational model and primary focus of early engineering efforts, this review actively covers the expanding frontier of targeting broader immune populations, including B cells, dendritic cells (DCs), macrophages, and natural killer (NK) cells, for diverse therapeutic applications. Comparison of conventional ex vivo and emerging in vivo CAR-T cell therapies. (Left) Conventional ex vivo CAR-T cell manufacturing is an autologous, individualised, and multi-step process. It begins with the patient’s blood collection, followed by T cell isolation, activation viral-mediated gene transduction, and extensive ex vivo expansion (typically 3–4 weeks). After standard quality control, the engineered CAR-T cells are infused back into the patient. This approach is associated with substantial clinical burden and complex individualised logistics. (Right) In vivo CAR-T therapy provides a potentially streamlined, off-the-shelf, and scalable alternative. Genetic cargo-loaded vehicles, such as mRNA-encapsulated lipid nanoparticles (mRNA-LNPs), are directly administered to the patient. These targeted delivery vehicles are taken up by endogenous T cells to enable rapid CAR protein expression within hours to days. The reprogrammed T cells then recognise and eliminate tumour cells. By circumventing prolonged ex vivo manipulation, this emerging strategy aims to reduce clinical logistics and offer broader applicability. Created in BioRender. Wang, D. (2026). The feasibility of in vivo reprogramming was recently validated by pioneering industry pipelines utilising engineered viral vectors, particularly lentiviruses and adeno-associated viruses, to generate potent in vivo CAR-T cells (Bot et al. 2026). However, because these integrating vectors deliver DNA, they carry inherent risks of long-term insertional mutagenesis and persistent, uncontrolled target-cell depletion. Drawing on the success of COVID-19 vaccines, transient messenger RNA (mRNA) has emerged as the ideal payload for in vivo cell therapy. Its transient and tunable expression kinetics avoid the unlimited proliferation of engineered T cells, thereby mitigating long-term toxicities such as prolonged B-cell aplasia (Zhu et al. 2022). Unlike DNA-based therapeutics, mRNA functions in the cytoplasm, which entirely eliminates the risk of genomic integration. In addition, mRNA can be rapidly designed and synthesised via in vitro transcription (IVT), enabling swift development cycles (Wang et al. 2023). Despite these advantages, mRNA therapeutics face intrinsic delivery barriers, including rapid nuclease degradation, immunogenicity triggered by pattern recognition receptors (PRRs), and restricted cellular uptake due to their large size and negative charge (Aldosari et al. 2021; Zhu et al. 2022). While chemical modification of nucleotides, such as substituting uridine with analogues like pseudouridine, N1-methylpseudouridine, or 5-methoxyuridine, can effectively enhance mRNA stability, reduce immunogenicity, and boost translational efficiency (Karikó et al. 2005, 2008; Svitkin et al. 2017), a robust delivery vehicle remains the major bottleneck. Among the explored nonviral vectors, lipid nanoparticles (LNPs) have become the most extensively validated and widely adopted platform. Their application is driven by highly scalable manufacturing processes and unparalleled modularity that allows for precise architectural engineering to targeted specific immune cell subsets (Aldosari et al. 2021). This review summarises strategies for targeted mRNA delivery in immune cell reprogramming. We start by outlining the biological barriers that determine nanoparticle fate from administration to functional protein expression. Next, we critically examine the evolution of delivery platforms. While various delivery modalities have been explored for genetic cargo, LNPs have emerged as the most clinically validated platform. Therefore, we focus on the engineering of LNPs and provide an in-depth analysis, including passive biodistribution, active targeting via surface-functionalized ligands and emerging stimuli-responsive systems. The discussion then highlights expanding clinical applications in oncology, protein replacement, autoimmunity, and tissue regeneration. Finally, we address remaining translational hurdles regarding safety, scalability, and regulation. Particular emphasis ia place on emerging optimisation strategies to pave the way for making these advanced therapies widely available. Biological and translational hurdles for in vivo mRNA delivery Extracellular and intracellular barriers The clinical advantages of in vivo immune cell engineering, particularly the elimination of lymphodepletion and the preservation of naive cellular phenotypes, depend on systemic mRNA delivery. Despite the success of LNP vaccines, precise mRNA delivery to specific immune cells remains a major challenge. Specifically, delivery vehicles must sequentially evade initial hepatic sequestration, bypass the splenic red pulp macrophages to access white pulp compartments, and ultimately breach the physical defenses of target lymphocytes before protein translation can occur. Understanding these anatomical and biophysical barriers is essential for rational carrier design to achieve sufficient mRNA delivery and realise the clinical potential of in vivo immune cell therapy. A systematic examination of these barriers begins with the molecular fate of the mRNA cargo itself. As a large, negatively charged biomacromolecule, naked mRNA is rapidly hydrolysed by ubiquitous serum ribonucleases (RNases) (Tsui et al. 2002; Houseley and Tollervey 2009). This results in extremely low delivery efficiency, with an estimated 0.01% of the administered dose reaching the cytosol of target cells (Huang et al. 2024). While LNP encapsulation provides structural protection against enzymatic degradation, systemic administration immediately introduces a second formidable obstacle, the formation of a protein corona. Within seconds of injection, serum proteins adsorb onto the nanoparticle surface, forming a dynamic biological layer that redefines the particle's identity and governs its subsequent biodistribution (Figure 2, Step 1) (Bashiri et al. 2023; Mahmoudi et al. 2023). Biological barriers and intracellular processing of systemically delivered mRNA-LNPs. Following intravenous administration, mRNA-LNPs encounter sequential physiological hurdles: (1) Circulation and opsonization: LNPs enter the bloodstream, where serum proteins adsorb onto the surface to form a protein corona that may alter pharmacokinetics. (2) MPS clearance: A fraction of circulating LNPs is recognised and cleared by the MPS residing in the liver and spleen. (3) Target binding: LNPs that evade systemic clearance reach the target tissue and engage specific receptors on the target cell membrane. (4) Endocytosis: LNPs are internalised via receptor-mediated uptake and become entrapped in endosomes. (5) Endosomal escape: Following endosomal acidification (pH < 6.5), the protonation of ionisable lipids within LNPs triggers membrane disruption, enabling mRNA release into the cytosol. (6a) Translation and expression: Successfully released mRNA is translated by host ribosomes to produce the encoded protein. (6b) Immune surveillance: Concurrently, exogenous mRNA or LNP components may be detected by intracellular pattern recognition receptors, including endosomal TLR 3/7/8 and cytosolic RIG-I-like receptors (RLRs; e.g. RIG-I, MDA5), potentially inducing type I interferon (IFN) expression and secretion. Abbreviations: MPS, mononuclear phagocyte system; PEG, polyethylene glycol; TLRs, Toll-like receptors; RLRs, RIG-I-like receptors; IFNs, interferons. Created in BioRender. Wang, D. (2026). The composition of protein corona acts as the primary determinant of biodistribution. Apolipoprotein E (ApoE) is a critical corona component, which endogenously directs LNPs to low-density lipoprotein (LDL) receptors on hepatocytes (Akinc et al. 2010). This process facilitates liver targeting but sequesters LNPs away from immune cells and favours hepatic clearance via the mononuclear phagocyte system (MPS) (Figure 2, Step 2) (Zelepukin et al. 2024). Furthermore, repeated LNP dosing can trigger the production of anti-PEG antibodies, which causes accelerated blood clearance (ABC) and shifts the biodistribution from target tissues to splenic antigen-presenting cells (Lila et al. 2013; Ibrahim et al. 2022). For nanoparticles that evade hepatic sequestration, the next barrier is the vascular endothelial. Crossing this barrier to reach lymphoid organs or interstitial tumour spaces remains an inefficient process governed by organ-specific vascular phenotypes and interstitial fluid pressure (Li et al. 2024). Once at the target site, cellular uptake heterogeneity poses another challenge. Phagocytic cells, such as macrophages and DCs, readily internalise LNPs (Shimosakai et al. 2022). In contrast, T cells are highly resistant to transfection, a property rooted in their biophysical architecture. T cells possess a high density of cortical actin and elevated membrane tension, both of which disfavour the membrane deformation required for endocytosis (Rahimmanesh et al. 2020). Their poor capacity for nonspecific pinocytosis or phagocytosis renders non-targeted lipid carriers inefficient. Therefore, it is necessary to exploit active targeting that engages specific surface receptors with high internalisation efficiency to initiate receptor-mediated endocytosis (Zeng et al. 2026). Alternatively, incorporation of biomimetic lipids can remodel nanoparticle architecture to facilitate direct membrane fusion (Rurik et al. 2022; Zhang et al. 2025). This biophysical constraint necessitates active targeting architectures for effective lymphocyte reprogramming. Following endocytosis, mRNA remains trapped within membrane-bound vesicles. Endosomal escape is widely regarded as the most pivotal rate-limiting step in delivery, with less than 1% of internalised mRNA is estimated to reach the cytosol of human epithelial (HTB-177) cells (Maugeri et al. 2019). The efficiency of endosomal escape relies on the precise pKa of ionisable lipids, which must undergo protonation in the acidic environment of the late endosome (pH < 5.5) to destabilise the membrane and release the cargo (Figure 2, Step 5). However, this process is highly cell-type dependent. A critical failure mechanism involves pH mismatch, where different immune cells acidify their endosomes at different extents. This difference may prevent certain cells from reaching the protonation threshold for endosomal escape, leading to lysosomal degradation of the cargo. For instance, DCs maintain a more alkaline phagosomal environment compared to the rapid acidification in macrophages. This divergence poses a challenge for conventional pH-responsive LNPs, addressing the need for rapid-response escape chemistries to prevent cargo degradation (Foote et al. 2019). After endosomal escape, mRNA must evade cytosolic surveillance mechanisms to initiate translation. The cytosol is equipped with PRRs, such as TLR3, TLR7, and RIG-I, which evolved to detect viral RNA (Figure 2, Step 6b). Activation of these sensors provokes immune signalling cascades that suppress therapeutic mRNA translation and promote the expression of type I interferons and pro-inflammatory cytokines (Diebold et al., 2004; Pichlmair et al. 2006; Pardi et al. 2018; Weidensee and Sahu 2025). Additionally, the E3 ubiquitin ligase TRIM25 and endoribonucleases, such as N4BP1, form a cytosolic surveillance system that targets exogenous mRNA for rapid turnover (Kim et al. 2025). Consequently, bypassing these innate defenses requires dual innovation, chemical modification of the mRNA cargo using N1-methylpseudouridine to evade detection and sequence engineering to minimise secondary structures recognised by sensors (Svitkin et al. 2017). Effective mRNA delivery requires a comprehensive design that navigates the cargo through both extracellular and intracellular barriers. Future developments should continue to refine LNP chemistry to address the specific bottlenecks of protein corona formation, endosomal escape, and cytosolic stability. Targeting specificity and biodistribution Although overcoming the aforementioned biological and physical barriers is a crucial step for cellular entry, the final therapeutic effect depends on the exact location of this entry. Achieving therapeutic efficacy in immunotherapy requires precise cellular reprogramming rather than systemic bioavailability. However, the inherent hepatic tropism of conventional LNPs, driven by the uptake mediated by ApoE and the clearance action carried out by the MPS, severely limits their access to immune cell subsets present in the spleen, lymph nodes, or tumour microenvironment (Shi et al. 2022). To bypass this hepatic filter without relying on exogenous ligands, the field has advanced toward a selective organ targeting (SORT) platform. This strategy operates on the principle of internal charge modulation, where incorporating a defined percentage of a permanently charged cationic, anionic, or ionisable lipid alters the global surface charge of the nanoparticle (Cheng et al. 2020). Crucially, this charge modification reshapes the protein corona rather than acting directly on cells. Specifically, neutral, cationic, and anionic SORT LNPs preferentially recruit apolipoprotein E (ApoE), vitronectin, and distinct circulating opsonins to mediate targeted delivery to liver, lung, and spleen, respectively (Wang et al. 2023). This mechanism enables the escape from the liver and reprogrammes the biological identity of the carrier. However, correcting organ tropism addresses only macroscopic biodistribution, leaving an important physiological paradox where tissue accumulation does not equate to cellular transfection. This divergence is most evident in the spleen. While nanoparticles may physically accumulate within the splenic architecture via passive filtration, they are predominantly trapped in the red pulp, which is populated by scavenger macrophages and monocytes designed to clear debris (Cheng et al. 2020; Kim et al. 2025). In contrast, the therapeutic targets, including T and B cells, reside in the white pulp, structurally segregated by the marginal zone. Standard formulations often lack the physiochemical properties or targeting ligands to traverse this anatomical barrier. As a direct result, particles are phagocytosed by red pulp macrophages before they can interact with the adaptive immune cells in the white pulp, highlighting a profound divergence between physical biodistribution and functional delivery. Bridging the gap between accumulation and transfection requires overcoming the phenomenon known as PEG dilemma. Polyethylene glycol (PEG) lipids are essential for preventing aggregation and extending circulation half-life, yet their steric hydration layer hinders direct membrane contact required for cellular uptake and creates a stealth effect that makes LNPs invisible to the immune system (Hatakeyama et al. 2011; Carrasco et al. 2021). To resolve this, active targeting strategies using high-affinity ligands such as antibodies or peptides are employed. Nevertheless, the engineering of these ligands introduces new complexities. Traditional random conjugation often results in the misalignment of the Fc region of the antibody, potentially triggering clearance by Fc-receptor-positive phagocytes rather than binding to the target antigen (Huang et al. 2024). Furthermore, excessively high ligand densities can compromise the fusogenicity of the lipid bilayer, preventing endosomal escape even after successful uptake. To overcome these engineering complexities, more sophisticated surface engineering approaches, such as oriented ligand conjugation strategies, will be explored in Section 4. Delivery platforms for targeted mRNA delivery to immune cells To overcome the biological barriers and achieve the precise delivery to specific immune cells, various delivery systems have been developed for transient mRNA delivery. While historical paradigms relied heavily on DNA-integrating viral vectors, such as adeno-associated virus, lentivirus (LV) and retrovirus, the potential risks of insertional mutagenesis have driven the field toward non-integrating mRNA systems. These transient platforms primarily include lipid-based systems, polymer nanoparticles, exosomes, and biomimetic vehicles (Figure 3a). Among them, LNPs have become the prime and most clinically advanced delivery platform (Figure 3b). Therefore, this section will predominantly focus on the architectural and chemical engineering of LNPs, followed by a comprehensive overview of alternative synthetic and biomimetic nanocarriers. Representative nanocarrier platforms and detailed structural composition of LNP. (a) Overview of lipid-based, synthetic, and biomimetic nanocarriers for mRNA delivery, including virus-like particles (VLPs), dendrimers, peptide-based carriers, polymer nanoparticles, lipid nanoparticles (LNPs), and exosomes. (b) Detailed schematic of a targeted LNP formulated for mRNA delivery. The LNP core encapsulates mRNA molecules and consists of four key components: PEGylated lipids, ionisable lipids, helper lipids, and cholesterol. The surface of the LNP is functionalized with targeting antibodies or ligands to facilitate targeted delivery to distinct cell types or tissues. Abbreviations: VLPs, virus-like particles; LNP, lipid nanoparticle; PEG, polyethylene glycol. Created in BioRender. Wang, D. (2026). Lipid nanoparticles (LNPs) Lipid nanoparticles serve as the cornerstone of nonviral mRNA delivery, with a modular architecture in which core components and their supramolecular organisation critically determine tropism, transfection efficiency, and therapeutic efficacy. The canonical formulation consists of four elements, ionisable lipids which drive endosomal escape via pH-responsive protonation, cholesterol which modulates membrane fluidity and fusion, helper phospholipids which facilitate lipid polymorphism, and PEG-lipids which provide steric stabilisation to extend systemic circulation (Figure 3c) (Song et al. 2024). These components define the physicochemical identity of the particle, including its size, surface charge, stability and protein corona composition, thereby determining its biological fate (Jain et al. 2024). Chemical evolution of ionisable lipids The functional efficacy of the carrier relies heavily on the ionisable lipid, which comprises a headgroup, a linker, and hydrophobic tails, that determine pKa, fusogenicity, and biodegradability (Eygeris et al. 2021). In acidic buffers, ionisable lipids become protonated to electrostatically condense mRNA, while their neutral state at physiological pH minimises cytotoxicity and prolongs stability. Upon cellular uptake, endosomal acidification triggers re-protonation, facilitating membrane destabilization and cytosolic release (Algarni et al. 2022). Traditionally, optimisation of ionisable lipids has depended on iterative structure-activity relationship studies by tuning the amine pKa, introducing biodegradable linkers, and modifying hydrophobic tails, which is labour intensive and time consuming. Recent advances in high-throughput screening have accelerated structure-activity relationship discovery. For instance, Hamilton and colleagues synthesised a library of lipids varying in tail length and polyamine core structure, revealing that subtle modifications, such as combining C16 alkyl tails with specific amine cores, altered tropism from the liver to the spleen and lymph nodes (Hamilton et al. 2024). Complementing synthetic libraries, nature-inspired designs offer a refined approach to cell-specific targeting. Drawing inspiration from natural terpenes, C. Zhang and colleagues developed a library of unsaturated lipids via Michael addition and identified the farnesol-derived lipid H4T3 as a potent vehicle comparable to clinical benchmarks (Zhang et al. 2025). Similarly, to address the biophysical recalcitrance of T cell membranes, researchers developed cardiolipin-mimetic phosphoramidate lipids, referred to as CAMPs. These analogues remodel the nanoparticle architecture by increasing particle rigidity and promoting membrane phase separation, enhancing internalisation by hard-to-transfect T cells (Zhang et al. 2025). The optimisation efforts are increasingly focused on the engineering of specific lipid geometries and functional groups to coordinate the intracellular behaviour of the carrier. Fenton and colleagues found that cis-alkene geometry in linoleic acid-derived lipids is a pivotal determinant of intracellular trafficking efficiency (Fenton et al. 2016). This enhanced membrane fusogenicity is particularly crucial for immune cell reprogramming, as primary lymphocytes, notably T cells, exhibit a high membrane rigidity and dense actin cortex that render them inherently resistant to conventional lipid-mediated endocytosis and endosomal escape (Billingsley et al. 2020). In parallel, to counter the self-adjuvanting immunogenicity of these carriers, Gong and colleagues introduced an antioxidant lipid bearing a phenolic moiety known as C-a16. This design actively quenches reactive oxygen species to mitigate innate immune activation and significantly prolong protein expression (Gong et al. 2025). Finally, helper lipids have evolved from passive stabilisers to active functional components. The engineering of cubic phase-inducible zwitterionic phospholipids exemplifies this trend, where specific ion-pairing with endosomal lipids promotes non-lamellar phase transitions essential for cytosolic release (Iwakawa et al. 2025). Stoichiometry and architectural engineering Beyond ionisable lipid structure, the stoichiometry of helper lipids, cholesterol, and PEG-lipids is a fundamental, independently tunable parameter in LNP engineering. Using a Taguchi L16(44) orthogonal design, Hamilton and colleagues demonstrated that optimal excipient combinations are inherently ionisable-lipid-dependent, underscoring the necessity of lipid-matched formulation screening to maximise immune cell uptake and transfection (Hamilton et al. 2024). The precise titration of helper lipids can also mediate organ tropism. For example, Guerrero and colleagues revealed that increasing the phospholipid molar ratio altered nanoparticle morphology and protein corona composition, further enhancing membrane fusogenicity and refining tissue targeting specificity (Guerrero et al. 2026). Similarly, tuning the DSPC molar percentage in ionisable tri-oleoyl-Tris (iTOT)-LNPs shifted the tropism from liver to spleen, with a formulation containing 15 mol% DSPC achieving maximal uptake by splenic B cells for effective mRNA vaccine delivery (Suzuki et al. 2025). The lipid/mRNA mass ratio is another key formulation parameter for organ tropism. Zhao and colleagues revealed that adjusting the polymer-to-mRNA mass ratio in the mass-ratio-controlled organ-selective (MACO) platform reversibly switches transfection among the spleen, liver, and lungs. This feature arises from mass-ratio-dependent pKa shifts and protein corona fingerprints operating independently of particle size (Zhao et al. 2025). In addition to the standard four-component systems, non-canonical formulations have also been explored to increase the transfection efficiency and delivery specificity. A landmark example is the SORT platform. As previously discussed in Section 2.2 regarding macroscopic biodistribution, Cheng and colleagues demonstrated that simply incorporating a fifth component to modulate the internal charge balance could redirect tissue tropism (Cheng et al. 2020). From a formulation perspective, this stoichiometric addition predictably alters the tropism without requiring chemical redesign of the core ionisable lipid. In the context of immune cell therapy, this architectural tuning is not merely about achieving organ-level accumulation; it is a prerequisite for bypassing the non-specific scavenger function of hepatic Kupffer cells and splenic red pulp macrophages, thereby enabling the intact LNP to penetrate lymphoid tissues and actively transfect target T and B cells resident in the white pulp (Kim et al. 2025). Building on this formulation principle, Dual SORT systems have recently been engineered to simultaneously target the liver and lungs, enabling multi-organ genome editing for systemic diseases (Kim et al. 2025). Formulation simplification represents a similar trend challenging this assumption. Zhang and colleagues developed a phospholipid-free, three-component system that achieved exclusive hepatocyte transfection with minimal splenic accumulation (Zhang et al. 2025). Reducing this multi-component paradigm to its minimal form, another study developed one-component secondary amine lipids that function as standalone carriers. These lipids exhibit enhanced mRNA binding and mediate potent spleen-specific T cell transfection. This finding is particularly disruptive, as it fundamentally challenges the decades-long dogma that a multi-component architecture is necessary for functional mRNA delivery, potentially ushering in an era of minimalist carrier design (Zhang et al. 2024). Synthetic and biomimetic nanocarriers Beyond lipid-based formulations, alternative nanocarriers have been developed to address the specific challenges of immune cell targeting. These platforms, ranging from synthetic polymers to biologically derived vehicles, offer distinct mechanisms for mRNA protection, cellular uptake, and intracellular release. Synthetic polymeric platforms Alternative nanoparticle delivery systems have been developed to complement LNPs, by focusing on immune cell-specific targeting and reduced systemic toxicity. Polymeric systems offer a versatile alternative with tunable chemical structures. A prominent example is the charge-altering releasable transporter (CART) platform. These poly(alpha-amino ester)-based materials efficiently encapsulate mRNA via electrostatic interactions and, upon cellular entry, undergo self-immolative degradation that releases the mRNA payload. Studies have demonstrated efficient delivery to lymphocytes (T cells ~1.6%, B cells ~11%), dendritic cells (~60%) and macrophages (~60%) following intravenous injection (McKinlay et al. 2017). Similarly, biodegradable poly(beta-amino ester)s (PBAE) formulated with PEG-lipid have been optimised for lung-specific targeting (Kaczmarek et al. 2018). Polyethylenimine (PEI), although traditionally limited by toxicity, has been reengineered through modification with biocompatible components such as PEG and redox-sensitive disulphide linkers to ensure intracellular cargo release (Kowalski et al. 2019). To further enhance structural precision, dendrimers, characterised by their highly branched and star-shaped architecture, have been developed for targeted immunotherapy. Compared to linear polymers, dendrimers possess a monodisperse architecture enabling precise control over size and surface functionality. A pioneering study by Chahal et al. demonstrated that a modified amphiphilic dendrimer system could deliver self-replicating mRNA replicons effectively to lymph nodes, generating protective immunity against lethal viral challenges with a single dose (Chahal et al. 2016). The rational design and screening of a polyvalent ionisable cationic dendrimer library has led to the development of dendrimer-based apolipoprotein nanoparticles (aNPs), significantly enhancing systemic mRNA delivery to immune cells (Trines et al. 2025). Complementing the synthetic constructs, natural polysaccharide-based polymers provide an intrinsic targeting advantage. Hyaluronic acid (HA), a natural glycosaminoglycan, specifically binds to the CD44 receptor that is upregulated on activated T cells and macrophages in the tumour microenvironment (Platt and Szoka 2008). Tang et al. developed an inhaled mRNA delivery system by coating cationic lipid nanoparticles with HA, successfully delivering mRNA encoding the p53 tumour suppressor to both CD44-overexpressing lung cancer cells and inflammatory macrophages in vivo. This work underscores the potential of HA-functionalized hybrid systems to overcome biological barriers and achieve cell-specific treatment in the lung immune microenvironment (Tang et al. 2023). Peptide-based carriers and capsid assemblies In parallel with synthetic polymers, peptide-based nanoparticles harness intrinsic biological properties for mRNA delivery. Protamine, an arginine-rich cationic polypeptide, represents a pioneering nonviral carrier. It effectively condenses mRNA into compact and stable nanocomplexes, shielding the payload from serum RNases. Despite lower transfection efficiency than modern LNPs for systemic delivery, protamine-based RNA vaccines, such as the RNActive® platform, are notable for inducing robust Th1-biased immune responses due to their inherent adjuvant properties (Fotin-Mleczek et al. 2011). To circumvent biological barriers, cationic cell-penetrating peptides (CPPs), which are short sequences of 8 to 30 amino acids, have been employed. Compared to carriers that rely exclusively on endocytosis, CPPs can facilitate direct transmembrane penetration or endosomal disruption via receptor-independent mechanisms, thereby promoting the cytosolic release of antigen-encoding mRNA to prime cytotoxic T lymphocytes (CTLs) (Yokoo et al. 2021). Moving beyond simple peptides, alternative supramolecular assemblies take direct inspiration from nature. VLPs are self-assembled viral capsid proteins and naturally recognised by PRRs on innate immune cells, making them ideal dual-function platforms that serve as both a delivery vehicle and a potent immunostimulant. Early efforts utilised simple exogenous protein capsids, such as bacteriophages and plant viruses. For instance, recombinant bacteriophage MS2 VLPs have been engineered to protect and deliver mRNA encoding prostate cancer antigens to DCs, eliciting substantial immune activation (Li et al. 2014). Similarly, cowpea chlorotic mottle virus (CCMV) capsids have been successfully reassembled in vitro to encapsulate heterologous mRNA. Due to their distinct repetitive surface structures, these plant-derived nanoparticles can facilitate uptake by mammalian antigen-presenting cells and elicit effective B-cell responses, positioning them as attractive candidates for cancer vaccine applications (Azizgolshani et al. 2013). By contrast, the exploration of endogenous mammalian retrovirus-like proteins represents a distinct conceptual transition in delivery. Segel et al. identified PEG10, a protein derived from a domesticated retrotransposon in the human genome, which naturally forms virus-like capsids. This selective endogenous encapsidation for cellular delivery, termed SEND, can be reprogrammed to package and secrete specific mRNA cargoes by recognising distinct flanking 5′ and 3′ untranslated regions (Segel et al. 2021). This phenomenon has also been observed in the neuronal gene Arc, which forms retrovirus-like capsids for RNA transfer between neurons, suggesting a broader biological role for retroviral elements in intercellular RNA communication (Pastuzyn et al. 2018). This endogenous VLP approach offers a compelling strategy to mitigate the immunogenicity associated with foreign viral vectors, providing potentially stealth properties for systemic gene delivery. Biomimetic & hybrid systems To bridge the gap between synthetic tunability and biological functionality, biomimetic and hybrid systems adopt complex biological features and represent a frontier in mRNA delivery. Extracellular vesicles (EVs), particularly exosomes, are naturally secreted lipid bilayers that possess favourable biocompatibility and the capacity to cross biological barriers, including the blood-brain barrier. To address inefficient loading of exogenous mRNA, engineering strategies have focused on generating hybrid vesicles by fusing engineered nanoparticles with exosomes. A recent study introduced a programmable fusion strategy by employing complementary lipidated DNA (LiNA) sequences on both EVs and liposomes that hybridise in a zipper-like manner to bring the membranes into close proximity and mediate fusion to form an mRNA-loaded single hybrid particle. This approach combines the high mRNA encapsulation efficiency of synthetic lipids with the natural homing capabilities of EVs, creating hybrid vectors that facilitate effective mRNA delivery with reduced cytotoxicity (Malle et al. 2024). Complementary to intact vesicles, a top-down biomimetic strategy involves cloaking synthetic nanoparticles with natural cell membranes, thereby translocating the complex antigenic profile of the source cells onto the nanoparticle surface to confer specific biological properties. Initial applications primarily used erythrocyte membranes to display self-markers, such as CD47, that mitigate phagocytic clearance and extend circulation half-life. Expanding this concept to active targeting, cancer cell membranes have been utilised for their distinct homotypic targeting capability, defined as the innate tendency of tumour cells to self-aggregate, enabling preferential accumulation at the primary tumour site (Fang et al. 2014). Representing a rapidly evolving frontier for immunotherapy, immune cell membranes are increasingly being exploited to target inflammation and lymphoid tissues. For example, Li et al. developed a DC membrane-coated nanoparticle system for mRNA vaccination. These nanoparticles retained the antigen-presentation machinery and co-stimulatory signals from the parent DCs. In vivo studies indicated that this biomimetic coating enhanced uptake by lymph node-resident DCs and elicited robust antigen-specific CD8+ T cell responses compared to uncoated controls, effectively leveraging the host's immune recognition mechanisms to achieve therapeutic effects (Li et al. 2024). While synthetic polymers, virus-like particles, and biomimetic systems each offer unique biological mechanisms for circumventing delivery barriers, their broader clinical translation is constrained by intrinsic limitations, such as immunogenicity, manufacturing scalability, or product consistency. In contrast, LNPs have emerged as the most clinically validated platform for mRNA delivery, owing to their favourable safety profile, scalable manufacturing, and highly modular architecture that enables precise structural engineering. Consequently, the following sections focus on advanced LNP engineering strategies, including surface functionalization and stimuli-responsive modifications to enhance targeted mRNA delivery to specific immune cell populations. Strategies to enhance targeted delivery Within the specific context of LNP platforms, although passive targeting adjusts organ-level tropism by the physicochemical tuning of lipid compositions, achieving cellular precision requires active surface engineering. Conventional lipid nanoparticles exhibit intrinsic hepatocyte tropism due to ApoE adsorption. Redirecting the biodistribution to extrahepatic immune subsets requires decorating the particle surface with high-affinity moieties (Mitchell et al. 2021). This engineering challenge has evolved from simple surface conjugation into a sophisticated design matrix involving three core parameters: ligand multivalence, orientational control, and receptor internalisation kinetics. Ligand and aptamer-mediated targeting Ligand-based targeting offers distinct advantages in chemical simplicity, low immunogenicity, and the ability to cross biological barriers that exclude bulky molecules. This approach is particularly effective for targeting antigen-presenting cells, because these cells constitutively express C-type lectin receptors for carbohydrate-based recognition and uptake. However, unlike antibodies, individual carbohydrate ligands typically exhibit weak affinity. Therefore, successful targeting relies on the cluster glycoside effect, which utilises the multivalent presentation of ligands at high density to trigger receptor clustering and enhance receptor binding. To exploit this effect, researchers have developed sophisticated molecular and structural multivalent strategies. For instance, researchers developed an advanced hybrid lipid polymer shell mRNA nanoparticle equipped with a multivalent trimannose sugar tree. This branched ligand architecture structurally mimics complex pathogenic glycans, significantly increasing the binding affinity to the mannose receptor (CD206) on dendritic cells. The enhanced receptor-specific engagement not only facilitated receptor-mediated endocytosis but also promoted the maturation of antigen-presenting cells, thereby potentiating the immune response (Van Der Jeught et al. 2018). In addition to structurally branched molecules, the dense spatial array of monomeric sugars on a nanoparticle surface can similarly trigger the cluster glycoside effect. Lei et al. developed mannosylated lipid nanoparticles (LNPs) that leverage a high-density multivalent presentation of mannose lipids. This spatial multivalence achieved exceptional antigen presentation efficiency, requiring only one-fifth of the standard therapeutic dose for mRNA cancer vaccines. Specifically, this multivalent engagement acquired unexpected biological benefits. By physically occupying multiple CD206 receptors, these LNPs actively blocked the CD206/CD45 signalling axis, which subsequently leads to the downregulation of the immune checkpoint CTLA-4 on T cells (Lei et al. 2024). Similarly, Yin and colleagues utilised engineered Sindbis virus glycoproteins to pseudotype virus-like particles. By mimicking the repetitive geometric array of viral surface proteins, this system achieved precise recognition of the DC-specific C-type lectin DC-SIGN through multivalent glycan-receptor interactions akin to he cluster glycoside effect described above. This structural mimicry elicited IgG titre nearly an order of magnitude higher than non-targeted controls, providing protection against Spike protein of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and herpes simplex virus 1 (HSV-1) (Yin et al. 2024). In addition to natural ligands, nucleic acid aptamers represent a class of synthetic ligands combining high specificity with chemical evolvability. Cui and colleagues developed the selective cell in organ targeting (SCORT) platform, incorporating an E3 aptamer targeting Transferrin Receptor 1. By optimising the PEG-linker length to ensure the aptamer extended beyond the protein corona, they achieved preferential accumulation in metastatic prostate cancer lesions over healthy liver tissue (Cui et al. 2025). These examples illustrate that when engineered with appropriate spatial density and linker architecture, small-molecule ligands can effectively navigate biological barriers (Zhang et al. 2025). Antibody conjugation and modular assembly Antibodies provide unparalleled specificity for distinguishing extremely similar immune subsets. However, their clinical translation has been hampered by the conjugation paradox. Conventional chemistries, such as N-hydroxysuccinimide (NHS)-ester coupling, attach antibodies randomly via lysine residues. This stochastic process often occludes the antigen-binding domains impairing binding affinity, or exposes the Fc region triggering clearance by Fc-receptor-positive phagocytes rather than binding to the target cell (Puertas et al. 2010; Xu et al. 2025). To resolve this, recent innovations exploit specific adaptor domains to capture antibodies in a defined orientation. Xu and colleagues proposed a modular assembly platform for lipid nanoparticles. This engineered system utilises monomeric streptavidin fused to Fc-binding domains, creating a universal linker that enables the direct plug-and-play capture of unmodified antibodies, achieving up to 98% transfection efficiency in primary mouse T cells (Xu et al. 2025). In parallel, Chen and colleagues developed a versatile antibody-capture system using the TP1107 nanobody, which binds IgG1 Fc domains in a side-on configuration. This orientational control increased target binding by 8-fold relative to random coupling, which enabled rapid screening of CD3, CD4, and CD5 targets in human PBMCs (Chen et al. 2025). Extending this modular concept to off-the-shelf utility, Park and colleagues introduced the Grab Antibody (GrAb) platform, where an antibody is fused to apolipoprotein A1 (ApoA1), integral to high-density lipoproteins (HDL). Utilising the strong lipid-binding affinity of ApoA1, these fusion proteins spontaneously assemble into a functional corona on the LNP surface, thereby circumventing complex chemical conjugation steps and offering a scalable manufacturing path (Park et al. 2025). A critical insight emerging from recent comparative studies is that receptor internalisation kinetics, rather than surface abundance or binding affinity alone, largely govern mRNA delivery efficiency. Zeng and colleagues systematically compared LNPs targeting CD4, CD5, CD8, and CD7. Counterintuitively, formulations targeting CD7, a marker with lower basal expression, consistently outperformed others in transfecting primary T cells, which was attributed to the rapid endocytic turnover of the CD7-antibody complex, which shuttles the payload intracellularly before clearance (Zeng et al. 2026). This kinetic advantage translated into remarkable potency, with CD7-targeted LNPs requiring only 2.5 μg of mRNA for robust in vivo CAR-T generation, substantially lower than the doses needed for CD5- or CD8-targeted benchmarks (Rurik et al. 2022; Hunter et al. 2025). These findings indicate that effective targeted delivery requires not only high-affinity binding but also strategic receptor selection. Moreover, the relationship between ligand density and delivery outcome is not monotonic. Hashiba and colleagues recently established a quantitative framework by conjugating single-domain antibodies (VHHs) to LNPs with site-specific orientation, isolating surface density as an independent design variable. Single-particle analysis revealed a bell-shaped dependence of mRNA delivery on VHH density. An optimal density of approximately 0.1 VHH per 100 nm2 maximised transfection and enabled functional in vivo CAR-T generation at doses of 10 to 30 μg/kg, whereas supra-optimal densities paradoxically reduced cellular uptake (Hashiba et al. 2026). This decline was driven by continuous receptor internalisation and proteolytic degradation rather than transient receptor blockade, indicating that successful antibody-functionalized LNPs require a careful balance between multivalent target engagement and preservation of receptor availability. From a translational perspective, such rapid receptor downregulation may transiently reduce target accessibility during repeated dosing, thereby posing an additional challenge for sustained therapeutic efficacy. Spatiotemporal regulation and microenvironmental modulation Despite advanced antibody-conjugation strategies, the precise spatiotemporal control of mRNA expression, particularly restricted translation within diseased tissues, remains an unmet need requiring further investigation. Utilising stimuli-responsive activation enables separating general tissue biodistribution from specific therapeutic action, thereby mitigating off-target expression while increasing target tissue efficacy. Exogenous physical triggers for spatiotemporal precision To achieve on-demand control over protein expression, external physical stimuli offer superior spatiotemporal resolution. A notable strategy is the light-induced transfection system (LITS) introduced by Chen et al., which utilises photosensitising polymers to form PEGylated nanoparticles, thereby maintaining stability during systemic circulation. Once endocytosis, the acidic endosomal environment renders the photosensitizer responsive to specific wavelengths of light. Irradiation with approximately 650 nm red light triggers localised hyperthermia and reactive oxygen species (ROS) generation, which disrupt endosomal membranes. Specially, this stimuli-responsive mechanism facilitated rapid cytosolic escape, reducing endosomal co-localisation of mRNA by approximately 25% within 20 minutes. The confined endosomal escape and protein translation within irradiated tumour sites effectively minimised off-target systemic toxicity (Chen et al. 2025). Optical methods provide high resolution, but their application is constrained by limited tissue penetration. To address this challenge in deep-seated organs, focused ultrasound (FUS) has emerged as a powerful alternative. By synergising microbubbles and thermosensitive nanocarriers, FUS can mechanically disrupt cellular barriers and induce local hyperthermia at precise focal points deep within the body. This acoustic method has been successfully applied to transiently permeabilize the blood-brain barrier, facilitating the site-specific delivery of mRNA therapeutics to the central nervous system with minimised systemic exposure (Kwak et al. 2024). Furthermore, ultrasound-responsive formulations utilising fluorinated PEG-lipids enable on-demand PEG shedding. Upon sonication at the target site, the protective steric layer detaches to expose the reactive core, facilitating rapid uptake by localised immune subsets such as splenic antigen-presenting cells (Chen et al. 2025). These research findings demonstrate an advancement from static chemical configurations to dynamically stimulus-responsive systems. Supported by cutting-edge microfluidic technologies and machine learning that predict complex structure-function relationships, such platforms integrate external control over transfection kinetics to maximise therapeutic precision (Waheed et al. 2022; Prazeres et al. 2025). Endogenous microenvironmental gating Complementing external triggers, intrinsic microenvironmental gradients provide a robust mechanism for autonomous activation, particularly in deep-seated tissues where applying external stimuli is still challenging. Taking the advantage of specific physiological hallmarks, including acidosis, redox potential imbalances, and enzymatic overactivity, recent strategies have advanced beyond passive accumulation to achieve actively logic-gated payload release. First, the acidic tumour microenvironment (pH 6.5–6.8), driven by the Warburg effect, represents a fundamental target for metabolic gating. Compared to the stringent pH responsiveness required for endosomal escape, TME-responsive systems often employ charge-reversal strategies to distinguish malignant tissues from physiological circulation (pH 7.4). For example, an innovative study demonstrated that polymers modified with acid-labile amides (e.g. 2,3-dimethylmaleic anhydride) can maintain a shielded profile in the bloodstream yet undergo selective hydrolysis within the acidic TME. This transition facilitates the shedding of protective PEG layers (de-PEGylation) and regenerates cationic surface charges to enhance interaction with tumour cell membranes, thereby acting as a prerequisite for internalisation (Du et al. 2011). Second, redox dysregulation offers a distinct intracellular trigger. The TME is characterised by elevated levels of ROS and glutathione (GSH). To exploit oxidative stress, Zhou et al. developed a ROS-responsive polymeric nanoparticle platform incorporating thioketal linkers. These linkers are designed to preserve stability in the bloodstream, but undergo selective cleavage upon exposure to the high intracellular ROS levels in cancer cells. This degradation mechanism triggers nanoparticle disassembly and subsequent release of mRNA payload. When deliver p53-encoding mRNA, this system demonstrated robust antitumor efficacy in lung cancer models by restoring p53 expression in tumour cells while minimising off-target effects in normal tissues (Zhou et al. 2023). Alternatively, bioreducible strategies exploit the steep cytosolic GSH gradient (2−10 mM intracellular vs <10 µM extracellular) for selective payload release (Schafer and Buettner 2001). By incorporating disulphide-linked lipidoids, these systems facilitate mRNA payload release predominantly within the reducing cytosol of target cells (Qiu et al. 2021). Finally, enzymatic gating offers perhaps the highest degree of biological specificity. Matrix metalloproteinases (particularly MMP-2 and MMP-9), which are markedly upregulated in metastatic niches, act as specific molecular triggers for mRNA carriers (Kessenbrock et al. 2010). Advanced systems functionalized with MMP-cleavable peptide linkers (e.g. GPLGVRG sequences) shield the mRNA-LNP interface during circulation. They selectively exposed cationic fusogenic domains upon enzymatic cleavage within the TME (Hatakeyama et al. 2007). Crucially, next-generation therapeutics are evolving towards AND logic gates that integrate concurrent signals to distinguish malignant tissues from benign inflammatory sites. For instance, hierarchical designs employ sequential activation: an outer MMP-sensitive shield is initially removed, revealing an inner pH-responsive core. This strategy restrictes fusogenic activity to microenvironments where both enzymatic and acidic criteria are met, thereby maximising the therapeutic index (Torchilin 2014). Host cellular state modulation The therapeutic potency of mRNA is governed not only by the delivery vehicle but fundamentally by the translational competency of the host cell itself. Even with efficient cytosolic delivery, exhausted or metabolically quiescent immune cells often exhibit a compromised capacity to generate therapeutic proteins at functional thresholds. Addressing this biological bottleneck, Tilsed et al. demonstrated that temporal modulation of the cellular state via cytokine pretreatment can markedly potentiate the intracellular processing of delivered mRNA. Their finding revealed that IL-7 signalling activates the mTORC1 axis, a master regulator of protein synthesis, thereby specifically enhancing ribosomal translation efficiency in T cells without necessarily altering nanoparticle uptake. When combined with targeted lipid nanoparticle (tLNP) administration, this biological conditioning elicited robust protein expression in both CD4+ and CD8+ T cells in vitro and substantially enhanced transgene expression in vivo (Tilsed et al. 2024). It manifests that synchronising mRNA delivery with a cytokine-primed cellular state can synergistically amplify therapeutic outcomes, which indicates the importance of modulating the host cells for optimal efficacy. Collectively, the successful in vivo engineering of immune cells relies on a multifaceted and synergistic optimisation of delivery systems (Figure 4). As illustrated, these enhancement strategies primarily focus on three interconnected pillars: (a) Targeting precision, which overcomes the inherent resistance of immune cells to conventional transfection through high-throughput lipid screening, rational formulation optimisation, and oriented antibody conjugation to facilitate rapid receptor-mediated endocytosis; (b) Spatiotemporal control, which employs exogenous physical triggers (e.g. light, ultrasound) or endogenous microenvironmental cues (e.g. low pH, reactive oxygen species, and matrix metalloproteinases) to achieve on-demand endosomal escape, deep tissue focusing, and logic-gated activation; and (c) Host cell modulation and persistence, which leverages co-delivery of cytokines (e.g. IL-7) to activate the mTORC1 axis for enhanced translational efficiency or incorporates anti-apoptotic elements to disrupt the FAS ligand-FAS autoregulatory circuit, thereby mitigating T-cell exhaustion and apoptosis following transient mRNA expression. The integration of these engineering principles is progressively transforming passive delivery vehicles into intelligent, programmable in vivo bioreactors, laying a foundation for the diverse therapeutic applications discussed in the following sections. Engineering strategies to enhance targeted mRNA delivery. The diagram illustrates three complementary strategies for improving targeted in vivo mRNA-LNP delivery and immune cell engineering. (a) Targeting precision involves lipid screening and formulation optimisation to identify optimal LNP compositions, coupled with antibody-mediated targeting. Oriented antibody conjugation on the LNP surface enables specific receptor-mediated uptake by target immune cells, whereas random antibody conjugation may lead to suboptimal orientations and reduced binding efficiency. (b) Spatiotemporal control encompasses two classes of triggers for mRNA release. External triggers utilise light to activate a photosensitizer for endosomal escape and focused ultrasound to target LNPs in deep tissues. Endogenous triggers exploit characteristic features of the TME, including acidic pH, ROS, and MMPs, which can be integrated through AND logic-gated peptide linkers for conditional payload release. (c) Host cell modulation focuses on optimising intracellular translation and engineered-cell persistence after mRNA delivery. IL-7 signalling activates mTORC1 to enhance translational efficiency and CAR expression, while modulation of the FAS-L/FAS autoregulatory circuit enhances CAR-T cell persistence and prevents activation-induced cell death. Abbreviations: MMP, matrix metalloproteinase; ROS, reactive oxygen species; FAS-L, FAS ligand; mTORC1, mechanistic target of rapamycin complex 1. Created in BioRender. Wang, D. (2026). Therapeutic applications of targeted mRNA delivery The widespread deployment of mRNA vaccines during the COVID-19 pandemic validated the safety and scalability of nonviral delivery platforms. Yet, the translational utility of this technology extends well beyond infectious disease prophylaxis. mRNA therapeutics are rapidly evolving from transient antigen expression toward programmable nanomedicine, in which the host functions as a bioreactor for therapeutic intervention. Realising this potential requires overcoming substantial physiological barriers to achieve highly specific delivery to defined cell subsets while minimising nonspecific systemic activation. The precision targeting strategies discussed previously, including molecular interface remodelling, antibody-directed targeting, and stimuli-responsive delivery, establish the technological foundation necessary to address these challenges. By refining delivery precision, these advances circumvent the logistical constraints and economic burdens of ex vivo cell manufacturing, thereby enabling the broader clinical translation of in vivo mRNA reprogramming across diverse therapeutic applications, including oncology, protein replacement therapy, autoimmune diseases, and tissue regeneration. Oncology: potentiating antitumor immunity mRNA therapeutics in oncology have developed two synergistic strategies: the activation of endogenous T cell responses via personalised vaccination and the direct reprogramming of immune cells within the host to express the receptors. Personalised cancer vaccines Personalised cancer vaccines (PCVs) represent the forefront of mRNA oncology, designed to elicit high-avidity T cell responses against tumour-specific neoantigens. Unlike off-the-shelf vaccines targeting shared antigens, PCVs employ next-generation sequencing of patient biopsies to identify non-synonymous somatic mutations. AI-driven algorithms then predict which mutated peptides optimise presentation by the patient’s specific human leucocyte antigen alleles, creating a bespoke immunogen profile unique to each individual. In high-mutational burden tumours such as melanoma, the combination of mRNA-4157 (V940) vaccine with anti-PD-1 checkpoint blockade has demonstrated marked clinical synergy. The vaccine primes a de novo pool of neoantigen-specific T cells, while the checkpoint inhibitor prevents functional exhaustion within the tumour microenvironment. Data from the Phase 2b KEYNOTE-942 trial indicated that this regimen reduced the risk of recurrence or death by 44% compared to pembrolizumab monotherapy in resected high-risk melanoma (Khattak et al. 2023). These findings underscore the ability of mRNA vaccination to provide the immunological priming often absent in tumours with low baseline T cell infiltration, effectively converting immunologically ‘cold’ tumours into ‘hot’ microenvironments responsive to checkpoint inhibition. This combination approach holds particular promise for immune-desert tumours, where checkpoint blockade monotherapy typically fails. Supporting this strategy, Rojas et al. reported that a uridine-based mRNA-lipoplex vaccine delivering neoantigens derived from resected pancreatic ductal adenocarcinoma (PDAC) induced high levels of neoantigen-specific T cells in half of the treated patients. Importantly, single-cell analysis revealed that these vaccine-expanded T cells were not transient effectors but persisted in peripheral blood for up to two years, establishing a durable immunological memory that correlated with delayed disease recurrence (Rojas et al. 2023). In vivo immune cell engineering While vaccines amplify the endogenous repertoire of tumour-specific T cells, in vivo engineering facilitates the direct equipping of immune cells with synthetic recognition domains. Conventional ex vivo therapies face profound hurdles including high production costs, prolonged vein-to-vein timelines, and product variability arising from T-cell exhaustion during expansion. Emerging strategies leverage targeted LNPs or viral vectors to reprogramme circulating lymphocytes directly within the host, thereby circumventing these limitations. A critical determinant of therapeutic efficacy is the functional persistence of engineered cells. Unlike viral integration, which results in permanent genomic alteration, mRNA-mediated expression is inherently transient. Recent mechanistic insights reveal that persistence is governed not only by expression kinetics but also by intrinsic cellular circuits. Yi et al. identified a FAS ligand (FAS-L)-FAS autoregulatory circuit as a central switch controlling CAR-T cell longevity. Strong CAR signalling upregulates FAS expression, while the same cells concomitantly express FAS-L. Engagement of FAS by FAS-L induces apoptosis, thereby limiting their persistence (Yi et al. 2025). To counteract this, next-generation mRNA constructs are being designed to co-deliver dominant-negative FAS variants or gene-editing payloads to disrupt this feedback loop, potentially enabling the transition of in situ generated CAR-T cells from a self-limiting to a more persistent phenotype. Complementing these approaches, cytokine co-expression can enhance cell proliferation and survival. mRNA-LNPs encoding both a CAR and IL-7 successfully edited T cells in vivo, enhancing expansion and antitumor efficacy against melanoma models (Li et al. 2024). To mitigate systemic cytokine toxicity, Ravichandran et al. developed a protease-activatable masked IL-12 mRNA. The engineered cytokine remains inert in systemic circulation but is cleaved and activated by metalloproteases specific to the tumour microenvironment. This TME-response approach achieved potent antitumor efficacy with a 30-fold improvement in tolerability compared to unmasked cytokines, illustrating the potential of mRNA to deliver complex synthetic biology circuits (Ravichandran et al. 2025). Collectively, these studies demonstrate that durable in vivo CAR engineering depends not only on efficient mRNA delivery but also on rational modulation of intrinsic survival pathways and functional support programmes. While T cells dominate the current clinical landscape of in vivo engineering, their efficacy in solid tumours is frequently hindered by dense physical barriers and the immunosuppressive tumour microenvironment. Consequently, the scope of targeted delivery is rapidly expanding beyond adaptive immunity to utilise innate immune subsets. Myeloid Therapeutics has advanced the MT-302 programme, utilising LNPs to directly reprogramme myeloid cells in vivo to express Trophoblast cell-surface antigen 2 (TROP2)-targeting CARs. Compared with conventional CAR-T cells, these engineered myeloid cells exhibit superior tumour infiltration and can remodel the immunosuppressive microenvironment (Lemech et al. 2025). Research by Gu et al. established an intraperitoneal mRNA-LNP delivery system to reprogramme macrophages into CAR-M in situ. Functionalized with CD3ζ or Dectin-1 signalling domains, these macrophages demonstrated enhanced phagocytosis of peritoneal metastases in colorectal and pancreatic cancer models (Gu et al. 2025). Most recently, a fluorinated ionisable lipid platform (F5-LNPs) enabled systemic intravenous delivery of human prostate-specific membrane antigen (PSMA)-targeted CAR (F5-CAR) mRNA to macrophages, achieving complete and durable tumour regression when combined with anti-PD-L1 blockade (Xu et al., 2026). Extending the therapeutic scope beyond oncology, Huang et al. further established a CD163-antibody-conjugated LNP platform to deliver fibroblast activation protein (FAP)-specific CAR mRNA to hepatic macrophages for the treatment of liver fibrosis, underscoring the versatility of in vivo CAR-M engineering (Huang et al. 2026). These advances indicate that rational immune-cell selection can extend in vivo CAR engineering beyond lymphocytes to overcome tissue-specific barriers. Alongside myeloid cell engineering, natural killer (NK) cells represent another critical frontier. Although CAR-NK therapies have been primarily explored ex vivo, emerging evidence suggests that in vivo reprogramming is feasible. A recent study rationally designed a library of carbonate-bearing ionisable lipids (CAILs) and successfully delivered functional mRNA to 21% of splenic NK cells in mice by intravenous administration (Wu et al. 2025). Although this efficiency remains modest compared to T-cell engineering, it provides the foundation to utilise the innate antitumor activity and favourable safety profile of NK cells. Moving beyond individual immune subsets, an innovative LNP-panCAR was developed to deliver CAR mRNA simultaneously to innate and adaptive immune cells, including NK cells, T cells, macrophages and dendritic cells. This strategy generated functional panCAR cells across multiple immune organs, remodelling the tumour microenvironment and producing potent antitumor responses (Yin et al. 2026).Taken together, these findings indicate that in vivo mRNA cell therapy will move toward the coordinated programming of diverse immune cell populations to achieve durable and multifaceted therapeutic responses. Beyond oncology: protein replacement, autoimmunity and tissue regeneration The impact of targeted mRNA delivery extends far beyond oncology, offering transformative approaches of chronic and congenital disorders by enabling transient protein replacement, permanent genomic correction, and immunological tolerance induction. Protein replacement For metabolic diseases caused by enzyme deficiencies, systemic mRNA administration aims to restore functional protein expression. LNP-delivered mRNA enables endogenous protein production with physiological post-translational modifications, offering advantages over traditional enzyme replacement therapies, which are frequently limited by short half-life, poor tissue distribution, and immunogenicity. Clinical translation is now underway for several genetic disorders, including methylmalonic acidemia, propionic acidemia and ornithine transcarbamylase (OTC) deficiency, where repeat administration has demonstrated promising treatment benefit (Shi et al. 2022). Due to the transient nature of mRNA, protein replacement therapy requires repeated administration to maintain therapeutic protein levels, renderings long-term tolerance of the delivery system a primary engineering challenge. Repeated dosing increases the cumulative immunogenicity of the carrier, including the potential induction of anti-PEG antibodies that may trigger accelerated blood clearance and compromise the efficacy of subsequent LNP doses (Lila et al. 2013; Ibrahim et al. 2022). Consequently, engineering stealth or biodegradable LNP formulations specifically optimised for multi-dosing regimens has become a major focus in advancing systemic replacement therapies. Inducing immune tolerance and autoimmunity Unlike oncology, where therapeutic efficacy relies on activating antitumor immunity, autoimmune therapy seeks to re-establish immune tolerance by selectively suppressing pathogenic immune responses while preserving protective immunity. This can be accomplished by transiently activating specific regulatory pathways or resetting aberrant immune cells. mRNA therapeutics are engineered to function as inverse vaccines. These constructs present autoantigens in a non-inflammatory context to expand regulatory T cells (Tregs) rather than effector cells. Krienke et al. showed that non-inflammatory mRNA delivery could suppress antigen-specific T cell responses in a model of multiple sclerosis without compromising general immune competence (Krienke et al. 2021). Beyond restoring antigen-specific tolerance, transient mRNA engineering can also reset pathogenic immune cell populations responsible for sustaining the autoimmune responses. The therapeutic potential of this strategy has been clinically validated in a Phase 2b trial of Descartes-08, an mRNA-engineered CAR-T therapy targeting B cell maturation antigen (BCMA) for myasthenia gravis (MG). Because CAR expression is transient rather than permanently integrated into the genome, these engineered cells effectively depleted autoantibody-producing plasma cells without inducing permanent B-cell aplasia. The therapy reduced pathogenic autoantibodies while preserving total immunoglobulin levels and vaccine-induced immunity, addressing key safety concerns associated with chronic autoimmunity (Fedak et al. 2026). Similarly, a CD8-targeted LNP delivering the CD19 CAR mRNA therapeutic HN2301 induced remission in patients with refractory systemic lupus erythematosus (SLE). This trial achieved rapid B-cell depletion and sustained clinical improvement, validating that transient, nonviral cell engineering can reset the immune rheostat in humans without the toxicities associated with conventional lymphodepleting chemotherapy conditioning regimens (Wang et al. 2025). Emerging frontier: tissue regeneration The versatility of mRNA therapeutics extends to reversing fibrosis across multiple organs through two different strategies. One strategy employs in vivo generation of antifibrotic CAR-T cells. Rurik and coworkers innovatively employed CD5-targeted LNPs to generate antifibrotic CAR-T cells in situ, resulting in reduced fibrosis and restored cardiac function after injury (Rurik et al. 2022). Following this approach, Zhao et al. utilised CD5-targeted LNPs to generate platelet-derived growth factor receptor beta (PDGFRβ)-specific CAR-T cells that eliminated activated myofibroblasts, achieving reversal of multi-organ fibrosis in hypertensive chronic kidney disease models (Zhao et al. 2025). An alternative strategy focuses on transient cellular reprogramming rather than selective cell elimination. Zhang et al. introduced in situ partial cellular reprogramming by delivering mRNA encoding Yamanaka factors (Oct4, Sox2, Klf4) to hepatocytes via optimised LNPs. This induced a transient progenitor-like state that attenuated liver fibrosis, exploiting the transient nature of mRNA, which allows a temporary pulse of reprogramming factors sufficient to stimulate regeneration without crossing the threshold of oncogenic transformation (Zhang et al. 2025). Addressing inflamm-aging, Z. Zhang et al. developed a senolytic strategy using cardiolipin-mimetic LNPs to deliver mRNA encoding a CAR targeting the urokinase-type plasminogen activator receptor (uPAR). These engineered CAR-T cells selectively cleared senescent cells in models of liver fibrosis and rheumatoid arthritis. The use of circular RNA (circRNA) in this study extended the duration of CAR expression, enhancing the window for senolytic activity while maintaining a favourable safety profile (Zhang et al. 2025). In recent years, the clinical application of in vivo mRNA-LNP therapies has accelerated considerably (Table 1). Applications now extend beyond CAR-T therapies for haematologic tumours to myeloid cell-based strategies for solid tumours, immune cell reprogramming for autoimmune diseases, as well as CAR-T therapy for immune-mediated kidney diseases. However, most current in vivo mRNA-LNP therapies targeting immune cells remain in early-phase clinical evaluation, and details regarding mRNA dosing and pre-medication regimens are largely undisclosed because of proprietary considerations. Table 1.Clinical trials of in vivo mRNA-LNP therapies targeting immune cells.Registration IDProductPhaseIndicationTargetVectorsVector targetTarget cellCAR constructStudy startLocationSponsorNCT06618313JCXH-213Not ApplicableRelapsed/refractory B-cell non-Hodgkin lymphomaCD19tLCNPNot reportedT cells, NK cells, macrophagesNot reportedMar-2025ChinaBeijing GoBroad HospitalNCT06917742CPTX2309Phase 1Healthy volunteer study andmoderate to severe RA or SLECD19CD8-L829-tLNPCD8T cellsCD19 scFv/CD28Apr-2025AustraliaCapstan TherapeuticsNCT06801119HN2301Phase 1Autoimmune diseases (SLE, SSc, RA)CD19CD8-tLNPCD8T cellsCD19 scFvMar-2025ChinaShenzhen MagicRNA BiotechnologyNCT07333677HN2301Early Phase 1Refractory Graves’ diseaseCD19CD8-tLNPCD8T cellsCD19 scFvJan-2026ChinaShanghai Zhongshan HospitalNCT06965309* HN2301Not ApplicableRefractory myasthenia gravisCD19CD8-tLNPCD8T cellsCD19 scFvJan-2026ChinaShenzhen MagicRNA BiotechnologyNCT07556120HN2301Phase 1Generalised myasthenia gravisCD19CD8-tLNPCD8T cellsCD19 scFvMay-2026ChinaShenzhen MagicRNA BiotechnologyNCT05969041MT-302Phase 1Epithelial tumoursTROP2LNPNot reportedMyeloid cellsTROP2 scFv/CD89Aug-2023AustraliaMyeloid TherapeuticsNCT06478693MT-303Phase 1Advanced or metastatic GPC3-expressing cancersGPC3LNPNot reportedMyeloid cellsGPC3 scFv/CD89Jul-2024Australia, South Korea, TaiwanMyeloid TherapeuticsNCT07334119MT-304Phase 1Advanced HER2-expressing solid tumoursHER2LNPNot reportedNK cells, myeloid cellsNKp44-CARNov-2025AustraliaMyeloid TherapeuticsNCT07333118GT801Early Phase 1Moderate-to-severe refractory autoimmune diseasesCD19T-LNPsNot reportedT cellsNot reportedNov-2025ChinaGrit BiotechnologyNCT07205315GT801Early Phase 1Relapsed/refractory CD19-positive B-cell haematologic malignancies and autoimmune haemolytic anaemiaCD19T-LNPsNot reportedT cellsNot reportedSep-2025ChinaGrit BiotechnologyNCT07143617STR-P004Not ApplicableRelapsed or refractory autoimmune diseasesCD19csLNPNot reportedT cellsNot reportedSep-2025ChinaBeijing GoBroad HospitalNCT07003178STR-P004Early Phase 1Relapsed/refractory CD19-positive B-lymphocytic non-Hodgkin's lymphomaCD19csLNPNot reportedT cellsNot reportedJun-2025ChinaStarna TherapeuticsNCT07363954STR-P004Not ApplicableImmune-mediated kidney diseasesCD19csLNPNot reportedT cellsNot reportedFeb-2026ChinaStarna TherapeuticsNCT07349849 in vivo CAR-T drug based on LNP-mRNAEarly Phase 1Haematologic MalignanciesCD19LNPNot reportedT cellsNot reportedDec-2025ChinaXinqiao Hospital of ChongqingNCT07362758 in vivo CAR-T drug based on LNP-mRNAEarly Phase 1Refractory autoimmune diseasesCD19LNPNot reportedT cellsNot reportedJan-2026No location dataXinqiao Hospital of ChongqingNCT07408336AFN50Early Phase 1Autoimmune diseasesNot reportedT-LNPNot reportedT cellsNot reportedFeb-2026ChinaAlphaNa Bioscience Company LimitedNCT07448298AFN50Early Phase 1Relapsed/refractory autoimmune diseasesNot reportedT-LNPNot reportedT cellsNot reportedMar-2026ChinaBeijing Boren HospitalNCT07332663WGb-0301Early Phase 1Relapsed and refractory B-cell malignanciesCD19CD8-tLNPCD8T cellsNot reportedJan-2026No location dataSichuan UniversityNCT07349823WGb-0301Early Phase 1Relapsed and refractory immune-related diseasesCD19CD8-tLNPCD8T cellsNot reportedJan-2026No location dataWest China HospitalNCT07622329JCXH-213Phase 1Primary immune thrombocytopeniaCD19tLCNPCD8T cellsNot reportedJun-2026No location dataThe General Hospital of Western Theatre CommandNCT07629596Arnovie101Early Phase 1B cell-mediated autoimmune diseases (SLE and AIHA)CD19tLNPCD8T cellsNot reportedApr-2026ChinaCircunited BioPharma (Shenzhen) Co., Ltd.NCT07388212CD19/BCMA in vivo CAR-TEarly Phase 1Relapsed and refractory autoimmune diseasesCD19/​BCMALNPNot reportedT cellsNot reportedFeb-2026No location dataWest China HospitalNCT07362602 In vivo CAR-T drug targeting CD20 based on mRNA-LNPEarly Phase 1Haematological MalignanciesCD20LNPNot reportedT cellsNot reportedJan-2026No location dataThe 923rd Hospital of Joint Logistics Support Force of People's Liberation ArmySLE, Systemic Lupus Erythematosus; SSc, Systemic Sclerosis; RA, Rheumatoid Arthritis; AIHA, Autoimmune Haemolytic Anaemia; GPC3, glypican 3; TROP2, Trophoblast Cell Surface Antigen 2; HER2, Human Epidermal Growth Factor Receptor 2. Data were obtained from ClinicalTrials.gov (accessed June 2026). * Trial terminated due to change of registrant. Clinical trials of in vivo mRNA-LNP therapies targeting immune cells. SLE, Systemic Lupus Erythematosus; SSc, Systemic Sclerosis; RA, Rheumatoid Arthritis; AIHA, Autoimmune Haemolytic Anaemia; GPC3, glypican 3; TROP2, Trophoblast Cell Surface Antigen 2; HER2, Human Epidermal Growth Factor Receptor 2. Data were obtained from ClinicalTrials.gov (accessed June 2026). Trial terminated due to change of registrant. Preliminary clinical data presented at the American Society of Clinical Oncology (ASCO) 2026 Annual Meeting have provided valuable insights into dosing paradigms. For instance, first-in-human data for GT801 demonstrated that an ultra-low dose of 0.01–0.03 mg/kg was sufficient to achieve antitumor activity in relapsed/refractory B-cell non-Hodgkin lymphoma patients without requiring lymphodepletion preconditioning (Liu et al., 2026). In the autoimmune disease setting, patients with autoimmune diseases received escalating doses of STR-P004 ranging from 0.03 to 0.1 mg/kg for up to 8 infusions (Pan et al. 2026). While this rapid clinical development is promising, the transition of preclinical efficacy to standard-of-care is still hindered by challenges in immunotoxicity, delivery precision, and manufacturing scalability, which are discussed in the subsequent section. Current limitations and future perspectives mRNA therapeutics have advanced significantly, while translating in vivo cell engineering into standard clinical practice requires addressing ongoing challenges in safety, biological precision, and manufacturing processes. Compared to local vaccine administration, systemic delivery of genetic coding payloads requires a higher safety threshold and distinct regulatory considerations. Safety, scalability, and manufacturing challenges Immunotoxicity and systemic reactogenicity The immune response to both the mRNA payload and the lipid carrier remains a primary safety concern. Chemical modifications, such as N1-methylpseudouridine, reduce TLR-mediated mRNA recognition, but the delivery vehicle itself can still cause inflammation (Nance and Meier 2021). For example, ionisable lipids and PEGylated components can trigger complement activation-related pseudoallergy (CARPA) and induce anti-PEG antibodies, causing accelerated blood clearance (ABC) after repeated dosing (Wadhwa et al. 2020; Yang et al. 2024). In in vivo CAR-T therapies, LNP-induced systemic inflammation, combined with rapid CAR-T cell expansion, may increase the risk of severe adverse events (Fatahichegeni et al. 2025; Testa et al. 2025; Yang et al. 2025). However, recent studies in non-human primates show that optimised lipid formulations can achieve therapeutic editing without dose-limiting hepatotoxicity or acute systemic inflammation, indicating a potential therapeutic window (Hunter et al. 2025). Biodistribution and dose-limiting toxicity Nonspecific tissue accumulation also restricts therapeutic application. Conventional LNPs typically accumulate in liver via apolipoprotein E mediated LDL receptor uptake by hepatocytes and nonspecific phagocytosis by Kupffer cells. This sequestration requires higher systemic doses to transfect lymphoid organs, increasing the risk of hepatotoxicity (Paunovska et al. 2022). Upon systemic administration, the biomolecular protein corona can mask targeting ligands or promote opsonization, reducing in vivo targeting efficiency despite successful in vitro binding. Furthermore, anti-vector antibodies or complement activation accelerate clearance. While established murine systems, such as Balb/c mice utilised for pancreatic cancer modelling, provide a robust immunological baseline for evaluating initial tumour responses, they exhibit distinct physiological differences from human biology. Discrepancies in apolipoprotein affinity, serum proteomic profiles, and mononuclear phagocyte clearance dynamics indicate that formulations optimised in rodents frequently display altered biodistribution and immune responses in nonhuman primates and clinical patients (Mitchell et al. 2021). To improve clinical predictability and overcome these inherent model limitations, researchers increasingly utilise advanced evaluation platforms, including humanised mouse models, to bridge the translational gap between synthetic nanoparticle design and clinical therapeutic application (Chuprin et al. 2023; Chen et al. 2025). Payload capacity and construct complexity Besides the physiological barriers of biodistribution, the size and structural complexity of the mRNA payload impose a major biophysical and manufacturing challenge. Conventional CAR constructs are typically 2 to 3 kilobases (kb), which are readily accommodated by standard LNP formulations. However, next-generation strategies often demand much larger payloads, including dual-target CAR constructs, CRISPR-Cas systems and multicistronic vectors. Extended mRNA sequences, particularly those harbouring abundant secondary structure, may interfere nanoparticle formation and result in heterogeneous particle sizes (Li et al. 2025). The steric bulk and high negative charge density significantly decrease encapsulation efficiency, increase the possibility of degradation, and alter the physicochemical properties of the LNP. These alterations can severely compromise both the in vivo biodistribution and the efficiency of transfecting hard-to-reach primary immune cells. Large mRNAs also exhibit reduced translational efficiency and increased susceptibility to innate immune sensors, requiring optimised 5′ and 3′ UTRs and modified nucleosides to maintain expression kinetics. These challenges are further amplified in multicistronic or co-delivery regimens, such as simultaneous administration of CAR mRNA with immunostimulatory cytokine mRNAs to enhance the activity of engineered cells against the suppressive tumour microenvironment. In such scenarios, differential encapsulation ratios, uneven biodistribution, and different expression kinetics can compromise therapeutic synergy and batch-to-batch consistency. The inherent modularity of the LNP–RNA platform is exceptionally well-suited for combinatorial applications, yet precise control over co-encapsulation stoichiometry and coordinated in vivo engineering of multiple immune lineages remains an active frontier. CMC and regulatory complexity Finally, the transition from bench-scale synthesis to GMP-grade manufacturing introduces significant complexity. Unlike traditional biologics, the scalability of LNP production relies on microfluidic parallelisation rather than simple volumetric enlargement. It requires strict control over flow rates and mixing dynamics to maintain batch-to-batch consistency in size and polydispersity index (PDI) (Tenchov et al. 2021). Moreover, formulations with novel ionisable lipids or active targeting ligands also need complex conjugation and purification processes, which increase production costs, reduce colloidal stability, and rely on ultra-low temperature cold chains (Crommelin et al. 2021). From the regulatory perspective, classifying these multi-component systems necessitates new frameworks for establishing critical quality attributes (CQAs). A major analytical bottleneck lies in the lack of high-resolution methods to verify ligand density and mRNA encapsulation efficiency at the single-particle level. Current bulk analysis techniques often fail to distinguish between functional nanoparticles and nonfunctional byproducts in heterogeneous solutions. The integration of AI-driven formulation screening with high-throughput quality control offers a promising path to accelerate optimisation (Zhao et al. 2025). Optimisation strategies: from empiricism to precision engineering Although current delivery platforms have achieved clinical proof-of-concept, most of them rely on formulations originally optimised for vaccines or hepatic delivery. To address the specific bottlenecks in immune cell transfection, the delivery systems must evolve beyond empirical screening toward mechanism-driven precision engineering. High-throughput screening and AI-driven discovery The chemical space for ionisable lipids is vast, but traditional single variable optimisation is rate-limiting. Recent advances have demonstrated the power of high-throughput formulation screening, combined with DNA barcoding, to evaluate thousands of nanoparticle variants simultaneously in vivo, enabling unbiased identification of structure-function relationships that govern extrahepatic tropism (Hamilton et al. 2024). The advancement of experimental instruments could also accelerate the screening of LNP. Cui and colleagues developed a fully automated high-throughput platform employing a laboratory-based liquid-handling system. It can process up to 384 different formulations per plate and multiple plates per run, suitable for high-throughput operations. The precise mixing and repeatable mRNA LNP preparation ensure that hundreds or even thousands of new LNPs can be directly compared, reducing the number of new ionisable lipids that need to be evaluated in vivo (Cui et al. 2022). Recently, the integration of artificial intelligence has transformed this process from discovery to rational design. Machine learning algorithms trained on large datasets that link physicochemical parameters such as pKa, lipid tail length, and PEG content to biological outcomes can directly predict novel lipid structures with desired organ-specific accumulation properties (Amoako et al. 2025). Formulation optimisation could also be achieved by a computational framework. A random forest regression model trained on simulated lipid nanoparticle formulations could accurately predict immune activation values. This model was also embedded into a genetic algorithm to identify optimal design parameters, including size, charge, formulation and targeting (Di Salvatore et al. 2025). The computational evolution has the potential to decouple the biological distribution of LNPs from empirical formulation, leading to the reverse design of carriers for difficult-to-transfect targets, such as quiescent T cells. Currently, some commercial enterprises are building end-to-end platforms aimed at significantly shortening the development time of new therapies by utilising these AI-based methods (Hutchinson et al. 2026). Deep engineering of the mRNA cassette The mRNA construct determines the magnitude and duration of the therapeutic effect. Optimisation should not only focus on the simple nucleoside modification but also on engineering the entire mRNA cassette. AI-designed untranslated regions (UTRs) and codon optimisation tuned to tRNA abundance of target cells can maximise translation efficiency (Zhao et al. 2025). Furthermore, compared to linear mRNA, circRNA preventing exonuclease degradation, diminishing immunogenicity and extending translation duration, provides an alternative coding template (Reautschnig et al. 2025; Bahrulolum et al. 2026). Its enhanced stability enables sustained protein expression for applications that require prolonged therapeutic windows, such as senescence clearance or chronic enzyme replacement, without the risks of genomic integration (Wang et al. 2025). Co-optimisation of host and carrier Increasing evidence suggests that delivery efficiency fundamentally depends on the host cellular state. Incorporating bioactive excipients or metabolic modulators into the delivery formulations can actively condition the intracellular environment to synergise with the mRNA payload. As discussed in Section 4.3.3, temporal modulation strategies such as IL-7 pretreatment to prime T cells via mTORC1 activation, enhancing ribosomal translation efficiency without altering nanoparticle uptake (Tilsed et al. 2024). This suggests that activating the host cell can amplify the efficacy of targeted mRNA therapeutics. Trends and future directions The field is rapidly evolving from passive delivery vehicles to intelligent, programmable nanomedicines. The ultimate goal is to create in situ drug factories that operate with the logic and precision of biological systems. Programmable logic and autonomous control Stimuli-responsive systems often rely on single triggers, such as pH, ultrasound, or light, which may limit sensitivity or tissue penetration. Advanced delivery systems utilise Boolean logic gating to ensure tumour-specific activation by incorporating AND, OR, and NOT gate designs (Umadevi et al. 2026). A recent breakthrough demonstrated the engineering of logic-gated intracellular network CAR-T cells. This rapid and reversible Boolean-logic AND-gated platform outperforms other systems in both efficacy and prevention of on-target off-tumour toxicity. Unlike traditional CARs that rely on constitutive CD3ζ signalling, this approach re-engineers T cell signalling pathways by incorporating proximal intracellular molecules, such as LAT and SLP-76 (Tousley et al. 2023). The AND-gated design requires the coincident activation of two distinct antigens to initiate the signalling cascade, offering a hierarchical activation threshold that differentiate between the tumour microenvironment and benign inflammation, thereby improving safety. Expanding therapeutic paradigms: beyond oncology The success of mRNA technologies has paved the way for broader applications. For genetic disorders, the transient peak of mRNA-encoded editors minimises the window for off-target effects, offering a safer alternative to prolonged expression mediated by viral vectors for precision genome editing (Gillmore et al. 2021). In the context of autoimmunity, mRNA-based therapeutics are being designed to reprogramme the immune system through an inverse vaccine strategy. They present autoantigens in a non-inflammatory context to expand regulatory T cells and restore peripheral tolerance (Krienke et al. 2021). In parallel, the engineered CAR-T cells selectively deplete pathogenic B cell populations to achieve an immune reset, offering a more direct intervention (Hunter et al. 2025). Clinical translation of this concept has accelerated, with ongoing clinical trials investigating mRNA-engineered CAR-T cells targeting BCMA for myasthenia gravis and CD19 for refractory systemic lupus erythematosus (Wang et al. 2025). Furthermore, mRNA therapeutics are emerging as powerful tools for chronic viral infections that establish latency within immune cells, such as HIV and human T-lymphotropic virus type 1 (HTLV-1). A principal obstacle to curing HIV lies in the persistence of integrated proviruses within resting CD4+ T cells that maintain a transcriptionally silent state. Recent studies have utilised mRNA-LNP technology to deliver mRNA encoding the HIV Trans-activator of Transcription (Tat) protein, which potently overcomes blocks in viral transcription elongation ex vivo without inducing global T cell activation (Pardons et al. 2023). Moreover, LNPs can co-encapsulate CRISPR activation (CRISPRa) machinery to achieve highly HIV-specific transcriptional activation (Cevaal et al. 2025). In addition to direct therapeutic interventions, these LNP platforms serve as powerful precision tools for the deep molecular characterisation of the translation-competent HIV reservoir across diverse anatomical compartments, including blood and lymphoid tissues (Pardons et al. 2025). Similarly, for HTLV-1, mRNA-LNP platforms encoding viral envelope glycoproteins are emerging as potent strategies to elicit robust cellular immunity against infected immune cells (Bano 2026). This demonstrates that overcoming the delivery barrier to resting immune cells opens entirely new avenues for eradicating persistent viral reservoirs. Finally, in the application of regenerative medicine, the ability of mRNA-based nanoparticles to deliver reprogramming factors as a temporary pulse enables tissue rejuvenation and fibrosis reversal. Senolytic strategies targeting inflamm-aging represent another frontier, with engineered mRNA constructs enabling selective clearance of senescent cells in models of liver fibrosis and rheumatoid arthritis (Zhang et al. 2025). The path to clinical translation As the technology matures, the bottleneck shifts from scientific discovery to manufacturing and regulation. The complexity of multi-component, actively targeted lipid nanoparticles poses significant chemistry, manufacturing, and control (CMC) challenges, particularly due to the heterogeneity in ligand density across the nanoparticle population. Establishing standardised, high-resolution assays to quantify surface functionalization and encapsulation efficiency at the single-particle level remains urgent to ensure batch-to-batch consistency (Kamanzi et al. 2025). The integration of microfluidics and automated high-throughput mixing platforms is expected to overcome these stochastic assembly limitations, enabling the highly reproducible synthesis of multi-ligand architectures (Hwang et al. 2024). Regulatory frameworks must evolve to accommodate platform technologies. Given the established safety profile of lipid carrier, the approval of new mRNA cargos using the same platform could potentially be accelerated (Skerritt et al. 2024). The integration of rational material design, artificial intelligence, and deep biological insight will transform in vivo mRNA delivery from a niche tool into a clinical application. Conclusion Direct in vivo programming of immune cells using mRNA provides a viable alternative to complex ex vivo cell manufacturing. Because mRNA is transient, non-integrating, and programmable, it avoids several safety risks linked to stable viral vectors while simplifying the therapeutic workflow. However, as this review highlights, major physiological barriers persist, which hinder the efficient mRNA delivery. The most widely used lipid platforms often default to liver accumulation, struggle to penetrate target microenvironments, and can trigger unwanted immune responses. To overcome these limitations, the field is shifting from broad empirical screening toward rational, mechanism-based engineering. Achieving successful clinical translation will require improved active targeting techniques, such as oriented ligand conjugation, combined with scalable manufacturing processes and rigorous single-particle-level quality control. Although treating the human body as an in vivo bioreactor is a promising concept, making it a clinical reality depends on integrating targeted delivery innovations with strict safety and manufacturing standards.