IntroductionMessenger RNA (mRNA) vaccines have become an important platform in modern vaccinology because their programmable design enables rapid antigen updating and flexible vaccine development.1 While the conceptual foundation for mRNA-based protein expression was established in the 1990s,2,3 its translational implementation was hindered by critical barriers, including mRNA degradation and inefficient cellular delivery.4 Breakthroughs in RNA stabilization and lipid nanoparticle (LNP) delivery systems5,6 catalyzed the field’s maturation, epitomized by the landmark success of the Pfizer-BioNTech and Moderna COVID-19 vaccines, with more than 13.64 billion doses administered globally by early 2026. This milestone underscores the unique advantages of mRNA vaccines: rapid adaptability to pathogen evolution, scalable manufacturing, and humoral/cellular immunity induction.7,8 With these attributes, the platform is now poised to revolutionize pandemic preparedness and therapeutic vaccine development beyond traditional infectious disease targets.The immune mechanisms of mRNA vaccines begin with overcoming extracellular degradation and achieving intracellular delivery via LNPs.9,10 LNPs protect mRNA from nucleases and promote delivery to antigen-presenting cells (APCs) through endocytosis.1 Following endosomal escape, the released mRNA is translated in the cytosol to produce antigen transiently and is subsequently degraded without a risk of genomic integration.11,12 This intracellular antigen expression engages complementary immune pathways: major histocompatibility complex (MHC) class I presentation supports priming of CD8⁺ cytotoxic T cells, whereas secreted and/or processed antigens presented on MHC class II activate CD4⁺ helper T cells,13 thereby promoting B-cell differentiation and antibody responses.14 By coupling intracellular antigen production with antigen processing and presentation, mRNA vaccines can induce coordinated humoral and cellular immune responses and support the development of antigen-specific B and T cell memory without exposure to replicating pathogens (Fig. 1).15,16Fig. 1Full size imageCellular mechanisms of mRNA vaccines inducing immune responses and the challenges faced in vaccine development. Injected mRNA vaccines are first taken up by local transfected cells and antigen-presenting cells (APCs). After escaping the endosome and entering the cytosol, mRNA is translated into protein by ribosomes, and the translated antigenic proteins further stimulate the immune system in several ways. Intracellular antigens are then broken down into smaller fragments by the proteasome complex, and the fragments can enter the MHC class I pathway. In addition to direct antigen presentation, MHC class I complexes derived from antigen-expressing cells may be transferred to dendritic cells in secondary lymphoid tissues through cross-dressing. In this setting, both conventional type 1 dendritic cells (cDC1s) and cDC2s can participate in CD8+ T cell priming, and this process is supported by type I interferon signaling. Activated cytotoxic T cells kill infected cells by secreting cytolytic molecules, such as perforin and granzyme. Additionally, secreted antigens can be taken up by cells, degraded inside endosomes, and presented on the cell surface to helper T cells by MHC class II proteins. Helper T cells facilitate the clearance of circulating pathogens by stimulating B cells to produce neutralizing antibodies and by activating phagocytes through inflammatory cytokines. During this process, mRNA vaccines still face major challenges, including instability of mRNA molecules, limited cell targeting and delivery efficiency of delivery systems, and the need to tailor immune responses. This figure is created with BioRender.comAs the field expands beyond acute viral respiratory pathogens, mRNA vaccine development increasingly faces indication-specific demands that are not adequately addressed by current one-size-fits-all formulations.1,17 Rapidly evolving viruses require broad neutralization and durable recall responses; persistent infections may require sustained T follicular helper (Tfh) cell support and prolonged germinal center activity; and cancer immunotherapy requires effective cytotoxic T cell priming despite immune suppression and tolerance within the tumor microenvironment.14,16 Emerging applications in cardiovascular, neurodegenerative, autoimmune/inflammatory, and metabolic disorders further impose constraints on tissue targeting, dosing frequency, and safety, especially when repeated administration is needed.1 Across these settings, a recurrent challenge is that vaccine performance depends not only on delivering sufficient antigen, but also on controlling how innate signals shape downstream immunity.8,18 Limited RNA stability, inefficient endosomal escape, imperfect biodistribution, and off-target expression can reduce antigen availability, whereas excessive or poorly timed innate activation can suppress translation and distort adaptive immune programming.11,19 Immune signal regulation is therefore a central determinant of mRNA vaccine immunogenicity.The same innate pathways that license APCs can also restrict translation when activation is excessive, poorly timed, or sustained. Consequently, the key design objective is to calibrate innate immune signaling so that it supports antigen expression while guiding adaptive immunity toward indication-matched phenotypes (such as high-quality neutralizing antibodies, polyfunctional CD8⁺ T cells, or balanced helper responses). This calibration is increasingly approached as a systems engineering problem in which multiple controllable variables—RNA molecular features, formulation composition, particle biophysics, and dosing regimens—jointly determine antigen kinetics, innate activation profiles, and downstream immune programming.20In this review, we organize the major mRNA vaccine platforms, molecular engineering strategies, and delivery systems within a unified bioengineering framework centered on one core challenge: how to coordinate antigen-expression efficiency with innate immune regulation to optimize vaccine immunogenicity across different disease settings. Within this framework, we first compare non-replicating mRNA, self-amplifying RNA (saRNA), and circular RNA (circRNA) in terms of antigen-expression kinetics and innate immune activation. We then discuss RNA engineering strategies that improve stability, translation, and immune regulation, including platform-specific optimization of saRNA and circRNA. Next, we examine emerging delivery platforms that function not only in cargo transport but also in immune modulation, cytosolic delivery, and tissue or immune-cell targeting. And the applications and clinical implications of mRNA vaccines across infectious diseases, cancer, cardiovascular and neurodegenerative disorders, and immune-related and metabolic diseases are further summarized. Finally, we discuss manufacturing, stability control, and the key translational challenges that shape the development of next-generation mRNA vaccines.Immune mechanism of mRNA vaccine platformsCurrently, the main mRNA vaccine platforms are broadly categorized as non-replicating mRNA, saRNA, and circRNA. These modalities differ in how they generate antigen and how they are sensed by innate immune receptors, resulting in distinct translation efficiency and innate immune activation profiles. This part will summarize and compare these platforms based on their RNA structural characteristics, antigen production mechanisms, and intracellular innate immune activation mechanisms (Figs. 2, 3) (Table 1).Fig. 2Full size imageInnate immune responses to mRNA. Non-replicating mRNA, saRNA, and circRNA engage overlapping but distinct endosomal and cytosolic RNA-sensing pathways in APCs. Non-replicating mRNA is sensed mainly through delivered RNA and IVT byproducts. saRNA generates additional replication-associated immunostimulatory species, including negative-strand RNA and double-stranded RNA intermediates, which strengthen TLR3/7/8-, RIG-I-, and MDA5-linked signaling. Innate activation of circRNA is shaped largely by preparation-related impurities, including residual linear RNA and intron self-cutting fragments. These pathways can support type I IFN production, DC maturation, co-stimulation, cytokine release, and adaptive immune priming, but excessive activation can also drive antiviral restriction through ISG overactivation, PKR/OAS/ADAR-mediated responses, RNA degradation, and translational inhibition. This figure is created with BioRender.comFig. 3Full size imageAntigen expression in different types of mRNA vaccines. Non-replicating mRNA, saRNA, and circRNA differ in both translation mode and antigen-expression kinetics after cytosolic delivery. Non-replicating mRNA undergoes direct translation, typically producing a rapid but transient burst of antigen expression. saRNA first requires replicase translation and intracellular RNA amplification, leading to delayed onset but higher and more sustained antigen output. circRNA does not replicate, but its covalently closed structure resists exonuclease degradation and supports prolonged antigen availability, although translation depends on cap-independent initiation and is often less efficient at early time points. Together, these platforms define distinct trade-offs among onset speed, peak output, and expression duration. This figure is created with BioRender.comTable 1 Comparison of different types of mRNA vaccinesFull size tableNon-replicating mRNA vaccineNon-replicating linear mRNA is currently the most widely used mRNA vaccine platform. It underpinned the first-generation COVID-19 mRNA vaccines and was deployed at scale during the pandemic, with clinical and real-world studies reporting high effectiveness against infection and, importantly, severe disease across multiple populations and settings.21,22,23,24 Beyond COVID-19, the same platform has been broadly applied in prophylactic vaccines and therapeutic cancer vaccines because the antigen sequence can be changed without altering the overall production framework, enabling rapid updates when new targets or variants emerge.1,25The authorized COVID-19 mRNA vaccines (for example, BNT162b2 and mRNA-1273) are representative non-replicating mRNA vaccines.26 Their structure mimics endogenous eukaryotic mRNA and typically comprises a 5′ cap, 5′ and 3′ untranslated regions (UTRs), an open reading frame (ORF), and a poly(A) tail.4 After entering the cytosol, the delivered mRNA is translated directly to produce antigen and then degraded. As to translation efficiency, non-replicating mRNA is typically characterized by two linked parameters: peak expression and expression duration. Protein expression typically begins shortly after sufficient mRNA is available in the cytoplasm, and then decreases as mRNA is degraded (Fig. 3).27 Thus, antigen output depends on the translation competence of the input mRNA and its intracellular stability. Consequently, non-replicating mRNA vaccines require control of innate sensing to balance antigen production with innate activation and costimulation.7,8 One established approach is nucleoside modification. Incorporation of pseudouridine (Ψ) and related analogues reduces Toll-like receptor (TLR)-mediated recognition of RNA and decreases dendritic cell (DC) activation.2,28 In addition, immunostimulatory by-products generated during in vitro transcription (IVT) can contribute to innate activation (Fig. 2), and strategies that reduce such by-products, such as transcription-enzyme engineering, can decrease immunostimulatory contaminants in IVT mRNA preparations.29Self-amplifying RNA (saRNA) vaccinesaRNA vaccines constitute a distinct linear RNA platform in which the antigen sequence is combined with viral replication machinery encoding an RNA-dependent RNA polymerase (RdRp).30 This design enables intracellular amplification of antigen-encoding subgenomic RNAs and can sustain cytosolic antigen expression at relatively low input doses.31 Preclinical studies have shown that saRNA vaccines can induce protective immune responses against a broad range of pathogens, including SARS-CoV-2, influenza virus, respiratory syncytial virus (RSV), rabies virus, Ebola virus, and HIV-1.32,33,34,35 In the field of RNA-based therapy, Madigan et al. reported that a single intramuscular administration of saNppa-LNP generated sustained atrial natriuretic peptide (ANP) expression and improved cardiac repair after myocardial infarction.36 This study shows saRNA can function beyond vaccination as a single-dose therapeutic platform for sustained protein expression and tissue repair. The clinical development has also expanded across infectious diseases and cancer. In 2023, ARCT-154 became the first saRNA vaccine to receive regulatory approval, marking an important milestone in the translational development of this platform.37,38saRNA vaccines are typically built on viral replicon systems, most commonly alphavirus-derived replicons such as those based on Venezuelan equine encephalitis virus (VEEV), Sindbis virus (SINV), or Semliki Forest virus (SFV).39 In their native form, alphaviruses contain a 10–12 kb positive-sense single-stranded RNA genome organized into two cistrons: one encodes the non-structural proteins (nsP1–nsP4), which form the replication machinery, whereas the other encodes the structural proteins.40 To generate saRNA vaccine vectors, the structural gene region is replaced with an antigen or gene of interest, while the replicase module is retained. Similar to conventional mRNA, synthetic saRNA is produced by IVT and typically contains a 5′ cap, UTRs flanking two ORFs that encode the viral replicase and the gene of interest, and a poly(A) tail. However, because saRNA molecules are substantially longer than conventional mRNAs, transcript integrity, IVT yield, and by-product formation become more important considerations during platform development (Fig. 4).30 After saRNA obtained through IVT is delivered into the cytoplasm, the 5′ replicase is translated first, producing the non-structural proteins (nsP1–nsP4) that assemble the replication complex. Then, the replication complex generates a full-length negative-strand intermediate, which serves as the template for synthesis of multiple full-length positive-strand genomic RNAs and thereby expands the replicon template pool. Consequently, the complex recognizes the internal subgenomic promoter and produces shorter subgenomic RNAs that directly encode antigens. The former process amplifies the number of replicon templates, whereas the latter provides the major RNA species used for antigen translation.31Fig. 4Full size imageLimitations and optimization strategies of saRNA vaccines. saRNA vaccines achieve high antigen output through intracellular replicase translation and self-amplification, but this process is constrained by large RNA size, excessive innate immune activation, and translation suppression. Current engineering strategies therefore focus on redesigning replicon backbones, introducing compatible nucleotide modifications, and transiently modulating antiviral signaling through co-delivery of innate immune inhibitory proteins (IIPs) or related immunomodulatory elements. Together, these approaches aim to preserve self-amplification while limiting premature antiviral shutdown and improving antigen expression. This figure is created with BioRender.comIn terms of antigen-expression kinetics, saRNA differs from non-replicating mRNA because antigen production is coupled to intracellular RNA amplification (Fig. 3).33 After delivery, saRNA first enters a replicase-dependent phase, during which the replicon is amplified and antigen-encoding subgenomic RNAs are produced. Antigen expression may not peak immediately, but can increase over time as replication proceeds. This kinetic feature allows saRNA to achieve substantial antigen output at lower input doses and underlies its dose-sparing potential. Thus, the peak level and duration of antigen expression in saRNA are determined not only by the stability of the delivered RNA, but also by the efficiency of replicase translation, replication-complex formation, subgenomic RNA production, and the extent to which innate antiviral pathways restrict these processes.30The activation of innate immunity by saRNA vaccines is closely related to their intracellular replication process (Fig. 2). Similarly, internalized saRNAs can be sensed by pattern recognition receptors (PRRs) in endosomes and cytoplasmic compartments. Endosomal receptors such as TLR3, TLR7 and TLR8 can detect exogenous RNA species, whereas in the cytosol, replication-associated RNA structures are recognized by RIG-I and MDA5.41 In addition, cytosolic double-stranded RNA can activate NLRP3 inflammasome signaling.42 Compared with non-replicating mRNA, saRNA generates additional immunostimulatory RNA species during replication, particularly negative-strand and double-stranded RNA intermediates, which provide an important mechanistic basis for its stronger innate activation and self-adjuvant effect. Consistent with this, delivery of saRNA vaccines also induces type I interferon (IFN) production and upregulation of IFN-stimulated genes (ISGs), which is accompanied by increased immune-cell infiltration and contributes to adaptive immune priming.41 Replication-derived double-stranded intermediates can contribute to self-adjuvancy and promote antigen-specific CD4+ and CD8+ T cell responses with a T helper 1-skewed phenotype.43 At the same time, innate immune activation is also a major determinant of the balance between amplification and expression. Type I IFNs induce ISGs and antiviral effectors, including double-stranded RNA-dependent protein kinase (PKR), 2’-5’-oligoadenylate synthetase (OAS), and adenosine deaminase acting on RNA (ADAR), which promote RNA degradation, inhibit translation initiation, and destabilize RNA structure.41,44 Therefore, although innate activation can enhance immune priming, excessive or prolonged antiviral signaling may suppress saRNA replication and reduce antigen output. saRNA differs from non-replicating mRNA by integrating antigen expression with intracellular RNA amplification. This feature can increase antigen output and prolong the expression window at lower doses, but it also introduces additional replication-associated RNA species that strengthen innate immune activation and tighten the balance between self-adjuvancy and antiviral restriction.45Circular RNA (circRNA) vaccinecircRNA vaccines represent an emerging RNA platform of interest for vaccine development because their covalently closed structure can support prolonged RNA persistence and extended antigen expression. Unlike traditional linear mRNA vaccines, circRNAs are single-stranded RNA molecules with a covalently closed-loop structure, lacking both 5′ caps and 3′ poly(A) tails.46 This covalently closed structure improves resistance to exonuclease-mediated degradation and prolongs intracellular RNA persistence, thereby extending the window of antigen expression.47 These properties may partially alleviate the rapid degradation and limited expression duration associated with linear mRNA.48 circRNA is of interest because its closed-loop structure can prolong antigen coding capacity, although effective expression still depends on cap-independent translation elements such as internal ribosome entry sites (IRESs).49The core functional module of circRNA vaccines consists of the ORF and IRES. The ORF determines the antigen to be expressed, whereas the IRES determines whether the circular transcript can be efficiently translated.50 ORF length is a critical consideration. While long ORFs can encode complex antigens, they also increase the circularization difficulty and reduce translation efficiency.51 Emerging research suggests that short ORFs (sORFs), which encode small peptide antigens, may offer a balance between high translation efficiency and synthesis ease.52,53 These sORFs can be customized to express neoantigens or minimal epitopes that elicit strong immune responses. IRES are specialized mRNA segments located within UTRs that can recruit ribosomes and initiate translation independently of the conventional 5′ cap-dependent translation initiation mechanism.54 Commonly used IRES elements include those derived from poliovirus 1 (PV1), human rhinovirus A1 (HRV-A1), encephalomyocarditis virus (EMCV), hepatitis C virus (HCV), cricket paralysis virus (CrPV), and coxsackievirus B3 (CVB3), although their activity can vary across cell types and species.55,56 The IRES-ORF cassette is therefore the core functional unit of circRNA vaccines, and translation can continue as long as the circRNA remains intact.57 In addition, translation efficiency can be further modulated by the topology of the circular transcript, the spacing between the IRES and ORF, and the incorporation of RNA-binding protein motifs or m6A-related sequence motifs that recruit translational factors.49In terms of antigen-expression kinetics, circRNA differs from both non-replicating mRNA and saRNA. Non-replicating mRNA typically produces a rapid but time-limited burst of expression, whereas saRNA couples antigen production to intracellular RNA amplification. By contrast, circRNA does not replicate intracellularly, but its covalently closed structure can preserve coding capacity for a longer period by increasing resistance to exonuclease degradation. However, IRES-dependent translation is often less efficient than canonical cap-dependent initiation.55 circRNA may therefore show a lower initial rate of protein production but a longer expression window (Fig. 3). This trade-off explains why circRNA optimization has focused on IRES selection, ORF engineering, and reduction of unfavorable secondary structures.Currently, RNA circularization is commonly achieved by enzymatic ligation, chemical ligation, or ribozyme-mediated self-splicing.58 Among these approaches, the permuted intron–exon (PIE) system based on group I introns is widely used because it can efficiently circularize relatively long RNA precursors (Fig. 5).59 In this strategy, the target RNA is flanked by intron-derived sequences, and group I introns mediate exon ligation through self-splicing.60,61 However, these circularization processes can also generate residual linear precursors, excised intron fragments, exogenous sequence remnants, and duplex-containing side products.62,63 For this reason, compared with non-replicating mRNA and saRNA, the innate immune activation of circRNA is shaped more strongly by preparation-related byproducts and purification quality (Fig. 2). When sufficiently purified, circRNA can support prolonged translation with relatively low immune stimulation, but residual duplex structures, linear precursors, exogenous sequence remnants, and splicing by-products in in vitro-generated preparations may still trigger innate immune activation.64 Mechanistically, these immunostimulatory species can engage endosomal and cytosolic RNA sensors, including TLRs and RIG-I-like receptors, with RIG-I identified as an important sensor in circRNA recognition.62,65 RNA modification can further modulate this process, but its effects in circRNA are not directly equivalent to those in linear mRNA. For example, m6A has been reported to reduce RIG-I-mediated sensing, whereas m1Ψ does not appear to confer the same advantages in engineered circRNA as in conventional linear mRNA. Accordingly, circularization strategy and purification should be regarded not only as manufacturing variables but also as major determinants of the innate immune profile of circRNA vaccines. High-performance liquid chromatography and related purification strategies are important for reducing contaminants and minimizing unwanted innate sensing. The activation of innate immunity by circRNA vaccines depends on the interaction between circRNA secondary structure, ORF design, sequence composition, nucleotide modification and purity.66Fig. 5Full size imageLimitations and optimization strategies of circRNA vaccines. circRNA vaccines are commonly generated through circularization strategies such as the group I intron-based PIE system, which enables cap-independent antigen expression from a covalently closed RNA template. However, circRNA performance is often limited by inefficient IRES-dependent initiation, incomplete circularization, residual linear RNA species, and purification-associated product heterogeneity. Current optimization strategies therefore focus on improving the circularization and translation unit as an integrated system, including initiation-module optimization, Kozak- and IRES-related design, m6A tuning, and cleaner production approaches that improve circularization quality and reduce contaminating byproducts. Together, these strategies aim to enhance translationally competent circRNA yield while minimizing preparation-related functional variability and innate immune activation. This figure is created with BioRender.comOverall, circRNA vaccines represent a platform in which prolonged antigen availability is achieved mainly through enhanced RNA stability rather than replicase-driven amplification. Their main mechanistic advantage lies in the covalently closed structure, which supports prolonged persistence, whereas their main constraints arise from cap-independent translation efficiency and from the need to control innate sensing through rational sequence design, circularization strategy, and purification. These features distinguish circRNA from both non-replicating mRNA and saRNA and position it as a complementary platform for tuning the relationship between antigen-expression duration and innate immune activation.Comparative safety considerations of mRNA vaccine platformsNon-replicating mRNA, saRNA, and circRNA share the principle of RNA-directed antigen expression, but they differ in molecular architecture, expression kinetics, innate immune recognition, manufacturing requirements, and clinical maturity. Therefore, their safety considerations should be evaluated in a platform-specific and indication-dependent manner rather than through a single generalized framework.4 From a regulatory and product-development perspective, safety is determined not only by the RNA format itself, but also by RNA integrity, product-related impurities, delivery formulation, biodistribution, dose, administration route, target population, and the intended clinical use.67Non-replicating mRNA has the most extensive clinical experience and generally supports a transient and relatively controllable expression window.68,69 Its main safety considerations are related to RNA purity, IVT-derived impurities, innate immune activation, LNP-associated reactogenicity,70 off-target tissue exposure, and repeated-dose tolerability.71 These issues are particularly relevant for booster vaccination and for non-infectious disease settings that may require repeated administration.saRNA provides dose-sparing potential through intracellular amplification, but this advantage introduces additional variables. Replicon-dependent RNA synthesis can generate negative-strand RNA and double-stranded RNA intermediates that activate antiviral innate immune pathways. Moderate innate activation may contribute to self-adjuvanticity, whereas excessive activation can restrict amplification, suppress translation, and increase inflammatory reactogenicity.72 The larger transcript size and, in some designs, the need to control more than one RNA component further increase manufacturing and quality-control complexity.30circRNA supports prolonged expression through its covalently closed structure without requiring intracellular replication.55 However, its safety assessment remains less clinically established and is strongly affected by circularization efficiency and product purity. Residual linear RNA, nicked circles, intron-derived fragments, duplex by-products, and translation-control elements may influence innate immune sensing, expression predictability, and the controllability of persistent antigen production.49The three RNA platforms should not be ranked by a simple safety hierarchy. Non-replicating mRNA offers clinical maturity but remains formulation and dose dependent; saRNA improves dose efficiency but requires control of amplification-associated innate sensing; circRNA extends expression durability but requires stringent control of circularization quality and prolonged-expression kinetics. These platform-specific limitations define the engineering priorities discussed in the following sections, including RNA modification, impurity removal, innate immune tuning, delivery targeting, and manufacturing control.Molecular engineering of mRNA to regulate translation and innate immune signalingA central challenge in improving the immunogenicity of mRNA vaccines is to achieve sufficient antigen expression without triggering excessive innate immune restriction. Different mRNA platforms exhibit distinct expression kinetics and innate sensing profiles because they differ in RNA structure, intracellular processing, and the immunostimulatory RNA species generated during expression. Vaccine performance is therefore determined not by RNA format alone, but by a set of molecular features that collectively regulate translation efficiency, RNA stability, and immune recognition. From an engineering perspective, the goal is to balance antigen expression with innate immune activation: expression should be strong enough to support effective antigen presentation, whereas innate signaling should be calibrated to promote immune priming without excessively suppressing translation. Although non-replicating mRNA, saRNA, and circRNA share these general objectives, the specific optimization strategies differ across platforms. In this section, we summarize molecular engineering approaches that improve translation, stabilize RNA, and fine-tune innate immune activation in different mRNA vaccine systems (Table 2).Table 2 Comprehensive impacts of mRNA engineering strategies on stability, translation efficiency, and immune modulationFull size table5′-Cap structure and poly(A) tailAll eukaryotic mRNAs are characterized by a 7-methylguanosine (m7G) cap structure at their 5′ terminus, which plays a crucial role in facilitating ribosomal recruitment and translation initiation.73 In the absence of this cap structure, mRNAs fail to activate scanning-dependent initiation pathways, severely compromising translational efficiency.74 Current mRNA vaccine platforms mainly use either enzymatic capping or co-transcriptional capping to generate optimized 5′ cap structures.75,76 Enzymatic capping can achieve very high efficiency and is used in the production workflow reported for Moderna’s mRNA-1273.22 Co-transcriptional capping is widely used in IVT mRNA production, but early generations of synthetic cap analogs could be incorporated in the reverse orientation, which reduced translational efficiency because reverse-capped mRNA is not efficiently recognized by the ribosome.74 To address this limitation, anti-reverse cap analogs and optimized co-transcriptional capping strategies have been developed to improve correct cap orientation and increase the proportion of translationally competent capped transcripts. In parallel, cap analogs equipped with removable purification handles have enabled enrichment of fully capped mRNA bearing diverse cap structures, further linking cap chemistry with manufacturability and translational performance.77 Warminski et al. developed a novel 5′-cap modification for mRNA, termed AvantCap (m7GpppBn6AmpG),78 which incorporates a benzyl group at the m6Am position. Compared to traditional caps, AvantCap-capped mRNA showed enhanced translation efficiency in certain cell lines, including CT26 cells and human DCs, with protein production up to 10 times higher. In vivo, it resulted in 6-fold higher protein expression in mice and demonstrated improved therapeutic potential, particularly in cancer models, by promoting T cell proliferation and inhibiting tumor growth. The modification also showed resistance to fat mass and obesity associated protein (FTO)-mediated demethylation, ensuring greater stability. These findings suggest that AvantCap can improve both translational performance and transcript stability, supporting its further evaluation in mRNA therapeutics and vaccine design. Chen et al. developed a method called “ligation-enabled mRNA-oligonucleotide assembly (LEGO) ” to enhance mRNA translation.79 They investigated branched, dual-capped mRNA, which attached multiple caps to the 5′ end of the mRNA. The results showed that dual-capped mRNA significantly enhanced protein production in vivo, with up to a 10-fold increase in translation and a 17-fold higher antibody production in vaccine settings compared to control groups. This approach was especially beneficial in applications like vaccines for SARS-CoV-2. The LEGO method allows for extensive modification of mRNA to enhance its translational capacity and stability. Therefore, implementing optimized capping protocols represents an essential component in maximizing mRNA vaccine potency and therapeutic efficacy. A recent study reported a streamlined LciRNA architecture designed to reduce reliance on complex enzymatic capping procedures and expensive modified nucleotides.80 In this system, an IRES drives non-canonical translation initiation, and a 5′ protective RNA sequence composed of a viral exonuclease-resistant RNA (xrRNA) element and a poly(A)-binding protein (PABP)-recruiting motif is introduced to improve transcript performance. Mechanistically, this design not only limits 5′-to-3′ exonuclease-mediated degradation but also promotes recruitment of RNA-binding factors, thereby supporting sustained antigen expression in vivo. The same architecture was also associated with activation of PRR pathways in DCs and enhanced antigen-presentation-related signaling. In tumor models, these combined effects supported stronger antigen-specific T cell responses and improved antitumor activity. Rather than simply replacing conventional capped mRNA, this work suggests that cap-independent RNA designs may offer an alternative route for coordinating expression persistence and immune activation in selected vaccine settings.The incorporation of an extended 3′ poly(A) tail, typically comprising ~100–200 adenosine residues, serves as a critical determinant of mRNA stability and translational efficiency.81 This structural feature effectively mimics endogenous mRNA polyadenylation patterns and provides a crucial binding platform for poly(A)-binding proteins, which collectively shield the 3′ terminus from exonuclease-mediated degradation. Chen et al. studied the effects of branched chemically modified poly(A) tails on the translation capacity of mRNA.82 By engineering poly(A) tails with branched structures and chemical modifications, they observed a significant improvement in translation efficiency compared with conventional linear poly(A) tails. In control comparisons, the modified mRNAs produced markedly higher protein output in cells and in vivo, demonstrating that structural optimization of the poly(A) tail can directly enhance translational capacity, thereby providing a new strategy for improving the performance of mRNA therapeutics and vaccines.5′ and 3′ UTRThe 5′ UTR directly influences translation initiation efficiency and ribosomal loading. In mRNA vaccine design, 5′ UTRs are optimized to facilitate efficient context-dependent scanning along linear mRNA while minimizing the formation of inhibitory secondary structures that could impede the progression of 40S ribosomal subunits (Fig. 2c).83,84 Current designs for mRNA vaccines typically incorporate short 5′ UTR sequences derived from highly translated genes, such as α and β-globin genes, which enhance ribosomal loading onto mRNAs by maintaining optimal accessibility to both the 5′ cap structure and the initiation codon.85,86 In addition, incorporation of a Kozak consensus sequence (gccRccAUGG) improves start-codon recognition and supports efficient scanning during translation initiation.87,88 This strategic optimization of the initiation context contributes significantly to the overall translational efficiency of mRNA vaccines.10The 3′ UTR of mRNA plays a fundamental role in translation efficiency and mRNA stability. Naturally derived 3′ UTRs contain cis-acting motifs that recruit RNA-binding proteins which either stabilize the transcript or promote its decay. Hence, engineering the 3′ UTR to improve transcript stability and sustain productive translation has become an important strategy in mRNA vaccine design.83,84 The first-generation mRNA vaccines against SARS-CoV-2, including Pfizer-BioNTech′s BNT162b2 and Moderna′s mRNA-1273, utilized endogenous human 3′ UTR sequences to achieve reliable translation and stability. Moderna’s vaccine incorporated the human β-globin 3′ UTR, while Pfizer-BioNTech used an ɑ-globin 3′ UTR variant, both of which are known to confer moderate mRNA stability and translation efficiency.86 These UTRs were selected based on prior evidence of their effectiveness in supporting gene expression in mammalian cells, particularly in erythroid tissues. While adequate for emergency vaccine deployment, these sequences were not specifically optimized for immunological applications, leaving room for performance enhancement.83 Lewis et al. identified short 3–6 nucleotide motifs within 3′ UTRs that increased antigen expression by up to 200-fold, especially in the presence of modified nucleotides such as N1-methylpseudouridine (N1mΨ).87Recent trends in mRNA vaccine research have shifted toward the use of computational models and high-throughput screening to design and evaluate UTR sequences.84 Deep learning frameworks such as Optimus 5-Prime and web-based tools such as mRNA designer are capable of generating de novo UTRs tailored for high translational output, based on training data from tens of thousands of experimentally tested UTR variants.85 These platforms optimize the interaction of the UTR with RNA-binding proteins. This has led to a new generation of modular mRNA designs that can be rapidly adapted for different disease targets and patient populations.Nucleotide modificationChemical modification of uridine- and cytidine-containing residues is a widely used strategy to improve the stability of exogenous mRNA, reduce excessive innate sensing, and support antigen expression. The inherent enzymatic vulnerability of unmodified mRNA significantly constrains both the temporal persistence and yield of antigen protein expression. This limitation consequently attenuates the robustness of antigen presentation and subsequent immune responses.29 Notably, incorporation of chemically modified nucleosides, such as Ψ and N1mΨ, into the mRNA sequence has emerged as an effective strategy to improve RNA stability, reduce excessive innate immune sensing, and enhance protein expression.89,90PseudouridineΨ is a naturally occurring uridine isomer found in multiple classes of cellular RNA, including tRNA, rRNA, sn/snoRNA, mRNA, and lncRNA.91 This uridine analog nucleoside is characterized by a distinct C-C bond between the uracil base and ribose sugar, diverging from the conventional C-N glycosidic linkage.92 This structural modification confers enhanced stability by circumventing the recognition sites of RNA endo- and exonucleases that typically target the C-N glycosidic bond. Moreover, Ψ-containing RNAs are poor substrates for RNase T2 and the exonucleases PLD3/PLD4, which are required to degrade RNA into the small ligands activating TLR7 and TLR8.93,94 As a result, Ψ-containing RNA is inefficiently converted into immunostimulatory products such as uridine-like ligands for TLR8 and 2′,3′-cGMP-related ligands for TLR7. These findings indicate that Ψ suppresses innate immune sensing through a dual mechanism involving both defective lysosomal processing and reduced receptor engagement.95 However, nucleoside-modified mRNA is not entirely immunologically inert. Although Ψ-containing RNAs are primarily introduced to reduce excessive innate immune activation and enhance translation, recent evidence suggests that such modified mRNAs can still induce a low yet functionally relevant type I IFN response in vivo.96 This signal acts predominantly on DCs, promoting their maturation and thereby facilitating Tfh cell differentiation and germinal center formation. This finding refines the conventional view that nucleoside modification simply suppresses innate sensing, and instead suggests that such modifications reprogram the magnitude and quality of innate immune activation to optimize adaptive immunity.A seminal breakthrough emerged when Kariko et al. demonstrated that the strategic incorporation of modified nucleosides, including m5C, m6A, m5U, s2U, or Ψ, effectively shields the synthesized mRNA from immune surveillance and subsequent clearance.2 Further investigations revealed that mRNA containing complete substitution with Ψ exhibited a 4-fold enhancement in protein expression within murine DCs, while simultaneously suppressing inflammatory cytokine production. This finding highlighted the synergistic advantages of combining Ψ with additional nucleoside modifications. Moreover, Anderson et al. investigated the strategic Ψ incorporation specifically within the 5′ UTR.28 This targeted modification approach maintained the global mRNA structural integrity, reduced immune activation and enhanced translational efficiency within human DCs. Taken together, these findings suggest that Ψ should not be viewed simply as a means of silencing innate sensing. Rather, it can be used to reshape the magnitude and quality of innate activation so that antigen expression is preserved while functionally useful immune priming is retained.4N1-methylpseudouridineN1mΨ is characterized by an additional methyl group at the N1 position of Ψ, conferring greater resistance to enzymatic degradation than unmodified Ψ.97 N1mΨ-modified mRNA demonstrates substantial resistance to a diverse spectrum of endonucleases and exonucleases, including Ψ-specific endonucleases. In addition, N1mΨ enhances translation by preserving eIF2 function (preventing stress-induced shutdown) and also by enabling backup eIF2-independent pathways to function more efficiently.98 This dual mechanism increases ribosome density and ensures robust protein synthesis even under conditions that normally suppress translation.97 This enhanced stability, combined with the ability to evade innate immune activation pathways, facilitates protein translation efficiency from N1mΨ-modified mRNA templates. Recent work further indicates that this advantage is not limited to dampening canonical RNA-sensing pathways. Kim et al. showed that N1mΨ also reduces the binding of proton-activated TRIM25 to exogenous RNAs delivered through acidified endosomes, thereby alleviating TRIM25-dependent RNA turnover and increasing transgene output.99 Pardi et al. incorporated N1mΨ into an mRNA vaccine encoding influenza hemagglutinin (HA) and showed that this design elicited strong stalk-specific antibody responses.100 Compared with unmodified mRNA or traditional vaccine controls, the nucleoside-modified mRNA immunization induced 10-fold antibody responses in ferrets, particularly targeting the conserved HA stalk domain. These responses included neutralizing antibodies capable of cross-protection against diverse influenza strains, along with enhanced germinal center B cell activity. In 2020, Moderna′s mRNA-1273 vaccine incorporated N1mΨ substitution coupled with codon optimization, and induced potent neutralizing antibody responses to both wild-type (D614) and D614G mutant SARS-CoV-2 as well as CD8+ T cell responses, and protected against SARS-CoV-2 infection in the lungs and noses of mice.101These modifications effectively exploit the preferential enzymatic activity toward native nucleotides, thereby reducing susceptibility to cellular degradation pathways. The modification density can be precisely modulated; however, while increasing the modified nucleotide proportion extends mRNA half-life, it may adversely affect translation efficiency beyond an optimal threshold owing to various factors, including impaired exportin-5 binding and reduced ribosomal processivity.102 Consequently, the positioning of modified bases in vulnerable regions, while maintaining the integrity of critical codon binding sites, is crucial for an optimal balance between mRNA stability and translational efficiency, ultimately enhancing immunogenicity of mRNA vaccine.m5Cm5C is a structurally modified cytidine analog characterized by methylation at the C5 position of its pyrimidine ring. This modification can improve RNA stability and reduce susceptibility to degradation-related processing, thereby preventing the enzymatic conversion of cytidine to uridine nucleotides in RNA sequences, a common degradation pathway that typically compromises RNA stability.103 The prevention of deamination substantially extends the intracellular half-life of m5C-modified mRNA molecules. In addition, this naturally modified nucleoside evades TLR recognition in vivo, improving the stability of mRNA. Kariko et al. revealed that mRNA containing substitution with m5C exhibited a 2-fold enhancement in protein expression in DCs.93In summary, nucleotide modification should be viewed as a rational engineering strategy to optimize, rather than simply suppress the immunological performance of mRNA vaccines.89 By introducing chemically modified nucleosides such as Ψ, N1mΨ, and m5C, mRNA can be protected from premature enzymatic degradation, exhibit improved translational efficiency, and avoid excessive activation of innate immune pathways that would otherwise restrict antigen expression.2 At the same time, accumulating evidence suggests that the most effective designs do not require complete elimination of innate sensing, but instead benefit from a controlled level of immune stimulation that remains sufficient to support DC activation, Tfh cell priming, germinal center formation, and durable antibody responses.18,96 In addition, the rational nucleotide modification engineering strategy depends not only on the proportion of modified bases but also on their positional distribution within the transcript. Recent work has shown that site-defined nucleoside-ribose modifications can exert markedly different effects depending on where they are introduced.102 For example, 2′-fluoro substitution within the ORF was well tolerated when located at the first nucleotide of codons, whereas the same modification at the second or third codon position substantially impaired translation.90 In contrast, terminal modifications in the 5′ UTR and poly(A) tail were more favorable for enhancing expression and stability.74 These findings indicate that the engineering of nucleotide-modified mRNA should move beyond uniform global substitution toward region- and position-selective design. Future engineering of nucleotide-modified mRNA should prioritize precise control of nucleotide identity, density, and positional distribution. This optimization should be coordinated with sequence, purification, and delivery design to balance stability, translation, tolerability, and immunogenicity (Table 2).Codon optimizationCodon composition significantly influences the efficiency of mRNA translation into antigenic proteins in mammalian cells. Consequently, codon optimization has emerged as a vital molecular strategy to maximize protein expression from vaccine mRNA templates in host cells.84,104 Codon optimization involves bioinformatics-guided sequence modifications that align with the target expression system′s codon usage preferences while preserving the encoded amino acid sequence.105,106 Owing to the species-specific nature of codon bias, substituting rare human codons in mRNA vaccine with frequently occurring synonymous codons can enhance multiple aspects of translation.107High-resolution studies tracking multiple ribosomes on individual mRNA transcripts have revealed substantial traffic jams and collisions between leading and lagging ribosomes at rare codon bottlenecks.108,109 This ribosomal congestion substantially impairs translational processivity. Codon optimization eliminates these bottlenecks, ensuring efficient ribosomal loading and elongation kinetics.110 Moderna′s Spikevax mRNA-1273 vaccine incorporates a fully codon-optimized SARS-CoV-2 spike immunogen sequence to maximize expression following intramuscular administration.101 This optimization strategy generated robust neutralizing antibody titers exceeding those of conventional vaccines, along with potent CD8+ T cell responses. Alimohammadi et al. reported COReNAPCIN®, a novel SARS-CoV-2 vaccine incorporating a codon-optimized DNA sequence encoding the full-length spike glycoprotein with mutations (L986P and V987P) to stabilize the prefusion conformation.111 Preclinical studies demonstrated robust humoral and cellular immune responses in both murine and non-human primate models, with effective protection against SARS-CoV-2 infection. The vaccine is anticipated to demonstrate comparable efficacy to established mRNA vaccines in clinical trials. Zhang et al. used artificial intelligence (AI) and machine learning to develop “LinearDesign”, an algorithm for optimizing the stability and codon usage of mRNA molecules that can efficiently design mRNA sequences with higher chemical stability, protein expression efficiency and immunogenicity in a short time.112 Experiments have shown that compared with the vaccine designed by traditional methods, the COVID-19 mRNA vaccine designed by LinearDesign induced up to 128 times higher antibody levels in mice, and also showed significant advantages in other vaccines (such as varicella-zoster virus vaccines). This algorithm provides a powerful and versatile tool for developing mRNA vaccines and therapeutic drugs.Controlling innate immune recognition of saRNAsaRNA poses a distinct challenge in innate immune control because replication-dependent RNA amplification generates highly immunostimulatory intermediates.113 In contrast to conventional non-replicating mRNA, saRNA not only enters cells as a large exogenous RNA molecule but also undergoes intracellular replication, generating double-stranded RNA intermediates that are highly stimulatory to host antiviral sensors.30 As a result, saRNA is more prone to activate pattern recognition pathways such as RIG-I, MDA5, PKR, and OAS-RNase L, leading to type I IFN production, translational arrest, and RNA degradation.45 Although such responses may contribute to adjuvant-like effects, excessive innate activation can prematurely suppress replicon amplification and limit antigen expression. Therefore, the key objective in saRNA design is not to abolish innate sensing entirely, but to control its magnitude and timing so that self-amplification and vaccine immunogenicity can be simultaneously optimized (Fig. 4) (Table 2).One major engineering strategy is nucleotide modification (Fig. 4). Beyond Ψ-derived analogs, emerging studies indicate that other noncanonical nucleotides may also attenuate innate immune activation in saRNA.44 In particular, complete substitution with m5C was shown to reduce IFN signaling in human PBMCs in a dose-dependent manner, while enabling prolonged protein expression and improved in vivo potency.72 Other modified nucleotides, including 5-methyluridine (m5U) and 5-hydroxymethylcytidine (hm5C), have likewise demonstrated compatibility with certain saRNA replicons and were associated with reduced IFN output and/or sustained expression in vivo.72 Notably, saRNAs generated with m5C or m5U were reported to mediate strong in vivo expression and antigen-specific immune responses, whereas uridine analogs such as N1mΨ were ineffective in some VEEV-derived replicon systems because of impaired RNA synthesis.72 These findings suggest that the benefits of nucleotide modification in saRNA extend beyond a single nucleoside analog, but remain highly dependent on replicon backbone and polymerase compatibility, with m5C currently emerging as one of the most promising candidates for achieving innate immune attenuation without compromising self-amplification.41Another important optimization strategy is redesign of the replicon architecture (Fig. 4). Recent studies indicate that the replication module itself can be engineered at multiple levels to improve potency and safety. One direction is the development of trans-amplifying RNA (taRNA), in which the replicase and antigen-encoding RNA are separated into two transcripts.114 This split-vector design reduces the burden imposed by a single long saRNA molecule and allows the antigen-coding transreplicon to be minimized to essential cis-acting elements. In a further refinement, removal of the subgenomic promoter and redesign of the 5′ UTR generated a shortened transreplicon, and directed evolution subsequently produced an evolved variant with accelerated replication, more than 10-fold lower transreplicon dose requirements, and enhanced antibody responses in mice.115 Beyond split-replicon designs, the replication machinery itself can be engineered to enable external control of saRNA amplification. A recent study incorporated drug-responsive degradation domains into alphavirus non-structural proteins and used the FDA-approved small molecule trimethoprim to regulate replicon activity, achieving reversible and temporally programmable expression in vivo.116 This strategy illustrates a synthetic-biology route for controlling the magnitude and timing of saRNA amplification, which may help decouple antigen output from excessive antiviral activation. Importantly, replicon origin is also a major determinant of performance. Alphavirus-derived saRNA systems are not limited to the commonly used VEEV framework, but can also be built from genetically distinct old or new backbones, such as SFV, SINV, CHIKV.117 Comparative work has shown that these alternative backbones differ substantially in replication efficiency, protein expression kinetics, cytotoxicity, and immunogenicity, indicating that the viral origin of the nsP/cis-regulatory module directly influences how the replicon interfaces with host antiviral pathways. Feng et al. developed an engineered viral protein genome-linked (VPg) saRNA platform that enables cap-independent, low-immunogenic, and precise encoding of therapeutic proteins in vivo.118 The significance of this work lies in showing that innate immune control in saRNA need not be limited to reducing impurities or substituting nucleotides; it can also be achieved by reprogramming how the transcript is organized and translated at the 5′ UTR. By decoupling expression from conventional cap-dependent logic and simultaneously lowering immune dysfunction, VPg-linked saRNA provides a new framework in which translation initiation mode and innate immune regulation are jointly engineered.Co-delivery of immunosuppressive agents with the target saRNA has emerged as a practical strategy to improve saRNA performance when excessive early innate sensing limits replicon amplification and translation (Fig. 4).44 B18R, a vaccinia virus-derived soluble decoy receptor for type I IFN, has been used to transiently attenuate IFN signaling during saRNA delivery. Recent studies showed that co-delivery of B18R-encoding mRNA with saRNA improved antigen expression and enhanced vaccine-induced immune responses, suggesting that early IFN blockade may be beneficial when innate restriction becomes a major barrier to saRNA performance.32 Blakney et al. demonstrated that incorporating innate immune inhibitory proteins (IIPs) into saRNA design can significantly enhance protein expression and immunogenicity.44 More recent studies have further extended this to host-targeted modulation, such as transient suppression of IFNAR1-associated signaling, although such approaches remain at an early stage of evaluation.119 Collectively, these findings suggest that co-delivery of immunomodulatory agents may provide a useful means to fine-tune the early antiviral environment of saRNA. However, because excessive suppression of innate sensing may also weaken the intrinsic self-adjuvanting properties of saRNA, the current objective is not broad immunosuppression, but selective and transient modulation of pathways that otherwise induce premature antiviral shutdown.120Translation efficiency improvement for circRNAcircRNA is relatively stable, but its translational efficiency is often limited.48 In the absence of a 5′ cap and a poly(A) tail, circRNA mainly relies on cap-independent initiation, typically through IRES. However, IRES-driven circRNA translation is frequently inefficient and strongly influenced by IRES types and RNA structure. Therefore, improving translation efficiency has become a key objective in circRNA engineering for vaccines and therapeutic applications (Table 2).Protein production from circRNA is first limited by translation initiation (Fig. 5). Wesselhoeft et al. showed that protein-coding circRNAs generated by group I intron-mediated circularization could support sustained expression in eukaryotic cells after HPLC purification.55 This study established engineered circRNA as a protein-expression platform, while also indicating that prolonged expression alone does not resolve the central translational constraint of circRNA, because protein output remains dependent on efficient cap-independent initiation and careful optimization of construct design and product purity. Chen et al. systematically evaluated synthetic circRNA design variables and found that protein output could be markedly improved by combined optimization of the initiation region, including the organization of IRES-containing modules and UTR-like elements.50 These results indicate that circRNA translation is determined by the overall architecture of the initiation unit (Fig. 5). This also explains why the same IRES can behave differently across constructs and cell types. In practice, efficient circRNA translation often requires empirical optimization of the full initiation sequence together with the encoded ORF. A more recent study further showed that translation can also be improved by simplifying the circular transcript itself. Zhang et al. developed small circRNA vaccines composed of minimal peptide-coding sequences together with a short IRES and a Kozak sequence.53 By comparing different short IRES elements, they identified a 71-nt crTMV IRES that supported more efficient antigen presentation and T cell priming than LINE1- or RBM3-derived short IRESs, whereas deletion of the IRES or Kozak sequence markedly reduced activity. In mice, these small circRNA vaccines induced strong and durable antigen-specific T cell responses and outperformed several modified mRNA, unmodified mRNA, and large circRNA controls. These findings indicate that improving circRNA translation requires not only stronger initiation modules, but also lower transcript complexity and a circular architecture that better supports translation.121,122 In addition, studies on the modification of circRNA endogenous properties have shown that efficient circRNA translation reduces dependence on conventional IRES initiation.123 Wang et al. showed that m6A-mediated translation of circRNA requires YTHDF3 and eIF4G2, is enhanced by METTL3/14, and is suppressed by FTO.124 These data indicate that cap-independent translation from circRNA is not restricted to classical viral IRES elements. Internal RNA modifications can also contribute to initiation, although their effects depend on the position of the m6A motif and the availability of the corresponding cellular factors (Fig. 5).Translation efficiency is also influenced by the coding region and the overall architecture of the circle. After ribosome loading, productive protein synthesis still requires efficient elongation across circRNA. Stable local structures, unfavorable sequence organization, and poorly designed regions around the ligation junction can all reduce output. Du et al. developed trans-splicing-based methods to generate synthetic circRNAs longer than 8,000 nucleotides without relying on bacterial sequence elements required by conventional PIE-based circularization.125 Compared with conventional PIE-derived constructs, these circRNAs showed lower immunogenicity and more efficient translation. In the same study, redesign of the circular template markedly improved rolling-circle translation. These findings indicate that translation from circRNA depends not only on initiation, but also on how the circle is assembled and structured.126 Product quality is another factor that directly affects apparent translation efficiency. Compared to linear mRNA, circRNA is more likely to generate incomplete circularization, nick circles, and residual linear precursors during IVT, which can more easily inhibit efficient translation and cause severe innate immune recognition. Purification should therefore be regarded as part of translation engineering rather than as a purely manufacturing step (Fig. 5).123Current studies suggest that future optimization of circRNA should focus on several linked aspects rather than on a single translation-enhancing element. First, translation initiation remains the primary bottleneck and therefore requires more systematic engineering of initiation modules, including not only IRES selection but also the arrangement of adjacent untranslated-region-like sequences and other ribosome-recruiting elements, because protein output is determined by the architecture of the initiation unit as a whole rather than by one IRES alone. Second, alternative cap-independent initiation strategies deserve further attention, particularly those based on internal chemical cues such as m6A or cap-mimetic/internal-cap designs, which may help reduce the dependence on inefficient conventional IRES-driven translation.127 Third, circRNA optimization should extend beyond translation-initiation modules to the design of the coding sequence and the overall circRNA architecture, because sequence arrangement, junction-region design, and circularization strategy can all influence translational efficiency. In addition, product quality should be treated as part of functional design, since incomplete circularization, residual linear precursors, and other by-products directly reduce the fraction of translationally competent molecules. These findings indicate that future circRNA engineering should move beyond element-by-element empirical adjustment toward an integrated optimization framework. In such a framework, translation-initiation efficiency, circRNA architecture, transcript miniaturization, and the rational selection of short translation-initiation elements should be considered together at the design level. Chemical modification and manufacturing quality should likewise be incorporated as coordinated determinants of stable and efficient protein expression.Dual-function of mRNA delivery platformDual-function mRNA delivery platforms are not merely carriers that encapsulate, protect and transport mRNA into target cells;128 they also participate in shaping the ensuing immune response.19 Beyond determining mRNA stability in vivo, cellular uptake and intracellular trafficking, the composition and structural features of the carriers, together with its interactions with biological interfaces, can influence innate immune sensing, antigen expression kinetics, antigen presentation efficiency, and the magnitude and quality of downstream adaptive immunity.129 Delivery systems are therefore not passive transport vehicles, but active determinants of therapeutic outcome (Fig. 6a). This issue has become increasingly important with the rapid expansion of mRNA-based vaccines and immunotherapies, in which carrier-dependent differences in inflammatory programming, antigen expression and tissue distribution can materially affect both efficacy and tolerability. A systematic analysis of delivery platforms that integrate mRNA delivery with immune regulation is thus needed not only to clarify the functional differences among material systems, but also to guide the rational design of platforms for distinct therapeutic settings.130Fig. 6Full size imageDelivery systems of mRNA vaccines. a Representative structural feature of major mRNA delivery systems, including LNPs, exosomes, polymer-based systems, VLPs. b The intracellular delivery route is a major source of functional divergence across platforms. LNPs support efficient uptake but remain limited by incomplete endosomal escape. Exosomes may support cytosolic delivery after endocytic uptake through relatively efficient natural endosomal escape, although release efficiency remains less controllable than in chemically engineered systems. Polymer-based systems offer chemically tunable uptake and release behavior, but efficient cargo liberation is often constrained by endosomal entrapment or incomplete unpacking. VLPs exploit biologically evolved entry pathways, which can improve cell entry efficiency but also impose structural and engineering constraints. c Targeting specificity is achieved through platform-dependent engineering strategies rather than a single universal design rule. LNPs are commonly redirected by lipid-composition redesign, exosomes by membrane-display engineering, polymeric systems by ligand conjugation, and VLPs by capsid or envelope reprogramming. As a result, targeting performance depends on how each platform integrates material composition with receptor recognition and in vivo biodistribution control. d These platforms should therefore be evaluated not only by delivery potency, but also by the balance they achieve among immune activity, biocompatibility, and engineering complexity. LNPs currently provide the most mature overall delivery performance, exosomes offer strong biological compatibility, polymers retain broad chemical flexibility, and VLPs preserve highly efficient biological entry logic. Together, these systems illustrate that mRNA carriers function as active regulators of delivery efficiency, immune signaling, and application-specific translational potential. This figure is created with BioRender.comEfficient delivery generally depends on strong nucleic acid packaging, robust cellular uptake and effective endosomal escape; however, these features are also often associated with greater membrane perturbation, amplified inflammatory signaling and increased tissue toxicity. Excessive emphasis on biocompatibility and low immunogenicity, by contrast, may improve tolerability but compromise intracellular release, reduce antigen expression and fail to support effective immune activation. This trade-off is particularly important in mRNA systems because innate immune sensing is closely linked to translational output. Nucleoside modification was originally introduced to reduce aberrant innate sensing and improve protein expression, whereas later studies showed that both the mRNA component and the lipid carriers can shape early inflammatory programs, with LNP-driven pro-inflammatory responses and mRNA-induced type I IFN signaling jointly influencing vaccine performance. The desired degree and mode of immune stimulation also vary across applications. Prophylactic and therapeutic vaccines generally require controlled innate immune activation to enhance antigen presentation and lymphocyte priming, whereas strategies aimed at sustained protein expression or in vivo cell reprogramming place greater value on limiting non-specific inflammation to avoid translational inhibition and functional loss. Achieving a controllable balance among delivery efficiency, immune regulation and safety has therefore become a central issue in the evaluation and optimization of mRNA delivery systems (Fig. 6d, Table 3).Table 3 Comparison between delivery systemsFull size tableLipid nanoparticles (LNPs)LNPs are the leading delivery platform for mRNA vaccines because they combine efficient RNA encapsulation, protection from extracellular degradation, and scalable manufacture with the capacity to support intracellular delivery after administration.9 Their role in vaccination extends beyond transport. LNPs also shape early immune events that influence antigen presentation, Tfh differentiation, and germinal center output.131,132 Recent evidence further shows that the immune activity of LNP-based mRNA vaccination extends beyond humoral programming and includes a distinct mode of CD8⁺ T cell priming.133 In contrast to protein- and DNA-based vaccines, which depend strongly on cDC1-mediated cross-presentation, mRNA–LNP vaccines can prime CD8⁺ T cells through both cDC1 and cDC2 cells in a partially redundant manner. Moreover, this process is not explained solely by classical direct presentation or WDFY4-dependent cross-presentation, because cross-dressing of peptide–MHC-I complexes from non-haematopoietic cells contributes substantially to CD8⁺ T cell activation in a type I IFN-dependent manner. Alameh et al. showed that empty LNPs enhanced both mRNA and protein subunit vaccines by promoting Tfh responses and humoral immunity.134 Castaño et al. further dissected this effect and showed that the mRNA and LNP components make distinct but cooperative contributions to vaccine immunity: nucleoside-modified mRNA induced type I IFN acting on DCs, whereas LNPs instructed a Tfh-promoting program in DCs, with both components required for efficient germinal center responses.96 The same study also showed that mRNA-LNP uptake by DCs occurred predominantly in draining lymph nodes rather than being driven mainly by migratory uptake from the injection site. These findings indicate that LNPs should be regarded as immune-active delivery materials rather than inert carriers.This dual function arises from the modular composition of LNPs.135 Current formulations usually contain ionizable lipids, helper phospholipids, sterol, and PEG-lipids, whereas conventional permanently cationic lipids such as DOTMA and DOTAP primarily established the feasibility of lipid-mediated RNA delivery but were limited by higher toxicity.136 Ionizable lipids like SM-102 and ALC-0315 are central components because they determine RNA condensation, pH-responsive charge conversion, endosomal membrane destabilization, and part of the inflammatory profile of the formulation.137 Helper phospholipids such as DOPE and DSPC, sterols such as cholesterol and related analogues, and PEG-lipids such as DSPE-PEG, DMG-PEG, and ALC-0159 are also not passive excipients.138 They influence membrane organization, colloidal stability, serum interactions, lymphatic drainage, intracellular trafficking, and the extent to which immune cells can productively access the payload. The key implication is that the same parameters that improve cytosolic delivery can also alter immune programming, tissue exposure, and tolerability.11The first challenge of LNPs is that endosomal escape is required for mRNA delivery but remains inefficient (Fig. 6b).139 This has made endosomal release a major focus of LNP optimization. One strategy has been structural refinement of ionizable lipids.140 The progression from DLin-DMA to DLin-MC3-DMA, SM-102, and ALC-0315 illustrates how changes in headgroups, linkers, and hydrophobic tails can improve functional delivery.141 He et al. further showed that ionizable lipid isomerism altered intermolecular organization within LNPs, shifted apparent pKa, changed transfection profiles across liver cell types after intravenous dosing, and improved muscle-selective delivery after intramuscular administration.142 These results indicate that subtle chemical changes can alter both delivery efficiency and tissue selectivity.143 The second strategy has been to quantify escape more directly and then redesign formulations around the limiting steps.144 Jozić et al. used LysoTag mice and lysosomal barcoding to show that even the optimized BiP-20 formulation delivered only about 8% of LNPs to the cytosol within 30 min after administration, and that endosomal maturation and recycling pathways regulate productive escape.145 Loss of Rab7 increased escape, indicating that late endosomal maturation itself constrains cytosolic delivery. In addition, Meng et al. incorporated the influenza HA2 fusion peptide into LNPs and showed improved mRNA delivery in vitro and in vivo, with enhanced gene editing in the liver and lungs and increased FAH expression in a hereditary tyrosinemia type 1 model.146 Together, these studies indicate that current work on endosomal escape is shifting from simply increasing membrane disruption to improving productive cytosolic access.The second challenge is that the intrinsic adjuvanticity of LNPs is useful for vaccination but can also become excessive. This issue is central to the dual-function nature of LNPs because the same formulation features that support intracellular delivery can amplify inflammatory signaling.147 The LNP component does not act only through nonspecific inflammation. Instead, LNPs induced a DC program associated with Tfh-cell promotion, including changes linked to activation, positioning, and control of IL-2 availability.96,134 This mechanistic distinction is important because it shows that LNP-mediated adjuvanticity can influence the quality of the immune response rather than only its magnitude. Han et al. modified TLR7/8 agonist with adipose chains to obtain adjuvant lipids, which were then inserted into LNPs to enhance the immunogenicity of SARS-CoV-2 mRNA.148 Li et al. developed an Am80-containing LNP that co-delivered mRNA and a retinoic acid receptor agonist, improved lymph node accumulation and mRNA delivery, preserved gut-homing imprinting, and enabled parenteral vaccination to induce intestinal IgA and protective mucosal immunity against rotavirus.149 These studies support two complementary strategies for immune control: tuning the inflammatory burden of the lipid formulation itself, and incorporating immunoregulatory cues into the particle when a defined immune output is required.The third challenge is that efficient delivery does not necessarily translate into selective delivery. In vaccines, LNPs should direct mRNA to APCs in lymphoid tissues or draining lymph nodes to support efficient antigen presentation and lymphocyte priming, whereas off-target expression in non-immune tissues may increase reactogenicity without improving protective immunity.150 In therapeutic applications, selective delivery is even more critical because many indications require mRNA expression in a disease-relevant tissue or cell type rather than in the liver by default.151 This problem is a combination of passive, active, and endogenous targeting, in which particle size, surface charge, stiffness, ligand display, protein corona formation, route of administration, and physiological barriers jointly determine organ and cell selectivity.152 Cheng et al. showed that inclusion of a supplemental selective organ targeting (SORT) component could redirect mRNA expression from the liver toward the lung or spleen, indicating that organ tropism can be tuned by formulation rather than treated as a fixed property of LNPs.153 More recent studies have extended this principle to harder biological barriers. Lei et al. reported a pancreatic-targeted LNP that undergoes protein-dependent size enlargement and exploits capsule-filter-mediated accumulation together with receptor-mediated uptake, enabling pancreatic delivery of Cas9 mRNA/sgRNA and therapeutic cytokine mRNA in mice and non-human primates.154 In the lung, Liu et al. showed that charge-assisted stabilization improved LNP integrity during nebulization and enabled efficient pulmonary mRNA delivery after inhalation, a design that addresses the instability and airway barriers that otherwise limit mucosal delivery.155 For central nervous system delivery, Wang et al. designed blood–brain-barrier-crossing lipids that enabled intravenous LNP-mediated mRNA delivery to neurons and astrocytes across broad brain regions.156 These studies indicate that targeted LNP delivery depends on coordinated control of formulation chemistry, endogenous biomolecular interactions, administration route, and tissue-specific barrier navigation rather than on particle accumulation alone.157The fourth challenge is that formulation stability can conflict with biological performance. PEG-lipids illustrate this trade-off clearly. They improve colloidal stability, reduce aggregation during formulation and storage, and limit premature interactions with serum components after administration, but excessive surface shielding can reduce cellular uptake and intracellular release. Repeated exposure can also induce anti-PEG antibodies and hypersensitivity-related concerns, making the “PEG dilemma” a persistent issue in LNP development.158,159 Similar trade-offs apply to other components. Barbieri et al. reported that helper lipid identity altered LNP stability, transfection efficiency, and immunogenicity, with DSPC providing better storage stability than DOPE-containing formulations in their saRNA-LNP system.160 Sterol composition also affects performance beyond particle integrity. Patel et al. showed that naturally occurring cholesterol analogues, including C-24 alkyl phytosterols, enhanced transfection relative to cholesterol-containing LNPs, indicating that sterol structure can alter membrane behavior and endosomal release.161 These findings indicate that helper phospholipids, sterols, and PEG-lipids should not be treated as fixed background ingredients. Their selection needs to be aligned with the intended route, tissue target, and immune objective, because the same changes that improve shelf stability or circulation behavior may reduce productive delivery or alter immunogenicity.162,163Current LNP development increasingly emphasizes design logic and functional purpose, focusing on how defined material choices and structural features shape delivery efficiency, biodistribution, immune regulation, and safety, rather than merely listing particle components.164,165 Recent studies have pursued several complementary directions to resolve the tensions inherent to LNP design. Xu et al. developed the LUMI-lab platform, which synthesized and tested more than 1,700 ionizable lipids, identified brominated tails as a favorable motif for pulmonary delivery, and achieved 20.3% lung epithelial gene editing in vivo with the top-performing formulation.166 Another study has focused on improving productive endosomal release through biomimetic design, including fusion peptide-incorporated LNPs.146 Parallel efforts have also introduced targeting modules to improve cell selectivity and reduce off-target expression.154 These studies suggest that the central objective of current LNP research is no longer to maximize transfection as an isolated endpoint, but to coordinate cytosolic delivery, immune regulation, tissue selectivity, and tolerability within the same system (Fig. 6d). In this case, LNPs remain a useful model for understanding how an mRNA carrier can simultaneously determine delivery efficiency and immune outcome. This dual-function nature creates a recurring challenge: endosomal escape is necessary but inefficient, adjuvanticity is beneficial but can become excessive, stability can reduce accessibility, and efficient delivery may still be off-target (Fig. 6d). A strategy-oriented framework is therefore more informative than a purely component-based description for understanding how LNPs are being adapted to different vaccine and therapeutic settings.ExosomesExtracellular vesicles (EVs) are a class of nano-sized particles with a lipid bilayer membrane structure secreted by cells, which are widely involved in intercellular material transfer and signal communication.167 According to their biological generation pathways and size ranges, EVs are usually divided into exosomes (~30–150 nm diameter), microvesicles (~50–1000 nm diameter) and apoptotic bodies (~50–5000 nm diameter).168,169 Exosomes are mainly derived from the fusion and release of multivesicular bodies (MVBs) and cell membranes, while microvesicles are directly formed by budding outward from the cell membrane.170 The membrane structure of exosomes is rich in phospholipids, cholesterol and various membrane proteins, and they encapsulate various bioactive molecules including mRNA, miRNA, DNA fragments, proteins and metabolites (Fig. 6c).171 These characteristics enable exosomes to play a key role in regulating intercellular communication, immune regulation, signal transduction and metabolic homeostasis.172 In addition, exosomes exhibit high biocompatibility owing to their natural origin, can circulate in the body for a long time, and efficiently target immune organs such as lymph nodes owing to their size.173 These unique biological properties make exosomes an ideal candidate for next-generation drug delivery systems, particularly in RNA delivery and vaccine development.174As naturally secreted nano-scale vesicles, exosomes provide a dual protection mechanism for mRNA delivery with their unique lipid bilayer structure and biological membrane coverage characteristics: they effectively block mRNA degradation by extracellular enzymes and significantly prolong the in vivo circulation duration by escaping macrophage clearance.175 These vesicles can cross selected biological barriers and, in some contexts, support cytosolic delivery after cellular uptake through endosomal escape (Fig. 6b).176 Their endogenous origin may improve biocompatibility relative to artificial synthetic carriers (Fig. 6d). More importantly, the surface of exosomes can inherit donor cell-specific protein receptors and can be precisely targeted through genetic engineering, chemical modification and other means.177 This combination of intrinsic biological compatibility and engineering flexibility enables more precise tissue-targeted delivery and expands the translational potential of mRNA vaccines.178,179 An innovative inhalable dry powder mRNA vaccine system based on lung-derived exosomes (Lung-Exos) has been developed, which has better biodistribution and retention properties than traditional synthetic LNPs after pulmonary administration.180 Moreover, it can be administered by both atomization and dry powder inhalation, as well as maintain stability for 28 days at room temperature. Lung-Exos loaded with SARS-CoV-2 spike protein-encoding mRNA can induce stronger IgG and secretory IgA immune responses.181 This platform may help address storage and route-of-administration constraints that limit conventional mRNA formulations in respiratory applications, providing an innovative drug delivery platform for treating respiratory diseases. Dong et al. developed an engineered sEV platform that combined active mRNA loading, modular retargeting, and immune reprogramming within the same vesicle.182 By loading IFN-γ mRNA and displaying CD64 to dock targeting antibodies, the authors generated sExosomes that preferentially targeted glioblastoma cells and showed antitumor activity in vivo. The main significance of this study lies in showing that exosome-based systems can be engineered not only for biocompatible RNA delivery, but also for programmable targeting and immunomodulation through the vesicle surface itself. In addition, Huo et al. developed an engineered EV platform that uses the M. tuberculosis lipoprotein LpqH to confer macrophage-directed targeting.183 By displaying LpqH on vesicle surfaces, the LpqH-tagged microvesicles achieved more efficient macrophage delivery and mRNA loading than the corresponding exosome formulation, while also eliciting lower inflammatory toxicity than the LNP.Exosome-based mRNA delivery platforms should not be regarded as natural substitutes for LNPs, but as a distinct class of biologically derived carriers whose main value lies in integrating selective cell engagement, immune modulation, and mRNA delivery within the same system.184 The value lies not only in biocompatible cargo transport, but also in the possibility of modifying vesicle surface components to influence cell targeting and immune responses. This means that future platform design should move beyond improving loading efficiency alone and instead prioritize APCs-oriented delivery, programmable surface engineering, and coordinated control of intracellular delivery and immune output.174 Future engineering of exosome-based mRNA delivery systems should focus on several practical aspects, including surface ligand modification, membrane protein engineering, receptor-directed uptake, and selection of the exosome subtype best suited to the intended immune application. However, this approach still faces major obstacles to translation, particularly in manufacturing complexity, batch consistency, loading reproducibility, and large-scale production. The key issue for further development is therefore not to identify a more “natural” vesicle platform, but to establish standardized loading strategies, reproducible surface-engineering methods, and application-specific evaluation criteria, particularly for APCs targeting and immune instruction in vaccine settings.167PolymersPolymer-based mRNA carriers are typically formed through electrostatic assembly between cationic polymers and negatively charged RNA, generating polyplex-like nanostructures capable of RNA condensation and protection.185 These delivery platforms have emerged as promising vectors for mRNA vaccine applications, demonstrating remarkable capabilities in addressing multiple critical aspects of mRNA delivery.186 Their multifaceted functionality encompasses comprehensive nuclease protection, enhanced cellular internalization, efficient endosomal escape mechanisms, and precisely controlled release kinetics of the mRNA payload.187,188 Several cationic polymers have demonstrated exceptional utility in polyplex formation for non-viral vector development.189 The earliest polymer-associated branch focused on electrostatic complexation: cationic macromolecules were used to condense and protect mRNA, provide some degree of self-adjuvancy, and enable intradermal or intramuscular vaccination.190 This route reached clinical application before the emergence of most synthetic polymer nanocarriers. The best-documented examples are the RNActive® cancer vaccines CV9103/CV9104 and CV9201, as well as the rabies vaccine CV7201, all of which showed that polymer-associated or polycation-complexed mRNA vaccines could be administered safely and could induce measurable antigen-specific immune responses in humans.190,191 However, these studies also defined the limits of the first generation: immunogenicity was achievable, but control over biodistribution, intracellular release, and route-specific delivery remained limited, and clinical performance was still less consistent than what later became possible with optimized LNPs.192The transition from historical to modern polymer systems was driven by a different set of requirements. The field moved away from relying mainly on strong static RNA binding and toward programmable release chemistry, biodegradability, route adaptability, and cell-selective delivery.193 This shift responded to several problems that were difficult to solve with early polycationic carriers: excessive charge-related toxicity, overly tight mRNA binding, incomplete intracellular release, poor extra-muscular delivery, and weak control over which immune or tissue compartments actually expressed the encoded antigen. Current polymer platforms such as charge-altering releasable transporters (CARTs), Poly (β-amino ester) (PBAEs), and poly(amine-co-ester) (PACE) were developed to address those needs directly.186 CARTs introduced a release mechanism in which the carrier changes chemically after delivery and loses cationic character, thereby helping to decouple extracellular stabilization from intracellular release.194 PBAEs introduced a degradable scaffold whose backbone and end-group chemistry can be adjusted to tune transfection, immune-cell access, and tissue exposure.195,196 PACE polymers extended this logic to inhaled delivery, where aerosol stability, airway compatibility, and lung-specific performance become central design constraints.197 This change in design logic also changed what polymer systems are expected to do in vaccines. In current studies, polymer carriers are no longer used only to protect mRNA; they are increasingly designed to influence which cells take up the payload, where the formulation remains after administration, and which immune signals are generated during antigen expression. This is where polymer systems become mechanistically interesting. Ben-Akiva et al. showed that biodegradable lipophilic polymeric nanoparticles could deliver mRNA to splenic DCs and co-deliver adjuvants that triggered costimulatory signaling needed for antigen-specific CD8+ T cell activation.198 Kim et al. identified an immunostimulatory branched PBAE formulation for gastrointestinal mRNA vaccination that elicited B cell and T cell activation together with antigen-specific humoral and cellular responses, while maintaining stability under refrigeration.199 Suberi et al. showed that inhalable PACE polyplexes could deliver mRNA broadly through the lung, including to epithelial cells and APCs, and could induce systemic and lung-resident adaptive immunity after pulmonary vaccination.197 These studies indicate that polymer platforms can be linked directly to immune programming through APC access, mucosal route selection, local residence, and co-delivery of adjuvant signals, rather than acting only as neutral condensation agents.200Current studies have also begun to connect carrier structure more directly to immune programming and safety. Huang et al. developed a series of alternating copolymers, termed PHTA, in which ortho-hydroxy tertiary amine repeating units were used to condense mRNA, stabilize polymeric nanoparticles, and prolong circulation.201 In their cancer-vaccine study, PHTA-based nanoparticles showed negligible inflammatory side effects in vivo across intradermal, intranasal, and intravenous administration settings, yet the lead formulation PHTA-C18 still supported in vivo mRNA vaccine delivery and induced CD8⁺ T cell-mediated antitumor immunity. Guo et al. extended this logic further by screening a PEI-derived carrier library and identifying POctS, a formulation that combined site-specific mRNA delivery with inherent STING activation.202 POctS mediated transfection mainly in DCs and macrophages in leg muscle tissue and draining lymph nodes after intramuscular administration, and it also showed pancreas-selective expression after intraperitoneal delivery. In validation studies in vivo, this system showed antitumor efficacy in both a tumor-antigen mRNA vaccine setting in melanoma and an in situ MLKL mRNA vaccine setting in an orthotopic pancreatic cancer model.However, the central problem is that polymer performance depends strongly on structure–function balance.203 The polymer should bind mRNA tightly enough to protect it, yet release it efficiently in the cell; promote uptake and endosomal escape, yet avoid membrane injury and inflammatory toxicity; and remain stable during storage or aerosolization, yet still expose the mRNA to the target cells. The second challenge is translation. Current synthetic polymer nanocarriers have not yet produced the same level of clinical maturity as LNPs, in part because their broad chemical tunability makes standardization, structure–activity mapping, batch reproducibility, and scale-up more difficult.204 The third challenge is the limited ability to predict how polymer structure influences biodistribution, intracellular trafficking, and immune output. Small changes in backbone chemistry, end-group modification, or formulation architecture can alter tropism, intracellular trafficking, and immune output, but these effects are still not predictable enough to support rapid clinical selection.205,206Future development of polymer-based mRNA vaccine carriers is likely to proceed along several linked directions. Chemically programmed mRNA release remains a priority, because stable extracellular complexation and efficient intracellular unloading still need to be reconciled within the same material system; charge-altering and degradable polymers were developed to address this problem directly.207 Application-specific polymer design, in which tissue selectivity, route compatibility, and immune programming are considered together, is likely to become a major direction for future development.208Virus-like particles (VLPs)VLPs represent self-assembling nanostructures that recapitulate the morphological and antigenic characteristics of native viruses while maintaining non-infectious and non-replicating properties.209 These highly organized molecular assemblies comprise precisely engineered viral structural proteins, including capsid or envelope proteins, which can be strategically modified to present specific antigens or immune relevant epitopes. When mRNA encodes appropriately designed structural proteins, VLP-like assemblies can also be generated in situ after vaccination.210 Based on their structural composition and assembly mechanisms, VLPs can be systematically classified into two distinct categories: enveloped and non-enveloped architectures.211 Enveloped VLPs are derived from the organization of lipid bilayers and structural proteins characteristic of viruses such as influenza, HIV, and coronaviruses. In contrast, non-enveloped VLPs are constructed through the precise self-assembly of viral capsid proteins and natural human proteins, such as those derived from human papillomavirus (HPV), hepatitis B virus (HBV), and human paraneoplastic Ma antigen (PNMA).212,213,214 The selection of an appropriate VLP platform is governed by multiple factors, including the specific characteristics of the target pathogen and the desired immunological response profile.VLPs have emerged as a promising platform for RNA vaccine delivery because they can both protect encapsulated RNA from extracellular degradation and promote efficient uptake by target cells.215 Their delivery mechanism is based on the intrinsic ability of viral structures to package and transport genetic material into host cells.216 Segel et al. developed the selective endogenous encapsidation for cellular delivery (SEND) VLP-platform (Fig. 6c).217 This innovative system, based on the mammalian retrovirus-like protein PEG10, demonstrated exceptional capabilities in packaging, secreting, and delivering mRNA molecules. Their research underscored the remarkable modularity of SEND as a highly efficient mRNA vaccine delivery system. More recently, Yin et al. developed dendritic-cell-targeting virus-like particles as potent mRNA vaccine carriers by pseudotyping mRNA-packaging VLPs with an engineered Sindbis-virus glycoprotein.218 This dendritic-cell-directed design enhanced antigen-specific humoral and cellular immune responses and improved protection in mouse infection models.Despite the considerable promise exhibited by VLP-based mRNA vaccine platforms, several critical challenges remain to be addressed before their full clinical potential can be realized. A fundamental consideration lies in VLP architecture optimization for the optimal equilibrium between RNA encapsulation efficiency, structural stability, and immunogenic potency. The physicochemical properties of VLPs, including their dimensional parameters, morphological characteristics, and surface biochemistry, influence their capacity for mRNA packaging and delivery efficacy, while simultaneously modulating their interactions with the host immune system.219To address these challenges, innovative rational design strategies are being systematically investigated. These approaches encompass the strategic incorporation of RNA-binding domains and precise modulation of VLP surface charge distributions to enhance the overall performance metrics of VLP-based mRNA vaccine platforms.220 Such engineering efforts require careful consideration of multiple parameters to maintain the delicate balance between functionality and stability. Moreover, comprehensive evaluation of safety profiles and long-term therapeutic efficacy through rigorous clinical trials remains paramount. While utilizing non-replicating VLPs substantially mitigates the risk of insertional mutagenesis commonly associated with viral vector-based vaccine platforms, the potential occurrence of off-target effects or adverse immune responses cannot be definitively excluded.218 This requires thorough evaluation in well-designed clinical studies with extended follow-up.Recent advances in bioengineering and synthetic biology have catalyzed the development of innovative VLP architectures specifically engineered to enhance mRNA vaccine immunogenicity. These state-of-the-art approaches systematically address the optimization of VLP structural parameters, compositional elements, and functional characteristics for improved antigen presentation, enhanced immune stimulation, and improved vaccine efficacy. A particularly promising strategy involves the strategic co-delivery of VLPs with carefully selected immunostimulatory molecules, including TLR agonists and specific cytokines, designed to amplify both innate and adaptive immune responses. In a landmark study, Lu et al. demonstrated the efficacy of an mRNA cocktail encoding three essential structural proteins: S (spike), M (membrane), and E (envelope), which successfully generated SARS-CoV-2 VLP (RQ3013-VLP) in vivo and resulted in robust immune responses.221 Furthermore, self-assembling protein nanoparticles (SAPNs) have been widely explored as an advanced VLP platform for mRNA vaccine delivery.222,223,224 These engineered protein scaffolds exhibit remarkable capabilities in autonomous assembly into precisely defined nanostructures, enabling control over antigen presentation and immunogenic properties. In 2024, Zhang et al. comprehensively investigated human PNMA proteins, revealing their exceptional capacity to form highly ordered icosahedral capsids.214 They demonstrated that these structures can be strategically modified to efficiently encapsulate and deliver mRNA molecules. This pivotal discovery has opened new avenues for developing highly effective mRNA vaccine delivery systems utilizing engineered endogenous capsid protein nanoparticles.Current optimization strategies are increasingly focused on rational control of the VLP architecture rather than simple improvement of a single formulation parameter. One major direction is to engineer RNA-binding modules and packaging signals that increase cargo selectivity and encapsulation efficiency while preserving release competence after cellular entry.225 The second direction is fine control over surface electrostatics and membrane-proximal features to improve particle stability, cell interaction, and endosomal processing.226 The third strategy is programmable surface functionalization, including APC-targeting ligands, membrane-active peptides, or structurally arranged antigenic domains, which allows the particle exterior to be tuned as an active immunological interface rather than an inert shell.227Other novel delivery systemsOther novel delivery systems expand mRNA delivery engineering by integrating additional biological functions into carrier design. These platforms address delivery bottlenecks that are difficult to resolve through conventional material reformulation and move the field toward more active control of biodistribution, membrane engagement, intracellular trafficking, and local immune context. As a result, these parameters are increasingly treated as engineerable determinants of delivery performance.One emerging direction is the active reprogramming of biological identity to redirect tissue distribution. Albumin-binder lipids, for example, altered protein corona composition to enhance lymphatic drainage, lymph-node accumulation, and dendritic-cell uptake while reducing liver exposure, thereby making biodistribution itself a programmable property.228 Another direction is the introduction of active membrane-fusion capacity into the carrier. Proteolipid vehicles achieved this by embedding fusion-associated small transmembrane proteins into a well-tolerated lipid formulation, shifting intracellular access away from disruption-dependent endosomal escape toward fusion-enabled membrane engagement.229 A third direction is the design of self-reinforcing intracellular delivery circuits. In one hybrid system, intracellularly released gold nanoparticles modulated subsequent trafficking, inhibited endolysosomal fusion, and promoted repeated uptake, thereby amplifying transfection efficiency.230,231,232Beyond these examples, biomaterial-based systems have also shown dual functions in both mRNA delivery and immune modulation. A weakly alkaline nano-aluminum adjuvant tethered with antigen mRNA was reported to neutralize the acidic tumor microenvironment, suppress regulatory T cells, and enhance antigen-specific cytotoxic T cell responses.233 Overall, these systems suggest that the next stage of mRNA delivery engineering may depend less on incremental optimization of established carriers and more on integrating new biological functions into delivery design.Comparison between delivery systemsEndosomal escape efficiencyEfficient cytosolic delivery remains the primary determinant of functional mRNA activity, linking material design directly to biological output. Across currently developed delivery platforms, LNPs, polymer-based systems, exosomes, and VLPs display distinct performance characteristics in both in vitro transfection assays and in vivo protein expression studies (Table 3). In vitro systems typically emphasize cellular uptake and translation capacity, where LNP formulations consistently achieve high reporter-gene expression across diverse cell types, reflecting optimized particle size, ionizable lipid composition, and membrane fusion capacity.234 Polymer-based systems exhibit broader structural tunability but often show greater variability in transfection efficiency depending on polymer architecture, charge density, and degradability.185 Exosomes and VLP-derived carriers are often internalized through receptor-mediated endocytic pathways, but the efficiency of endosomal escape and productive cytosolic release appears to be more context-dependent and is generally less well defined than in ionizable LNP systems.176,218 In vivo, these differences become more pronounced, as delivery efficiency is further shaped by serum stability, biodistribution, immune recognition, and tissue penetration. Among current platforms, LNP systems generally produce the highest systemic protein expression levels following intravenous or intramuscular administration, while polymer systems show heterogeneous performance profiles depending on chemical composition.235 Exosome-based and VLP-derived systems demonstrate advantages in biological targeting and tolerability but often require further optimization to reach expression levels comparable to LNP formulations.236A major source of divergence between in vitro and in vivo delivery efficiency lies in the endosomal escape (Fig. 6b). Internalization alone does not ensure functional delivery, as the majority of internalized nucleic acids remain confined within endosomal compartments and are ultimately degraded in lysosomes.11 Ionizable lipid-containing LNPs facilitate partial disruption of endosomal membranes through pH-dependent protonation, enabling limited cytosolic release of mRNA molecules.237 However, quantitative imaging studies indicate that less than 5% on average of mRNA delivered by LNPs can escape endosomes, suggesting that endosomal escape represents a key efficiency bottleneck even in clinically validated platforms.238 Polymer-based systems similarly rely on endosomal destabilization mechanisms, often mediated through proton buffering capacity or membrane-disruptive motifs, yet the magnitude and kinetics of escape vary substantially depending on polymer composition.239 Exosomes and VLPs-derived carriers are often internalized through receptor-mediated pathways, but the extent to which they promote productive cytosolic release appears to be more context-dependent and is generally less well defined than in ionizable LNP systems.240,241 Consequently, differences in endosomal escape efficiency provide a mechanistic explanation for the variability observed across delivery systems in functional protein expression output (Fig. 6b, Table 3).Although all delivery systems ultimately face the same requirement of cytosolic access, they differ substantially in how efficiently and controllably this step is achieved. Current optimization strategies across delivery platforms converge on the same objective: increasing the efficiency of productive cytosolic release while minimizing membrane damage, inflammatory burden, and systemic toxicity.239 In comparative terms, LNPs remain the most validated system for pH-responsive endosomal escape and currently provide the most consistent balance between delivery efficiency and translational feasibility.238,242 Polymer-based systems offer greater chemical tunability and can be tailored for environmentally responsive intracellular release, but their performance is more formulation-dependent and less uniform across biological settings.234 In contrast, exosome- and VLP-based platforms are more strongly shaped by biologically guided uptake and membrane interactions, yet their productive transition from endocytic compartments to the cytosol remains less well defined and generally less controllable.184,218,241,243 These differences indicate that endosomal escape should not be viewed as a single universal parameter, but as a platform-dependent performance constraint that must be optimized together with tolerability, targeting behavior, and immune signaling outcomes.236,244Specific targetingAcross delivery systems, differences in targeting performance are most clearly reflected in how precisely in vivo distribution can be controlled (Fig. 6c) (Table 3). Current evidence suggests that LNP remain the most mature platform for targeted delivery research among the four systems. On the one hand, traditional LNPs exhibit a significant liver bias;245 on the other hand, adjustments to the ratios of ionized lipids, cholesterol, cofactor phospholipids, and PEG lipids have gradually shifted the organ distribution of LNP from passive enrichment to programmable regulation.246 The SORT strategy further demonstrates that by introducing additional lipid components, LNP can achieve reproducible distribution redirection among the liver, lungs, and spleen.153 The targeting capability of polymer delivery systems relies heavily on the designability of their chemical structures. Lin et al. reported a peptide ionizable lipid-driven organ-targeting (PILOT) platform based on amino-acid-modular design. By incorporating specific amino acids and functional groups into peptide ionizable lipids, the authors generated LNPs with tissue-selective mRNA delivery to the lungs, liver, spleen, thymus, and bone. The same platform enabled co-delivery of PEmax mRNA and engineered pegRNA for in vivo prime editing, indicating that rational lipid design can simultaneously improve organ selectivity and editing function.247 Feng et al. reported an albumin-recruiting lipid nanoparticle system with high lymphatic drainage properties and no accumulation in liver tissue.228 They constructed an ionizable lipid library with albumin binding ability as an alternative to traditional polyethylene glycol-coupled lipids, and identified an EB-modified LNP formula with high in vivo expression levels. With the help of albumin, it is transported to the lymph nodes through the intramuscular lymphatic vessels, is highly internalized by DCs, and has low permeability to intramuscular blood vessels, thus avoiding accumulation in liver tissue and solving the hepatotoxicity problem of traditional LNP preparations from the delivery route. Polymer-based systems offer high chemical tunability for active targeting, but in vivo targeting performance remains more formulation-dependent and less predictable than that of LNPs.197Exosomes may provide favorable biocompatibility and natural tropism, but their targeting performance remains highly dependent on source, preparation, and surface composition. Current efforts therefore focus on engineered exosomes with improved cell-selective recognition (Fig. 6c).174,248 VLP-based targeting is shaped more strongly by receptor-directed entry and tropism engineering than by passive physicochemical distribution, which gives these systems distinctive targeting potential but also makes their performance highly design specific.225,249Immune signaling regulationRecent studies indicate that, beyond enabling cytosolic mRNA delivery, the intrinsic adjuvant activity and innate immune profile of the carrier are also major determinants of vaccine immunogenicity (Table 3).134 Innate immune activation by LNPs arises through multiple coupled processes, including receptor engagement determined by surface chemistry, sensing of membrane perturbation during endosomal escape, and changes in dendritic-cell localization or differentiation.239,250 This creates a central design tension: stronger membrane activity may improve cytosolic delivery, but the same changes can also increase inflammatory burden and reactogenicity.251,252 In addition, different applications require different immune biases. Infectious-disease and tumor vaccines generally benefit from controlled IL-1, type I IFN, and T cell priming, whereas protein replacement, gene editing, and repeated administration require lower background inflammation.253,254,255 Accordingly, current LNP design increasingly treats innate immune output as a programmable property that can be tuned through lipid chemistry, formulation composition, and immunomodulatory cargo selection.252,253,254Polymer systems more often function as low-inflammatory delivery scaffolds with limited intrinsic immunostimulatory activity. Their immune effects are therefore commonly introduced through added adjuvants, built-in stimulatory motifs, or site-specific triggering after organ or cell targeting.202,256 This gives polymer platforms greater flexibility for spatially restricted immune programming and may help reduce systemic toxicity. However, their immunological performance remains highly dependent on the material structure, release behavior, and application context, making rational polymer design a key determinant of whether delivery efficiency and immune activation can be effectively coordinated.186 Exosomes typically exhibit low innate immunostimulatory activity, but demonstrate good biocompatibility, cross-barrier ability, and source plasticity, confining immune signals to local generation within target cells or lesions.257Immune regulation by exosomes is primarily achieved through local delivery followed by local generation of innate immune signals (such as engineered surface ligands or replacement with exosomes from different biological sources). Dong et al. used CD64 surface anchoring to display anti-CD71 and anti-PD-L1 antibodies on the surface of mRNA-loaded sExosomes and further loaded IFN-γ mRNA as the therapeutic cargo.182 This design illustrates how immune regulation can be achieved through three coordinated levels: first, surface antibodies enhance tumor localization; second, IFN-γ mRNA generates local inflammatory signals within target cells; and third, MHC-I upregulation, increased CD8⁺ T cell infiltration, and increased CD86⁺ macrophages amplify the intratumoral immune response. Jang et al. developed ExoSTING, which encapsulates the STING agonist into Exosomes, enabling them to preferentially target APCs in the tumor microenvironment.258 The result was a more than 100-fold increase in in vivo potency, enhanced local Th1 response and CD8⁺ T cell recruitment, while therapeutically active doses did not induce systemic inflammatory cytokines. Chen et al. combined tumor-derived Exosomes with an in situ formed hydrogel scaffold vaccine, allowing the EV antigen and adjuvant to form a long-lasting depot locally, continuously recruiting DCs into the scaffold, thereby generating a more durable tumor-specific CD8⁺ T cell response.259 Cao et al. constructed platelet–Treg hybrid exosome nanovesicles and then locally released anti-IL-23 antibody via an MMP-cleavable linker.260 Treg-exosomal cargo reprograms DCs into tolerogenic DCs and promotes Treg amplification, leading to localized tolerance-mediated immune remodeling characterized by increased IL-10/TGF-β, decreased TNF-α, and decreased levels across multiple pro-inflammatory axes.VLPs retain the size, multivalent surface structure, and efficient cell entry capability of viruses, making them more readily perceived by the immune system as “particulate exogenous substances requiring processing”.217 In vaccine scenarios, this property typically enhances antigen presentation and antibody responses. However, because off-target delivery to APCs can trigger innate sensing and strong adaptive immunity, and intravenous or local administration easily generates anti-carrier neutralizing antibodies, it directly limits the scope for repeated dosing.261 Studies have found that even after deleting non-structural proteins, which reduce viral response-related genes such as RIG-I, MDA5, IFNA, and IFNB1, intravenous injection of SFV-based VLPs still rapidly generates neutralizing antibodies and renders repeat dosing after 6 weeks ineffective.225,262,263 The strategy for immune regulation using VLPs lies in reducing adverse immune stimuli while pursuing controllable maturation, multivalent display, and targeted enhancement of immune stimulation. Zhang et al. comprehensively investigated human PNMA proteins, revealing their exceptional capacity to form highly ordered icosahedral capsids.214 They demonstrated that these structures can be strategically modified to efficiently encapsulate and deliver mRNA molecules. This pivotal discovery has opened new avenues for developing highly effective and safe mRNA vaccine delivery systems utilizing engineered endogenous capsid protein nanoparticles. Another study incorporated retroviral protease into Gag-based VLPs-forming mRNA vaccines, allowing Gag to mature and form more mature VLPs. This resulted in increased binding and neutralizing antibodies in both HIV-1 and SARS-CoV-2 models.264mRNA vaccines in disease prevention and treatmentThe clinical expansion of mRNA vaccines has made clear that this platform is not confined to pandemic control, but is increasingly relevant across a broader therapeutic landscape (Table 4).265 Their programmability, rapid manufacturing, and capacity to encode structurally optimized antigens have enabled broad deployment in prophylactic vaccination and increasingly informed therapeutic applications. However, the translational demands of different disease settings are not equivalent (Table 4). In infectious diseases, the central objective is the rapid induction of durable protective immunity at the population scale.266 In cancer, efficacy depends on productive antigen presentation, effective T cell priming, and the ability to overcome local immune suppression.267 In autoimmune and other non-classical settings, by contrast, the relevant goal may be to attenuate inflammatory signaling while preserving sufficient expression of the encoded cargo.268 Accordingly, the expansion of mRNA vaccination across diseases should be understood not as a simple increase in indications, but as a series of distinct immunological design problems that require different balances among antigen selection, translation efficiency, tissue targeting, and innate immune activation. In this section, we discuss how these design principles are translated across major disease settings, with emphasis on how pathological context reshapes the requirements for antigen selection, formulation, delivery precision, and immune modulation (Table 4).Table 4 Representative clinical/preclinical studies of mRNA vaccines in various diseasesFull size tableInfectious diseasesInfectious diseases remain the most clinically advanced application of mRNA vaccines and provide the clearest evidence for the translational value of this platform (Table 4).269 The field has already expanded across multiple viral targets, including SARS-CoV-2, Zika virus, HIV, influenza virus, cytomegalovirus (CMV), and RSV, thereby establishing a broad prophylactic application landscape.266 The major advantages of mRNA vaccines in this setting include rapid antigen design, scalable cell-free manufacturing, and the ability to iteratively update encoded antigens in response to pathogen evolution.270 However, as the field moves beyond initial proof-of-concept, the key challenges are shifting from simple platform feasibility to antigen breadth, immune durability, mucosal protection, and rational multicomponent design for genetically diverse or immunologically complex pathogens.The development of SARS-CoV-2 mRNA vaccines has shifted from the early stages of the pandemic, focusing on rapid validation of the original full-length spike, to a new phase centered on immune evasion, complex prior immune backgrounds, and upgrades to second-generation platforms. While updated mRNA vaccines can still provide additional protection against severe outcomes in the real world, their protective advantage primarily lies in reducing disease severity rather than providing stable blocking of the infection itself.271,272 In the interim evaluation for the 2024–2025 season, updated vaccines provided ~33% protection against COVID-19-related emergency/urgent visits in adults and ~45%–46% protection against hospitalization elderly individuals.273 This suggests that current strategies primarily maintain realistic protection against ongoing antigenic drift, rather than achieving a breakthrough in broad-spectrum, durable, and near-sterile immunity. Against this backdrop, second-generation mRNA vaccines have begun substantial clinical translation. Chalkias et al. reported in the NextCOVE Phase III trial that domain-focused mRNA-1283, which encodes only the NTD and RBD of Spike, achieved non-inferiority compared to the first-generation mRNA-1273 and induced a higher neutralizing antibody response.274,275 Meanwhile, the saRNA route has also shown clear clinical feasibility. Oda et al. demonstrated that ARCT-154, as a fourth booster dose, achieved immunogenicity non-inferiority against SARS-CoV-2 and superior efficacy against BA.4/5.276 Hồ et al. further showed in a combined phase 1/2/3 study that ARCT-154 provided 56.6% protection against any type of COVID-19 and 95.3% protection against severe COVID-19.277 These studies collectively demonstrate that the focus of current clinical research on SARS-CoV-2 mRNA vaccines has shifted from proving the platform’s effectiveness to how to improve immune quality, dose efficiency, and protective durability in the context of continuous mutation evolution.However, the key bottlenecks of existing platforms have not fundamentally changed. One of the most prominent limitations is that intramuscular injection of mRNA vaccines has difficulty effectively establishing upper respiratory tract mucosal immunity. Lasrado et al. showed that while the XBB.1.5 mRNA booster significantly increased serum neutralizing antibodies and peripheral IgG, it did not significantly enhance nasal mucosal neutralizing activity.271 The response gain was mainly limited to systemic circulation, which provides a direct immunological explanation for the current vaccine’s “better performance in preventing severe illness but less stable infection prevention.” At the same time, “immune imprinting” continues to influence the remodeling of responses to newer antigens. Tortorici et al. demonstrated that even after XBB.1.5 mRNA booster, humoral responses were still significantly dominated by memory B cells induced by previous SARS-CoV-2 spike exposure, indicating that simply relying on annual strain changes is insufficient to completely escape the constraints of previous immune trajectories.278 Therefore, the core challenge facing current SARS-CoV-2 mRNA vaccines is how to rebuild broader and more durable protection at the respiratory mucosal front under conditions of continued antigenic drift and persistent immune imprinting.Currently, second-generation strategies are being advanced at three interconnected levels: antigen engineering, platform engineering, and delivery engineering. In terms of antigen engineering, the domain-focusing strategy represented by Moderna′s mRNA-1283 essentially reduces the antigen burden while directing the response further toward the immunogenic neutralizing region, thereby increasing the immune output per unit dose.274 In terms of delivery engineering, Liu et al. developed CAS-LNP, which improved the integrity of LNP nebulization through charge-assisted stabilization, achieving effective lung mRNA delivery in mice, dogs, and pigs and inducing immune responses containing both systemic and mucosal components.155 Correspondingly, McMahan et al. further demonstrated in rhesus monkeys that mucosal enhancement significantly improved local respiratory protection, and the protective effect was more closely related to mucosal humoral and cellular immunity.279 In addition, antigen design itself is beginning to shift from “passively following the prevalent strain” to “actively predicting future escape directions.” Youssef et al. proposed EVE-Vax, which can generate Spike designs that simulate future immune escape trends, providing a new computational framework for prospectively evaluating vaccine antigens.280 Overall, the key to the next phase of SARS-CoV-2 mRNA vaccines is no longer simply to accelerate the pace of strain replacement, but to further advance the protective endpoint from primarily preventing severe illness to balancing infection prevention and transmission blocking through more focused antigen design, more efficient expression platforms, and delivery methods more geared toward the respiratory front lines.281Beyond SARS-CoV-2, the clinical development of mRNA vaccines for infectious diseases has shown significant divergence: RSV is one of the few mRNA vaccine targets to have completed advanced late-stage clinical validation and achieved substantial translational progress.282 In contrast, mRNA vaccines against influenza, CMV, HIV, Epstein–Barr virus (EBV), malaria, and bacterial pathogens remain largely at an earlier stage of clinical expansion. For these indications, field advancement has been driven mainly by high-impact preclinical studies. For example, the pre-fusion F protein mRNA vaccine for RSV has demonstrated clear protective effects in the elderly population, indicating that mRNA platforms can move from emergency epidemic applications into routine infectious disease vaccine systems.283 In contrast, the key issue for other pathogens is no longer whether the platform can be used in human studies, but whether a truly translational design framework can be established around the immunological limitations of different pathogens.284 For influenza, this limitation mainly manifests as the tension between antigenic breadth and persistent variability;285 correspondingly, multivalent, cross-subtype, and component rebalancing mRNA design has become a core direction, and related studies have shown broad protective potential covering all known influenza A subtypes and influenza B lineages in animal models.286 For CMV, the research focus has shifted to multi-antigen co-expression. The strategy of co-encoding gB and the pentamer complex has induced potent and durable humoral and cellular immunity in preclinical models,287 suggesting that complex viruses are more likely to benefit from combined immunogens with complementary structures and functions, rather than single-target substitution.288This trend is even more pronounced in more complex infection scenarios such as HIV, EBV, and malaria. HIV mRNA vaccine development has shifted from the early approach of “enhancing antigen expression” to B-cell lineage-directed programming centered on germline targeting, sequential enhancement, and membrane-anchored Env structures.289 The goal is not simply to induce a response, but to promote the maturation of broad-spectrum neutralizing antibody precursors along a pre-defined pathway.290,291 EBV and malaria studies further demonstrate that the advantage of mRNA platforms lies not only in antigen expression efficiency but also in their ability to support immune remodeling that matches the biological stage of the pathogen, such as inducing EBV neutralizing antibodies that protect humanized mice or preferentially shaping resident memory T cells against liver-stage infections.292,293 Even in the field of bacterial vaccines, studies on mRNA-LNP against Y. pestis have shown that, after optimization of antigen conformation and sequence, a single dose of immunization can provide effective protection against lethal infection.294 Therefore, the core of developing mRNA vaccines for non-COVID-19 infectious diseases lies in truly integrating antigen engineering, immunomodulatory design, and delivery system optimization with the pathogenic mechanisms and protective endpoints of specific pathogens.17,269,270CancersTherapeutic mRNA cancer vaccines have entered a development stage shaped by both clinical validation and tumor-specific biological constraints (Table 4). There is no single design paradigm across cancer types. Instead, vaccine design depends on antigen type, mutational burden, immune microenvironment, disease stage, and compatibility with existing treatment systems.295 Tumors with high mutational burden and sensitivity to immune checkpoint blockade are more suitable for personalized neoantigen vaccines, whereas tumors with stable shared antigens are more amenable to off-the-shelf multi-antigen platforms.296 In tumors with marked immunosuppression or limited antigen presentation, greater emphasis is placed on delivery efficiency and innate immune activation.297 Current development therefore increasingly focuses on the coordinated matching of antigen selection, RNA design, delivery strategy, combination therapy, and clinical setting.298Melanoma remains one of the most clinically advanced indications. Intismeran autogene (Moderna/Merck’s mRNA-4157/V940) uses a personalized neoantigen design encoding up to 34 neoantigens and, in combination with pembrolizumab, has induced sustained neoantigen-specific CD4⁺ and CD8⁺ T cell responses while showing the potential to reduce recurrence risk in high-risk resected melanoma.299,300 Personalized neoantigen vaccination has also shown promise in pancreatic ductal adenocarcinoma, where BioNTech’s autogene cevumeran uses an RNA–LPX platform to encode up to 20 selected neoantigens and has been associated with durable vaccine-induced T cell responses and longer recurrence-free survival after surgery.301 In contrast, non-small cell lung cancer illustrates the value of fixed shared-antigen strategies. BioNTech’s BNT116 encodes six shared tumor-associated antigens and is designed for rapid clinical use without individualized manufacturing, while also taking advantage of APC-directed delivery by the RNA-LPX platform (Table 4).For tumors with low antigenicity and strong immunosuppression, vaccine design increasingly depends on delivery-driven immune reprogramming. In glioma, RNA-LP/RNA-LPA platforms loaded with tumor-derived antigen RNA have been used to amplify innate immune activation, recruit immune cells, and promote inflammatory remodeling of the tumor microenvironment in both preclinical and early clinical settings.302 In hepatocellular carcinoma, off-the-shelf vaccine development has shifted from rare mutated neoantigens toward aberrant splicing-derived neoantigens with broader coverage and greater relevance to immune escape.303 More broadly, recent studies indicate that the mRNA-LNP platform itself can function as an immune-modulating tool by inducing transient type I IFN signaling, activating APCs, and enhancing subsequent CD8⁺ T cell priming.96,134 These findings suggest that next-generation mRNA cancer vaccines should focus not only on antigen identity but also on the integration of innate immune activation, antigen presentation, and T cell expansion to improve efficacy in poorly immunogenic solid tumors.304,305Cardiovascular diseasesCardiac-directed mRNA therapy is currently being explored in ischemic heart disease, heart failure, conduction disorders, and selected hereditary cardiomyopathies. The most clinically mature project is AstraZeneca’s AZD8601, which encodes VEGF-A as a naked modified mRNA.306,307 Instead of using LNPs, it is prepared with citrate buffer and administered by local epicardial injection at multiple sites during CABG. Clinical reporting indicators suggest an improving trend in left ventricular function and heart failure biomarkers. Another approach is Moderna′s mRNA-0184, which encodes a relaxin-2–variable light chain kappa fusion protein.308 This mRNA is encapsulated in LNPs for systemic administration, aiming to achieve anti-fibrotic and heart failure-improving effects through short-term in vivo expression of secreted proteins. Preclinical studies have further expanded the functional boundaries of cardiac-targeted mRNA. hPKM2 modified mRNA (modRNA) promoted cardiomyocyte division and improved contractile function in a porcine ischemia reperfusion model;309 VEGF-C mRNA-LNP reduced inflammation and improved repair after myocardial infarction by promoting lymphangiogenesis;310 and PSAT1 modRNA promoted cardiomyocyte proliferation, reduced scarring, and improved cardiac function in a mouse myocardial infarction model.311 Despite their different cargos, these studies share a common design logic. The encoded products include pro-angiogenic factors, secreted anti-fibrotic proteins, electrophysiological regulators, and missing structural proteins, but all are intended for short-term expression, local or targeted delivery, and avoidance of permanent genomic modification.17However, the primary limitation currently facing this approach remains how to actually deliver RNA to the heart.312 Traditional mRNA-LNPs are usually guided into the liver by apoE after intravenous administration, making it difficult to achieve effective enrichment in the myocardium.313 Shuvaev et al. constructed a cardiac-oriented cLNP and demonstrated that it significantly accumulated in the heart within 30 min after intravenous injection in Apoe − /− mice.312 This phenomenon was also verified in Apoe− /− rats, and similar results were observed in wild-type mice after pre-administration of plasma apoE with siRNA. To demonstrate that this delivery did not stop at the vascular level, the study further used siRNA-ATP2A2 as a proof-of-concept, where ATP2A2 encodes the key cardiac calcium cycle protein SERCA2A. The results showed that SERCA2A in the myocardium could be almost completely knocked down after cLNP delivery; the contraction amplitude of isolated cardiomyocytes decreased, and the contraction and relaxation dynamics slowed, indicating that the mRNA had entered the function-related cardiomyocytes and generated function. This study suggests that intravenously administered mRNA nanoparticles can directly enter the myocardium and transiently regulate cardiomyocyte function under specific conditions, providing important verification for cardiac-directed RNA therapy. However, its limitation lies in its high dependence on the Apoe − /− background or pre-clearance of apoE, indicating that this platform has not yet truly overcome the barrier of preferential uptake by the liver under normal physiological conditions. Further validation with luciferase mRNA and Cre mRNA in mice revealed that the in vitro screening results were consistent with cardiac transfection performance in vivo. These acid-degradable PEG-LNPs significantly enhanced transfection in the heart, while hepatic non-target uptake was lower than that of standard LNPs. This suggests that acid-degradable PEG-lipids may be an important formulation direction for improving the diffusion and transfection efficiency of cardiac LNPs.Neurodegenerative disordersNeurodegenerative diseases, including Alzheimer’s disease, Parkinson’s disease, amyotrophic lateral sclerosis (ALS), and Huntington’s disease, are a group of chronic diseases characterized by progressive neuronal dysfunction and loss.235 With population aging, the burden of neurodegenerative disorders continues to rise. Current treatments are primarily symptomatic, with limited options capable of truly stabilizing and altering disease progression. One long-standing key obstacle is that most therapeutic drugs struggle to effectively cross the blood brain barrier (BBB) and achieve sufficient and controlled distribution and expression within the brain.314 Current mRNA-based approaches for neurodegenerative disorders mainly follow three strategies.315 The first strategy is active immunization against pathogenic protein aggregates, particularly amyloid-β. Although mRNA-based Alzheimer’s disease vaccines remain at an early stage, the clinical validation of antibody-mediated amyloid-β targeting by lecanemab supports the biological rationale for antigen-directed immunotherapy in early Alzheimer’s disease.316 The second is protein function compensation or aggregate clearance mRNA therapy. Shi et al. introduced DSPE-PMPC into the conventional LNP component (PLNP) to enhance BBB permeability and reduce non-target tissue loss, enabling systemic delivery of TRIM11 mRNA to the brain for tau disaggregation and neuroinflammation alleviation in Alzheimer’s disease models.317 These results showed that after the PLNP system delivered TRIM11 mRNA, the cumulative amount in the hippocampus was approximately 8.1 times higher than that of the control LNP, and the neuronal transfection efficiency was increased by more than 30 times. The third strategy is immunomodulatory and neuroprotective mRNA therapy. In Parkinson’s disease models, intramuscular GM-CSF mRNA-LNP injection increased peripheral GM-CSF levels, promoted Treg expansion, reduced neuroinflammation, and protected dopaminergic neurons in both MPTP-induced and α-synuclein overexpression models.318However, the decisive challenge currently facing this field remains BBB delivery.314 Neurodegenerative diseases require RNA to cross the BBB after systemic administration and further achieve sufficient brain region exposure, cell type selectivity, and controlled expression duration.156 Dong et al. reported an OS4T LNP in which a BBB-crossing small-molecule motif and Tat peptides were incorporated into ionizable lipids.319 After systemic administration, this system achieved a more than 50-fold increase in intrabrain mRNA translation compared with Onpattro/MC3 LNP, and achieved expression in neurons, astrocytes, microglia, and endothelial cells. A further study screened 72 BBB-crossing lipids to obtain MK16 BLNP, which can deliver mRNA to neurons and astrocytes in a wide range of regions of the whole brain after intravenous administration and maintain delivery capability in ex vivo human brain samples, suggesting that systemic whole-brain expression has moved from concept to platform. Overall, the next key optimization directions for mRNA therapy for neurodegenerative diseases focus on improving cross-BBB efficiency, reducing off-target effects in the liver and periphery, enhancing selectivity for specific brain cell subsets, extending the effective expression window, and supporting repeated dosing.315Autoimmune diseasesAutoimmune diseases include multiple sclerosis, type 1 diabetes, rheumatoid arthritis, systemic lupus erythematosus, and psoriasis, which are chronic diseases characterized by immune tolerance disruption and persistent inflammatory damage.320 Although current treatments can suppress inflammation and relieve symptoms, most still rely on long-term broad-spectrum immunosuppression, making it difficult to achieve true antigen-specific disease modification.321 Current mRNA-based strategies in autoimmune disease mainly follow three directions. The first is antigen-specific tolerance induction.322 In multiple sclerosis, m1Ψ-modified mRNA, combined with the lipoplex (LPX) platform encoding myelin antigens such as MOG35–55 and PLP139–151, can induce antigen-specific regulatory T cells and inhibit the Th1/Th17 response in experimental autoimmune encephalomyelitis (EAE) after systemic administration.268 Low-inflammatory delivery systems such as PS-LNP and DOPG-incorporated LNP further favor the mRNA toward tolerant APCs and reduce innate immune activation.268,323 In type 1 diabetes, β-cell-directed PD-L1 mRNA-LNP upregulates PD-L1 on the surface of β cells in a prediabetic model, alleviating pancreatitis and delaying disease progression.324 The second is immunomodulatory mRNA therapy. Kenney et al. developed a membrane-anchored IDO1 mRNA-LNP platform for transient in vivo expression of an immunometabolic regulator.325 By enhancing the tryptophan–kynurenine pathway, this strategy reduced inflammation in rheumatoid arthritis, experimental autoimmune encephalomyelitis (EAE), and graft-versus-host disease (GVHD) models. This work expands the role of mRNA beyond antigen delivery and supports its use for short-term expression of metabolic immunomodulatory molecules. The third is cell-directed in vivo reprogramming. In a clinical study of refractory systemic lupus erythematosus, intravenous administration of CD8-targeted LNPs carrying CD19 CAR mRNA enabled transient in vivo engineering of endogenous CD8⁺ T cells, which subsequently expressed CD19 CAR and eliminated pathogenic B cells.326The major challenges in this field remain the intrinsic inflammatory properties of many delivery systems, insufficient disease-specific targeting, and the difficulty of achieving repeatable dosing while preserving tolerogenicity. Traditional mRNA-LNP platforms are naturally biased toward inducing innate immune activation, while autoimmune therapies often require low-inflammation, tolerability, or cell-specific immune reprogramming.327 One study showed that mRNA-LNP can induce substantial myeloid activation, cytokine release, and complement activation across different donors.328 Notably, the response patterns to mRNA-LNP were not identical between patients with autoimmune diseases and healthy individuals. These findings suggest that the safety and efficacy of this platform are strongly influenced by the disease context. One strategy to address these challenges is to reduce the inherent immunostimulatory nature of the mRNA platform, such as through m1Ψ modification, anionic lipid incorporation, or tolerant lipid design, reducing inflammatory signals such as IL-6 and TNF-α and biasing the IL-10/TGF-β environment.131,329 At the same time, a platform system that can meet the requirements of low inflammation, targeted delivery, repeatable dosing feasibility and selective regulation of disease-related immune circuits will be established.Metabolic disordersMetabolic disorders comprise a broad group of inherited and acquired diseases caused by defects in key enzymes, transporters, or metabolic regulatory pathways. They are commonly associated with abnormal metabolite accumulation, deficiency of essential products, and progressive organ damage.330 Standard management for many of these conditions still relies on dietary restriction, cofactor supplementation, removal of toxic metabolites, or supportive care, but these measures often fail to prevent recurrent metabolic decompensation, neurological complications, liver injury, and impaired growth.331 The mRNA platform is well suited to this field because it can restore missing protein expression in vivo without genomic integration, and existing LNP systems already provide efficient hepatic delivery. This combination has made liver-directed inherited metabolic disease one of the earliest non-vaccine indications to enter clinical development.332Current mRNA strategies in metabolic diseases mainly follow two directions. The first is liver-directed enzyme replacement for inherited disorders such as propionic acidemia, methylmalonic acidemia, phenylketonuria, ornithine transcarbamylase deficiency, and glycogen storage disease type Ia.332 A representative example is Moderna’s mRNA-3927, a dual-mRNA therapeutic for propionic acidemia that delivers two N1mΨ-modified mRNAs encoding propionyl-CoA carboxylase subunits alpha (PCCA) and beta (PCCB) in LNPs.333 In the phase 1/2 clinical study, 16 participants received a total of 346 intravenous doses across the dose-optimization and extension phases. Early analyses showed reductions in disease biomarkers, including 2-MC, 3-HP, and the C3/C2 ratio, and suggested an approximately 70% reduction in the risk of metabolic decompensation among participants with prior events. These data support the feasibility of mRNA-based enzyme replacement in inherited metabolic disease, particularly in liver-directed settings.The second direction is the use of mRNA to remodel the metabolic and immune microenvironment in MAFLD/MASH and related hepatocellular carcinoma. In this setting, persistent metabolic stress, oxidative injury, and immune dysregulation continuously reshape the liver microenvironment and reduce the efficacy of conventional treatment and immunotherapy. Patient samples showed downregulation of T cell protein tyrosine phosphatase (TCPTP) and excessive activation of STAT1/STAT3 in MAFLD liver tissue, identifying the TCPTP–STAT axis as an important node linking metabolic dysfunction, inflammatory imbalance, and tumor progression.334,335 However, direct delivery of TCPTP using conventional mRNA-LNP systems may aggravate hepatic oxidative stress and inflammation. To address this problem, Yu et al. developed a Def-LNP formulation based on a vitamin E-derived phosphatidylcholine helper lipid (VEPC).336 Compared with conventional LNPs, this system showed improved colloidal stability, faster lysosomal escape, reduced oxidative damage, and more sustained hepatocyte mRNA expression. In mice, liver expression lasted approximately 3 days, versus ~24 h for conventional LNPs, and favorable liver enrichment, biocompatibility, and safety were further supported in pigs and non-human primates. This study illustrates a co-design strategy in which the delivery system and encoded cargo are optimized together to correct metabolic dysfunction while reshaping the immunosuppressive liver microenvironment.Indication-specific design principles for mRNA vaccinesThe application of mRNA vaccines across disease settings should be guided by indication-specific design principles rather than a uniform strategy of maximizing antigen expression or immunogenicity. Different indications impose distinct requirements on antigen selection, RNA format, delivery route, expression duration, innate immune activation, and safety tolerance. Therefore, the central design goal is to match the magnitude, location, duration, and immunological context of encoded cargo expression to the biological needs of each disease.10,265For infectious diseases, mRNA vaccines are mainly designed for prophylactic immunization at the population scale. The desired outcome is rapid induction of neutralizing antibodies, coordinated T cell help, and durable immune memory, with controlled innate activation serving as an adjuvant-like signal.337 Key priorities include rapid antigen updating, multivalent or variant-adapted design, scalable manufacturing, distribution stability, and acceptable reactogenicity.338,339 Major challenges include antigenic variation, immune escape, insufficient mucosal immunity, immune imprinting, and waning protection.266 Cancer vaccines require a different design logic. Their purpose is to amplify tumor-specific cellular immunity in patients with established disease. Therefore, neoantigen selection, antigen presentation, dendritic-cell targeting, CD8+ T cell priming, and combination with immune checkpoint blockade are central design considerations. The main barriers include tumor heterogeneity, antigen loss, weak immunogenicity, impaired antigen presentation, and the immunosuppressive tumor microenvironment.267,301 Autoimmune diseases and metabolic disorders represent settings in which stronger immunogenicity is not necessarily desirable. Autoimmune applications require antigen-specific tolerance induction or immune reprogramming, which demands low-inflammatory RNA chemistry, tolerogenic delivery, cell-selective targeting, controlled expression, and repeat-dose compatibility.268 Metabolic disorders are closer to mRNA therapeutics or protein-replacement strategies, where the priority is predictable, tissue-selective, and repeatable protein expression with minimal innate activation and long-term toxicity.332,333These differences indicate that next-generation mRNA vaccine design should be based on indication-matched engineering. Infectious-disease vaccines require rapid and scalable immune priming; cancer vaccines require personalized antigen selection and potent T cell activation; autoimmune applications require tolerogenic immune modulation; and metabolic disorders require controlled tissue-selective protein expression. Accordingly, the encoded cargo, RNA format, delivery tropism, innate immune set-point, and expression kinetics should be co-optimized for each clinical scenario.Manufacturing and stability control of mRNA vaccinesThe rapid development and large-scale manufacturing of mRNA-based SARS-CoV-2 vaccines represent a major advance in the translation of nucleic acid therapeutics into clinically validated platforms. The trajectory from laboratory validation to commercial-scale implementation, however, presents multifaceted technical challenges that extend beyond preliminary research achievements. Clinically viable mRNA vaccine development requires coordinated control of IVT, purification, LNP formulation, and stability management throughout manufacturing and distribution (Fig. 7). This critical scale-up transition demands systematic optimization of multiple parameters that significantly influence vaccine efficacy, including production variables, storage conditions, and distribution requirements. Defining and controlling these critical quality attributes has therefore become essential for preserving both the molecular integrity and immunological potency of mRNA vaccines across the full manufacturing and delivery cycle.340Fig. 7Full size imageFormulating IVT mRNA into LNP vaccines using a cell-free production pipeline. mRNA vaccine development includes a series of steps, including mRNA sequence design, IVT, purification, microfluidics, filtration, and storage or transportation through the cold-chain system. This figure is created with BioRender.comYield and purification optimization of mRNA in IVTThe IVT process, which relies on T7 RNA polymerase to synthesize mRNA from DNA templates, faces inherent limitations in reaction efficiency and scalability. Traditional batch methods often suffer from enzyme inhibition owing to pyrophosphate accumulation and nucleotide depletion, constraining yields. A three-phase fed-batch strategy with optimized nucleoside triphosphates (NTPs) (3–7.5 mM) and Mg²⁺ (10–38 mM) concentrations increased mRNA yield by 45% (to 367.8 μg/180 min) while reducing enzyme consumption.341 Key variables influencing IVT efficiency include Mg²⁺: NTP stoichiometry, temperature, and ionic strength. Excess Mg²⁺ stabilizes the RNA-polymerase complex but promotes nonspecific interactions, while insufficient Mg²⁺ impedes nucleotide binding. In early reaction phases, an Mg²⁺: NTP ratio of approximately 4:1 can help prevent premature termination and maintain transcriptional fidelity. Additionally, maintaining temperatures below 37 °C mitigates enzyme denaturation, and adjusting ionic strength through additives like spermidine enhances template binding. These refinements highlight the importance of dynamic parameter control during scale-up to maintain product consistency across bioreactor systems.342 Post-IVT purification is critical because residual contaminants—including DNA templates, enzymes, nucleotides (NTPs), and double-stranded RNA—can compromise product quality, provoke innate immune responses, and hinder large-scale manufacturing. Removal of these impurities is necessary to avoid immunogenicity and preserve translational fidelity.343 Chromatographic methods suitable for large-scale, high-precision purification have been developed for mRNA vaccine manufacturing. Affinity chromatography (AC) using poly(dT) resins binds mRNA via poly(A) tails, effectively separating it from DNA, enzymes, and truncated RNAs. With dynamic binding capacities of approximately 2 mg/mL resin, AC is scalable for industrial production but requires stringent wash conditions to minimize non-specific binding.344 Anion exchange chromatography (AEX) separates mRNA based on charge differences, selectively retaining dsRNA and RNA-DNA hybrids. Elution with chaotropic agents such as 6 M urea at elevated temperature can improve purity, but it also increases the risk of mRNA degradation.344,345Scalable precision manufacturing of LNPsLNPs, which typically contain ionizable lipids, phospholipids, cholesterol, and PEG-lipids, require tight compositional and process control to achieve uniform particle size (50–100 nm), high encapsulation efficiency (>90%), and acceptable storage stability.346,347 Although small-scale microfluidic mixing enables fine control over particle formation, industrial-scale production introduces additional variability through altered flow regimes and shear stress.340 For instance, deviations in mixing speed or lipid ratios can lead to particle aggregation or mRNA leakage, compromising vaccine efficacy.342 A study comparing lab-scale and pilot-scale LNP production reported a 15% reduction in encapsulation efficiency when scaling from 1 mL to 10 L batches, underscoring the need for standardized protocols.348 Recent advancements in microfluidic technology address these challenges by enabling continuous manufacturing with precise mixing control. A universal platform capable of operating at 0.1–75 mL/min flow rates facilitates both exploratory research and industrial production.342 By employing coaxial flow with triangular microstructures, the system generates consistent vortices across Reynolds numbers, ensuring uniform nanoparticle self-assembly. This technology has been validated for synthesizing mRNA-LNPs, polyplexes, and polymer nanoparticles, achieving polydispersity indices 85% at scales up to 75 mL/min. Such systems reduce batch-to-batch variability and support rapid process optimization, both of which are essential for large-scale vaccine manufacturing.Integrated stability control in pharmaceutical cold-chain systemsBecause mRNA-LNP formulations are thermolabile, they usually require storage at −20 °C to −80 °C, which creates major distribution barriers in resource-limited settings (Fig. 7).340 Lyophilization (freeze-drying) offers a solution by removing water content and stabilizing the LNPs at 4 °C for over 12 months.349 Optimizing cryoprotectant blends, such as 10% sucrose and 5% trehalose in Tris buffer, prevents nanoparticle aggregation during freezing and reconstitution. A lyophilized monkeypox mRNA vaccine retained full bioactivity after 12 months at 4 °C, with size (85 nm vs. 83 nm), polydispersity (0.12 vs. 0.10), and immunogenicity of reconstituted particles matching those of fresh formulations.350 Scaling this process to 2000 vials/batch demonstrated consistent quality attributes, validating its industrial feasibility. Temperature excursions during distribution degrade mRNA integrity and LNP structure. A kinetic model integrating real-time temperature data has been used to predict critical quality attributes (CQAs), including the mRNA degradation rate and dsRNA content.351 For example, exposure to 25 °C for 8 h reduces mRNA integrity by 20%, but pre-lyophilized formulations show