## Main

Vaccines prevent many millions of illnesses and save numerous lives every year1. As a result of widespread vaccine use, the smallpox virus has been completely eradicated and the incidence of polio, measles and other childhood diseases has been drastically reduced around the world2. Conventional vaccine approaches, such as live attenuated and inactivated pathogens and subunit vaccines, provide durable protection against a variety of dangerous diseases3. Despite this success, there remain major hurdles to vaccine development against a variety of infectious pathogens, especially those better able to evade the adaptive immune response4. Moreover, for most emerging virus vaccines, the main obstacle is not the effectiveness of conventional approaches but the need for more rapid development and large-scale deployment. Finally, conventional vaccine approaches may not be applicable to non-infectious diseases, such as cancer. The development of more potent and versatile vaccine platforms is therefore urgently needed.

Nucleic acid therapeutics have emerged as promising alternatives to conventional vaccine approaches. The first report of the successful use of in vitro transcribed (IVT) mRNA in animals was published in 1990, when reporter gene mRNAs were injected into mice and protein production was detected5. A subsequent study in 1992 demonstrated that administration of vasopressin-encoding mRNA in the hypothalamus could elicit a physiological response in rats6. However, these early promising results did not lead to substantial investment in developing mRNA therapeutics, largely owing to concerns associated with mRNA instability, high innate immunogenicity and inefficient in vivo delivery. Instead, the field pursued DNA-based and protein-based therapeutic approaches7,8.

Over the past decade, major technological innovation and research investment have enabled mRNA to become a promising therapeutic tool in the fields of vaccine development and protein replacement therapy. The use of mRNA has several beneficial features over subunit, killed and live attenuated virus, as well as DNA-based vaccines. First, safety: as mRNA is a non-infectious, non-integrating platform, there is no potential risk of infection or insertional mutagenesis. Additionally, mRNA is degraded by normal cellular processes, and its in vivo half-life can be regulated through the use of various modifications and delivery methods9,10,11,12. The inherent immunogenicity of the mRNA can be down-modulated to further increase the safety profile9,12,13. Second, efficacy: various modifications make mRNA more stable and highly translatable9,12,13. Efficient in vivo delivery can be achieved by formulating mRNA into carrier molecules, allowing rapid uptake and expression in the cytoplasm (reviewed in Refs 10,11). mRNA is the minimal genetic vector; therefore, anti-vector immunity is avoided, and mRNA vaccines can be administered repeatedly. Third, production: mRNA vaccines have the potential for rapid, inexpensive and scalable manufacturing, mainly owing to the high yields of in vitro transcription reactions.

The mRNA vaccine field is developing extremely rapidly; a large body of preclinical data has accumulated over the past several years, and multiple human clinical trials have been initiated. In this Review, we discuss current mRNA vaccine approaches, summarize the latest findings, highlight challenges and recent successes, and offer perspectives on the future of mRNA vaccines. The data suggest that mRNA vaccines have the potential to solve many of the challenges in vaccine development for both infectious diseases and cancer.

## Basic mRNA vaccine pharmacology

mRNA is the intermediate step between the translation of protein-encoding DNA and the production of proteins by ribosomes in the cytoplasm. Two major types of RNA are currently studied as vaccines: non-replicating mRNA and virally derived, self-amplifying RNA. Conventional mRNA-based vaccines encode the antigen of interest and contain 5′ and 3′ untranslated regions (UTRs), whereas self-amplifying RNAs encode not only the antigen but also the viral replication machinery that enables intracellular RNA amplification and abundant protein expression.

The construction of optimally translated IVT mRNA suitable for therapeutic use has been reviewed previously14,15. Briefly, IVT mRNA is produced from a linear DNA template using a T7, a T3 or an Sp6 phage RNA polymerase16. The resulting product should optimally contain an open reading frame that encodes the protein of interest, flanking UTRs, a 5′ cap and a poly(A) tail. The mRNA is thus engineered to resemble fully processed mature mRNA molecules as they occur naturally in the cytoplasm of eukaryotic cells.

Complexing of mRNA for in vivo delivery has also been recently detailed10,11. Naked mRNA is quickly degraded by extracellular RNases17 and is not internalized efficiently. Thus, a great variety of in vitro and in vivo transfection reagents have been developed that facilitate cellular uptake of mRNA and protect it from degradation. Once the mRNA transits to the cytosol, the cellular translation machinery produces protein that undergoes post-translational modifications, resulting in a properly folded, fully functional protein. This feature of mRNA pharmacology is particularly advantageous for vaccines and protein replacement therapies that require cytosolic or transmembrane proteins to be delivered to the correct cellular compartments for proper presentation or function. IVT mRNA is finally degraded by normal physiological processes, thus reducing the risk of metabolite toxicity.

## Recent advances in mRNA vaccine technology

Various mRNA vaccine platforms have been developed in recent years and validated in studies of immunogenicity and efficacy18,19,20. Engineering of the RNA sequence has rendered synthetic mRNA more translatable than ever before. Highly efficient and non-toxic RNA carriers have been developed that in some cases21,22 allow prolonged antigen expression in vivo (Table 1). Some vaccine formulations contain novel adjuvants, while others elicit potent responses in the absence of known adjuvants. The following section summarizes the key advances in these areas of mRNA engineering and their impact on vaccine efficacy.

### Optimization of mRNA translation and stability

This topic has been extensively discussed in previous reviews14,15; thus, we briefly summarize the key findings (Box 1). The 5′ and 3′ UTR elements flanking the coding sequence profoundly influence the stability and translation of mRNA, both of which are critical concerns for vaccines. These regulatory sequences can be derived from viral or eukaryotic genes and greatly increase the half-life and expression of therapeutic mRNAs23,24. A 5′ cap structure is required for efficient protein production from mRNA25. Various versions of 5′ caps can be added during or after the transcription reaction using a vaccinia virus capping enzyme26 or by incorporating synthetic cap or anti-reverse cap analogues27,28. The poly(A) tail also plays an important regulatory role in mRNA translation and stability25; thus, an optimal length of poly(A)24 must be added to mRNA either directly from the encoding DNA template or by using poly(A) polymerase. The codon usage additionally has an impact on protein translation. Replacing rare codons with frequently used synonymous codons that have abundant cognate tRNA in the cytosol is a common practice to increase protein production from mRNA29, although the accuracy of this model has been questioned30. Enrichment of G:C content constitutes another form of sequence optimization that has been shown to increase steady-state mRNA levels in vitro31 and protein expression in vivo12.

Although protein expression may be positively modulated by altering the codon composition or by introducing modified nucleosides (discussed below), it is also possible that these forms of sequence engineering could affect mRNA secondary structure32, the kinetics and accuracy of translation and simultaneous protein folding33,34, and the expression of cryptic T cell epitopes present in alternative reading frames30. All these factors could potentially influence the magnitude or specificity of the immune response.

### Modulation of immunogenicity

Exogenous mRNA is inherently immunostimulatory, as it is recognized by a variety of cell surface, endosomal and cytosolic innate immune receptors (Fig. 1) (reviewed in Ref. 35). Depending on the therapeutic application, this feature of mRNA could be beneficial or detrimental. It is potentially advantageous for vaccination because in some cases it may provide adjuvant activity to drive dendritic cell (DC) maturation and thus elicit robust T and B cell immune responses. However, innate immune sensing of mRNA has also been associated with the inhibition of antigen expression and may negatively affect the immune response9,13. Although the paradoxical effects of innate immune sensing on different formats of mRNA vaccines are incompletely understood, some progress has been made in recent years in elucidating these phenomena.

Studies over the past decade have shown that the immunostimulatory profile of mRNA can be shaped by the purification of IVT mRNA and the introduction of modified nucleosides as well as by complexing the mRNA with various carrier molecules9,13,36,37. Enzymatically synthesized mRNA preparations contain double-stranded RNA (dsRNA) contaminants as aberrant products of the IVT reaction13. As a mimic of viral genomes and replication intermediates, dsRNA is a potent pathogen-associated molecular pattern (PAMP) that is sensed by pattern recognition receptors in multiple cellular compartments (Fig. 1). Recognition of IVT mRNA contaminated with dsRNA results in robust type I interferon production13, which upregulates the expression and activation of protein kinase R (PKR; also known as EIF2AK2) and 2′-5′-oligoadenylate synthetase (OAS), leading to the inhibition of translation38 and the degradation of cellular mRNA and ribosomal RNA39, respectively. Karikó and colleagues13 have demonstrated that contaminating dsRNA can be efficiently removed from IVT mRNA by chromatographic methods such as reverse-phase fast protein liquid chromatography (FPLC) or high-performance liquid chromatography (HPLC). Strikingly, purification by FPLC has been shown to increase protein production from IVT mRNA by up to 1,000-fold in primary human DCs13. Thus, appropriate purification of IVT mRNA seems to be critical for maximizing protein (immunogen) production in DCs and for avoiding unwanted innate immune activation.

Besides dsRNA contaminants, single-stranded mRNA molecules are themselves a PAMP when delivered to cells exogenously. Single-stranded oligoribonucleotides and their degradative products are detected by the endosomal sensors Toll-like receptor 7 (TLR7) and TLR8 (Refs 40,41), resulting in type I interferon production42. Crucially, it was discovered that the incorporation of naturally occurring chemically modified nucleosides, including but not limited to pseudouridine9,43,44 and 1-methylpseudouridine45, prevents activation of TLR7, TLR8 and other innate immune sensors46,47, thus reducing type I interferon signalling48. Nucleoside modification also partially suppresses the recognition of dsRNA species46,47,48. As a result, Karikó and others have shown that nucleoside-modified mRNA is translated more efficiently than unmodified mRNA in vitro9, particularly in primary DCs, and in vivo in mice45. Notably, the highest level of protein production in DCs was observed when mRNA was both FPLC-purified and nucleoside-modified13. These advances in understanding the sources of innate immune sensing and how to avoid their adverse effects have substantially contributed to the current interest in mRNA-based vaccines and protein replacement therapies.

In contrast to the findings described above, a study by Thess and colleagues found that sequence-optimized, HPLC-purified, unmodified mRNA produced higher levels of protein in HeLa cells and in mice than its nucleoside-modified counterpart12. Additionally, Kauffman and co-workers demonstrated that unmodified, non-HPLC-purified mRNA yielded more robust protein production in HeLa cells than nucleoside-modified mRNA, and resulted in similar levels of protein production in mice49. Although not fully clear, the discrepancies between the findings of Karikó9,13 and these authors12,49 may have arisen from variations in RNA sequence optimization, the stringency of mRNA purification to remove dsRNA contaminants and the level of innate immune sensing in the targeted cell types.

### Progress in mRNA vaccine delivery

Efficient in vivo mRNA delivery is critical to achieving therapeutic relevance. Exogenous mRNA must penetrate the barrier of the lipid membrane in order to reach the cytoplasm to be translated to functional protein. mRNA uptake mechanisms seem to be cell type dependent, and the physicochemical properties of the mRNA complexes can profoundly influence cellular delivery and organ distribution. There are two basic approaches for the delivery of mRNA vaccines that have been described to date. First, loading of mRNA into DCs ex vivo, followed by re-infusion of the transfected cells58; and second, direct parenteral injection of mRNA with or without a carrier. Ex vivo DC loading allows precise control of the cellular target, transfection efficiency and other cellular conditions, but as a form of cell therapy, it is an expensive and labour-intensive approach to vaccination. Direct injection of mRNA is comparatively rapid and cost-effective, but it does not yet allow precise and efficient cell-type-specific delivery, although there has been recent progress in this regard59. Both of these approaches have been explored in a variety of forms (Fig. 2; Table 1).

Ex vivo loading of DCs. DCs are the most potent antigen-presenting cells of the immune system. They initiate the adaptive immune response by internalizing and proteolytically processing antigens and presenting them to CD8+ and CD4+ T cells on major histocompatibility complexes (MHCs), namely, MHC class I and MHC class II, respectively. Additionally, DCs may present intact antigen to B cells to provoke an antibody response60. DCs are also highly amenable to mRNA transfection. For these reasons, DCs represent an attractive target for transfection by mRNA vaccines, both in vivo and ex vivo.

Although DCs have been shown to internalize naked mRNA through a variety of endocytic pathways61,62,63, ex vivo transfection efficiency is commonly increased using electroporation; in this case, mRNA molecules pass through membrane pores formed by a high-voltage pulse and directly enter the cytoplasm (reviewed in Ref. 64). This mRNA delivery approach has been favoured for its ability to generate high transfection efficiency without the need for a carrier molecule. DCs that are loaded with mRNA ex vivo are then re-infused into the autologous vaccine recipient to initiate the immune response. Most ex vivo-loaded DC vaccines elicit a predominantly cell-mediated immune response; thus, they have been used primarily to treat cancer (reviewed in Ref. 58).

Injection of naked mRNA in vivo. Naked mRNA has been used successfully for in vivo immunizations, particularly in formats that preferentially target antigen-presenting cells, as in intradermal61,65 and intranodal injections66,67,68. Notably, a recent report showed that repeated intranodal immunizations with naked, unmodified mRNA encoding tumour-associated neoantigens generated robust T cell responses and increased progression-free survival68 (discussed further in Box 2).

Physical delivery methods in vivo. To increase the efficiency of mRNA uptake in vivo, physical methods have occasionally been used to penetrate the cell membrane. An early report showed that mRNA complexed with gold particles could be expressed in tissues using a gene gun, a microprojectile method69. The gene gun was shown to be an efficient RNA delivery and vaccination method in mouse models70,71,72,73, but no efficacy data in large animals or humans are available. In vivo electroporation has also been used to increase uptake of therapeutic RNA74,75,76; however, in one study, electroporation increased the immunogenicity of only a self-amplifying RNA and not a non-replicating mRNA-based vaccine74. Physical methods can be limited by increased cell death and restricted access to target cells or tissues. Recently, the field has instead favoured the use of lipid or polymer-based nanoparticles as potent and versatile delivery vehicles.

Protamine. The cationic peptide protamine has been shown to protect mRNA from degradation by serum RNases77; however, protamine-complexed mRNA alone demonstrated limited protein expression and efficacy in a cancer vaccine model, possibly owing to an overly tight association between protamine and mRNA36,78. This issue was resolved by developing the RNActive vaccine platform, in which protamine-formulated RNA serves only as an immune activator and not as an expression vector52.

Cationic lipid and polymer-based delivery. Highly efficient mRNA transfection reagents based on cationic lipids or polymers, such as TransIT-mRNA (Mirus Bio LLC) or Lipofectamine (Invitrogen), are commercially available and work well in many primary cells and cancer cell lines9,13, but they often show limited in vivo efficacy or a high level of toxicity (N.P. and D.W., unpublished observations). Great progress has been made in developing similarly designed complexing reagents for safe and effective in vivo use, and these are discussed in detail in several recent reviews10,11,79,80. Cationic lipids and polymers, including dendrimers, have become widely used tools for mRNA administration in the past few years. The mRNA field has clearly benefited from the substantial investment in in vivo small interfering RNA (siRNA) administration, where these delivery vehicles have been used for over a decade. Lipid nanoparticles (LNPs) have become one of the most appealing and commonly used mRNA delivery tools. LNPs often consist of four components: an ionizable cationic lipid, which promotes self-assembly into virus-sized (~100 nm) particles and allows endosomal release of mRNA to the cytoplasm; lipid-linked polyethylene glycol (PEG), which increases the half-life of formulations; cholesterol, a stabilizing agent; and naturally occurring phospholipids, which support lipid bilayer structure. Numerous studies have demonstrated efficient in vivo siRNA delivery by LNPs (reviewed in Ref. 81), but it has only recently been shown that LNPs are potent tools for in vivo delivery of self-amplifying RNA19 and conventional, non-replicating mRNA21. Systemically delivered mRNA–LNP complexes mainly target the liver owing to binding of apolipoprotein E and subsequent receptor-mediated uptake by hepatocytes82, and intradermal, intramuscular and subcutaneous administration have been shown to produce prolonged protein expression at the site of the injection21,22. The mechanisms of mRNA escape into the cytoplasm are incompletely understood, not only for artificial liposomes but also for naturally occurring exosomes83. Further research into this area will likely be of great benefit to the field of therapeutic RNA delivery.

The magnitude and duration of in vivo protein production from mRNA–LNP vaccines can be controlled in part by varying the route of administration. Intramuscular and intradermal delivery of mRNA–LNPs has been shown to result in more persistent protein expression than systemic delivery routes: in one experiment, the half-life of mRNA-encoded firefly luciferase was roughly threefold longer after intradermal injection than after intravenous delivery21. These kinetics of mRNA–LNP expression may be favourable for inducing immune responses. A recent study demonstrated that sustained antigen availability during vaccination was a driver of high antibody titres and germinal centre (GC) B cell and T follicular helper (TFH) cell responses84. This process was potentially a contributing factor to the potency of recently described nucleoside-modified mRNA–LNP vaccines delivered by the intramuscular and intradermal routes20,22,85. Indeed, TFH cells have been identified as a critical population of immune cells that vaccines must activate in order to generate potent and long-lived neutralizing antibody responses, particularly against viruses that evade humoral immunity86. The dynamics of the GC reaction and the differentiation of TFH cells are incompletely understood, and progress in these areas would undoubtedly be fruitful for future vaccine design (Box 3).

## mRNA vaccines against infectious diseases

Development of prophylactic or therapeutic vaccines against infectious pathogens is the most efficient means to contain and prevent epidemics. However, conventional vaccine approaches have largely failed to produce effective vaccines against challenging viruses that cause chronic or repeated infections, such as HIV-1, herpes simplex virus and respiratory syncytial virus (RSV). Additionally, the slow pace of commercial vaccine development and approval is inadequate to respond to the rapid emergence of acute viral diseases, as illustrated by the 2014–2016 outbreaks of the Ebola and Zika viruses. Therefore, the development of more potent and versatile vaccine platforms is crucial.

Preclinical studies have created hope that mRNA vaccines will fulfil many aspects of an ideal clinical vaccine: they have shown a favourable safety profile in animals, are versatile and rapid to design for emerging infectious diseases, and are amenable to scalable good manufacturing practice (GMP) production (already under way by several companies). Unlike protein immunization, several formats of mRNA vaccines induce strong CD8+ T cell responses, likely owing to the efficient presentation of endogenously produced antigens on MHC class I molecules, in addition to potent CD4+ T cell responses56,87,88. Additionally, unlike DNA immunization, mRNA vaccines have shown the ability to generate potent neutralizing antibody responses in animals with only one or two low-dose immunizations20,22,85. As a result, mRNA vaccines have elicited protective immunity against a variety of infectious agents in animal models19,20,22,56,89,90 and have therefore generated substantial optimism. However, recently published results from two clinical trials of mRNA vaccines for infectious diseases were somewhat modest, leading to more cautious expectations about the translation of preclinical success to the clinic22,91 (discussed further below).

Two major types of RNA vaccine have been utilized against infectious pathogens: self-amplifying or replicon RNA vaccines and non-replicating mRNA vaccines. Non-replicating mRNA vaccines can be further distinguished by their delivery method: ex vivo loading of DCs or direct in vivo injection into a variety of anatomical sites. As discussed below, a rapidly increasing number of preclinical studies in these areas have been published recently, and several have entered human clinical trials (Table 2).

### Self-amplifying mRNA vaccines

Most currently used self-amplifying mRNA (SAM) vaccines are based on an alphavirus genome92, where the genes encoding the RNA replication machinery are intact but the genes encoding the structural proteins are replaced with the antigen of interest. The full-length RNA is ~9 kb long and can be easily produced by IVT from a DNA template. The SAM platform enables a large amount of antigen production from an extremely small dose of vaccine owing to intracellular replication of the antigen-encoding RNA. An early study reported that immunization with 10 μg of naked SAM vaccine encoding RSV fusion (F), influenza virus haemagglutinin (HA) or louping ill virus pre-membrane and envelope (prM-E) proteins resulted in antibody responses and partial protection from lethal viral challenges in mice93. The development of RNA complexing agents brought remarkable improvement to the efficacy of SAM vaccines. As little as 100 ng of an RNA replicon vaccine encoding RSV F, complexed to LNP, resulted in potent T and B cell immune responses in mice, and 1 μg elicited protective immune responses against RSV infection in a cotton rat intranasal challenge system19. SAM vaccines encoding influenza virus antigens in LNPs or an oil-in-water cationic nanoemulsion induced potent immune responses in ferrets and conferred protection from homologous and heterologous viral challenge in mice94,95,96. Further studies demonstrated the immunogenicity of this vaccine platform against diverse viruses in multiple species, including human cytomegalovirus (CMV), hepatitis C virus and rabies virus in mice, HIV-1 in rabbits, and HIV-1 and human CMV in rhesus macaques50,87,97. Replicon RNA encoding influenza antigens, complexed with chitosan-containing LNPs or polyethylenimine (PEI), has elicited T and B cell immune responses in mice after subcutaneous delivery98,99. Chahal and colleagues developed a delivery platform consisting of a chemically modified, ionizable dendrimer complexed into LNPs89. Using this platform, they demonstrated that intramuscular delivery of RNA replicons encoding influenza virus, Ebola virus or Toxoplasma gondii antigens protected mice against lethal infection89. The same group recently demonstrated that vaccination with an RNA replicon encoding Zika virus prM-E formulated in the same manner elicited antigen-specific antibody and CD8+ T cell responses in mice88. Another recent study reported immunogenicity and moderate protective efficacy of SAM vaccines against bacterial pathogens, namely Streptococcus (groups A and B) spp., further demonstrating the versatility of this platform100.

One of the advantages of SAM vaccines is that they create their own adjuvants in the form of dsRNA structures, replication intermediates and other motifs that may contribute to their high potency. However, the intrinsic nature of these PAMPs may make it difficult to modulate the inflammatory profile or reactogenicity of SAM vaccines. Additionally, size constraints of the insert are greater for SAM vaccines than for mRNAs that do not encode replicon genes, and the immunogenicity of the replication proteins may theoretically limit repeated use.

### Dendritic cell mRNA vaccines

As described above, ex vivo DC loading is a heavily pursued method to generate cell-mediated immunity against cancer. Development of infectious disease vaccines using this approach has been mainly limited to a therapeutic vaccine for HIV-1: HIV-1-infected individuals on highly active antiretroviral therapy were treated with autologous DCs electroporated with mRNA encoding various HIV-1 antigens, and cellular immune responses were evaluated101,102,103,104,105,106. This intervention proved to be safe and elicited antigen-specific CD4+ and CD8+ T cell responses, but no clinical benefit was observed. Another study in humans evaluated a CMV pp65 mRNA-loaded DC vaccination in healthy human volunteers and allogeneic stem cell recipients and reported induction or expansion of CMV-specific cellular immune responses107.

### Direct injection of non-replicating mRNA vaccines

Directly injectable, non-replicating mRNA vaccines are an appealing vaccine format owing to their simple and economical administration, particularly in resource-limited settings. Although an early report demonstrated that immunization with liposome-complexed mRNA encoding influenza virus nucleoproteins elicited CTL responses in mice108, the first demonstration of protective immune responses by mRNA vaccines against infectious pathogens was published only a few years ago18. This seminal work demonstrated that intradermally administered uncomplexed mRNA encoding various influenza virus antigens combined with a protamine-complexed RNA adjuvant was immunogenic in multiple animal models and protected mice from lethal viral challenge.

Immunization with the protamine-based RNActive platform encoding rabies virus glycoprotein has also induced protective immunity against a lethal intracerebral virus challenge in mice and potent neutralizing antibody responses in pigs56. In a recently published seminal work, Alberer and colleagues evaluated the safety and immunogenicity of this vaccine in 101 healthy human volunteers91. Subjects received 80–640 μg of mRNA vaccine three times by needle-syringe or needle-free devices, either intradermally or intramuscularly. Seven days after vaccination, nearly all participants reported mild to moderate injection site reactions, and 78% experienced a systemic reaction (for example, fever, headache and chills). There was one serious adverse event that was possibly related to the vaccine: a transient and moderate case of Bell palsy. Surprisingly, the needle-syringe injections did not generate detectable neutralizing antibodies in 98% of recipients. By contrast, needle-free delivery induced variable levels of neutralizing antibodies, the majority of which peaked above the expected protective threshold but then largely waned after 1 year in subjects who were followed up long term. Elucidating the basis of the disparate immunogenicity between the animals and humans who received this vaccine and between the two routes of delivery will be informative for future vaccine design using this platform.

Other infectious disease vaccines have successfully utilized lipid- or polymer-based delivery systems. Cationic 1,2-dioleoyloxy-3-trimethylammoniumpropane (DOTAP) and dioleoylphosphatidylethanolamine (DOPE) lipid-complexed mRNA encoding HIV-1 gag generated antigen-specific CD4+ and CD8+ T cell responses after subcutaneous delivery in mice109. Two other studies demonstrated that PEI-complexed mRNAs could be efficiently delivered to mice to induce HIV-1-specific immune responses: subcutaneously delivered mRNA encoding HIV-1 gag elicited CD4+ and CD8+ T cell responses, and intranasally administered mRNA encoding the HIV-1 envelope gp120 subunit crossed the nasal epithelium and generated antigen-specific immune responses in the nasal cavity110,111. Kranz and colleagues also performed intravenous immunizations in mice using lipid-complexed mRNA encoding influenza virus HA and showed evidence of T cell activation after a single dose59.

Nucleoside-modified mRNA vaccines represent a new and highly efficacious category of mRNA vaccines. Owing to the novelty of this immunization platform, our knowledge of efficacy is limited to the results of four recent publications that demonstrated the potency of such vaccines in small and large animals. The first published report demonstrated that a single intradermal injection of LNP-formulated mRNA encoding Zika virus prM-E, modified with 1-methylpseudouridine and FPLC purification, elicited protective immune responses in mice and rhesus macaques with the use of as little as 50 μg (0.02 mg kg−1) of vaccine in macaques20. A subsequent study by a different group tested a similarly designed vaccine against Zika virus in mice and found that a single intramuscular immunization elicited moderate immune responses, and a booster vaccination resulted in potent and protective immune responses85. This vaccine also incorporated the modified nucleoside 1-methylpseudouridine, but FPLC purification or other methods of removing dsRNA contaminants were not reported. Notably, this report showed that antibody-dependent enhancement of secondary infection with a heterologous flavivirus, a major concern for dengue and Zika virus vaccines, could be diminished by removing a cross-reactive epitope in the E protein. A recent follow-up study evaluated the same vaccine in a model of maternal vaccination and fetal infection112. Two immunizations reduced Zika virus infection in fetal mice by several orders of magnitude and completely rescued a defect in fetal viability.

Another recent report evaluated the immunogenicity of LNP-complexed, nucleoside-modified, non-FPLC-purified mRNA vaccines against influenza HA 10 neuraminidase 8 (H10N8) and H7N9 influenza viruses in mice, ferrets, non-human primates and, for the first time, humans22. A single intradermal or intramuscular immunization with low doses (0.4–10 μg) of LNP-complexed mRNA encoding influenza virus HA elicited protective immune responses against homologous influenza virus challenge in mice. Similar results were obtained in ferrets and cynomolgus monkeys after immunization with one or two doses of 50–400 μg of a vaccine containing LNP-complexed mRNA encoding HA, corroborating that the potency of mRNA–LNP vaccines translates to larger animals, including non-human primates.

On the basis of encouraging preclinical data, two phase I clinical trials have recently been initiated to evaluate the immunogenicity and safety of nucleoside-modified mRNA–LNP vaccines in humans for the first time. The mRNA vaccine encoding H10N8 HA is currently undergoing clinical testing (NCT03076385), and interim findings for 23 vaccinated individuals have been reported22. Participants received a small amount (100 μg) of vaccine intramuscularly, and immunogenicity was measured 43 days after vaccination. The vaccine proved to be immunogenic in all subjects, as measured by haemagglutination inhibition and microneutralization antibody assays. Promisingly, antibody titres were above the expected protective threshold, but they were moderately lower than in the animal models. Similarly to the study by Alberer et al.91, most vaccinated subjects reported mild to moderate reactogenicity (injection site pain, myalgia, headache, fatigue and chills), and three subjects reported severe injection site reactions or a systemic common cold-like response. This level of reactogenicity appears to be similar to that of more traditional vaccine formats113,114. Finally, the Zika virus vaccine described by Richner et al.85,112 is also entering clinical evaluation in a combined phase I/II trial (NCT03014089). Future studies that apply nucleoside-modified mRNA–LNP vaccines against a greater diversity of antigens will reveal the extent to which this strategy is broadly applicable to infectious disease vaccines.

## mRNA cancer vaccines

mRNA-based cancer vaccines have been recently and extensively reviewed115,116,117,118,119. Below, the most recent advances and directions are highlighted. Cancer vaccines and other immunotherapies represent promising alternative strategies to treat malignancies. Cancer vaccines can be designed to target tumour-associated antigens that are preferentially expressed in cancerous cells, for example, growth-associated factors, or antigens that are unique to malignant cells owing to somatic mutation120. These neoantigens, or the neoepitopes within them, have been deployed as mRNA vaccine targets in humans121 (Box 2). Most cancer vaccines are therapeutic, rather than prophylactic, and seek to stimulate cell-mediated responses, such as those from CTLs, that are capable of clearing or reducing tumour burden122. The first proof-of-concept studies that not only proposed the idea of RNA cancer vaccines but also provided evidence of the feasibility of this approach were published more than two decades ago123,124. Since then, numerous preclinical and clinical studies have demonstrated the viability of mRNA vaccines to combat cancer (Table 3).

### DC mRNA cancer vaccines

As DCs are central players in initiating antigen-specific immune responses, it seemed logical to utilize them for cancer immunotherapy. The first demonstration that DCs electroporated with mRNA could elicit potent immune responses against tumour antigens was reported by Boczkowski and colleagues in 1996 (Ref. 124). In this study, DCs pulsed with ovalbumin (OVA)-encoding mRNA or tumour-derived RNAs elicited a tumour-reducing immune response in OVA-expressing and other melanoma models in mice. A variety of immune regulatory proteins have been identified in the form of mRNA-encoded adjuvants that can increase the potency of DC cancer vaccines. Several studies demonstrated that electroporation of DCs with mRNAs encoding co-stimulatory molecules such as CD83, tumour necrosis factor receptor superfamily member 4 (TNFRSF4; also known as OX40) and 4-1BB ligand (4-1BBL) resulted in a substantial increase in the immune stimulatory activity of DCs125,126,127,128. DC functions can also be modulated through the use of mRNA-encoded pro-inflammatory cytokines, such as IL-12, or trafficking-associated molecules129,130,131. As introduced above, TriMix is a cocktail of mRNA-encoded adjuvants (CD70, CD40L and constitutively active TLR4) that can be electroporated in combination with antigen-encoding mRNA or mRNAs132. This formulation proved efficacious in multiple preclinical studies by increasing DC activation and shifting the CD4+ T cell phenotype from T regulatory cells to T helper 1 (TH1)-like cells132,133,134,135,136. Notably, the immunization of patients with stage III or stage IV melanoma using DCs loaded with mRNA encoding melanoma-associated antigens and TriMix adjuvant resulted in tumour regression in 27% of treated individuals137. Multiple clinical trials have now been conducted using DC vaccines targeting various cancer types, such as metastatic prostate cancer, metastatic lung cancer, renal cell carcinoma, brain cancers, melanoma, acute myeloid leukaemia, pancreatic cancer and others138,139 (reviewed in Refs 51,58).

A new line of research combines mRNA electroporation of DCs with traditional chemotherapy agents or immune checkpoint inhibitors. In one trial, patients with stage III or IV melanoma were treated with ipilimumab, a monoclonal antibody against CTL antigen 4 (CTLA4), and DCs loaded with mRNA encoding melanoma-associated antigens plus TriMix. This intervention resulted in durable tumour reduction in a proportion of individuals with recurrent or refractory melanoma140.

### Direct injection of mRNA cancer vaccines

The route of administration and delivery format of mRNA vaccines can greatly influence outcomes. A variety of mRNA cancer vaccine formats have been developed using common delivery routes (intradermal, intramuscular, subcutaneous or intranasal) and some unconventional routes of vaccination (intranodal, intravenous, intrasplenic or intratumoural).

Intranodal administration of naked mRNA is an unconventional but efficient means of vaccine delivery. Direct mRNA injection into secondary lymphoid tissue offers the advantage of targeted antigen delivery to antigen-presenting cells at the site of T cell activation, obviating the need for DC migration. Several studies have demonstrated that intranodally injected naked mRNA can be selectively taken up by DCs and can elicit potent prophylactic or therapeutic antitumour T cell responses62,66; an early study also demonstrated similar findings with intrasplenic delivery141. Coadministration of the DC-activating protein FMS-related tyrosine kinase 3 ligand (FLT3L) was shown in some cases to further improve immune responses to intranodal mRNA vaccination142,143. Incorporation of the TriMix adjuvant into intranodal injections of mice with mRNAs encoding tumour-associated antigens resulted in potent antigen-specific CTL responses and tumour control in multiple tumour models133. A more recent study demonstrated that intranodal injection of mRNA encoding the E7 protein of human papillomavirus (HPV) 16 with TriMix increased the number of tumour-infiltrating CD8+ T cells and inhibited the growth of an E7-expressing tumour model in mice67.

The success of preclinical studies has led to the initiation of clinical trials using intranodally injected naked mRNA encoding tumour-associated antigens into patients with advanced melanoma (NCT01684241) and patients with hepatocellular carcinoma (EudraCT: 2012-005572-34). In one published trial, patients with metastatic melanoma were treated with intranodally administered DCs electroporated with mRNA encoding the melanoma-associated antigens tyrosinase or gp100 and TriMix, which induced limited antitumour responses144.

Intranasal vaccine administration is a needle-free, noninvasive manner of delivery that enables rapid antigen uptake by DCs. Intranasally delivered mRNA complexed with Stemfect (Stemgent) LNPs resulted in delayed tumour onset and increased survival in prophylactic and therapeutic mouse tumour models using the OVA-expressing E.G7-OVA T lymphoblastic cell line145.

Intratumoural mRNA vaccination is a useful approach that offers the advantage of rapid and specific activation of tumour-resident T cells. Often, these vaccines do not introduce mRNAs encoding tumour-associated antigens but simply aim to activate tumour-specific immunity in situ using immune stimulatory molecules. An early study demonstrated that naked mRNA or protamine-stabilized mRNA encoding a non-tumour related gene (GLB1) impaired tumour growth and provided protection in a glioblastoma mouse model, taking advantage of the intrinsic immunogenic properties of mRNA146. A more recent study showed that intratumoural delivery of mRNA encoding an engineered cytokine based on interferon-β (IFNβ) fused to a transforming growth factor-β (TGFβ) antagonist increased the cytolytic capacity of CD8+ T cells and modestly delayed tumour growth in OVA-expressing lymphoma or lung carcinoma mouse models147. It has also been shown that intratumoural administration of TriMix mRNA that does not encode tumour-associated antigens results in activation of CD8α+ DCs and tumour-specific T cells, leading to delayed tumour growth in various mouse models148.

Systemic administration of mRNA vaccines is not common owing to concerns about aggregation with serum proteins and rapid extracellular mRNA degradation; thus, formulating mRNAs into carrier molecules is essential. As discussed above, numerous delivery formulations have been developed to facilitate mRNA uptake, increase protein translation and protect mRNA from RNases10,11,79,80. Another important issue is the biodistribution of mRNA vaccines after systemic delivery. Certain cationic LNP-based complexing agents delivered intravenously traffic mainly to the liver21, which may not be ideal for DC activation. An effective strategy for DC targeting of mRNA vaccines after systemic delivery has recently been described59. An mRNA–lipoplex (mRNA–liposome complex) delivery platform was generated using cationic lipids and neutral helper lipids formulated with mRNA, and it was discovered that the lipid-to-mRNA ratio, and thus the net charge of the particles, has a profound impact on the biodistribution of the vaccine. While a positively charged lipid particle primarily targeted the lung, a negatively charged particle targeted DCs in secondary lymphoid tissues and bone marrow. The negatively charged particle induced potent immune responses against tumour-specific antigens that were associated with impressive tumour reduction in various mouse models59. As no toxic effects were observed in mice or non-human primates, clinical trials using this approach to treat patients with advanced melanoma or triple-negative breast cancer have been initiated (NCT02410733 and NCT02316457).

A variety of antigen-presenting cells reside in the skin149, making it an ideal site for immunogen delivery during vaccination (Fig. 3). Thus, the intradermal route of delivery has been widely used for mRNA cancer vaccines. An early seminal study demonstrated that intradermal administration of total tumour RNA delayed tumour growth in a fibrosarcoma mouse model65. Intradermal injection of mRNA encoding tumour antigens in the protamine-based RNActive platform proved efficacious in various mouse models of cancer36 and in multiple prophylactic and therapeutic clinical settings (Table 3). One such study demonstrated that mRNAs encoding survivin and various melanoma tumour antigens resulted in increased numbers of antigen-specific T cells in a subset of patients with melanoma150. In humans with castration-resistant prostate cancer, an RNActive vaccine expressing multiple prostate cancer-associated proteins elicited antigen-specific T cell responses in the majority of recipients151. Lipid-based carriers have also contributed to the efficacy of intradermally delivered mRNA cancer vaccines. The delivery of OVA-encoding mRNA in DOTAP and/or DOPE liposomes resulted in antigen-specific CTL activity and inhibited growth of OVA-expressing tumours in mice152. In the same study, coadministration of mRNA encoding granulocyte–macrophage colony-stimulating factor (GM-CSF) improved OVA-specific cytolytic responses. Another report showed that subcutaneous delivery of LNP-formulated mRNA encoding two melanoma-associated antigens delayed tumour growth in mice, and co-delivery of lipopolysaccharide (LPS) in LNPs increased both CTL and antitumour activity153. In general, mRNA cancer vaccines have proved immunogenic in humans, but further refinement of vaccination methods, as informed by basic immunological research, will likely be necessary to achieve greater clinical benefits.

The combination of mRNA vaccination with adjunctive therapies, such as traditional chemotherapy, radiotherapy and immune checkpoint inhibitors, has increased the beneficial outcome of vaccination in some preclinical studies154,155. For example, cisplatin treatment significantly increased the therapeutic effect of immunizing with mRNA encoding the HPV16 E7 oncoprotein and TriMix, leading to the complete rejection of female genital tract tumours in a mouse model67. Notably, it has also been suggested that treatment with antibodies against programmed cell death protein 1 (PD1) increased the efficacy of a neoepitope mRNA-based vaccine against metastatic melanoma in humans, but more data are required to explore this hypothesis68.

## Therapeutic considerations and challenges

### Good manufacturing practice production

mRNA is produced by in vitro reactions with recombinant enzymes, ribonucleotide triphosphates (NTPs) and a DNA template; thus, it is rapid and relatively simple to produce in comparison with traditional protein subunit and live or inactivated virus vaccine production platforms. Its reaction yield and simplicity make rapid mRNA production possible in a small GMP facility footprint. The manufacturing process is sequence-independent and is primarily dictated by the length of the RNA, the nucleotide and capping chemistry and the purification of the product; however, it is possible that certain sequence properties such as extreme length may present difficulties (D.W., unpublished observations). According to current experience, the process can be standardized to produce nearly any encoded protein immunogen, making it particularly suitable for rapid response to emerging infectious diseases.

All enzymes and reaction components required for the GMP production of mRNA can be obtained from commercial suppliers as synthesized chemicals or bacterially expressed, animal component-free reagents, thereby avoiding safety concerns surrounding the adventitious agents that plague cell-culture-based vaccine manufacture. All the components, such as plasmid DNA, phage polymerases, capping enzymes and NTPs, are readily available as GMP-grade traceable components; however, some of these are currently available at only limited scale or high cost. As mRNA therapeutics move towards commercialization and the scale of production increases, more economical options may become accessible for GMP source materials.

GMP production of mRNA begins with DNA template production followed by enzymatic IVT and follows the same multistep protocol that is used for research scale synthesis, with added controls to ensure the safety and potency of the product16. Depending on the specific mRNA construct and chemistry, the protocol may be modified slightly from what is described here to accommodate modified nucleosides, capping strategies or template removal. To initiate the production process, template plasmid DNA produced in Escherichia coli is linearized using a restriction enzyme to allow synthesis of runoff transcripts with a poly(A) tract at the 3′ end. Next, the mRNA is synthesized from NTPs by a DNA-dependent RNA polymerase from bacteriophage (such as T7, SP6, or T3). The template DNA is then degraded by incubation with DNase. Finally, the mRNA is enzymatically or chemically capped to enable efficient translation in vivo. mRNA synthesis is highly productive, yielding in excess of 2 g l−1 of full-length mRNA in multi-gram scale reactions under optimized conditions.

Once the mRNA is synthesized, it is processed though several purification steps to remove reaction components, including enzymes, free nucleotides, residual DNA and truncated RNA fragments. While LiCl precipitation is routinely used for laboratory-scale preparation, purification at the clinical scale utilizes derivatized microbeads in batch or column formats, which are easier to utilize at large scale156,157. For some mRNA platforms, removal of dsRNA and other contaminants is critical for the potency of the final product, as it is a potent inducer of interferon-dependent translation inhibition. This has been accomplished by reverse-phase FPLC at the laboratory scale158, and scalable aqueous purification approaches are being investigated. After mRNA is purified, it is exchanged into a final storage buffer and sterile-filtered for subsequent filling into vials for clinical use. RNA is susceptible to degradation by both enzymatic and chemical pathways157. Formulation buffers are tested to ensure that they are free of contaminating RNases and may contain buffer components, such as antioxidants and chelators, which minimize the effects of reactive oxygen species and divalent metal ions that lead to mRNA instability159.

Pharmaceutical formulation of mRNAs is an active area of development. Although most products for early phase studies are stored frozen (−70 °C), efforts to develop formulations that are stable at higher temperatures more suitable for vaccine distribution are continuing. Published reports suggest that stable refrigerated or room temperature formulations can be made. The RNActive platform was reported to be active after lyophilization and storage at 5–25 °C for 3 years and at 40 °C for 6 months91. Another report demonstrated that freeze-dried naked mRNA is stable for at least 10 months under refrigerated conditions160. The stability of mRNA products might also be improved by packaging within nanoparticles or by co-formulation with RNase inhibitors161. For lipid-encapsulated mRNA, at least 6 months of stability has been observed (Arbutus Biopharma, personal communication), but longer-term storage of such mRNA–lipid complexes in an unfrozen form has not yet been reported.

### Regulatory aspects

There is no specific guidance from the FDA or European Medicines Agency (EMA) for mRNA vaccine products. However, the increasing number of clinical trials conducted under EMA and FDA oversight indicate that regulators have accepted the approaches proposed by various organizations to demonstrate that products are safe and acceptable for testing in humans. Because mRNA falls into the broad vaccine category of genetic immunogens, many of the guiding principles that have been defined for DNA vaccines162 and gene therapy vectors163,164 can likely be applied to mRNA with some adaptations to reflect the unique features of mRNA. A detailed review of EMA regulations for RNA vaccines by Hinz and colleagues highlights the different regulatory paths stipulated for prophylactic infectious disease versus therapeutic applications165. Regardless of the specific classification within existing guidelines, some themes can be observed in what is stated in these guidance documents and in what has been reported for recently published clinical studies. In particular, the recent report of an mRNA vaccine against influenza virus highlights preclinical and clinical data demonstrating biodistribution and persistence in mice, disease protection in a relevant animal model (ferrets), and immunogenicity, local reactogenicity and toxicity in humans22. As mRNA products become more prominent in the vaccine field, it is likely that specific guidance will be developed that will delineate requirements to produce and evaluate new mRNA vaccines.

### Safety

The requirement for safety in modern prophylactic vaccines is extremely stringent because the vaccines are administered to healthy individuals. Because the manufacturing process for mRNA does not require toxic chemicals or cell cultures that could be contaminated with adventitious viruses, mRNA production avoids the common risks associated with other vaccine platforms, including live virus, viral vectors, inactivated virus and subunit protein vaccines. Furthermore, the short manufacturing time for mRNA presents few opportunities to introduce contaminating microorganisms. In vaccinated people, the theoretical risks of infection or integration of the vector into host cell DNA are not a concern for mRNA. For the above reasons, mRNA vaccines have been considered a relatively safe vaccine format.

Several different mRNA vaccines have now been tested from phase I to IIb clinical studies and have been shown to be safe and reasonably well tolerated (Tables 2, 3). However, recent human trials have demonstrated moderate and in rare cases severe injection site or systemic reactions for different mRNA platforms22,91. Potential safety concerns that are likely to be evaluated in future preclinical and clinical studies include local and systemic inflammation, the biodistribution and persistence of expressed immunogen, stimulation of auto-reactive antibodies and potential toxic effects of any non-native nucleotides and delivery system components. A possible concern could be that some mRNA-based vaccine platforms54,166 induce potent type I interferon responses, which have been associated not only with inflammation but also potentially with autoimmunity167,168. Thus, identification of individuals at an increased risk of autoimmune reactions before mRNA vaccination may allow reasonable precautions to be taken. Another potential safety issue could derive from the presence of extracellular RNA during mRNA vaccination. Extracellular naked RNA has been shown to increase the permeability of tightly packed endothelial cells and may thus contribute to oedema169. Another study showed that extracellular RNA promoted blood coagulation and pathological thrombus formation170. Safety will therefore need continued evaluation as different mRNA modalities and delivery systems are utilized for the first time in humans and are tested in larger patient populations.

## Conclusions and future directions

Currently, mRNA vaccines are experiencing a burst in basic and clinical research. The past 2 years alone have witnessed the publication of dozens of preclinical and clinical reports showing the efficacy of these platforms. Whereas the majority of early work in mRNA vaccines focused on cancer applications, a number of recent reports have demonstrated the potency and versatility of mRNA to protect against a wide variety of infectious pathogens, including influenza virus, Ebola virus, Zika virus, Streptococcus spp. and T. gondii (Tables 1,2).

While preclinical studies have generated great optimism about the prospects and advantages of mRNA-based vaccines, two recent clinical reports have led to more tempered expectations22,91. In both trials, immunogenicity was more modest in humans than was expected based on animal models, a phenomenon also observed with DNA-based vaccines171, and the side effects were not trivial. We caution that these trials represent only two variations of mRNA vaccine platforms, and there may be substantial differences when the expression and immunostimulatory profiles of the vaccine are changed. Further research is needed to determine how different animal species respond to mRNA vaccine components and inflammatory signals and which pathways of immune signalling are most effective in humans.

Recent advances in understanding and reducing the innate immune sensing of mRNA have aided efforts not only in active vaccination but also in several applications of passive immunization or passive immunotherapy for infectious diseases and cancer (Box 4). Direct comparisons between mRNA expression platforms should clarify which systems are most appropriate for both passive and active immunization. Given the large number of promising mRNA platforms, further head-to-head comparisons would be of utmost value to the vaccine field because this would allow investigators to focus resources on those best suited for each application.

The fast pace of progress in mRNA vaccines would not have been possible without major recent advances in the areas of innate immune sensing of RNA and in vivo delivery methods. Extensive basic research into RNA and lipid and polymer biochemistry has made it possible to translate mRNA vaccines into clinical trials and has led to an astonishing level of investment in mRNA vaccine companies (Table 4). Moderna Therapeutics, founded in 2010, has raised almost US$2 billion in capital with a plan to commercialize mRNA-based vaccines and therapies172,173. The US Biomedical Advanced Research and Development Authority (BARDA) has committed support for Moderna's clinical evaluation of a promising nucleoside-modified mRNA vaccine for Zika virus (NCT03014089). In Germany, CureVac AG has an expanding portfolio of therapeutic targets174, including both cancer and infectious diseases, and BioNTech is developing an innovative approach to personalized cancer medicine using mRNA vaccines121 (Box 2). The translation of basic research into clinical testing is also made more expedient by the commercialization of custom GMP products by companies such as New England Biolabs and Aldevron175. Finally, the recent launch of the Coalition for Epidemic Preparedness Innovations (CEPI) provides great optimism for future responses to emerging viral epidemics. This multinational public and private partnership aims to raise$1 billion to develop platform-based vaccines, such as mRNA, to rapidly contain emerging outbreaks before they spread out of control.

The future of mRNA vaccines is therefore extremely bright, and the clinical data and resources provided by these companies and other institutions are likely to substantially build on and invigorate basic research into mRNA-based therapeutics.

## Publisher's note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.