mRNA vaccines in oncology: personalized cancer immunization and neoantigen targeting

ABSTRACT

Precision oncology is evolving with personalized mRNA neoantigen vaccines, although long-term clinical responses vary. These mRNA-based vaccines have facilitated the development of patient-specific neoantigens. Clinical success is dependent not only on immunogenicity but also on tumor neoantigen clonality, expression, and presentation by the vaccines. Evidence from trials conducted from 2020 to 2025 shows that indicators like minimal residual disease are crucial. The mRNA-4157 (V940) combined with pembrolizumab demonstrated improved recurrence-free survival in resected high-risk melanoma patients (18-month RFS 79% vs 62%; HR 0.56). Similarly, the autogene cevumeran triggered significant neoantigen-specific T cell responses in 8 out of 16 patients with resected pancreatic ductal adenocarcinoma, leading to a delayed recurrence for immune responders (not reached vs 13.4 months; HR 0.08). This highlights a translational model focusing on tumor clonality, antigen quality, and immune accessibility. The review also addresses (i) clonality aware neoantigen selection; (ii) AI-based predictions of antigen presentation and immunogenicity, including issues of false positives; (iii) alternative delivery systems beyond lipid nanoparticles; and (iv) real-world challenges such as turnaround time, batch variability, regulatory frameworks, and operational costs that impact implementation. A structured model for crucial events and validation plans is proposed to bridge the gap between predictions and actual clinical benefits, utilizing techniques like immunopeptidomics and functional T cell assays.

Introduction

Immunotherapy of cancer has come a long way, starting with primitive approaches, such as bacterial toxins, to modern treatments, such as checkpoint blockade and CAR-T.¹ The concept of therapeutic cancer vaccination has gained popularity, particularly in the application of tumor-specific neoantigens.² The most recent discovery in the creation of mRNA vaccines against COVID-19 was the astounding success that demonstrated that synthetic mRNAs in lipid nanoparticles (LNPs) could be created and generated with previously unheard-of speed and elicit a potent immune response. This has triggered the use of mRNA technology in oncology, where individualized cancer vaccines against patient-specific neoantigens can be designed within a short period.^2,3 In fact, over 150 clinical trials are presently examining mRNA-based cancer vaccines (usually in combination with checkpoint inhibitors), and these investigations have demonstrated encouraging results. For instance, the Phase IIb KEYNOTE-942 trial demonstrated that an mRNA neoantigen vaccination in conjunction with pembrolizumab reduced recurrence in resected melanoma more effectively than pembrolizumab alone.^3-5 The development of high-throughput sequencing and AI has also improved the discovery of neoantigens and vaccine design.^2,3

Immunotherapy of cancer has diverted attention from traditional cytotoxic drugs to precision vaccines that stimulate the host immune system to tumor antigens.⁶ Neoantigens (peptides formed by tumor-specific mutations) are naturally non-self and are selectively expressed by malignancies, and are therefore the best vaccine targets.⁷ The mRNA vaccine method was validated by the COVID-19 pandemic, which demonstrated that in vitro-transcribed (IVT) mRNA encapsulated in lipid nanoparticles (LNPs) may produce a robust immunity more quickly than ever before.⁸ These findings have sparked a novel idea in oncology: the rapid development and production of customized mRNA vaccines based on an individual's tumor mutations that comprise a variety of neoantigens (and even immunomodulators).⁹ The mRNA platform has three key features: (i) Speed and Flexibility, individual tumor antigen constructs can be expressed in about 2–4 weeks of tumor sequencing; (ii) Multicomponent Encoding, mRNA can encode more than one tumor antigen (shared or unique) and induce other adjunct cytokine genes with the same construct; (iii) Broad Immune Activation, expression of mRNA into the tumor activates antigen-presenting cells (APCs) and leads to both MHC class I and II presentation.^9,10

However, the following difficulties are linked to the lack of clinical translation: the manufacture of customized vaccines is logistically challenging, and the heterogeneity of tumors and immunosuppressive milieu may limit the effectiveness.⁹Even though the field is rapidly advancing, there are still some biological and translational questions that put the clinical efficacy of personalized mRNA vaccines into doubt. Recent reviews, such as , have fully outlined the technological history of the development of personalized mRNA cancer vaccines.¹¹ Nevertheless, the majority of previous syntheses focused on immunogenicity and platform development as opposed to the translational determinants of clinical success. The current review will be a step in the right direction, as it incorporates clonality-aware neoantigen selection, AI-driven prioritization constraints, and real-world deployment challenges into a single-purpose critical-path framework. This review offers a translational viewpoint that goes beyond the current descriptive literature by moving the dialogue away to clinical durability.

Molecular mechanisms

Neoantigen generation and T cell priming

Neoantigens are produced by tumor-specific mutations, such as non-synonymous single-nucleotide variations (SNVs), insertions/deletions (INDELs), gene fusions, aberrant splice variants, or post-translational modifications that alter peptide sequences.⁷ Taking SNVs or INDELs in an oncogene (e.g. KRAS^ G12D) as an example, the resultant mutant epitopes are not present in normal cells.¹² The ubiquitinated mutant proteins are degraded by the proteasome into peptides (usually 820 aa). TAP transports high-affinity peptides into the endoplasmic reticulum, where they load MHC class I molecules and appear on tumor cell surfaces.¹³ CD8+ cytotoxic T cells (CTLs) use their T cell receptors (TCRs) to detect these peptide-MHC-I complexes. When the TCR recognizes the mutated peptide and thinks it is foreign, the CD8+ T cell is activated to destroy the tumor cell. There is also a spontaneous CTL response against neoantigens, which is not commonly observed in the absence of vaccination.^7,14

Neoantigens may also be processed and cross-presented using MHC class II by professional antigen-presenting cells (APCs), particularly dendritic cells (DCs). The CD4+ helper T cells are primed by a smaller number of neo-epitopes on MHC-II molecules (approximately 0.5% of the anticipated).^15,16 Neoantigen-recognizing CD4+ T cells, which are less common but significant, contribute to anti-tumor immunity by both directly eliminating tumor cells that express MHC-II and secreting cytokines (such as IFN-γ and IL-2) to enhance the effect of CTL activity.¹⁷ Neoantigen-specific CD4+ assistance may be essential in influencing long-term responses, according to recent data. Interestingly, effective neo-antigens typically alter TCR-exposed residues to promote foreignness while maintaining MHC anchor regions, as in the wild-type peptide.^16,18 This, in practice, implies that such a mutant epitope frequently has to bind the patient HLA tightly (to appear on MHC), but non-self at other residues, such that there is no central tolerance.¹⁸ In fact, the immunogenic neoantigens frequently contain important anchor motifs that overlap those of self-peptides but vary in exposed residues, as recognized by the TESLA consortium as a characteristic of vaccine targets.¹⁹

Innate sensing and mRNA vaccine action

When mRNA-LNP vaccines are administered (typically intramuscularly or intradermally), APCs, especially DCs that live in the lymph nodes, absorb them. LNPs protect mRNA from extracellular RNases and facilitate their cellular uptake.⁸ Ionizable lipids in the LNP become positively charged in the endosome (pH ~6), rupturing the membrane and releasing mRNA into the cytoplasm.²⁰ The mRNA vaccine has a poly(A) tail, an open reading frame (ORF) that encodes the antigen or antigens, optimized 5′ and 3′ untranslated regions (UTRs), and a 5′ cap. By binding eIF4E, the 5′ cap structure protects the mRNA from decapping enzymes and attracts ribosomes for translation.²¹ Sequence selection or artificial intelligence design is used to create flanking UTR sequences that optimize translational efficiency and stability.²² To dampen innate RNA sensors (like TLR7 and RIG-I) and avoid excessive type I interferon responses that might otherwise impede protein translation, chemical nucleotide modifications (like N^1-methylpseudouridine) are frequently added to the mRNA.²³

Ribosomes convert the mRNA into complete antigenic proteins once it is in the cytoplasm. These proteins, whether intracellular or secreted, are processed in a manner akin to that of endogenous proteins.²⁴ The proteasome breaks down cytosolic proteins into peptides, which are then attached to MHC-I molecules and displayed on the DC surface.¹³ To activate CD4⁺ T cells, peptides from extracellular or endocytosed proteins can be loaded onto MHC-II concurrently.²⁵ Importantly, DCs can prime CD4⁺ Th1 cells with MHC-II and activate CD8⁺ T cells by “cross-presenting” vaccine-encoded antigens via MHC-I. IFN-γ, IL-2, and other cytokines secreted by CD4⁺ cells stimulate B cell antibody production, increase CTL cytotoxicity, and foster memory development.^26,27 In the end, CD4⁺ helper cells increase and maintain the immune response, whereas activated CD8⁺ CTLs travel to tumors and directly kill cells exhibiting the neoantigen (Figure 1).²⁷

Figure 1.

Mechanism of action of mRNA Vaccines: mRNA vaccines increase the synthesis of antigens, priming immunological responses in lymph nodes and activating CD8 and CD4 T cells via MHC pathways. The figure is adapted from Ref no.²⁸

AI and bioinformatics for neoantigen prediction

mRNA neoantigen vaccination Starting with tumor-normal sequencing (usually WES with RNA-seq) and HLA typing, computational prioritization of candidate neoepitopes is performed.²⁹ Even though better prediction of peptide-MHC binding is necessary, clinical translation is limited by the continued presence of false positives: predicted binders are not naturally processed/presented, and peptides that are presented do not typically cause functional T cell responses.³⁰ Consortium benchmarking (TESLA) proved that the neoantigen immunogenicity is dictated by a cascade of events — variant expression, processing, presentation, and TCR recognition, and therefore ranking based on binding affinity is not enough.³¹

The use of ligand data derived by immunopeptidomics gradually supplants more modern predictors of presentation: NetMHCpan-4.1/NetMHCIIpan-4.0, both of which combine binding affinity and MS eluted ligand data, with enhanced pan-allelic prediction of both classes I and II.³² Complementary models are like MHCflurry 2.0, which includes antigen processing parts, and MS-driven models such as HLAthena, which can expand allele coverage and limit bias on well-investigated HLA types.³² Practically, pipelines with the best performance are provided with ensemble methods or consensus (e.g. pVACtools-based workflows) and filters of expression and clonality to reduce noise. Algorithms, despite the success in the laboratory, the predicted immunogenicity is not perfectly correlated with clinical success, as host factors and the tumor environment take center stage: HLA loss of heterozygosity, microenvironmental exclusion, and subclonal selection can, despite an immune-relevant target, make it clinically ineffective.³³ Thus, the strategy of validation must be presented as phase-appropriate: (i) genomic/transcriptomic evidence of clonal, expressed mutations; (ii) confirmation of presentation by immunopeptidomics whenever possible; (iii) functional T cell verification (ELISpot/ICS, multimer or TCR-based confirmation).³⁴ Significantly, in resected PDAC, high-threshold assays detected de novo high-magnitude vaccine-specific responses in 8/16 patients and immune responders had delayed recurrence, demonstrating the clinical importance of so-called validated immunogenic targets, rather than predicted binders, as the most clinically significant.³⁵

Peptide–MHC binding

Initial AI systems relied on neural networks to analyze binding affinity data (e.g. NetMHC). Recent deep learning state-of-the-art models represent more complicated features, such as peptide structure and flanking residues.³⁶ As an illustration, NeoaPred builds 3D scaffolds of peptide HLA complexes and was ~82% accurate at separating immunogenic peptides.³⁷ DeepHLApan and the CNN/RNN models involve the full context of antigen processing (including the flanking pattern of the proteasome) to contextualize binding predictions. Compared to classical methods that work on motifs, these deep models often perform better on benchmark data.³⁸

TCR recognition and immunogenicity

A peptide that binds MHC with high affinity does not necessarily activate a T cell when the TCR repertoire does not match, or when the peptide is self-like. Predicting peptide-TCR interactions, AIs are being developed.³⁹ For instance, ERGO uses paired peptide–TCR sequences and an LSTM network to predict binding probability. To discover actual peptide–TCR couples with high specificity, NetTCR-2.0 employs CNNs on the sequence concatenation of peptide and TCR α/β CDR3 sequences.^40,41 One of them is a transfer-learning model, pMTnet, which is trained on large datasets of viral epitopes and TCRs and then fine-tuned on tumor neoantigens to make new TCR–pMHC binding predictions. The objective of these tools is to estimate the neoantigen–HLA complexes that T cells consider to be seen.⁴²

Moreover, integrative predictors of immunogenicity are equations that integrate several features. As an example, Zhang et al. constructed a pipeline using distinct deep models of MHC binding, antigen processing, and immunogenicity and combined the scores to rank neoantigens.⁴³ NUCC works better than NetMHCpan in a gastric cancer dataset and has found experimentally validated epitopes. AI can learn to assess immunogenic epitopes, which are typically medium in size and moderately to poorly hydrophobic, and exhibit an ideal divergence from self.⁴⁴

To implement these models, many have been integrated into web servers and pipelines that are easily accessible. One example of a sequence and structure deep learning predictor of peptide immunogenicity is DeepNeo, while pVACtools generates a sorted list of neoantigens by combining several predictors (binding, expression filters, and proteasome cleavage).^45,46 An autoencoder termed DeepVACPred is based on developing ideal sets of epitopes for use in vaccinations. Clinical procedures have already begun to be improved by such AI-enhanced systems; a study revealed that trials using these technologies had better predicted neoantigen validations. In general, AI will be essential in optimizing the process of neoantigen selection, greatly decreasing false-positive targets and accelerating the development of vaccines.⁴⁷

In addition to epitope prediction, AI can be used in designing the mRNA sequence. Deep learning models are optimized to codon usage and UTR sequences, maximizing protein production.²³ For instance, GANs and LSTMs conditioned on massive expression datasets can generate synthetic 5′UTRs (like Smart5UTR) or simultaneously optimize code sequences and UTRs (like LinearDesign2).⁴⁸ These approaches have shown that AI-designed UTRs can increase protein output (as much as 2–3x) in relation to native ones. Likewise, balanced, high-level antigen translation in DCs can be achieved by AI-guided codon optimization (e.g. BiLSTM-CRF models).⁴⁹ With multivalent vaccines, these strategies can be applied in practice to achieve high and uniform expression levels of multiple peptide neoantigens. Therefore, AI optimizes the detection of antigens and the creation of mRNA constructs.⁴⁸

mRNA design and delivery technologies

mRNA molecule engineering

The synthetic mRNA used in vaccines is carefully engineered for stability, translation, and immunogenicity. Notably, each mRNA contains four main parts: (i) a 5′ cap (typically a modified guanine nucleotide like CleanCap or anti-reverse cap analog (ARCA)), (ii) 5′ and 3′ UTRs, (iii) the ORF encoding the antigenic peptide(s), and (iv) a 3′ polyadenylated (poly(A)) tail.⁸ The 5′ cap is crucial for recruiting ribosomes and protecting against exonucleases. Modern vaccines often use cap analogs (e.g. Cap1, 2′-O-methylated) that further enhance translation and evade innate sensors.^8,50 The poly(A) tail, usually ~100–150 adenosines, increases mRNA stability and translation efficiency by interacting with poly(A)-binding proteins.⁵¹ UTRs are taken from highly expressed genes (such as alpha-globin) or designed de novo (via AI) to maximize ribosome loading and minimize decay. For example, engineered 5′ UTRs with reduced secondary structures improve initiation, while optimized 3′ UTRs can extend the mRNA half-life.⁵²

Codon usage in the ORF is another design variable. Synonymous codon optimization aligns with abundant tRNAs of the production system (e.g. human or mouse DCs) to accelerate translation.⁵³

Traditional heuristics (codon adaptation index) are now often supplanted by AI models that learn from ribosome profiling; these can tune codon choice to boost expression and avoid depletion of rare tRNAs. In practice, codon and UTR optimization are integrated (e.g. LinearDesign2) to balance translation speed and mRNA folding stability.⁴⁸ Chemical nucleotide modifications—most commonly N^1-methylpseudouridine in place of uridine—are incorporated during IVT. Such modifications blunt the activation of RIG-I/TLR7 sensors and reduce type I IFN responses, thereby preventing the shutdown of antigen expression. Altogether, these design principles ensure that the mRNA vaccine can produce high antigen levels in DCs while evading excessive innate immune inhibition.^23,24

Lipid nanoparticle and other delivery platforms

Lipid nanoparticles (LNPs) play a key role in the in vivo delivery of mRNA, especially in mRNA cancer vaccines, and in the transition from immunogenicity research to randomized trials, particularly in minimal residual disease. The LNPs are approximately 100 nm in diameter, and they may be composed of an ionizable lipid (e.g. DLin-MC3-DMA, SM-102, ALC-0315), a helper lipid (e.g. DSPC), cholesterol and a PEG-lipid to increase circulation.^20,54 These parts autofold to entrap mRNA, which can be co-loaded during multifunctional vaccination approaches.⁵⁴ PEGylation enhances blood stability and targeting to lymphoid tissues and antigen-presenting cells (primarily dendritic cells), promoting cytoplasmic mRNA release and preventing nuclease degradation and access to the translation machinery.⁵⁵ This drastically improves the antigen expression, immunogenicity and stability of the vaccine. The effectiveness of LNP-based mRNA COVID-19 vaccines, with billions produced cheaply, presents their scalability, high manufacturing, and regulatory compliance.⁵⁶

Despite these benefits, LNPs must be viewed as a dynamic technological platform rather than a solution set in stone. Several parameters can greatly determine vaccine potency and safety, including biodistribution, lymph node targeting, endosomal escape efficiency, innate reactogenicity, and repeat-dosing tolerability.⁵⁷ Preferential hepatic accumulation, which can be up to 80%–90% and is predominantly mediated by ApoE interactions, is also a significant liability that could limit the successful delivery of extrahepatic tumors but can be beneficial to liver-directed therapies.⁵⁷ Furthermore, the inflammatory reactions linked with lipid elements should be optimized carefully, and the formation of anti-PEG antibodies after repeated administrations may decrease the effectiveness or cause infusion-related reactions.⁵⁸ Such limitations have led to the production of the next generation of LNPs with targeting ligands, alternative lipid chemistries, or biodegradable materials to enhance tumor specificity and immune compatibility.²⁰

At the same time, new shipping solutions are expanding the translational environment and could be used to address the existing drawbacks. Given their tunable physicochemical properties, polymeric nanoparticles and lipid‒polymer hybrid systems permit controlled release of antigens, enhance structural stability and may permit targeted delivery to tissues in a tissue-specific way.⁵⁹ Nanoparticles made of proteins, virus-like and exosomes and other nanocarriers have different immune-interaction profiles that can potentially improve antigen presentation and reduce systemic toxicity.⁶⁰ Localized and sustained antigen exposure. Injectable depots and hydrogel platforms have the potential to co-deliver molecular adjuvants, which enhance the activation of dendritic cells and induce long-lasting immune memory.⁶¹ In the meantime, cell-based methods, including the dendritic cells transfected ex vivo with mRNA, go around many of the delivery barriers of in vivo delivery, and permit more specific development of antigen presentation, although with higher complexity of individualized manufacturing.⁶² The most effective platforms will necessarily be delivery-agnostic in this prospective view: they must be able to consistently address professional antigen-presenting cells, facilitate a coordinated prime of MHC class I and II, and integrate with combination therapies aimed at re-modeling the immunosuppressive tumor microenvironment.⁶³ Further developments in biomaterials engineering, targeted delivery and immune modulation should see next-generation delivery systems become a key determinant of the ultimate clinical efficacy of personalized mRNA cancer vaccines (Figure 2).²⁰

Figure 2.

Delivery mechanism of LNP@mRNA. Image adapted from Ref. no.⁶⁸

Table 1 could compare LNPs with alternative carriers (viral vectors, polymers): LNPs currently dominate due to manufacturing maturity and high encapsulation rates.

Table 1.

Clinically relevant delivery platforms for mRNA cancer vaccines.

Delivery platform	Delivery efficiency	Immunogenicity and efficiency	Manufacturing maturity	Clinical trial use	Key limitations
Lipid nanoparticles (LNPs)20,64	High encapsulation and transfection	Moderate innate activation; generally manageable	High (established GMP, scalable)	Predominant platform in phase I–III mRNA cancer vaccine trials	Liver accumulation; reactogenicity; cold-chain dependence
Viral vectors65	Very high gene expressions	High vector immunogenicity; anti-vector immunity	Moderate	Limited use for cancer vaccination; more common in gene therapy	Safety concerns; limited repeat dosing
Polymeric nanoparticles66	Moderate, formulation-dependent	Lower innate activation than viral vectors	Low-moderate	Early phase/pre-clinical studies	Lower efficiency; lack of standardization
Exosomes-based systems67	Moderate; loading dependent	Low intrinsic immunogenicity	Low	Preclinical and exploratory translational studies	Manufacturing and scale-up challenges

Preclinical and clinical evidence

Preclinical models

In animal models, mRNA neoantigen vaccines have consistently elicited strong immune responses and tumor control. For example, mice bearing transplantable tumors (melanoma, colon carcinoma, etc.) have been vaccinated with neoantigen-encoding mRNA–LNP, resulting in robust CD8⁺ T cell activation and prolonged survival compared to controls.⁶⁹ These studies confirmed that chemically stabilized, codon-optimized mRNA can prime functional CTLs without exogenous adjuvant.⁷⁰ Preclinical work has also demonstrated the feasibility of multiplexing dozens of neoantigens in one vaccine, owing to the large coding capacity of mRNA. Such animal data laid the groundwork for human trials.⁶⁹

Clinical trials and evidence

Recent clinical experience with personalized mRNA neoantigen vaccines points to two broad conclusions that individualized vaccination can be achieved and induce de novo and polyfunctional T cell responses, but clinical benefit can be meaningful only in cases where the tumor burden is limited and where immunomodulatory combinations are used.⁷¹ Initial proof-of-concept investigations on synthetic long peptides and RNA platforms in melanoma demonstrated that mutation-directed immunization could be safely achieved to produce robust CD4+ and CD8+ T cells, and that the therapeutic rationale of personalized immunization could be fulfilled despite heterogeneous tumor regression.⁷² This paradigm was improved in subsequent trials.⁷² In the mRNA-4157 (V940) trial (KEYNOTE-942) randomized phase IIb trial with pembrolizumab in patients with resected high-risk melanoma, a combination of pembrolizumab and mRNA-4157 (V940) showed an 18-month recurrence-free survival (79 vs. 62; HR 0.56) with no surplus immune-mediated toxicity, despite the trial being an open-label design.⁵⁴ In resected immunologically cold tumor pancreatic ductal adenocarcinoma, autogene cevumeran with atezolizumab and chemotherapy induced high-magnitude neoantigen-specific responses in about 50% of patients, and responders had delayed recurrence (not reached median RFS vs. 13.4 months).⁷³ It is worth noting that there was an increased tumor clonality in the responders, which is a strong argument in favor of the need to target truncal, uniformly expressed neoantigens.⁷⁴

On the other hand, initial basket trials in progressive metastatic disease indicate immunogenicity can be observed with low objective response rates, which is indicative of biological inhibitory capacity, including immune exclusion, antigen presentation defects, T cell exhaustion, and subclonal heterogeneity.⁷⁵ Antigen clonality, tumor immune contexture, and timing of vaccination seem to be strong forces of clinical response; subclonal targeting might permit antigen-negative escape, whereas therapeutic pressure can dynamically remodel the neoantigen landscape between sequencing and dosing.⁷⁶ These findings suggest that the effectiveness of the vaccine does not only depend on the choice of the epitopes but also on the level of interactions between the evolution of tumors and host immunity.⁷⁷

Taken together, existing data suggest that minimal residual disease environments should be prioritized, that rational combinations of these settings should be used, and that structured clinical endpoints not based solely on immunogenicity, such as recurrence-free survival, distant metastasis-free survival, tumor DNA dynamics in the blood, and intratumor clone expansion, should be embraced.⁷⁸ Timelines of manufacturing will be standardized, and patient disappearance between sequencing and dosing will also help further explain translational feasibility.⁷⁹ Further rationalization of antigen prioritization, patient selection and combination strategies will be necessary to move from a strong immunogenicity to a high-quality clinical response (Table 2).⁷⁸

Table 2.

Comparative summary of key mRNA neoantigen vaccine trials and programs (2020–2025).

Trial/program	Cancer setting	Design features	Endpoints (as reported)	Key results	Key limitations and translational lessons
KEYNOTE-942 (mRNA-4157/V940 + pembrolizumab)54	Resected high-risk cutaneous melanoma (adjuvant/MRD)	Randomized, open-label phase 2b; 2:1 allocation; 157 patients; ~23–24 mo median follow-up	Primary: RFS; also DMFS; safety	18-mo RFS 79% vs 62%; HR 0.561 (95% CI 0.309–1.017), p = 0.053; grade ≥ 3 TRAEs 25% vs 18%; immune-mediated AEs similar (36% vs 36%)	Open-label; follow-up for OS claims is still restricted; borderline p-value and CI surpassing 1. Nevertheless, it offers the most convincing randomized evidence that tailored mRNA neoantigen treatment can enhance MRD results
Autogene cevumeran + atezolizumab + mFOLFIRINOX (phase I)74	Resected PDAC (adjuvant/MRD)	Single-arm; sequential regimen; safety cohort n = 19; biomarker-evaluable n = 16; manufactured/dosed ~9 weeks from surgery	Endpoints included high-threshold vaccine-induced neoantigen-specific T cells; feasibility; 18-mo RFS	De novo high-magnitude neoantigen-specific T cells in 8/16; responders had longer RFS (median not reached vs 13.4 mo; HR 0.08); feasibility: median time to vaccine ~9.4 weeks; limited grade ≥ 3 AEs (notably 6% grade 3 with vaccine)	Limited diversity (only white persons are noted in the trial); correlation-based outcome analysis; non-randomized; small cohort; highlights the necessity of randomized follow-up trials and biomarkers (clonality/immune competence)
RO7198457 (autogene cevumeran/BNT122) ± atezolizumab (basket)80	Advanced/metastatic solid tumors	Phase 1a/1b dose escalation/expansion (basket); reported in meeting/early updates	Safety; immunogenicity; ORR/stable disease in early analyses	Early reports describe manageable safety, ORR ~8% with combination, and ~50% stable disease	Results are mostly from abstracts and press summaries rather than complete peer-reviewed papers; the advanced-disease context probably penalizes response because of tumor burden and TME; this shows why MRD settings would be better for vaccinations
NCT05968326 (autogene cevumeran + atezolizumab + mFOLFIRINOX vs mFOLFIRINOX)81	Resected PDAC (randomized adjuvant; follow-up program)	Randomized design planned/ongoing	Efficacy and safety; (trial registry-level details)	No outcomes yet (trial ongoing/registered)	Signifies the essential next step, which is improved effect magnitude and generalizability estimation as well as randomized validation of the PDAC immune–RFS connection

Discussion

The growing preclinical and clinical data demonstrate the potential of personalized mRNA neoantigen vaccines to revolutionize cancer immunotherapy by triggering tumor-specific adaptive immune responses.⁸² Immuno-surveillance: in contrast to shared tumor-associated antigens, the neoantigens are tumor-specific and can be expressed using personalized mRNA repertoires, which allow specific activation of the immune system with minimal adverse off-target toxicity.⁸³ The clinical trials in melanoma and pancreatic cancer show that the vaccines have the potential to induce long-lasting polyfunctional T cell responses and enhance the ability to survive longer without recurrence, especially when used together with immune checkpoint inhibitors.² However, the therapeutic efficacy is still inconsistent because biological and technical limitations are still present. A significant factor of vaccine success is tumor heterogeneity because neoantigens reside in an evolutionary structure: those based on clones, or subclonal variants, are expressed in subsets and have the capacity to evade antigen-negative selection.⁸⁴ Further evolution of temporal tumor-specific adaptations in response to therapeutic pressure has the potential to redefine the neoantigen space between sequencing and vaccination, making some of the designed antigen repertoires displaced.⁸⁵ In line with this paradigm, the results of resected pancreatic ductal adenocarcinoma treated with autogene cevumeran indicate that immune responders had high tumor clonality, delayed relapse, and longer-lasting increase in neoantigen-specific CD8+ T cell clones that have the capability of homing to micro metastatic locations.⁷⁴ The immune editing provides another layer of defense because tumors can downregulate the expression of neoantigens, impair antigen-processing mechanisms, and disallow major histocompatibility complex presentation by mechanisms, including HLA loss of heterozygosity, decoupling vaccine immunogenicity and effective tumor killing.⁸⁶ At the same time, one should also consider that the prediction of neoantigens is constrained by the advances in AI because integrative selection models that integrate antigen processing, T cell receptor recognition, clonality, and host immune competence are needed.⁸⁷ Large-scale implementation is further limited by translational difficulties such as the lipid nanoparticle biodistribution, reactogenicity and the high logistical challenge and cost associated with individualized production.³ The future in this field will probably rely on clonality-conscious antigen prioritization, longitudinal genomic surveillance, next-generation delivery, self-amplifying RNA scaffolds, and biomarker-directed trial designs and make neoantigen vaccines a promising platform of multimodal precision oncology.⁸⁸

Challenges and limitations

Numerous challenges hinder the clinical efficacy of personalized mRNA neoantigen vaccines, despite progress in the field. A key issue is tumor heterogeneity, where genetic variations lead to differing neoantigen expressions, allowing non-targeted subclones to evade immune responses.⁸⁹ Clonal neoantigens might be more reliable targets, yet many are subclonal mutations prone to immune editing. Additionally, T cell responses are compromised in an immunosuppressive tumor microenvironment, necessitating combination therapies with immune checkpoint inhibitors for effectiveness.⁹⁰ High false-positive rates in prediction pipelines complicate accurate neoantigen identification, as they are often biased toward well-characterized HLA alleles; improved validation systems that incorporate antigen processing and T cell receptor recognition are needed.⁹¹

Manufacturing scalability also poses challenges owing to intricate schedules and stringent legal standards compared to fixed-sequence biologics. The individualized production process is time-consuming, delaying treatments, and quality control for each vaccine lot remains difficult under strict quality criteria.⁸⁴ Current delivery systems, like lipid nanoparticles, face issues with reactogenicity and stability, signaling the need for thermostable systems and alternative delivery methods.⁹²

Regulatory models grounded in platform technology could streamline development, and robust regulation of bioinformatics and data integrity is essential. To enhance responsiveness, clinical applications should prioritize environments with strong biological and economic rationales, such as minimal residual disease.⁹³ These interrelated barriers call for greater collaboration across computational biology, manufacturing, regulatory frameworks, and clinical trial designs, along with innovations in automation and adaptive strategies, to facilitate the advancement of personalized mRNA vaccines in precision oncology.⁸⁴

Future perspectives

The field of mRNA cancer vaccines is rapidly evolving. Several emerging trends are poised to enhance their impact:

• •Self-amplifying RNA (saRNA): to boost antigen expression, saRNA constructs encode both the target antigen and viral replicase proteins. These replicate the RNA intracellularly, potentially yielding much higher protein output from a lower dose. SaRNAs have shown efficacy in preclinical cancer models and are entering early trials. They promise greater potency and dose-sparing.
• •Advanced neoantigen discovery: the scope and precision of next-generation sequencing (long-read, single-cell) are still growing. Neoantigen lists will be improved by integrating multi-omics (transcriptomics, proteomics, and genomes). Deep learning and AI will become more advanced, perhaps by integrating patient-specific variables (TCR repertoire, HLA variety) to customize epitopes. For semi-personalized “off-the-shelf” vaccines, extensive retrospective data may reveal conserved neoantigens across patient groupings.
• •Combination therapies: it is increasingly acknowledged that synergy with other modalities may be necessary for cancer vaccinations. Neoantigen vaccines are being tested in conjunction with oncolytic viruses, co-stimulatory agonists (OX40, 4-1BB), or ICIs (PD-1/PD-L1, CTLA-4). Vaccines may also be used in conjunction with adoptive cell therapies (CAR-T or TIL therapy) to grow certain T cell clones. For instance, PD-1 inhibition combined with immunization has already improved results for NSCLC and melanoma.
• •Enhanced delivery platforms: novel delivery vehicles are being investigated in addition to regular LNPs. Polymeric nanoparticles, such as PLGA-PEI hybrids, can be engineered for controlled release and lymph node targeting. mRNAs may be directed to particular APC subsets using ligand-directed or cell-penetrating peptide formulations. Even replicon RNA-carrying bacterial and viral vectors are being studied. Exosome-based delivery, which involves loading mRNA into patient-derived vesicles, is an interesting field that may lessen the carrier's immunogenicity.
• •Clinical innovations: the evaluation is being improved by adaptive trial design, biomarker-driven patient selection and real-time immune monitoring. For example, sequential biopsies and liquid biopsies can measure epitope coverage and on-treatment T cell growth to make iterative vaccine revisions. In an attempt to trigger a T cell response before resection, clinical trials are being carried out for a neoadjuvant vaccination that delivers mRNA vaccines before tumor excision. These techniques can increase the vaccine's efficacy and provide an opportunity to make it more dynamic.
• •Industrial and regulatory progress: scaling up standardized GMP processes of custom mRNA is partly being enabled, in part owing to the COVID-19 vaccine infrastructure. Regulatory bodies are developing systems of individualized treatments. Distribution would be revolutionized through the achievement of room-temperature stable formulations.

All of this means that mRNA cancer vaccines will be more effective, more specific and more feasible. The sector is heading to the point where a newly diagnosed cancer patient may have their tumor sequenced, their neoantigens chosen by artificial intelligence and their own mRNA vaccine printed in a few weeks, all made possible by automated and modular systems.

Conclusion

The field of personalized mRNA neoantigen vaccines has transitioned out of feasibility to evidence of clinical advantage, but more than the generation of quantifiable T cell responses is required to translate into clinical practice. Minimal residual disease cases are so far the most convincing indicators on the signal best supporting a model whereby a lower tumor burden, lower subclonal complexity and maintained host immune competence leads to the higher likelihood of vaccine-expanded T cells eliminating micrometastatic disease. Three priorities are likely to be the driving force behind future progress. The first step to reduce false-positive detection targets requires selecting better antigens through clonality-based antigen selection, which uses presentation-based validation methods that include immunopeptidomics analysis whenever possible. The second step involves microenvironment intelligent therapy, which uses multi-regimen treatments to improve antigen presentation and T cell activity in tumors beyond the effects of checkpoint blockade through rational molecular adjuvants and different treatment combinations. The third step requires implementation preparedness through regulatory platforms and standardized analytics methods, which enable faster manufacturing processes that decrease turnaround times, batch variability, and patient losses. In case these priorities are achieved, personalized mRNA neoantigen vaccines might transform into niche and highly specialized interventions into scalable parts of precision immunotherapy, particularly with MRD trial designs, longitudinal immune monitoring and adaptive biomarker strategies refining.

Acknowledgments

Not applicable. JR and MP helped with the first draft. DSP and JR collaborated to develop the conceptual framework. MP and DSP were responsible for collecting the information. The final evaluation was completed by JR. The integrity and correctness of each contributor's individual contributions to the analysis are their own personal responsibility. The work and its contents have been reviewed and approved for publication by all the authors. All authors have read and agreed to the published version of the manuscript.

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Funding

This research received no external funding.

Data availability statement

Data sharing does not apply to this article as no new data were created or analyzed in this study.

Informed consent statement

Not applicable.

Institutional review board statement

Not applicable.