A first-in-human dose has two jobs. It has to protect the participant, and it should put the programme in a position to learn something useful. For an mRNA therapeutic, both become more complicated because the administered dose sits several biological steps upstream of the pharmacology we ultimately want to achieve.
The active substance is the mRNA, while the pharmacological effector is produced in the patient. Following administration, the formulation has to reach the relevant tissue, enter the appropriate cells, release sufficient mRNA into the cytoplasm and support translation. The resulting protein can then have its own distribution, turnover, target interaction and downstream pharmacology. Recent reviews of clinical and quantitative pharmacology for mRNA therapeutics describe this as a linked set of kinetic and pharmacodynamic processes rather than a conventional plasma concentration-response relationship (PMID: 41085467; PMID: 40083288).
That creates a deceptively simple development question. If an animal receives 0.3 mg/kg of an mRNA-LNP therapeutic, what does that tell us about the dose that should be given to the first human?
Toxicology provides one boundary
Animal toxicology remains central to first-in-human development. NOAEL-based approaches, human-equivalent dose calculations and safety factors provide a well-established framework, while modern risk-based approaches can also integrate pharmacology, target biology, species relevance and other information when selecting the starting dose (PMID: 32757299).
For an mRNA therapeutic, interpretation of those data requires an understanding of where an adverse effect originates. Relevant effects can arise from the delivery system, innate immune stimulation, the distribution or level of mRNA expression, the biological activity of the encoded protein, or interactions between these components. Current clinical pharmacology frameworks therefore consider the mRNA and its LNP delivery system together when evaluating exposure, pharmacology and tolerability (PMID: 41085467).
This creates two different boundaries around the dose range. The minimum dose capable of producing useful pharmacology may be determined by how much functional protein can be produced in the relevant cells, while the upper end of dosing may be constrained by quite different properties of the formulation or encoded product. A large safety margin relative to an animal NOAEL is valuable, but it does not establish where the starting dose sits on the expected human pharmacological dose-response curve.
Work backwards from the biology
The pharmacological side of the calculation starts with what the therapeutic is intended to achieve. For an enzyme-replacement mRNA, that might be sufficient functional enzyme activity to alter a disease-relevant metabolic pathway. For an mRNA encoding a secreted protein, it might be a defined circulating protein exposure or degree of receptor engagement. An immune-modulatory protein could require a more conservative entry point if relatively small amounts of protein can produce substantial downstream effects.
A useful simplified chain is:
administered mRNA dose → delivery to target cells → intracellular mRNA → translated protein → target engagement or functional activity → biological response.
Each transition introduces uncertainty. Species can differ in biodistribution, cellular uptake, translation, protein turnover and downstream biology, and those differences will not all scale in the same way. Translational modelling provides a structured way to combine the available information, quantify key assumptions and make a prediction that the first human cohorts can subsequently test.
A real example: mRNA-3927 in propionic acidaemia
The mRNA-3927 programme in propionic acidaemia provides one of the clearest published examples. mRNA-3927 contains mRNAs encoding the PCCA and PCCB subunits of propionyl-CoA carboxylase. Investigators developed a semi-mechanistic translational PK/PD model using data from a disease-model mouse, rats and cynomolgus monkeys, linking administered dose to PCCA/PCCB mRNA exposure and then to changes in disease-associated metabolites. The interspecies model was scaled to humans to guide the first-in-human dose range and dosing regimen (PMID: 36577040).
Subsequent work applied related translational models across three metabolic mRNA programmes. The predicted first-in-human doses were 0.3 mg/kg for mRNA-3927 in propionic acidaemia, 0.1 mg/kg for mRNA-3705 in methylmalonic acidaemia and 0.4 mg/kg for mRNA-3210 in phenylketonuria. These projected doses sat below the respective nonclinical NOAELs, giving reported safety margins of 10-, 50- and 7.5-fold (PMID: 38714648).
The published dose-response plots from that work are particularly useful because they make the decision visible. Model-predicted changes in disease biomarkers are shown across a range of human mRNA doses, while the selected first-in-human doses can be interpreted in the context of both expected pharmacology and nonclinical safety. Figure 5a-b from the paper is therefore a useful illustration of the dose-selection problem for mRNA-3927 and mRNA-3705 (Baek et al., Nature Communications 2024, Figure 5).
Did the prediction translate into humans?
For mRNA-3927, there is now a published clinical answer. Interim analyses from a first-in-human Phase 1/2 dose-optimisation study reported 16 participants across five dose cohorts and 346 intravenous administrations. No dose-limiting toxicities occurred, exposure increased with dose escalation, and among the eight participants who had experienced metabolic decompensation events during the 12 months before treatment, the interim analysis estimated a 70% reduction in event risk during treatment (PMID: 38570682).
Those results do not establish that every assumption in the original translational model was correct, and interim Phase 1/2 data cannot establish the ultimate clinical value of the therapy. They do show that the modelling supported a clinically usable entry point from which the programme could escalate dose, collect human pharmacological data and refine its understanding of the dose-response relationship.
That distinction is important. The purpose of the starting dose is not to predict the final therapeutic dose perfectly. It is to enter the clinic with an appropriate margin of safety and a rational route towards an informative exposure range, while collecting the data needed to replace preclinical assumptions with human observations.
Prospective prediction is more useful than retrospective explanation
Another instructive example comes from mRNA-6231, which encodes an albumin-fused IL-2 mutein designed to expand regulatory T cells. A semi-mechanistic kinetic-pharmacodynamic model connected administered mRNA dose to expression of the encoded HSA-IL2 mutein and then to T-regulatory-cell expansion in non-human primates. The model was translated to humans using allometric and physiological principles and used prospectively to inform the dose selection and design of the first-in-human study (PMID: 38676306).
Once Phase I human data became available, the investigators compared the observed pharmacodynamic response with the earlier predictions and reported that the modelling approach credibly predicted the clinical response. This is a particularly useful form of validation because the model had to make a prediction before the human data were known, rather than being built afterwards to describe an already observed result.
Where MABEL can contribute
MABEL, the minimum anticipated biological effect level, was developed particularly for situations in which a starting-dose strategy based largely on conventional toxicology might provide inadequate protection against potent pharmacology. The approach can integrate target biology, human in-vitro activity, receptor occupancy, PK/PD and relevant animal data to estimate a dose expected to produce a minimal biological effect in humans. Much of the foundational thinking came from monoclonal antibody development, where highly potent or novel target biology can make this type of pharmacology-led approach particularly valuable (PMID: 19896825).
FDA brought the subject back into focus in June 2026 with draft guidance on the use of quantitative systems pharmacology to support MABEL-based first-in-human dose selection. The guidance is not specific to mRNA therapeutics, but its emphasis on mechanistically linking administration, distribution, target engagement, signalling and downstream biological response is highly relevant to a modality in which the administered material can be several steps removed from the final pharmacological effect. The document also emphasises model uncertainty, sensitivity analysis and updating the model when early human PK and PD data become available (FDA draft guidance, June 2026).
Consider an mRNA encoding a potent cytokine. A biologically meaningful threshold might ultimately be defined by a relatively low cytokine concentration, receptor occupancy or downstream cellular response. Working back to the administered dose requires estimates of how much encoded protein will generate that response, how much protein will be produced from a given amount of intracellular mRNA, and how the administered mRNA dose relates to delivery into the relevant cells.
MABEL will not automatically be the appropriate pharmacological anchor for every mRNA programme. An enzyme-replacement therapy may have a defined level of functional enzyme activity or biomarker correction that provides a more informative target, while an encoded antibody may have a clinically interpretable circulating concentration. The selection framework should follow the biology of the encoded protein and the uncertainties that could cause the greatest harm if the human prediction is wrong.
Some encoded proteins are much harder to follow than others
The metabolic-disease examples are attractive for translational pharmacology because they offer relatively clear chains between mRNA, functional enzyme and measurable disease biomarkers. Many mRNA programmes will be less accommodating, particularly when the active protein remains intracellular or acts locally within a tissue.
A secreted protein gives us a circulating analyte that can potentially be measured repeatedly. Human experience with mRNA-1944, an mRNA therapy encoding a chikungunya-neutralising antibody, illustrates how different the kinetics of the administered mRNA and its encoded product can become. A single administration generated measurable antibody rapidly, while the encoded antibody had a mean terminal half-life of approximately 69 days and remained above a predicted therapeutically relevant concentration for at least 16 weeks at the higher dose levels studied (PMID: 34887572).
For an intracellular enzyme or structural protein, circulating protein exposure may be irrelevant. The programme then needs evidence that sufficient functional protein has been produced in the correct cells and intracellular compartment, and the most useful pharmacodynamic readout may sit downstream of the protein itself.
Transcription factors expose the limits of a simple PK/PD framework
A transcription factor adds another level of complexity because the encoded protein can act as a regulator of an entire biological programme. After translation, it may enter the nucleus and change expression across a network of downstream genes, so the pharmacological response cannot necessarily be represented by a single circulating protein concentration or one proximal target-engagement measurement.
Preclinical work with HNF4A mRNA illustrates this problem. Delivery of HNF4A mRNA to hepatocytes restored hepatocyte-associated functions and attenuated fibrosis across preclinical models of chronic liver injury. Mechanistic work included gene-expression profiling, single-cell RNA sequencing and chromatin immunoprecipitation to understand how transient HNF4A expression altered downstream transcriptional programmes (PMID: 34453962).
For this type of therapeutic mechanism, first-in-human translation raises different questions from an enzyme-replacement programme. The programme needs to understand how much expression is required in the target cell population to shift the desired transcriptional programme, which downstream changes can serve as early pharmacodynamic markers, how those changes relate to a therapeutically relevant phenotype, and what level or distribution of expression could create unwanted biology. The proportion of cells reached by the delivery system may become as important as the average amount of protein expressed in the tissue.
The assay strategy has to be ready before the first dose
The translational hypothesis becomes substantially more valuable when the first clinical cohorts can test it. If a model predicts measurable encoded protein at the starting dose but the clinical assay cannot reliably quantify protein at that concentration, one of the most useful early checks on the model has been lost. Where direct measurement of the protein is impossible, a sufficiently proximal downstream pharmacodynamic marker may be needed instead.
Sampling schedules also need to reflect the biology being measured. The kinetics of circulating mRNA, intracellular translation, encoded-protein turnover and downstream pharmacodynamics can differ markedly, and a sampling schedule designed around conventional plasma drug PK can miss the useful window for protein expression or biological response. The clinical protocol, bioanalytical plan and translational model therefore need to be developed together.
Published pharmacometric work in the mRNA field remains relatively young. A 2025 systematic review identified 15 published quantitative modelling studies supporting preclinical or clinical development of mRNA-LNP modalities through its October 2024 search cut-off, spanning PK/PD, kinetic-PD, PBPK, QSP and vaccine immunodynamic approaches (PMID: 40933709). The available toolkit is expanding quickly, although the value of a model still depends on whether it addresses the specific development decision facing the programme.
The first dose and the escalation strategy belong together
Selecting the starting dose cannot be separated completely from deciding what happens afterwards. A dose that is reassuringly low but requires many cohorts before reaching biologically meaningful exposure may provide limited information in patients with serious disease, while rapid escalation can be difficult to justify when pharmacology is steep, human translation is uncertain or the consequences of overshooting could be substantial.
The same translational understanding should therefore inform escalation increments, dosing interval, pharmacodynamic measurements, stopping criteria and the timing of decisions between cohorts. FDA’s 2026 draft QSP/MABEL guidance similarly emphasises examining the predicted dose-response around the starting dose and updating models as human PK and PD data accumulate. The recently finalised ICH M15 guideline on model-informed drug development also provides a broader framework for planning, evaluating and documenting model-based evidence in drug development (ICH M15, final guidance, June 2026).
This is where model-informed drug development is most useful. The model is a quantitative representation of the programme’s current understanding, and the first human data provide an opportunity to challenge and improve that representation. Before entering the clinic, I would want to understand the relationship between administered dose and the relevant biological effect, the safety constraints and where they arise, the assumptions used in animal-to-human translation, the uncertainty around those assumptions, the biomarkers and assays that can test them, and the rules by which emerging clinical data will change the next dosing decision.
A first-in-human dose is one number in the protocol, but the quality of the development strategy sits in the reasoning behind that number. For mRNA therapeutics, that reasoning has to connect delivery, intracellular expression, encoded-protein pharmacology, safety and the measurements that will tell us whether the human biology is behaving as predicted.