Key points#
- Starting-dose selection combines toxicology, toxicokinetics, pharmacology, product quality, species relevance, and a prediction of human concentrations.
- FDA's classic maximum recommended starting dose method starts with an animal no-observed-adverse-effect level, converts it to a human equivalent dose, and applies a safety factor.
- A pharmacology-based estimate such as the minimum anticipated biological effect level can be lower and more protective for potent, novel, or human-specific mechanisms.
- The chosen dose must fit the people enrolled, route and rate of administration, escalation plan, maximum planned concentration, monitoring, and stopping rules.
- Every important assumption should be traceable and tested against the clinical data as they arrive.
The decision begins by defining what could go wrong#
The first administered dose is often described as a conversion problem: take an animal dose, scale it to a human, then divide. That calculation can be useful, but it is not the decision itself. Before you choose a number, you need a product-specific account of uncertainty.
The questions you have to answer include:
- What does the target do in healthy and diseased tissue?
- Is target binding reversible, saturable, long lasting, or capable of amplification?
- Are the relevant target, immune pathways, and metabolites represented in the test species?
- Which adverse findings appeared, at what concentrations, in which organs, and did they reverse?
- Could the manufacturing process create variants, impurities, aggregates, or potency variation that alter risk?
- What human pharmacokinetics are predicted, and how uncertain are those predictions?
- Would healthy volunteers receive no plausible benefit while carrying a mechanism-related risk that patients might accept differently?
EMA's first-in-human guideline treats this as an integrated risk assessment. The analysis should connect product quality, nonclinical evidence, participant selection, clinical monitoring, and escalation. A calculation that is internally correct can still be inappropriate if it rests on an irrelevant species or ignores a potent human biological effect.
The NOAEL route produces a ceiling, not a command#
FDA's 2005 guidance describes a structured route for many systemically administered therapeutics first studied in adult healthy volunteers, and it begins with the no-observed-adverse-effect level, or NOAEL, from a suitable animal toxicology study. A NOAEL is the highest tested dose at which the study did not identify an adverse effect under its design. It is not proof of zero harm. It depends on what the study measured, its duration, and sample size. It also depends on dosing route and species.
The selected animal NOAEL is converted to a human equivalent dose. FDA's default conversion commonly uses body surface area, represented by species-specific Km factors, rather than copying the animal milligram-per-kilogram dose directly. In simplified form:
human equivalent dose = animal NOAEL × (animal Km / human Km)
A safety factor is then applied:
maximum recommended starting dose = human equivalent dose / safety factor
Ten is a common default safety factor in the guidance, but it is neither a universal guarantee nor a fixed entitlement; greater uncertainty can support a larger divisor and therefore a lower dose. Factors include steep dose-response behavior, severe or irreversible toxicity, and uncertain species relevance. They include variable bioavailability, nonlinear kinetics, a novel mechanism, or weak ability to monitor early injury. Strong, coherent data can affect the rationale in the other direction, but any departure needs a documented scientific basis.
The result is called a maximum recommended starting dose, or MRSD. The word “maximum” matters: the method identifies a ceiling generated by that route. It does not say that the protocol should begin at that ceiling.
Concentration links the animal finding to the human prediction#
Nominal dose can hide large differences between species, and the more informative bridge often asks what systemic concentration accompanied the adverse finding and what concentration the proposed human dose is expected to produce.
Toxicokinetic measurements such as maximum concentration and area under the concentration-time curve help make that bridge. The comparison may need to include active metabolites, free rather than total drug, and tissue distribution. It may need to include accumulation after repeated dosing, route-specific delivery, and nonlinear clearance. For an infused product, the rate of administration can change peak concentration even when total dose is unchanged; for a long-acting product, delayed accumulation may matter more than the first peak.
A useful dossier does not present one precise human prediction as fact. It examines a plausible range, identifies the inputs driving that range, and asks what happens under conservative alternatives. The trial's planned upper concentration should also be bounded. A protocol should not continue through a preset dose sequence after observed human concentrations approach a pharmacological or toxicological limit that the plan was meant to avoid.
Pharmacology may set the lower boundary#
Some products can produce meaningful biology at concentrations far below those associated with animal toxicity. A minimum anticipated biological effect level, or MABEL, estimates where a biological response could first become detectable in humans. Inputs can include human target affinity, receptor occupancy, and concentration-response curves. They can include cell-based studies, target abundance, and relevant animal pharmacology. They can include predicted human kinetics and the relationship among those pieces.
MABEL is not one standardized equation. It is a model-based synthesis whose credibility depends on the model and data. Teams should examine multiple plausible models and parameter assumptions, particularly when signaling can amplify, binding is prolonged, or different effects begin at different concentrations. The most sensitive relevant biological effect may be more informative than the most convenient assay.
Compare a toxicology-derived MRSD with pharmacology-based estimates even when one method will ultimately control your choice. A large gap shows which assumption is doing the work. If the MABEL sits well below the MRSD, the lower estimate may be the prudent starting point. Other approaches can also matter for modalities not well served by the adult healthy-volunteer algorithm, including some endogenous substances, vaccines, cell or gene therapies, and oncology products.
FDA issued a draft guidance in June 2026 on using quantitative systems pharmacology to support MABEL dose selection, but as of July 15, 2026, that document is for comment and explicitly not for implementation. It signals a direction of regulatory thinking, but it should not be described as a final requirement.
A worked reasoning example#
Consider a hypothetical molecule whose relevant-species toxicology study identifies a NOAEL. Surface-area conversion and a safety factor produce an MRSD of 2 milligrams. Human-cell pharmacology, however, predicts 10 percent target occupancy at 0.2 milligrams and a steep rise beyond that point. The model also has wide uncertainty because the target is expressed differently across species.
Starting at 2 milligrams merely because the arithmetic permits it would ignore the more sensitive evidence. A defensible plan might select a dose below 0.2 milligrams, confirm that the assay can measure the resulting concentration, dose one sentinel participant, observe for a biologically justified interval, and require review before the rest of the cohort. The exact choice cannot be derived from this example, but the logic is clear: the starting dose follows the most relevant credible risk boundary, not the largest available number.
The protocol completes the safety argument#
Starting low does not substitute for a controlled trial design. EMA's guideline asks sponsors to justify safeguards within and between cohorts. Depending on the product, these can include:
- sequential or sentinel administration within a cohort;
- intervals long enough to detect plausible immediate and delayed effects;
- facilities and staff able to recognize and manage mechanism-related emergencies;
- prespecified review of safety, pharmacokinetics, and pharmacodynamic findings;
- smaller dose increments as the active range approaches;
- limits tied to observed concentration, not dose alone;
- individual, cohort, and study-level stopping criteria;
- rules for restarting after a pause or resolving discordant data;
- separate transition criteria for single-dose, multiple-dose, food-effect, interaction, or patient portions of an integrated protocol.
The observation interval must match the biology. A few hours may be inadequate for a product whose effects develop over days. Likewise, enrolling all participants at once can defeat the purpose of a low starting dose if there is no chance to learn from the first participant before dosing the rest.
The best record is an auditable chain of decisions#
A reviewable dose rationale should state the selected studies and species, the critical findings, the NOAEL and human conversion, the safety factor, pharmacology-based estimates, predicted human concentration range, product-quality uncertainties, participant rationale, and reasons for the final choice. It should also show how those assumptions set escalation increments, concentration limits, sampling, monitoring, and stopping rules.
That record should remain live. Human pharmacokinetic and pharmacodynamic data can invalidate a preclinical prediction after the first cohort, and continuing safely means updating the model and your decision, not defending the number you started with. First-in-human dose selection is therefore a controlled-learning plan whose first number is only its opening decision.
Sources and further reading
- FDA, Estimating the Maximum Safe Starting Dose in Initial Clinical Trials for Therapeutics in Adult Healthy Volunteers, final guidance, July 2005 (accessed 2026-07-15)
- EMA, Strategies to Identify and Mitigate Risks for First-in-Human and Early Clinical Trials, Revision 1, effective February 2018 (accessed 2026-07-15)
- FDA, ICH M3(R2) Nonclinical Safety Studies for the Conduct of Human Clinical Trials and Marketing Authorization, final guidance, January 2010 (accessed 2026-07-15)
- FDA, QSP-Based Dose Selection for MABEL in First-in-Human Trials, draft guidance, June 2026, not for implementation (accessed 2026-07-15)
Questions and answers
Is the maximum recommended starting dose the dose a trial must use?
No. It is an upper result from one toxicology-based method. Pharmacology, uncertainty, product features, and participant risk can justify a lower starting dose.
When can a MABEL approach matter?
It is especially informative when important biological effects may begin well below a toxicology-derived ceiling or when animal models do not represent human target biology well.
Does a low first dose make a first-in-human trial safe?
No. Participant selection, sentinel dosing, observation intervals, monitoring, escalation limits, stopping rules, and real-time review remain part of the safety system.