First-in-Human Starting Doses for Immune-Cell-Depleting and T-Cell-Engaging Antibodies Outside Oncology

First-in-human starting doses for immune-cell-depleting and T-cell-engaging antibodies outside oncology: what sponsors argued and what regulators accepted

Key takeaways

• For the approved non-oncology depleting antibodies (ocrelizumab, ofatumumab, ublituximab, inebilizumab, alemtuzumab, teplizumab), FDA and EMA marketing-application reviews almost never reconstruct a MABEL, receptor-occupancy or NOAEL/HED derivation of the very first human dose. The first human dose usually came from an earlier program: oncology, transplant, or a patient study in another autoimmune disease.
• Three pathways recur. The first is an inherited oncology or transplant dose (alemtuzumab, ublituximab, ofatumumab, teplizumab). The second is a low, weight-based single ascending dose in patients, escalated on B-cell depletion (inebilizumab, ocrelizumab). The third is a ramp-up or split first dose to control first-exposure cytokine and infusion reactions (teplizumab, ocrelizumab, ublituximab, ofatumumab).
• At licensure, regulators judged dose justification mainly on pharmacodynamics (the depth and duration of B- or T-cell depletion, and target-mediated disposition) and on chronic-toxicology margins. They did not look for a formal FIH calculation. They accepted PD-anchored regimens even when dose-response data were thin, and they recorded their reservations.
• No CD3 T-cell engager is yet approved outside oncology. The current non-oncology trials are in patients, use priming or step-up doses, and in some cases use affinity-attenuated CD3 arms. The regulatory benchmarks are the FDA and EMA MABEL/PAD frameworks and the oncology T-cell engager precedents.

The regulatory frameworks sponsors are arguing against

FDA

FDA's default algorithm takes the NOAEL in each species, converts it to a human equivalent dose (HED), picks the most appropriate species, and divides by a safety factor to derive the maximum recommended starting dose (MRSD) ⁴⁹⁵². The default safety factor is 10. It should be increased for steep dose-response curves, severe or unmonitorable toxicities, novel targets, and animal models of limited utility ⁵¹⁵³. FDA also expects sponsors to compare the MRSD with a pharmacologically active dose (PAD). The guidance specifically names monoclonal antibodies as a class where exaggerated pharmacology can make the PAD more sensitive than the NOAEL ⁴⁴⁴⁷.

For biologics that stimulate immune responses directly or indirectly, FDA states that a MABEL- or PAD-based starting dose may be more appropriate than a toxicology-based dose. Cytokine-release assay results can shape the starting dose, the escalation scheme and the stopping rules ⁶⁸. FDA defines MABEL as the dose expected to produce a minimal biological effect in humans, estimated mainly from human-cell pharmacology and receptor binding with PK integrated ⁶⁰.

For bispecifics, the 2021 Bispecific Antibody Development Programs guidance says a MABEL approach should be considered for agonistic constructs. It cites a retrospective analysis of 17 CD3 bispecifics that found an FIH dose at 10 to 30% pharmacologic activity to be acceptable ⁶¹. The same guidance warns that healthy volunteers may be inappropriate because of immunogenicity and toxicity risk ⁶¹. FDA's monoclonal antibody guidance adds that the decision to use healthy volunteers depends on the antibody, the target antigen and the intended use ⁶⁴².

The June 2026 draft guidance on quantitative systems pharmacology (QSP)-based MABEL dose selection names in vitro human-cell receptor occupancy, receptor activation and cytokine-release data as model inputs ⁶¹⁷. It sets no universal receptor-occupancy threshold ⁶¹⁷. Where QSP and traditional estimates disagree for a novel target, sponsors should choose the lowest estimate ⁵⁹. A steep predicted low-dose response supports a lower start and more cautious escalation ⁶²⁰. The draft also says that in life-threatening diseases a starting dose above the absolute MABEL may sometimes be justified to avoid prolonged subtherapeutic exposure ⁵⁹. Most autoimmune indications will not have that option.

ICH

ICH M3(R2) treats the NOAEL in the most appropriate species as generally the most important input to the starting dose, weighed alongside PD and molecule-specific factors ⁵⁹⁶. ICH S6(R1) requires pharmacologically relevant species. Where none exists, it points to homologous molecules or human-target transgenic models, and notes that these are generally not useful for quantitative risk assessment ⁶⁰⁹⁵⁹⁸. MABEL is named explicitly in ICH S9 for immune-agonistic biopharmaceuticals ⁵⁹⁷. S9 is an oncology guideline, so non-oncology sponsors rely on the FDA and EMA FIH documents instead.

EMA

The EMA FIH guideline asks sponsors to estimate NOAEL-based human exposure, MABEL, PAD and anticipated therapeutic dose. Depending on uncertainty, the starting dose should be related to the MABEL, PAD or NOAEL ⁸¹. MABEL and PAD estimates should draw on in vitro target binding and receptor occupancy in human and animal target cells, integrated by modelling where possible ¹. In healthy volunteers, the starting exposure should ordinarily be below the PAD ¹. Safety factors should reflect novelty, the shape of the dose-response curve, the relevance of the animal model and how easily the toxicity can be monitored ¹. The guideline also stresses that toxicity from exaggerated pharmacology must not be discounted ¹².

On conduct, it requires sentinel dosing in each cohort ⁶²². It requires pre-specified maximum dose increments, which should be smaller when curves are steep or toxicity may not be detectable before it becomes serious ⁶²⁸. The choice between healthy volunteers and patients must be justified case by case ⁶²⁷.

What sponsors did for approved non-oncology depleting antibodies

Inherited oncology doses: ublituximab, ofatumumab, alemtuzumab

Ublituximab. The first human study was a dose escalation in haematologic malignancies. It started at 450 mg and escalated through 600, 900 and 1,200 mg ⁹⁶. The EPAR records no MABEL, receptor-occupancy or NOAEL/HED derivation of that starting dose. The only assessor comment on the study was that it did not appear to support dose proportionality from 450 to 1,200 mg ⁹⁶. The MS regimen came instead from the Phase IIa RMS201 program. All cohorts in that program achieved rapid, deep CD19+ depletion (95.8% reached at least 95% depletion by Week 4), so the study did not discriminate among regimens ³⁹³³⁹². CHMP asked why the exact 4 h/1 h/1 h infusion scheme had never been tested as a single Phase II regimen. It accepted the answer that the regimen matched Phase III and most closely resembled Cohort 2 ⁹⁷.

Ofatumumab. The first program was in CLL, starting at 300 mg IV followed by 2,000 mg doses ¹⁴⁰. The MS dose was built from PD rather than toxicology. Exposure-PD analyses linked maximal MRI suppression to mean CD19+ counts of 8 cells/µL or fewer ⁴¹⁰. A K-PD model predicted that three weekly 20 mg loading doses would reach that target in more than 95% of patients ⁴⁰⁷⁴⁰⁸. Target-mediated drug disposition supported front-loading, because clearance falls as B cells are depleted ⁴¹⁵⁴⁴⁵. CHMP said the rationale was "not entirely clear" and "not entirely convincing", since the MIRROR study did not separate doses clinically. It nevertheless accepted 20 mg as the lowest dose that reliably depleted B cells, noting its better tolerability ⁴¹⁴.

The two agencies reached different conclusions on toxicology. EMA agreed that 100 mg/kg was the NOAEL in the 7-month monkey study ⁷⁶ and cited AUC margins of at least 570-fold from the bridging studies ⁷². FDA's MS review said no clear NOAEL was identified in that study because of deaths from infection and haemolytic anaemia ¹³¹¹³³.

Alemtuzumab. There was no formal MS dose finding. Early MS dosing was empirical and drew on B-CLL and rheumatology experience ³⁷⁶. CHMP noted that B-CLL cumulative exposure was more than tenfold higher. It added that a lower dose cannot be assumed safer for a biological immunomodulator, since the autoimmune effects seen in MS had not been predicted by the oncology experience ³⁷⁹. CAMMS223 bracketed the 20 mg/day pilot dose with 12 and 24 mg/day. CHMP accepted that design and preferred 12 mg/day because efficacy was comparable and tolerability better ^376388380.

The oncology heritage is visible in FDA's Campath review. Clinical exploration covered 2.5 to 240 mg across several schedules ²⁶². In cynomolgus monkeys, 1 and 3 mg/kg IV produced lymphocytopenia and 0.1 mg/kg did not. Alemtuzumab bound monkey CD52 with roughly 16-fold lower affinity than human CD52 ²⁵⁰. That affinity gap is the kind of species difference that argues for a MABEL approach in any new program.

Predecessor-molecule heritage and cytokine-driven ramp-up: teplizumab

Teplizumab is a humanized, Fc-modified (Leu234Ala/Leu235Ala) successor to OKT3. The Fc changes were intended to reduce the Fc-receptor-mediated cytokine release that made OKT3's key toxicity cytokine-release syndrome ²⁹¹²⁷. Early renal-transplant regimens copied OKT3's 10 to 14 days of daily IV dosing ²⁷. The EPAR does not identify the actual first human dose ²⁷.

Cytokine release shaped the dose that followed. In psoriatic arthritis, toxicity led to a ramp-up period, and one participant developed CRS with raised liver enzymes after a first 1 mg dose ²⁷. Nonclinical data were consistent with reduced but not eliminated cytokine release: it occurred in chimpanzees, and the Fc modification did not remove it in a mouse surrogate system ³⁶⁹. In a later study, the sponsor targeted the steady-state concentration seen in earlier work, about 99 ng/mL on Days 10 to 12. When a 25% dose increase was followed by grade 4 hyperbilirubinaemia, the lower regimen was kept and two ramp-up days were added ³⁶⁵.

The regimen FDA's clinical pharmacology review assessed, the one used in the TN-10 trial, is 51, 103, 207 and 413 µg/m² on Days 1 to 4, then 826 µg/m² daily, for a total of about 9 mg/m². FDA described the first four doses (under 10% of the total) as a precaution against cytokine-release reactions and found the regimen appropriate ²⁹²²⁹³. The current label's Stage 2 regimen uses different numbers: 65, 125, 250 and 500 µg/m² on Days 1 to 4, then 1,030 µg/m² on Days 5 to 14 ⁶⁴³. The label (revised 06/2026) states that total exposure in TN-10 was comparable to exposure at this recommended dosage ⁶⁴⁴. Clinical PK/PD modelling estimated maximum CD3 occupancy of 86.8% and an EC50 of 117 ng/mL ³¹. These were post hoc characterizations, not the basis of an FIH calculation. EMA assessors judged the dose-development overview adequate ²⁵. They considered CRS an important identified risk that can be mitigated by the regimen, premedication and monitoring ³⁶⁸.

Low-dose patient SAD escalated on B-cell depletion: inebilizumab and ocrelizumab

Inebilizumab is the clearest non-oncology example of a conventional patient SAD. The Phase 1 systemic sclerosis study (MI-CP200) gave single IV doses of 0.1, 0.3, 1, 3 and 10 mg/kg. Median maximal B-cell reduction rose with dose: 89.8% at 0.3 mg/kg and 99.7% at 10 mg/kg ²⁹⁶. After the first 3 mg/kg participant had a moderate infusion reaction, premedication was added ²⁹⁶. The public reviews do not state how 0.1 mg/kg was derived.

Human CD19 is not bound in standard species, so the huCD19 transgenic mouse was the only relevant model ³⁰⁸⁸⁴. EMA recorded that the applicant converted the 6-month mouse NOAEL of 30 mg/kg/week to an HED using FDA's allometric factor. The resulting HED-based margin was 6.9-fold and the exposure-based margin 55-fold, and assessors endorsed both ⁸³⁸⁴. FDA's review, by contrast, reported that B-cell depletion occurred at every repeat-dose level and that no no-effect level was found for offspring immunotoxicity ^297308317. FDA concluded the findings were expected pharmacology and could be handled in labeling ³⁰²³¹⁰.

The 300 mg Day 1/Day 15 dose was predicted from systemic sclerosis PK/PD to deplete B cells fully for 28 weeks ³⁴⁹. NMOSD exposure-response analyses placed it on the efficacy plateau ³⁵⁴³⁵⁵.

Ocrelizumab's early RA study (ACT2847g) gave two infusions 14 days apart, totalling 20 mg (2 × 10 mg) up to 2,000 mg. B-cell depletion occurred at every dose, and depletion and repletion plateaued at about 400 mg total ³³⁰³³⁵. In monkeys, two doses of 0.05 mg/kg already produced partial depletion ²⁷³. Because RA efficacy was best at 1,000 mg despite full depletion at lower doses, the sponsor took 600 and 2,000 mg into MS Phase II. It chose 600 mg when MRI outcomes did not differ ³²⁴³²⁸.

FDA accepted 600 mg as reasonable but recorded that it had encouraged testing a lower dose. Because Phase III studied only one dose, exposure-response relationships could not be established ³²⁶³²⁸. The split first dose (300 mg × 2) was justified partly as a way to reduce dose-dependent first-infusion reactions and early anti-drug antibodies ³²³³²⁵.

What regulators accepted, and where they pushed back

• Pharmacodynamic depletion as the dose anchor. Both agencies accepted B-cell depletion targets and depletion-plateau arguments as the main dose rationale for anti-CD20 and anti-CD19 products ^280324414349. FDA cautioned that equal peripheral depletion did not guarantee equal clinical outcomes ³²⁴³³⁰.
• Toxicology when pharmacology is the toxicity. Where depletion occurred at every dose, the agencies sometimes differed on whether a NOAEL existed (ofatumumab ⁷⁶¹³¹; inebilizumab ⁸⁴³⁰⁸). FDA also found that no NOAEL or safety margin could be set in ublituximab's single-dose pregnant-monkey study ¹⁰⁵¹¹⁶. This matters for FIH planning. If the pharmacologic effect is the adverse effect, a NOAEL-based MRSD may be uninformative. FDA and EMA guidance then point to PAD- or MABEL-based starting doses ⁴⁴⁴⁷¹.
• Inherited doses need bridging, not re-derivation. For ublituximab, alemtuzumab and ofatumumab, regulators did not ask for a new MABEL calculation for the first autoimmune exposure. They did challenge whether oncology experience predicted autoimmune-population safety ³⁷⁹. They also questioned regimens that had not been tested exactly as proposed ⁹⁷.
• First-exposure mitigation is expected. Ramp-up (teplizumab), split first doses (ocrelizumab), a lower first infusion given over a longer time (ublituximab) and weekly loading at a tolerable dose (ofatumumab) were all accepted as ways to manage first-dose cytokine and infusion reactions ^29232397414.

T-cell engagers outside oncology: where the evidence stands

No CD3 bispecific has yet been approved for a non-oncology indication, so there is no public review of an autoimmune FIH starting dose. ClinicalTrials.gov shows several design patterns in current programs:

• Priming and step-up dosing. Examples are the CD19×CD3×CD28 trispecific CC312, with a lower priming dose before therapeutic dosing in a 3+3 escalation across autoimmune diseases ⁴⁷¹, and GSK5926371, a CD19/CD20 T-cell engager whose Part 1 is designed to identify a suitable priming dose ⁶³⁷. The GSK protocol requires eligibility for corticosteroid, antihistamine and antipyretic prophylaxis and permits tocilizumab rescue ⁶⁴⁰.
• Affinity-engineered CD3 arms. GB261 (CD20×CD3) was designed with very low CD3 affinity and high CD20 affinity to reduce CRS risk in refractory SLE ⁴⁶⁸.
• Patient-only populations. Programs include BCMA×CD3 agents in SLE (F182112), autoimmune bullous disease and ITP (CM336), warm autoimmune haemolytic anaemia (enatumab) and transplant desensitization (OM336) ^{465472474473466}. There are also repurposed oncology T-cell engagers: mosunetuzumab in SLE and lupus nephritis, and teclistamab in anti-MDA5 interstitial lung disease ⁶³⁵⁶¹⁶. The registry records do not disclose numerical starting doses or their derivation ^465635616.

The oncology T-cell engager reviews show what regulators have accepted as evidence:

• Glofitamab. The sponsor translated a monkey PK/PD model of IL-6 excursions to humans and kept predicted IL-6 below a predefined threshold ⁵⁶⁷. FDA then confirmed the 2.5 mg starting (step-up) dose, using an exposure-response analysis that linked Day 1 CD20 receptor occupancy to grade 2 or higher CRS ^566568569.
• Tebentafusp. There was no pharmacologically relevant species ⁵⁸⁵. The review cites an implied MABEL of about 1 pM, based on in vitro melanocyte reactivity at 10 to 100 times MABEL ⁵⁷¹. IFN-γ EC50 values ranged from 0.89 to 121 pM ⁵⁷²⁵⁸⁵.
• Epcoritamab. EMA noted that the nonclinical data did not directly support the human regimen. The monkey data did suggest that a low starting dose followed by an intermediate dose could mitigate severe cytokine release ⁴⁸⁴.

A non-oncology sponsor should be cautious about borrowing these precedents. The QSP draft's allowance for starting above the absolute MABEL applies to life-threatening diseases ⁵⁹, and EMA ordinarily expects healthy-volunteer exposure to stay below the PAD ¹. For a CD3 engager in a chronic autoimmune disease, the defensible position is a human-cell MABEL (or the 10 to 30% pharmacologic-activity benchmark FDA cites for CD3 bispecifics ⁶¹). That estimate should be cross-checked against cytokine-release assays and, where available, receptor-occupancy modelling ⁶¹⁷. The first study should enroll patients rather than healthy volunteers ⁶¹⁶²⁷ and use sentinel dosing and small increments on the steep part of the curve ⁶²²⁶²⁸.

Practical implications for FIH packages

1. State which anchor drives the dose. Regulators accept MABEL, PAD or NOAEL anchors. They expect the protocol and Investigator's Brochure to explain the choice and any safety factor ¹⁵¹.
2. Do not rely on a NOAEL when depletion is the toxicity. Precedent shows that agencies may disagree on whether a NOAEL exists for depleting antibodies ⁷⁶¹³¹. Build a PD- or MABEL-based estimate alongside it ⁴⁴⁶⁸.
3. Bridge inherited doses explicitly. If the first autoimmune exposure uses a dose from oncology or transplant, address differences in population, cumulative exposure and immune-mediated toxicity. CHMP raised exactly this point for alemtuzumab ³⁷⁹.
4. Plan the first-exposure mitigation into the regimen. Ramp-up, split or priming doses and premedication were accepted for every approved product discussed here ^29232397414. They are standard in current autoimmune T-cell engager protocols ⁴⁷¹⁶³⁷.
5. Use modelling for the steep low-dose region. FDA's 2026 QSP draft explicitly asks for simulations around the proposed starting dose and a lower start when the predicted response is steep ⁶²⁰.