References for the TCE-TMDD working document

What to read, in what order, and what has actually been checked

The reading queue and source list for the T-cell engager TMDD project, with a status marker on every entry recording whether the claim drawn from it has been verified against its source.

Two sources have been read in full and every transcribed number is unverified against a printed table. Pearce et al. and Kovalenko et al. were read as open-access full text on 2026-09-11. The teclistamab, mosunetuzumab and elranatamab parameters came from nlmixr2lib model files, which are a second-hand but heavily annotated source: each one cites the table it was taken from, and none of those citations has been followed here. The epcoritamab, talquetamab, glofitamab and blinatumomab entries are characterizations from abstracts and label text. The odronextamab simulation in Section 3 of the specification reproduces a baseline total clearance of 7.1 L/day where the paper reports a population median of 5.4 L/day, and that discrepancy is unresolved.

The working document is the specification, the parameter survey is PK parameters, and the folder’s documents are listed on the project index.

Status markers

• ❌ Not checked. No claim in the working documents has been verified against this source. Where a working document characterizes it, the characterization comes from an abstract or a search result.
• ⚠️ Transcribed, unverified. A number, table or model structure in a working document came from this source but has not been checked against the source’s own printed table.
• ✅ Checked. Verified against the source.

On a reading-queue entry, ✅ means the question posed against that entry has been answered, by reading the paper or by deciding it is out of scope. On a source entry below the queue, ✅ keeps the stricter meaning: the claim drawn from it has been checked against the printed table. Two source entries are ✅ for having been read; none is ✅ for having its numbers checked, which is why every one of them carries ⚠️ instead.

How to read these

The general method is on the Reading Papers page. Read against the question under each entry rather than through the paper front to back.

Two of the entries need the full hour, and for opposite reasons. Kovalenko et al. needs it because its parameter table is the input to every simulation in the specification and a transcription error there propagates everywhere. Miao et al. needs it because of one sentence: the authors tested a quasi-steady-state TMDD parameterization and a time-varying soluble BCMA covariate and rejected both. Why those failed is the strongest available evidence about whether the empirical decaying-clearance form is a convenience or a necessity, and it decides whether Section 4 of the specification is describing a modelling habit or a data limitation.

Watch for the answer that runs the other way. If Miao et al. rejected the quasi-steady-state form on numerical grounds rather than on fit, then the class-wide preference for time-dependent clearance is an estimation artifact, Section 4’s verdict softens from “the sink is destroyed” to “the sink is destroyed and nobody could identify the concentration dependence anyway”, and epcoritamab becomes the only entry in this project that saw the mechanism directly.

Notes go one line per question, carrying an answer and the table or page number it came from, plus anything that contradicts the working document.

Reading queue

Ordered so that each entry makes the next easier to read.

1. Pearce et al. 2025 — predicting TCE pharmacokinetics as a function of TMDD ✅

CPT: Pharmacometrics & Systems Pharmacology. PMC12597969. AstraZeneca authors. Open access.

Read 2026-09-11. A model in which TCE half-life is the sum of an intrinsic clearance, a CD3-mediated clearance and a tumor-associated-antigen-mediated clearance, applied to 29 TCEs. Predicts human half-life within two-fold for 16 of 18, against 10 of 17 for allometric scaling alone.

The sentence the whole project turns on, in Section 3.2: at doses below 0.1 mg/kg the pharmacokinetics are essentially linear, which they identify as the fourth phase of TMDD in the Peletier and Gabrielsson classification. This is explanation 1 of the specification, stated generically rather than for one drug.

Still open. Their Figure 2 simulation, half-life against CD3 affinity from 1000 nM to 1 nM at 0.01 mg/kg, is reproducible from the equations in their methods and has not been reproduced. Their generic parameters assume blood T-cell counts throughout, which is the assumption Section 7 of the specification questions.

2. Kovalenko et al. 2026 — odronextamab population pharmacokinetics ✅

CPT: Pharmacometrics & Systems Pharmacology. PMC12823311. Open access.

Read 2026-09-11. 507 patients, 14,618 observations, doses 0.03 to 320 mg. Two-compartment with parallel first-order and Michaelis-Menten elimination, where the maximum velocity declines exponentially toward an asymptote. Table 2 carries the full parameter set with bootstrap confidence intervals.

Why it leads the source list. It is the only published TCE model that reports a fitted Michaelis constant, so it is the only one that can answer “is the priming dose above or below the kink” with a number.

Open, and it is the first item in the order of work. The specification’s simulation gives a baseline total clearance of 7.1 L/day for the reference patient; the paper reports a population median of 5.43 L/day for DLBCL and 4.67 L/day for follicular lymphoma. Covariate effects and between-subject variability on \(V_{\max,0}\) (variance 0.403) plausibly account for the gap, and that has not been demonstrated.

3. Epcoritamab population pharmacokinetics, 2025 ❌

Clinical Pharmacokinetics. doi:10.1007/s40262-024-01464-2 · PubMed 39708278. Paywalled; the publisher redirected an unauthenticated fetch.

The highest-value unread entry. The only TCE model published with a quasi-steady-state TMDD structure in its own declared terms. Three questions against it.

1. What are \(K_{ss}\), \(R_{\rm tot}\) (or \(R_{\max}\)) and \(k_{\rm int}\), in what units, and how do they compare with the CD3 pool computed in Section 7 of the specification?
2. Is \(R_{\rm tot}\) constant or does it carry its own turnover? If it is constant, epcoritamab’s model disagrees with every other model in this project about whether the sink is destroyed.
3. What is \(K_a\), and what is the elimination half-life it should be compared against, so that the flip-flop table in Section 5 gains a third row?

4. Miao et al. 2023 — teclistamab population pharmacokinetics ⚠️

Target Oncol 2023;18(5):667-684. doi:10.1007/s11523-023-00989-z · PMC10518021.

Parameters transcribed from nlmixr2lib’s Miao_2023_teclistamab, which cites Table 2 and its footnotes a to d. Not checked against the paper.

The question. The authors tested a quasi-steady-state TMDD parameterization and a time-varying soluble BCMA covariate and rejected both, in favour of \(CL(t)=CL_1+CL_2e^{-K_{DES}t}\). On what grounds? See the warning above about the answer running the other way.

A second question while there. \(Q\) = 0.0390 L/day with a 55.5% relative standard error makes the peripheral compartment nearly invisible, and the two-compartment terminal slope computed from the printed parameters is about 26 days, against a reported half-life of 3.8 days. Which quantity does the paper call the half-life?

5. Talquetamab recommended phase II doses, 2025 ❌

Clinical Pharmacology & Therapeutics. PMC12439005. Open access; the parameter table is in the supplement, which has not been retrieved.

The question. The supplementary population pharmacokinetic table, for the absorption and time-dependent clearance columns of PK parameters. A second one: baseline total T-cell count was tested as a covariate and rejected, which is evidence against CD3-mediated clearance being dominant at the recommended doses. On what was it tested, and with what power?

6. Bender et al. 2024 — mosunetuzumab population pharmacokinetics ⚠️

Clin Transl Sci 2024;17(5):e13825. doi:10.1111/cts.13825.

Parameters transcribed from nlmixr2lib’s Bender_2024_mosunetuzumab, which cites Table 2 and reproduces the NONMEM control stream from Table S3. Not checked against the paper.

The question. The competitive CD20 receptor-occupancy submodel carries dissociation constants for mosunetuzumab, rituximab and obinutuzumab. Those are binding constants for the same target that the time-dependent clearance is supposed to represent, so the model already contains the ingredients of a mechanistic TMDD term and does not use them that way. Did the authors say why?

7. Zhu et al. — blinatumomab clinical pharmacology ❌

Clinical Pharmacokinetics. Cited from the CD20×CD3 review; full reference not retrieved.

Read only to confirm the format contrast: 54 kDa, no Fc domain, continuous infusion, 2.11-hour half-life. Low priority, because blinatumomab is in the parameter table as a contrast rather than as a comparable entry.

Sources

Everything the working documents draw on, whether or not it is in the queue.

Population pharmacokinetic models

• ⚠️ Kovalenko et al. 2026, odronextamab. Queue entry 2. Source of every parameter in the specification’s Sections 3 and 4 figures.
• ⚠️ Miao et al. 2023, teclistamab. Queue entry 4. Via nlmixr2lib.
• ⚠️ Bender et al. 2024, mosunetuzumab. Queue entry 6. Via nlmixr2lib.
• ⚠️ Poels et al. 2025, elranatamab. npj Syst Biol Appl 2025;11:102, doi:10.1038/s41540-025-00585-z. Via nlmixr2lib’s Poels_2025_elranatamab_qsp, whose pharmacokinetic parameters are a preliminary population analysis fixed inside a systems model, cited to its Supplementary Table 3. The CD3 and BCMA binding constants in the specification’s Section 5 and the parameter page’s Section 3.4 come from here. A preliminary analysis fixed inside somebody else’s model is the weakest parameter source in this project.
• ❌ Epcoritamab, 2025. Queue entry 3.
• ❌ Talquetamab, 2025. Queue entry 5.
• ❌ Djebli et al. 2020, glofitamab. Blood 2020;136(Suppl 1):1. Conference abstract. Characterized in the specification’s Section 4 as a linear model with no time-varying term, from the abstract text. That characterization carries the weight of the “remove the target and the time dependence goes away” argument and rests on one sentence in an abstract.
• ❌ Blinatumomab label. BLINCYTO US prescribing information. Clearance, half-life and \(V_z\) in the parameter page’s Section 3.8.

Mechanism and review

• ⚠️ Pearce et al. 2025. Queue entry 1. Source of the half-life table in the specification’s Section 1, the CD3 expression and internalization values in Section 7, and the fourth-phase argument in Section 3.
• ❌ Peletier and Gabrielsson, the four-phase classification of TMDD pharmacokinetics. Cited through Pearce et al. rather than read. The primary reference has not been located, and the specification’s Section 3 attributes the classification to them on Pearce’s authority alone. Use it for vocabulary only. The 2012 work derives its phases for a one-compartment model with no distribution phase and no peripheral compartment. Every model in this project is two-compartment, where a bend in a log-concentration profile can be distribution rather than target saturation, and Section 9.5 of the specification shows the two-compartment version reversing the conclusion. Locating and reading the primary source, and checking whether a two-compartment treatment exists, is now ahead of several entries in the queue above.
• ⚠️ CD20×CD3 bispecific antibodies in B-NHL, 2025. Clin Transl Sci, PMC12139688. Read 2026-09-11 for the preclinical TMDD observations and the obinutuzumab pretreatment rationale for glofitamab. Its Table 1 records TMDD in cynomolgus monkey at doses at or below 0.1 mg/kg for three of the four CD20×CD3 drugs, which is the preclinical counterpart of the specification’s explanation 1.

Software and libraries

• ✅ nlmixr2lib, nlmixr2.github.io/nlmixr2lib, version 0.3.2.9000. Cloned and searched on 2026-09-11; the model counts and the present/absent lists in the parameter page’s Section 4 were produced by that search and are checked. Maintainer Bill Denney.
• ✅ rxode2, used for the odronextamab simulations.

Where to start

If one hour is available, read Kovalenko et al. Table 2 against the specification’s Section 3, and resolve the 7.1 versus 5.4 L/day discrepancy. Everything downstream depends on those eight numbers being right.

If a second hour is available, read Miao et al. on why the quasi-steady-state parameterization was rejected. That answer decides whether this project is describing pharmacology or estimation practice.