Elranatamab Population Pharmacokinetics and Exposure–Response for Cytokine Release Syndrome in Patients with Relapsed or Refractory Multiple Myeloma Authors: Jennifer E Hibma, Donald Irby, Alan Liu, Mohamed Elmeliegy, Lindsay E King, David Gifondorwa, Sibo Jiang, Kamrine E Poels, Pooneh Soltantabar, Hoi-Kei Lon, Blerta Shtylla, Diane Wang, Jason H Williams, Timothy Nicholas Published: 2026 ABSTRACT AbstractBackground and ObjectiveA population pharmacokinetics (PopPK) model was developed for elranatamab, a bispecific antibody targeting both B-cell maturation antigen (BCMA) and cluster of differentiation 3 (CD3), to characterize the PopPK of elranatamab and guide clinical development. Elranatamab is approved in several countries for the treatment of relapsed or refractory multiple myeloma.MethodsThe PopPK model incorporated data from four clinical studies and described both free and total elranatamab as well as soluble BCMA (sBCMA) using a semi-mechanistic two-compartment target-binding model. In total there were 13,233 observations from 321 participants who received intravenous (IV) or subcutaneous (SC) elranatamab monotherapy with doses ranging from 0.1 to 1000 µg/kg.ResultsElranatamab exhibited approximately linear PK over the dose range evaluated (i.e., 6–76 mg SC), with limited cellular target-mediated drug disposition and generally proportional exposure, with modest deviations at very high baseline sBCMA. None of the potential covariates evaluated, including age, sex, body weight, baseline sBCMA, and immunogenicity, were found to be clinically significant predictors of elranatamab exposure. Exposure–response analyses identified early peak exposure and tumor burden as key predictors of cytokine release syndrome (CRS), primarily after the first step-up dose. Simulations supported restarting treatment without re-priming after dose interruptions of up to 12 weeks.ConclusionThese findings support the approved fixed-dose regimen and provide a framework for clinical management of elranatamab in relapsed or refractory multiple myeloma.Clinical Trial RegistrationNCT03269136, NCT04798586, NCT04649359, NCT05014412.Supplementary InformationThe online version contains supplementary material available at 10.1007/s40262-026-01663-z. Key Points Total and free elranatamab pharmacokinetics (PK) were well described using a semi-mechanistic two-compartment target-binding model.None of the potential covariates, including body weight, were clinically significant predictors of elranatamab exposure; therefore, fixed dosing of elranatamab is supported.Exposure–response (ER)–cytokine release syndrome (CRS) by dose analyses indicate that CRS was primarily associated with peak concentrations following initial exposure from the first step-up dose of elranatamab.Population PK simulations supported 12-week dose interruptions without re-priming. Introduction Multiple myeloma (MM) is the second most common hematologic cancer characterized by dysregulated proliferation of the bone marrow plasma cells [1, 2]. In MM, abnormal accumulation of cancerous plasma cells can crowd out normal blood cells, forming tumors and leaving a person vulnerable to infection [3]. Standard available treatments consist of several therapeutic classes, including proteasome inhibitors (e.g., bortezomib, carfilzomib), immunomodulatory agents (e.g., lenalidomide, pomalidomide), and monoclonal antibodies (e.g., daratumumab, isatuximab) [4]. Despite targeting different pathways in MM, many patients eventually relapse and become refractory to multiple drug classes, known as relapsed or refractory MM (RRMM), and the disease remains incurable [5]. B-cell maturation antigen (BCMA), which is a transmembrane protein that is widely and selectively overexpressed on malignant plasma cells, has emerged as a validated target for the treatment of RRMM [6]. Elranatamab is a T-cell redirecting bispecific antibody (BsAb) targeting both BCMA and the T-cell co-receptor cluster of differentiation 3 (CD3). BsAbs offer a convenient ‘off-the-shelf’ alternative to cell-based therapies such as chimeric antigen receptor T-cell (CAR-T) therapies [7]. In August 2023, elranatamab received accelerated approval from the US Food and Drug Administration (FDA) for the treatment of adult patients with RRMM based on its high overall response rate (ORR) and the durability of those responses in heavily pre-treated patients [8]. Other BsAbs have also been approved for RRMM, including teclistamab and linvoseltamab, which target BCMA, and talquetamab, which targets GPRC5D. These agents reflect the growing role of T-cell–redirecting therapies in the evolving treatment landscape for RRMM. With the growing interest in BsAbs in oncology, understanding the population pharmacokinetics (PopPK) of elranatamab is important for optimizing dosing regimens and ensuring therapeutic efficacy while minimizing adverse events such as cytokine release syndrome (CRS) [9–11]. During the clinical development, the PopPK of free and total elranatamab provided valuable insights with regards to dose selection, characterization of steady-state kinetics, and support for clinical decision making, such as restarting treatment following dose delays. The objectives of this study were to characterize the PopPK of elranatamab monotherapy in patients with RRMM based on clinical data from four studies [MagnetisMM-1 (ClinicalTrials.gov identifier: NCT03269136), MagnetisMM-2 (NCT04798586), MagnetisMM-3 (NCT04649359), and MagnetisMM-9 (NCT05014412)] and to evaluate exposure–response (ER) relationships for CRS in MagnetisMM-3 among participants who received the two–step-up priming dose regimen to support clinical interpretation of CRS risk. Methods Clinical Studies Included in Population Pharmacokinetics (PK) Model Development and Exposure–Response (ER) Analysis The current PopPK and ER analyses were conducted based on studies MagnetisMM-1, -2, -3, and -9, which varied in design and population but collectively encompassed a range of relevant dosing regimens and clinical settings. MagnetisMM-1 was a first-in-human, phase I dose-escalation and dose-expansion study evaluating intravenous (IV) (1-hour infusion) and subcutaneous (SC) administration of elranatamab, weight-based (µg/kg) and fixed-dose regimens, as well as step-up priming regimens. MagnetisMM-2 assessed safety and tolerability in Japanese patients using a similar SC regimen. MagnetisMM-3 was a phase II registrational study with two cohorts based on prior BCMA-targeted therapy exposure, incorporating a step-up dosing strategy and a switch to biweekly dosing after 24 weeks in responders. MagnetisMM-9 explored an alternative step-up regimen with extended dosing intervals. Detailed study designs and primary outcomes have been previously published [7, 12–14]. Details of the clinical trials, including study descriptions, dosing regimens, participant numbers, and sampling schemes, are provided in Table S1 (see electronic supplementary material [ESM]). The study protocols were approved by local ethics committees or institutional review boards, and all patients provided written informed consent prior to participation. Bioanalytical Methodology Validated electrochemiluminescent (ECL) ligand binding assays were used to determine free and total elranatamab and free (free from elranatamab) shed BCMA (sBCMA) concentration in plasma, and an immunoaffinity liquid chromatography-tandem mass spectrometry (LC-MS/MS) assay was validated for the quantification of total sBCMA (all forms of sBCMA). The lower limit of quantification (LLOQ) values for total elranatamab were 0.5 ng/mL (~0.0034 nM) and 50 ng/mL (~0.34 nM) for free drug. For sBCMA, the LLOQ values were 1 ng/mL (~0.185 nM) for total sBCMA and 0.08 ng/mL (~0.015 nM) for free sBCMA. For the development of the target-binding model, the elranatamab and sBCMA concentrations and dose amounts were converted from ng/mL to molar units (nM) using the molecular weights of elranatamab (148 kilodalton [kDa]) and sBCMA (5.4 kDa). Anti-drug antibodies (ADA) were assessed using a validated ligand binding assay following a multi-tier approach (screening, confirmation, and titer determination) in accordance with FDA and EMA guidance [15, 16]. The assay incorporated drug tolerant procedures to enable detection of ADA in the presence of circulating drug. Software The PopPK analysis was performed using the stochastic approximation expectation-maximization (SAEM) and importance sampling method (IMP) estimation methods in NONMEM version 7.5.0 (ICON, Dublin, Ireland). Perl-speaks-NONMEM (PsN) version 5.2.6 (Uppsala, Sweden), and R software version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria) were used for model evaluation and graphical assessment. The ER analyses were performed using R version R.4.1.3 and PopPK simulations for treatment restart were conducted using mrgsolve v1.41. Population PK Model Development Model development began with a standard two-compartment model with linear clearance, followed by systematic evaluation of nonlinear elimination pathways to account for potential target-mediated or time-dependent clearance. The base structural model used to describe the PK of both free and total elranatamab was a semi-mechanistic model that incorporated the binding dynamics of elranatamab to sBCMA. The CD3 target interaction, while crucial for T-cell activation, was not modeled here due to its relatively lower expression and weaker binding affinity compared with sBCMA. Although a full target-mediated drug disposition (TMDD) framework is theoretically identifiable when both free and total drug and target concentrations are available, exploratory evaluations using a full TMDD parameterization resulted in unstable estimation and high parameter uncertainty. In the absence of CD3 or tissue level BCMA measurements, a reduced TMDD structure focused on sBCMA was considered the most appropriate representation of the observable system. The model was parameterized in terms of the three entities: free elranatamab, free sBCMA, and the elranatamab–sBCMA complex, each with their own clearance and volume of distribution [17]. Binding dynamics were modeled on a molar amount scale, and volumes of distribution were estimated to link model-predicted amounts to observed concentration data. The base model equations are shown below. Residual variability (intraindividual random effect) was modeled using a proportional model. Interindividual variability (IIV) in the PK parameters was modeled using multiplicative exponential random effects. For SAEM-IMP estimation with Mu-referencing, random effect variances estimated to be near zero were fixed to a small positive value (0.025) to maintain numeric stability of the stochastic approximation. Population PK Covariate Analysis and Final Model Selection Based on mechanistic plausibility, exploratory analysis, and scientific interest, the following demographics, baseline laboratory values, disease and treatment-related variables were evaluated as potential covariates: for elranatamab clearance (CL), age, sex, race, baseline body weight, baseline eGFR, baseline immunogenicity, immunogenicity incidence on treatment, baseline sBCMA, and hepatic function (baseline values for albumin, bilirubin, and aspartate aminotransferase [AST]) were tested; for elranatamab central volume (Vc), age, sex, race, baseline body weight, immunogenicity, and baseline sBCMA were assessed; and for absorption rate constant (ka), age and baseline body weight. Potential continuous and categorical covariates were evaluated for correlation and none of the potential covariates in Table 1 were considered highly correlated (r2 > 0.6). Missing baseline continuous covariate values were set to the population median if the proportion of missing values represented < 10% of the analysis population; if > 10%, the covariate was excluded. For categorical covariates, missing values were treated as their own category. Table 1Demographic and baseline characteristics of participants included in the population pharmacokinetic analysis (N = 321)Variable/category1–11–1.1QW1–1.1Q2W1–2A23–A3–B9–19–2ATotalN53613154123643310321Sex, n (%) Male26 (49)2 (33)9 (69)8 (53)3 (75)68 (55)30 (47)14 (42)7 (70)167 (52) Female27 (51)4 (67)4 (31)7 (47)1 (25)55 (45)34 (53)19 (58)3 (30)154 (48)Race, n (%) Black11 (21)2 (33)2 (15)1 (7)09 (7)2 (3)2 (6)029 (9) Asian3 (6)01 (8)2 (13)4 (100)16 (13)1 (2)19 (58)3 (30)49 (15) White36 (68)4 (67)10 (77)10 (67)072 (59)44 (69)11 (33)6 (60)193 (60) Missing3 (6)002 (13)026 (21)17 (27)1 (3)1 (10)50 (16)Baseline ADA, n (%) Negative42 (79)4 (67)10 (77)12 (80)4 (100)102 (83)42 (66)30 (91)6 (60)252 (79) Positive8 (15)2 (33)1 (8)1 (7)011 (9)13 (20)2 (6)2 (20)40 (12) Not evaluable3 (6)02 (15)2 (13)010 (8)9 (14)1 (3)2 (20)29 (9)ADA on-treatment, n (%) Negative42 (79)6 (100)10 (77)13 (87)2 (50)101 (82)52 (81)30 (91)8 (80)264 (82) Positive8 (15)01 (8)02 (50)12 (10)3 (5)2 (6)028 (9) Not evaluable3 (6)02 (15)2 (13)010 (8)9 (14)1 (3)2 (20)29 (9)Tumor burden, n (%) Low or intermediate–––––131 (72)–––– High–––––38 (21)–––– Missing–––––14 (7.7)––––Age (years) Mean (SD)63.55(9.61)59.50(13.31)67.38(7.26)66.73(9.65)64.00(10.03)67.07(9.45)65.69(9.42)60.42(10.14)64.70(8.38)65.27(9.68) Median (range)64.00(46.00–82.00)58.50(42.00–74.00)67.00(53.00–79.00)69.00(50.00–80.00)68.50(49.00–70.00)68.00(36.00–89.00)67.00(41.00–84.00)63.00(36.00–79.00)66.00(49.00–78.00)66.00(36.00–89.00)Body weight (kg) Mean (SD)79.13(20.71)68.83(13.45)73.57(14.93)75.17(19.25)68.78(5.17)74.05(18.62)73.23(16.73)68.19(15.24)75.38(14.04)74.04(17.95) Median (range)75.10(42.00–120.0)64.50(56.00–87.30)71.10(51.30–99.40)67.50(55.20–121.9)70.10(61.40–73.50)72.00(40.90–159.6)69.30(36.50–119.3)66.30(44.50–101.5)79.60(54.50–91.70)71.50(36.50–159.6)eGFR (mL/min/1.73m2) Mean (SD)76.33(20.04)87.84(23.40)77.81(21.01)81.87(21.20)92.52(13.02)75.07(21.98)77.91(24.47)86.67(20.73)92.22(21.28)78.46(22.18) Median (range)74.46(36.58–107.47)94.02(44.58–110.63)79.97(44.01–103.27)92.52(36.29–103.99)95.63(74.12–104.70)74.83(22.63–122.77)81.86(22.50–114.86)96.52(42.09–124.58)97.38(33.66–108.49)81.21(22.50–124.58)Baseline sBCMA (nM) Nm0121400019 Mean (SD)16.24(19.24)38.41(64.72)21.73(23.07)14.31(25.02)NA16.31(30.43)21.24(29.26)8.72(9.10)13.99(14.30)16.90(27.01) Median (range)8.44(0.00–74.63)12.04(2.83–153.5218.70(0.00–81.67)4.19(0.42–95.19)NA7.52(0.00–266.67)10.66(0.00–161.67)5.91(0.41–48.52)6.35(0.67–38.52)8.29(0.00–266.67)Albumin (g/dL) Mean (SD)3.54 (0.64)3.53 (0.39)3.56 (0.71)3.41 (0.53)3.48 (0.26)3.61 (0.54)3.55 (0.53)3.67 (0.54)3.76 (0.68)3.59 (0.56) Median (range)3.70(2.00–4.60)3.40(3.20–4.30)3.70(2.40–4.50)3.30(2.60–4.20)3.50(3.20–3.70)3.60(2.10–4.70)3.50(2.60–4.50)3.70(1.90–4.80)3.75(2.30–4.80)3.60(1.90–4.80)Bilirubin (mg/dL) Mean (SD)0.53 (0.29)0.38 (0.31)0.62 (0.35)0.47 (0.23)0.77 (0.43)0.53 (0.41)0.52 (0.29)0.56 (0.22)0.52 (0.33)0.53 (0.34) Median (range)0.50(0.10–1.40)0.25(0.20–1.00)0.50(0.30–1.40)0.40(0.20–1.00)0.65(0.40–1.40)0.40(0.10–3.80)0.40(0.20–1.40)0.50(0.20–0.90)0.45(0.10–1.10)0.40(0.10–3.80)AST (U/L) Mean (SD)30.96(33.75)33.00(30.92)26.00(10.80)19.47(9.66)19.50(7.85)24.60(12.84)30.20(12.78)22.06(10.48)23.60(10.85)26.39(18.22) Median (range)23.00(9.00–247.00)20.50(12.00–93.00)22.00(13.00–49.00)18.00(8.00–40.00)16.50(14.00–31.00)21.00(10.00–86.00)28.00(12.00–79.00)19.00(12.00–73.00)21.00(10.00–42.00)23.00(8.00–247.00)Tumor burden was only relevant for the ER analysis, which only included participants in MagnetisMM-3: High tumor burden at baseline = plasma cell infiltrate in the bone marrow ≥80% or serum M-spike ≥5 g/dL or serum free light chain ≥5000 mg/L. Low tumor burden at baseline = plasma cell infiltrate in the bone marrow <50% and serum M-spike <3 g/dL and serum free light chain <3000 mg/L. Missing tumor burden at baseline = any or all of the following lab values are missing: plasma cell infiltrate in the bone marrow, serum M-spike, serum free light chain. Patients who did not fit either criterion were categorized as having intermediate tumor burden. Not evaluable: Subjects with only baseline samples and no post-treatment samples, or with assay interference preventing ADA assessmentADA anti-drug antibody, AST aspartate aminotransferase, eGFR estimated glomerular filtration rate, ER Exposure–response, NA not available, Nm number missing, QW every week, Q2W every 2 weeks, sBCMA soluble B-cell maturation antigen, SD standard deviation, – indicates not applicable, 1-1 MagnetisMM-1 Part 1, 1QW MagnetisMM-1 Part 1.1 QW, 1Q2W MagnetisMM-1 Part 1.1 Q2W, 1-2A MagnetisMM-1 Part 2A, 2 MagnetisMM-2, 3A MagnetisMM-3 Cohort A, 3B MagnetisMM-3 Cohort B, 9-1 MagnetisMM-9 Part 1, 9-2A MagnetisMM-9 Part 2A Demographic and baseline characteristics of participants included in the population pharmacokinetic analysis (N = 321) 63.55 (9.61) 59.50 (13.31) 67.38 (7.26) 66.73 (9.65) 64.00 (10.03) 67.07 (9.45) 65.69 (9.42) 60.42 (10.14) 64.70 (8.38) 65.27 (9.68) 64.00 (46.00–82.00) 58.50 (42.00–74.00) 67.00 (53.00–79.00) 69.00 (50.00–80.00) 68.50 (49.00–70.00) 68.00 (36.00–89.00) 67.00 (41.00–84.00) 63.00 (36.00–79.00) 66.00 (49.00–78.00) 66.00 (36.00–89.00) 79.13 (20.71) 68.83 (13.45) 73.57 (14.93) 75.17 (19.25) 68.78 (5.17) 74.05 (18.62) 73.23 (16.73) 68.19 (15.24) 75.38 (14.04) 74.04 (17.95) 75.10 (42.00–120.0) 64.50 (56.00–87.30) 71.10 (51.30–99.40) 67.50 (55.20–121.9) 70.10 (61.40–73.50) 72.00 (40.90–159.6) 69.30 (36.50–119.3) 66.30 (44.50–101.5) 79.60 (54.50–91.70) 71.50 (36.50–159.6) 76.33 (20.04) 87.84 (23.40) 77.81 (21.01) 81.87 (21.20) 92.52 (13.02) 75.07 (21.98) 77.91 (24.47) 86.67 (20.73) 92.22 (21.28) 78.46 (22.18) 74.46 (36.58–107.47) 94.02 (44.58–110.63) 79.97 (44.01–103.27) 92.52 (36.29–103.99) 95.63 (74.12–104.70) 74.83 (22.63–122.77) 81.86 (22.50–114.86) 96.52 (42.09–124.58) 97.38 (33.66–108.49) 81.21 (22.50–124.58) 16.24 (19.24) 38.41 (64.72) 21.73 (23.07) 14.31 (25.02) 16.31 (30.43) 21.24 (29.26) 8.72 (9.10) 13.99 (14.30) 16.90 (27.01) 8.44 (0.00–74.63) 12.04 (2.83–153.52 18.70 (0.00–81.67) 4.19 (0.42–95.19) 7.52 (0.00–266.67) 10.66 (0.00–161.67) 5.91 (0.41–48.52) 6.35 (0.67–38.52) 8.29 (0.00–266.67) 3.70 (2.00–4.60) 3.40 (3.20–4.30) 3.70 (2.40–4.50) 3.30 (2.60–4.20) 3.50 (3.20–3.70) 3.60 (2.10–4.70) 3.50 (2.60–4.50) 3.70 (1.90–4.80) 3.75 (2.30–4.80) 3.60 (1.90–4.80) 0.50 (0.10–1.40) 0.25 (0.20–1.00) 0.50 (0.30–1.40) 0.40 (0.20–1.00) 0.65 (0.40–1.40) 0.40 (0.10–3.80) 0.40 (0.20–1.40) 0.50 (0.20–0.90) 0.45 (0.10–1.10) 0.40 (0.10–3.80) 30.96 (33.75) 33.00 (30.92) 26.00 (10.80) 19.47 (9.66) 19.50 (7.85) 24.60 (12.84) 30.20 (12.78) 22.06 (10.48) 23.60 (10.85) 26.39 (18.22) 23.00 (9.00–247.00) 20.50 (12.00–93.00) 22.00 (13.00–49.00) 18.00 (8.00–40.00) 16.50 (14.00–31.00) 21.00 (10.00–86.00) 28.00 (12.00–79.00) 19.00 (12.00–73.00) 21.00 (10.00–42.00) 23.00 (8.00–247.00) Tumor burden was only relevant for the ER analysis, which only included participants in MagnetisMM-3: High tumor burden at baseline = plasma cell infiltrate in the bone marrow ≥80% or serum M-spike ≥5 g/dL or serum free light chain ≥5000 mg/L. Low tumor burden at baseline = plasma cell infiltrate in the bone marrow <50% and serum M-spike <3 g/dL and serum free light chain <3000 mg/L. Missing tumor burden at baseline = any or all of the following lab values are missing: plasma cell infiltrate in the bone marrow, serum M-spike, serum free light chain. Patients who did not fit either criterion were categorized as having intermediate tumor burden. Not evaluable: Subjects with only baseline samples and no post-treatment samples, or with assay interference preventing ADA assessment ADA anti-drug antibody, AST aspartate aminotransferase, eGFR estimated glomerular filtration rate, ER Exposure–response, NA not available, Nm number missing, QW every week, Q2W every 2 weeks, sBCMA soluble B-cell maturation antigen, SD standard deviation, – indicates not applicable, 1-1 MagnetisMM-1 Part 1, 1QW MagnetisMM-1 Part 1.1 QW, 1Q2W MagnetisMM-1 Part 1.1 Q2W, 1-2A MagnetisMM-1 Part 2A, 2 MagnetisMM-2, 3A MagnetisMM-3 Cohort A, 3B MagnetisMM-3 Cohort B, 9-1 MagnetisMM-9 Part 1, 9-2A MagnetisMM-9 Part 2A Covariate selection began with an exploratory phase where potential covariates were assessed graphically by plotting against individual η estimates. Next, during the verification phase, a univariate analysis was conducted to identify covariates meeting the minimum verification criteria for inclusion in the formal covariate analysis, namely statistical significance (α = 0.01), confidence intervals excluding zero, and reduction in parameter variability. The formal covariate analysis was then performed using the stepwise covariate modeling (SCM) procedure implemented in PsN (Uppsala University), employing forward inclusion (α = 0.05) and backward elimination (α = 0.001) based on likelihood ratio tests [18–20]. Categorical covariates were modeled using a linear approach, while continuous covariates used a power model centered on the median covariate value. Model Evaluation During model building, the goodness of fit (GOF) of different models to the data was evaluated using the following criteria: change in OFV, visual inspection of different diagnostic plots, precision and plausibility of the parameter estimates, as well as decreases in both inter-individual variance (IIV) and residual variability. These criteria were used only when the minimization step was successful and standard errors of parameter estimates were obtained using the covariance step. The difference in minimum objective function (MOF) between two hierarchical models has an approximate chi-squared probability distribution with the number of degrees of freedom (df) equal to the difference in the number of parameters between the models. Based on the chi2 distribution with df = 1, a change in OFV of 3.84 corresponds to a p-value of 0.05 and a change in OFV of 10.83 corresponds to a p-value of 0.001. Traditional prediction-corrected visual predictive checks were used to assess overall model performance for the base and final models. To account for early discontinuation in non-responders, simulated data for individuals with elranatamab clearance >0.3 L/d were excluded after approximately three cycles (84 days), reflecting observed dropout patterns. Additionally, given observed differences in exposure among patients remaining on treatment at later timepoints, individual model predicted elranatamab and sBCMA levels were compared with observed data for the final models. These complimentary approaches demonstrate that the model captures both population variability and observed trends in patients remaining on treatment at week 24, that is, mostly responders given the rapid responses observed with elranatamab treatment. Prediction-corrected visual predictive checks (pcVPCs) were generated using 100 replicates, consistent with standard practice, and a sensitivity analysis with 1000 replicates was performed during model evaluation; results were essentially identical, confirming robustness of percentile coverage and overall model fit. From these simulation-based diagnostics using final parameter estimates (fixed and random effects), concentration–time data were summarized using median, 5–10th and 90–95th percentiles and the concordance between observations and simulated values, as well as the distribution, were evaluated. For IPRED-based diagnostics, the 5th, median, and 95th percentiles of the model-predicted individual concentrations were overlaid with the 5th, median, and 95th percentiles of the observed concentrations to evaluate the individual level fit. Furthermore, to provide temporal validation, the model performance was evaluated over a longer follow-up period by pooling the original development dataset (cutoff 2022) with additional data from MagnetisMM-3 as it became available (cutoff March 2024). This provided an opportunity to assess the model robustness using an expanded longitudinal dataset. ER–Safety Analysis: CRS By Dose Population ER analyses were performed for the incidence of any-grade CRS on a given dose event following the two–step-up priming dose regimens tested in MagnetisMM-3 (cutoff March 2024) [21]. Binomial logistic regression analyses were conducted using the glm() function in the R programming language for each ER analysis by dose event, including first step-up dose, second step-up dose, and first full dose. Potential covariate effects on CRS incidence for a given dose event, including exposure, were evaluated. To identify the most predictive exposure metric, each candidate metric was tested in separate univariate base models as a predictor of the CRS event. Exposure metrics were derived from individual post-hoc PK parameters estimated and simulated within NONMEM using the final PopPK model and included free elranatamab Cmax 24 (i.e., maximum elranatamab concentration 24 h after the dose) and Ctrough (trough elranatamab concentration prior to the next dose), calculated on both the linear and log-transformed scales. Model selection was guided by evaluation of the p-value for the change in deviance between nested models. The most significant exposure metric was carried into the full multivariable model, which also evaluated the following additional prespecified baseline covariates: age, body weight, creatinine clearance, platelet count, sBCMA, lactate dehydrogenase, sex, race, myeloma, immunogenicity, premedication, steroid route, prior CAR-T, prior sBCMA treatment, and tumor burden. Final model selection was performed using a backwards elimination procedure with a significance threshold of α = 0.01. Model performance was assessed using the Hosmer–Lemeshow GOF test, the receiver under the operating characteristic (ROC) curves, and standard residual diagnostics. PopPK Simulations Simulations were performed using the final model parameters to predict dose proportionality of free and total elranatamab exposure across various dosing regimens, as well as to estimate half-life, the accumulation ratio, and steady-state exposure. Simulations using the final PopPK model parameters were also performed to predict free elranatamab concentration–time profiles for various dose interruption scenarios. This helped guide restart treatment strategies to mitigate the risk of CRS events. Simulations for each dose interruption scenario assumed a hypothetical dose interruption after each of the following dose events: the first dose (step-up dose 1: 12 mg on Day 1), the second dose (step-up dose 2: 32 mg on Day 4), the third dose (first full treatment dose: 76 mg on Day 8), the fourth dose (76 mg on Day 15), the sixth dose (76 mg on Day 29), steady-state weekly (QW) dosing through Week 24 (last dose on Day 162), steady-state every-2-week (Q2W) dosing through Week 48 (last dose on Day 323), and steady-state every-4-week (Q4W) dosing through Week 72 (last dose on Day 477). For each scenario, the time required for the median free elranatamab concentration to fall below a reference threshold, which was based on the ER-CRS by dose analysis, was calculated. Simulations were run using individual post-hoc estimates from the analysis population (N = 321), assuming everyone received the approved 12/32/76 dosing regimen. Results Patient Characteristics and Observed Data The baseline demographic and laboratory variables of the study population from the MagnetisMM-1, -2, -3, and -9 studies are provided in Table 1. Overall, 167 (52%) of the participants were male, 193 (60%) of the participants were White, the median (range) baseline age was 66 years old (36–89 years), and the median (range) baseline body weight was 71.5 kg (37–160 kg). For the ER analysis in MagnetisMM-3, there were 183 participants who received the two–step-up priming dose regimen. For the analysis of any grade CRS, there were a total of 79 CRS events (43%) after the first step-up priming dose, 35 CRS events (19%) after the second step-up priming dose, and 13 CRS events (7%) after the first full dose. In total, there were 13,233 PK and pharmacodynamic observations: 3739 total elranatamab PK observations, 2947 free elranatamab observations, 3812 total sBCMA observations, and 2735 free sBCMA observations. The dataset contained a low percentage of values below the limit of quantification (BLQ), with <10% overall for total and free elranatamab and total and free sBCMA. BLQ observations were excluded from the analysis, as the proportion of BLQs was low and primarily occurred at early time points or lower doses. The percentage of BLQ values was lowest for total drug (0.08%) but higher for free drug (13%), with 35% of free drug BLQs occurring at lower IV doses. In contrast, BLQ values for total and free sBCMA were more common at higher doses, consistent with target saturation dynamics. For the temporal validation, the expanded dataset included additional PK observations from MagnetisMM-3 study participants, including data from the Q4W dosing frequency. In total, there were 14,002 non-BLQ observations; 4080 total elranatamab PK observations, 3162 free elranatamab observations, 3910 total sBCMA observations, and 2789 free sBCMA observations. The observed concentration versus time profiles are presented in Figure 1 by study and color coded by dosing frequency. Observed concentration versus time stratified by protocol for free (a) and total (b) elranatamab as well as free (c) and total (d) sBCMA. QW weekly, Q2W every 2 weeks, Q4W every 4 weeks, sBCMA soluble B-cell maturation antigen Structural Population PK Model The base structural model developed to describe the PopPK of free and total elranatamab was a semi-mechanistic two-compartment target-binding model with linear clearance and first order absorption. Time-varying clearance (i.e., incorporating an exponential decay function) and saturable elimination (i.e., using Michaelis-Menten kinetics as an alternative to linear clearance) were explored but not included in the base model. The saturable elimination term was evaluated prior to and in combination with the implementation of mechanistic target-binding models to capture potential non-linearities associated with drug-target interactions. TMDD model and various approximations were evaluated and the semi-mechanistic target-binding base model structure described in Sect. 2.4 was determined to best fit the observed data. Final Model and Covariate Effects The covariate analysis was performed using the base model to investigate whether any potential covariates, listed in Sect. 2.5, significantly influenced elranatamab PK. Graphical summaries of empirical Bayes estimates for inter-individual variability (ETAs) versus covariates examined potential relationships for elranatamab CL, Vc, and first-order ka. Of all the covariate relationships assessed, only the following were statistically significant and reduced parameter variability: age on ka, sex on CL, and body weight on Vc. The full model was reached in three forward selection steps (α = 0.05). First, the effect of sex on CL of elranatamab was added to the base model using a linear function. Next, baseline body weight on Vc of elranatamab was included using a power function. And finally, age on ka was added, also using power functions. Final model development began with the full model, as described in the previous section, and then was subjected to a backward elimination algorithm. Backward elimination (α = 0.001) was finalized after one step as none of the covariates selected during forward addition were eliminated. The equations below describe the final model estimation of typical values of CL, Vc, and ka before IIV: where CLcov SEXF is 1 if applicable to each participant and 0 otherwise. TV represents the typical individual (population mean) value of the parameter, BWT represents the baseline body weight, and AGE represents the baseline age. The IIV for parameters for Vp, Q, F, and Kd were fixed to a small number (~15%), which is common practice when using the SAEM method. The final model parameters with confidence intervals are summarized in Table 2. All model parameters were estimated with good precision (relative standard error [RSE] <30%), were within the 95% CIs, and are in agreement with published values for monoclonal antibodies (mAbs) [4]. Table 2Final population PK model parameter estimatesParameterEstimateRSE (%)Shrinkage (%)95% CICLelranatamab (L/day)0.3249.114–(0.266; 0.382)Vc,elranatamab (L)4.7775.745–(4.239; 5.315)Vc,sBCMA (L)15.41810.904–(12.123; 18.713)CLsBCMA (L/day)0.27321.03–(0.161; 0.386)CLcomplex (L/day)0.1649.201–(0.134; 0.193)Vc,complex (L)3.8024.848–(3.441; 4.164)BLsBCMA (nM)6.9147.793–(5.858; 7.970)Vp,elranatamab (L)2.831.766–(2.732; 2.928)Q (L/day)0.2251.933–(0.217; 0.234)F0.5620.564–(0.556; 0.568)ka (day–1)0.2874.432–(0.262; 0.311)Kd (nM)3.1383.312–(2.934; 3.342)Residual error total elranatamab0.4223.038–(0.397; 0.447)Residual error total sBCMA0.3474.028–(0.320; 0.375)Residual error free elranatamab0.3472.61–(0.330; 0.365)Residual error free sBCMA0.5183.503–(0.482; 0.553)Sex on CLelranatamab−0.49212.533–(−0.613; −0.371)BWT on Vc,elranatamab1.01722.351–(0.571; 1.462)Age on ka−1.45920.025–(2.031; −0.886)IIVCV (%)RSE (%)Shrinkage95% CIIIV on CLelranatamab (%)100.153.99620.17(24.08; 143.6)IIV on Vc,elranatamab (%)68.5613.34631.267(58.91; 77.01)IIV on Vc,sBCMA (%)135.9831.9124.118(83.25; 173.38)IIV on CLsBCMA (%)448.0715.70226.128(372.8; 512.41)IIV on CLcomplex (%)79.1819.33826.252(62.45; 93.01)IIV on Vc,complex (%)70.0717.26123.128(57.01; 81.06)IIV on θ BLsBCMA (%)134.6113.0780.276(116.1; 150.9)IIV on Vp elranatamab (%)14.83#–––IIV on Q (%)14.83#–––IIV on F (%)14.83#–––IIV on ka (%)68.4114.53818.344(57.88; 77.59)IIV on Kd (%)14.83#–––Condition number150.125–––#IIV for Vp, Q, F, and Kd was fixed to a small positive value (≈15% CV) to maintain numeric stability during model estimationBL baseline, BWT body weight, CI confidence interval, CL clearance, complex elranatamab-sBCMA, CV coefficient of variation, F bioavailability, IIV inter-individual variability, ka first-order absorption rate constant, Kd dissociation constant of elranatamab:sBCMA, Prop proportional, Q inter-compartmental clearance, RSE relative standard error, sBCMA soluble B-cell maturation antigen, Vc central volume of distribution, Vp peripheral volume of distribution, – indicates not applicable Final population PK model parameter estimates #IIV for Vp, Q, F, and Kd was fixed to a small positive value (≈15% CV) to maintain numeric stability during model estimation BL baseline, BWT body weight, CI confidence interval, CL clearance, complex elranatamab-sBCMA, CV coefficient of variation, F bioavailability, IIV inter-individual variability, ka first-order absorption rate constant, Kd dissociation constant of elranatamab:sBCMA, Prop proportional, Q inter-compartmental clearance, RSE relative standard error, sBCMA soluble B-cell maturation antigen, Vc central volume of distribution, Vp peripheral volume of distribution, – indicates not applicable None of the covariate effects resulted in clinically relevant changes in exposure, even when evaluated at extreme values (5th and 95th percentiles) compared with the typical patient. The exposure between male versus female was indistinguishable (Fig. 2a), the exposure between extreme values for body weight did not differ (Fig. 2b), and the exposure between extreme values for age were similar to exposure for typical participants (Fig. 2c). Furthermore, consistent with the forest plot (Fig. S1, see ESM), median free elranatamab AUCtau,steady-state values were 0.579 mg/L (95% range 0.0788–2.25) in males and 1.33 mg/L (0.122–2.71) in females, demonstrating substantial overlap despite a ~49% difference in typical clearance. Similar overlap was observed for steady-state maximum concentration (Cmax,ss). For age (<65 years versus ≥65 years) and body weight (quartiles as well as 10th vs 90th percentile), changes in PK parameters also produced overlapping exposure distributions that were not clinically meaningful. These findings indicate that although certain covariate effects are detectable at the parameter level, they do not translate into clinically relevant differences in therapeutic exposure. Exposure by sex, body weight, and age: (a) assuming a typical patient with median values for age (66 years) and weight (71.5 kg); (b) assuming a typical male patient with a median value for age (66 years); (c) assuming a typical male patient with median value for weight (71.5 kg). Each point represents the exposure metric for a single participant included in the PopPK analysis (N = 321). Caverage average concentration, Cmax maximum concentration Model Evaluation The PopPK model demonstrated good predictive performance as assessed by standard GOF plots—prediction-based diagnostic plots of the observed concentrations (DV) versus population predictions (PRED) and individual predictions (IPRED); residual-based diagnostic plots of NPDE and IWRES versus PRED and time (TAFD). IPRED-based diagnostic plots demonstrate that the final model performed well in predicting the observed data, with the percentiles of the observed data mostly lying within the prediction intervals of the simulated data. For IPRED-based diagnostics (Fig. 3), a comparison of the model-predicted individual concentration versus the observed concentration–time profiles indicating good individual-level fit and consistency between model-predicted post-hoc estimates and observed trajectories. As shown in Fig. 3, the model predicted individual elranatamab and sBCMA levels and closely matched the observed data. IPRED-based diagnostic plots showing observed concentrations and individual predictions summarized by median and 5th/95th percentiles for (a) total elranatamab, (b) free elranatamab, (c) total sBCMA, and (d) free sBCMA following 12/32/76 QW of elranatamab. These plots illustrate individual-level fit; classical simulation-based VPCs are provided in Fig. S2 (see ESM). The red lines represent the median (solid) and the 5th and 95th percentiles (dashed) of the observed data. The blue lines represent the median (dashed) and 5th and 95th percentiles (dotted) of the IPRED concentration values. IPRED individual prediction, QW once weekly, sBCMA soluble B-cell maturation antigen, VPCs visual predictive checks As a temporal validation, additional data from a later cutoff date in study MagnetisMM-3 were added to the original dataset and evaluated without refitting the model. The final model adequately described these new data, supporting temporal consistency of model performance within the same population (Fig. 4). Population and individual predicted concentrations aligned well with observed data across the full range of dosing frequencies (Fig. 4a). Graphical evaluations indicated that the previously identified covariates remained sufficient to explain variability (Fig. 4b, c). Visual predictive checks based on individual predictions further confirmed the model’s ability to capture the temporal dynamics and variability in the new dataset (Fig. 4d). Diagnostic pcVPCs for total and free elranatamab demonstrated good agreement between observed and model-predicted concentrations. Adjusting for early discontinuation in non-responders improved predictive performance (see Methods Sect. 2.7); plots are provided in the ESM (Fig. S2). Additional standard GOF plots are provided in Fig. S3 (ESM). Elranatamab exposure metrics are summarized using final model post-hoc estimates from the analysis population, shown in Table 3. Temporal validation. (a) Prediction-based goodness-of-fit plots for free elranatamab. The blue solid line represents unity. The red dashed line represents linear smooth. (b,c) Final model ETAs on clearance of elranatamab versus baseline covariates. (d) Model predicted individual vs observed profiles for free elranatamab. The red lines represent the median (solid) and the 5th and 95th percentiles (dashed) of the observed data. The blue lines represent the median (dashed) and 5th and 95th percentiles (dotted) of the IPRED concentration values. ADAB baseline anti-drug antibody, ADAT treatment-induced or treatment-boosted anti-drug antibody, CL clearance, eGFR estimated glomerular filtration rate, ETA empirical Bayes estimate for inter-individual variability, IPRED individual prediction, PRED population prediction, QW weekly, Q2W every 2 weeks, Q4W every 4 weeks, RegFreq regimen frequency, sBMCA baseline soluble B-cell maturation antigen Table 3Summary exposure metrics for free elranatamab in subjects with RRMMRegimenCaverage ng/mL GM(CV%)Cmax ng/mL GM(CV%)Ctrough ng/mL GM(CV%)First full dose (Day 8)3083.46 (94.08)3794.53 (94.42)323.35 (102.28)SSQW (Week 24)32,025.26 (45.96)33,020.03 (45.80)30,460.11 (47.90)SSQ2W (Week 48)17,688.78 (52.96)19,493.64 (51.36)15,099.10 (60.23)SSQ4W (Week 72)8788.19 (58.17)11,519.69 (53.53)5921.46 (77.88)Caverage average concentration, Cmax maximum concentration, Ctrough trough concentration, CV coefficient of variation, GM geometric mean, RRMM relapsed or refractory multiple myeloma, SSQW steady-state for once-weekly schedule, SSQ2W steady-state for every-2-week schedule, SSQ4W steady-state for every-4-week schedule Summary exposure metrics for free elranatamab in subjects with RRMM Caverage average concentration, Cmax maximum concentration, Ctrough trough concentration, CV coefficient of variation, GM geometric mean, RRMM relapsed or refractory multiple myeloma, SSQW steady-state for once-weekly schedule, SSQ2W steady-state for every-2-week schedule, SSQ4W steady-state for every-4-week schedule PK Simulations for Model Applications Simulations were performed with the final model parameters to predict dose proportionality of free and total elranatamab exposure across various dosing regimens, as well as to estimate the accumulation ratio and half-life. On average, total and free elranatamab exhibited approximately linear PK over the dose range evaluated via SC route (fixed doses of 6–76 mg QW); slope values from linear regression analysis based on simulations from the population PK model were 0.98–1.2. Across different baseline sBCMA levels, total elranatamab exposure is expected to increase in an approximately dose proportional manner over the clinical dose range (slope = 0.98–1.1). Free elranatamab is generally dose-proportional (slope value = 1–1.2) across the clinical dose range, except in patients with high baseline sBCMA (>90th percentile), where supra-proportional increases in exposure occur (slope = 1.5–1.6). This pattern reflects the impact of high shed target burden on free drug levels, as elevated sBCMA may reduce cellular target-mediated clearance and decrease circulating free drug, that is, greater than dose proportionality of free drug exposure in these subjects reflects saturation of shed BCMA sink as dose increases. See Fig. 5 for free elranatamab and ESM for total elranatamab (Fig. S4). Dose proportionality for free elranatamab exposure as area under the concentration–time curve over the dosing interval following a single dose (AUCtau.single-dose; top) and at steady-state (AUCtau.steady-state; bottom) at extreme baseline sBCMA levels. The black line is the linear regression. The x-axis and y-axis are on the logarithmic scale. AUC area under the concentration–time curve, BL baseline, sBCMA soluble B-cell maturation antigen, tau dosing interval of 7 days The median accumulation ratio for the 76 mg QW dosing regimen of elranatamab was calculated for Cmax (4.8 for total and 6.6 for free), as well as for AUCtau (8.0 for total and 11.2 for free). The half-life of elranatamab is approximately 22 (64%) days at the 76-mg dosage, with clearance of 0.324 L/day (100%) following 24 weeks’ dosing. ER–Safety Analysis—CRS by Dose The analysis of CRS incidence for a given dose event was evaluated for the first step-up dose, the second step-up dose, and the first full dose in MagnetisMM-3. PK and CRS data were available from adult patients (N = 183) who received elranatamab monotherapy as a two–step-up priming dose regimen of 12/32/76 mg C1D1/C1D4/C1D8 then QW. The first step-up priming dose analysis of any-grade CRS included a total of 79 events and 104 non-events from 183 study participants. Log(Ctrough [ng/mL]) of free elranatamab on Day 4 of the observation window following the first step-up priming dose and prior to the second step-up priming dose (which also corresponds to the overall Cmax for the same interval) showed the largest change in deviance and was the exposure metric evaluated in the full model. Of the potential covariates evaluated, including prior CRS, dosage delay, and prior tocilizumab use, two covariates were identified as significant in the final model—elranatamab exposure and tumor burden at baseline. Patients with higher exposure, and those having higher baseline tumor burden, were associated with higher probability of any grade CRS after the first step-up priming dose (OR 3.86 and 3.11, respectively). For the analysis of the second step-up priming dose of any-grade CRS, there were not any statistically significant ER relationships identified. And finally, the ER analysis of any-grade CRS after the first full dose could not be conducted since over 90% of participants did not have CRS after the first full dose. Thus, the ER-safety analysis indicated an increased probability of CRS grade at higher early exposures after the first step-up dose only. Estimates from the final ER model for CRS (any grade) after the first step‑up priming dose are summarized in the ESM in Table S2, with model discrimination evaluated using ROC analysis (Fig. S5). Fig. 6 shows the predicted probability curve as well as box plots comparing the exposure between participants with any-grade CRS and without after each dose event evaluated. Probability of any-grade CRS after the first step-up priming dose and boxplots comparing exposure for each dose event. (a) The solid blue line is the predicted probability of a CRS event, and the shaded blue area is the 95% confidence interval. The vertical dashed lines represent the observed 0.05, 0.25, 0.5, 0.75, and 0.95 quantiles of the dependent variable. This plot was generated assuming ‘reference’ tumor burden (i.e., low/intermediate). Boxplots comparing exposure of free elranatamab for participants without (No) and with (Yes) any-grade CRS after (b) the first step-up priming dose, (c) the second step-up priming dose, and (d) the first full dose Simulation to Inform Treatment Restart Guidance Based on the ER-safety analysis of CRS by dose, it was determined that the elranatamab exposure following the first dose was associated with CRS, whereas elranatamab exposures following subsequent priming doses were not. Therefore, the geometric mean of the peak free elranatamab concentration following the first priming dose (i.e., 212 ng/mL [90% CV]) was considered an appropriate threshold to mitigate the risk of a potential CRS event after dose interruption. Simulated concentration–time profiles for six potential dose interruption scenarios showed that the time for the median free elranatamab drug exposure to fall below the threshold was >84 days (12 weeks) across all evaluated scenarios (Fig. 7, Table 4). Thus, PK modeling simulations support a strategy for restarting treatment at the first step-up dose when dose interruption exceeds 12 weeks. Simulated median concentration–time profiles of free elranatamab concentration following various dose delay scenarios. The solid green line represents the geometric mean of free elranatamab after the first step-up 12-mg dose. Scenario 1 (first black dashed line): concentration–time profile after receiving 12 mg on Day 1, 32 mg on Day 4, 76 mg on Day 8 followed by a delay. Scenario 2 (second black dashed line): concentration–time profile after receiving 12 mg on Day 1, 32 mg on Day 4, 76 mg QW on Day 8 and Day 15 followed by a delay. Scenario 3 (third black dashed line): concentration–time profile after receiving 12 mg on Day 1, 32 mg on Day 4, 76 mg QW on Day 8, 15, 22 and 29 followed by a delay. Scenario 4 (blue dashed line): concentration–time profile after receiving 12 mg on Day 1, 32 mg on Day 4, 76 mg QW for 6 cycles (24 weeks) followed by a delay. Scenario 5 (green dashed line): concentration–time profile after receiving 12 mg on Day 1, 32 mg on Day 4, 76 mg QW for 6 cycles (up to week 24), 76 mg Q2W for 6 cycles (up to week 48), followed by a delay. Scenario 6 (purple dashed line): concentration–time profile after receiving 12 mg on Day 1, 32 mg on Day 4, 76 mg QW for 6 cycles (up to week 24), 76 mg Q2W for 6 cycles (up to week 48), 76 mg Q4W for 6 cycles (up to week 72), followed by a delay. QW once a week, Q2W every 2 weeks, Q4W every 4 weeks Table 4Summary of time since last dose when median free elranatamab concentration at threshold and after 12-week delayTime of last 76-mg dose before doseNumber of 76-mg doses before dose delayTime for median elranatamab concentration to be ≤Cmax after 1st step-up dose (days)Concentration of elranatamab 12 weeks (84 days) after last dose (ng/mL)Cycle 1 Day 8187.3231.9Cycle 1 Day 152103.5390.5Cycle 2 Day 1 (Study Day 29)4119.6697.1Steady-state QW (Study Day 162)23143.11668.0Steady-state Q2W (Study Day 323)35123.9863.0Steady-state Q4W (Study Day 477)41107449.0Cmax maximum concentration, QW once weekly, Q2W every 2 weeks, Q4W every 4 weeks Summary of time since last dose when median free elranatamab concentration at threshold and after 12-week delay Cmax maximum concentration, QW once weekly, Q2W every 2 weeks, Q4W every 4 weeks Discussion The PopPK of elranatamab, a BsAb targeting BCMA and CD3, was well characterized using pooled data from 321 patients receiving elranatamab monotherapy in MagnetisMM-1, MagnetisMM-2, MagnetisMM-3, and MagnetisMM-9. A semi-mechanistic two-compartment target-binding model with linear clearance and first order absorption adequately described the PK of free and total elranatamab, as well as target analytes, free and total sBCMA. The typical value for elranatamab CL was 0.324 L/d (100% CV), sBCMA clearance was 0.273 L/d (448% CV), and the clearance of the elranatamab:sBCMA complex was 0.164 L/d (79% CV). The volume of distribution for elranatamab was 4.78 L for the central compartment (Vc) and 2.83 L for the peripheral compartment (Vp). Similarly, the volume of distribution for the complex (elranatamab:sBCMA) was 3.80 L. The volume of distribution for sBCMA was somewhat larger, 15.4 L, which is expected given its smaller size (5.4 kDa for sBCMA vs 148 kDa for elranatamab). The accumulation ratio for AUC and Cmax were approximately 4.8 and 8.0 for total elranatamab, respectively, and were 6.6 and 11.2 for free elranatamab, respectively. The exposure of free and total elranatamab was dose proportional over the dose range evaluated (6–76 mg SC; Section 3.5) and the elranatamab CL and half-life values are generally consistent with large molecules (i.e., mAbs) with linear PK [22]. Patients with very high baseline sBCMA (>90th percentile) exhibited modest more-than-dose-proportional increases in free elranatamab exposure (slope ≈ 1.5–1.6), as shown in Fig. 5. This phenomenon reflects shed target-binding dynamics rather than a need for higher dosing or a change in systemic pharmacokinetics. The major elimination pathway of elranatamab in humans is expected to be via non-specific pathways (e.g., proteolysis, catabolic degradation) rather than complex-mediated clearance (i.e., TMDD). This is further supported by the estimated elimination rate constant for free drug (0.068) being greater than that for the drug-target complex (elranatamab:sBCMA 0.043), suggesting that the free drug is cleared via non-specific pathways while the complex may be more stable in circulation. The elimination rate constant for the drug-target complex was faster than the rate for free target, which implies that the formation of the complex increases the clearance rate of the drug (i.e., target depletion). The observed decline in total and free sBCMA concentrations following elranatamab administration is consistent with the hypothesis that sBCMA is shed from malignant plasma cells, and that treatment-induced tumor cell depletion reduces the source of sBCMA over time. This is supported by the faster decline of free sBCMA relative to total elranatamab, particularly after the first dose, suggesting rapid target engagement and clearance of circulating sBCMA. These dynamics are in line with prior observations in the literature where sBCMA levels were shown to correlate with tumor burden and response [23]. While the current model does not explicitly distinguish between shed and membrane-bound BCMA, the inclusion of total and free sBCMA compartments captures the net effect of target engagement and depletion. The large IIV estimated for sBCMA clearance likely reflects biological heterogeneity in sBCMA turnover across patients, including differences in tumor burden and treatment response, combined with the limited ability to disentangle production and elimination processes from sparsely sampled biomarker data. In patients who experience tumor reduction, circulating sBCMA concentrations typically decline rapidly, driven primarily by reduced production and shedding from BCMA-expressing tumor cells, whereas individuals with limited response tend to exhibit more stable sBCMA concentrations over time. These distinct biological profiles reflect differences in underlying target burden and turnover dynamics and contribute to the observed variability in sBCMA kinetics. In addition, therapy-related effects on sBCMA turnover, such as changes in shedding dynamics or reductions in BCMA-expressing tumor cells, cannot be determined with the available data and are therefore captured as unexplained variability rather than explicit structural components. High variability in soluble target kinetics has been reported previously for BCMA and other soluble receptors and is thus not unexpected. Similarly, the somewhat higher %RSE for IIV on elranatamab CL likely reflects variability in target-mediated disposition and sparse late-phase sampling rather than model overparameterization. The proportional residual error identified for free sBCMA is consistent with expected assay-related variability and the challenges associated with quantifying low concentration biomarkers. Overall, model stability and parameter identifiability were supported by acceptable η-shrinkage (<32%), robust GOF diagnostics, and sensitivity analyses; alternative simplified structures were explored but did not reduce variability without compromising model fit. Although the model was developed using data from four clinical studies with a 2022 cutoff date, its predictive performance was further evaluated using an expanded dataset that included additional observations from MagnetisMM-3, the registrational phase II study, with a later cutoff date in 2024. While not formally an external validation, as the data were derived from the same population, the consistent model performance in the presence of new longitudinal data supports the reliability of the model over time. Notably, the model retained good predictive performance without the need for re-estimation, highlighting its stability across different regimens and data cutoffs. These findings increase confidence in the model’s suitability for describing elranatamab PK for broader clinical datasets and its utility in guiding dosing decisions. None of the potential covariates evaluated, including immunogenicity, eGFR, albumin, sBCMA, age, sex, and body weight, were found to be clinically significant predictors of elranatamab exposure. In the SCM analysis, three variables (i.e., age, sex, and body weight) were statistically significant yet their clinical impact on elranatamab exposure was considered not relevant, which is consistent with the lack of strong trends observed in the graphical summaries of covariate effects in the base model for CL, Vc, and ka. Therefore, since the impact of body weight on exposure was not considered clinically relevant, weight-based dosing of elranatamab is not considered necessary, and fixed dosing of elranatamab is therefore supported. No other significant relationships were identified for elranatamab CL, Vc, and ka, including potential covariates sBCMA and immunogenicity. Anti-drug antibodies (ADA), detected in approximately 9.5% (16/168 evaluable) of participants, were mostly treatment-induced, low titer, and transient. Evaluation of ETA plots for CL showed no discernible trend by baseline or on‑treatment ADA status. Incidence of baseline ADA, treatment-induced ADA, and ADA assessed over time were tested separately as potential covariates on elranatamab CL. None of these ADA variables were identified as statistically significant covariates on elranatamab PK parameters. Furthermore, the exposure between ADA-positive versus ADA-negative was similar for typical participants (see Fig. S6 in the ESM). Thus, immunogenicity (baseline, treatment-induced, or longitudinal) is not considered to significantly affect elranatamab PK. The effect of renal and hepatic impairment on elranatamab PK was also assessed. For the assessment of renal function, baseline eGFR was not found to have a statistically significant influence on elranatamab CL. Similarly, for the assessment of hepatic function, none of the baseline associated safety laboratory measurements (albumin, AST, or bilirubin) were found to be statistically significant covariates in the final model. Furthermore, based on the categorical assessment of renal and hepatic function, the results show that the CL of elranatamab is similar for normal function compared with all evaluable renal impairment and hepatic dysfunction categories. Therefore, no dose adjustments are warranted for renal impairment or hepatic dysfunction. PopPK simulations and exposure-CRS analysis by dose support the need for repriming only if elranatamab exposure drops below Cmax after the first dose. Exposure–safety analyses indicated that overall, CRS risk was highest following the first dose and was mitigated by step-up dosing; therefore, the maximum free elranatamab concentration following the first dose (Cmax following the first step-up dose of 12 mg) was considered a reasonable threshold to mitigate the risk of CRS given that this threshold represents the typical exposure level achieved during initial priming. The PopPK model was used to simulate concentration–time profiles under different dose interruption scenarios, and to determine the time required for free elranatamab exposure to drop below this threshold. In all evaluated dose delay scenarios, exposure was maintained above the specified threshold for up to 84 days (i.e., 12 weeks). Thus, a PK rationale was provided for the allowance of 12-week dose interruptions without repriming. This PopPK model provided a data-driven framework to guide clinical recommendations for restarting therapy after dose interruptions or delays. Future analyses may refine these thresholds based on more detailed safety data or alternative markers of CRS risk. Of note, while logistic regression per dose–event was used for simplicity and regulatory familiarity, we acknowledge that this approach does not fully capture the temporal dynamics of CRS. Advanced longitudinal methods, such as time-to-event or Markov models, provide a more comprehensive framework and have been explored by our group to inform dose optimization [9]. TMDD models and various approximations have been used to describe the PK of bispecific drugs and other protein drugs with disposition influenced by receptor interaction [24, 25]. In this analysis, the relationship between the three entities, free elranatamab, free sBCMA, and elranatamab:sBCMA complex were assumed to always be at equilibrium. Given the timescale of elranatamab disposition (weeks) and the short equilibration time for the complexation reaction, such an assumption is considered appropriate. Furthermore, the influence of CD3 on elranatamab PK was assumed to be minimal and was not evaluated here, which is in alignment with modeling of other BsAbs where simplifications are justified to balance model complexity with interpretability and prioritize the most clinically relevant pharmacokinetic interactions. Moreover, the individual prediction-based diagnostics closely matched the observed data (Fig. 2), indicating that the model is sufficient for predicting post-hoc total and free elranatamab exposure metrics that can be used for subsequent ER analyses. Evolving data for BCMA BsAbs suggests the need for less frequent dosing to align with myeloma disease assessment schedules, improve convenience, reduce cumulative dosing, and potentially toxicity. Thus, a response-guided Q4W switch was implemented via a protocol amendment for patients in the MagnetisMM-3 study. Nonetheless, the predictive performance of the final model was assessed using an expanded dataset with additional observations from the registrational study, MagnetisMM-3, including data from Q4W dosing frequency (Fig. 4). The model performance in the presence of new longitudinal data supports the robustness and generalizability of the model over time. The initial simulation-based VPCs suggested further improvements in the model may be considered, but attempts to implement these so far have resulted in overparameterization of the final model. At later time points, observed PK values were higher than predicted by the model. This may reflect enrichment bias, where patients with higher exposure or better tolerability remained on treatment longer. In this case, enrichment may be partly driven by target binding dynamics, where CL decreases over time due to saturation or depletion of target. Diagnostic checks, including pcVPCs accounting for early discontinuation in non-responders, confirmed that the final model adequately described both total and free PK. This supports retention of a parsimonious structure while acknowledging that clinical recovery may contribute to the observed trend. This analysis has several limitations, and the findings should be interpreted in the context of the following methodological considerations. First, the PopPK framework assumes a simplified structure, including rapid equilibrium between free elranatamab, soluble BCMA, and their complex, as well as exclusion of CD3-mediated processes from the PK model. These assumptions were considered appropriate given the relevant timescales of elranatamab disposition but may not capture all underlying biological complexity. Second, large IIV in sBCMA kinetics reflects biological heterogeneity and limited identifiability of biomarker production and elimination processes from sparsely sampled data, which were therefore captured as unexplained variability rather than explicit structural components. Despite these considerations, diagnostic assessments supported the adequacy of the final model for its intended clinical decision-making applications. Although the model was not formally externally validated, its predictive performance was evaluated using an expanded dataset that included additional longitudinal observations from MagnetisMM-3, including data from the Q4W dosing regimen. Consistent performance across dosing frequencies without re-estimation supports the robustness of the model for its intended clinical applications. Conclusion This analysis aimed to characterize the PK of elranatamab to inform optimal clinical use. The developed model provides clinically meaningful insights into elranatamab PK and supports key aspects of drug development, including dose selection, treatment reinitiation, and safety management. These findings reinforce the robustness of the approved fixed subcutaneous dosing regimen across demographic covariates and highlight the utility of the target-binding model in capturing PK behavior. Future work may build on these insights to support lifecycle management and continued clinical development across broader patient populations. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 2560 KB) Author Contributions J.E.H. wrote the manuscript; T.N., J.H.W., M.E., D.W., and J.E.H. designed the research; J.E.H., L.E.K., D.G., D.I., and A.L. performed the research; J.E.H., D.I., A.L., P.S., M.E., D.W., J.H.W., L.E.K., D.G., S.J., H.-K.L., B.S., K.E.P., and T.N. analyzed the data. All authors read and approved the final version. Funding This work was sponsored by Pfizer. Declarations Conflict of Interest J.E.H., D.I., A.L., M.E., L.E.K., D.G., S.J., K.E.P., P.S., H.-K.L., B.S., and T.N. are full-time employees of Pfizer, Inc., and own Pfizer stock. D.W. and J.W were full-time employees of Pfizer, Inc. at the time of conducting this analysis and may own Pfizer stock. Availability of Data and Material Upon request, and subject to review, Pfizer will provide the data that support the findings of this study. Subject to certain criteria, conditions, and exceptions, Pfizer may also provide access to the related individual de-identified participant data. See https://www.pfizer.com/science/clinical-trials/trial-data-and-results for more information. Model scripts may be made available upon reasonable request to the corresponding author. Code Availability Not applicable. Ethics Approval Each study site’s local independent ethics committee or institutional review board approved the study protocol. The study was conducted in accordance with the principles of the Declaration of Helsinki and the International Conference on Harmonization Good Clinical Practice guidelines. All patients provided written informed consent. Consent to Participate Written informed consent was obtained from all participants prior to study participation. Consent for Publication Not applicable. Footnotes Affiliation listed for Daine Wang reflects the institution at the time of contribution. Currently at Blossom Hill Therapeutics. Affiliation listed for Jason H. Williams reflects the institution at the time of contribution. Currently at Exelixis. References Associated Data Supplementary Materials Supplementary file1 (DOCX 2560 KB)