Clinical Validation of a QSP Model for ISB 2001, a Trispecific T Cell Engager to Support Optimal FIH Study Design in RRMM Patients Authors: Vinu Chandralayam Ayyappa Menon, Tomomi Matsuura, Beata Holkova, Girish S. Gudi, Adam Drake, Maria Pihlgren, Piet H. van der Graaf, Sunitha GN, Andrew Garton, Mario Perro, Cyril Konto, Lida Pacaud Published: 2026 ABSTRACT ISB 2001 is a trispecific antibody that binds to two tumor‐associated antigens (TAA), CD38 and BCMA, and crosslinks to CD3 on T cells resulting in T cell‐mediated cytotoxicity of tumor cells expressing these TAAs. ISB 2001 is currently being tested in relapsed and/or refractory multiple myeloma (RRMM) patients in the first‐in‐human (FIH) TRIgnite‐1 study (NCT05862012). The CD3ε‐binding domain of ISB 2001 does not cross‐react with CD3 in cynomolgus monkeys. Therefore, no pharmacokinetic, efficacy, or toxicology data could be generated in a fully cross‐reactive preclinical species to facilitate prediction of optimal FIH dose or the clinical efficacy dose range. For T cell engagers (TCEs), the minimum anticipated biological effect level (MABEL) approach is widely used to derive FIH doses. While this method is considered safe, this standard approach may result in starting doses that are sometimes too low to elicit meaningful clinical activity, potentially leading to several dose escalation cohorts at subtherapeutic levels in seriously ill patients. We used a novel quantitative systems pharmacology (QSP) model to inform FIH dose selection and predict the clinical efficacy dose range for ISB 2001. This modeling strategy enabled an efficient dose escalation design for the TRIgnite‐1. This approach is consistent with the emerging FDA roadmap to reduce animal testing by utilizing computational modeling methodologies for FIH dose selection. Finally, we demonstrate the validation of the QSP model and approach by comparing its predictions with evolving preliminary safety, PK, and antimyeloma activity data from the ongoing TRIgnite‐1 clinical trial. Study Highlights WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC? The recommended FIH dose for TCEs is derived typically using a conservative MABEL approach from in‐vitro assays, leading to a very low starting dose leading to multiple escalation cohorts at non‐therapeutic doses to seriously ill patients. WHAT QUESTION DID THIS STUDY ADDRESS? Can mechanistic mathematical modeling integrate relevant preclinical data and literature information to select an optimal FIH dose and design an efficient phase 1 study for trispecific TCE to expedite clinical development and minimize the number of patients exposed to non‐therapeutic doses? WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE? A QSP model was developed for ISB 2001, a trispecific TCE, which accurately predicted a safe FIH dose, PK profile, and clinical efficacy dose range. This approach significantly reduced the number of patients exposed to non‐therapeutic doses and accelerated drug development to rapidly reach efficacy dose levels. HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE? QSP model guided FIH dose and phase 1 dose escalation design can overcome critical limitations of the traditional MABEL‐based approaches to design more efficient phase 1 study design, as well as optimize animal experiments in alignment with emerging regulatory advocacy in reducing animal usage. Study Highlights WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC? WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC? WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC? The recommended FIH dose for TCEs is derived typically using a conservative MABEL approach from in‐vitro assays, leading to a very low starting dose leading to multiple escalation cohorts at non‐therapeutic doses to seriously ill patients. WHAT QUESTION DID THIS STUDY ADDRESS? WHAT QUESTION DID THIS STUDY ADDRESS? WHAT QUESTION DID THIS STUDY ADDRESS? Can mechanistic mathematical modeling integrate relevant preclinical data and literature information to select an optimal FIH dose and design an efficient phase 1 study for trispecific TCE to expedite clinical development and minimize the number of patients exposed to non‐therapeutic doses? WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE? WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE? WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE? A QSP model was developed for ISB 2001, a trispecific TCE, which accurately predicted a safe FIH dose, PK profile, and clinical efficacy dose range. This approach significantly reduced the number of patients exposed to non‐therapeutic doses and accelerated drug development to rapidly reach efficacy dose levels. HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE? HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE? HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE? QSP model guided FIH dose and phase 1 dose escalation design can overcome critical limitations of the traditional MABEL‐based approaches to design more efficient phase 1 study design, as well as optimize animal experiments in alignment with emerging regulatory advocacy in reducing animal usage. Harnessing the patient's immune system to target tumor cells is revolutionizing cancer therapy. 1 T cell engagers (TCEs), a highly effective immunotherapy modality, crosslink CD3 on T cells with tumor‐associated antigens (TAA) to form an immune synapse which activates T cells and induces tumor cytotoxicity. During 2022–2024, FDA approved 8 TCEs for treatment of various cancers, including teclistamab (CD3‐BCMA), elranatamab (CD3‐BCMA), and talquetamab (CD3‐GPRC5D) for the treatment of RRMM patients. 2 While treatment advances have improved progression‐free survival in multiple myeloma, relapse remains common due to resistance mechanisms including TAA downregulation. Trispecific TCE targeting two TAAs is a promising approach to overcome emerging resistance and enhance tumor‐specific cytotoxicity. ISB 2001 is a trispecific TCE targeting CD38, BCMA, and CD3 with optimal affinities for strong tumor cell binding, while minimizing on‐target off‐tumor binding. ISB 2001 showed superior potency in various in‐vitro and in‐vivo assays compared with monovalent or bivalent BCMA‐targeting TCE benchmarks, with similar proinflammatory cytokine secretion, suggesting improved safety margins. 3 FIH dose selection of TCE has typically been conservative, based on MABEL from in‐vitro assay data corresponding to 10–50% pharmacologic activity. 4 This approach has resulted in low FIH doses exposing seriously ill cancer patients to subtherapeutic dose levels and lengthy dose escalation, resulting in delays in bringing efficacious TCEs to patients (Table S1 ). Recently, a comprehensive retrospective analysis of 55 drug candidates including 11 TCEs, reported that all TCEs utilized MABEL based FIH dose selection and instead proposed an integrated approach to avoid significant underdosing in patients. 5 Computational approaches with varying complexity levels are being used to model the interaction of TCEs with on‐tumor and off‐tumor target antigens along with the T cells for answering multiple drug development questions. 6 A mechanistic PKPD model characterized in‐vitro cytotoxicity as a function of trimer concentration for a bispecific P‐cadherin LP‐DART, ultimately MABEL human dose was projected based on trimer EC20. 7 A translational QSP model was reported for PF‐06671008 describing the xenograft mouse TGI to the trimer between PF‐06671008, P‐cadherin, and T cells, and was used for clinical predictions. 8 Another QSP model for a trispecific TCE binding to CD38, CD28, and CD3 on T cells describing the in‐vitro cytotoxicity and T cell activation assays proposed effective receptor occupancy as a metric to connect TCE efficacy. 9 QSP model strategies were used for FIH dose selection for a bispecific, and tetra‐specific TCE utilizing in‐vitro cytotoxicity and in‐vivo monkey PK data. 10 , 11 Most of the QSP publications appeared to rely only on the in‐vitro data for FIH determination, and none of them demonstrated clinical translation. Instead of relying only on conservative in‐vitro data, we used a novel QSP‐guided approach integrating in‐vitro, in‐vivo, and clinical benchmark data to derive a safe, efficacy‐proximal FIH dose for ISB 2001. This approach was accepted by the US FDA and Australian Human Research Ethics Committee and is now validated by preliminary TRIgnite‐1 clinical data. MATERIALS AND METHODS The workflow, key data sets feeding to each model and the decision tree adopted for deriving ISB 2001 FIH dose and efficacy dose range are presented in Figure 1 . All animal experiments were in accordance with the corresponding Institutional Animal Care and Use Committee protocols. 3 Standard Goodness of Fit (GOF) Plots were used to evaluate the adequacy of the model at each step. QSP modeling guided FIH derivation workflow. QSP‐guided FIH selection and efficacy dose range prediction was done in 2 parts. Part 1 was done using only ISB 2001 data sets, which demonstrated the need for benchmarking. In part 2, preclinical benchmarking data of teclistamab and ISB 2001 along with published clinical data of teclistamab facilitated prediction of MPAD. Additional correction to account for potential CRS based on cytokine release data resulted in an optimal FIH dose. Further preemptive empirical fractionated step‐up dose was added to attenuate potential CRS. This resulted in a safe FIH dose regimen close to the efficacious dose range, now validated with preliminary clinical safety, PK, and antimyeloma activity. CYTK, cytokine; MPAD, minimum pharmacologically active dose; mPBPK, minimal PBPK; PAD, pharmacologically active dose; RDL, Redirected Lysis Assay; TE, target engagement; TGI, tumor growth inhibition. Software Matlab/Simbiology version R2021a (The Mathworks, Natick, MA, USA) was used for all PKPD/QSP analyses, simulations, and digitizing the literature figures. R version 4.2.1 was used for data formatting and additional visualizations. In‐vitro modeling In‐vitro tumor cytotoxicity and T cell model Redirected lysis assays were performed as previously described. 3 In brief, labeled MM cell lines were incubated for 72 h with human PBMCs at an E:T ratio of 5:1 with increased concentration of antibodies. Tumor cytotoxicity was measured by flow cytometry as percentage of dead target cells or as the decrease of live target cells normalized to the untreated condition. A one‐compartment target engagement (TE) model for ISB 2001 was developed using in‐vitro binding and redirected lysis (RDL) assays across multiple cell lines. This simulated in‐vitro experiments with CD4+ and CD8+ T cells expressing CD3 and tumor cells expressing BCMA and CD38. ISB 2001 interactions with CD3, BCMA, and/or CD38 were modeled using ordinary differential equations (ODE) forming: (1) ISB 2001 dimers with CD3, CD38, or BCMA; (2) trimers involving CD38 and BCMA; (3) ISB 2001‐CD3 trimers with CD38 or BCMA; and (4) tetramers forming synapses with both CD38 and BCMA (Figure S1A ). In part 2, this model was adapted for teclistamab as well using only BCMA and CD3 binding parameters, while the system parameters were kept identical for ISB 2001 and teclistamab. Representative ODEs below describe sequential binding of targets (T1, T2, & T3) with D (TCE) with respective Kon and Koff rates. (1) dD:T1dt=KonT1×D×T1−KOffT1×D:T1, (2) dD:T1:T2dt=KonT2×D:T1×T2−KOffT2×D:T1:T2, (3) dD:T1:T2:T3dt=KonT3×D:T1:T2−KoffT3×D:T1:T2:T3. Key TE model parameters included binding kinetics, receptor densities (CD38, BCMA), tumor, and T cell counts from in‐vitro cytotoxicity assays. Target engagement was assumed to occur sequentially without cooperative binding. Only non‐crosslinking complexes were assumed to be internalized (Table 1 , Table S2 ). CD3 expression 7 and internalization rates for CD3, CD38, and BCMA were sourced from literature. 12 , 13 , 14 Trimers and tetramers crosslinking T cells (CD4+ or CD8+) to tumor cells were considered equipotent, and their sum defined as active species (ACT) driving cytotoxicity (Figure S1A ). Normalized ACT (nACT) was calculated by dividing ACT by tumor cell number. In the in‐vitro cytotoxicity model, the nACT was linked to tumor cell death through an E max model. An empirical tumor cell death rate constant (Kdeg1) was added to capture the initial 24‐hour decline in tumor cell counts, likely due to spontaneous PBMC‐mediated killing. (4) dNtdt=R×1−NtNmax×Nt−Kdeg+Kdeg1×1+DE×Nt, (5) DE=Emax×nACTHEC50H+nACTH, where Nt: tumor cell counts; definition of parameters in Table 1 and Table S2 . Table 1Estimated parameters in each step of modelingModelParameterUnitValueDefinitionSourcePartTarget engagementkon CD31/(nm × second)1.80E‐03ISB 2001 association rate constant to CD3Calculated1&21/(nm × second)2.69E‐04Teclistamab association rate constant to CD3Calculated2kon BCMA1/(nm × second)2.40E‐04ISB 2001 association rate constant to BCMACalculated1&21/(nm × second)1.53E‐03Teclistamab association rate constant to BCMACalculated2kon_CD381/(nm × second)1.00E‐03Association rate constant to CD38Calculated1&2 In‐vitro modelR1/h0.0332Tumor cell growth rateEstimateda 1&2K deg1/h0.0345; 0.0133Tumor cell death rate constantEstimateda 1; 2K deg11/h0.0420Additional tumor cell death rate constant turned off after 24 hrEstimateda 2NmaxCell2.00E+08Maximal tumor cellCalculated1&2kprol8_01/day0.00049; 0.0005Rate constant changing to proliferation state for CD8 w/o drugEstimateda 1; 2kprol4_01/day0.00014; 0.0009Rate constant changing to proliferation state for CD4 w/o drugEstimateda 1; 2kact8_01/day0.00095; 0.0011Rate constant changing to CD25+ CD8+ activation w/o drugEstimateda 1; 2kact4_01/day0.00064; 0.0011Rate constant changing to CD25+ CD4+ activation w/o drugEstimateda 1; 2kprol_81/day0.031; 0.0242CD8 T cell proliferation rate constant in proliferation stateEstimateda 1; 2kprol_41/day0.034; 0.0314CD4 T cell proliferation rate constant in proliferation stateEstimateda 1; 2 E max_prol.CD8No unit33.17; 33.08ISB 2001 E max for CD8 activation to proliferation stateEstimateda 1; 2No unit20.65Teclistamab E max for CD8 activation to proliferation stateEstimateda 2EC50_prol.CD8Molecule/cell0.047; 0.0154ISB 2001 EC50 for CD8 activation to proliferation stateEstimateda 1; 2Molecule/cell0.2169Teclistamab EC50 for CD8 activation to proliferation stateEstimateda 2 E max_prol.CD4No unit89.33; 83.63ISB 2001 E max for CD4 activation to proliferation stateEstimateda 1; 2No unit44.8Teclistamab E max for CD4 activation to proliferation stateEstimateda 2EC50_prol.CD4Molecule/cell0.129; 0.0432ISB 2001 EC50 for CD4 activation to proliferation stateEstimateda 1; 2Molecule/cell0.3542Teclistamab EC50 for CD4 activation to proliferation stateEstimateda 2 E max_CD25CD8No unit36.93; 94.63ISB 2001 E max for CD25+ CD8+ activationEstimateda 1; 2No unit82.91Teclistamab E max for CD25+ CD8+ activationEstimateda 2EC50_CD25CD8Molecule/cell0.0125; 0.0033ISB 2001 EC50 for CD25+ CD8+ activationEstimateda 1; 2Molecule/cell0.0493Teclistamab EC50 for CD25+ CD8+ activationEstimateda 2 E max_CD25CD4No unit26.45; 23.85ISB 2001 E max for CD25+ CD4+ activationEstimateda 1; 2No unit17.14Teclistamab E max for CD25 + CD4+ activationEstimateda 2EC50_CD25CD4Molecule/cell0.0469; 0.0063ISB 2001 EC50 for CD25+ CD4+ activationEstimateda 1; 2Molecule/cell0.0565Teclistamab EC50 for CD25+ CD4+ activationEstimateda 2 E max_tc_killNo unit1.62; 2.475ISB 2001 E max for tumor cell killingEstimateda 1; 2No unit2.558Teclistamab E max for tumor cell killingEstimateda 2EC50_tc_killMolecule/cell0.0051; 0.0027ISB 2001 EC50 for tumor cell killingEstimateda 1; 2Molecule/cell0.117Teclistamab EC50 for tumor cell killingEstimateda 2HillNo unit3.31; 1.60ISB 2001 Hill coefficient for tumor cell killingEstimateda 1; 2No unit1.43Teclistamab hill coefficient for tumor cell killingEstimateda 2Mouse PBPK modelVpL0.00094; 0.00159ISB 2001 plasma volume in KMS‐12‐BM mouse modelEstimatedb 1; 2L0.00131Teclistamab plasma volume in KMS‐12‐BM mouse modelEstimatedb 2L0.0011ISB 2001 plasma volume in NCI‐H929 mouse modelEstimatedb 1σ2No unit0.657Leaky‐tissue reflection coefficient in KMS‐12‐BM mouse modelEstimatedb 1No unit0.413Leaky‐tissue reflection coefficient in NCI mouse modelEstimatedb 1CLL/day0.000272hFcRn clearanceEstimatedb 1; 2L/day0.000044; 0.000535ISB 2001 clearance in KMS‐12‐BM mouse modelEstimatedb 1; 2L/day0.000415Teclistamab clearance in KMS‐12‐BM mouse modelEstimatedb 2L/day0.00011ISB 2001 clearance in NCI‐H929 mouse modelEstimatedb 1Mouse PBPK‐PD modelTCCells/g tumor1.29E+07Tumor cell counts per gram tumor in KMS‐12‐BM xenograftCalculated1&2Cells/g tumor9.82E+06Tumor cell counts per gram tumor in NCI‐H929 xenograftCalculated1Pcm/s2.8E‐07; 2.24E‐07Permeability of ISB 2001/teclistamab into tumor34; estimated in part 21; 2Dcm2/s1.3E‐07; 1.35E‐08Diffusivity of ISB 2001/teclistamab into tumor1; 2 E max 1/day0.63; 1.2Maximum tumor killingEstimatedb 1; 2EC50 Molecule/cell27; 3.1ISB 2001 nACT for 50% killing in KMS‐12‐BM xenograftEstimatedb 1; 2Molecule/cell91ISB 2001 nACT for 50% killing in NCI‐H929 xenograftEstimatedb 1Molecule/cell137Teclistamab nACT for 50% killing in KMS‐12‐BM xenograftEstimatedb 2ktr1/day0.45; 0.49Transit rate constant of tumor cell transduction stateEstimatedb 1; 2r01/day0.44; 0.29Tumor growth rate constantEstimatedb 1; 2 M max mm3 1813; 1,456Maximum tumor sizeEstimatedb 1; 2Kt_d01/day0.19; 0.037Tumor death rate constantEstimatedb 1; 2Human QSP modelCLL/day0.253ISB 2001 clearanceScaled from hFcRn mouse1&2L/day0.60Teclistamab clearanceCalibrated with IV data2KA1/day0.20Teclistamab absorption rate constant after SC injectionCalibrated With SC data2BANo unit0.85Teclistamab bioavailability after SC injection2Kon, koff/kd; TC, sum of TC_CD38+ & TC_BCMA – TC_CD38 + BCMA+; tumor cell counts were determined by multiplying the tumor volume with cell counts/g assuming density of 1 g/mL. a Average of individual estimates (Tables S3 and S4 ). b Average of individual estimates (Tables S7 and S8 ). Estimated parameters in each step of modeling Kon, koff/kd; TC, sum of TC_CD38+ & TC_BCMA – TC_CD38 + BCMA+; tumor cell counts were determined by multiplying the tumor volume with cell counts/g assuming density of 1 g/mL. Average of individual estimates (Tables S3 and S4 ). Average of individual estimates (Tables S7 and S8 ). In part 1, ISB 2001 cytotoxicity data from KMS‐12‐BM, MOLP‐8, and NCI‐H929 cell lines were used to calibrate the model. In part 2, the model was adapted and calibrated with cytotoxicity data for both ISB 2001 and teclistamab in the KMS‐12‐BM cell line (Table 1 , Table S2 , Figures S1C,D and S2A–C ). The T cell activation and proliferation model details along with equations are provided in Data S2 . In‐vivo model Mouse minimal physiologically based pharmacokinetic‐pharmacodynamic (PBPK‐PD) model In part 1, single dose intravenous (IV) PK data of ISB 2001 in hFcRn Tg32 SCID, NSG, and NCG mice were fitted to a minimal PBPK (mPBPK) model consisting of blood, lymph, leaky, and tight tissues using nonlinear regression with the fminsearch algorithm from Matlab (Figures S3A,C,D and S4C ). 15 Clearance (CL) was estimated for hFcRn mice; CL, Vp, and σ2 were estimated for NSG and NCG mice; remaining parameters were sourced from literature. 15 , 16 , 17 A mPBPK‐PD model with a tumor compartment (TCO) was fitted to TGI data from NSG and NCG mice xenografted with KMS‐12‐BM/NCI‐H929 tumor cells and systemically administered human PBMC. The ISB 2001 distribution between plasma and TCO was described by Eq. (6) 8 (6) TD=2×P×RcapRkrough2+6×DRtumor2×C1−C3EPS, where TD = tumor disposition; C1, C3 = TCE concentration in central compartment and TCO, respectively (Table S2 , Table 1 ). T cell counts in the central compartment and its distribution into TCO after systemic PBMC injection were defined as a function of initial PBMC amount and T cell trafficking rate constant. 8 The TE module described the binding dynamics in plasma and TCO, with nACT driving tumor cytotoxicity. The target in each compartment is the product of target density with target expressing cells. The tumor volume change was described by a cell distribution transduction model 8 (Figure S3B ). (7) dM0dt=r0×1−M0Mmax×M0−ktd0+Kkill×M0, (cell state 1) dM1dt=ktd0+Kkill×M0−ktr×M1, (cell state 2) dM2dt=ktr×M1−ktr×M2, (cell state 3) dM3dt=ktr×M2−ktr×M3, Kkill=Emax×nACTEC50+nACT, TTV=M0+M1+M2+M3, where M0 = living tumor cell volume; M1, M2, and M3 = dead tumor cell volumes; TTV = total tumor volume. Rest of the parameters are defined in Table 1 , Table S2 . In part 2, this model was recalibrated to fit PKPD data generated for both ISB 2001 and teclistamab in KMS‐12‐BM xenografted NSG mice. Donor‐dependent cell dynamics parameters were fixed, and drug‐tumor disposition parameters were refined to fit both ISB 2001 and teclistamab. The EC50 as a function of nACT was estimated for ISB 2001 and teclistamab (Table 1 , Table S2 ). Human QSP model A previously published mPBPK model with blood, tight tissues, and lymph compartments was modified by separating the leaky‐tissue compartment into bone marrow (BM) and other leaky tissues, 15 enabling simulation of TCE and nACT profiles in BM, the main efficacy site (Figure 2 a ). Physiological parameters were based on typical IgG values. 16 , 17 , 18 , 19 ISB 2001 hFcRn mice CL was allometrically scaled to 70 kg human. Teclistamab CL was estimated from digitized IV PK profiles at doses 270 and 720 μg/kg; the absorption rate constant and bioavailability following SC injection were derived from 720 and 1,500 μg/kg SC PK data 20 (Table 1 , Table S2 ). Human PBPK & QSP model. (a) Human PBPK schematic; V tight, volume of interstitial fluid (ISF) in tight tissues; V bone marrow, ISF volumes in bone marrow; V leaky woBM, ISF volume in leaky tissue without bone marrow; V lymph, lymph volume; L1, lymph flow to tight tissue; L2, lymph flow to leaky tissue; L, total lymph flow as sum of L1 and L2; σ1, vascular reflection coefficient to V tight; σ2, vascular reflection coefficient to V leaky; σL, lymphatic capillary reflection coefficient; CL, plasma clearance; V ratio1, V leaky‐BM/V leaky; V ratio2, V bone marrow/V leaky; (b) Human PBPK simulated serum concentration and BM concentration–time course without target binding; (c) human QSP model. Schematic shown predominantly ISB 2001 interaction, however, applicable for teclistamab as well. The complexes that are not relevant for teclistamab are indicated with the help of legends; T2, T3, and Tet1 in BM are active species driving killing; Black dashed line: Elimination of free ISB 2001. For visual clarity, only sBCMA is shown in the schematic, but sCD38 is also integrated in the model and is only applicable for ISB 2001 interaction. The human QSP model integrated mPBPK and TE modules and included 5 compartments with sequential target binding under well‐stirred conditions (Figure 2 c ). The target engagement was assumed only in plasma and BM compartments, where plasma cells represented CD38/BCMA expressing tumor cells. CD4+/CD8+ T cells, monocytes, neutrophils, NK cells, sCD38, and sBCMA were included in both compartments to describe target‐mediated drug disposition (TMDD). 21 , 22 , 23 Cell counts and target densities were sourced from literature. 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 This model described ISB 2001 and teclistamab PK, distribution, target engagement, and nACT profiles in patients (Table 1 , Table S2 ). Sensitivity analyses of the key model parameters (target density, soluble target, plasma cell, and T cell counts) were performed to understand the impact on BM nACT profile at a predicted robust efficacy dose level (500 μg/kg) (Table S8 ). For each simulation, only 1 parameter was changed while others were fixed to the typical value (Table S9 ). Clinical dose prediction Prediction of MABEL dose The standard MABEL dose was derived such that the QSP model‐simulated Cmax were similar to in‐vitro cytotoxicity EC20, EC30, and EC50 concentrations. Prediction of MPAD and efficacy dose range The potential efficacy dose range was predicted for ISB 2001 and teclistamab from in‐vitro cytotoxicity as well as the in‐vivo TGI data. It was assumed that maintaining BM nACT concentrations within in‐vitro cytotoxicity EC50–EC90 or in‐vivo TGI EC50–EC90 range is required for clinical efficacy. Patient BM nACT profiles following once‐weekly subcutaneous (SC) dosing were simulated to identify doses that keep trough nACT within these effective ranges. Benchmarking with teclistamab The human QSP model was used to simulate patient BM nACT profiles for the approved teclistamab regimen. The mouse ECx range yielding tumor nACT levels matching patient BM nACT following teclistamab served as the reference for ISB 2001 dose predictions. During dose escalation, teclistamab showed first clinical response at 38.4 μg/kg (IV). 37 The mouse ECx with equivalent tumor nACT to patient BM nACT at 38.4 μg/kg (IV) was used as reference to derive ISB 2001 MPAD. FIH dose selection The MPAD was refined by incorporating a safety margin based on comparative cytokine release data for ISB 2001 and teclistamab in KMS‐12‐BM cell line, 3 published phase 1 results for teclistamab 37 and QSP model‐simulated ISB 2001 serum Cmax. To mitigate cytokine release syndrome (CRS), empirical step‐up dosing regimens were implemented during the first week in all dose escalation cohorts. Virtual patient simulation ISB 2001 serum concentration–time profiles were simulated in a virtual patient (VP) population (n = 100) generated assuming normal body weight distribution (40–100 kg), SC bioavailability (0.5–1.0), and 30% CV around typical sBCMA, CD38, and CD3 densities of the human QSP model (Table S2 ) and compared with the observed profiles. First‐in‐human phase 1 study Study design The phase 1, first‐in‐human (FIH) dose‐escalation study of ISB 2001 monotherapy in patients with RRMM (NCT05862012) has been previously reported. 37 , 38 , 39 , 40 This study was approved by the institutional review and ethics boards, and conducted in accordance with the Declaration of Helsinki. ISB 2001 was administered SC with priming and step‐up doses on days 1 and 4, followed by weekly target doses from day 8 onwards. Based on evolving data, the priming dose was fixed to 15 μg/kg from DL4 onwards to minimize CRS incidence and severity (Figure S7A ). Safety, tolerability, PK, immunogenicity, and preliminary clinical activity were assessed as part of this study. 38 , 39 , 40 , 41 , 42 , 43 Blood samples were collected at prespecified time points for serum ISB 2001 quantification (Figure S7B ). Free ISB 2001 concentrations in the serum samples were quantified using a validated electrochemiluminescence‐based immunoassay on the Meso Scale Discovery (MSD®) platform. RESULTS In‐vitro tumor cytotoxicity model and T cell model The dimer, trimer, tetramer, and total active‐species concentrations increased with rising ISB 2001 and teclistamab concentrations; however, a decreasing trend was observed for ISB 2001‐BCMA‐CD38 trimer and ISB 2001‐CD4+/CD8+ T cell‐CD38‐BCMA tetramers at 20 nM (Figure S1B ). In part 1, model parameters were estimated for ISB 2001 using individual experiments from 3 cell lines (Table S3 ), whereas in part 2, parameters were estimated for both ISB 2001 and teclistamab in KMS‐12‐BM cell line (Table S4 ). Tumor growth and basal T cell kinetic parameters were fixed for each donor, and only drug‐related parameters were estimated for ISB 2001 and teclistamab. GOF plots demonstrated that the model adequately described the observed data (Figures S1C,D and S2A,C ). Average estimated parameters are summarized in Table 1 . The model‐derived average killing EC50 for ISB 2001 in KMS‐12‐BM differed by approximately two‐fold between parts 1 and 2. Because ISB 2001 and teclistamab were evaluated under identical experimental conditions using the same PBMC donors in part 2, their parameters were compared within this data set. ISB 2001 demonstrated ~40‐fold greater cytotoxic potency (EC50) than teclistamab, with similar E max and hill coefficients. In‐vivo mouse PBPK‐PD model The serum‐time course of ISB 2001 showed bi‐exponential decline in hFcRn, NSG, and NCG mice (Figure S3C,D ). ISB 2001 CL was estimated for individual hFcRn mouse (Table S7 ), yielding an average CL of 0.272 mL/day, consistent with IgG‐like CL (Table 1 ). In NSG and NCG mice, CL, Vp, σ2 were estimated individually (Table S7 ), and their averages are summarized in Table 1 . GOF plots demonstrated adequate model performance (Figures S3E,F and S4A,B,D,E ). In part 1, PD parameters were estimated for each PBMC donor (Table S8 ) and the average is summarized in Table 1 . The nACT EC50 for TGI was more potent in the KMS‐12‐BM cell line than in NCI‐H929. In part 2, this model was recalibrated using TGI data from KMS‐12‐BM xenografted NSG mice, following ISB 2001 or teclistamab IV injection. ISB 2001 showed approximately 44‐fold more potent nACT EC50 for TGI than teclistamab. GOF plots demonstrated that the model adequately captured the PKPD profile of both molecules (Figures S3G–J and S5A–D ). Human QSP model Human nonspecific CL, scaled from hFcRn mice, was 0.253 L/day. PK simulations without target engagement showed a t1/2 of approximately 17 days, consistent with IgG profiles 44 (Figure 2 b ). Teclistamab CL was estimated at 0.6 L/day, and the SC absorption rate constant and bioavailability were estimated as 0.20 d−1 and 0.85, respectively 20 (Table 1 ). Clinical dose prediction MABEL dose ISB 2001 EC20, EC30, and EC50 estimated from in‐vitro cytotoxicity assay in KMS‐12‐BM cell line were approximately 0.05, 0.07, and 0.11 ng/mL, respectively. The simulations showed that SC doses to achieve these concentrations at the Cmax were 0.045, 0.06, and 0.1 μg/kg, respectively (Figure S6A,B ). MPAD and efficacy dose range In part 1, the QSP model estimated in‐vitro nACT EC50 – EC90 of 0.0034–0.011/tumor cell and in‐vivo nACT EC50 – EC90 of 27–243/tumor cell in KMS‐12‐BM cell‐based models. The corresponding weekly SC efficacy dose range was 0.045–0.15 μg/kg based on the in‐vitro model and 200–600 μg/kg based on the in‐vivo model (Table S5 ). In part 2, the QSP model predicted teclistamab clinical efficacy dose range using the nACT EC50‐EC90 values from both in‐vitro cytotoxicity and in‐vivo mouse PKPD models. The predicted teclistamab dose referenced to in‐vitro nACT (0.13–0.7 μg/kg/week) was substantially lower than the observed clinically efficacious dose range of 80–1,500 μg/kg/week 37 (Figure S6C ). By contrast, the QSP model closely predicted the teclistamab clinical efficacy dose range when referenced to the in‐vivo nACT from the mouse PKPD model (Figure S6E ). The in‐vivo nACT associated with approximately 13%, 40%, and 73% efficacy in the xenograft model corresponded to model‐simulated patient BM nACT levels at 60, 300, and 1,500 μg/kg SC teclistamab, respectively (Table 2 ). Table 2Predicted clinical doseCompoundParameterValueUnitExperimentTarget exposure Metric after SCCorresponding predicted clinical dose (μg/kg)CommentISB 2001EC20 0.000257nMIn‐vitro cytotoxicitySerum ISB 2001 Cmax 0.045MABELEC30 0.000350Serum ISB 2001 Cmax 0.06EC50 0.000585Serum ISB 2001 Cmax 0.1nACT EC50 0.00251 Molecule /cell BM SS Ctrough 0.035Predicted clinical efficacy dose rangenACT EC90 0.00919BM SS Ctrough 0.13nACT EC50 3.1 In‐vivo mouse PKPDBM SD; SS Ctrough 70; 45nACT EC90 28BM SD; SS Ctrough 320; 180nACT EC13 0.5BM SD Cmax 8.0Based on teclistamab benchmarkingnACT EC17 0.63BM SD Cmax 10.5MPAD based on teclistamab benchmarkingnACT EC40 2.1BM SD Cmax 32Based on teclistamab benchmarkingnACT EC70‐75 7.0–9.0BM SD Ctrough 130–160Based on teclistamab benchmarkingTeclistamabEC20 0.00219nMIn‐vitro cytotoxicitySerum teclistamab Cmax 0.06MABELEC30 0.00322Serum teclistamab Cmax 0.085EC50 0.00604Serum teclistamab Cmax 0.16nACT EC50 0.0946 Molecule /cell BM SS Ctrough 0.13Predicted clinical efficacy dose range for benchmarkingnACT EC90 0.494BM SS Ctrough 0.7nACT EC50 137 In‐vivo mouse PKPDBM SD; SS Ctrough 450; 200nACT EC90 1,233BM Ctrough Not achievablenACT EC13 24BM SD Cmax 60Teclistamab SUD1nACT EC17 27*BM SD Cmax 38.4MPADnACT EC40 91BM SD Cmax 300Teclistamab SUD2nACT EC75 320BM SD Cmax 1,500Teclistamab TDSD, single dose; SS, steady state; *, after IV. Predicted clinical dose Molecule /cell Molecule /cell SD, single dose; SS, steady state; *, after IV. During teclistamab FIH IV dose escalation, the first clinical response occurred at 38.4 μg/kg, 37 where the predicted bone marrow nACT Cmax was 27 per tumor cell, equivalent to the EC17 from the mouse model. In the same model, ISB 2001 achieved EC17 at an nACT of 0.63 per tumor cell. QSP modeling predicted that a 10.5 μg/kg SC dose of ISB 2001 would reach this nACT in human bone marrow (MPAD) (Figure S6F , Table 2 ). QSP simulations showed that patient BM nACT plateaued at ~400/tumor cell (EC75) with teclistamab SC at 1500 μg/kg, whereas ISB 2001 showed no plateauing (Figure 3 a,b ). ISB 2001 at 320 μg/kg SC was predicted to achieve and maintain nACT EC90 in patient BM at trough from the first dose and at 180 μg/kg SC once weekly (Figure 3 c,d ). Insights into ISB 2001 potential efficacy dose range and FIH. (a) and (b) nACT profiles in patient bone marrow show plateauing for teclistamab, while no plateauing for ISB 2001 with dose increment. (c) ISB 2001 SC single dose administration to achieve nACT over EC50 to EC90 (d) ISB 2001 SC Q1W dose range to achieve steady state nACT over EC50 to EC90 (e) ISB 2001 serum concentration profiles at MPAD and at lower doses vs. in‐vitro cytokine release EC30 (Table S6 ). Sensitivity analyses showed that the CD3 density and T cell counts were the most sensitive parameters, followed by the plasma cell count for the nACT formation. However, at 500 μg/kg weekly SC injections (3‐fold lower than teclistamab RP2D), these did not yield the nACT below 3.1 molecule/cell. Although the influence of sBCMA on nACT was greater than that of sCD38, the nACT was >3.1 molecule/cell (Figure S8A–C ). FIH dose regimen The MPAD was refined using in‐vitro cytokine release data of ISB 2001 and teclistamab 3 (Table S6 ). FIH target dose of 5.0 μg/kg was selected where the QSP simulated serum C max is approximately at the cytokine release EC30, thereby anticipated to minimize CRS risk (Figure 3 e ). Additionally, an empirical step‐up dosing regimen with priming dose (10% of the target dose) on C1D1 and step‐up dose (30% of the target dose) on C1D4 was implemented to further reduce the CRS risk at C1D8 (Figure S7A ). Clinical validation of the QSP model ISB 2001 pharmacokinetics The preliminary serum‐concentration data from 30 patients from DL1 to DL8 demonstrated slow absorption following SC injections of ISB 2001, and generally sustained concentrations between weekly dose intervals. The serum concentration–time profiles were comparable with the VP simulations (Figure 4 a ). Clinical validation of the ISB 2001 QSP model. (a) QSP model simulated serum concentration profiles of ISB 2001 vs. observed serum concentration profiles up to DL8 showed good predictability of the QSP model. Solid thick blue line shows the median simulated profile for each dose regimen. The shaded region corresponds to model simulated minimum and maximum concentration profiles. Solid lines with black markers depict the observed individual subject concentration profiles. (b) Observed ORR (%) vs. model predicted median nACT demonstrated a robust response when BM nACT is approximately 4.3 molecules/tumor cell, consistent with QSP model prediction from in‐vivo TGI data with corresponding back calculated mouse ECx approximating to 60%. Preliminary cytokine release syndrome incidence and severity No CRS incidences were observed in the first 3 cohorts at C1D1, where the C1D1 doses ranged from 0.5 to 5.0 μg/kg. This was consistent with the prediction based on PK simulations and the in‐vitro cytokine release data. In DL4, with a C1D1 dose of 15 μg/kg, 3 out of 4 patients experienced low‐grade CRS (two grade 1 and one grade 2) (Table 3 ). Based on this, the C1D1 dose was fixed to 15 μg/kg for the remaining dose escalation. Table 3Cytokine release syndrome incidenceDose OccasionDL1 N = 1DL2 N = 1DL3 N = 1DL4 N = 4DL5 N = 5DL6 N = 7DL7 N = 8DL8 N = 3Total N = 30C1D10/10/10/13/43/53/74/81/314/30C1D40/10/10/11/40/51/74/80/36/30C1D80/10/10/10/41/50/70/70/31/29 Cytokine release syndrome incidence Preliminary anti‐myeloma activity ISB 2001 demonstrated strong clinical response in patients starting at 50 μg/kg (DL3), consistent with the QSP predictions. 38 , 39 , 40 The observed overall response rate (ORR) vs. QSP predicted BM nACT in VP population demonstrated strong potency for ISB 2001, where BM nACT ≥4.3 molecules/tumor cell was associated with a saturating ORR (>75%) in TRIgnite‐I study. This nACT is equivalent to an ECx of approximately 58% in the in‐vivo mouse efficacy model (Figure 4 b ). DISCUSSION ISB 2001 was designed to overcome tumor resistance mechanisms including TAA downregulation. Dual tumor targeting to BCMA and CD38 increased the binding avidity to tumor cells, expected to enhance cytotoxicity, even when the expression of either BCMA or CD38 is low. The published in‐vitro cytotoxicity data in multiple MM cell lines with a range of CD38, BCMA expression showed superior potency for ISB 2001 viz‐a‐viz teclistamab and other bispecific BCMA‐targeting TCEs, due to its dual targeting. 3 ISB 2001 was also less affected by soluble BCMA and APRIL when compared to bispecific BCMA‐targeting TCEs. 3 ISB 2001 has a lower human CD3 affinity or a faster off‐rate relative to the benchmarked BCMA targeted TCEs. 3 This along with its unique architecture where the anti‐CD38 domain distal to the anti‐CD3 domain is expected to attenuate the CRS from off‐tumor target binding. 3 Overall, ISB 2001 was expected to have good safety; therefore, the conservative MABEL approach was not optimal for FIH dose selection. Since ISB 2001 is not fully cross reactive to preclinical species (e.g., cynomolgus monkey), no PK, PD, or toxicology data were generated, precluding the use of empirical scaling methods to predict clinical PK for FIH dose and clinical efficacy dose range. Here, we generated a QSP model that described in‐vitro and in‐vivo data of both ISB 2001 as well as teclistamab. The mouse and human PBPK model parameters were fixed to the published typical IgG parameters or were estimated from experimental data (Table S2 , Table 1 ). The drug permeability and diffusivity into mouse xenograft was obtained from literature in part 1, 34 while these were estimated in part 2 to fit the PKPD data of both teclistamab and ISB 2001. These estimated parameters were within the reported range for IgGs for different tumor cell lines. 34 The in‐vivo nACT EC50 for ISB 2001 was approximately 9‐fold lower in part 2 vs. part 1. We believe this is majorly due to 10‐fold lower drug diffusivity estimated in part 2. This QSP model, benchmarked to teclistamab, showed that relying on in‐vitro cytotoxicity data will underpredict ISB 2001's clinical efficacy dose, whereas in‐vivo TGI data more accurately predicts the efficacy dose range and MPAD (Figure S6 ). Further, we refined the MPAD by integrating the predicted serum C max with the cytokine release EC30 from the in‐vitro RDL assay, resulting in a final starting dose estimate of 5 μg/kg, approximately 50 to 100‐fold higher than that based on the standard MABEL approach. We included C1D1 and C1D4 step‐up doses empirically as 10% and 30% of the target dose, respectively, to further attenuate CRS risk. Mechanistic models linking T cell activation and IL‐6 spikes to CRS, along with VP simulations for step‐up dosing optimization are documented. 45 We did not use mechanistic models to predict CRS, as in‐vivo data sets for such model calibration could not be generated because ISB 2001 is non‐cross reactive to CD3 in preclinical species. Consistent with the predictions, no CRS incidences were observed at C1D1 for DL1 (0.5 μg/kg) to DL3 (5 μg/kg), while 2 CRS events were observed at DL4, where the C1D1 dose was 15 μg/kg (Table 3 ). The observed serum concentration profiles of ISB 2001 were comparable with the predicted virtual patient serum concentration profiles (Figure 4a). No clinical responses were observed in the single‐patient cohorts at DL1 and DL2; however, DL3 patient treated at 50 μg/kg achieved a stringent complete response with MRD negativity (10−5 sensitivity by NGS, ClonoSEQ). 39 As predicted, robust antimyeloma activity was seen at 50 μg/kg and higher, where bone marrow trough nACT was predicted ≥3.1 molecules/tumor cell (Table 2 ). Comparison of observed ORR with VP‐simulated median BM nACT confirmed robust responses emerging at ~4.3 molecules/tumor cell (Figure 4 b ). Lack of response at 1.1 molecules/tumor cell vs. robust activity at 4.3 molecules/tumor cell suggests the response threshold may lie between these values, and single‐patient at lower dose levels may have limited resolution of this range. Although we did not model the avidity, these results suggest no substantial impact on the predictions. Overall, the evolving preliminary safety, PK, and efficacy data from the TRIgnite‐1 study (NCT05862012) support and validate the translational QSP model for ISB 2001. To the best of our knowledge, this study marks the first clinical validation of a QSP model referencing in‐vivo efficacy data for optimal FIH dose selection and efficacy dose range derivation for trispecific TCE. Considering the advances in designing safer TCEs, we propose a QSP modeling approach for FIH derivation of future TCEs focusing on optimal balance of efficacy and safety rather than a conservative MABEL approach. This strategy is both timely and aligned with the emerging regulatory advocacy for stepwise reduction of animal testing such as the FDA's new approach methodologies initiative for greater use of computational methods to support preclinical to clinical translation. FUNDING This study was funded by IGI Inc. CONFLICT OF INTEREST V.M, B.H, G.S.G, A.D, M.Pihlgren, A.G, M.Perro, C.K, and L.P. are employees or former employees of IGI Inc. T. Matsuura and P.H.v.d.G are employees of Certara UK. S.GN is an employee of Glenmark Pharmaceutical Ltd. AUTHOR CONTRIBUTIONS V.C.A.M., M.Pihlgren, A.G., M.Perro, L.P., C.K., and P.H.v.d.G. wrote the manuscript; V.C.A.M., T.M., G.S.G., A.D., M.Pihlgren, P.H.v.d.G., B.H., A.G., M.Perro, and L.P. designed the research; V.C.A.M., T.M., B.H., A.D., M.Pihlgren, and S.GN. performed the research; V.C.A.M., T.M., B.H., G.S.G., A.D., M.Pihlgren, and S.GN. analyzed the data. Supporting information Data S1. Table S1. Figure S1. Data S2.