A Novel Step‐Up Dosage Regimen for Enhancing the Benefit‐to‐Risk Ratio of Mosunetuzumab in Relapsed or Refractory Follicular Lymphoma

Abstract

Mosunetuzumab, a T‐cell engaging bispecific antibody targeting CD20xCD3, is approved for treating relapsed/refractory follicular lymphoma. This research supports the approved intravenous clinical dose regimen, summarizing the exposure–response relationships for clinical safety and efficacy. A population pharmacokinetic model and E _max logistic regression exposure–response models for safety and efficacy were developed using data from 439 patients with relapsed/refractory non‐Hodgkin lymphoma and 159 patients with relapsed/refractory follicular lymphoma, respectively, from a Phase I/II study (NCT02500407). Data from 0.2 to 60 mg across fixed dosing (Cohort A) and Cycle 1 step‐up dosing (Cohort B) were used. Exposure–response models, using two‐cycle area‐under‐the‐concentration curve (AUC_0–42) as the primary exposure endpoint, accurately depicted the complete response and objective response rate data across a 600‐fold AUC_0–42 range. The approved clinical dose regimen of 1/2/60/30 mg achieved near‐maximal efficacy, with model‐estimated CR and ORR (90% confidence interval) of 63.1% (49.7–75.0) and 79.1% (69.1–87.7), respectively. The exposure–response analysis for Grade ≥ 2 cytokine release syndrome identified receptor occupancy (%) within the first two cycles as a driver, with CRS dissipating beyond the first dosing cycle. No exposure‐dependent increases were observed for other serious adverse events, including neutropenia and infections. The approved intravenous step‐up dose regimen (i.e., step doses of 1 and 2 mg on Day 1 and 8, respectively) mitigated severe CRS risk, allowing safe administration of loading (60 mg) and target doses (30 mg every 3 weeks) to achieve a favorable benefit–risk profile.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Mosunetuzumab, a first‐in‐class T‐cell engaging bispecific antibody targeting CD20 and CD3, has gained accelerated/conditional approval in the U.S. and European Union for treating relapsed/refractory follicular lymphoma in patients who received two or more prior therapies.

• WHAT QUESTION DID THIS STUDY ADDRESS?

How to dose T‐cell engagers to optimize efficacy without compromising safety? Can dose regimens decouple the PK exposure driver for cytokine release from clinical efficacy? How to integrate pharmacological modeling into real‐time drug development to streamline and inform appropriate dose finding in line with Project Optimus?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This study illustrates the crucial role of dose/regimen in the therapeutic benefit/risk of T‐cell bispecifics and its implications for development decisions. It details pharmacological characterizations, including exposure–response relationships for clinical biomarkers and efficacy/safety, guiding the selection of step‐up dosing, loading doses, and maintenance regimens of mosunetuzumab.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

The development of T‐cell bispecifics presents unique challenges, including the nonlinearity, time dependency, and multi‐dimensionality of safety and efficacy E–R relationships linked through the MOA. The success of future novel agents hinges on efficiently characterizing pharmacological drivers and integrating model‐based insights with innovative Phase I/II dose‐finding designs. Our patient‐centric approach emphasizes reducing lower grade safety events and implementing a fixed duration of treatment with a favorable long‐term safety profile on top of achieving deep and durable efficacy. This sets a high standard for the future development of effective and safe bispecific antibody therapies.

Follicular lymphoma (FL) remains incurable and is the most common indolent B‐cell lymphoma. Despite advances in first‐line immunochemotherapy, most patients relapse, especially those with heavily pretreated relapsed/refractory (R/R) FL, highlighting a substantial unmet medical need. ¹ Current treatments for R/R FL with ≥ 2 prior therapies often have low complete response (CR) rates and poor tolerability. Chimeric antigen receptor T cell (CAR‐T) therapy is an option but is limited by severe toxicities and logistical challenges.

Mosunetuzumab, an off‐the‐shelf, humanized immunoglobulin (IgG1) CD20 × CD3 T‐cell engaging bispecific antibody (BiSp), redirects T cells to eliminate B cells. It is approved as a first‐in‐class BiSp in the U.S. and EU for R/R FL with ≥ 2 prior therapies, using a Cycle 1 step‐up dosing to mitigate severe cytokine release syndrome (CRS) and allow escalation to maximal efficacious doses. ² , ³ The fixed‐duration intravenous (IV) regimen includes initial doses of 1 and 2 mg on Cycle 1 Day 1 and 8, 60 mg loading doses on Day 15 and Cycle 2 Day 1, followed by 30 mg every 3 weeks for eight cycles for patients with a CR, and up to 17 cycles for those with a partial response (PR) or stable disease (SD). Regulatory approvals were based on the Phase II FL expansion cohort (n = 90) of the GO29781 study, where mosunetuzumab showed high CR (60%; 95% CI: 49.1–70.2) and objective response (80%; 95% CI: 70.3–87.7) rates, with durable remissions and a manageable safety profile. ⁴ The response rate exceeded the 14% CR rate for copanlisib, a historical control, and was comparable to CAR‐T therapies. ⁵ , ⁶ , ⁷

Mosunetuzumab's mechanism of action (MOA) involves conditional agonism for T‐cell activation through immune‐synapse formation, effective at sub‐saturating receptor occupancy levels. This potent MOA negates the “more is better” philosophy ⁸ , ⁹ for optimal dose determination. It showed broad activity across B cells and lymphoma cell lines with diverse CD20 expression, even in the presence of rituximab. ⁹ The target‐dependent T‐cell activation triggers immune activation, including cytokine release, which can lead to CRS and neurotoxicity. ⁵ , ¹⁰ , ¹¹ Managing these on‐target adverse events (AE) is crucial for T‐cell engaging immunotherapies.

Translational quantitative systems pharmacology (QSP) modeling was used to design dosing strategies for mosunetuzumab, aiming to reduce acute cytokine increases and improve the therapeutic index. ¹² , ¹³ This guided Phase I dose‐finding modifications to identify the minimal maximally efficacious dose while minimizing low‐grade safety risks, rather than determining the maximum tolerated dose (MTD), which is no longer sufficient for modern drug development of targeted and immunotherapies. ¹⁴ , ¹⁵

This article outlines the clinical dosing rationale, exposure–response (E–R) characterizations for efficacy and safety, and benefit–risk evaluations for regulatory decisions. Additionally, it discusses the impact of prior anti‐CD20 antibody treatments (e.g., rituximab and obinutuzumab) on the response to CD20xCD3 BiSp in R/R non‐Hodgkin lymphoma (NHL) patients. A novel model‐based technique for estimating clinical receptor occupancy (RO%) was reported by Bender et al. ¹⁶ and used to characterize the dose–response relationship, especially with target competition from residual anti‐CD20 antibodies. ¹⁷ RO%‐based E–R characterization provides critical insights into mosunetuzumab's clinical pharmacology and dose regimen selection.

PATIENTS AND METHODS

Study design

The GO29781 study (NCT02500407) is a Phase I/II, multicenter, open‐label, dose‐escalation, and dose‐expansion study of mosunetuzumab in patients with R/R NHL expressing CD20 and an Eastern Cooperative Oncology Group performance status of 0–1. Details regarding the study population, design, and procedures are described by Budde et al. ⁴ , ¹⁸ The GO29781 study evaluated fixed dosing (Group A; 0.05–2.8 mg IV q3w) and various step‐up dosing regimens (Group B; 0.4/1/2.8–1/2/60 mg IV q3w). PK samples for mosunetuzumab serum concentrations were collected at various points. The study protocol was approved by institutional review boards at each center. The trial adhered to the Declaration of Helsinki, ICH Guidelines for Good Clinical Practice, and applicable laws, with informed patient consent. For Roche's Global Policy on data sharing and access to clinical study documents, see (https://www.roche.com/innovation/process/clinical‐trials/data‐sharing).

PK and E–R analysis

Key analysis populations supporting clinical pharmacology analyses are summarized in Table S1 and Figure S2 . The PK‐evaluable population (n = 439, data cutoff of December 4, 2020) included all enrolled patients with R/R NHL (Groups A and B) who received at least one dose of mosunetuzumab and had one measurable drug concentration. Data from 418 patients were analyzed for anti‐drug antibodies (ADAs).

E–R analyses used clinical efficacy and safety data until the clinical cutoff date (CCOD) of August 27, 2021. The efficacy E–R analysis included patients with R/R FL and ≥ 2 prior therapies (N = 159), using investigator (INV)‐assessed response results. INV‐ and independent review facility (IRF)‐assessed clinical responses showed high concordance. The INV‐assessed CR rate (CRR) and objective response rate (ORR; CR or PR) were 56.7% and 77.8% for B11 cohorts, and 47.8% and 63.0% for B7 cohorts, respectively. INV data were used for efficacy E–R characterization due to broader data availability across dose groups. Patient doses ranged from 0.2 to 60 mg, with seven patients from Group A receiving 0.2–2.8 mg every 3 weeks, and 152 patients from Group B receiving step‐up dosing from 0.4/1.0/2.8 to 1.0/2.0/60/30 mg. Safety E–R analyses used data from patients with R/R NHL (n = 439) who received doses from 0.05 to 60 mg. For the E–R analysis of CRS, a RO‐based exposure endpoint was considered.

The primary exposure metric for efficacy E–R characterization was the model‐estimated two‐cycle area‐under‐the‐concentration curve (AUC_0–42), reflecting the average exposure of mosunetuzumab and accounting for initial loading exposure. This choice was supported by a superior Akaike information criterion compared to other metrics evaluated (e.g., three‐cycle, four‐cycle, steady state AUC), as well as the rapid tumor debulking within the first two cycles (Figure S5 ). For safety, AUC_0–42 was used for E–R evaluation for neutropenia, infections/infestations, and any Grade ≥ 3 or ≥ 1 AEs. For CRS E–R analyses, a model‐estimated maximum CD20 RO (RO_max%) within the first two cycles (42 days) was used as the primary exposure endpoint, as most Grade ≥ 2 CRS events occurred within this period. During early cycles, residual anti‐CD20 monoclonal antibodies (mAbs) such as rituximab and obinutuzumab from prior treatments transiently competed with mosunetuzumab target engagement, altering kinetics (Figure S7 ). Detailed RO% derivation was published by Bender et al. ¹⁶ based on PK and the binding affinities (KDs) of all anti‐CD20 agents (including mosunetuzumab) and summarized in Figure S7 .

The primary E–R analysis for INV‐assessed CRR and ORR used an E _max logistic regression model:

$$\begin{array}{c}\text{logit}\left(p\right)=\log \left(\frac{p\left(\text{Event}=1\right)}{1-p\left(\text{Event}=1\right)}\right)={\beta }_0+\frac{E_{\max }·{\mathrm{AUC}}_{0-42}}{E_{50}+{\mathrm{AUC}}_{0-42}}\hfill \\ +{\beta }_1·x_1+\dots +{\beta }_n·x_n\hfill \end{array}$$

here, p is the likelihood of event (CR or OR); β ₀ reflects the likelihood of event without any other variables included in the model; E _max is the maximal effect of AUC_0–42 on p; E ₅₀ is the value of AUC_0–42 which produces an effect equal to 50% of E _max; x _1…n are drug‐independent variables; and β _1…n are coefficients for effects of drug‐independent variables on p.

Kaplan–Meier (KM) plots for progression‐free survival (PFS, n = 159) and duration of CR (DOCR, n = 86) were stratified by the two expansion cohorts (B7: 1/2/13.5 and B11: 1/2/60/30 mg) and AUC_0–42 exposure tertiles. KM plots for neutropenia, infections, any Grade ≥ 3 AE, and any Grade ≥ 1 AE (by National Cancer Institute Common Terminology Criteria for AE, version 4) were stratified by AUC_0–42 exposure tertiles.

Linear logistic E–R modeling was used for the occurrence of at least one Grade ≥ 2 CRS event (by American Society for Transplantation and Cellular Therapy [ASTCT] and Lee 2014 grading criteria) ¹⁹ , ²⁰ :

$$\begin{array}{c}\text{logit}\left(p\right)=\log \left(\frac{p\left(\text{Event}=1\right)}{1-p\left(\text{Event}=1\right)}\right)={\beta }_0+{\beta }_{{\text{ROmax}}_{0-42}}·{\text{ROmax}}_{0-42}\hfill \\ +{\beta }_1·x_1+\dots +{\beta }_n·x_n\hfill \end{array}$$

here, β _ROmax0–42 is the coefficient describing the effect of RO_max0–42 on the likelihood of an event, and other parameters are as previously defined.

The E–R analysis of plasma IFN‐γ used data with a cutoff date of August 27, 2021 from 446 evaluable R/R NHL patients (33 in Group A and 413 in Group B), defined as having measurable IFN‐γ concentration before Cycle 8. Four hundred eighteen NHL patients were RO_max‐evaluable (31 in Group A receiving 0.2–2.8 mg q3w, and 387 in Group B receiving 0.4/1.0/2.8 to 1.0/2.0/60 mg). The Log₂‐maximal‐fold change from pre‐dose baseline of IFN‐γ (Log₂‐IFN‐γ, CFBL_max), defined as Log₂ (IFN‐γ_max) – Log₂ (IFN‐γ baseline) within respective dosing windows, was used as the pharmacodynamic (PD) biomarker endpoint.

Exploratory covariate modeling

Exploratory covariate analyses assessed the impact of patient baseline factors on the E–R relationships for CR and Grade ≥ 2 CRS, using AUC_0‐42 or RO_max0–42. Upon identifying a positive E–R relationship, additional covariates were evaluated using a stepwise forward addition and backward deletion process with the criteria of reduction in the objective function value (ΔOFV) ≥ 3.84 (P ≤ 0.05) for forward inclusion and a criterion of ΔOFV ≤6.63 (P ≥ 0.01) for backward deletion. Given the model's complexity and low sample size, the exploration scope was predefined and restricted to clinically significant predictors (Tables S8 and S9 ).

Integrated efficacy‐safety evaluation

A model‐based assessment of the overall clinical benefit‐to‐risk profile was performed by overlaying predicted efficacy (CRR, ORR) and safety (probability of Grade ≥ 2 CRS) for selected regimens. The results compared dose levels of two expansion cohorts (1/2/13.5 and 1/2/60/30 mg) and a simulated regimen of 1/2/30 mg (excluding 60 mg loading doses), supporting the adequacy of loading and target doses in the approved regimen. Four hundred thirty nine simulated concentration‐time profiles across dose levels were generated from the final population PK (popPK) model using nominal dosing and time points. AUC_0–42 was used for CR and OR, and RO_max0–42 for Grade ≥ 2 CRS. Nonparametric bootstraps of the models (N = 1,000 iterations, resampling with replacement) were performed. Simulated PK endpoint distributions predicted event probabilities, which were plotted with CIs and stratified by dose regimen groups to illustrate the model‐predicted therapeutic window.

RESULTS

The demographics of the E–R efficacy population (n = 159 patients with R/R FL who received ≥ 2 prior therapies) and E–R safety population (n = 439 patients with R/R NHL) are detailed in Tables S10 and S11 .

Clinical dose regimen

The IV dosing regimen for mosunetuzumab (Figure 1 ) included “step doses” of 1 and 2 mg on Cycle 1 Day 1 and 8 to mitigate acute CRS. “Loading doses” of 60 mg on Cycle 1 Day 15 and Cycle 2 Day 1 aimed to induce rapid tumor debulking (Figure S5 ) and maximize clinical response in R/R FL patients, resulting in high clinical response rates (60% CRR; 80% ORR) and durable responses (median duration of response: 22.8 months; median DOCR: not reached). ⁴ “Target doses” of 30 mg on Day 1 of each cycle from Cycle 3 onwards were intended to sustain the clinical response. The lower target dose of 30 mg was chosen to minimize overdosing and promote better long‐term tolerability after the initial tumor burden reduction.

Figure 1.

Clinical dose regimen of mosunetuzumab (1/2/60/30 mg) for treatment in R/R FL ≥ 2 lines of prior therapies. Intravenous mosunetuzumab was administered in 21‐day cycles with C1 step‐up dosing: 1 mg on C1D1, 2 mg on C1D8, 60 mg on C1D15 and C2D1, and 30 mg on D1 of C3 and onwards. Patients with a complete response by investigator assessment using the International Harmonization Project criteria completed treatment after C8, whereas patients with a partial response or stable disease continued treatment for up to 17 cycles. C, Cycle; CRS, cytokine release syndrome; D, day; FL, follicular lymphoma; R/R, relapsed/refractory.

Mosunetuzumab PK and ADA

Mosunetuzumab PK characteristics are described in detail by Bender et al. ¹⁶ The PK is linear, dose‐proportional, and time‐dependent in the dose range studied (0.2–60 mg) and in the clinically active dose range (≥ 1.2 mg). The serum popPK was well‐described by a two‐compartment model with time‐dependent clearance (CL) (Figure S6 ), reflecting reduction of target‐mediated drug disposition (TMDD) and cancer‐associated cachexia ¹⁷ over time, consistent with drug's therapeutic response. The terminal half‐life estimate was 16.1 days at steady state. No clinically meaningful baseline covariates were found for mosunetuzumab PK requiring dose adjustments (Table S4 ). No ADA were detected in 418 ADA‐evaluable patients. PopPK model parameters and PK metrics at the clinical dose are summarized in Tables S3 and S5 .

E–R for efficacy

E–R analyses for CRR and ORR were conducted in 159 patients with R/R FL across a wide dose range and AUC values (i.e., 0.2–60 mg and across 600‐fold of AUC_0–42 values [0.4–251 day*μg/mL]). The E–R model suggested that response increased with exposure (AUC_0–42) and plateaued at E ₉₀ of AUC_0–42 values around 92.7 and 38.4 day*μg/mL for CRR and ORR, respectively. The theoretical maximal clinical response (E _max) estimated was 68.6% and 81.4% for CRR and ORR, respectively (Figure 2 , Table S7 ).

Figure 2.

E–R (AUC_0–42) analyses for investigator‐assessed clinical response following IV administrations of mosunetuzumab monotherapy (GO29781 study). (a) Complete response rate. (b) Objective response rate. Filled circles represent the individual patient AUC_0–42 (green = 1/2/60/30 mg; blue = lower dose cohorts) and response assessment (0 = no event, 1 = event); orange solid line represents the E–R curve based on the final parameter estimates; orange shaded area represents the E–R model – estimated 90% CI. Percentages indicate the observed response rate (%; x/y = x responders out of y patients) within each exposure quartile. Open circles overlaid on top of the E‐R curve are the observed median probability of patients having a clinical response; error bars are the SE [sqrt(P*(1−P)/N)] at each exposure quartile. Horizontal lines at the bottom represent simulated AUC distributions (median with 95% CI) at selected dose levels (green = 1/2/60/30 mg; orange = 1/2/30 mg; black = 1/2/13.5 mg); simulations are based on the popPK model Empirical Bayes Estimates (EBEs). AUC, area‐under‐the‐concentration curve; AUC_0–42, AUC from time equal to 0–42 days post‐dose; CI, confidence interval; CR, complete response; E ₅₀, AUC_0–42 at 50% of maximal response; E ₉₀, AUC_0–42 at 90% of maximal response; E–R, exposure response; FL, follicular lymphoma; IV, intravenous; OR, objective response; N, sample size; Obs, observations; P, proportion; popPK, population pharmacokinetics. SE, standard error; sqrt, square root.

Overlaying E–R with simulated AUC_0–42 distributions at selected dose levels indicated the regimen of 1/2/60/30 mg is expected to achieve exposure levels corresponding to the efficacy plateau (i.e., E ₉₀) of the E–R curves (Figure 2 ). Observed clinical responses suggested higher response rates at the 1/2/60/30 mg dose compared with a lower dose of 1/2/13.5 mg. Model‐based simulations support the potential clinical benefits of 60 mg loading doses, compared with a simulated regimen of 1/2/30 mg (without the loading doses), with a higher CRR and ORR. The model‐predicted CRR was 63% and 58%, and ORR was 79% and 76% at 1/2/60/30 and 1/2/30 mg, respectively (Figure 4 ).

Figure 4.

Integrated understanding of mosunetuzumab clinical dose–response relationships predicted by modeling for CR, OR and Grade ≥2 CRS. (a) Predicted probability of CR and CRS (Grade ≥2) by dose. (b) Predicted probability of OR and CRS (Grade ≥ 2) by dose. Note: base models (i.e., without covariates) used for prediction. Filled circles represent the model‐predicted probability of response (blue for CR, green for OR, red for Grade ≥ 2 CRS) at selective dose levels; points connected by solid lines are medians; dashed lines connect geometric means; shaded areas represent 50%, 90%, and 95% confidence intervals. For the 1/2/60/30 mg dose regimen, the modeling predicts: a median CRR of 63.1% (95% prediction interval, 49.7–75.0); a median ORR of 79.1% (69.1–87.7); and a median Grade ≥ 2 CRS rate of 11.8% (5.27–23.2). For the 1/2/30 mg dose regimen, the modeling predicts: a median CRR of 58.3% (95% prediction interval, 39.90–68.8); a median ORR of 76.4% (60.9–84.4); and a median Grade ≥ 2 CRS rate of 9.75% (4.90–16.3). For the 1/2/13.5 mg dose regimen, the modeling predicts: a median CRR of 50.2% (95% prediction interval, 27.7–62.0); a median ORR of 71.0% (43.9–79.8); and a median Grade ≥ 2 CRS rate of 8.27% (4.60–12.3). CR, complete response; CRR, complete response rate; CRS, cytokine release syndrome; INV, investigator; OR, objective response; ORR, objective response rate.

PFS (INV‐assessed) dose–response analyses indicated a trend for prolonged survival at the higher mosunetuzumab dose, although the CIs overlapped at the dose levels assessed. Accordingly, increased exposure (AUC_0–42) correlated with longer PFS (Figure 5 a,b ). DOCR was similar across the expansion dosing regimens (B7: 1/2/13.5 and B11: 1/2/60/30 mg) (Figure 5 c ), while longer response duration was observed in the highest exposure (AUC_0–42) tertile (Figure 5 d ).

Figure 5.

Clinical efficacy; PFS and DOCR by dose (for B11 vs. B7 cohorts). (a) Investigator‐assessed PFS for the R/R FL population stratified by cohorts B11 (1/2/60/30 mg) and B7 (1/2/13.5 mg). (b) Investigator‐assessed PFS for the R/R FL population stratified by tertiles of mosunetuzumab AUC_0–42. Dose range included 0.2–2.8 mg (fixed q3w) in Group A and 0.4/1.0/2.8 to 1.0/2.0/60/30 mg (Cycle 1 step‐up dosing) in Group B. For PFS analysis: T1 of mosunetuzumab AUC_0–42 exposure (≤ 29.9 day*μg/mL); T2 of mosunetuzumab AUC_0–42 exposure (> 30.2 day*μg/mL and ≤ 131 day*μg/mL); and T3 of mosunetuzumab AUC_0–42 exposure (> 131 day*μg/mL). (c) Investigator‐assessed DOCR for the R/R FL population by cohorts B11 and B7. (d) Investigator‐assessed DOCR for the R/R FL population by mosunetuzumab AUC_0–42 tertiles. For DOCR analysis: T1 of mosunetuzumab AUC_0–42 exposure (≤ 44.8 day*μg/mL); T2 of mosunetuzumab AUC_0–42 exposure (> 46.8 day*μg/mL and ≤ 142 day*μg/mL); and T3 of mosunetuzumab AUC_0–42 exposure (> 142 day*μg/mL). AUC, area‐under‐the‐concentration curve; AUC_0–42, AUC from time 0–42 days; CI, confidence interval; DOCR, duration of complete response; FL, follicular lymphoma; NR, not reached; PFS, progression‐free survival; q3w, every 3 weeks; R/R, relapsed/refractory; T, tertile.

Exploratory E–R covariate analyses indicated that patients with larger baseline tumor sum of products of diameters (SPD) values or higher pre‐dose anti‐CD20 concentrations may have a lower probability of attaining CR (Figure S3 ). However, patients with the most extreme baseline anti‐CD20 (≥ 12.1 μg/mL) and baseline tumor SPD values (≥ 10,816 mm² [= 104 × 104 mm]), arguably a specific subpopulation with a lower inherent response probability and more aggressive disease, were rare in the study (N = 2/159 [1.3%]) (Figure S3F ).

E–R for safety

The safety of mosunetuzumab has been reported to be tolerable and manageable. ⁴ CRS were the most frequently reported AE, predominantly Grade 1–2, limited to Cycle 1, and were manageable and reversible. The frequency of observed Grade ≥ 2 CRS was low across dose groups (Figure 3 a, Figure S9 , Table S12 ). E–R analyses stratified by dosing groups showed that the positive E–R in Grade ≥2 CRS at the lower fixed dose regimens (0.05–2.8 mg) was mitigated by Cycle 1 step‐up dosing (Figure S9 ), allowing further escalation to >20‐fold higher levels in Group B without MTD.

Figure 3.

E–R relationships for occurrence of Grade ≥2 CRS (by ASTCT grading) in Cycle 1 by dosing windows. (a) Overlay of Cycle 1 E–R curves between dosing windows for occurrence of Grade 2 CRS following Cycle 1 step‐up dosing regimen (by ASTCT grading). Solid lines represent the model fit; shaded regions represent the 90% prediction intervals. Analysis based on N = 418 of patients with R/R NHL (GO29781 study, Group A and B). Note that the x‐range of each curve represents the available data range as a result of the dose ranging tested (i.e., Day 1 dose from 0.4 to 1 mg; Day 8 dose from 1 to 2 mg; Day 15 dose from 2.8 to 60 mg). (b) Overlay of E–R relationships between mosunetuzumab RO_max and plasma IFN‐γ (Log₂‐ IFN‐γ CFBL) modulations following Cycle 1 step‐up dose administrations on Day 1 (first dose), Day 8 (second dose), and Day 15 (third dose) in 387 patients with R/R NHL (GO29781 study; Group B; colored by dose). Blue solid line and blue circles represent the LOESS regression and the Log₂ maximal fold change IFN‐γ at corresponding RO_max level for patients with NHL at the first dose; purple solid line and purple circles represent the LOESS regression and the Log₂ maximal fold change IFN‐γ at corresponding RO_max level for patients with NHL at the second dose; red solid line and red circles represent the LOESS regression and the Log₂ maximal fold change IFN‐γ at corresponding RO_max level for patients with NHL at the third dose; note that the x‐range of each curve represents the available data range as a result of the underlying rituximab and/or obinutuzumab level and the dose range tested (i.e., Day 1 dose from 0.4 to 1 mg; Day 8 dose from 1 to 2 mg; Day 15 dose from 2.8 to 60 mg). ASTCT, American Society for Transplantation and Cellular Therapy; CFBL, change from baseline; CRS, cytokine release syndrome; E–R, exposure–response; LOESS, locally weighted scatterplot smoothing; NHL, non‐Hodgkin lymphoma; RO_max (%), model‐predicted maximum CD20 receptor occupancy (RO%) for each dosing window; R/R, relapsed/refractory.

E–R modeling shows a significant relationship between RO_max% and the occurrence of Grade ≥ 2 CRS within the first two cycles (42 days) (Figure S4 ). Further stratification by dosing cycles revealed a slight trend towards increased CRS with RO_max% following dose one (Cycle 1 Day 1), no trend following dose two (Cycle 1 Day 8), and an increased trend following dose three (Cycle 1 Day 15) towards the higher tested RO_max% range. This “rightward shift” in the E–R curve indicates a reduced mosunetuzumab effect for CRS over time and dosing administrations, dissipating by the end of Cycle 2 (Figure 3 a ). By ASTCT grading, CRS Grade ≥2 events were observed for dose Cohorts B7 (1/2/13.5 mg; N = 1/73; 1.37%) and B11 (1/2/60/30 mg; N = 33/198; 16.7%) for RO‐evaluable patients. At the clinical approved regimen of 1/2/60/30 mg, the corresponding RO_max% at the selected dose levels mapped to the bottom of the E–R CRS curves (Figures 3 a and 4 ), indicating the appropriate selection of each respective dose level for CRS mitigation.

The exploratory E–R covariate analysis for CRS identified RO_max0–42, “other” tumor histology (e.g., Mantle Cell Lymphoma [MCL] and Richter's Syndrome [RS]), and age (64 years) as significant covariates for the E–R relationship for Grade ≥ 2 CRS (Figure S4 ). Increased risk was positively correlated with increasing RO_max0–42 and “other” histology, and negatively correlated with age. The E–R curves for each subset are displayed in Figure S10 . The overall rate of Grade ≥ 2 CRS remained low regardless of age. The E–R relationships indicated a mild relationship emerging at the RO_max0–42 range achieved with the intended registration dose and schedule of 1/2/60/30 mg, supporting the target dose selection with minimized safety risks.

Step‐up dosing of mosunetuzumab did not lead to exposure‐dependent increases in Grade ≥ 3 neutropenia, infections, or AEs (Figure 6 , Table S6 ). Higher exposure did not lead to earlier onset of AEs. The step‐up dosing strategy successfully mitigated neurological AEs (NAEs). Data from Group B in GO29781 indicated a low risk of NAEs with mosunetuzumab, with no apparent relationship between exposure metrics and the frequency or onset of NAEs (data not shown).

Figure 6.

Time to event analyses for selected Grade ≥ 3 AEs by AUC_0–42 tertiles in patients with R/R NHL following IV mosunetuzumab administration using the step‐up dosing regimen (group). For analysis: T1 of mosunetuzumab AUC_0–42 exposure (≤ 28.5 day μg/mL); T2 of mosunetuzumab AUC_0–42 exposure (> 28.5 day μg/mL and ≤ 121 day μg/mL); and T3 of mosunetuzumab AUC_0–42 exposure (> 121 day μg/mL). AE, adverse event; AUC, area‐under‐the‐concentration curve; AUC_0–42, AUC from time 0–42 days; CI, confidence interval; IV, intravenous; NHL, non‐Hodgkin lymphoma; R/R, relapsed/refractory; T, tertile.

E–R for PD biomarkers

IFN‐γ, a proximal marker for target engagement following T‐cell activation, was selected as the primary PD biomarker for E–R characterization. Similar patterns were observed for other cytokines like interleukin‐6 (IL‐6) (data not shown). In patients receiving the Cycle 1 step‐up dosing regimen, cytokine modulation was acute, transient, and primarily limited to the first cycle, despite ongoing mosunetuzumab accumulation. Initial peak PD modulation of IFN‐γ was evident following the Cycle 1 Day 1 dose (0.4 mg and higher), with a smaller second peak at the highest doses on Cycle 1 Day 15 (Figure S8 ). Median Log₂ maximal fold change from baseline for IFN‐γ ranged from 0.4 to 2.5 following the Cycle 1 step‐up dosing regimen. Logistic regression showed a positive relationship between RO_max% and Log₂‐IFN‐γ CFBL following Cycle 1 doses, independent of NHL histology (Figure 3 b ). This relationship was not observed at subsequent doses (Cycle 1 Day 8 or Cycle 1 Day 15). Figure 3 shows the rightward shifts in the E–R curves throughout the step‐up dosing regimen of mosunetuzumab in Cycle 1. The dissipating E–R effect over time is consistent with the CRS mitigation using the step‐up dosing regimen.

Integrated E–R for efficacy and safety

Integrated E–R analyses of efficacy and safety indicated a broad therapeutic window for mosunetuzumab using the Cycle 1 step‐up dosing regimen (Figures 2 , 3 and 4 ). The 1/2/60/30 mg dose achieved PK exposures at the E–R plateau, maximizing response with model‐estimated CRR and ORR (90% CI) of 63.1% (49.7–75.0) and 79.1% (69.1–87.7), respectively (Figure 4 ). PFS and DOCR further established the clinical benefit of this dose compared with lower doses (B11 at 1/2/60/30 mg [n = 90], and B7 at 1/2/13.5 mg [n = 46]; Figure 5 ). At the clinical approved regimen of 1/2/60/30 mg, the corresponding RO_max% at the selected dose levels mapped to the bottom of the E–R CRS curves (Figures 3 a and 4 ), indicating the appropriate selection of each respective dose level from an acute safety perspective. The clinical dose regimen had a favorable safety profile with a less than 12% probability of a patient experiencing transient Grade ≥2 CRS (Figure 4 ), <3% Grade ≥3 CRS, and minimized to the first 1–2 cycles. No exposure‐dependent increases in onset or occurrence of Grade ≥3 neutropenia, infections, or AEs (Figure 6 , Table S6 ).

DISCUSSION

This study presents the first comprehensive E–R assessment of mosunetuzumab in patients with R/R FL with ≥ 2 prior therapies. As an immune agonist, mosunetuzumab is active at low target engagement levels (≥ 1% RO%, data not shown) and clinically active at a low dose of 1 mg (q3w) and higher. The approved regimen of 1/2/60/30 mg achieved PK exposures (AUC_0–42) at the plateau of the E–R curves for rapid tumor debulking and maximal clinical response. Mosunetuzumab was administered for eight cycles for patients with CR, and up to 17 cycles for those with PR or SD at the end of Cycle 8. Durable clinical responses were evident beyond the fixed treatment duration. More than half of all responders had ongoing responses at 18 months after their first response. ¹⁸ Out of 16 patients with R/R FL who received more than eight cycles, six displayed a late initial response or deepening of response, including five who achieved CR after more than eight cycles. ⁴ These data support the fixed duration of mosunetuzumab and the potential benefit of continuing treatment in patients who demonstrate disease control but have not reached CR at Cycle 8.

The risk of CRS is a common challenge for all T‐cell engaging therapies and represents a multifactorial issue which is molecule‐ and disease‐dependent. ¹² Multiple clinical approaches are being actively explored to mitigate clinical CRS risks, including the use of steroids, blockade of myeloid‐derived cytokines (e.g. tocilizumab), blockade of T‐cell derived cytokines (e.g., TNF‐α), and kinase inhibitors. ²¹ In the case of mosunetuzumab, CRS was predominantly low grade and confined to Cycle 1. ¹⁸ The use of Cycle 1 step‐up dose regimen and corticosteroid premedication effectively mitigated the risk of CRS which differentiates from other T‐cell engaging BiSp. This favorable safety profile reduces the healthcare burden of administering T‐cell engaging BiSp therapies to patients. The selection of the low and intermediate step‐doses of 1 and 2 mg was based on achieving low exposures, contributing to low levels of cytokine secretion, posing minimal risk of Grade ≥ 2 CRS (i.e., at the bottom of the E–R curves for CRS). Importantly, these doses produced sufficient exposures to induce CRS desensitization for each subsequent dosing (Figure 3 ; E–R curve shifted to less potency with repeated dosing), permitting the administration of the 60 mg loading doses for maximal efficacy without increasing the CRS risk. At the clinical regimen, the risk of CRS was low (< 12% Grade ≥ 2 CRS (Figure 4 ), < 3% Grade ≥ 3 CRS), the use of tocilizumab was low (7/90) patients in the B11 expansion cohort, ¹⁸ contrasting with higher rates observed with CAR‐T cells. ⁵ , ⁷ , ²² No exposure‐dependency was identified for other AEs, including Grade ≥ 3 neutropenia, Grade ≥ 3 infections, or Grade ≥ 3 AEs (Figure 6 , Table S6 ). Dosage interruptions or modifications of mosunetuzumab due to an AE occurred in 37% of patients, primarily due to neutropenia, infection, and CRS. ² MTD was not reached.

E–R analyses have elucidated the different PK drivers for the clinical safety and efficacy of mosunetuzumab, supporting the future design of dosage regimens. Acute safety concerns, such as CRS, are primarily due to high initial target engagement levels (RO_max), necessitating a step dose strategy to mitigate early risk. In contrast, clinical responses are driven by cumulative PK exposure (AUC_0–42), suggesting that initial loading doses can expand the therapeutic window and optimize mosunetuzumab's benefit‐to‐risk ratio, particularly in patients with potential risk factors, such as higher baseline tumor burden and higher residual anti‐CD20 concentrations. These findings are consistent with the temporal profiles of clinical data, highlighting two distinct timelines for efficacy and safety observations: an initial rapid onset of tumor response in the first two cycles (Figure S5 ) followed by a durable effect, contrasting with CRS which occurs in a transient and acute fashion. Such observations align with prior preclinical and translational research, suggesting initial cytokine release which diminished with repeated doses, in contrast to the continued cytotoxicity via release of cytotoxic granules (e.g., perforin and granzyme B). ¹³ These pharmacological principles also apply to subcutaneous monotherapy administrations and could guide future development of mosunetuzumab or other T‐cell engaging BiSp in R/R FL. For combination therapies, the pharmacological drivers would depend on the combination agent and would need case‐by‐case evaluation.

Exploratory multivariate covariate modeling for efficacy and safety identified higher risk populations and areas for further research. The analyses predicted lower CR rates in populations with increased residual anti‐CD20 concentrations and larger baseline tumor sizes. In rare cases (< 1.5% of patients) with residual anti‐CD20 serum concentration ≥ 12.1 μg/mL and baseline tumor SPD ≥ 10,816 mm², the model predicted a maximum CR rate of 50% (Figure S3 ). Future combination strategies, such as mosunetuzumab plus lenalidomide, could enhance treatment benefits for these patients. ²³

Covariate modeling for CRS suggested an elevated risk of Grade ≥ 2 CRS in patients with MCL and RS histologies, increased RO_max%, and younger age (≤64 years) (Figure S4 ). No mosunetuzumab dose adjustments are recommended based on these covariates. Sehn et al. ²⁴ found additional factors like total metabolic tumor volume (TMTV), increased bone marrow TMTV, and higher white blood cell count as predictors of Grade ≥ 2 CRS. These findings align with previous QSP modeling and CAR‐T induced CRS observations. ⁷ , ¹⁰ , ¹¹ , ²⁵ A composite risk score for CRS was developed for glofitamab to inform CRS risk stratification for patient monitoring. ²⁶ Early changes in TNF‐α was identified as one of the predictors of CRS. Future studies may integrate wider datasets across BiSp and NHL histologies, utilizing machine learning and deep learning algorithms to analyze multi‐dimensional, nonlinear systems over time. This approach could better link PK, target engagement, biomarker modulation, and CRS risk, enabling a more comprehensive exploration of patient baseline factors and CRS predictors.

Mosunetuzumab is eliminated via non‐specific intracellular catabolism and target‐mediated elimination, similar to conventional antibodies. ²⁷ , ²⁸ , ²⁹ Its larger volume of distribution (V1 of 5.49 L) suggests binding to accessible targets post‐IV administration. Unlike higher doses of anti‐CD20 mAbs like rituximab or obinutuzumab, ²⁸ , ³⁰ , ³¹ mosunetuzumab exhibits linear PK, consistent across disease histologies. No ADA was observed, aligning with its B‐cell depleting MOA. No dosing modifications are recommended based on intrinsic factors such as body weight, age, sex, race, organ impairment, or disease histology. ² , ³ , ¹⁶

Due to challenges in obtaining tumor tissue samples, we explored peripheral cytokine data as an alternative pharmacodynamic biomarker for response. Mosunetuzumab stimulated cytokine release, including IFN‐γ and IL‐6, demonstrating its MOA. Cytokine elevations were RO_max‐dependent and aligned with CRS occurrence, supporting their use in studying CRS (Figure 3 ). Early‐cycle cytokine surges indicated initial target engagement with peripheral B cells. However, no peripheral cytokine release was observed in later cycles despite ongoing clinical response. This is plausibly due to the low tissue‐to‐blood ratio (< 1% for lymph nodes), as estimated by the mosunetuzumab QSP model and published by Hosseini et al. ¹² Thus, while peripheral biomarkers support the MOA, they may not accurately reflect cytotoxic T‐cell activity at tumor sites.

The model‐derived RO% metric effectively characterized inter‐patient variability in acute target engagement relative to residual anti‐CD20 therapy levels (e.g., rituximab or obinutuzumab) from prior treatments. Residual anti‐CD20 mAbs compete with mosunetuzumab for CD20 binding. Using clinical RO% as an exposure metric instead of dose or concentration revealed crucial exposure–response relationships for dose selection. This clarified the absence of dose–response associations for PD biomarker modulations and CRS due to competing pharmacological binding, which was overcome by higher dosages (e.g., loading doses) and over time given its reversible nature. The clinical RO% methodology can also be applied in other contexts, particularly for aggressive NHL histologies and understanding initial target engagement kinetics for acute safety events. ¹⁶

Study limitations include a relatively modest sample size (n = 159) for covariate modeling of E–R efficacy. Also, restricted dose variations of the Cycle 1 Day 8 step‐up dose limited the investigation of the benefit–risk potential using a quicker ramp‐up regimen, an aspect currently under examination for subcutaneous mosunetuzumab. ³² The comprehensive cytokine profiles across cycles were primarily confined to earlier cycles and specific cytokines, thereby restricting the scope of PK‐PD modeling for PD effects.

In conclusion, mosunetuzumab represents a well‐tolerated, effective T‐cell engaging BiSp, exhibiting a high response rate and durability, akin to CAR‐T cell therapies. Mosunetuzumab, however, can be administered as an outpatient regimen, ² , ³ and offers an immediately available, fixed‐duration alternative without the need for ex vivo manipulation, cytoreductive chemotherapy and risks associated with genetic modification. The novel model‐directed development strategy, coupled with the key insights on dose selection and methodology outlined here, may be applicable to similar upcoming T‐cell engaging BiSp programs.

FUNDING

The clinical trial conducted and the research is funded by Roche.

CONFLICT OF INTEREST

C.‐C.L., B.B., F.L., D.T., B.W., Z.L., S.V., A.K., K.P., E.P., L.‐Y.H., P.C., C.L., S.Y., M.W. are currently employed by Genentech, Inc., F. Hoffmann‐La Roche Ltd or were employed by Genentech during the conduct of this research and hold company stocks. H.H. is employed by Hoffmann‐La Roche Ltd. R.D. and J.W. are paid consultants to Genentech, Inc., F. Hoffmann‐La Roche Ltd. The authors confirm that these affiliations do not influence their scientific judgments in relation to the work described in this paper. Genentech, Inc. has filed patent applications related to TDBs.

AUTHOR CONTRIBUTIONS

C.‐C.L., B.B., J.W., F.L., D.C.T., B.W., R.D., Z.L., A.K., H.H., E.P., P.C., S.Y., and M.C.W. wrote the manuscript; C.‐C.L., B.B., J.W., A.K., H.H., K.P., E.P., P.C., C.L., S.Y., and M.C.W. designed the research; C.‐C.L., B.B., J.W., F.L., D.C.T., B.W., R.D., Z.L., A.K., H.H., K.P., E.P., L.‐Y.H., P.C., C.L., S.Y., and M.C.W. performed the research; C.‐C.L., B.B., J.W., F.L., B.W., Z.L., S.V., A.K., H.H., K.P., E.P., L.‐Y.H., S.Y., and M.C.W. analyzed the data.

Supporting information

Data S1