Clinical Pharmacology of Cytokine Release Syndrome with T-Cell–Engaging Bispecific Antibodies: Current Insights and Drug Development Strategies

Immune-targeting therapeutics have changed the cancer treatment landscape over the past three decades, the newest of which are T cell–engaging bispecific antibodies (T-BiSp). Since the first T-BiSp approval (blinatumomab) in 2014, seven additional T-BiSp have been approved by the FDA, all as monotherapy in metastatic or relapsed/refractory settings. The present T-BiSp landscape includes more than 100 T-BiSp in clinical development across various tumor targets, making T-BiSp poised to become major therapeutic modalities in oncology (1).

A major safety concern for T-BiSp is cytokine release syndrome (CRS; ref. 2). CRS is an on-target adverse response following T-cell activation or engagement, resulting in elevated systemic release of cytokines (2). Clinically, CRS presents as fever with or without hypotension, hypoxia, and organ dysfunction and is graded per the 2019 American Society of Transplantation and Cellular Therapy criteria (3, 4).

With growing interest and investment in T-BiSp as novel anticancer agents, dose regimen optimization to manage CRS risk is imperative. This review provides a snapshot of the rapidly evolving understanding of T-BiSp-induced CRS focused on clinical pharmacology and drug development aspects, including mitigation strategies, characterization, and prediction. Knowledge gaps and future research opportunities are highlighted.

Historically, the recommended phase II dose (RP2D), a key objective of phase I trials, for anticancer agents was determined by defining an MTD (5). For molecular targeted agents, however, the MTD is often not reached; i.e., maximal dose-dependent efficacy is below tolerability thresholds (5, 6). For immune agonists like T-BiSp, the optimal efficacious dose can be much lower than the MTD. The 2021 FDA-initiated Project Optimus placed greater emphasis on dose optimization during drug development and encouraged a robust evaluation of doses and schedules in order to optimize the benefit–risk profile of novel oncologic therapeutics (7).

Figure 1A shows the efficacy and CRS profiles of approved T-BiSp molecules. Recently authorized liquid tumor–targeting T-BiSp had 60% to 80% with overall response rates (ORR) of 39% to 79%. Tebentafusp, authorized for metastatic uveal melanoma, had a CRS occurrence of 89% (72% grade 2+) and 9% ORR, but further dose optimization was not requested by FDA (8). The acceptable CRS risk for T-BiSp molecules seems to vary depending on the disease setting and available therapeutic options.

Dose optimization requires considering the totality of evidence including safety, efficacy, patient burden, and treatment landscape for a disease. For T-BiSp molecules, optimization involves step-up dosing (SUD), target dose selection, dosing frequency, duration of treatment, and administration route, aimed at improving short- and long-term tolerability (including CRS risk) while maximizing efficacy. Below we discuss strategies and their outcomes for T-BiSp molecules approved and in development.

SUD involves the administration of small doses with submaximal efficacy prior to introducing the intended target dose. SUD attenuates the cascade of cytokine release and reduces overall CRS risk (2). This is explained by immune desensitization of cytokine release with repeated exposure (9, 10) and reduced T-cell activation following target depletion (11, 12). Importantly, SUD has facilitated decoupling of CRS risk and treatment activity such that similar CRS risk can be obtained following different target doses (Table 1).

All approved T-BiSp include SUD, and, consequently, CRS occurrence is low or minimal beyond the first targeted efficacious dose (Fig. 1B). Various dose-finding strategies are used during T-BiSp development, including increasing the number of steps with escalating target doses, varying SUD schedules for a single target dose, and split dosing, which involves splitting a step dose into two equal or unequal doses given closely together (Table 1). Two-step SUD schedules are most common among approved T-BiSp, administered 3 days apart for once weekly regimens and 7 days apart for every 3 weeks regimens (Fig. 1B).

In approved SUD schedules, CRS rates either increase with time, peaking at the first full dose (FD1), or decrease with time, peaking at the first step dose (SUD1; Fig. 1B). The prospective strategies used by investigators for dose selection are unknown. Retrospective evaluation shows that molecules with ascending CRS rates during SUD tend to have greater fold increases from SUD1 to FD1 (1 to 48–80 fold) compared with molecules with descending CRS rates (1 to 3.4–25 fold; Supplementary Table S1). This difference in CRS risk progression may reflect varying degrees of desensitization, which depends on target level and immune status. A mechanistic explanation of this phenomenon suggests that greater SUD1 amounts deplete the immune cell pool faster, reducing cytokine secretory capacity with future doses (10). Molecule, disease, and dosing interval factors may also play a role.

Subcutaneous administration of T-BiSp was developed to improve safety and convenience for patients and caregivers (13, 14). Subcutaneous administration delays and reduces peak drug concentrations compared with intravenous administration, potentially lowering cytokine levels and CRS risk as found for alnuctamab (15, 16). A common development strategy is to develop subcutaneous administration in parallel with or after intravenous development, pivoting as early as phase I (Table 1). Epcoritamab is a unique example in which investigators initiated dosing in humans with a subcutaneous formulation (15, 17).

The extent of CRS risk mitigation by subcutaneous administration is not fully established and varies across T-BiSp molecules (18). Although investigators have alluded to better CRS profiles as the basis for pursuing subcutaneous dose expansion, external evaluation is difficult given differences in dose amounts and exposures of subcutaneous and intravenous arms. Without adjusting for exposure differences, a meta-analysis of B-cell maturation antigen–T-BiSp for multiple myeloma found subcutaneous formulations had higher grade ≤2 but lower grade 3+ CRS events than intravenous formulations (19). Preliminary phase II results with mosunetuzumab’s subcutaneous formulation reported 20% CRS (6.6%, grade 2+) with the subcutaneous RP2D compared with 39% (17%, grade 2+) with the intravenous RP2D (20, 21). To our knowledge, this is the best exposure-equivalent intramolecule comparison of subcutaneous versus intravenous CRS risk available publicly; however, administration of the FD1 on day 8 with subcutaneous versus day 15 with intravenous may confound the comparison. Overall, the available evidence suggests that subcutaneous administration has promising potential to lower high-grade CRS risk.

Evidence of the effect of maintenance dose frequency on CRS risk in T-BiSp therapy has not been extensively published. More frequent dosing leads to higher drug exposure and increases CRS risk, whereas longer intervals between doses may lead to loss of immune desensitization, potentially causing CRS resurgence. Therefore, minimizing CRS risk while keeping the disease controlled is a primary goal for maintenance dose regimen design. Dose-ranging was performed for every 2 weeks and once weekly teclistamab, but CRS by dosing schedule was not reported (14). Talquetamab is the only approved T-BiSp with two maintenance dose frequencies. Overall CRS and CRS following the full dose were similar between once weekly and every 2 weeks dosing (Fig. 1; ref. 22). Elranatamab is currently administered once weekly (23); a phase I/II study with grade 2+ CRS as the primary endpoint is evaluating every 2 weeks and monthly schedules and alternative SUD (24). This is supported by a prior mechanistic model demonstrating preserved effectiveness when transitioning from once weekly to every 2 weeks dosing (23).

CRS is an immunologic cascade triggered by direct target cell lysis or therapeutic activation of T cells. These T cells produce cytokines like IFN-γ and TNFα, which activate innate immune cells such as macrophages and endothelial cells. These cells then release large amounts of proinflammatory cytokines like IL-6, perpetuating a cytokine storm through a positive feedback loop (2). Therapeutic interventions include blocking monocyte-released IL-6 (e.g., tocilizumab) and IL-1 (e.g., anakinra), blocking upstream T cell-released TNFα, and using JAK inhibitors, which prevent cytokine release without compromising T-cell cytotoxicity. Early implementation of prophylactic cytokine blockade is also recommended to reduce early-cycle cytokine release without impacting T-cell cytotoxicity (9). The field is rapidly evolving, with new agents and combinations being explored.

All approved T-BiSp require pretreatment with corticosteroids, antipyretics (acetaminophen), prehydration, and antihistamines, typically through the first 1 to 2 full doses (Supplementary Table S2). Dexamethasone is the preferred corticosteroid for glofitamab and elranatamab, supported by lower CRS occurrence (23, 25). Administration of targeted antibodies to deplete peripheral B cells prior to T-BiSp administration in B cell lymphomas may also decrease CRS risk. Despite strong preclinical evidence, SUD was still required to reduce rates of grade 2+ CRS in obinutuzumab-pretreated patients receiving glofitamab (25–27). Emerging data suggests dexamethasone may offer more CRS protection compared with other corticosteroids, but the optimal role, context, and dosing schedule with or without other pretreatment agents remain unknown (28, 29).

Tocilizumab prophylaxis is predicted to block the IL-6 signaling pathway for 10 days after a single 8 mg/kg dose (30). When administered prophylactically 4 hours prior to the first or second teclistamab step dose, CRS occurrence reduced from 73% to ∼30% (31, 32). When administered reactively as a CRS rescue agent, the dosing regimen of tocilizumab has been borrowed from CAR-T-induced CRS protocols and experience, which may not fully apply to T-BiSp cases (33). Modeling of mosunetuzumab and glofitamab data suggest that ≤2 tocilizumab doses per event, not exceeding three doses over 6 weeks, is sufficient to maintain tocilizumab activity and safe exposure (34), compared with up to four doses per event following CAR-T therapy (33). The pharmacologic rationale for administering multiple doses of tocilizumab in managing CRS is not strongly established, in general; it largely depends on clinician discretion and the resolution of CRS symptoms rather than a definitive pharmacologic necessity.

Kinase inhibitors have CRS mitigation potential by suppressing cytokine release downstream of TNFα signaling and offering antitumor activity (35, 36). In vitro assays revealed potent cytokine release inhibition while preserving T cell killing of PI3K, JAK, and mTOR inhibitors (36, 37). Itacitinib (JAK1 inhibitor) reduced overall CRS by 22% and grade 2+ CRS by 39% in a randomized placebo-controlled trial of axicabtagene ciloleucel in relapsed/refractory large B-cell lymphoma (38). POLB-001, a novel p38 MAPK inhibitor, significantly reduces cytokines and is in clinical development for inflammatory diseases and CRS, with plans for multiple myeloma (39).

Multiple cytokines including IL-6, IL-1, IL-2, IL-8, IL-10, IFN-γ, and TNFα are elevated in patients with CRS (Fig. 2A; ref. 40). IL-6, IL-10, and IFN-γ dynamics are included in most T-BiSp licensing applications and typically increase after step-up and target doses to varying degrees (Supplementary Fig. S1). Lower peaks are generally observed with subcutaneous administered T-BiSp.

IL-6 is highly elevated after dosing and has emerged as a key biomarker for the pharmacologic characterization of CRS (40). However, mean dose-period IL-6 peaks do not consistently trend with CRS rates for approved molecules (Fig. 2B). IL-6 trended with CRS for mosunetuzumab tebentafusp and tarlatamab, but IL-10 and IFN-γ better trended with CRS for glofitamab and elranatamab. Although post–step-dose IL-6 concentrations were not reported for epcoritamab recent cycle 1 dose optimization efforts have shown a reduction in CRS, which is associated with lower IL-6 levels (41). Tocilizumab administered prophylactically or for treatment increases IL-6 levels by saturating receptor-mediated IL-6 clearance and may alter the IL-6 to CRS relationship as found in teclistamab patients (30, 42, 43). Therefore, tocilizumab needs to be accounted for in IL-6 to CRS associations. TNFα peaks earlier than IL-6 and may be a better marker of CRS severity (44, 45). A multicytokine model including IL-6 and IL-8 as proinflammatory and IL-10 as anti-inflammatory effects may also aid CRS grade prediction (46).

Other biomarkers may aid CRS understanding and prediction. MCP-1, CXCL10, and MIP-1a(b) expression is induced by CD20/CD3 bispecifics (35). Clinical observations following CD19 chimeric antigen receptor T-cell (CAR-T) therapy and cell assays with CD3/carcinoembryonic antigen (CEA) bispecific suggest a role for angiopoietin-2 and von Willebrand factor in severe CRS (36, 47). Furthermore, the immune status for CD4⁺, CD8⁺, and regulatory T-cell populations may also affect individual CRS responses (48, 49).

Quantitative clinical pharmacology (qCP) approaches are powerful tools in drug development. They identify key contributors to clinical efficacy and safety, provide insights into untested clinical scenarios, and design safe and effective clinical trials with an increased probability of success. Table 2 provides examples of qCP approaches in drug development for CRS mitigation. Applying multiple methods synchronously in a parallel, sequential, or cross-talk manner is important to drive efficient and strategic decision-making during development (12, 48).

Mechanistic translational modeling (i.e., “bottom-up” approaches), such as quantitative systems pharmacology (QSP) and physiologically based pharmacokinetics (PBPK), is valuable when clinical experience is lacking or to leverage mechanistic insights. When linked with drug pharmacokinetics (PK) and trimer formation, QSP has been applied to predict cytokine and tumor dynamics under various SUD schedules to guide dose regimen optimization (10, 12, 50, 51). Additionally, it can be adapted to predict drug and disease interactions with combination regimens (52) and offer insights at the tissue level when combined with PBPK. Applications beyond SUD objectives include dose prediction in a new disease (11), formulation (53), and patient population (53). The future of T-BiSp drug development will likely focus on combination regimens to expand the therapeutic index and improve the benefit/risk profile, such as co-stimulatory agonistic bispecifics to enhance T-cell proliferation and cytotoxicity (54), agents to attenuate CRS response (36), and dual or triple T-BiSp therapy to diversify target engagement. Multitherapy QSP models will enable simultaneous prediction of efficacy and CRS risk with combination regimens, as applied for T-BiSp plus checkpoint inhibitors [e.g., anti-PD1/PDL1 (52, 55)], to streamline dose, regimen, and sequencing (56).

Empirical/statistical modeling of CRS (i.e., “top-down” approaches), such as logistic regression, is commonly applied to support dose selection in drug approvals. Investigators assess CRS endpoints using pharmacokinetic (PK) metrics [e.g., maximum concentration (Cmax) or derived receptor occupancy (RO)] at early dose periods, aligning with the CRS mechanism induced by T-BiSp. In contrast, efficacy endpoints have been described with early-cycle or steady-state PK metrics (i.e., cumulative AUC and average concentration) or trimer formation kinetics (Supplementary Fig. S2; refs. 10, 11). For most T-BiSp, there was a positive trend between Cmax/ROmax and the probability of CRS at SUDs and FD1 if enough CRS events were present (Table 3; Supplementary Fig. S2). No CRS E-R relationships were detected with talquetamab and epcoritamab levels (22, 57, 58) and may be due to a limited dose/concentration range and high response variability. Notably, glofitamab and mosunetuzumab used RO% instead of drug levels, resulting in a wider range of input exposure values given the interpatient variability in circulating anti-CD20 drugs (20, 25). Simulated trimer levels could be used to characterize CRS E-R as done for epcoritamab ORR E-R (57), but examples do not exist publicly. Postauthorization, time-to-CRS event modeling aided further optimization of epcoritamab SUD (59).

Multivariate predictive modeling could further inform risk-stratified patient monitoring or individualized dosing. Analyses for mosunetuzumab, glofitamab, teclistamab, and elranatamab reported various covariates associated with CRS including disease status and treatment history, target level, and effector cell status (Table 3). A CRS predictive model from 715 patients treated with T-BiSp identified ALL disease, prior infections, recent diagnosis, lower lactate dehydrogenase, and others associated with very high CRS risk (60). A risk stratification tool for grade 2+ CRS with the first glofitamab dose in aggressive non–Hodgkin lymphoma histologies excluding mantle cell lymphoma identified baseline lactate dehydrogenase, leukocyte count, age, cardiac morbidity, bone marrow and peripheral blood infiltration, Ann Arbor stage, and tumor burden as risk factors (61). The 5-parameter CRS risk prediction tool has been prospectively validated in aggressive non–Hodgkin lymphomas across several mono and combo glofitamab phase I trials with 94% true-positive prediction. The Cmax of the prior dose and tocilizumab use have also been shown to be important factors for predicting CRS in patients treated with teclistamab and elranatamab (43, 62). Other potential predictive factors include a proinflammatory tumor microenvironment (63), serum inorganic phosphate and magnesium level changes (64), and endothelial activation and stress index (EASIX) score (65), which have demonstrated predictive value in CRS post–CAR-T.

Platform modeling to further establish the link between drug exposure, trimer formation, biomarkers (e.g., IL-6), and CRS would aid RP2D optimization and translational predictions across different molecules. Hosseini and colleagues (12) developed a QSP model using preclinical mosunetuzumab data, calibrated to clinical blinatumomab data, to predict cytokine dynamics with untested mosunetuzumab dosing schedules. This could be extended to include an IL-6 to CRS relationship (59) to further predict CRS risk. A pan-molecule PK-cytokine-CRS model could thus inform discovery and clinical development by providing optimal PK targets and biomarker thresholds, enabling early decision-making.

Lastly, AI/ML methods are flexible in handling complex data, including correlated, nonlinear, and interdependent covariates thus could support CRS characterization and prediction. These methods can facilitate large-scale covariate searches and enable the identification of higher-dimensional relationships over time, which is challenging to achieve with traditional methods. Examples include time-to-event analysis (66), pharmacogenetic biomarker identification (67), survival prediction from tumor dynamics (68), and deep learning that conserves the time-dependent state (69). Such applications require large-scale clinical trial data, necessitating careful consideration for data handling (e.g., missing data) and practical judgment when interpreting data pooled from multiple clinical trials.

Our review highlights that CRS remains a key safety concern for T-BiSp and integral to efficient drug development. Despite the recent availability of guidelines for assessing and managing CRS in T-BiSp therapies (70), it is acknowledged that creating high-level evidence-based guidelines is still premature. Balancing the benefit-risk ratio for this on-target adverse effect requires careful consideration. Clinical development goals aim to minimize CRS frequency and severity, thereby circumventing the need for intensive clinical interventions and hospitalization. SUD has effectively mitigated CRS with T-BiSp, limiting most events to grades 1 to 2. Despite early promising data, the effectiveness of subcutaneous administration alone in mitigating CRS is not yet proven. Pretreatment with steroids, tocilizumab, and potentially kinase inhibitors are additional tools, but the optimal dose and timing for CRS prevention requires further study of the pharmacologic interactions relevant to the dosing schedule, molecule, and disease contexts.

QCP has significantly influenced dose and regimen design to deliver safe and effective T-BiSp agents. Mechanistic-based approaches (i.e., QSP, PBPK, and mPKPD) have accelerated early clinical development, designed SUD schedules to mitigate CRS, selected first-in-human doses, and extrapolated intravenous to subcutaneous doses. CRS E-R has been successfully characterized with drug levels and RO. Creating a universal predictive algorithm for CRS is challenging because of its multifactorial nature. CRS occurrence and patterns are influenced by molecule design (e.g., modality and affinity), disease context (e.g., target expression, microenvironment, and tissue distribution), patient population (e.g., tumor types), and clinical practices (e.g., prior treatments, dosing regimens, premedications, pretreatments, and rescue). These factors result in unique exposure–response and dose–response relationships. Developing a universal modeling platform that accounts for these variables could be valuable for future research. Future efforts using trimer formation may further integrate efficacy/risk predictions. AI/ML can expedite predictive modeling platforms, facilitate multivariate covariate searches, and enable large-scale use of longitudinal PK-biomarker-CRS data.

Solid tumor–targeting T-BiSp represent a large proportion of T-BiSp development (1). A key challenge is the broader target expression patterns and the mitigation of on-target off-tumor toxicities, including CRS (71). Preliminary data suggested higher CRS risk for some solid tumor T-BiSp. Mitigation strategies successful in liquid tumors may be applied to solid tumor development for further risk optimization (72). Tebentafusp, one of the earliest approved T-BiSp, was authorized with high-grade 2+ CRS occurrence (77%), yet the FDA deemed further dose optimization unnecessary (8). Newly authorized tarlatamab for small cell lung cancer reported 55% CRS (19% grade 2+), in line with CRS rates in liquid tumor T-BiSp (72). Clinical studies and QSP model simulations indicate that CRS in tebentafusp-treated patients is primarily driven by cytokine production in the skin rather than in tumor tissue, due to a larger immune cell pool in the skin (10). This complicates CRS prediction and management for solid tumor targets with broader expression patterns in both healthy and tumor tissues. Given that direct measurements of RO may not be feasible with solid tumors, QSP modeling offers a practical method for CRS prediction by incorporating tumor-associated antigen expression levels and simulating various dosing or intervention scenarios.

Regulatory policies will likely have a significant influence on CRS mitigation and management standards. Established frameworks like Project Optimus and the Model Informed Drug Development pilot foster productive engagement between drug developers and the FDA to address dose-related development questions and apply qCP for dose optimization. Because CRS is profoundly influenced by molecular properties and the specific disease, a comprehensive discussion around pharmacologic context is crucial. This can guide appropriate dose-finding strategies while maintaining a focus on the swift delivery of effective treatments to patients. Furthermore, real-world clinical infrastructure and CRS management guidelines are anticipated to evolve over the coming years. This underscores the need for collaboration between industry, regulatory bodies, and the healthcare community to ensure safe administration of T-BiSp for the benefit of the patients.

K.K. Radtke reports employment with Genentech/Roche and ownership of Genentech/Roche shares. B.C. Bender reports current employment with Roche and ownership of Roche stock. Mosunetuzumab, a Roche molecule, is highlighted in the review. D.C. Turner reports other support from Roche/Genentech during the conduct of the study. S. Roy reports other support from Roche/Genentech outside the submitted work. A. Belousov reports personal fees from Hoffmann-La Roche Ltd. during the conduct of the study, as well as a patent for P36683-US-1 issued to Hoffmann-La Roche. C.-C. Li reports current employment with and stock ownership in Genentech/Roche. No disclosures were reported by the other author.

The authors would like to thank Monica Susilo for her early contributions and overall knowledge of applying quantitative clinical pharmacology strategies for CRS mitigation of T-cell–engaging bispecifics.