Model‐Informed Development of a Subcutaneous Formulation of Tislelizumab: Phase 1 Pharmacokinetics in Patients With Cancer

ABSTRACT

Although tislelizumab is approved across multiple malignancies and dosing schedules, intravenous administration imposes logistical burdens. A subcutaneous formulation may improve convenience and reduce resource demands if comparable exposure is achieved. This study evaluated the pharmacokinetics, bioavailability, and injection‐site performance of subcutaneous tislelizumab and used population pharmacokinetic (PopPK) modeling and simulations to support phase 3 dose selection. In the phase 1 BGB‐A317‐103 study, treatment‐naïve patients with locally advanced or metastatic non‐small cell lung cancer received tislelizumab via subcutaneous thigh, subcutaneous abdominal, or intravenous administration. A PopPK model was developed with subcutaneous compartments to characterize site‐specific absorption and covariate effects. Simulations compared subcutaneous 300 mg every 3 weeks (Q3W) with intravenous 200 mg Q3W regimens to estimate the probability of achieving comparable pharmacokinetic exposure. The PopPK model demonstrated greater bioavailability and lower variability with subcutaneous thigh vs. abdominal administration. Estimated absorption rates were 0.189 1/d for thigh and 0.247 1/d for abdomen, with corresponding bioavailability estimates of 86.6% and 73.1%. The model described absorption and exposure across administration routes and supported the simulation of 1,000 trials. Subcutaneous 300 mg thigh administration achieved pharmacokinetic exposures comparable to intravenous 200 mg across key parameters, including trough concentration and area under the curve (AUC) at Cycle 1 and steady state, whereas abdominal administration showed reduced probability of achieving target exposure thresholds for Cycle 1 AUC. Tislelizumab 300 mg Q3W subcutaneous thigh administration is predicted to provide exposure comparable to intravenous 200 mg Q3W, supporting its selection for phase 3 evaluation. Trial registration: ClinicalTrials.gov identifier: NCT06091943.

Study Highlights

• What is the current knowledge on the topic?

  • ○
    Programmed cell death protein‐1 (PD‐1) and programmed death‐ligand 1 (PD‐L1) inhibitors are typically administered intravenously, requiring prolonged infusion times and healthcare resource utilization. Subcutaneous formulations are available or in development for some agents and may offer a more convenient route of administration. The PD‐1 inhibitor tislelizumab is currently delivered intravenously, and a subcutaneous formulation has been developed as an alternative route of administration; however, pharmacokinetic and clinical validation are required to ensure comparable exposure, efficacy, and safety.
• What question did this study address?

  • ○
    Using data from the phase 1 BGB‐A317‐103 study of subcutaneous tislelizumab in patients with advanced or metastatic non‐small cell lung cancer, this population pharmacokinetic analysis evaluated whether subcutaneous administration could achieve target exposure and whether model‐informed approaches could support dose selection and phase 3 study design.
• What does this study add to our knowledge?

  • ○
    Population pharmacokinetic modeling and clinical trial simulations demonstrated that subcutaneous administration of tislelizumab in the thigh is predicted to achieve favorable and reproducible exposure in larger trials. These findings supported the design of the ongoing phase 3 study and provided evidence that subcutaneous delivery may achieve exposure comparable to intravenous administration.
• How might this change clinical pharmacology or translational science?

  • ○
    If validated in phase 3, subcutaneous tislelizumab could represent a meaningful advancement in immunotherapy delivery by improving convenience, workflow efficiency, and patient experience without compromising efficacy. This work highlights the value of model‐informed drug development in enabling alternative administration strategies and accelerating translational optimization of established therapies.

1. Introduction

Despite advances in cancer therapy, malignancies such as non‐small cell lung cancer (NSCLC) remain major contributors to the nearly 10 million deaths due to cancer reported worldwide each year, highlighting the need for treatments that more effectively overcome tumor immune evasion [1]. Tumors frequently suppress antitumor immunity by exploiting inhibitory checkpoint pathways such as programmed cell death protein‐1 (PD‐1)/programmed death‐ligand 1 (PD‐L1), impairing T‐cell activation and allowing malignant cells to proliferate unchecked [2]. Although immune checkpoint inhibitors can restore antitumor immune function by blocking these pathways, variability in response and persistent mechanisms of resistance underscore the need for strategies that achieve more consistent and durable immune activation across a wide range of solid tumors [2].

Tislelizumab—a humanized immunoglobulin G4 monoclonal antibody directed against PD‐1—is one such immune checkpoint inhibitor that has demonstrated durable clinical benefit, leading to regulatory approval across multiple regions and countries for various malignancies, including NSCLC, small cell lung cancer, nasopharyngeal carcinoma, gastric cancer, and esophageal squamous cell carcinoma [3, 4, 5, 6, 7]. Tislelizumab is uniquely engineered to reduce Fcγ receptor binding on macrophages, which minimizes the likelihood of antibody‐dependent phagocytosis of T cells and, as shown in preclinical models, enhances the durability of antitumor immune responses [8].

Tislelizumab is approved for intravenous administration using multiple fixed‐dose regimens, including the originally approved 200 mg every 3 weeks (Q3W) schedule [9]. This regimen was selected based on phase 1/2 studies demonstrating a flat exposure–response relationship for both efficacy and safety across a broad dosing range (2–5 mg/kg or fixed 200 mg Q3W) [10, 11, 12]. Comparable response rates and tolerability with weight‐based and fixed dosing further supported its approval across multiple global indications. In addition, a pooled population pharmacokinetic (popPK) analysis in > 2,500 patients showed no clinically meaningful impact of age, body weight, sex, hepatic or kidney function, or tumor burden on drug exposure, reinforcing the appropriateness of fixed dosing across diverse patient populations [13]. Additional fixed‐dose regimens (e.g., 150 mg every 2 weeks, 300 mg every 4 weeks, and 400 mg every 6 weeks) have been evaluated using popPK and exposure–response analyses and subsequently approved based on demonstrated comparability of systemic exposure to the originally approved 200 mg Q3W regimen [14].

Despite the established efficacy and safety of intravenous infusion, there is growing clinical and operational interest in subcutaneous administration of monoclonal antibodies [15, 16, 17, 18]. Subcutaneous formulations can substantially reduce administration time, reduce reliance on infusion centers, and enable treatment in outpatient or even home‐based settings [19, 20, 21, 22]. These changes offer meaningful benefits for patients and healthcare systems by improving workflow efficiency, increasing patient autonomy, and potentially enhancing adherence [19]. Emerging evidence for other immunotherapies—including subcutaneous formulations of the PD‐(L)1 monoclonal antibodies nivolumab and atezolizumab—demonstrates comparable efficacy and tolerability to their intravenous counterparts, along with high bioavailability and low rates of injection‐site reactions [16, 17]. Furthermore, patient‐reported preference data increasingly show strong favorability toward subcutaneous administration due to its convenience, reduced treatment burden, and minimized disruption to daily activities [21, 23, 24].

Compared with intravenous delivery, subcutaneous administration of monoclonal antibodies can have distinct pharmacokinetic profiles, immunogenicity, and safety characteristics, influenced by depot absorption, lymphatic transport, and injection site selection. Therefore, robust early‐phase clinical data and model‐informed strategies are essential to characterize subcutaneous bioavailability, define optimal dosing regimens, and ensure therapeutic equivalence to the established intravenous regimen.

To support the development of a more convenient subcutaneous formulation, the foundational BGB‐A317‐103 open‐label, phase 1 trial evaluated the pharmacokinetics, safety, and tolerability of subcutaneous tislelizumab in treatment‐naïve patients with locally advanced or metastatic NSCLC [25]. The key objectives were to characterize systemic exposure and bioavailability, identify the optimal subcutaneous injection site, and assess exposure comparability with the approved 200 mg intravenous regimen. The study evaluated subcutaneous administration at two injection sites (abdomen and thigh) in comparison with intravenous dosing and included an expansion phase to further assess a recommended subcutaneous regimen, all in combination with platinum‐based induction chemotherapy and maintenance therapy per tumor histology. PopPK modeling was performed to characterize subcutaneous absorption and systemic exposure and to evaluate injection site effects and exposure comparability with the approved intravenous 200 mg regimen. Model‐informed clinical trial simulations supported dose selection for the subcutaneous formulation and informed the design of the ongoing BGB‐A317‐316 phase 3 trial, which is evaluating subcutaneous tislelizumab administered via a pre‐filled syringe in combination with chemotherapy in first‐line treatment of gastric or gastroesophageal junction adenocarcinoma [26].

This manuscript uses the primary pharmacokinetic findings from the BGB‐A317‐103 study [25]. These results provide essential support for the development of subcutaneous tislelizumab as a more convenient, patient‐friendly formulation that maintains pharmacological integrity. If validated in larger clinical trials, subcutaneous administration has the potential to broaden access to immunotherapy and reduce treatment burden for patients across multiple indications.

2. Methods

2.1. Study Design

The BGB‐A317‐103 study (NCT06091943) was a phase 1, open‐label, multicenter trial conducted at 15 clinical sites across China and Europe. The study has been published, and the methods are briefly summarized [25].

Eligible patients were adults (≥ 18 years) with histologically or cytologically confirmed locally advanced or metastatic NSCLC who had not received prior systemic therapy for advanced or metastatic disease. Additional inclusion criteria were the presence of at least one measurable lesion per Response Evaluation Criteria in Solid Tumors v1.1, an Eastern Cooperative Oncology Group performance status of 0 or 1, adequate organ function, and a life expectancy of > 12 weeks.

The study consisted of two sequential parts: Part 1 (dose and injection‐site exploration) and Part 2 (dose expansion). In Part 1, patients were randomized to one of six predefined administration sequences over the first three treatment cycles. These sequences included subcutaneous tislelizumab 300 mg in the thigh, subcutaneous tislelizumab 300 mg in the abdomen, and intravenous administration of tislelizumab 200 mg, with SC sequences administered in two of the first three cycles (Table S1). Following completion of Cycle 3, all patients received intravenous administration of tislelizumab 200 mg Q3W beginning with Cycle 4. Cycle 5 was considered steady state. Intensive pharmacokinetic sampling was performed during Cycles 1–3 for both subcutaneous and intravenous administration, including pre‐dose and multiple post‐dose timepoints (including end of infusion for intravenous administration) to characterize the pharmacokinetic profile. In later cycles (Cycles 4, 5, and 9), sampling was limited to pre‐dose timepoints. Primary endpoints for Part 1 included pharmacokinetic characteristics after subcutaneous administration of tislelizumab: Area under the concentration‐time curve (AUC), trough serum concentration (C_trough), and bioavailability. Secondary endpoints included the maximum serum concentration, immunogenicity, safety, and tolerability.

In Part 2, patients received the recommended subcutaneous regimen of tislelizumab 300 mg Q3W in the thigh, the preferred injection site identified in Part 1 based on higher estimated bioavailability compared with abdominal administration, for the duration of treatment. Intensive pharmacokinetic sampling was conducted during Cycles 1 and 5, while sparse pre‐dose sampling was performed in subsequent cycles (Cycles 2, 6, 9, and 17) to assess trough concentrations. Primary endpoints included pharmacokinetic parameters after subcutaneous administration of tislelizumab and safety. Secondary endpoints included immunogenicity, overall response rate, duration of response, and progression‐free survival, determined by the investigator from tumor assessments per Response Evaluation Criteria in Solid Tumors v1.1.

The study was conducted in accordance with the principles of the Declaration of Helsinki, International Council for Harmonization Good Clinical Practice guidelines, and all applicable regulatory requirements. The protocol was approved by an independent ethics committee or institutional review board at each participating site, and all patients provided written informed consent before enrollment.

2.2. PopPK Model Development and Evaluation

Nonlinear mixed‐effects modeling of pharmacokinetic data from BGB‐A317‐103 was performed to characterize the absorption phase of subcutaneous tislelizumab. This model was developed based on the established PopPK model of tislelizumab after intravenous administration, which was a three‐compartment model with first‐order elimination [13]. Accordingly, all systemic PK parameters (CL/F, Vc/F, V2/F, V3/F, Q2/F, Q3/F) and associated covariate effects were fixed to the final intravenous model estimates (Table 1). Because tislelizumab disposition is expected to be independent of route of administration, subcutaneous dosing was assumed to primarily introduce absorption‐related processes, including absolute bioavailability (F) and absorption rate constant (k_a) parameters. Significant covariates in the intravenous PopPK model included baseline body weight, albumin concentration, tumor size, tumor type, and post‐treatment anti‐drug antibody (ADA) status, while baseline body weight, sex, and age were identified as significant predictors of central volume of distribution.

TABLE 1.

Key pharmacokinetic parameter estimates of the PopPK model.

Parameter	Estimate	Parameter fixed to PopPK model estimate (IV)	Relative SE	Shrinkage
CL/F, L/d	0.153	Fixed
Vc/F, L	3.05	Fixed
V2/F, L	1.27	Fixed
V3/F, L	2.1	Fixed
Q2/F, L/d	0.74	Fixed
Q3/F, L/d	0.092	Fixed
Effect of baseline weight on CL/F	0.565	Fixed
Effect of baseline albumin on CL/F	−0.457	Fixed
Effect of baseline tumor burden on CL/F	0.0735	Fixed
Effect of treatment‐emergent ADA status on CL/F	0.111	Fixed
Effect of gastric cancer status on CL/F	0.069	Fixed
Effect of classic Hodgkin lymphoma status on CL/F	−0.216	Fixed
Effect of baseline body weight on Vc/F	0.397	Fixed
Effect of sex on Vc/F	−0.116	Fixed
Effect of age on Vc/F	0.0966	Fixed
Ka thigh, 1/d	0.189		6.1%
Bioavailability, thigh	0.866		3.2%
Ka abdomen, 1/d	0.247		10.9%
Bioavailability, abdomen	0.731		5.9%
ω (on SD scale)—interindividual variability, %
CL/F	26.3	Fixed
Vc/F	16.7	Fixed
V2/F	74.7	Fixed
V3/F	99.9	Fixed
Ka, thigh	38.2		13.5	17.1
Bioavailability, thigh	13.7		25.9	38.9
Ka, abdomen	56.8		20.7	34.5
Bioavailability, abdomen	26.4		23.6	36.3
σ (residual error)
Proportional error	12.6%	Fixed
Additive error	2.09	Fixed

Abbreviations: 1/d, per day; ADA, antidrug antibody; CL/F, apparent clearance; Ka, absorption rate constant; PopPK, population pharmacokinetics; Q₂/F, distributional clearance between the central and first peripheral compartment; Q₃/F, distributional clearance between the central and second peripheral compartment; SE, standard error; V₂/F, volume of distribution of the first peripheral compartment; V₃3/F, volume of distribution of the second peripheral compartment; V_c/F, apparent volume of distribution of the central compartment.

To describe subcutaneous absorption, alternative absorption models were evaluated, including single‐ and two‐depot structures. A two‐depot absorption model was ultimately selected based on a statistically significant improvement in model fit compared with the single‐depot model (p < 0.001), as well as improved parameter stability and physiological plausibility. The final model incorporated two depot absorption compartments corresponding to thigh and abdominal administration, with parameters for k_a and F (Figure S1). Covariates of interest were evaluated on absorption parameters using a stepwise procedure consisting of univariate screening (p < 0.05) followed by backward elimination (p < 0.01).

Model performance was assessed using standard goodness‐of‐fit diagnostic plots and prediction‐corrected visual predictive checks (pcVPCs). Individual exposure metrics, including C_trough and AUC at Cycle 1 and steady state, were derived from the empirical Bayes estimates of the PopPK parameters. Parameter estimation and model evaluation were performed using NONMEM version 7.4 (ICON Development Solutions; RRID:SCR_016986), Perl‐Speaks‐NONMEM version 5.2.6 (Uppsala University) [27], and R version 4.3.3 (RRID:SCR_001905).

2.3. Model‐Based Simulations

Model‐based simulations were performed to support the dose recommendation for the phase 3 study in gastric cancer. One thousand trials (cohorts of subcutaneous abdominal, subcutaneous thigh, and intravenous administration; n = 100 virtual patients per trial) were simulated for subcutaneous abdominal or thigh administration of 300 mg Q3W to estimate the probability of achieving pharmacokinetic exposures comparable to intravenous administration of tislelizumab 200 mg Q3W. The simulated intravenous infusion was assumed to be given over 60 min.

The demographic and disease characteristics of the virtual population were assumed to resemble those of patients with gastric cancer in the first‐line setting in the BGB‐A317‐305 study (Table S2). Sex and treatment‐emergent ADA status were sampled randomly from four categories (male ADA−, male ADA+, female ADA−, and female ADA+). Albumin, body weight, age, and baseline tumor burden were each assumed to follow normal distributions. Potential correlations among continuous and categorical baseline covariates were evaluated during simulation setup, and no clinically meaningful or systematic associations were identified. Accordingly, covariates were simulated independently. Interindividual variability, residual unexplained variability, and uncertainty in the PopPK model parameter estimates were also incorporated into the clinical trial simulations.

For each simulated trial, the geometric mean ratio (GMR; subcutaneous/intravenous) was calculated for predefined pharmacokinetic metrics, including Cycle 1 C_trough (Cycle 2 predose), C_trough at steady state (Cycle 5 predose), and AUC from day 0 to day 21 (AUC_0‐21) at Cycle 1 and steady state. The proportion of trials in which the lower bound of the 90% CI for GMR was > 0.8 (noninferiority margin) was tabulated to estimate the likelihood of achieving pharmacokinetic exposures comparable to intravenous administration under each subcutaneous dosing scenario. These probabilities informed the selection of the subcutaneous tislelizumab dose most likely to achieve exposures comparable to intravenous tislelizumab in the planned phase 3 study.

3. Results

3.1. PopPK Analysis

Of the 62 patients enrolled in the BGB‐A317‐103 study (39 from Part 1 and 23 from Part 2), 60 had evaluable pharmacokinetic data at the data cutoff date (July 31, 2025). These 60 patients included 39 from Part 1 and 21 from Part 2, all of whom received tislelizumab and had at least one measurable post‐dose pharmacokinetic sample. Patient demographics and baseline characteristics have been published elsewhere [25]. All evaluable patients were included in the PopPK analysis dataset.

The extended PopPK analysis was conducted using 922 data points from the 60 patients with evaluable pharmacokinetic data who received subcutaneous and intravenous tislelizumab (Table 1). Individual plasma concentration–time profiles for subcutaneous and intravenous tislelizumab administration were reported previously [25]. For subcutaneous administration, k_a was estimated separately by injection site. K_a was 0.189 1/d for thigh injections and 0.247 1/d for abdominal injections, with corresponding bioavailability estimates of 86.6% and 73.1%, respectively, indicating greater bioavailability following thigh administration. Interindividual variability in bioavailability was also lower for thigh injections (13.7%) compared with abdominal injections (26.4%), further supporting greater consistency of exposure with thigh administration. Clinically relevant covariates, including body weight, age, sex, race, albumin, and ADA status, were systematically evaluated; however, none resulted in a statistically significant improvement in model fit or improved parameter precision and stability. Therefore, no covariates were retained in the final model.

3.2. Model Evaluation

The predicted vs. observed concentration plots demonstrated good agreement between model‐predicted and observed tislelizumab concentrations, indicating that the final PopPK model adequately captured the central tendency of the data (Figure 1). In addition, the residual diagnostic plots showed no apparent bias in the conditional weighted residuals over time or across predicted concentrations, supporting the appropriateness of the residual error structure and confirming the overall adequacy of the final model (Figure 2).

FIGURE 1.

Predicted vs. observed concentration diagnostic plots for the final PopPK model. Observed vs. individual predicted concentrations (IPRED, right) and observed vs. population predicted concentrations (PRED, left) for the final pharmacokinetic model. Red solid lines represent the unit diagonal, and blue dashed lines represent the loess smooth curves. PopPK, population pharmacokinetics.

FIGURE 2.

Residual diagnostic plots for the final PopPK model. Conditional weighted residuals (CWRES) vs. time after previous dose (left) and PRED (right). Points are individual data. Red solid lines represent the unit line at zero. Green dashed lines represent |CWRES| of 5. Blue dashed lines are smooth curves (loess) to show the relationship between two variables. PopPK, population pharmacokinetics; PRED, population predicted; TFDS, time after previous dose.

The predictive performance of the final PopPK model was assessed using pcVPCs generated from 1000 simulated datasets. Across all administration routes—subcutaneous thigh, subcutaneous abdominal, and intravenous—the model adequately reproduced both the central tendency and variability of the observed plasma tislelizumab concentrations, demonstrating that the final PopPK model reliably captured the distribution of tislelizumab concentrations over time (Figure 3).

FIGURE 3.

PCVPC of tislelizumab plasma concentration–time profiles with different administration routes: (a) subcutaneous abdomen, (b) subcutaneous thigh, and (c) intravenous. Points are observed concentrations, the solid red line represents the median observed value, and dashed red lines represent the 2.5th and 97.5th percentiles of the observed values. Pink shaded areas represent the spread of the median predicted values, and purple shaded areas represent the spread of the 2.5th and 97.5th predicted percentile concentrations. pcVPC, prediction‐corrected visual predictive check; TFDS, time after previous dose.

3.3. Simulation Results With the Selected Subcutaneous Dosing Regimen

Exposure metrics from the clinical trial simulations demonstrated that tislelizumab 300 mg administered subcutaneously in the thigh consistently achieved exposures comparable to the intravenous 200 mg reference dose (Table 2). For thigh administration, the probability of meeting the noninferiority criterion was 100% across all pharmacokinetic parameters for both Cycle 1 and Cycle 5. Corresponding GMs ranged from 1.03 to 1.64, indicating exposure levels similar to or exceeding those with the intravenous regimen.

TABLE 2.

Probabilities of phase 3 outcomes and geometric mean (%CV) with 90% CI for exposure metrics from clinical trial simulations with subcutaneous Q3W dosing.

Parameters		Tislelizumab Q3W
		Subcutaneous		Intravenous
		Abdomen 300 mg	Thigh 300 mg	200 mg
Cycle 1 Ctrough, mg/L	Probability, %	100	100	—
	GMGMR (%CV)	1.33 (5.7)	1.64 (5.0)	—
	GM (%CV)	19.28 (4.4)	23.86 (3.5)	14.52 (3.7)
	90% CI	15.19–23.73	20.20–27.65	11.61–16.89
Cycle 1 AUC0‐21, mg·d/L	Probability, %	88.6	100	—
	GMGMR (%CV)	0.91 (4.5)	1.03 (3.7)	—
	GM (%CV)	532.24 (3.84)	603.23 (2.9)	587.58 (2.3)
	90% CI	444.91–647.67	525.89–691.54	528.64–652.00
Cycle 5 Ctrough, mg/L	Probability, %	100	100	—
	GMGMR (%CV)	1.29 (6.0)	1.59 (5.5)	—
	GM (%CV)	42.72 (4.6)	52.52 (3.8)	33.08 (4.2)
	90% CI	34.05–53.07	43.77–63.50	27.18–39.62
Cycle 5 AUC0‐21, mg·d/L	Probability, %	100	100	—
	GMGMR (%CV)	1.08 (4.8)	1.28 (4.3)	—
	GM (%CV)	1226.81 (4.0)	1451.90 (3.2)	1135.77 (2.9)
	90% CI	1028.29–1504.50	1258.37–1681.78	986.39–1279.61

Abbreviations: AUC0–21, area under the concentration–time curve from day 0 to day 21; Ctrough, trough concentration; CV, coefficient of variation; GM, geometric mean; GM_GMR, geometric mean ratio of geometric means; Q3W, every 3 weeks.

In contrast, subcutaneous administration in the abdomen did not consistently achieve comparable exposures (Table 2). Notably, for the model‐predicted Cycle 1 AUC_0‐21, the probability of meeting the noninferiority criterion was only 88.6%, substantially lower than the 100% probability observed with thigh administration. Greater variability was also observed with abdominal injections, reflected by a lower GM value (0.91) for Cycle 1 AUC_0‐21, although other exposure metrics were generally similar to or greater than those with the intravenous regimen.

Simulated pharmacokinetic profiles showed that the subcutaneous 300 mg Q3W regimen administered in the thigh produced exposures comparable to those achieved with the corresponding intravenous 200 mg Q3W regimen, whereas abdominal administration demonstrated greater variability, particularly during Cycle 1 (Table 2, Figure 4).

FIGURE 4.

Simulated pharmacokinetic profiles of tislelizumab subcutaneous administration in Cycle 1 and steady state, overlaid with intravenous 200 mg Q3W. Solid lines are the medians of geometric means for each simulated dose; shaded areas are the 5th and 95th percentiles of geometric means for each simulated dose. IV, intravenous; PI, predicted interval; Q3W, every 3 weeks; SC, subcutaneous.

4. Discussion

The phase 1 BGB‐A317‐103 study demonstrated that subcutaneous administration of tislelizumab 300 mg achieved exposure comparable to that with intravenous administration of tislelizumab 200 mg, with a safety profile consistent with prior studies of tislelizumab in combination with chemotherapy [25]. Preliminary PopPK modeling supported the selection of the thigh as the preferred injection site, as it yielded higher bioavailability (86.6%) than the abdomen (73.1%) and more consistent C_trough, as demonstrated by lower interindividual variability. These findings are consistent with observations for other subcutaneous monoclonal antibodies, for which thigh injection has demonstrated superior absorption kinetics relative to abdominal injection [28].

The observed differences in absorption and bioavailability between thigh and abdominal administration are biologically plausible and have been reported for other monoclonal antibodies administered subcutaneously [28]. Regional differences in subcutaneous tissue composition, local blood flow, lymphatic drainage, and injection‐site thickness may contribute to variability in absorption kinetics and systemic exposure [29]. In the present analysis, thigh administration was associated with higher bioavailability and lower variability relative to abdominal administration, supporting selection of the thigh as the preferred injection site for further development.

Transitioning from an intravenous to a subcutaneous regimen requires a detailed understanding of absorption kinetics, variability, and exposure relative to the approved dose. PopPK modeling was therefore essential to integrate data across routes of administration, characterize site‐specific absorption, and estimate bioavailability and interindividual variability. This analysis focused on characterizing key pharmacokinetic parameters, including C_trough, maximum serum concentration, and AUC, all of which demonstrated favorable accumulation and consistent exposure with repeated subcutaneous dosing. In Part 2 of the study, Cycle 5 median C_trough reached approximately 65 mg/L, reflecting approximately three‐fold accumulation from Cycle 1 and exceeding the exposure typically observed with the approved 200 mg intravenous regimen. These findings indicate that the subcutaneous 300 mg Q3W regimen not only achieves pharmacokinetic equivalence to intravenous dosing but may also provide higher and more sustained systemic exposure, supporting its suitability for further clinical development.

Safety and tolerability profiles of subcutaneous tislelizumab were consistent with intravenous data, and no new safety signals or injection‐site reactions were observed [25]. Rates of immune‐mediated and treatment‐emergent adverse events were similar to those previously reported [3]. Despite a higher treatment‐emergent ADA rate of up to 52.4% in Part 2 compared with 19.2% reported for intravenous tislelizumab [30], no apparent effect on pharmacokinetic exposure was observed, and no patient tested positive for neutralizing antibodies.

Comparative data for other subcutaneous PD‐(L)1 inhibitors further support the feasibility of subcutaneous administration in oncology. Subcutaneous atezolizumab has demonstrated bioavailability of approximately 82.9% when administered to the thigh and approximately 71.1% when administered to the abdomen [28]. Similarly, subcutaneous nivolumab has demonstrated an estimated bioavailability of approximately 76.7% [31].

To ensure that the PopPK model was suitable for characterizing tislelizumab pharmacokinetics and supported subsequent simulation analyses, a comprehensive model evaluation was performed. The predicted‐vs.‐observed plots demonstrated close agreement between observed concentrations and model predictions, indicating that the structural model accurately captured the central tendency of the pharmacokinetic data across routes of administration. In addition, the conditional weighted residual plots showed no systematic trends over time or across predicted concentrations, supporting the adequacy of the residual error structure and suggesting that unexplained variability was well accounted for. Together, these diagnostics demonstrate that the model reliably represents the observed data and provides a sound foundation for subsequent simulation‐based analyses. This is particularly important for evaluating exposure comparability between intravenous and subcutaneous administration routes, identifying the preferred injection site, and supporting dose selection for future clinical studies. The consistency of these diagnostic outcomes also increases confidence that the model can be applied to new patient populations, such as those simulated for phase 3 gastric cancer treatment development.

PopPK modeling also enabled simulation of clinical dosing scenarios to evaluate whether the subcutaneous 300 mg regimen could achieve exposures comparable to the intravenous 200 mg regimen. PcVPCs generated from 1000 simulated datasets confirmed that the model accurately reproduced the distribution of observed tislelizumab concentrations over time, with strong alignment between the observed data and the simulation‐based 95% CIs for the model‐predicted median and variability. Importantly, this agreement was consistent across all administration routes—subcutaneous thigh, subcutaneous abdomen, and intravenous—indicating that the model reliably captured both the central tendency and the dispersion of pharmacokinetic profiles regardless of route or injection site. Collectively, these findings confirm that the final PopPK model reliably represents tislelizumab pharmacokinetics and is appropriate for use in exposure comparisons and dose‐selection simulations essential for subsequent dose selection in the subcutaneous development program.

This modeling‐based approach, using the established intravenous PopPK model as a foundation [13], illustrates the value of model‐informed drug development in accelerating clinical decision‐making. By integrating data across subcutaneous and intravenous administration routes, the PopPK model enabled quantitative predictions of exposure comparability that could not be derived from observed data alone. These simulations provide justification for dose‐bridging between intravenous and subcutaneous formulations and played a central role in informing dose selection and the overall design of the BGB‐A317‐316 phase 3 trial, which is evaluating the pharmacokinetic comparability of subcutaneous and intravenous tislelizumab administration in patients with advanced gastric or gastroesophageal junction adenocarcinoma.

From a clinical practice perspective, the transition from intravenous to subcutaneous dosing offers meaningful advantages, including shortened drug administration duration, improved infusion center throughput, and potential at‐home administration [19, 20, 21, 22]. For patients, this shift can lessen travel burden, increase autonomy, and may improve treatment adherence and satisfaction [21, 23, 24]. Subcutaneous dosing may also expand access to immunotherapy in regions where intravenous infusion capacity is limited, supporting more equitable cancer care delivery.

These findings should be interpreted in the context of the relatively limited sample size of the phase 1 dataset. Injection‐site effects estimated from smaller datasets may be sensitive to variability and may evolve as additional clinical data become available. Similar observations have been reported during the development of other subcutaneous immune checkpoint inhibitors, including atezolizumab, where early injection‐site differences became less pronounced following the expansion of the clinical pharmacology dataset [32]. Therefore, although the current analysis supports thigh administration as the preferred subcutaneous injection site, continued evaluation in larger clinical studies remains important to further characterize injection‐site effects and exposure variability.

In summary, the BGB‐A317‐103 study indicated that subcutaneous tislelizumab is a viable alternative to intravenous administration, offering high bioavailability, consistent exposure, and a favorable safety profile. The findings also identify the thigh as the optimal administration site, providing greater and more consistent absorption compared with the abdomen. These results strongly support further development of subcutaneous tislelizumab and have informed its selection for the pivotal BGB‐A317‐316 phase 3 trial.

Author Contributions

X. Liu, Y.G., and N.B. wrote the manuscript; R.A., Z.N., S.S., W.D.H., and N.B. designed the research; X. Liu, Z.N., X. Lin, X.Y., Y.W., S.S., and N.B. performed the research; X. Liu, Y.G., Y.L., T.Y., Y.W., and S.S. analyzed the data.

Funding

This study was funded by BeOne Medicines, Ltd.

Conflicts of Interest

X. Liu, Y. Guan, R. Abdrashitov, Z. Niu, X. Lin, W.D. Hanley, and N. Budha report being employees of BeOne Medicines, Ltd. X. Liu, T. Yu, Z. Niu, X. Lin, W.D. Hanley, and N. Budha report owning stocks in BeOne Medicines, Ltd. R. Abdrashitov reports owning stock in BeOne Medicines, Ltd., AstraZeneca, Geron, Takeda, Revolution Medicines, and Syndax. S. Sahasranaman reports being an employee of BeOne Medicines, Ltd. at the time of the study, owning stocks in BeOne Medicines, Ltd., and employment, stock or other ownership, travel, accommodations, and expenses from Prelude Therapeutics. All other authors declared no competing interests for this work.

Supporting information

Table S1: Sequence of administration sites assessed in Part 1.

Table S2: Assumed demographic and baseline disease characteristics of virtual patients in the clinical trial simulations.

Figure S1: PopPK model diagram for tislelizumab SC administration.

Liu X., Guan Y., Liu Y., et al., “Model‐Informed Development of a Subcutaneous Formulation of Tislelizumab: Phase 1 Pharmacokinetics in Patients With Cancer,” Clinical and Translational Science 19, no. 7 (2026): e70649, 10.1111/cts.70649.

Data Availability Statement

On request, and subject to certain criteria, conditions, and exceptions, BeOne Medicines, Ltd. will provide access to individual deidentified participant data from applicable BeOne Medicines‐sponsored studies. BeOne Medicines shares data only when permitted by applicable data privacy and security laws and regulations, shares when it is feasible to do so without compromising the privacy of the study participants, and other considerations. Data requests may be submitted to ClinicalTrials@beonemed.com.