Model‐based population pharmacokinetic analysis of tislelizumab in patients with advanced tumors

Abstract Tislelizumab, a humanized immunoglobulin G4 monoclonal antibody, is a programmed cell death protein 1 (PD‐1) inhibitor designed to minimize Fc gamma receptor binding on macrophages to limit antibody‐dependent phagocytosis, a potential mechanism of resistance to anti–PD‐1 therapy. The pharmacokinetic (PK) profile of tislelizumab was analyzed with population PK modeling using 14,473 observed serum concentration data points from 2596 cancer patients who received intravenous (i.v.) tislelizumab at 0.5–10 mg/kg every 2 weeks or every 3 weeks (q3w), or a 200 mg i.v. flat dose q3w in 12 clinical studies. Tislelizumab exhibited linear PK across the dose range tested. Baseline body weight, albumin, tumor size, tumor type, and presence of antidrug antibodies were identified as significant covariates on central clearance, whereas baseline body weight, sex, and age significantly affected central volume of distribution. Sensitivity analysis showed that these covariates did not have clinically relevant effects on tislelizumab PK. Other covariates evaluated, including race (Asian vs. White), lactate dehydrogenase, estimated glomerular filtration rate, renal function categories, hepatic function measures and categories, Eastern Cooperative Oncology Group performance status, therapy (monotherapy vs. combination therapy), and line of therapy did not show a statistically significant impact on tislelizumab PK. These results support the use of tislelizumab 200 mg i.v. q3w without dose adjustment in a variety of patient subpopulations.


INTRODUCTION
The programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1) axis plays a central role in suppressing antitumor activity. 1 Binding of PD-1 to PD-L1 on tumor cells downregulates cytotoxic T-cell responses. Blockade of this interaction with PD-1/PD-L1 inhibitor therapy releases T cells from the inhibitory effects of PD-1, thereby inducing an antitumor immune response. [2][3][4] In recent years, immunotherapy targeting the PD-1/PD-L1 pathway has become an important strategy for cancer treatments, and immune checkpoint inhibitors have demonstrated substantial clinical benefits for cancer patients treated with mono-or combination immunotherapies. 5 However, multiple mechanisms of primary and secondary resistance to PD-1/PD-L1 pathway blockade exist, including antibody clearance (CL) via antibody-dependent cellular phagocytosis (ADCP) through macrophage Fc gamma receptor (FcγR) binding. 6 Tislelizumab (BGB-A317) is a humanized immunoglobulin (Ig) G4 monoclonal antibody with high affinity and binding specificity for PD-1. 6,7 Tislelizumab was designed to minimize FcγR binding on macrophages to limit ADCP. 6,8 In preclinical studies, binding to FcγR on macrophages has been shown to compromise the antitumor activity of PD-1 antibodies through activation of antibodydependent, macrophage-mediated killing of T effector cells. 6 Tislelizumab demonstrated high target affinity and a slow dissociation rate from PD-1 and showed a longer binding time, different binding orientation, and more complete blockade of PD-1/PD-L1 interaction compared with pembrolizumab and nivolumab in preclinical models. 9 Tislelizumab is being developed as a monotherapy and in combination with other therapies for the treatment of a broad array of both solid tumors and hematologic cancers.
Tislelizumab has shown robust antitumor activity and was generally well tolerated in patients with advanced tumors, 1,10 which led to conditional approvals in China for previously treated classical Hodgkin lymphoma (cHL), 11 previously treated urothelial carcinoma (UC), 12 and previously treated hepatocellular carcinoma (HCC). 13 Full approvals have been granted in China for advanced squamous and nonsquamous non-small cell lung cancer (NSCLC) in combination with chemotherapy in the firstline setting and as a second-or third-line treatment for patients with locally advanced or metastatic NSCLC. [13][14][15] To inform the clinical usage of tislelizumab, it is necessary to better understand its pharmacokinetic (PK) properties across different patient populations and to ensure an appropriate dose is given to these populations. The aim of this analysis was to describe the PK of tislelizumab in patients with cancer and estimate typical values and interpatient variability in PK parameters. The study also aimed to evaluate the effect of patient demographics, pathophysiologic factors, immunogenicity, and organ function on tislelizumab PK to identify the clinical factors that might affect tislelizumab exposure in individual patients. In addition, we aimed to evaluate the impact of body weight (WT) distribution on flat dosing and WT-based dosing regimens for tislelizumab using a population PK (PopPK) simulation approach.

Analysis dataset
This analysis was based on data from the following 12 studies, in which patients were treated with tislelizumab in a dose range of 0.5 mg/kg to 10 mg/kg every 2 weeks (q2w) or every 3 weeks (q3w), or 200 mg q3w

WHAT QUESTION DID THIS STUDY ADDRESS?
The analysis characterized tislelizumab pharmacokinetics (PK) and explored the effects of covariates on tislelizumab PK. The feasibility of a flat dose regimen was also assessed in comparison with simulated body weight-based dosing.

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?
The analysis shows that tislelizumab PK is linear across the dose range tested. Tislelizumab exposures are similar across various covariates, including body weight and tumor type, and clinical factors such as hepatic and renal status have no significant effect on tislelizumab PK. These results support the use of tislelizumab 200 mg intravenously every 3 weeks without dose adjustment in a variety of patient subpopulations.
Patients were defined as evaluable for PopPK analysis if they had at least one adequately documented tislelizumab administration and a corresponding PK sample collection after the dose. Tislelizumab serum concentrations in the samples were quantified using validated bioanalytical assays using an enzyme-linked immunosorbent assay method. The lower limit of quantification for tislelizumab was 400 ng/ml. The antidrug antibody (ADA) binding assay employed the electrochemiluminescence immunoassay method to detect the presence of anti-tislelizumab antibodies in human serum.
A small percentage of postbaseline concentration data (0.358% [52/14,543]) were below the limit of quantification and were therefore omitted from the PopPK dataset. Suspected data errors were identified by data examination prior to modeling and excluded from the subsequent PopPK analysis. Outliers were primarily evaluated by conditional weighted residual (CWRES) error of the final base model, and observations for which |CWRES| were >5 were considered outliers. The PopPK analysis was performed with outliers omitted. Missing covariates were imputed if missing for ≤15% of the patients included in the analysis.

Modeling approach
The PopPK analysis was performed using the nonlinear mixed-effects modeling approach with the firstorder conditional estimation with interaction method in NONMEM. Model parameter estimation and evaluation were implemented with NONMEM  Serum tislelizumab concentrations versus time profiles were evaluated to compare the PK results across studies and doses, and to identify potential outliers for exclusion. These graphical analyses and prior knowledge of tislelizumab PK 16,17 provided initial direction for the structural model selection.

Base model and random-effects model development
Based on the known PK properties of tislelizumab, 16,17 the default structural model is a three-compartment model with first-order elimination, as illustrated in Figure 1. Alternative model structures (e.g., one-compartment, two-compartment, and three-compartment models with time-varying CL) were also explored, as appropriate. The PopPK model was parameterized in terms of CL from the central compartment, volume of the central compartment (V c ), CL of distribution from the central to the peripheral compartments (Q 2 and Q 3 ), and volume of the peripheral compartments (V 2 and V 3 ). The final base model was chosen based on the objective function value (OFV), goodness-of-fit (GOF) plots, and reliability of model parameter estimates.
Assuming a log-normal distribution, the interindividual variability (IIV) on PK parameters was described by an exponential model (Equation 1): θ i = individual parameter value for the i th patient, θ T = natural logarithm of the typical value of the parameter in the population, and η i = interindividual random effect with a mean of 0 and variance of ω 2 .
Residual error was described using a combined additive and proportional error model (Equation 2): C(t) ij = j th observed serum concentration of individual i, Ĉ(t) ij = j th model-predicted value (serum concentration) for individual i, and ε pij /ε aij = normally distributed residual random errors with a mean of 0 and variances of σ1 2 and σ2 2 , respectively.

Covariate model development
Following base model development, covariates likely to impact tislelizumab PK were explored for a possible correlation with key tislelizumab post hoc PK parameters. The covariates of interest tested for inclusion in the PopPK model were baseline WT, age, sex, race, estimated glomerular filtration rate (eGFR), alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (TBIL), albumin, lactate dehydrogenase (LDH), baseline Eastern Cooperative   Table 2 and presented by study in Tables S1 and S2.
The covariate effects of WT on PK parameters were tested as part of the base model development, whereas other covariates were tested after WT effects were incorporated in the base model. Once the base model was developed, covariate screening was conducted by examining the correlations between the other covariates and relevant PK parameters graphically, followed by linear regression (for continuous covariates) and analysis of variance testing (for categorical covariates) using R Version 3.5.3. These analyses were conducted using the individual empirical Bayesian estimates (EBEs) of interindividual random effects of PK parameters (η values) obtained from the final base model. Only those covariates that showed a significant (p < 0.05) correlation with the relevant PK parameters that could be meaningfully explained from both clinical and scientific perspectives were examined further in covariate modeling using NONMEM. Covariates were selected using a stepwise forward addition and backward elimination method based on a significance level of p < 0.01 for the forward steps and p < 0.001 for the backward steps.

Model evaluation
After completion of model development, robustness of the final PopPK model was evaluated with multiple qualification/validation methods, including diagnostic GOF plots, prediction-corrected visual predictive check (pcVPC), 18 numerical predictive check (NPC), 19 nonparametric bootstrap, 20,21 and shrinkage assessments. 22

Covariate sensitivity analysis
The covariate sensitivity analysis was performed for the final PopPK model to examine the influence of statistically significant covariates on the predicted exposure of tislelizumab, including the area under the concentration-time curve at steady state (AUC ss ), the maximum concentration at steady state (C max,ss ), and the minimum concentration at steady state (C min,ss ) after the target dose of 200 mg q3w for 30 weeks. Tornado plots were generated for different scenarios (10 th and 90 th values of continuous covariates or possible group of categorical covariates) to show the influence of each covariate on expected exposure compared with the reference value, which was set as the predicted exposure in a typical male patient with tumors except cHL and GC aged 60 years, WT of 65 kg, albumin of 41 g/L, baseline tumor size of 63 mm, and ADA negative after the target dose.

PopPK model simulations
To predict the exposure of tislelizumab in the target patient population, the tislelizumab concentration-time profiles were simulated using the EBEs of individual PK parameters based on the final PopPK model for all patients following treatment with tislelizumab 200 mg q3w for 30 weeks. The predicted steady-state exposure metrics (AUC ss , C max,ss , and C min,ss ) were compared among covariate subgroups to evaluate the need for dose adjustment in patient subgroups The system is described by the following differential equations: To evaluate the effect of WT distribution on WT-based versus flat dosing regimens, simulations were also performed for all patients following a 3 mg/kg q3w dose regimen.

Base model development
Of the model structures tested, a three-compartment model with first-order elimination from the central compartment and redistribution into the peripheral compartments best characterized tislelizumab PK following i.v.
administration. An Ω matrix with IIV on CL, V c , V 2 , and V 3 and covariance between CL and V c was selected based on the model fitness as well as ETA correlation results. A combined additive and proportional error model was selected based on model fitness. An empirical model of time-varying CL was investigated during the base model development. The relationship was modeled as follows (Equations 3 and 4): CL i = clearance for the i th patient, CL T = typical value of clearance in the population, η CL,i = random interindividual effect for clearance, E time = time-varying clearance, T tmax = log maximum magnitude change in clearance, η Tmax = random interindividual effect for T max , T 50 = time when half T max is achieved, and γ = sigmoid factor of timevarying clearance.
The results indicated that a time-varying sigmoid maximum effect (E max )-type model of CL did not improve the model fit.
After the base model structure was established, the model was rerun excluding 18 outlier data points with |CWRES| >5 (nine of these outliers were from Study BGB-A317-001 [2 mg/kg, n = 3; 5 mg/kg, n = 6]). Subsequently, given the previously known effect of WT on the CL and volumes of other antibodies and tislelizumab, the effects of baseline WT on CL, V c , V 2 , and V 3 were examined. The results indicated that adding the WT effect on all four parameters significantly improved the model fit (p < 0.01; OFV decreased by 497). However, removing the WT effect from V 2 and V 3 resulted in a nonsignificant change in OFV (p = 0.876; increased by 0.264). Comparison of the IIV estimates between the base model and the corresponding model without the WT effects indicated that inclusion of WT as a covariate reduced the IIV (coefficient of variation [CV]) by 12.9% and 20.8% for CL and V c , respectively.

Covariate model development
Based on an examination of PK parameter-covariate relationships, the following had a statistical significance of p < 0.01 and were therefore included in the forward covariate search in NONMEM: baseline age, sex, race, eGFR, albumin, AST, LDH, ECOG PS, ADA status, tumor size, and TUMTP on CL, and age, sex, and race on V c.
Testing of the covariates one at a time using a stepwise forward addition method in NONMEM showed that the effect of albumin, tumor size, ADAs, TUMTP, and ECOG PS on CL, and sex and age on V c were significant (p < 0.01). The full PopPK model included all significant covariate relationships. ECOG PS was removed in the backward elimination process.

Final PopPK model
The original dataset included 14,786 measurable serum samples from 2601 subjects. Of these data points, 2% (313/14,786) were excluded from the PK analysis. As a result, the final PopPK analysis dataset comprised 14,473 observed serum concentrations from 2596 patients enrolled in 12 clinical studies of tislelizumab in cancer patients. A three-compartment model with first-order elimination from the central compartment and redistribution into the peripheral compartments best characterized tislelizumab PK following i.v. administration (Figure 1).
In the final PopPK model, baseline WT, age, sex, albumin, tumor size, TUMTP, and ADAs were identified as covariates with a statistically significant effect on the PK of tislelizumab, and the estimated covariate coefficients were well estimated (relative standard error ranging from 5. For a typical male patient (as described previously), the estimated CL was 0.15 L/day, V c was 3.05 L, Q 2 was 0.74 L/day, V 2 was 1.27 L, Q 3 was 0.09 L/day, and V 3 was 2.10 L. IIVs of CL, V c , V 2 , and V 3 were 26.3%, 16.7%, 74.7%, and 99.9%, respectively.
A summary of covariate effects (evaluated for the 10 th and 90 th percentiles of covariate distributions) on tislelizumab PK parameters (CL and V c ) are presented in Table S3. The steady-state volume of distribution was 6.42 L, and the geometric mean elimination half-life was 23.8 days with a CV of 31%. The time to reach 90% steadystate level was approximately 84 days (12 weeks no apparent bias was observed in the residual plots over time and across predicted concentrations (Figures S1 and S2). Specifically, there was no trend between individual weighted residuals versus time up to approximately 3 years posttreatment ( Figure S2), which indicated that the final model could describe the longitudinal observations well without the time-varying CL component. The distribution of IIV was centered at zero and was normally distributed for all parameters ( Figure S3). Pairwise correlations between the ηs showed a slight correlation between CL and V c , which is consistent with the Ω matrix structure of the model ( Figure S4). The pcVPC plots (Figure 2, Figures S5, S6, and S7) showed that the final PopPK model could adequately reproduce the central tendency and variability of the tislelizumab serum concentrations across different regimens, studies, and tumor types. The ηshrinkage for CL and V c was low (15.9% and 15.7%, respectively), suggesting that their EBEs could be used to describe the relationships between CL or V c and the relevant covariates ( Table 3). The NPC showed that the percentage of observed data points outside the 90% prediction interval was 6.51%, suggesting that the final model adequately reproduced the distribution of the observed tislelizumab concentration data.

Covariate sensitivity analysis
The covariate sensitivity analysis showed that WT was the most influential covariate on tislelizumab exposure ( Figure 3). Compared with a typical patient with a WT of 65 kg, patients with WT at the 10 th and 90 th percentiles of the overall population were expected to have 13.8%-18.6% higher and 11.4%-15.0% lower steady-state exposures (AUC ss , C max,ss , and C min,ss ), respectively. TUMTP had a modest effect on tislelizumab exposure: the steady-state exposures in cHL patients were up to 31.4% higher, and exposures in GC patients were 8.82% lower. The effects of albumin, tumor size, ADA status, age, and sex on tislelizumab exposure were relatively small (3.27%-10.8% for albumin, 2.81%-10.1% for tumor size, 5.39%-13.9% for ADAs, 0%-1.48% for age, and 0%-6.49% for sex). Overall, the differences in exposure due to these significant covariates were within the overall variability of exposure in the cancer patient population, which was −38.1% to +55.8%, −28.9% to 47.3%, and −48.9% to 70.8% for the 5 th to 95 th percentiles relative to the typical values of AUC ss , C max,ss , and C min,ss , respectively. Other covariates evaluated (in the covariate model development and/or post hoc covariate analysis), including race, LDH, eGFR, renal function categories, hepatic function measures (AST, ALT, and TBIL) and categories, ECOG PS, therapy (monotherapy vs. combination therapy), and line of therapy did not show a statistically significant impact on the PK of tislelizumab.  comparison as it corresponds to a 200 mg dose for a 65 kg patient, which is the median WT in the analysis dataset (Table S2). The extent and variability of exposure from these dose regimens were simulated from post hoc estimates from 2596 cancer patients included in the PopPK analysis. The flat dose of 200 mg tislelizumab showed a high degree of overlap in exposures with the hypothetical 3 mg/kg dose after a single dose and at steady state ( Figure S8).

PopPK model simulations
The model-predicted average concentrations after the first tislelizumab dose (C avg1 ) with 3 mg/kg q3w and 200 mg q3w across the WT range are presented in Figure 4a, and their distributions across tumor types are presented in Figure 4b. Despite the higher predicted exposures in patients with lower WT receiving the flat dosing regimen, tislelizumab exposures across the WT range were well below the median exposure observed with tislelizumab 10 mg/kg q2w, the clinically established safe and tolerable dose. 1 Moreover, the distribution of average concentration after the first dose (C avg1 ) and average concentration at steady state (C avg,ss ) following a tislelizumab 200 mg flat dose overlapped with the distribution with a 3 mg/kg dose (Figure 4b).
Moderate differences in the geometric mean simulated exposures were observed across TUMTP (up to 34.8% for cHL and 13.6% for GC; Table S4), WT quartiles (within ±17.5%), renal function categories (within ±18.5%), and hepatic function categories (within ±15.4%) compared with the overall population geometric mean, and between White and Asian (up to 21.1%), ADA negative and positive status (up to 20.5%), and male and female (up to 19.0%). Small differences were observed in the geometric mean simulated exposures by age (<4% for adults <65 years of age, elderly patients 65-75 years of age, and elderly patients ≥75 years of age), type of therapy (<9% for monotherapy and combination therapy), and line of therapy (<7%) relative to the overall variability of exposures.

DISCUSSION
Tislelizumab PK was confirmed as linear in the tested dose range (0.5 to 10 mg/kg and 200 mg) and can be   Similar to other Ig G monoclonal antibodies, tislelizumab has a low CL (0.15 L/day), limited volume of distribution (6.42 L at steady state), mainly within the extracellular fluid, and the derived elimination half-life is 23.8 days. 23 Tislelizumab had low IIV on PK parameters, which were in the typical range for monoclonal antibodies. 23 Although other marketed anti-PD-1 monoclonal antibodies, such as pembrolizumab and nivolumab, demonstrate a reduction in CL during the treatment period, as described by time-varying CL, 24,25 this phenomenon was minimal with tislelizumab based on our current pool of data from 12 clinical studies. Although there is no clear mechanistic understanding of the time-varying CL, it was hypothesized that the decrease in CL during the course of treatment may be associated with improvement in disease status. 24,25 A model allowing baseline covariates to vary over time (time-varying covariate model) provided a better fit to the atezolizumab (an anti-PD-L1 antibody) PK data than a model with empirical time-varying E max CL function. These results supported the hypothesis that variation in atezolizumab CL over time is associated with a patient's disease status, as shown with other checkpoint inhibitors. 26 It was also reported that individual patient pembrolizumab CL, by incorporating the impact of on-treatment covariates (albumin, tumor size, etc.), can decrease initially and later increase during the treatment period depending on how the disease progresses, as opposed to either decrease or increase only as limited by the E max model. 27 Nevertheless, it F I G U R E 4 Simulated tislelizumab C avg1 and C avg,ss distribution across (a) body weight and (b) tumor types. (a) Simulated tislelizumab C avg1 and C avg,ss across body weight in patients across tumor types given 3 mg/kg q3w and 200 mg q3w (observed median body weight 65 kg from all patients). (b) Simulated tislelizumab C avg1 and C avg,ss distribution in patients across tumor types given 200 mg q3w and 3 mg/kg q3w. C avg1 , model-predicted average concentration after the first dose; C avg,ss, model-predicted average concentration at steady state; P, percentile; Q2W, every 2 weeks; Q3W, every 3 weeks. needs to be acknowledged that time-varying CL is subject to survivorship bias, as PK samples were collected from a decreasing number of cancer patients over time and may be biased toward patients who remained in the study longer due to baseline disease characteristics, potentially better disease control, or nature of disease type with longer progression time. Therefore, variation in CL depends on a complex interplay of factors (treatment effect, course of disease, and patient characteristics). However, even for current marketed anti-PD-1/PD-L1 antibodies with observed time-varying CL, the magnitude of reduction in CL is relatively small (<25%). [25][26][27] Therefore, it is deemed not clinically relevant and does not warrant any clinical action in terms of dose modification. Given the prior knowledge of the effects of WT on monoclonal antibody CL and volumes of distribution, 28 the covariate effects of WT on PK parameters were tested first and incorporated into the base PopPK model. The exponents of WT effect on CL and volume of distribution were 0.565 and 0.397, respectively. In line with most monoclonal antibodies that perform similarly between WT-based dosing and flat dosing when the exponent is <0.5 or ~0.5, 29 both regimens provided comparable PK exposures, supporting flat dosing of tislelizumab.
Baseline albumin and ADA status were identified as statistically significant covariates on CL with other disease-related covariates, including baseline tumor size and TUMTP. TUMTP had a modest effect on tislelizumab exposure: the steady-state exposures in cHL and GC patients were up to 31.4% higher and 8.82% lower, respectively, compared with those in other cancer patients. These differences are small relative to the overall variability and hence are considered clinically not relevant. The effects of albumin, tumor size, and ADA status on tislelizumab exposure were also small relative to the variability (−38.1% to 55.8%) in the whole population. Baseline tumor size was associated with increasing CL, for example, patients with smaller tumor size had 7.69% higher exposure. ADA-positive patients had 13.9% lower exposure than ADA-negative patients, likely due to the higher CL caused by the formation of ADAs. However, this difference was relatively small compared with the overall variability and was not clinically significant. Sex and age were identified as significant covariates for V c : females had a slightly decreased V c compared with males, and the simulated exposures in elderly subjects were up to 4% lower relative to those in adults <65 years of age. In addition, none of the hepatic or renal function-related covariates (TBIL, ALT, AST, eGFR) had significant effects on tislelizumab PK, and hence no dose adjustment is needed for patients with renal or hepatic impairment.
Overall, based on the flat exposure-response relationships observed for efficacy and safety end points (data not shown), tislelizumab exposure variations of approximately ±20% in magnitude across WT quartiles, age groups, sex, race, ADAs, renal function categories, hepatic function categories, TUMTP, therapy, and line of therapy were not deemed clinically meaningful at the dose of 200 mg q3w. This dose of tislelizumab, a regimen tested in all 12 clinical trials that formed the basis of the PopPK analysis, was selected based on the findings from study BGB-A317-001 that the PK, efficacy, and safety profiles were comparable between 2 mg/kg and 5 mg/kg dose levels, including the fixed dose of 200 mg q3w. 1 Of note, this dose is the same as that recommended for pembrolizumab in the treatment of cancers, including cHL and GC. 30 A flat exposureresponse (efficacy, safety) relationship has been observed with tislelizumab. Therefore, dose adjustment of tislelizumab is not required for any special populations. Simulations of the extent and variability of exposure produced by tislelizumab showed that the PK profile of tislelizumab following flat doses of 200 mg q3w was comparable with the WT-based regimen (3 mg/kg q3w), thereby confirming that WT-based dosing does not confer additional advantages over the 200 mg q3w flat dose regimen.

CONCLUSION
PopPK modeling of tislelizumab in patients with solid tumors and hematological cancer demonstrates that tislelizumab PK is linear. The absence of covariate effects with clinical relevance supports a consistent dosing regimen for tislelizumab. Simulations with the flat dose regimen and WT-based dosing regimen showed comparable PK exposures with similar variabilities, indicating that a flat dose regimen of 200 mg q3w is appropriate in a variety of patient subpopulations.

ACKNOWLEDGMENTS
Medical writing support, under the direction of the authors, was provided by Kirsty Millar, MSc, and Simon Lancaster, BSc, of Ashfield MedComms, an Inizio company, and was funded by BeiGene, Ltd. The