A dose‐finding population pharmacokinetic/pharmacodynamic model of ginisortamab, an anti‐gremlin‐1 monoclonal antibody, in patients with solid tumours Authors: Yin Cheong Wong, Luisa Rojas Chavarro, Paul Jones, David Egging, Wanchana Ungphakorn, Marta Valle, Ivan Matthews Published: 2026 ABSTRACT AimsGinisortamab, a first‐in‐class human monoclonal antibody for the treatment of advanced solid tumours, binds to gremlin‐1 and restores bone morphogenetic protein signalling. We used pharmacokinetic/pharmacodynamic (PK/PD) modelling to characterize the relationship between ginisortamab dose and serum gremlin‐1 binding, using model‐based simulations to inform ginisortamab‐dose selection for ongoing clinical development.MethodsGinisortamab and total gremlin‐1 serum concentration data were obtained from the monotherapy dose‐escalation module of the ONC001 phase 1/2 multipart study in participants with solid tumours (NCT04393298). Population PK/PD models were generated using an iterative model‐building process with nonlinear mixed effects.ResultsA two‐compartment PK model with first‐order drug elimination adequately described dose‐proportional exposure for ginisortamab 100–2000 mg (administered every 2 weeks via intravenous infusion). PK/PD modelling incorporating total gremlin‐1 serum concentrations adequately described ginisortamab binding to serum gremlin‐1. Body surface area significantly affected central (V c, 1.3 [95% CI: 1.0–1.7]) and peripheral (V p, 2.2 [95% CI: 1.6–3.0]) volume of distribution of ginisortamab, but no other tested covariates impacted ginisortamab–gremlin‐1 binding kinetics. Estimated gremlin‐1 turnover time was rapid (~23 [95% CI: ~18–32] min). The developed PK/PD model was used to simulate ginisortamab dosing regimens; 1000 mg once‐weekly and 2000 mg every 2 weeks were predicted to maintain target suppression ≥90% over the dosing interval at steady state.ConclusionThis population PK/PD model adequately described ginisortamab PK and binding to serum gremlin‐1 in patients with advanced solid tumours. Dose simulations informed the level of gremlin‐1 suppression and supported further clinical evaluation of several dosing regimens. What is already known about this subject? High gremlin‐1 expression is associated with poor prognosis in various solid tumours.Ginisortamab, a first‐in‐class human monoclonal antibody, inhibits gremlin‐1 binding to bone morphogenetic proteins and demonstrated favourable preliminary safety, pharmacokinetics (PKs) and anti‐tumour activity in a first‐in‐human phase 1/2 study in participants with advanced solid tumours. High gremlin‐1 expression is associated with poor prognosis in various solid tumours. High gremlin‐1 expression is associated with poor prognosis in various solid tumours. Ginisortamab, a first‐in‐class human monoclonal antibody, inhibits gremlin‐1 binding to bone morphogenetic proteins and demonstrated favourable preliminary safety, pharmacokinetics (PKs) and anti‐tumour activity in a first‐in‐human phase 1/2 study in participants with advanced solid tumours. Ginisortamab, a first‐in‐class human monoclonal antibody, inhibits gremlin‐1 binding to bone morphogenetic proteins and demonstrated favourable preliminary safety, pharmacokinetics (PKs) and anti‐tumour activity in a first‐in‐human phase 1/2 study in participants with advanced solid tumours. What this study adds? A population PK/PD model was developed that adequately described the PK of ginisortamab and its binding to serum gremlin‐1.Simulations predicted the level of serum gremlin‐1 suppression, supporting the findings of the first‐in‐human study and identifying several dosing regimens for further clinical evaluation. A population PK/PD model was developed that adequately described the PK of ginisortamab and its binding to serum gremlin‐1. A population PK/PD model was developed that adequately described the PK of ginisortamab and its binding to serum gremlin‐1. Simulations predicted the level of serum gremlin‐1 suppression, supporting the findings of the first‐in‐human study and identifying several dosing regimens for further clinical evaluation. Simulations predicted the level of serum gremlin‐1 suppression, supporting the findings of the first‐in‐human study and identifying several dosing regimens for further clinical evaluation. INTRODUCTION The treatment of solid tumours remains a significant unmet medical need globally because of their high incidence and mortality rates, and the challenges associated with early detection and effective treatment. In 2022, there were approximately 20 million new cancer cases and 9.7 million cancer‐related deaths worldwide, with solid tumours such as lung, colorectal, breast, liver and stomach cancers accounting for most of these cases and associated deaths. 1 Gremlin‐1 is a high‐affinity bone morphogenetic protein (BMP) antagonist that binds extracellularly to transforming growth factor‐β (TGF‐β) family BMP‐2, BMP‐4 and BMP‐7. 2 The BMP pathway has dual tumour‐promoting and tumour‐suppressing activities depending on the cancer cell type, tumour microenvironment, and stage of tumour growth 3 ; however, high gremlin‐1 levels are associated with poor prognosis and outcomes across a range of solid tumours. 4 When secreted from tumour stroma or the tumour cells themselves, gremlin‐1 antagonism of BMP signalling promotes tumour ‘stemness’, hyperproliferation, and a more invasive mesenchymal phenotype. 5 , 6 Gremlin‐1 is known to be highly expressed in >60% of oesophageal, colorectal and pancreatic cancers, and in approximately 50% of bladder, breast and lung cancers. 4 Evidence suggests high gremlin‐1 expression in late‐stage breast and colorectal cancer can promote metastasis 7 , 8 , 9 , 10 and correlates with poor prognosis. 4 , 6 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 Ginisortamab is a first‐in‐class fully human IgG4 monoclonal antibody (mAb) that has demonstrated the ability to restore gremlin‐1‐suppressed BMP signalling in in vitro cancer cell culture assays through inhibition of gremlin‐1 binding to BMP‐2, −4 and ‐7. 20 In human colorectal cancer cell lines, ginisortamab fully blocked the activity of recombinant human gremlin‐1 and restored BMP signalling pathways. 20 In a human colorectal–fibroblast co‐culture system in which gremlin‐1 is produced by the fibroblasts, ginisortamab also restored BMP signalling. 20 There is also strong preclinical evidence to suggest that ginisortamab can limit tumour cell expansion through restoration of BMP signalling. A chimeric, murine analogue of ginisortamab (Ab7326) has demonstrated anti‐tumour activity in mouse models of multiple myeloma and colorectal cancer. 21 , 22 Additionally, treatment with Ab7326 in a murine Apc Min /+; Vil1‐GREM1 spontaneous genetic model of intestinal polyposis and cancer led to reduced polyp formation and prolonged mouse survival, supporting the potential for therapeutic benefit in gremlin‐1‐driven cancers through gremlin‐1 antagonism (data on file). The safety, pharmacokinetics (PKs), and anti‐tumour activity of ginisortamab in participants with advanced tumours have been investigated in ONC001 (NCT04393298), a first‐in‐human, phase 1/2 open‐label, multicentre study that evaluated ginisortamab administered via intravenous (IV) infusion in a monotherapy dose‐escalation module (Part A), in combination with selected standard of care in two further dose‐escalation modules (Parts B and C) and in a dose‐optimization module (Part A1). Monotherapy dose‐escalation data indicated a favourable safety and PK profile at the maximum tested dose of 2000 mg every 2 weeks (Q2W), with preliminary anti‐tumour activity and no observed antidrug antibody (ADA) formation. 23 The ONC001 study is complete, and at time of writing, a Phase 2 clinical study (ISRCTN63154609) evaluating ginisortamab in participants with pancreatic ductal adenocarcinoma (PDAC), the most common form of pancreatic cancer, is underway. The population PK and pharmacodynamic (PD) model presented here was designed to describe the time course of the effect of ginisortamab on total gremlin‐1 serum concentrations (free gremlin‐1 and ginisortamab‐bound gremlin‐1) using data from the ONC001 monotherapy dose‐escalation module. The objectives were to identify covariates that may explain variability in ginisortamab PK and total gremlin‐1 serum concentrations; to simulate different ginisortamab dosing regimens to evaluate their impact on free gremlin‐1 suppression levels; and to support and inform ginisortamab dose selection for upcoming clinical studies, including the Phase 2 study in PDAC. METHODS ONC001 clinical study Data for this population PK/PD analysis was obtained from the dose‐escalation module of ONC001, a non‐randomized first‐in‐human monotherapy dose‐escalation module which enrolled participants at several sites in the UK. The ONC001 study was conducted in accordance with the study protocol and consensus ethical principles derived from international guidelines, including the Declaration of Helsinki, the International Conference on Harmonization Good Clinical Practice, 24 and the Council for International Organizations of Medical Sciences International Ethical Guidelines, and was approved by institutional review boards, including the London–West London and Gene Therapy Advisory Committee (GTAC) Research Ethics Committee (REC reference 19/LO/1926). All participants in the ONC001 study provided written informed consent before participating. Eligible participants were ≥18 years old with recurrent or metastatic advanced solid tumours known to be associated with GREM1 mRNA expression (i.e. bladder urothelial carcinoma, breast invasive carcinoma, colorectal adenocarcinoma, oesophageal carcinoma, head and neck squamous cell carcinoma, lung adenocarcinoma, lung squamous cell carcinoma, pancreatic adenocarcinoma, prostate adenocarcinoma and gastric adenocarcinoma) 4 , 25 and resistant or refractory to standard anti‐cancer treatment. All participants had an estimated life expectancy of ≥12 weeks, access to approved therapies, measurable or non‐measurable disease as defined by the relevant Response Evaluation Criteria in Solid Tumours (RECIST), version 1.1, and an Eastern Cooperative Oncology Group performance status (ECOG PS) ≤ 1. Ginisortamab was administered as a 60‐min IV infusion Q2W until the occurrence of progressive disease, unacceptable toxicity, or withdrawal of consent. Five sequential dose cohorts were escalated through ginisortamab 100 (cohort 1), 250 (cohort 2), 500 (cohort 3), 1000 (cohort 4) and 2000 mg (cohort 5). All participants remained in the hospital or a standard monitoring facility of the study site for continuous monitoring over the first 24 h following the first dose of ginisortamab and, thereafter, at the investigator's discretion for all subsequent doses. Sampling and bioanalytical methods Blood samples for analysis of ginisortamab serum concentration were collected following the first (Day 1) and second (Day 15) doses during the first three treatment cycles, with a final sample obtained at the safety follow‐up visit within 30 days of the last dose. Ginisortamab serum concentrations were measured using a validated bridging immunoassay on the MSD QuickPlex platform, employing biotinylated and Ru‐sulfotagged anti‐idiotype antibodies to ginisortamab as capture and detection reagents, respectively. This assay format detects free (unbound) ginisortamab capable of simultaneously binding both reagents. Because the anti‐idiotype antibodies compete with gremlin‐1 for binding, ginisortamab molecules in which one or both antigen‐binding arms are occupied by gremlin‐1 are not quantified. During method development, target interference resulted in concentration‐dependent negative bias (up to −71% at the lower limit of quantification [LLOQ] and −15% at the high‐quality control level at investigated gremlin‐1 concentrations). The LLOQ was 20.0 ng mL−1 with an analytical range up to 5000 ng mL−1. Total gremlin‐1 serum concentrations were measured using a fit‐for‐purpose validated liquid chromatography–tandem mass spectrometry (LC/MS/MS) assay with an immunoprecipitation step. Gremlin‐1 and the internal standard gremlin‐1 peptide (H2N‐QTIHEEG [Cam]NSR–OH) were analysed in 130 μL human serum samples using immunoprecipitation. Data were collected using selected reaction monitoring turbo ion spray LC/MS/MS in the positive ion mode, with a concentration range of 0.45 (LLOQ) to 90 ng mL−1 (Upper LOQ), using single‐point calibration for determination of concentrations calculated from peak area ratios. The assay was evaluated for interference in the presence of ginisortamab at concentrations up to 1500 μg mL−1 and met the acceptance criteria. A homogeneous bridging assay (electrochemiluminescence immunoassay: ECLIA) was developed and validated to assess the presence of ADAs in the serum samples of study participants. A tiered approach to ADA sample analysis was used in this study, comprising screening, confirmatory and titre steps, where appropriate. The assay showed acceptable intra‐ and inter‐run precision, both in screening and in confirmation assays, with values for coefficient of variation (CV)below 20% for negative control and positive control levels. Investigated drug tolerance showed that at the highest ginisortamab concentration investigated (480 μg·mL−1), positive control could be detected at 6.07 and 9.76 ng mL−1 in screening and confirmatory assays, respectively (interpolated from positive control dilution series). The drug tolerance data suggest the assay can detect potential ADA in ONC001 study samples containing drug at relevant sampling timepoints. Drug and molecular target nomenclature conform to the International Union of Basic and Clinical Pharmacology (IUPHAR)/British Pharmacological Society (BPS) Guide to Pharmacology. Modelling and simulation approaches The PK/PD analysis population included all participants from the ONC001 monotherapy dose‐escalation module who received ≥1 dose of ginisortamab and had ≥1 post‐dose ginisortamab serum concentration and ≥1 total gremlin‐1 serum concentration measurement above the LLOQ at a known/recorded time. The population PK and PK/PD models were developed following an iterative model‐building process using a nonlinear mixed‐effects approach. The first‐order conditional estimation with η–ε interaction (FOCE‐INT) in NONMEM was employed for all model parameter estimations (Supporting Information Methods). Both PK and PD samples below the LLOQ were excluded from the analysis according to the ‘M1’ methodology. 26 The base and final population models were selected on the basis of overall model performance, including the examination of parameter precision, shrinkage, minimum objective function value (OFV), condition number, significant digits, successful covariance step and goodness‐of‐fit plots. The performance of the final population model was evaluated with prediction‐corrected visual predictive checks (pcVPC): 1000 replicates of the original dataset were simulated to evaluate if the final model adequately described the central tendency and variability in the observed data. Parameter uncertainty was assessed by evaluating the standard error of the parameters and confidence intervals around the parameters using sampling importance resampling (SIR). The final population PK/PD model was used to simulate and predict ginisortamab serum concentration, total gremlin‐1 serum concentration, free gremlin‐1 serum concentration and the level of target (gremlin‐1) suppression for various ginisortamab‐dosing regimens. Free gremlin‐1 concentrations were derived from the target‐mediated drug disposition (TMDD) model, and target suppression was calculated as percentage change relative to the model‐estimated baseline‐free gremlin‐1 concentration (BLR). Population PK/PD model development A two‐compartment disposition model with first‐order drug elimination from the central compartment was used as a starting point for development of a base PK model, and additional structural compartmental models (i.e. 1‐n compartment) were then tested. Inter‐individual variability (IIV) in the PK parameters was introduced in the model using exponential terms, assuming lognormal distribution of the parameters as θi = θ × exp (ηi), where θi is the PK parameter of the individual, θ is the typical population value of the parameter, and ηi is the inter‐individual random effect assumed to have a normal distribution, mean of zero and a variance of ω 2. When substantial correlation was present between individual random effects, covariance between these parameters was included in the model through a block covariance matrix (Ω). A proportional error model was tested to describe the residual variability for ginisortamab serum concentrations. After identification of the base PK model, participant baseline characteristics (age, race, body weight [BW], body mass index [BMI], body surface area [BSA]), sex and cancer type were considered for evaluation as covariates on model parameters (clearance [CL], central volume of distribution [V c ], inter‐compartmental clearance [Q] and peripheral volume of distribution [V p ]) to explain variability in ginisortamab serum concentrations. Dose was tested as a potential covariate graphically; however, because of the lack of any sign of relationship between dose and any of the PK parameters, no formal test was performed in NONMEM. Covariates that showed a correlation with the random effects and had possible clinical relevance were further evaluated for inclusion and retained if there was a statistically significant change in OFV using a stepwise approach with forward inclusion (P < .01) and backward elimination (P < .001). The final PK model, including base and covariate model, was used as a starting point to develop the structural PK/PD model. Total gremlin‐1 serum concentration–time data were included and evaluated using different approximations of TMDD. BLRs were parameterized as BLR = gremlin‐1 synthesis rate constant (K syn)/gremlin‐1 degradation rate constant (K deg). The quasi‐steady‐state (QSS) model approximation 27 best described the data and was retained in the model. More complex models, including binding of ginisortamab to gremlin‐1 in the peripheral compartment and concentration‐dependent K syn models, were also tested but not retained. 28 All PK and PD data were fitted simultaneously and IIV in the PD parameters was introduced using exponential terms assuming a lognormal distribution, as described for the population PK model development. A proportional error model was tested to describe the residual variability for total gremlin‐1 serum concentration. After the identification of the base population PK/PD model, participant baseline characteristics (age, race, BW, BMI, BSA) and cancer type were considered for evaluation as covariates on model parameters (CL, BLR, and K syn) to explain the variability in ginisortamab and total gremlin‐1 serum concentrations. Covariates that showed a correlation with random effects and that had a possible clinical relevance were further evaluated using the same method described for population PK model development. Simulations Simulations were performed using the final PK/PD model for the ginisortamab‐dosing regimens included in the ONC001 monotherapy dose‐escalation module (six cycles of ginisortamab 100, 250, 500, 1000 or 2000 mg administered as a 60‐min IV infusion Q2W over a 28‐day cycle) as well as those under investigation in the ONC001 dose‐optimization module (ginisortamab 2000 mg Q2W 30‐min IV infusion over a 28‐day cycle, 3000 mg Q3W eight cycles 90‐min IV infusion over a 21‐day cycle, and 4000 mg Q4W 120‐min IV infusion over a 28‐day cycle), plus an additional hypothetical dosing regimen of 1000 mg once weekly (QW) (60‐min IV infusion over a 28‐day cycle). Simulations were conducted to generate 1000 virtual subjects per scenario, incorporated IIV, and aimed to predict the drug exposure and level of target suppression achieved for each regimen. BSA values were sampled from the observed ONC001 dose‐escalation population to preserve demographic characteristics. PK exposure metrics (area under the curve over the dosing interval [AUCtau], and 14 days after the respective dose administration [AUC0–336h], maximum concentration [C max], trough concentration [C min] and average concentration [C av]; calculated as AUCtau/dosing interval) were calculated and summarized by dosing regimen. Steady‐state maximum target suppression, target suppression after the first dose in the first dose cycle, and target suppression at the end of the dosing period were also calculated. Software Exploratory data preparation, summary statistics and visualization were performed using R (version 4.1.1). 29 The population PK analysis, population PK/PD analysis and simulations were performed in an environment interfaced by Pirana (version 2.9.9, Certara) in which NONMEM (version 7.4.4, ICON Development Solutions), PsN (Perl‐speaks‐NONMEM, version 5.0.0), xpose4 (version 4.5.3) and R (version 4.1.1) were incorporated in an integrated manner for data manipulation, estimation, graphical and statistical summaries. RESULTS Population data Overall, 25 participants received ≥1 dose of ginisortamab and had data included in this analysis; baseline demographics and characteristics for these participants are included in Table S1. The most common tumour was colorectal adenocarcinoma (n = 11/25) followed by pancreatic adenocarcinoma (n = 6/25) and prostate adenocarcinoma (n = 3/25). 23 All participants were ADA negative at all visits based on the screening and confirmatory assays performed. A total of 371 ginisortamab serum concentrations and 270 total gremlin‐1 serum concentrations were included in model development (24 ginisortamab pre‐first dose observations below the limit of quantification [BLQ], 1 ginisortamab unexplained pre‐first dose observation above LLOQ, six total gremlin‐1 serum concentrations BLQ and two missing total gremlin‐1 serum concentrations were excluded, Table S2). Ginisortamab serum concentrations increased dose‐proportionally and declined in an approximately bi‐exponential manner. Variability of ginisortamab serum concentrations was low to moderate at all timepoints among participants across all doses (Figure 1A). Total gremlin‐1 serum concentrations after dosing were higher, following ginisortamab doses of ≥500 mg IV Q2W, compared with doses ≤250 mg IV Q2W. A large variability between participants in total gremlin‐1 serum concentrations was observed (Figure 1B). (A) Ginisortamab serum concentration and (B) total gremlin‐1 serum concentration–time profiles by dose. Solid lines are median concentrations per dose; data points represent individual observations per dose. Dropout data are excluded. Population PK/PD model evaluation The base population PK model that best described ginisortamab concentrations was a two‐compartment disposition model with linear first‐order elimination from the central compartment and IIV in CL, V c and V p. A proportional error model was used to describe random residual error. Baseline body size descriptors (BW, BMI, and BSA) were evaluated using a forward inclusion and/or backward elimination approach. BSA‐based empirical scaling provided superior model performance compared with theoretical allometric exponents in terms of statistical and model predictions improvement, with the largest reduction in OFV and IIV, and was the only statistically significant covariate. BSA (centred at the participant median value of 2 m2) was included on V c and V p, with the exponent estimated as 1.3 (95% CI: 1.0–1.7) and 2.2 (95% CI: 1.6–3.0), respectively. Ginisortamab and total gremlin‐1 serum concentrations were best described by a two‐compartment disposition model with TMDD in which the binding of ginisortamab to gremlin‐1 is given by the QSS approximation (Figure 2). The QSS model assumes that drug–target complex formation and elimination occur rapidly, relative to other PK and target‐turnover processes, that is d (drug–target complex)/dt ≈ 0. Under this assumption, the association rate (K on) of ginisortamab to gremlin‐1 is balanced by the sum of the dissociation rate constant (K off) and internalization rate constant (K int), expressed as QSS constant (K ss) parameter, which is defined as K ss = (K off + K int)/K on. Gremlin‐1 IIV was included on CL, K int, K deg, V c and V p. A proportional error model was used to describe random residual error. During the development of the model, different approaches were tested, including fixing the target‐turnover value to the elimination rate constant of ginisortamab and fixing K off to the value derived from an in vitro binding assay. However, better model fitting was obtained when the QSS approximation was implemented, with all parameters estimated from the clinical data. Population PK/PD final model equations and schematic. θ CL, population mean CL; θ Vc, population mean V c; θ Vp, population mean V p; θ Q, population mean Q; θ Ksyn, population mean K syn; θ Kdeg, population mean K deg; θ Kss, population mean Kss; θK int, population mean K int; CL, clearance; CLi, individual CL; 𝑒𝜂𝐶𝐿, inter‐individual random effects for CL; 𝑒𝜂𝑉𝑐, inter‐individual random effects for V c; 𝑒𝜂𝑉𝑝, inter‐individual random effects for V p; 𝑒𝜂𝐾𝑑𝑒𝑔, inter‐individual random effects for K deg; 𝑒𝜂𝐾𝑖𝑛𝑡, inter‐individual random effects for Kint; IV, intravenous; K, rate constant; K syni, individual K syn; K degi, individual K deg; K ssi, individual K ss; K inti, individual K int; K e, elimination rate constant; PK/PD, pharmacokinetic–pharmacodynamic; Q, inter‐compartmental clearance; Q i, individual Q; V C, central volume of distribution; V ci, individual V c; V p, peripheral volume of distribution; V pi, individual V p. All parameters were estimated with satisfactory precision (the estimated relative standard error, based on SIR for all parameters, was <19% for fixed effect and <22% for random effect parameters), and shrinkages were <29% (Table 1). Goodness‐of‐fit plots for observed concentrations versus population‐predicted concentrations demonstrated minimum bias, with most data evenly scattered around the line of unity for both ginisortamab and gremlin‐1 (Figure S1A and B). Observed concentrations versus the individual predicted concentration goodness‐of‐fit plots were tightly scattered around the line of unity with no apparent bias. The typical estimated ginisortamab half‐life (t 1/2) elimination for the typical subject was ~13.3 days, and typical estimated gremlin‐1 turnover time (which equates to 1/Kdeg) for the typical subject was ~23 min. TABLE 1Parameter estimates for the final population PK/PD model.ParametersEstimated value (%RSE from NONMEM)%RSE from SIR (95% CI)Shrinkage (%)Clearance (CL, L/day)0.433 (4.7)4.7 (0.395–0.474) ‐ Central volume of distribution (V c, L)4.38 (3.7)3.3 (4.12–4.69) ‐ Covariate effect of body surface area on V c 1.31 (14.2)13.7 (0.96–1.66) ‐ Inter‐compartmental clearance (Q, L/day)0.849 (12.8)9.6 (0.700–1.019) ‐ Peripheral volume of distribution (V p, L)3.23 (7.8)7.4 (2.80–3.76) ‐ Covariate effect of body surface area on V p 2.18 (21.3)13.9 (1.75–2.96) ‐ Gremlin‐1 synthesis rate constant (K syn, nmol/L/day)0.378 (22.2)10.5 (0.311–0.466) ‐ Gremlin‐1 degradation rate constant (K deg, 1/day)61.8 (28.5)14.1 (45.2–79.4) ‐ Quasi‐steady‐state constant (K ss, nmol/L)7.59 (35.7)18.2 (5.45–10.91) ‐ Complex internalization rate constant (K int, 1/day)3.23 (31.3)13.2 (2.45–4.09) ‐ IIV for CL, CV (%)22.9 (10.9)19.3 (18.2–27.2)−0.3IIV for V c, CV (%)12.9 (21.3)14.4 (11.1–14.7)2.8Correlation coefficient V c and V p a 0.97‐ ‐ IIV for V p, CV (%)22.0 (35.5)15.0 (18.3–25.1)5.1IIV for K deg, CV (%)38.1 (18.2)21.1 (29.7–46.9)28.7IIV for K int, CV (%)55.4 (16.3)18.1 (43.4–65.7)4.7RUV for ginisortamab concentrations (%)10.8 (6.4)4.1 (10.0–11.7)RUV for total gremlin‐1 concentrations (%)35.6 (4.0)5.3 (32.3–39.8) Notes: CV% of proportional error is calculated as 100* THETA, error is estimated as a THETA; CV% of exponential ETAs are calculated as 100 sqrt (exp [OMEGA]‐1). a Correlation coefficient V c and V p is calculated as ωVc,VpωVc*ωVp. Abbreviations: CI, confidence interval; CV, coefficient of variation; IIV, inter‐individual variability; PK/PD, pharmacokinetic–pharmacodynamic; RSE, relative standard error; RUV, residual unexplained variability; SIR, sampling importance resampling. Parameter estimates for the final population PK/PD model. Notes: CV% of proportional error is calculated as 100* THETA, error is estimated as a THETA; CV% of exponential ETAs are calculated as 100 sqrt (exp [OMEGA]‐1). Correlation coefficient V c and V p is calculated as ωVc,VpωVc*ωVp. Abbreviations: CI, confidence interval; CV, coefficient of variation; IIV, inter‐individual variability; PK/PD, pharmacokinetic–pharmacodynamic; RSE, relative standard error; RUV, residual unexplained variability; SIR, sampling importance resampling. The pcVPC of the population PK and final population PK/PD model demonstrated that the median, 10th and 90th percentiles of the prediction‐corrected data were contained within the 95% CIs of corresponding simulations (Figures 3 and S2), validating the models' predictive performances. The PK/PD model adequately described the time profiles of the observed ginisortamab and total gremlin‐1 serum concentrations (Figure 4), which allowed estimation of gremlin‐1 suppression at an individual patient level. Prediction‐corrected visual predictive check of final population PK/PD model for total gremlin‐1 serum concentrations (log scale). CI, confidence interval; PK/PD, pharmacokinetic–pharmacodynamic. Illustrations of the final population PK/PD model performance at an individual level: participant 1 Ginisortamab dose 100 mg; participant 2. Ginisortamab dose 2000 mg. Q2W 60‐min IV administration of ginisortamab. IV, intravenous; PK/PD, pharmacokinetic–pharmacodynamic; Q2W, once every 2 weeks. Simulations to predict target suppression Maximum target suppression and target suppression at the end of the dosing period after the first dose of the first cycle and after the last dose of the last cycle (i.e. steady state) are presented in Table 2. After using the final model to simulate time profiles of gremlin‐1 suppression of different ginisortamab dosing regimens, the PK/PD model predicted that the hypothetical 1000 mg QW and 2000 mg Q2W doses tested in the ONC001 dose‐escalation module, and the 3000 mg Q3W and 4000 mg Q4W doses that were subsequently evaluated in the ONC001 dose‐optimization module would produce a median target suppression close to or above 80% over the dosing interval (Figures 5 and S3). After simulation of repeated ginisortamab dosing, the model predicted that the hypothetical dose of 1000 mg QW and the 2000 mg Q2W regimen tested in the ONC001 dose‐escalation module would produce a median target suppression close to or above 90% over the dosing interval at steady state (Figure 6). TABLE 2PK/PD model predicted median (90% prediction interval) target suppression after administration of ginisortamab at first dose–first cycle and last dose–last cycle.Dose simulationsFirst dose–first cycleLast dose–last cycle (steady‐state)Target suppression at end of dosing period (%)Maximum target suppression (%)Target suppression at end of dosing period (%)Maximum target suppression (%)ONC001 monotherapy dose escalation100 mg Q2W, 60 min18 (7–41)87 (79–92)29 (10–57)77 (65–87)250 mg Q2W, 60 min35 (15–63)94 (90–97)51 (23–77)86 (76–93)500 mg Q2W, 60 min53 (26–78)97 (95–98)68 (37–87)91 (83–96)1000 mg Q2W, 60 min69 (42–88)98 (97–99)81 (55–93)95 (89–98)2000 mg Q2W, 60 min82 (59–93)99 (99–100)89 (71–96)97 (93–99)Hypothetical dose1000 mg QW, 60 min77 (52–92)98 (97–99)91 (77–97)96 (89–99)ONC001 monotherapy dose optimization2000 mg Q2W, 30 min82 (57–93)99 (99–100)89 (70–97)97 (93–99)3000 mg Q3W, 90 min81 (54–94)99 (99–100)86 (59–96)98 (94–99)4000 mg Q4W, 120 min80 (45–93)99 (99–100)83 (48–95)98 (96–99) Note: min, minutes; PK/PD, pharmacokinetic–pharmacodynamic; QW, once weekly; Q2W, once every 2 weeks; Q3W, once every 3 weeks; Q4W, once every 4 weeks. PK/PD model predicted median (90% prediction interval) target suppression after administration of ginisortamab at first dose–first cycle and last dose–last cycle. Note: min, minutes; PK/PD, pharmacokinetic–pharmacodynamic; QW, once weekly; Q2W, once every 2 weeks; Q3W, once every 3 weeks; Q4W, once every 4 weeks. Predicted percentage gremlin‐1 target suppression after IV ginisortamab administration—simulation of ONC001 monotherapy dose‐escalation module dosing regimens: (A) 100 mg Q2W (60‐min IV infusion), 28‐day treatment cycle, 6 cycles; (B) 250 mg Q2W (60‐min IV infusion), 28‐day treatment cycle, 6 cycles; (C) 500 mg Q2W (60‐min IV infusion), 28‐day treatment cycle, 6 cycles; (D) 1000 mg Q2W (60‐mi IV infusion), 28‐day treatment cycle, 6 cycles; and (E) 2000 mg Q2W (60‐mi IV infusion), 28‐day treatment cycle, 6 cycles. IV, intravenous; Q2W, once every 2 weeks. Predicted median % gremlin‐1 target suppression after repeated ginisortamab IV dosing (last dose–last cycle). IV, intravenous; min, minutes; Q2W, once every 2 weeks. The predicted exposure (AUCtau, AUC0–336h, C max, C min and C av) after the first and last dose of all the simulated dosing regimens is summarized in Table S3. Ginisortamab exposure appeared to increase proportionally with doses. Predicted median accumulation ratios based on AUC0–336h and C max were estimated to be 1.7 and 1.4, respectively. DISCUSSION This PK/PD model characterized the relationship between serum‐free ginisortamab concentrations and serum total gremlin‐1 concentrations, and enabled assessment of the level of target suppression after IV administration of different ginisortamab‐dosing regimens to patients with advanced solid tumours. The observed time profiles of ginisortamab and total gremlin‐1 serum concentrations were best described by a two‐compartment disposition with a TMDD population PK/PD model with QSS approximation. The QSS approximation of the TMDD has been suggested to accurately predict the phase when the amount of free target is small or close to zero, which is expected for an mAb, such as ginisortamab, given its high affinity for the target and its in vitro equilibrium dissociation constant for human gremlin‐1 of 87 pM; 20 this phase may represent most of the post‐dose period. This makes QSS a suitable approximation approach to adequately capture and model measured total gremlin‐1 serum concentrations. 28 More complex models, including binding of ginisortamab to gremlin‐1 in the peripheral compartment and concentration‐depended K syn models, were also tested to evaluate whether their inclusion led to a better explanation of the binding of ginisortamab to gremlin‐1; however, no significant differences in OFV or improvements in the goodness‐of‐fit plots were observed, and large relative standard errors were found compared with the QSS model approximation. Overall, the estimated PK and binding parameters for ginisortamab and gremlin‐1 appear similar to other mAb drugs evaluated in various therapeutic areas. 30 , 31 The dose‐dependent increase in total gremlin‐1 serum concentrations observed in the study, possibly attributed to target binding in which the ginisortamab‐bound gremlin‐1 had an extended t 1/2 compared with free gremlin‐1, was adequately captured by the model. Conversely, the estimated internalization rate constant (K int) for the ginisortamab–gremlin‐1 complex exceeded the apparent elimination rate of free (unbound) ginisortamab (CL/V c). This finding may indicate there is target‐mediated internalization and degradation of the complex, which can proceed via cellular uptake pathways distinct from FcRn‐mediated recycling of unbound IgG. Baseline BSA was identified as the only significant covariate on ginisortamab PK (P < .001); individuals with high BSA were estimated to have a higher volume of distribution than those with low BSA, although this was expected because measures of body size have been commonly identified as covariates that influence the PK of therapeutic mAbs. 30 , 32 No additional covariates were found to significantly influence the PK or PK/PD model, so only BSA was included in the final PK/PD model, using a power function centred to 2 m2 and with the exponent estimated as 1.3 for V c and 2.2 for V p, suggesting a nonlinear relationship between BSA and volume of distributions. The differing exponents estimated for V c and V p likely reflect the empirical model optimization rather than true physiological differences in scaling. In two‐compartment models, these volumes represent structural distribution parameters and are often highly correlated. Therefore, the estimated exponents should be interpreted as descriptive of the dataset rather than as mechanistic constants. Although cancer type was not found to be a significant covariate, this finding should be interpreted with caution, because of the small number of participants with certain cancer types in the ONC001 dose‐escalation module; most participants had colorectal adenocarcinoma (n = 11) or pancreatic adenocarcinoma (n = 6) with other tumour types present in three or fewer participants. Although measuring and comparing concentrations of free target before and after drug dosing can directly inform the level of target suppression, it is not always possible to develop a reliable free‐target assay, especially for targets of low abundance and rapid turnover, such as gremlin‐1. Challenges arise not only from the low concentration of the free‐target (for which an assay with high sensitivity is needed) relative to the high concentrations of drug and drug–target complex (for which an assay with high tolerance is needed) in post‐dose samples but also from the dissociation of the drug–target complex during storage and shipment, which can inflate the amount of free target in samples. Conversely, the measurement of total target concentration together with drug concentration could provide useful data, which can be analysed with PK/PD modelling approaches that help to infer the level of suppression of the free target. 33 For novel targets in which turnover times are not well known, PK/PD modelling allows the estimation of this parameter, which is critical in determining the level of target suppression and ultimately plays a role in the selection of the optimal dosing regimen. Estimation from our PK/PD model suggested that gremlin‐1 has a fast turnover time in serum (typically 23 min), and, to our knowledge, this is the first reported estimate of gremlin‐1 turnover time in humans. Soluble serum proteins may exhibit rapid systemic turnover, because of renal clearance or receptor‐mediated uptake, even when their biological effects within tissues are sustained through localized production or compartmental retention. Therefore, the serum turnover estimate should be interpreted as representing circulating gremlin‐1 dynamics. Nevertheless, a similarly rapid turnover of gremlin‐1 in the tissue microenvironment is feasible. The physiological antagonism of BMP signalling by gremlin‐1 produces highly localized morphogen gradients (for example, in the base intestinal crypts). Such systems require rapid morphogen turnover to achieve this level of localization, and a recent study has indeed shown that, after secretion by intestinal fibroblasts, gremlin‐1 is endocytosed by neighbouring epithelial cells. 34 Simulations with the final PK/PD model allow identification of dosing regimens that might achieve a high level of gremlin‐1 suppression in solid tumours associated with GREM1 expression. The PK and total gremlin‐1 serum concentrations measured following the administration of 2000 mg ginisortamab Q2W in the ONC001 dose‐escalation and dose‐optimization modules were generally in line with the simulations generated from this PK/PD model (data not shown). The predicted median ginisortamab accumulation ratio based on an AUC0–336h of 1.7 and the predicted median accumulation ratio based on a C max of 1.4 suggest low‐to‐moderate accumulation at steady state. The 1000 mg QW regimen and the 2000 mg Q2W regimen monotherapy were predicted to produce a median target suppression close to 90% over the dosing interval at steady state. Simulations of the dosing regimens used in the ONC001 dose‐optimization module suggest that applying a higher dose with an extended dosing interval of every 3 weeks (3000 mg Q3W) and every 4 weeks (4000 mg Q4W) would provide a slightly lower target suppression at the end of the dosing interval (attributable to a fast turnover time of free gremlin‐1), but these regimens may still be considered for future investigation. There are some limitations to this PK/PD analysis. Data were based on the modest number of participants in the ONC001 dose‐escalation module, which may have limited the ability to identify significant covariates, impacting the generalizability of the findings. For instance, five of the seven primary tumour types diagnosed in participants from the ONC001 dose‐escalation module were in only 1–3 participants, which prevented the evaluation of the potential impact of tumour type on PK, baseline gremlin‐1 level and gremlin‐1 turnover. Heterogeneity among participants in disease status, medical history, and prior and concomitant treatments was present, but its potential impact could not be fully evaluated because of the relatively small sample size. The relationship between serum gremlin‐1 suppression, tumour gremlin‐1 suppression and anti‐tumour effects of ginisortamab also remains to be explored. Lastly, in the absence of an assay that can reliably measure the concentration of free gremlin‐1 following ginisortamab administration, the level of gremlin‐1 suppression predicted by our simulations will have to be confirmed with the measurement of downstream markers of gremlin‐1 activity. To conclude, measurement of free PK and total target concentration in this first‐in‐human study of ginisortamab has allowed the determination of dose regimens to be considered for future Phase 2 studies using a population PK/PD modelling approach. Model‐based simulation of the suppression of gremlin‐1 supports the potential use of several ginisortamab dosing regimens for further evaluation of efficacy in patients with advanced solid tumours. In the ongoing Phase 2 clinical study evaluating ginisortamab in participants with PDAC, the most common form of pancreatic cancer, an initial starting‐dose regimen of 2000 mg Q2W, was adopted, based on the safety profile observed in ONC001 and our PK/PD simulations, which suggested that this regimen maintains a high level of gremlin‐1 suppression throughout the dosing interval. AUTHOR CONTRIBUTIONS Conception and design of the work: Yin Cheong Wong, Paul Jones, David Egging and Ivan Matthews. Development of methodology: Yin Cheong Wong, Luisa Rojas Chavarro, David Egging, Wanchana Ungphakorn and Ivan Matthews. Acquisition of data: Paul Jones. Analysis and interpretation of data: Yin Cheong Wong, Luisa Rojas Chavarro, David Egging, Wanchana Ungphakorn, Marta Valle and Ivan Matthews. Study supervision: Ivan Matthews. Writing, review and/or revision of the manuscript: All authors. Final approval of the manuscript: All authors. CONFLICT OF INTEREST STATEMENT Y.C.W and P. J are employees of UCB and have stocks/shares in UCB, and P. J is named as an inventor on ginisortamab‐related patents; D.E. is an employee of UCB; I.M. is an employee of UCB and has stocks/shares in AstraZeneca, Eli Lilly, Gilead, Haleon, Novo Nordisk and UCB; L.R.C., W.U. and M.V. are employees of Parexel International. Supporting information Table S1. PKPD model population summary demographic and baseline characteristics (participants from ONC001 monotherapy dose‐escalation). Table S2. (A) ginisortamab serum concentrations excluded from PK model development; (B) total gremlin‐1 serum concentrations excluded from PKPD model development; (C) available ginisortamab and total gremlin‐1 serum concentration records by dose. Table S3. PKPD model predicted median (90% prediction interval) PK parameters after administration of ginisortamab at first dose–first cycle and last dose–last cycle (steady‐state). Figure S1. Observed vs. predicted goodness‐of‐fit plots for: (A) ginisortamab serum concentrations; (B) total gremlin‐1 serum concentrations in the final PKPD model. Figure S2. Prediction‐corrected visual predictive check of final PK model for ginisortamab: (A) time after first dose; (B) first 14 days after first dose. Figure S3: Predicted percentage gremlin‐1 target suppression after IV ginisortamab administration (simulation of ONC001 monotherapy dose‐optimization module regimens and additional hypothetical dosing regimen).