The Pharmacokinetic–Pharmacodynamic (PKPD) Relationships of AZD3229, a Novel and Selective Inhibitor of KIT, in a Range of Mouse Xenograft Models of GIST

KIT belongs to a family of transmembrane tyrosine kinase growth factor receptors and gain-of-function mutations that result in constitutive KIT activation, which has an important pathogenic role in gastrointestinal stromal tumors (GIST; ref. 1). Existing tyrosine kinase inhibitors such as imatinib (approved first line), sunitinib (approved second line), and regorafenib (approved third line) may initially control GIST, but resistance develops due to emergence of secondary mutations in the ATP-binding pocket and activation loop of KIT (2–5). This presents a major challenge for targeted drug discovery efforts seeking improved small-molecule inhibitors, primarily due to the complex heterogeneity of oncogenic KIT mutations found in patients. The existing approved KIT inhibitors for GIST treatment inherently lack activity against a spectrum of both primary and secondary mutations and have numerous dose-limiting safety liabilities such as high-grade hypertension (6) that can lead to dose reductions and drug holidays (6, 7). Ripretinib (DCC-2618) and avapritinib (BLU-285) are KIT inhibitors undergoing investigation in clinical trials in patients with GIST (8–12). Our goal has been to develop a compound targeting a wide spectrum of known primary and secondary resistance mutations of KIT/PDGFRα, while ensuring a wide margin against the main antitarget, VEGFR-2 to minimize clinical toxicities such as high-grade hypertension. AZD3229 is an oral, potent KIT/PDGFRα inhibitor that is active against a wide spectrum of primary and secondary KIT/PDGFRα mutations that are known to confer resistance to standard-of-care (SoC) agents (13). It is also expected to avoid significant VEGFR-2 activity at clinically relevant exposures (13, 14).

Pharmacokinetic–pharmacodynamic (PKPD) modeling has emerged as an important capability in drug discovery and development to assist in the design of studies and the integration and analysis of datasets to quantitatively investigate the understanding of drug action (15). For Oncology targets, PKPD modeling is used to establish a quantitative understanding of the target/pathway modulation requirements for optimal antitumor activity and based on this understanding, it is possible to identify candidate drug molecules and predict and prioritize a human dose and schedule to test in the clinic (16, 17).

As far as we are aware, there are no published PKPD models exploring the relationship between exposure, target suppression, and antitumor activity in KIT-dependent mouse models of GIST. To explore the PKPD relationships, we used a number of GIST cell line–derived xenograft (CDX), allograft, and patient-derived xenograft (PDX) models available that harbor different primary and secondary mutations of KIT.

The objective of this PKPD modeling work was 4-fold: (i) For the in vivo models, investigate and establish the relationship between AZD3229 exposure and inhibition of phosphorylation of KIT; (ii) confirm the extent and duration of inhibition of KIT phosphorylation required for optimal antitumor activity, as defined by tumor regressions; (iii) using a prediction of human pharmacokinetics of AZD3229, estimate the dose anticipated to be required to deliver the extent and duration of inhibition of KIT phosphorylation necessary to control tumor growth as defined by the preclinical studies; (iv) where feasible use published data for other key KIT inhibitors to benchmark AZD3229 and evaluate the translational assumptions presented here.

The in vitro binding of AZD3229 to plasma protein from the mouse, and mixed human plasma was measured over a concentration range of 0.1 to 100 μmol/L AZD3229, using equilibrium dialysis method following 20-hour incubation. The summary of plasma protein binding (% unbound) of AZD3229 in the mouse and human (mean ± SD) is shown in Supplementary Table S1. The binding of AZD3229 to 10% of FCS was also determined (Supplementary Table S1).

The potency of AZD3229 along with approved (imatinib, sunitinib, and regorafenib) and investigational agents (ripretinib and avapritinib) have been reported previously in a cell viability assay across a Ba/F3 cell panel as the GI₅₀ (10, 13) and GI₉₀ (14). The known clinical exposures (Supplementary Table S2) for KIT inhibitors at the approved dose for SoC agents and the recommended phase II dose for the investigational agents have been taken from the literature and the plasma concentration at trough (C_trough) compared and color coded as green if the C_trough exceeds the GI₉₀ (Supplementary Table S3).

To define the plasma pharmacokinetic properties of AZD3229 in mouse, a multiple sampling PK study after intravenous (5 mg/kg) and oral gavage (10 mg/kg) dosing in nontumor-bearing mice (n = 2) was conducted for AZD3229. Plasma samples were taken by serial bleed at various time points up to 24 hours after dosing. The plasma concentrations of AZD3229 were determined using a protein precipitation procedure followed by LC/MS-MS detection. The lower limit of quantification was 0.0014 μmol/L. Calibration standards passed with percent coefficient of variation (CV) less than 6% and percent bias in the assay in the range of −16% to 8%. The animal PK studies were conducted under the condition of established Institutional Animal Care and Use Committee (IACUC) guidelines.

The in vivo studies used in these analyses are described in detail with the primary data, including tumor growth curves and immunoblots by Banks and colleagues (14). All animal studies were conducted under the condition of established IACUC guidelines. Briefly, the level and duration of AZD3229-driven inhibition of phosphorylation of KIT (pKIT) was determined by Western blot analysis from allograft Ba/F3 (KIT exon 11 del 557-558/D816H), CDX GIST430/V654 (KIT exon 11 del 560-578/V654A), and PDX tumor models HGiXF-105 (GS5108; KIT exon 11 del 557-558/Y823D; Crown Bioscience) and HGiXF-106 (GS11331; KIT exon 11 del 557-558/V654A; Crown Bioscience). Mice were orally dosed with AZD3229 at several dose levels between 0.1 and 40 mg/kg, either as a single dose or for 3 consecutive days. Plasma PK and tumor samples were taken at several time points up to 24 hours after the last dose. Plasma concentrations of AZD3229 and levels of phosphorylated KIT were measured in each sample. Supplementary Table S4 provides details of the doses tested in each model, the measured AZD3229 plasma concentrations, and change in phosphorylated KIT at the respective time points selected for measurement.

The antitumor activity was also explored in these models using AZD3229 (see details of the doses tested in each efficacy model in Supplementary Table S5), along with approved (imatinib, sunitinib, and regorafenib), and investigational (ripretinib and avapritinib) agents. Mice were dosed twice daily (AZD3229 and ripretinib) or once daily (imatinib, sunitinib, regorafenib, and avapritinib) by oral gavage and the tumor growth inhibition from start of treatment was assessed by calculating the mean percentage change in tumor size at the end of treatment compared with baseline (start of treatment) as shown in equation (1) below.

PK samples were taken at the end of the efficacy studies via serial bleeding of 3 animals per dose group after the last dose.

Visualization of the data by plotting AZD3229 plasma concentration against observed change in KIT phosphorylation relative to baseline for each of the four in vivo models did not reveal any significant hysteresis suggesting a direct relationship between plasma concentration and changes in KIT phosphorylation. Therefore, the pharmacodynamic model assumes a direct response relationship between plasma exposure and level of KIT phosphorylation. A schematic of the PKPD model structure is shown in Supplementary Fig. S1.

Equation (2) above shows the mathematical expression used to relate AZD3229 plasma concentration to pKIT level, where E_max is the maximal effect observed, EC₅₀ is the concentration that achieves 50% of E_max, base pKIT is the baseline pKIT level which was set to 100, γ is the hill coefficient, and C_p is the drug concentration in plasma. Optimal parameter estimates can be derived by fitting the model relating the observed plasma concentration against the observed change in KIT phosphorylation from baseline. A PK model, using pooled dataset, with first-order absorption and elimination was built using NONMEM 7.1 software (ICON Development Solutions). A sequential PKPD model was built using Phoenix 6.4 software (Certara) by fixing pooled PK model parameters (Supplementary Table S6) and estimating PD model parameters such as EC₅₀. The PKPD model was utilized to simulate the time course of decrease of KIT phosphorylation seen in tumor models.

The potency of a drug is quantified as the EC₅₀, and equation (3) can be used to calculate the concentration necessary to deliver any degree of decrease, where DEC% is the percentage change required and EC_DEC is the concentration necessary to deliver DEC% decrease.

For any translational calculations or comparisons (e.g., calculating potency to set a human dose), the potency values from the mouse tumor models were corrected for the difference in plasma protein binding.

In mouse, the plasma pharmacokinetics of AZD3229 is characterized as being a low volume of distribution (0.7 L/kg), low clearance (7 mL/min/kg) compound with good oral bioavailability and a half-life of approximately 2 hours (13).

Following oral dosing, there is a proportional increase in exposure with dose and a one-compartment model with first-order absorption and elimination is adequate to describe the concentration-time profile observed across the dose range (0.5–20 mg/kg) tested. Thus, a pooled PK model was parameterized (Supplementary Table S6) and model simulations are compared against representative data across the dose range in Supplementary Fig. S2. This PK model was subsequently used to simulate the plasma concentration–time course to drive the model of inhibition of phosphorylation of KIT.

In the four in vivo models explored here, AZD3229 drives rapid and extensive decrease in levels of phosphorylated KIT in a concentration-dependent manner following a single oral dose of AZD3229 (Fig. 1; Supplementary Table S4). The observed maximal inhibition of phosphorylation of KIT relative to baseline occurs around the time of maximum AZD3229 (2–3 hours), and the level of pKIT recovers to baseline levels tracking the PK profile. In the GIST430/V654 (KIT exon 11 del 560-578/V654A) and HGiXF-106 (KIT exon 11 del 557-558/V654A) models, the level of KIT phosphorylation reaches above the vehicle baseline at the 24-hour time point, but this was considered within assay variability, rather than representing significant PD rebound above baseline. The effects on KIT phosphorylation levels after 3 days of twice daily oral dosing was also investigated (except in the Ba/F3 allograft model), and the response mirrored that seen after a single dose, which was evidence that there was no time-dependent change for the inhibition of phosphorylation of KIT.

On the basis of the PD results described above, it was assumed that (i) a direct response relationship exists between plasma PK and changes in tumor pKIT; (ii) no change in PD response is observed on repeat dosing; and (iii) pKIT rebound above baseline is insignificant. Therefore, an E_max model was fitted to the data from each in vivo model. On the basis of initial visual inspection of the data, E_max and the hill coefficient were fixed to 1 and the EC₅₀ was varied to achieve an optimal relationship (Fig. 1). Once the EC₅₀ had been derived, the EC₉₀ was calculated [equation (3)]. The results of this analysis show that the EC₅₀ for decrease of KIT phosphorylation varies across models, with AZD3229 demonstrating lowest EC₅₀ in the HGiXF-106 model (KIT exon 11 del 557-558/V654A; 0.4 nmol/L) followed by Ba/F3 (KIT exon 11 del 557-558/D816H; 2.2 nmol/L), GIST430/V654 (KIT exon 11 del 560-578/V654A; 4.8 nmol/L), and HGiXF-105 (KIT exon 11 del 557-558/Y823D; 9 nmol/L; Table 1).

When combined with the mouse PK model for AZD3229, the time course of change in pKIT levels simulated for the doses tested in the respective in vivo models agrees adequately with the observed data (Fig. 1). Example goodness-of-fit plots are shown in Supplementary Fig. S3.

Following twice daily oral dosing of AZD3229, the antitumor activity observed in the four in vivo models increases with dose resulting in tumor regressions at the highest doses tested (Supplementary Table S5). At these higher doses, it is clear from the pKIT data that a decrease in the levels of pKIT greater than 90% is achieved for a substantial proportion of the dosing interval (Fig. 1). On this basis, and to establish the relationship between the level of pKIT decrease and degree of antitumor activity observed, the PK-pKIT model was used to calculate the time above the EC₉₀ over a 24-hour period at each dose used in the efficacy studies. The time above EC₉₀ was then plotted against the level of antitumor activity measured (Fig. 2) and a strong trend is observed, such that increased antitumor activity correlates with a longer time above EC₉₀. Thus, greatest depth of antitumor activity is seen at 20 mg/kg in the HGiXF-106 (KIT exon 11 del 557-558/V654A) model and at this dose, the time above EC₉₀ is approximately 15 hours. This correlation also suggests that the inhibition of phosphorylation of KIT to efficacy relationship is broadly consistent, showing the same trend across the four in vivo models tested. Therefore, duration of pKIT decrease can be used as a predictor of antitumor activity regardless of the KIT secondary mutation.

The PK–pKIT relationship has been explored in four in vivo models, and this covers a narrow range of known mutations. Additional mutations are known to confer resistance to imatinib, sunitinib, and regorafenib and the potency of these agents along with AZD3229 and investigational agents (ripretinib and avapritinib) have been previously reported in a cell viability assay across a Ba/F3 cell panel as the GI₅₀ (13) and GI₉₀ (Supplementary Table S3; ref. 14). It would be desirable to demonstrate a correlation between the cell viability GI₉₀ and in vivo EC₉₀ to be able to use the in vitro data to more fully assess relative activity and exposure requirements across a more extensive number of KIT mutations. For the GIST430/V654 (KIT exon 11 del 560-578/V654A) and Ba/F3 (KIT exon 11 del 557-558/D816H) models, the cell lines have been grown in vitro and a direct comparison with the in vivo potency is possible. HGiXF-105 (KIT exon 11 del 557-558/Y823D) and HGiXF-106 (KIT exon 11 del 557-558/V654A) are PDX models and corresponding cellular data are not available and so the closest equivalent Ba/F3-panel cell lines have been used (Table 1). There is a strong in vitro to in vivo correlation with the difference being within 1- to 4-fold: approximately 1-fold for HGiXF-106 (KIT exon 11 del 557-558/V654A) and Ba/F3 (KIT exon 11 del 557-558/D816H); approximately 2.5-fold for HGiXF-105 (KIT exon 11 del 557-558/Y823D); and approximately 4-fold for GIST430/V654 (KIT exon 11 del 560-578/V654A).

The PKPD insights described above for AZD3229 should be generalizable to other KIT inhibitors. For the SoC agents (imatinib, sunitinib, and regorafenib), the potency against the various KIT mutations in the Ba/F3 cell panel are available, along with in vivo antitumor activity measured at doses in a mouse that deliver clinically relevant exposures (Supplementary Table S2; ref. 14). The mouse equivalent doses for imatinib (300 mg/kg), sunitinib (80 mg/kg), and regorafenib (100 mg/kg) were calculated using a PK-guided approach by matching clinical dose normalized unbound exposure in human to mouse, rather than using an empirical factor (18).

The in vitro GI₉₀ values have been plotted alongside the mouse in vivo exposure observed in the in vivo studies (Fig. 3). The GI₉₀ values were used rather than EC₉₀ as a full characterization of the in vivo exposure–pKIT relationship has not been completed for these agents. These plots (Fig. 3) show that the in vivo antitumor activity is predictable from the in vitro GI₉₀ in the respective Ba/F3 cell line, combined with the observed exposure achieved in the in vivo study: Imatinib shows little duration above the respective in vitro GI₉₀, and minimal antitumor activity in vivo at a dose of 300 mg/kg across the four models tested [HGiXF-106 (KIT exon 11 del 557-558/V654A), HGiXF-105 (KIT exon 11 del 557-558/Y823D), Ba/F3 (KIT exon 11 del 557-558/D816H), and GIST430/V654 (KIT exon 11 del 560-578/V654A)] and this is consistent with the known lack of clinical activity in patients with secondary mutations. Sunitinib is clinically active in patients with secondary mutations of the ATP-binding pocket and it shows antitumor activity in the GIST430/V654 (KIT exon 11 del 560-578/V654A) and HGiXF-106 (KIT exon 11 del 557-558/V654A) models with an exposure profile at 80 mg/kg that delivers 6 hr [GIST430/V654 (KIT exon 11 del 560-578/V654A)] to 24 h [Ba/F3 (KIT exon 11 del 557-558/V654A)] coverage above the GI₉₀. Meanwhile regorafenib has been tested in vivo in the Ba/F3 (KIT exon 11 del 557-558/D816H) and HGiXF-105 (KIT exon 11 del 557-558/Y823D) models and delivers antitumor activity in these models with an exposure profile at 160 mg/kg that provides continuous coverage above the respective GI₉₀.

Avapritinib (13, 19), approved for PDGFRα-driven GIST, and ripretinib (13, 20) are structurally distinct inhibitors of KIT that are in clinical development and offer the opportunity to evaluate the PKPD insights derived from AZD3229. At a dose of 30 mg/kg, avapritinib shows limited tumor inhibition in GIST430/V654 (KIT exon 11 del 560-578/V654A), and tumor shrinkage in HGiXF-105 (KIT exon 11 del 557-558/Y823D), with an unbound plasma concentration that delivers no cover and 24 h cover above the respective GI₉₀ (Fig. 3). Ripretinib shows little tumor growth inhibition in the models tested at a dose of 50 mg/kg twice a day, and the mouse exposure provides no cover above the GI₉₀. It should be noted that this mouse dose does not achieve clinical exposures observed (10). Higher doses were not explored due to body weight loss at 50 mg/kg twice a day; therefore, it is unknown whether antitumor activity would have been achieved at higher doses that deliver exposures sufficient to exceed the GI₉₀. It was not known at the time the data were generated that ripretinib has an active metabolite in mouse and human that is equipotent and observed at equivalent concentrations to parent in mice (10), further complicating interpretation of the PKPD relationship. If we account for the active metabolite present in the plasma (assuming equivalent concentrations as parent), then the net plasma concentration of active KIT inhibition will be approximately double than that observed for parent, and this would approach the GI₉₀ at around the time of maximal plasma concentration, but fail to give significant duration of inhibition required for antitumor activity.

Overall, the results of this analysis for the SoC and investigational KIT inhibitors builds further evidence for the need for durable inhibition of KIT phosphorylation >90% and that the in vitro GI₉₀ of the relevant Ba/F3 cell line is predictive of the exposure (and thus dose) required in vivo for efficacy in the respective model.

The GI₉₀ values (Supplementary Table S3; ref. 13) across the mutant KIT and PDGFRα Ba/F3 cell panel for the SoC agents and the investigational drugs ripretinib and avapritinib have been color coded in such a way that if the observed clinical exposure (at C_trough) exceeds the GI₉₀ for the respective mutation then it is marked green. The aim is to provide a useful way with which to benchmark each compound and visualize how the in vitro potency values correlate with clinical activity in the different patient groups. The pattern observed for the SoC agents fits with the known clinical activity such that imatinib is active against primary mutations but not secondary mutations and it is only the primary mutations that are colored green. Meanwhile sunitinib provides good cover against secondary mutations in the ATP-binding pocket but does not do so for those secondary mutations in the A-loop domain. Regorafenib and ripretinib provide broad coverage (10), particularly for the A-loop domain, but the clinical data for ripretinib are not sufficiently mature to comment on the genotype-mediated efficacy. On the basis of the disclosed PK data for avapritinib, at the dose being used in phase III testing, the compound lacks the exposure to adequately cover the secondary mutations in the ATP-binding pocket. This fits with a lack of clinical activity observed for avapritinib in the KIT ATP-binding mutant population and the current testing of this compound in patients is restricted to KIT V654A and T670I negative GIST population in a phase III trial (21). The backtranslation of clinical data to correlate with the Ba/F3 cell-panel GI₉₀ values for the SoC agents builds confidence in the use of these in vitro data as a benchmark to assess the exposure requirements for KIT inhibitors—including predictions for AZD3229—to adequately cover the spectrum of mutations represented.

A plasma concentration greater than the EC₉₀ (derived from the mouse in vivo models) over the dosing interval was set as the minimal requirement to estimate an active clinical dose. Active human doses have been estimated by combining a prediction of the human pharmacokinetics of AZD3229 with the mouse PD model parameters (corrected for differences in plasma protein binding mouse to human; Supplementary Table S1). In preclinical species, the apparent volume of distribution of AZD3229 is low in mouse (0.7 L/kg), rat (0.6 L/kg), and dog (0.3 L/kg), and oral bioavailability is high in mouse (>90%), rat (79%), and dog (69%; ref. 13). On the basis of these data, the volume of distribution is predicted to be low in human (0.5 L/kg), and absorption following oral dosing high (80%). Hepatic metabolism is the main route of elimination in preclinical species and is assumed to be so in patients. Using human hepatocytes, the intrinsic clearance has been measured (<1 μL/minute/10⁶ cells; ref. 13), and scaled to the whole liver using scaling methodology (22) to provide a predicted clearance of 0.4 mL/minute/kg, resulting in a predicted half-life of 15 hours.

Using the predicted human PK, the doses (on a twice daily schedule) required to deliver plasma concentrations > EC₉₀ continuously were calculated on the basis of Ba/F3 (KIT exon 11 del 557-558/D816H; 8 mg), GIST430/V654 (KIT exon 11 del 560-578/V654A; 18 mg), HGiXF-105 (KIT exon 11 del 557-558/Y823D; 34 mg), and HGiXF-106 (KIT exon 11 del 557-558/V654A; 2 mg) models as shown in Table 1 and in Fig. 4. This represents a near 20-fold range, which mirrors the range in EC₉₀ derived across the models, and highlights the exposure required to drive adequate inhibition of KIT activity may differ depending on the mutation(s) that are present in the clinical tumor. The models explored here represent the most common ATP-binding mutation in the clinic (V654A) and the least sensitive A-loop mutation to imatinib (D816H) thus suggesting that AZD3229 is well placed to inhibit mutant KIT tumors in the clinic (23). However, four in vivo models limit our ability to assess the full spectrum of apparent sensitivity that is likely across the complement of known mutations and the Ba/F3 cell panel offers the opportunity to explore in vitro the relative sensitivity of the different mutations to KIT inhibition (Supplementary Table S3). The GI₉₀ values range from 1 nmol/L for the exon 11 del 557-558 primary mutation to 74 nmol/L for the exon 9 insertion AY502-503/D816H secondary mutation. This range is consistent with the range in EC₉₀ observed in vivo, and the upper predicted active dose of 34 mg would provide adequate exposure to cover all mutations tested in vivo and represented in the Ba/F3 cell panel. The analysis presented here gives confidence in taking the compound forward to test in patients with GIST. AZD3229 is also able to inhibit the KIT exon 9 primary mutation, thus differentiating it from imatinib that is subefficacious in the clinic for this mutation (Supplementary Table S3).

Imatinib provides front-line treatment for GIST and relapse on treatment is driven by the emergence of several secondary ATP-binding and A-loop KIT mutations. The current second- and third-line therapies—sunitinib and regorafenib—do not have strong activity against the full spectrum of secondary KIT mutations at clinically achievable doses/exposures (24) and a lack of selectivity results in a tolerability profile that requires management. This provides a clinical opportunity for a KIT inhibitor that delivers PKPD characteristics that enable strong activity across a wider spectrum of secondary KIT mutations. However, little has been published describing a systematic PKPD analysis to define the requirements for such a compound.

A combination of in vitro Ba/F3 cell panel and four mouse in vivo models have been used to characterize the PKPD understanding for what level of target suppression drives deep and durable antitumor activity. We have shown that following oral dosing in mice, AZD3229 drives rapid and extensive decrease of KIT phosphorylation in an exposure-dependent manner and the derived EC₅₀ and EC₉₀ varies depending on the mutation the model harbors. Changes in pKIT levels were measured at several time points and dose levels in the four models tested. This enabled the time course of pKIT level changes to be observed and confirm that the pKIT levels tracked changes in drug exposure without any significant time delay. This study design also facilitated robust model estimates for the in vivo potency in each model, as indicated by low CVs derived on the EC₅₀ estimates. Examination of the Ba/F3 cell-panel viability data suggests that there is a good correlation between the relevant GI₉₀ in vitro and the estimated EC₉₀ in the respective in vivo model (Table 1). Although the EC₉₀ varies across the models, optimal antitumor activity is consistently observed when the pKIT level is suppressed >90% durably over the dosing interval. This was further supported by experiments with a probe compound (Supplementary Fig. S4) in the Ba/F3 (KIT exon 11 del 557-558/V654A) mouse allograft model where different dosing schedules were used to control the duration of inhibition of KIT phosphorylation. Mice received the same daily dose as a single 100 mg/kg every day, twice daily 50 mg/kg, or three times daily 33 mg/kg, and it was found that 33 mg/kg three times daily showed the strongest antitumor activity, followed by the 50 mg/kg twice a day, with the least active being the 100 mg/kg every day dose. All three dose levels would be expected to deliver near maximal decrease in KIT phosphorylation, but the more frequent dosing ensures more durable coverage of the target over each day of dosing.

Taken together, the requirement for continuous inhibition of phosphorylation of KIT >90% has been used to anchor the predicted human dose. Because the exposure necessary to achieve this depends on the mutation the tumor harbors, the predicted dose also varies depending on the model used, with the lowest dose predicted to be 2 mg based on the HGiXF-106 (KIT exon 11 del 557-558/V654A) model and the highest dose predicted to be 34 mg based on HGiXF-105 (KIT exon 11 del 557-558/Y823D) model. Because the interpatient and intrapatient tumor heterogeneity is large in the GIST population, the exploration of the PKPD requirements in multiple models provides a means to provide some preclinical assessment of the diversity of response likely to be seen in the clinic. This is an advantageous approach compared with relying on a single model that may be very sensitive but underestimates the exposure requirements for effective treatment across a population of patients. The use of PDX models further increases the translational relevance compared with relying solely on CDX models (22, 23). Indeed, AZD3229 has been tested in two further PDX models (purchased from Crown Bioscience) as described by Banks and colleagues (14): the HGiXF-107 (GS5106; KIT K642E/N822K) and HGiXF-108 (GS11327; KIT exon 11 del K550fs). Both models are sensitive to AZD3229 showing tumor shrinkage at a dose of 4 mg/kg. These datasets are less comprehensive than the other models presented here and precluded full PKPD analysis. However, given that these models are on the sensitive end of the spectrum of models tested in vivo, they represent models that would have predicted doses at the lower end of the range, and in the context of predicting the human dose from preclinical insights, it is more important to understand the impact on less sensitive models and any resulting risk to the feasibility of dosing patients to deliver adequate target coverage. Indeed in a setting such as GIST where there is such a spectrum of mutations that exhibit different sensitivity to AZD3229 inhibition, it may be desirable to have the option to be able to dose beyond the exposure level predicted from the available preclinical datasets to offer the opportunity to inhibit mutations that have not been captured in the Ba/F3 cell panel, which may be less sensitive to inhibition by AZD3229. We believe that the overall pharmacokinetics of AZD3229 should allow dosing beyond 150 mg (simulation shown in Fig. 4) before the biopharmaceutical properties begin to limit and reduce to any significant extent the fraction of dose orally absorbed (Supplementary Fig. S5).

An additional benefit of AZD3229 is that it has a wide margin against VEGFR-2 (6, 14) and at the predicted clinical exposures, it may have a lower risk of causing significant hypertension. Sunitinib and regorafenib are associated with high-grade hypertension, which requires clinical management, and ripretinib has also been reported to show grade 3 and 4 hypertension in a recent phase I clinical study (11, 24).

Taken as a whole, our hypothesis is that AZD3229 offers a PKPD profile that, based on the translational modeling, would provide an exposure profile to deliver adequate inhibition of phosphorylation of KIT across a broad spectrum of primary and secondary KIT mutations seen in patients with GIST and warrants clinical testing.

It should be acknowledged that the use of preclinical data to predict clinical efficacy can be uncertain (25) but the use of Ba/F3 model has shown success (20, 22), and the PKPD datasets in vitro (Ba/F3 cell panel) and in mouse in vivo models generated for the SoC and investigational agents are consistent with the known clinical activity in this heterogenous GIST patient population. Moreover, the translational approach presented here is attempting to go beyond relying simply on mouse efficacy data to predict to the clinic by incorporating a more in-depth quantitative analysis of the PK and PD. This includes in vitro and mouse in vivo data and the backtranslation of data for SoC agents to define appropriate translational assumptions. More generally, building a quantitative PKPD model such as the one presented here for AZD3229 provides a modeling framework to integrate datasets including for SoC agents (in vitro, in vivo, and clinical), with the aim of better understanding translational uncertainties and risk (26). As a molecule moves toward and into the clinic, the PKPD modeling framework provides a translational bridge from nonclinical to clinical and back, which can be used to support the study design in terms of setting dose and schedule and sampling time points for PK and PD endpoints; set decision-making criteria for PK and PD in the clinic; benchmark emerging PK and PD data against competitor data and preclinical insights; and refine translational assumptions in the model.

V. Pilla Reddy and R. Anjum are employees/paid consultants for AstraZeneca. A. Smith is an employee/paid consultant for and holds ownership interest (including patents) in AstraZeneca. D. Bhavsar, S.M. Guichard, W. Shao, J.G. Kettle, and R.D.O. Jones are employees/paid consultants for AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: V. Pilla Reddy, R. Anjum, E. Barry, S.M. Guichard, W. Shao, J.G. Kettle, R.D.O. Jones

Development of methodology: V. Pilla Reddy, R. Anjum, R.D.O. Jones

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Pilla Reddy, R. Anjum, M. Grondine, A. Smith, D. Bhavsar, E. Barry, C. Brown

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Pilla Reddy, R. Anjum, A. Smith, W. Shao, R.D.O. Jones

Writing, review, and/or revision of the manuscript: V. Pilla Reddy, R. Anjum, M. Grondine, E. Barry, S.M. Guichard, W. Shao, R.D.O. Jones

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Pilla Reddy, C. Brown, R.D.O. Jones

Study supervision: S.M. Guichard, R.D.O. Jones

This study was sponsored by AstraZeneca. The authors wish to thank Camila de Almeida, for her scientific help in carrying out the PKPD-efficacy study procedures described.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.