Pharmacokinetic/Pharmacodynamic Analysis of Savolitinib plus Osimertinib in an EGFR Mutation–Positive, MET-Amplified Non–Small Cell Lung Cancer Model

Abstract

Osimertinib is a third-generation, irreversible, oral EGFR tyrosine kinase inhibitor (TKI) recommended as first-line treatment for patients with locally advanced/metastatic EGFR mutation–positive (EGFRm) non–small cell lung cancer (NSCLC). However, MET amplification/overexpression is a common acquired osimertinib resistance mechanism. Savolitinib is an oral, potent, and highly selective MET-TKI; preliminary data suggest that combining osimertinib with savolitinib may overcome MET-driven resistance.

A patient-derived xenograft (PDX) mouse model with EGFRm, MET-amplified NSCLC was tested with a fixed osimertinib dose [10 mg/kg for exposures equivalent to (≈)80 mg], combined with doses of savolitinib (0–15 mg/kg, ≈0–600 mg once daily), both with 1-aminobenzotriazole (to better match clinical half-life). After 20 days of oral dosing, samples were taken at various time points to follow the time course of drug exposure in addition to phosphorylated MET and EGFR (pMET and pEGFR) change. Population pharmacokinetics, savolitinib concentration versus percentage inhibition from baseline in pMET, and the relationship between pMET and tumor growth inhibition (TGI) were also modeled.

As single agents, savolitinib (15 mg/kg) showed significant antitumor activity, reaching ∼84% TGI, and osimertinib (10 mg/kg) showed no significant antitumor activity (34% TGI, P > 0.05 vs. vehicle). Upon combination, at a fixed dose of osimertinib, significant savolitinib dose-related antitumor activity was shown, ranging from 81% TGI (0.3 mg/kg) to 84% tumor regression (15 mg/kg). Pharmacokinetic–pharmacodynamic modeling showed that the maximum inhibition of both pEGFR and pMET increased with increasing savolitinib doses.

Savolitinib demonstrated exposure-related combination antitumor activity when combined with osimertinib in the EGFRm MET-amplified NSCLC PDX model.

Introduction

MET is a receptor tyrosine kinase that is a central regulator of tissue homeostasis (1). Dysregulation of MET signaling is associated with several human malignancies (2). Overexpression of MET or hepatocyte growth factor (HGF), MET-amplification, mutation or rearrangement, and changes in ligand-induced autocrine or paracrine signaling lead to cancer progression by promoting tumor growth, invasion, metastasis, and survival (3, 4). Therefore, molecules that inhibit the MET/HGF signaling pathway are of clinical interest for the treatment of cancers that are driven by aberrant MET expression, including small cell lung cancer and non–small cell lung cancer (NSCLC; ref. 4).

Savolitinib, an oral, potent, and highly selective MET-tyrosine kinase inhibitor (MET-TKI), was granted conditional approval in China for the treatment of patients with NSCLC with MET exon 14 skipping alterations who had disease progression following prior systemic therapy, or are unable to receive chemotherapy (5). It is currently in clinical development as a monotherapy for papillary renal cell carcinoma (PRCC) and in combination therapy for NSCLC (6–13). Savolitinib potently inhibits the tyrosine kinase activity of recombinant MET in vitro, with an IC₅₀ value of 4 nmol/L, and has more than 200-fold selectivity versus 267 other human kinases (7). In vitro, savolitinib inhibits MET phosphorylation, which correlates with inhibition of MAPK and PI3K/AKT signaling pathways, and MYC downregulation. In vivo, savolitinib inhibits these pathways and significantly decreases growth of MET-dependent patient-derived xenografts (PDX; ref. 4).

Single-agent savolitinib has demonstrated antitumor activity in several preclinical studies of MET-dysregulated malignancies, including NSCLC cell lines and PDX models (4, 7, 8). In a phase III study, single-agent savolitinib showed antitumor activity and acceptable tolerability in patients with PRCC with MET aberrations (10). Despite initial efficacy, resistance to savolitinib has been observed in MET-amplified NSCLC cell lines via reactivation of kinase signaling, which has been associated with overexpression of MYC, constitutive activation of the PI3K/AKT/mTOR pathway, and a context-specific reliance on EGFR signaling (4). Approximately a third of NSCLC tumors contain mutations in the EGFR tyrosine kinase domain (14). Overexpression of EGFR promotes sustained activation of signaling pathways and cell proliferation (15).

Osimertinib is a third-generation, irreversible, oral EGFR-TKI that potently and selectively inhibits both EGFR mutation–positive (EGFRm) and EGFR T790M, and has demonstrated efficacy in NSCLC with central nervous system metastases (16–20). In the FLAURA trial (NCT02296125), first-line osimertinib significantly lengthened progression-free survival and overall survival versus comparator EGFR-TKI therapies (gefitinib/erlotinib; refs. 20, 21). Furthermore, adjuvant osimertinib showed statistically significant improvement in disease-free survival versus placebo in patients with stage IB–IIIA resected EGFRm NSCLC in the ongoing ADAURA trial [NCT02511106, overall hazard ratio for disease recurrence or death: 0.20; 99.12% confidence interval (CI), 0.14–0.30; P < 0.001; ref. 22]. Osimertinib is the recommended first-line treatment for patients with EGFRm NSCLC (23, 24) and has recently been approved in many countries globally for use as an adjuvant treatment in patients with resectable EGFRm NSCLC (22).

Many patients with advanced NSCLC treated with a first-line EGFR-TKI eventually develop treatment resistance (25). Up to 15% of patients with acquired resistance to osimertinib, and more than 5% of patients who have progressed on first- or second-generation EGFR-TKIs, have MET-driven resistance (25–30). In vitro, in vivo, and preliminary clinical studies have shown that MET inhibition may restore sensitivity to treatment with EGFR-TKIs in cells with MET-amplification when used in combination with an EGFR-TKI (31–33). Savolitinib has demonstrated antitumor effects in combination with gefitinib in a phase Ib study (NCT02511106) in patients with EGFRm, MET-amplified NSCLC, and in combination with osimertinib in the phase Ib TATTON study (NCT02143466) in patients with MET-amplified, EGFRm, advanced NSCLC, who had disease progression on a previous EGFR-TKI (11–13). A recent study of savolitinib in PDX and cell line–derived xenograft models suggests that high and durable levels of MET inhibition are needed for optimal monotherapy in preclinical models (34).

This study analyzed the inhibition of phosphorylated MET (pMET) and phosphorylated EGFR (pEGFR) and antitumor activity in an EGFRm, MET-amplified, LG1208 NSCLC PDX mouse model dosed with osimertinib in combination with several schedules of savolitinib. The LG1208 NSCLC PDX model is derived from a patient on prior erlotinib therapy and harbors the L858R EGFR mutation (sensitive to osimertinib) with MET amplification. We aimed to determine whether savolitinib could overcome resistance to osimertinib and explored the potential doses required for combination benefit.

Materials and Methods

Objectives

The objectives of this study were: (i) to explore the combination effect on biomarkers of target engagement for savolitinib (pMET) and osimertinib (pEGFR); (ii) to demonstrate combination activity of savolitinib+osimertinib at clinically relevant exposures; (iii) to build a pharmacokinetic/pharmacodynamic (PK/PD) model relating the exposure of savolitinib+osimertinib to biomarker effects to quantify the combination effect and relate the biomarker effects to tumor growth inhibition (TGI). Savolitinib was manufactured by SynTheAll Pharmaceuticals (Changzhou, China) and distributed by AstraZeneca (Södertälje, Sweden; ref. 6); osimertinib was manufactured by AstraZeneca (Södertälje, Sweden; ref. 16).

Animals

Animal experimentation was carried out by In Vivo Services at The Jackson Laboratory Sacramento facility, an Office of Laboratory Animal Welfare (OLAW)-assured and Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC)-accredited organization (LG1208 PDX studies) or AstraZeneca's AAALAC accredited facility in Waltham, MA (NCI-H820 study). All studies were performed according to Institutional Animal Care and Use Committee (IACUC)-approved protocols and in compliance with the Guide for the Care and Use of Laboratory Animals (National Research Council, 2011). Details of the mice used and their welfare conditions are described in the Supplementary Materials.

Preliminary analysis to confirm the utility of the EGFRm, MET-amplified, LG1208 NSCLC PDX mouse model

Initial experiments were conducted in tumor-bearing mice implanted with LG1208 and NCI-H820 tumors. Both are models of NSCLC tumors and are MET-amplified, with the LG1208 model harboring the L858R mutation and the NCI-H820 model harboring the T790M mutation. These experiments were used to: (i) confirm how sensitive these models were, following treatment for 15 days with savolitinib (12.5 mg/kg once daily) and osimertinib (25 mg/kg once daily) alone, or in combination in the LG1208 model, and treatment for 21 days with savolitinib (25 mg/kg once daily) and osimertinib (25 mg/kg once daily) alone, or in combination in the NCI-H820 model; (ii) aid the study design of a more extensive experiment. Learnings from the initial experiment were used to set up the main efficacy and PK/PD study (details below). It was found that the LG1208 model was easier to work with, showing less variability in growth kinetics across animals compared with the NCI-H820 model, and was therefore prioritized to go forward in the main study.

Xenograft efficacy study

Mice were acclimated for 2 days and subsequently implanted with tumor fragments from the TM00784/LG1208 PDX model, which was derived from a 42-year-old female with prior erlotinib treatment. Bodyweight and clinical observations were recorded once to twice weekly, and digital caliper measurements were initiated to determine tumor volume twice weekly when tumors became palpable. Mice were randomized into 8 dose groups (n = 10 per group; Supplementary Table S1) once tumor volumes reached ∼60 to 120 mm³ (Study Day -1) and dosing started on Study Day 0. Bodyweight, clinical observations, and digital caliper measurements were recorded twice weekly post–dose-initiation at time of takedown. Animals that reached a body condition score of ≤2, a bodyweight-loss of ≥20%, a tumor volume >2,000 mm³, or had ulcerated tumors, were euthanized before study end. On Study Days 20, 21, 22, and/or 24 (1, 24, 48, or 96 hours post-final dose on Study Day 20), animals from each group were euthanized by CO₂ asphyxiation and tissues collected. Tumors were collected and separated into two fragments and flash-frozen. Whole blood was collected via cardiocentesis and processed to plasma. Tumor volume calculations are described in the Supplementary Materials.

PKPD study

Additional satellite groups were used to explore the time-course of exposure and PD effects following single and repeat dosing over 3 days. Details of the doses and time points sampled are shown in (Supplementary Table S2). Sample collection followed the same process as the efficacy study.

Formulation and dosing

Savolitinib was formulated in acidic carboxymethyl cellulose 0.5% (pH = 2.1) in sterile deionized water; various doses of savolitinib were chosen (0.02 mg/kg to 15 mg/kg). Osimertinib was formulated in 0.5% hydroxypropyl methylcellulose in deionized water; 10 mg/kg osimertinib fixed dose was selected. The pan CYP450 inhibitor, 1-aminobenzotriazole (ABT; 0.5% methylcellulose in sterile deionized water, 100 mg/kg intraperitoneal injection) was administered 2 hours before savolitinib and osimertinib to increase their half-lives in mice to better match the clinical half-life. The administration of 0 to 15 mg/kg savolitinib and ABT, matches 0 to 600 mg clinically; the administration of 10 mg/kg osimertinib and ABT matches the exposure of the 80-mg once-daily clinical dose.

A PK study was run to confirm the pharmacokinetics in the NOD/SCID gamma (NSG™) mouse strain. The study consisted of three groups (n = 3): a single savolitinib dose (15 mg/kg); a single osimertinib dose (10 mg/kg); a single dose of savolitinib (15 mg/kg) plus osimertinib (10 mg/kg). Samples were taken at various time points after dosing to follow the time-course of drug exposure.

As noted above, at the end of the efficacy study, following 20 days oral dosing of savolitinib, osimertinib, and ABT once daily, PK samples were taken at various time points after the final dose to follow the time-course of drug exposure in addition to pMET and pEGFR change (Supplementary Table S1). Additional satellite groups were used to explore the time-course of exposure and PD effects following single and repeat dosing over three days (Supplementary Table S2).

PK/PD models

Mouse population PK models for savolitinib and osimertinib previously developed were used as a starting point in these analyses (34, 35). A direct response maximum effect (E_max) model linking savolitinib plasma concentration to inhibition of pMET and TGI had been evaluated previously (34). Similarly, an indirect response model linking osimertinib plasma concentration to the reversible and irreversible inhibition of pEGFR and TGI was available (35). Here, previously reported work was used to build the modeling framework for the combination (34, 35). Supplementary Figure S1 shows a schematic of the mathematical model used to link savolitinib plasma concentration to MET phosphorylation, osimertinib plasma concentration to EGFR phosphorylation and the changes in these PD biomarkers that drives antitumor activity.

Human population PK models previously reported for savolitinib and osimertinib were used for the translational modeling (36, 37).

PK and PD sample analysis

PK and PD sample analysis was performed as described previously (38). Briefly, plasma from study animals was analyzed for savolitinib and osimertinib using a protein precipitation extraction procedure, followed by liquid chromatography with tandem mass spectrometry.

Tumor preparation, Western blot analysis, and quantification of change in pEGFR and pMET for the PD analysis are described in the Supplementary Materials.

Statistical analysis

Statistical methods were reported previously (34).

Data availability statement

Data underlying the findings described in this manuscript may be obtained in accordance with AstraZeneca's data sharing policy described at https://astrazenecagrouptrials.pharmacm.com/ST/Submission/Disclosure. Requests for specific data described in the manuscript can be submitted through https://vivli.org/members/enquiries-about-studies-not-listed-on-the-vivli-platform/ or to the corresponding author, Rhys DO Jones at Owen.Jones@astrazeneca.com.

Results

PK of savolitinib and osimertinib in mice and modifications to existing PK models

For savolitinib with ABT, the PK in the NSG mice showed a shorter half-life (2.5 hours) versus previous studies using nude mice (3.5 hours; ref. 34).

The existing mouse population PK model for savolitinib (34) was simplified from a two-compartment nonlinear model to a one-compartment linear clearance model (Supplementary Fig. S1) with clearance set to 0.43 (95% CI, 0.26–0.59) L/h/kg and a distribution volume of 1.44 (95% CI, 0.88–2.0) L/kg. This change was driven by the data i.e., nonlinearity was not observed to a significant extent, which was expected as doses did not exceed 15 mg/kg, whereas in the previous study, doses greater than 15 mg/kg were used, and nonlinearity became pronounced above 30 mg/kg in mouse (34). Model fits to the observed data from this study are shown in Supplementary Fig. S2A.

Group to group and inter-study variability was observed for the plasma exposure of osimertinib, which was accounted for by varying the parameters between experiments: relative bioavailability of dose absorbed was 43% to 100% depending on the experiment. In addition, the osimertinib clearance with ABT ranged from 0.5 to 1.3 L/h/kg. The largest variability was seen in rate of absorption (K_a) where the values ranged from 0.44 to unidentifiable (fixed to a high value). This variability was larger than seen previously (35). Consistent with previous work, the addition of ABT did not change the fraction of osimertinib metabolized to its active metabolite (AZ5104; ref. 35). Final model parameters are shown in Supplementary Table S3, and simulations versus observed data are shown in Supplementary Fig. S2B.

A comparison of the simulated exposure profiles in mice (0.02–15 mg/kg) versus model simulations in humans (10–600 mg), showed good overlap, confirming the doses used in mice were clinically relevant (Supplementary Fig. S2C).

Because the active metabolite AZ5104 potency and observed exposure contributed meaningfully to the overall pharmacology, the total concentration of osimertinib plus AZ5104 was used as the basis to compare versus the observed clinical exposures. Supplementary Figure S2D shows that the mouse C_max following 10 mg/kg was less than two-fold lower than that seen in patients following 80-mg doses of osimertinib, whilst the half-life was longer in patients than in mice.

No modifications to the PK models for savolitinib+osimertinib were required to describe the PK data when the two agents were combined.

Savolitinib showed dose-related combination antitumor activity with osimertinib in the EGFRm, MET-amplified, LG1208 NSCLC PDX model

To confirm the suitability of both the EGFRm, MET-amplified T790M NSCLC NCI-H820 and LG1208 NSCLC PDX models in our studies, we conducted initial experiments to determine response to savolitinib, osimertinib, and combination treatment. Treatment of NCI-H820 tumor-bearing mice with osimertinib (25 mg/kg, once daily for ∼21 days) showed no significant response (∼34% TGI at Day 26, P > 0.05 vs. vehicle), confirming osimertinib resistance in this MET-amplified model. However, significant antitumor activity was observed with daily savolitinib at 25 mg/kg (∼9% tumor regression, P < 0.05 vs. vehicle). Combination treatment with savolitinib+osimertinib demonstrated significantly greater benefit over single-agent treatment (∼94% tumor regression, P < 0.05 vs. vehicle, or either of the single agents; Fig. 1A).

Figure 1.

Antitumor activity of savolitinib and osimertinib as single agents and in combination in the EGFRm, MET-amplified, T790M NSCLC NCI-H820 and LG1208 NSCLC PDX models. A, Tumor inhibition time curves with single-agent osimertinib (25 mg/kg) and savolitinib (25 mg/kg), and savolitinib plus osimertinib (treatment once daily for ∼21 days) in the NCI-H820 model. B, Tumor inhibition time curves with single-agent osimertinib (25 mg/kg) and savolitinib (12.5 mg/kg), and savolitinib plus osimertinib in the absence of ABT (treatment once daily for 15 days) in the LG1208 model. C, Tumor inhibition time curves following repeat dosing of single-agent savolitinib, osimertinib, and various doses of savolitinib combined with a fixed dose of osimertinib in the presence of ABT in the LG1208 model. Abbreviations: ABT, 1-aminobenzotriazole; EGFRm, epidermal growth factor receptor mutation positive; NSCLC, non–small cell lung cancer; PDX, patient-derived xenograft; QD, once daily. *Not statistically significant to vehicle (P > 0.05); **Not statistically different to osimertinib alone (P > 0.05).

Treatment of LG1208 tumor-bearing mice with a commonly used and clinically relevant dose and schedule of osimertinib (25 mg/kg, once daily for 15 days) resulted in no significant antitumor activity (6% TGI, P > 0.05 vs. vehicle), confirming osimertinib resistance. Interestingly, daily treatment with savolitinib at 12.5 mg/kg also showed no significant antitumor response (18% TGI, P > 0.05 vs. vehicle), despite the presence of MET-amplification. In contrast, savolitinib+osimertinib resulted in a strong antitumor activity reaching ∼90% TGI (P < 0.05 vs. vehicle or either single agent group; Fig. 1B); for this experiment, ABT was not co-administered with savolitinib or osimertinib, and no plasma samples were taken for PK analysis.

Despite both EGFRm, MET-amplified NSCLC models demonstrating suitability, the LG1208 model was selected for further investigation as it was easier to work with and showed less variability in growth kinetics across animals. In the main efficacy experiment, savolitinib 15 mg/kg once daily in the presence of ABT showed significant antitumor activity, reaching ∼84% TGI, which was significantly stronger than that observed in the absence of ABT (Supplementary Table S1). Osimertinib 10 mg/kg once daily in the presence of ABT showed no significant antitumor activity (34% TGI, P > 0.05 vs. vehicle), confirming initial observations and resistance of this model to EGFR-only inhibition (Supplementary Table S1). Savolitinib+osimertinib in the presence of ABT showed significant dose-related antitumor activity reaching 84% tumor regression, 46% tumor regression, 96% TGI, and 81% TGI when savolitinib was dosed at 15, 2.5, 1, and 0.3 mg/kg once daily, respectively. Lowering the savolitinib dose to 0.02 mg/kg once daily no longer showed benefit in combination with osimertinib (TGI 50%, P > 0.05 vs. osimertinib alone) (Supplementary Table S1; Fig. 1C).

Inhibition of pMET and pEGFR by savolitinib and osimertinib in the EGFRm, MET-amplified, LG1208 NSCLC PDX model

The Western blot analyses for pMET, total MET, pEGFR, and total EGFR expression are shown in Supplementary Fig. S3. PD analysis of tumors after single-agent treatment with savolitinib or osimertinib showed a clear reduction in the phosphorylation of their respective receptors (Supplementary Fig. S3A). Following single and repeat dosing of savolitinib 1 or 15 mg/kg alone, pMET was inhibited in an exposure-dependent manner, with near complete inhibition observed at higher concentrations (Fig. 2) and return to baseline upon clearance of savolitinib. When dosed alone, savolitinib did not inhibit pEGFR.

Figure 2.

Model simulations compared with observed levels of pMET: after a single dose (A), after 3 daily doses (B) and at the end of efficacy study (C); and pEGFR: after a single dose (D), after three daily doses (E); and at end of efficacy study (F). Abbreviations: pEGFR, phosphorylated epidermal growth factor receptor; pMET, phosphorylated MET.

With osimertinib 10 mg/kg alone, pEGFR was inhibited in an exposure-dependent manner (Fig. 2), with a maximum inhibition of ∼50%. There was no change in PD response following 3 or 21 days of repeat dosing, versus a single dose. When administered alone, osimertinib did not inhibit pMET.

The addition of osimertinib to savolitinib did not appear to change the extent of pMET inhibition mediated by savolitinib. In contrast, the addition of savolitinib to osimertinib appeared to increase the extent and duration of pEGFR inhibition in a dose-dependent manner (Supplementary Fig. S3B), to a maximum of 90% with savolitinib 15 mg/kg once daily. This was most apparent visually when comparing savolitinib from 0.02 mg/kg to 15 mg/kg, where the level of maximum inhibition increased from ∼50% to ≥90% (Fig. 2F).

PK/PD efficacy modeling

PK/PD modeling was conducted to link exposure of savolitinib and osimertinib to inhibition of pMET and pEGFR, respectively.

In the case of savolitinib inhibition of pMET, initial parameter estimates were set to the values from the previous analysis (34). After reestimating E_max and the plasma concentration at which half-maximal effect was observed (EC₅₀) on the basis of these data, it was found that E_max increased from 97% to 98.5% and EC₅₀ increased from 1.1 nmol/L to 1.3 nmol/L.

Despite careful study design, several samples were taken with a quantifiable level of pMET, but the concentration of savolitinib was below the limit of quantification (LOQ). To use the samples with a PK below the LOQ, the mouse PK model was used to simulate the expected plasma concentration at that time point and was plotted onto the savolitinib versus pMET inhibition curve (Supplementary Fig. S4).

For modeling osimertinib inhibition of pEGFR, the model previously reported included a depletable pEGFR pool to account for a longer time returning to baseline for high and/or prolonged pEGFR suppression (35). As this was not observed with the current data, the model was simplified to a standard turnover model. Model parameters were estimated on the basis of osimertinib dosing and are shown in Table 1.

Table 1.

Final model parameters used in the mouse PK/PD, and efficacy models for savolitinib, osimertinib, and the combination of both.

Parameter	Unit	Estimate	95% CI
PK/PD models
pEGFR baseline	—	24,300	23,100–25,500
pEGFR EC50base	μmol/L	0.17	0.106–0.233
pEGFR turnover rate constant (Kout)	/h	0.0152	0.0136–0.0167
pEGFR Emax (proportional increase Kout)	—	9.06	8.50–9.62
pEGFR combination pMET effect slope on EC50	—	−0.0317	−0.0391 to −0.0243
pMET baseline	—	17,200	16,600–17,800
pMET EC50	nmol/L	1.3	1.2–1.4
pMET Emax	—	−0.98	−0.984 to −0.983
Tumor growth model
Baseline tumor size	mm3	90.2	82.3–98.1
Exponential tumor growth rate constant	/h	0.0254	0.0183–0.0324
Linear tumor growth rate	mm3/h	4.75	4.11–5.39
pEGFR effect on tumor growth (slope)	—	0.0837	0.0492–0.118
pMET effect on tumor growth (slope)	—	0.0179	0.0165–0.0193
Variability baseline (log-normal)	SD	0.624	0.447–0.801
Variability exponential growth (log-normal)	SD	0.201	0.0613–0.341

Abbreviations: EC₅₀, half maximal effective concentration; EC_50base, the EC₅₀ for inhibition of pEGFR by osimertinib, when no savolitinib is present; E_max, maximum effect; IC₅₀, half maximal inhibitory concentration; pEGFR, phosphorylated epidermal growth factor receptor; PK/PD, pharmacokinetic/pharmacodynamic; pMET, phosphorylated MET.

Savolitinib had an effect on pEGFR changes, but only when osimertinib was also present. To model the changes in pEGFR driven by osimertinib alone and in combination with savolitinib, the PK/PD model structure was modified to include an interaction term for the combination effect whereby pEGFR inhibition is a function of the extent of pMET inhibition. Greater pMET inhibition would therefore result in a lower EC₅₀ value. The overall combination effect resulted in the maximum inhibition of pEGFR increasing from ∼50% with no savolitinib to ∼90% with savolitinib 15 mg/kg, which equates to near complete inhibition of pMET over the dosing range.

Final model parameter values are shown in Table 1, along with model fits to the time course of changes in pMET and pEGFR in Fig. 2.

Finally, the model was used to link the pEGFR and pMET inhibitory effects to the antitumor activity. The model structure is shown in Supplementary Fig. S1 and is consistent with previous savolitinib models (34), except that TGI was dependent on both pMET and pEGFR levels. As shown in Fig. 3A, the model simulations compare closely with the observed data and describe the dose-dependent level of TGI.

Figure 3.

PK/PD modeling data showing the link between biomarker effects and antitumor activity. A, Comparison of tumor inhibition between observed data (green diamonds) for savolitinib 0 to 15 mg/kg plus osimertinib 10 mg/kg from the EGFRm, MET-amplified, LG1208 NSCLC PDX mouse model and model simulations (solid black line). B, Model simulations for the relationship between the degree of pMET inhibition and effect on the EC₅₀ for the pEGFR inhibition by osimertinib. Abbreviations: BID, twice daily; pEGFR, phosphorylated epidermal growth factor receptor; pMET, phosphorylated MET; QD, once daily.

Using the combination PK/PD efficacy model, the interaction between the level of pMET suppression and pEGFR EC₅₀ was determined. Fig. 3B shows that there is minimal effect on pEGFR EC₅₀ below 80% pMET suppression; ≥80% pMET suppression showed a significant effect on pEGFR EC₅₀, with maximal effect at ≥95% suppression.

Translation to patients

Using human PK models for savolitinib and osimertinib, simulations were run for a fixed dose of 80 mg osimertinib, combined with savolitinib regimens of 0 to 600 mg once daily and 0 to 300 mg twice daily. Supplementary Figure S5 shows the time-course for simulated pMET and pEGFR.

From these simulations, a structured analysis is provided in Fig. 4; Fig. 4A shows dose plotted against the average degree (±90% prediction intervals) of pMET and pEGFR inhibition over 24 hours after the first dose. On the basis of the PK of the two drugs, the level of pMET inhibition over the dosing interval was ≥95% at all doses, whilst the level of pEGFR inhibition increased from ∼65% without savolitinib to 85% and 90% when administered with savolitinib 50 mg and 300 mg, respectively.

Figure 4.

Model simulations for the translation to human, exploring the effects on pEGFR and pMET following combination dosing of savolitinib (0 to 600 mg) plus osimertinib (80 mg; N = 10). A, Relationship between dose and the degree of inhibition of pEGFR or pMET at 24 hours after dosing. B, Relationship between dose and the time above 80% inhibition of pEGFR or 95% pMET over a 24-hour interval. C, Relationship between dose and the probability of a single patient achieving adequate exposure to deliver >80% pEGFR and >95% pMET continuously over a 24-hour dosing interval. Abbreviations: BID, twice daily; pEGFR, phosphorylated epidermal growth factor receptor; pMET, phosphorylated MET; QD, once daily.

Using the model to simulate pMET and pEGFR changes following repeated daily dosing of savolitinib plus osimertinib showed that over the dosing interval, there were periods where the plasma exposure drives PD suppression above and below 95% for pMET and 80% for pEGFR; to further explore this, Fig. 4B plots dose against the percentage time above these thresholds. Following once-daily dosing, the model simulated that for a savolitinib 50-mg dose, the threshold was surpassed for 60% of the time for pMET and ∼84% of the time for pEGFR; at savolitinib 300 mg, this increased to >99% and 95%, respectively. Following twice-daily dosing of ≥50-mg savolitinib, the threshold was surpassed for both pMET and pEGFR ≥ 95% of the time. However, given the population PK variability observed, the model predicted that a proportion of patients may receive insufficient exposure to reach these thresholds, as indicated by the 90% prediction interval error bars. Therefore, Fig. 4C plots dose against the probability of a single patient achieving the criteria of >95% pMET inhibition and >80% pEGFR inhibition over the dosing interval. The model indicated that following savolitinib 50 mg once daily, the probability was only 25% for pMET and 35% for pEGFR. This increased up to a probability of 80% for both pMET and pEGFR with savolitinib 600 mg once daily. Following savolitinib 50 mg twice daily, ∼75% and 85% of patients achieved the criteria for pMET and pEGFR, respectively; this increased to ∼95% and 100%, respectively, with savolitinib 300 mg twice daily.

Discussion

This study aimed to analyze the inhibition of pMET and pEGFR, and antitumor activity in an EGFRm, MET-amplified NSCLC PDX mouse model dosed with savolitinib plus osimertinib.

Drug combinations present a challenge in selecting a dose and schedule that delivers optimal efficacy whilst retaining adequate tolerability, and fully exploring options in the clinic can be difficult to achieve in an efficient and effective way (39). Preclinical insights can, however, be used to narrow down the options to test (40). Preclinical PK/PD modeling of combinations has often focused on estimating the correlation between drug exposure and TGI (41, 42), but to our knowledge, there are fewer studies, such as the study presented here, linking plasma exposure to biomarker changes that drive TGI (43). Single agent PK/PD models, including that for savolitinib and osimertinib, increasingly incorporate PD biomarker data to demonstrate and quantitate the pharmacologic mechanism, and determine the degree and duration of target modulation required for efficacy in a mouse model, which is used as the basis for translational modeling to the clinic (40, 44). In addition to the above, for combinations, delivering a similar approach is desirable to provide a more mechanistic rationale for the combination benefit and to quantitate the combination mechanism. The data presented here for the PDX model enables the combination effects to be put into context against the PD effects, with translation to the clinic to determine the exposure requirements of both drugs. The model was configured such that the mouse PK models were substituted with the human population PK models for savolitinib and osimertinib and modified to adjust for potency values for pMET and pEGFR inhibition, reflecting the differences in plasma protein binding between mouse and human.

An initial small efficacy study confirmed the utility of the PDX model as only significant antitumor activity was observed for the savolitinib-osimertinib combination group and no activity was observed with osimertinib only. This initial study was also used alongside the existing PK/PD understanding for savolitinib (34) and osimertinib (35) to optimize the design of a more extensive efficacy study exploring multiple dose levels and PK/PD studies exploring changes in the PD over time. Taking this approach maximized the probability of a successful study and optimal use of animals (40, 45).

Savolitinib and osimertinib doses were chosen for use in the mouse model that would have an exposure profile consistent with that observed in patients (46). As the half-life of both drugs is known to be shorter in mice than in humans and the use of ABT has been previously shown to reduce the clearance of both drugs and produce plasma concentration-time profiles more consistent with that seen in the clinic, ABT was co-dosed with both drugs (34, 35).

All study groups were extensively sampled for PK to maximize the data available for the PK/PD modeling. It was notable that in the NSG mouse strain, the PK of savolitinib differed from that seen previously in studies using other mouse strains, including the nude mouse, requiring the existing mouse PK model to be updated (34). Given that there were no differences in the PK of savolitinib or osimertinib when given alone or in combination, it is unlikely that a drug–drug PK interaction would explain these differences versus previous work. The PK interpretation for osimertinib was complicated by significant variability across dose groups but this was accounted for in the PK model by varying the parameters between studies. The cause of this group-to-group variability is unclear, but given the variation in absorption rate [0.44/h–unidentifiable (fixed to a high value)], clearance (0.5–1.3 L/h/kg), and relative bioavailability (43%–100%), it was likely to be absorption related. Despite not quantifying the inter-study variability mechanistically in the PK model, it was captured empirically, and in the context of the model purpose as a driver to understand the exposure versus PD and efficacy relationships provides an adequate solution. These issues were overcome in our study by collecting comprehensive data, which demonstrates, as previously reported, the importance of obtaining adequate exposure data to ensure appropriate and accurate interpretation of pharmacologic results (45).

In the main efficacy study, osimertinib 10 mg/kg in mice showed no significant difference versus the vehicle group, whilst savolitinib 15 mg/kg in mice showed significant antitumor activity (TGI, 84%). This single agent activity of savolitinib is in contrast to that observed in the absence of ABT, where no significant antitumor activity was observed, and is in line with ABT reducing clearance and thereby increasing exposure, resulting in prolonged duration of MET pathway suppression, driving antitumor activity as reported previously (34). Savolitinib (0.3–15 mg/kg) in combination with osimertinib showed significant dose-dependent antitumor activity versus vehicle. Osimertinib plus savolitinib 0.3 mg/kg delivered approximately the same level of antitumor activity as savolitinib 15 mg/kg alone, confirming a strong benefit of combining osimertinib with savolitinib.

The changes in pEGFR and pMET were measured at the end of the efficacy study, and separate satellite groups were also used to explore PD changes after single and repeat dosing for 3 days. Due to the different PK properties and pharmacologic action of savolitinib and osimertinib, the doses and sampling points were carefully chosen to maximize the data obtained so that the exposure versus response relationship could be adequately captured and the model parameters estimated including EC₅₀ for pEGFR and pMET inhibition. Savolitinib drives the reversible inhibition of pMET and pMET returns to baseline rapidly after savolitinib clearance. On the other hand, osimertinib is a covalent inhibitor and the recovery of pEGFR is partly driven by the resynthesis of the protein; because osimertinib has a longer half-life than savolitinib, the effects on pEGFR are longer lasting than that on pMET. This necessitated sampling of some combination groups that were optimal for one drug but less optimal for the other. This was particularly evident for several samples that had measurable pMET, but the drug concentrations were around the LOQ or not measurable. To enable these samples to be included in the PK/PD modeling, the savolitinib PK model was used to simulate the plasma concentrations expected at the respective time points. Inclusion or not of these imputed concentrations into the PK/PD model had little effect on the final parameter estimates of EC₅₀ and E_max for pMET inhibition, providing support that the reported change in pMET in these samples was reasonable. This approach offers a valuable diagnostic step to integrate data, even when some information is unavailable.

Overall, it was concluded that the PK-pMET model was suitable for this dataset and closely matched previous findings (35). The maximum inhibition of pEGFR by osimertinib (55%), was lower than previously observed, but this is likely consistent with the PDX model being insensitive to EGFR inhibition (35). There was no evidence of a change in pMET or pEGFR inhibition over time, confirmed by measuring PD after a single dose, 3-day dosing, and at study end. When savolitinib and osimertinib were combined, the presence of osimertinib had no apparent effect on the degree or duration of pMET inhibition versus dosing savolitinib alone. In contrast, combining savolitinib+osimertinib resulted in an exposure-dependent increase in pEGFR inhibition from ∼55% without savolitinib to a maximum of >80% with savolitinib 15 mg/kg. Because savolitinib alone did not appear to have any obvious effect on pEGFR, it was concluded that the combination was necessary for this effect and cannot be simply explained by an off-target effect, or compensatory changes induced by savolitinib. This observation has been previously reported; in an in vitro study using an osimertinib-resistant lung cancer cell line, ErbB3 phosphorylation was minimally inhibited by osimertinib alone, but was fully inhibited when combined with a MET-inhibitor (47). ErbB3 is a key mediator of MET-dependent resistance to EGFR inhibitors, likely due to MET-amplification activating ErbB3/PI3K signaling (47). Both EGFR and MET can require dimerization for receptor phosphorylation and activation, and EGFR can form homodimers and heterodimers with its family members, ErbB2 and ErbB4, and more distant receptor tyrosine kinases including MET, insulin-like growth factor 1 receptor, and Axl (48). Studies have shown that cross-talk between MET and EGFR may mediate altered growth control in tumorigenesis (49) and response to MET inhibition correlates with EGFR-MET dimerization, which is influenced by EGFR genotype (50). Therefore, a potential explanation for the reduced EGFR phosphorylation with savolitinib plus osimertinib in our study, is savolitinib inhibition of MET-driven trans-phosphorylation of EGFR in EGFR/MET dimers. Indeed, no change in EGFR phosphorylation was seen with savolitinib only, suggesting that the EGFR/MET heterodimerization was a result of either osimertinib, or the osimertinib plus savolitinib combination.

In the PK/PD model, the combination effect on pEGFR was modelled as an interaction term on the EC₅₀ for pEGFR inhibition such that increasing levels of pMET inhibition resulted in a lower EC₅₀. The model was then used to quantitate how the single agent and combination effects on pEGFR and pMET translated to the antitumor activity observed in the PDX model. The model successfully showed that TGI profiles across doses tested can be explained by setting tumor growth to be dependent on pEGFR and pMET levels. Therefore, mechanistically, the combination effects on the PD biomarkers result in combination efficacy benefit. Successfully linking PD and efficacy quantitatively in this way also strengthens the validity of the model structure describing the combination effect on pEGFR.

One limitation of this work is the reliance on the use of a single PDX model to fully explore the PK/PD–efficacy combination effects. We did complete an efficacy study in the NCI-H820 model, a NSCLC model with the T790M EGFR mutation and MET amplification. The results in this model are consistent with the results presented here for the LG1208 model, showing that the model is not sensitive to osimertinib, but is sensitive to savolitinib, and even more sensitive to the combination. There are few appropriate models that are EGFR driven, but also resistant to osimertinib and MET-amplified. That said, the PK/PD results are consistent with previous single agent PK/PD insights and what is expected biologically for the LG1208 PDX model, and pharmacologic mechanism from the combination of the two agents. In addition, we did not extensively explore combination effects on downstream signaling pathways such as MAPK and PI3K–AKT–mTOR. This was down to the lack of robust data, particularly at the higher doses, where due to the small size of treated tumors, the available tissue was prioritized for measurement of pEGFR and pMET. However, data is available at the two lower doses of savolitinib (0.02 and 0.3 mg/kg) combined with osimertinib. The Western blots (Supplementary Fig. S3C) show increased suppression of both pERK and pAKT at 1 hour following 0.3 mg/kg, compared with the 0.02 mg/kg savolitinib dose, and this is consistent with the higher dose driving increased inhibition of tumor growth compared with the lower dose.

In summary, these results confirm that it is desirable to maximize the extent and duration of pEGFR and pMET inhibition for optimal efficacy with savolitinib+osimertinib; the model suggests that near maximal (>95%) inhibition of pMET is required to drive significant combination effects on pEGFR. For an average patient, the model predicts that for osimertinib 80 mg once daily plus a savolitinib dose 300 mg once daily or 50 mg twice daily, the levels of pMET inhibition >95% and pEGFR >80% are achieved continuously. However, at these doses, there are patients who will not achieve the level of exposure to reach this level of target modulation. Taking this variability into account, the model predicts that 80% and 100% of patients are likely to achieve the necessary drug coverage following savolitinib doses of 600 mg once daily or 300 mg twice daily, respectively, when combined with osimertinib. The study highlights the value of integrating both preclinical and clinical datasets, and considering the requirements at the individual patient level, rather than just the population level, to understand how to maximize the combination benefit. Translating these data to the clinic suggests that continuous dosing of savolitinib to maximize inhibition of pMET, when combined with osimertinib, is beneficial. Comparing the current clinical doses being tested (300 mg once daily/twice daily; 600 mg once daily), the twice-daily dose of 300 mg would be expected to deliver the best exposure profile needed to maximize the antitumor activity of the combination. These doses are being explored in the phase II SAVANNAH study (NCT03778229), which is investigating the combination of osimertinib plus savolitinib in patients with MET-amplified, EGFRm, locally advanced/metastatic NSCLC who had disease progression on osimertinib, and will provide further insights into the potential combination benefit to patients.

Supplementary Material

Supplementary Materials and Methods

Supplementary Data text

Supplementary Table S1

Supplementary Table S1

Supplementary Table S2

Supplementary Table S2

Supplementary Table S3

Supplementary Table S3

Supplementary Figure S1

Supplementary Figure S1

Supplementary Figure S2

Supplementary Figure S2

Supplementary Figure S3

Supplementary Figure S3

Supplementary Figure S4

Supplementary Figure S4

Supplementary Figure S5

Supplementary Figure S5

Authors' Disclosures

R.D.O. Jones, H. Tomkinson, and A. Schuller are employees of AstraZeneca and hold company stocks. K. Petersson is an employee of qPharmetra LLC and reports consultancy fees from AstraZeneca. A. Tabatabai and L. Bao are employees of AstraZeneca.

Authors' Contributions

R.D.O. Jones: Conceptualization, data curation, formal analysis, methodology, writing–original draft, writing–review and editing. K. Petersson: Formal analysis, visualization, writing–review and editing. A. Tabatabai: Data curation. L. Bao: Data curation. H. Tomkinson: Supervision, writing–review and editing. A.G. Schuller: Conceptualization, resources, data curation, supervision, writing–original draft, writing–reviewing and editing.