Preclinical Comparison of Osimertinib with Other EGFR-TKIs in EGFR-Mutant NSCLC Brain Metastases Models, and Early Evidence of Clinical Brain Metastases Activity

A number of EGFR tyrosine kinase inhibitors (TKI) are recommended for first-line treatment of patients with advanced non–small cell lung cancer (NSCLC) harboring an EGFR-TKI–sensitizing mutation (EGFRm; refs. 1, 2). However, more than 30% of patients with NSCLC experience disease progression during treatment with established EGFR-TKIs due to growth of synchronous or metachronous brain metastases (3, 4).

For successful treatment of brain metastases, a drug must first be able to cross the blood–brain barrier (BBB). BBB penetration is influenced by factors such as a drug's affinity for the ATP-binding cassette efflux transporters, permeability glycoprotein (P-gp), and breast cancer-resistance protein (BCRP), which are involved in the removal of toxins, drugs, and chemotherapies from the central nervous system (CNS; refs. 5–9). Chemotherapy agents and large monoclonal antibodies are generally unable to cross the BBB (7, 10). The ability of a molecule to cross the BBB is affected by multiple factors, including molecular weight (11).

Active or untreated brain metastases are often exclusion criteria in trials of EGFR-TKIs; however, there are reports documenting efficacy of EGFR-TKIs in treatment of, and/or preventing development of, brain metastases in patients with EGFRm NSCLC (12–15). In a retrospective analysis of 155 patients, initial gefitinib or erlotinib treatment was associated with a lower cumulative risk of CNS progression (1%, 6%, and 21% at 6, 12, and 24 months) compared with chemotherapy (7%, 19%, and 32%, respectively; ref. 12). In addition, pulsatile administration of high-dose erlotinib can control CNS metastases from EGFR-mutant NSCLC (16).

Despite reports of tumor responses, the TKIs gefitinib, erlotinib, and afatinib are considered to have generally poor biopharmaceutical properties for penetrating the BBB, perhaps attributable to interactions with P-gp and BCRP (17–19). However, penetration may be increased in patients with more advanced brain metastases where BBB disruption has already occurred (20–22). In addition, there is a cumulative increase in brain metastases incidence in patients with EGFRm NSCLC over time (23). Although many patients die of systemic progression, rather than brain lesion progression, quality of life is significantly worsened, both directly and as a result of whole brain radiotherapy (WBRT), which degrades cognitive function (24). In addition, as systemic therapies improve for patients with EGFRm NSCLC, the brain may increasingly become a sanctuary site where the BBB may offer protection from pharmacological agents (22). Therefore, there exists a clinical need for EGFR-TKIs with improved BBB penetration, and it is important that new mutant-selective agents, such as osimertinib (AZD9291), an oral, potent, irreversible EGFR-TKI selective for sensitizing and T790M-resistance mutations (25, 26), and rociletinib (CO-1686; ref. 27), are explored in this context.

Osimertinib was recently approved by the FDA for treatment of patients with NSCLC harboring a T790M mutation, and whose disease has progressed following treatment with another EGFR-TKI (28). We examined the brain exposure and distribution of osimertinib and the active metabolites AZ5104 and AZ7550 in the preclinical setting. We compared brain distribution, pharmacokinetics (PK), and in vivo brain xenograft efficacy of osimertinib with other EGFR-TKIs and simulated potential clinical efficacy based on these data. Furthermore, brain penetration of radiolabeled osimertinib and other EGFR-TKIs was examined in a nonhuman primate model. We also present early evidence of clinical efficacy of osimertinib against brain metastases as part of the ongoing AURA trial (NCT01802632).

Madin-Darby canine kidney (MDCK) epithelial cells were obtained from U.S. National Institutes of Health, Bethesda, MD, and Netherlands Cancer Institute [multidrug-resistance protein 1 (MDR1)-MDCK]; Absorption Systems and AstraZeneca (BCRP-MDCK); ATCC (Caco2). Cells were authenticated during each experiment by monitoring transepithelial electrical resistance, and with positive controls. H1975 cells were obtained from ATCC in 2004, and authenticated by short-tandem repeat analysis in November 2012. PC9 cells (exon 19 deletion) were obtained in November 2011 from Akiko Hiraide at Preclinical Sciences R&D, AstraZeneca, and tested and authenticated by short-tandem repeat analysis in May 2013.

Details on cell line maintenance and test compounds are presented as Supplementary Information.

P-gp (also known as MDR1) and BCRP substrate assessments of osimertinib, AZ7550, AZ5104, rociletinib, afatinib, and erlotinib were performed using transfected MDCK cells (MDR1-MDCK; BCRP-MDCK) and the Caco2 (colon carcinoma) cell line. Digoxin, cladribine, and Ko143 10 mmol/L, valspodar 1 mmol/L, and atorvastatin 0.075 mmol/L stock solutions in DMSO were prepared as positive controls or inhibitors.

Cell monolayers were grown onto collagen-coated, 12-well, polycarbonate membranes in Costar Transwell plates (1.13 cm² insert area, 0.4 μm pore size; Corning Life Sciences). Test and control compound substrate assessments were carried out in triplicate in each direction (apical [AP]-to-basolateral [BL], and BL-to-AP), co-dosed with 200 μmol/L Lucifer yellow (LY). Aliquots were taken from the test compound receiver compartment at preselected time points and replaced with an equal volume of fresh transport buffer. Positive control receiver samples were taken at 120 minutes. LC/MS-MS was used for analysis of the test compound, and LY concentration measured using a BMG microplate reader (excitation 485 nm; emission 540 nm). Donor sample aliquots were taken at selected time points without replacement.

Relative efflux ratios of compounds between MDR1-MDCK or BCRP-MDCK cells, compared with nontransfected MDCK cells, and co-dosing with 1 μmol/L valspodar or 10 μmol/L Ko143, were utilized to verify whether test compounds were P-gp and BCRP substrates, respectively.

For Caco2 drug transport assays, atenolol (low permeability marker), minoxidil (high permeability marker), and digoxin (control efflux marker) were prepared in DMSO to a test concentration of 10 μmol/L. To determine test and control compound transport rates in AP-to-BL and BL-to-AP directions, LY 100 μmol/L was run alongside all compounds. Donor samples were taken at 60 and 120 minutes, and receiver samples taken additionally at 30 and 60 minutes and replaced with an equal volume of fresh transport buffer.

Apparent permeability coefficients of test compounds across Caco2 monolayers was estimated using high-performance liquid chromatography (HPLC)/MS-MS; LY concentrations were measured using an Infinite 200 PRO microplate reader (485 nmol/L excitation; 530 nmol/L emission).

Tissue distribution of radioactivity in male partially pigmented (Lister-Hooded) rats was determined following a single oral dose of [¹⁴C]osimertinib (4 mg/kg, 7.4 MBq/kg) as a suspension in aqueous hydroxypropyl methylcellulose [HPMC; 0.5% (w/v)]. Rats were terminated under anesthesia by cold shock at 0.5, 1, 6, and 24 hours, and 2, 7, 21, and 60 days post-dose. Animals were immersed in a freezing mixture for ≥30 minutes and subjected to sagittal plane whole body sectioning (nominal thickness 30 μm). Concentration of radioactivity in tissues was determined using quantitative whole body autoradiography (QWBA) by Covance Laboratories Ltd. using a validated image analysis system.

In a separate experiment, the distribution of a single oral dose of [¹⁴C]gefitinib (5 mg/kg, 8.3 MBq/kg) in male Pievald Virol Glaxo pigmented rats was assessed. [¹⁴C]gefitinib was administered as a suspension in 0.5% HPMC in 0.1% polysorbate 80. Rats were terminated by CO₂ narcosis at 1, 2, 6, 24, 48, and 96 hours postdose and frozen and sectioned as above. Radioactivity was quantified using a Phosphor Imager SF (Molecular Dynamics).

Brain tissue from treatment-naïve nude mice was homogenized with PBS (1:3, w/w), and [³H]osimertinib spiked into the blank brain homogenate at a 5 μmol/L final concentration. Radioactive content of spiked brain homogenate was assessed by liquid scintillation counting (LSC) using Ultima Gold Scintillation cocktail. Samples were counted for 10 minutes, or until 10⁵ counts had accumulated; test compound concentrations were calculated using the radioactive content of each sample in conjunction with the specific activity of the material used. Spiked brain homogenate was dialyzed in triplicate against PBS for 4 hours at 37°C using Spectra/Por2 equilibrium dialysis membrane discs (molecular weight cut-off of 12–14 kDa; Spectrum Laboratories Inc.). Aliquots of brain homogenate and PBS were assessed by LSC; the free fraction of test compounds in brain homogenate was calculated and overall recovery of radioactivity from cells determined. The free fraction in undiluted brain was determined from these data.

The in vitro brain binding assays of unlabeled osimertinib was carried out on a HT-Dialysis plate (HTD 96 b, Cat. no. 1006). Blank brain homogenate (1:3 with Dulbecco's PBS pH 7.4) was spiked with 5 μmol/L test compound (in triplicate) and dialyzed (molecular weight cut-off 12–14 kDa) against 100 mmol/L PBS buffer (pH 7.4), slowly rotated at 37°C for 4 hours. Receiver and donor aliquots were taken following incubation, and donor samples further diluted. Paired samples were matrix-matched with buffer or blank brain homogenate, and precipitated with cold acetonitrile with internal standard. After centrifuging at 4000 rpm for 20 minutes, supernatant was diluted with 0.1% formic acid aqueous solution and analyzed by LC-MS/MS (API 4000; Applied Biosystems).

The in vitro brain binding assay of unlabeled gefitinib was carried out on a RED device system (Thermo Fisher Scientific) with a semipermeable membrane. Gefitinib 10 mmol/L was prepared in DMSO, and a 1 mmol/L working solution spiked in triplicate into blank brain homogenate (1:3 with PBS pH 7.4) to a 5 μmol/L final concentration; the final percentage volume of the organic solvent was 0.5%. Spiked brain homogenate was dialyzed against PBS pH 7.4 for 4 hours at 37°C with 5% CO₂ for 4 hours on a slowly rotating orbital shaker. Following incubation, a 50 μL receiver aliquot, and a 5 μL donor aliquot were taken; the donor sample was diluted with 45 μL of blank brain homogenate. Paired samples were matrix-matched with buffer or blank brain homogenate, mixed for 2 minutes, and precipitated with 150 μL cold acetonitrile containing internal standard. Samples were centrifuged at 4,000 × g for 20 minutes and supernatant diluted 1:2 with distilled water for LC/MS-MS (Waters Xevo TQ).

Osimertinib was administered orally to female CB17 SCID mice bearing H1975 tumor xenografts at Oncodesign Biotechnology. Three mice per dose group and time point received osimertinib 5 or 25 mg/kg (as suspension in 0.5% HPMC). Mice were terminated (isoflurane overdose) and blood, brain, and tumor samples collected at 0.5, 1, 2, 4, 6, and 24 hours postdose.

Osimertinib 25 mg/kg, gefitinib 6.25 mg/kg, rociletinib 100 mg/kg, or afatinib 7.5 mg/kg were dosed to naïve female nude mice, with animals terminated (rising CO₂ dose) and blood and brain samples collected at 1, 2, and 6 hours postdose.

Plasma was prepared from blood in lithium heparin chilled tubes immediately after collection. Test compounds were extracted by protein precipitation using acetonitrile, containing internal standard (AZ10024306) in a 4:1 solvent:sample ratio. Samples were centrifuged for 10 minutes at 3,000 rpm; 50 μL of supernatant was combined with 300 μL of water before HPLC/MS-MS.

Brain and tumor samples were cut into two portions prior to snap freezing in liquid nitrogen, stored at −80°C, and transferred to Oncology DMPK, AstraZeneca R&D. Following thawing, samples were transferred to FastPrep Lysing Matrix A tubes (MP Biomedicals LLC). Water was added (1:3 tissue:water w/v) and samples homogenized; homogenate was treated in the same way as plasma.

Exposure was expressed as area under the plasma or tissue–concentration time curve (AUC) from 0 to 6 or 0 to 24 hours according to the last concentration measured. Maximum observed concentration (C_max) and time to C_max (t_max) were direct observations of concentration versus time data. PK parameters were obtained by means of noncompartmental analysis using Phoenix32 version 6.4 (Pharsight Corporation).

A PET microdosing study was performed in four cynomolgus monkeys (Supplementary Table S2). Monkeys were anesthetized using ketamine, and anesthesia maintained by a mixture of O₂, air, and sevoflurane (2–8%). Anesthesia, heart rate, blood pressure, and body temperature were continuously monitored. Radiolabeled compounds were injected as a bolus into a sural vein. Injected radioactivity was 141 ± 25 MBq corresponding to <3 μg of radiolabeled osimertinib, AZ5104, gefitinib, and rociletinib. The distribution of radioactivity following intravenous microdosing was assessed using PET imaging more than 125 minutes with a Siemens molecular imaging high-resolution research tomograph system (29). Arterial blood samples were collected and analyzed for radioactivity in blood. Separate experiments were conducted to determine radioactivity distribution to the brain and abdomen. Radioactivity concentration in brain was calculated and normalized to arterial blood radioactivity concentration as described in the Supplementary Material.

Effects of chronic once daily (QD) dosing of osimertinib, gefitinib, and rociletinib were assessed in a mouse brain metastases model. The human NSCLC cell line PC9 (exon 19 deletion) was transfected with the GL4.50[luc2/CMV/Hygro] vector containing the luciferase gene (PC9_Luc) using lipofectamine LTX in order to monitor brain tumor growth by measuring bioluminescence signals in tumor cells (30, 31). Signal intensity was measured using the Bright-Glo Luciferase Assay System (Promega). The EGFR exon 19 deletion mutation was further confirmed in the PC9_Luc cell line.

For the brain metastases model, PC9_Luc cells were implanted into 6- to 8-week-old specific-pathogen-free immunodeficient nude mice purchased from Vital River by intra-internal carotid artery injection (32, 33). Bioluminescence signals were measured with an IVIS Xenogen imaging system weekly to monitor tumor growth. When signals reached the range of 10⁷ photons/second, mice were randomized to be treated orally with osimertinib 5 or 25 mg/kg QD, gefitinib 6.25 mg/kg QD, rociletinib 100 mg/kg QD, or vehicle for 8 weeks. Bioluminescence signals (antitumor efficacy) and body weight (tolerability) were measured weekly; tumor growth inhibition was assessed by comparing mean bioluminescence intensity change for control and treated groups.

A series of modeling studies was undertaken to predict active doses of osimertinib in humans by incorporating human PK data into a previously developed PK-pharmacodynamic (PKPD) PC9 subcutaneous xenograft efficacy model that linked phosphorylated EGFR (pEGFR) inhibition to reductions in tumor volume (34). The studies also accounted for reduced free concentrations in the brain due to higher levels of binding than observed in plasma.

The PD model represents pEGFR, relative to control, as a pool homogenously distributed across the cell and cells in the tumor. The reversible interaction (docking) of the molecule with the receptor is assumed to be more rapid than the covalent binding step and subsequent deactivation of the receptor. The system is simplified with the reversible interaction characterized by the binding affinity, and the irreversible interaction characterized by a binding constant for unbound parent and metabolite. Further information on the model is presented as supplementary material.

We present two clinical case studies from the dose escalation/expansion component of the ongoing phase I/II AURA study (ClinicalTrials.gov: NCT01802632; ref. 35). In these cases, presence of brain metastases was documented at study entry; patients underwent magnetic resonance imaging of the brain at baseline and every 6 weeks thereafter as part of the systemic evaluation of disease status according to RECIST version 1.1 until disease progression.

Two efflux transporters, P-gp and BCRP, prevent molecules from crossing the BBB, so it was important to determine the level of osimertinib substrate activity against these transporters. Assessments were carried out in transwell plates in AP-to-BL and BL-to-AP directions (Fig. 1). Relative efflux ratios of osimertinib or its metabolite AZ5104 between MDR1-MDCK and normal MDCK cells and between BCRP-MDCK and MDCK cells suggest that these agents are P-gp and BCRP transporter substrates. When dosed in the presence of transporter inhibitors valspodar or Ko143, efflux ratios were similar to that observed in nontransfected MDCK or BCRP-MDCK cells, respectively. Collectively, data confirm that osimertinib and AZ5104 are substrates of P-gp and BCRP. The apparent permeability of AZ7550 in the BL-to-AP direction of MDR1-MDCK cells in the presence of valspodar or BCRP-MDCK cells in the presence of Ko143 suggests that AZ7550 is also a substrate of P-gp and BCRP.

In MDCK-MDR1 cells, efflux ratios of test compounds dosed at 1 μmol/L were 13.4 for osimertinib, 5.38 for rociletinib, 4.62 for afatinib, and 4.63 for erlotinib. Efflux ratios in MDCK-BCRP cells were 5.4 for osimertinib, 54.6 for afatinib, and 6.39 for erlotinib.

Osimertinib demonstrated a passive permeability profile across Caco2 cell monolayers, with no concentration dependence when tested at 1, 10, and 50 μmol/L, and an efflux ratio <2.0. The high-binding propensity of osimertinib to cells and plasticware led to low recovery values in this in vitro study (data not shown). The present osimertinib data indicate an efflux ratio of 0.372 for osimertinib 50 μmol/L (Supplementary Table S1). In this cell line, the efflux ratios were 4.61 for rociletinib 10 μmol/L and 11.5 for afatinib 10 μmol/L.

To begin to directly explore the extent that osimertinib can achieve brain exposure following oral dosing (4 mg/kg), we utilized QWBA in rat with radiolabeled osimertinib. The tissue exposure pattern in male pigmented animals indicated that radioactivity associated with [¹⁴C]osimertinib was rapidly absorbed and distributed into the brain, suggesting that drug-related radioactivity may have crossed the BBB. Maximum concentrations of total radioactivity were recorded in the blood and various tissues, including the brain at 6 hours postdose; the concentration of total radioactivity in blood and brain was 0.552 and 1.02 nmol equivalents/g, respectively. The maximum brain:blood ratio was approximately 2.2, achieved by 60 minutes postdose (Supplementary Fig. S2). Distribution was maintained in the brain up to 21 days after a single dose, although the brain:blood ratio (0.2) had decreased, and was below the lower limit of quantification in blood and brain at 60 days.

In comparison, gefitinib achieved little brain exposure following oral dosing (5 mg/kg), with 0.36 nmol equivalents/g detected in brain and a brain:blood ratio of approximately 0.69, achieved by 6 hours postdose. Gefitinib levels in the brain were below the limit of quantification by 24 hours postdose.

Osimertinib was highly bound in mouse brain tissue in vitro. Dilution-corrected fractions of unbound osimertinib and [³H]osimertinib in brain homogenate were 0.0009 and 0.0012 (standard error for both = 0.0002), with no statistically significant difference between datasets (P = 0.3059). The fraction of unbound gefitinib was 0.0052 (SE = 0.0002).

Although protein binding and efflux activity are important parameters contributing to BBB penetrance, they may not predict overall brain distribution. The rat QWBA study indicated that osimertinib-related material could penetrate the brain, so we were interested to discover if this radioactivity could be attributed to osimertinib and/or its metabolites. As a consequence, we measured osimertinib brain distribution directly in vivo in mouse following oral dosing at 5 and 25 mg/kg. Osimertinib was highly distributed in brain, to a similar extent as in tumor, resulting in AUC tissue:plasma ratios of 1.7 to 2.8 (Table 1). Furthermore, osimertinib was more highly distributed to mouse brain than gefitinib, rociletinib, or afatinib (Table 2). The brain:plasma C_max ratio for osimertinib was 3.41. In comparison, ratios for gefitinib, rociletinib, and afatinib were only 0.21, <0.08, and <0.36, respectively, with brain concentrations of rociletinib and afatinib below the assay lower limit of quantification. The unbound brain-to-plasma partition ratio (Kp_uu,brain) was 0.39 and 0.02, for osimertinib and gefitinib, respectively, but could not be determined for rociletinib and afatinib.

Of interest, based on AUC from time 0 to time t (AUC_0–t), the osimertinib metabolites AZ5104 and AZ7550 were observed at 21% to 44% and 24% to 34% of osimertinib at 5 and 25 mg/kg, respectively. AZ5104 exposure in tumor was at somewhat lower levels than observed in plasma; tumor:plasma ratios ranged from 0.26 to 0.86 (Table 1). However, in contrast to parent, no AZ5104 concentrations above the assay limit of detection (0.08 μmol/L) were measurable in brain (Table 1). Similarly, AZ7550 exposure in tumor was at similar levels to that observed in plasma; tumor:plasma ratios ranged from 0.63 to 2.0 (Table 1). AZ7550 showed minimal brain distribution with a brain:plasma ratio of 0.1 determined at the 25 mg/kg dose; distribution in brain was below the assay lower limit of detection (0.02 μmol/L) at 5 mg/kg (Table 1).

To further investigate osimertinib brain exposure using a quantitative imaging of radiolabeled drug approach, we used PET microdosing in cynomolgus monkeys. Under microdosing conditions (total dose <3 μg), [¹¹C]osimertinib showed marked exposure in the cynomolgus monkey brain in contrast to its active metabolite [¹¹C]AZ5104, and other EGFR-TKIs (Fig. 2). Following intravenous administration of [¹¹C]osimertinib, distribution to the brain was fast, plateauing at 1.29% ± 0.42% (N = 3) of injected radioactivity within 10 minutes. In contrast, administration of [¹¹C]AZ5104 (0.17%; N = 2), [¹¹C]gefitinib (0.11%; N = 2), and [¹¹C]rociletinib (0.023%; N = 2) resulted in very low brain exposure, with no regional differences in brain radioactivity observed. Ratios of the area under the brain radioactivity concentration–time curve from 0 to 90 minutes to that for blood radioactivity were calculated for [¹¹C]osimertinib (2.62 ± 1.42), [¹¹C]AZ5104 (0.35), [¹¹C]gefitinib (0.28), and [¹¹C]rociletinib (0.025).

Assessment of whole body distribution up to 120 minutes postdose revealed extensive hepatobiliary excretion of [¹¹C]rociletinib; hepatobiliary excretion of [¹¹C]osimertinib and [¹¹C]gefitinib occurred to a lower degree at a slower rate (Fig. 2).

No adverse effects or significant changes in physiological or blood parameters related to administration of radioactive test compounds were observed.

As the PK and PET studies supported brain penetration of osimertinib, but not its metabolites, we explored how this translated into antitumor activity in a mouse PC9 (exon 19 deletion) xenograft brain metastases model. For this aggressively growing tumor model, there was only one control animal from a cohort of six still on study after day 50, as the tumor in this mouse grew slower than in other controls. Consequently, the control growth curve is observed to decrease after day 50 (Fig. 3A). A dose-dependent tumor regression was achieved with osimertinib (Fig. 3A), which correlated with overall survival (Fig. 3B). The dose of osimertinib 25 mg/kg QD, roughly equating to the 80 mg QD clinical dose of osimertinib in terms of exposure, was well tolerated and induced sustained tumor regression until study end at day 60, with a little weight loss at the initial time point and no subsequent decrease throughout the dosing period (Fig. 3C). Although the lower 5 mg/kg QD dose of osimertinib also caused tumor regression, it was more transient, only occurring in the first 3 weeks (Fig. 3A). In contrast, no tumor regression was achieved with rociletinib 100 mg/kg, approximately equivalent to a 500 mg twice daily human dose, and no survival benefit was observed (Fig. 3). A clinically relevant dose of gefitinib 6.25 mg/kg, approximating to the exposure of a 250 mg QD human clinical dose, also demonstrated only transient tumor regression, for up to 20 days (Supplementary Fig. S3).

Overall, preclinical data indicated that osimertinib could potentially achieve efficacy in mutant EGFRm brain metastases. We therefore wanted to translate this into a clinical context using a PKPD modeling approach. The plasma PKPD model for osimertinib was adjusted according to mouse BBB penetration and binding data. After adjusting for control growth, simulated efficacy suggested that the model adequately predicted efficacy (Fig. 4), validating the assumption of adjusting for free exposure in the brain metastases model. Simulation based on subcutaneous xenograft models and free brain exposure to osimertinib revealed that higher doses were needed to achieve the same percentage tumor growth inhibition of brain metastases as seen in the primary tumor (Fig. 4), potentially due to lower free exposure of osimertinib and AZ5104 in the brain than systemically. Tumor growth simulations using human exposure of osimertinib predicted that a human dose of at least osimertinib 80 mg QD could be sufficient to target EGFRm NSCLC brain metastases.

As clinical proof of principle to support the preclinical findings, we report two clinical case studies (Fig. 5.) demonstrating evidence of clinical activity of osimertinib in brain metastases observed in the AURA phase I/II study (NCT01802632) in patients with acquired resistance to current EGFR-TKIs (35).

Case study 1 is of a 62-year-old Asian female diagnosed with EGFRm (exon 19 deletion) advanced NSCLC. This patient had been previously treated with gemcitabine/cisplatin for four cycles [outcome: stable disease (SD)], gefitinib between June 2011 and October 2012 [partial response (PR)], and pemetrexed for 10 cycles between November 2012 and June 2013 (SD), with WBRT between December 2012 and January 2013. Biopsy in July 2013 identified T790M mutation. The patient started osimertinib 40 mg QD in a T790M positive expansion cohort of the AURA clinical study on August 7, 2013. PR was achieved from the 12-week Response Evaluation Criteria In Solid Tumors (RECIST) 1.1 scan in October 2013, with noncomplete response-nonprogressive disease (in effect, SD) reported in the nontarget lesions (including brain metastases). The patient was ongoing with systemic PR as of May 1, 2015, 632 days after starting osimertinib.

Case study 2 is of a 59-year-old Asian female diagnosed with EGFRm (L858R) advanced NSCLC. This patient was previously treated with erlotinib between October 2011 and October 2012 (outcome: PR), pemetrexed/cisplatin/carboplatin for five cycles between October 2012 and January 2013 (SD), erlotinib between January 2013 and March 2013 (nonevaluable [NE]), docetaxel for three cycles between April 2013 and June 2013 (SD), and gemcitabine for two cycles between June 2013 and July 2013 (NE). Biopsy in August 2013 identified a T790M mutation. The patient started osimertinib 80 mg QD in a T790M positive expansion cohort of the AURA clinical study on September 2, 2013, and PR was achieved from the 6-week RECIST 1.1 scan in October 2013, with noncomplete response-nonprogressive disease (in effect, SD) reported in the nontarget lesions (including brain metastases). PR remained for the 12- and 18-week systemic assessments. Disease progression was observed in the brain as nontarget lesions at the 24-week scan in February 2014, although extracranial target lesions remained in response. The patient discontinued osimertinib on March 10, 2014, after 189 days on study treatment, and commenced vinorelbine on March 10, 2014, and whole brain irradiation on March 17, 2014.

Identification of brain metastases in patients with NSCLC has increased over recent decades; contributing factors include improved survival as a result of more effective systemic therapies, and improved imaging quality and accessibility allowing detection of asymptomatic lesions (36). Although approximately one-third of patients with NSCLC progress during treatment by the development of brain metastases (3), there are few effective treatment options available.

Traditionally, WBRT has been regarded as the cornerstone of treatment; however, there are concerns regarding its long-term neurotoxicity profile (37). In recent years, brain metastases management has been refined, and now includes local therapies such as surgical resection for single brain lesions and stereotactic radiosurgery for oligometastatic lesions (22).

Incomplete BBB penetration is widely viewed as the reason for the high prevalence of brain metastases in patients with NSCLC who have achieved good systemic control with chemotherapy regimens (20). As currently available EGFR-TKIs have a limited ability to penetrate the BBB, there remains an unmet need for EGFR-TKIs with improved clinical efficacy against brain lesions. This is particularly important as patients with EGFRm advanced NSCLC are living longer, and managing long-term neurotoxicity related to WBRT is challenging. We therefore describe preclinical and clinical evidence supporting that osimertinib may be an EGFR-TKI with improved brain exposure for treatment of brain metastases in the EGFRm NSCLC setting.

Kp_uu,brain is well established as a good predictor of BBB permeability, with values greater than 0.3 indicative of good diffusion across the BBB (38). Osimertinib was shown to have a good Kp_uu,brain value (0.39) compared with other currently available TKIs and rociletinib, suggesting it has the potential to achieve good brain exposure. This was despite evidence that osimertinib is a substrate of P-gp and BCRP efflux transporters, which are involved in the removal of toxins, drugs, and chemotherapies from the CNS, ultimately leading to drug resistance in the brain, when measured in cell lines (MDCK-MDR1, MDCK-BCRP) overexpressing these transporters (6–9). In the Caco2 cell line, which expresses P-gp/BCRP at physiological levels, osimertinib was the only agent without an efflux ratio, suggesting that permeability of osimertinib is sufficient to overcome the efflux in this non-transfected human cell line. In contrast, the other EGFR-TKIs assessed were restricted by efflux. By analogy, this same phenomenon could be happening at the BBB, resulting in the superior brain penetration of osimertinib, compared with the other TKI agents; however, these data need to be confirmed before any firm conclusions can be drawn. Kp_uu data presented here consistently showed that osimertinib could achieve significant exposure in the brain. Moreover, preclinical data showed that osimertinib caused durable shrinkage in an in vivo EGFRm brain metastases model at clinically relevant doses, consistent with its efficacy in extracranial preclinical models (25).

Importantly, the improved brain exposure of osimertinib indicated by these preclinical studies may result in improved clinical activity compared with currently available EGFR-TKIs, and also rociletinib. In these studies, osimertinib was more highly distributed to the mouse brain than gefitinib, afatinib, and rociletinib, and penetration of the rat brain was greater than previously described for gefitinib (39). Interestingly, low uptake into brain was also observed for erlotinib in xenografted mice (40). Osimertinib also demonstrated markedly more penetration of the nonhuman primate brain than rociletinib and gefitinib at microdosing levels. In the PC9 EGFRm mouse brain metastases model, osimertinib 25 mg/kg QD induced sustained tumor regression, with the antitumor activity correlating with overall survival. Although a dose of gefitinib 6.25 mg/kg, which roughly equates to a 250 mg QD human clinical dose, also demonstrated tumor regression, this was only for up to 20 days. Interestingly, consistent with distribution studies, no tumor regression was achieved with rociletinib at a dose of 100 mg/kg, and no survival benefit observed. It should be noted that at a dose of 25 mg/kg, plasma exposure of the active metabolites AZ5104 and AZ7550 was ∼24% to 34% that of osimertinib, whereas human exposure of the metabolites has been reported as ∼10% (41). In addition, the plasma terminal half-life of osimertinib was ∼3 hours in mouse models, and reported as at least 50 hours in healthy volunteers (25).

An EGFR-TKI designed specifically to penetrate the BBB (AZD3759) is currently being investigated for treatment of patients with NSCLC with brain metastases. AZD3759 is active against EGFR-TKI sensitizing mutations, and preclinical evidence indicates that this compound shows good penetration of the BBB and induces profound tumor regression in animal models. In addition, in an ongoing phase I study in patients with EGFRm NSCLC (BLOOM; NCT02228369), AZD3759 was well tolerated and was associated with intracranial tumor shrinkage (42). Although data are encouraging, it is important to note that, unlike osimertinib, this compound is not selective for T790M resistance mutations.

PET microdosing has been shown to be a robust method of predicting brain exposure compared with pharmacological dosing (43) and is comparable to microdialysis for confirming adequate brain exposure of CNS drug candidates (44). The low extent of brain exposure for gefitinib in the PET studies is consistent with human clinical experience, lending support to this approach being predictive. Indeed, PET microdosing in nonhuman primates has been used to confirm adequate brain exposure with AZD3241, a drug targeting the CNS to support to the conduct of phase IIa studies in patients (45, 46).

The greater distribution of osimertinib in the brain has the potential to translate into clinical benefit versus other EGFR-TKIs. Based on PKPD modeling, doses of up to 240 mg QD were simulated for brain metastases. These tumor growth simulations predicted that osimertinib at the current clinically recommended dose of 80 mg QD could be sufficient to target human EGFRm NSCLC brain metastases, although 160 mg QD may be more effective. This potency modeling is based on the observed lack of metabolite exposure in preclinical brain models, which contrasts to systemic plasma levels; it will be important to determine whether metabolites have similar lack of exposure in the clinical setting. Indeed, in strong support of these preclinical predictions, we also present early evidence of clinical activity of osimertinib in brain metastases, observed in two case studies of patients enrolled in the phase I AURA study. Both of these patients' cases achieved benefit from osimertinib treatment for controlling brain lesion growth.

The collective preclinical results reported here are promising, and suggest that osimertinib could offer a new clinically significant treatment option for patients with EGFRm brain metastases. Nonetheless, further investigation of osimertinib in patients with EGFRm NSCLC and brain metastases is warranted. An analysis of osimertinib PK in cerebrospinal fluid is an exploratory objective in the ongoing AURA3 (NCT02151981) trial, in which patients with EGFRm advanced NSCLC and stable brain metastases have been enrolled. As patients with brain metastases frequently present with concurrent leptomeningeal metastases (LM; ref. 47), osimertinib and AZD3759 are also being investigated in patients with LM in a phase I study (NCT02228369).

P. Ballard, M. Cantarini, and P. Johnstrom have ownership interest (including patents) in AstraZeneca. J.C-H. Yang reports receiving speakers bureau honoraria from and is a consultant/advisory board member for AstraZeneca. P. Janne has ownership interest (including patents) in Gatekeeper Pharmaceuticals, is a consultant/advisory board member for ACEA Bioscience, Ariad, AstraZeneca, Boehringer Ingelheim, Chugai Pharmaceuticals, Merrimack Pharmaceuticals, Pfizer, and Roche/Genentech, reports receiving commercial research grants from Astellas Pharmaceuticals and AstraZeneca, and post-marketing royalties from Dana Farber Cancer Institute-owned intellectual property on EGFR mutations, which is licensed to Lab Corp. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P. Ballard, J.W.T. Yates, Z. Yang, D.-W. Kim, J.C.-H. Yang, M. Cantarini, P. Johnström, K.S. Thress, D. Cross

Development of methodology: J.C.-H. Yang, M. Box, P. Johnström, J. Malmquist

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.C.-H. Yang, M. Cantarini, A. Jordan, P. Johnström, J. Malmquist, P.A. Jänne

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Ballard, J.W.T. Yates, Z. Yang, D.-W. Kim, J.C.-H. Yang, M. Cantarini, K. Pickup, P. Johnström, K. Varnäs, K.S. Thress, D. Cross

Writing, review, and/or revision of the manuscript: P. Ballard, J.W.T. Yates, Z. Yang, D.-W. Kim, J.C.-H. Yang, M. Cantarini, K. Pickup, A. Jordan, M. Hickey, M. Grist, M. Box, P. Johnström, K. Varnäs, J. Malmquist, K.S. Thress, P.A. Jänne, D. Cross

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Ballard, M. Grist, J. Malmquist

Study supervision: P. Ballard, M. Cantarini, D. Cross

Other (synthesis of the chemical compounds for labeling and use in the studies): M. Box

We thank Jon Moran, PhD, from iMed Comms, an Ashfield Company, part of UDG healthcare plc, who provided medical writing support funded by AstraZeneca. We also thank Sue Ashton and Martine Mellor (plasma and brain in vivo), Ryan Bragg (radiolabels), and M. Raymond V. Finlay (coordination of chemistry) from AstraZeneca, members of the Karolinska Institutet PET group (PET micro-dosing), and Ziqiang Cheng and Kan Chen (ICC; gefitinib in vitro brain binding data) for their contributions to these studies.

This research was funded by AstraZeneca.

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.