Target-Agnostic P-Glycoprotein Assessment Yields Strategies to Evade Efflux, Leading to a BRAF Inhibitor with Intracranial Efficacy

Abstract

The blood–brain barrier (BBB) presents a major hurdle in the development of central nervous system (CNS) active therapeutics, and expression of the P-glycoprotein (P-gp) efflux transporter at the blood–brain interface further impedes BBB penetrance of most small molecules. Designing efflux liabilities out of compounds can be laborious, and there is currently no generalizable approach to directly transform periphery-limited agents to ones active in the CNS. Here, we describe a target-agnostic, prospective assessment of P-gp efflux using diverse compounds. Our results demonstrate that reducing the molecular size or appending a carboxylic acid in many cases enables evasion of P-gp efflux in cell-based experiments and in mice. These strategies were then applied to transform a periphery-limited ^V600EBRAF inhibitor, dabrafenib, into versions that possess potent and selective anti-cancer activity but now also evade P-gp-mediated efflux. When compared to dabrafenib, the compound developed herein (everafenib) has superior BBB penetrance and superior efficacy in an intracranial mouse model of metastatic melanoma, suggesting it as a lead candidate for the treatment of melanoma metastases to the brain and gliomas with BRAF mutation. More generally, the results described herein suggest the actionability of the trends observed in these target-agnostic efflux studies and provide guidance for the conversion of non-BBB-penetrant drugs into versions that are BBB-penetrant and efficacious.

Graphical Abstract

INTRODUCTION

Developing therapeutics for brain diseases remains a formidable challenge, complicated by the protective layer and physical gatekeeper, the blood–brain barrier (BBB).^1–3 Composed of brain endothelial microvascular cells, bound together by high expression of tight junction and adherent junction proteins and further buttressed by glial podocytes, the BBB protects the brain from xenobiotics. In addition to the tight junctions, the expression of P-glycoprotein (P-gp) and other efflux transporters ultimately excludes ~98% of drugs from the central nervous system (CNS).⁴ Even in situations where the BBB might be partially compromised, such as primary brain tumors or metastases to the brain, there is significant difficulty in getting a sufficient quantity of drug to the brain;^3,5,6 for example, virtually all cancer drugs are markedly more effective against a primary systemic tumor compared to a metastasis to the brain.⁵ Bypassing the BBB altogether with direct CNS administration and transient opening of the BBB via targeted ultrasound has garnered some interest,⁷ but this physical disruption can lead to vascular pathology and neurotoxicity.³ Other creative strategies to shuttle therapeutics, both small and large, into the brain are under active investigation.^8–10

Retrospective analyses of physicochemical properties of CNS-approved drugs suggest that generally small and lipophilic compounds have favorable BBB penetrance,^1–3 and as such, increasing passive diffusion remains the principal strategy when modifying compound structure to increase CNS penetrance. BBB penetrance of compounds may be predicted using various formulas, such as the method proposed by Young, utilizing lipophilicity and total polar surface area (TPSA) to calculate cLogBB,¹¹ and computational metrics such as the Multi-Parameter Optimization (MPO) score developed by Wager.¹² The MPO score utilizes six physicochemical properties easily calculated by widely available software, and this user-friendly calculation was followed by publications of other algorithms, such as the probabilistic MPO (pMPO) scores,¹³ the Technically Enhanced Multiparameter Optimization (TEMPO),¹⁴ and the BBB score.¹⁵ The input data sets informing these models are typically from retrospective studies, which are biased toward molecules developed for traditional CNS molecular targets, such as GPCRs and ion channels, and thus trends extrapolated from these studies may not be transferable to all compound types. Indeed, CNS-penetrating compounds that engage more contemporary targets, such as kinases, have starkly different properties than those of CNS-approved drugs (Figure S1).¹⁶ As such, these predictive algorithms can fall short when applied to kinase inhibitors and some other classes of compounds.

The general impermeability of small molecules into the brain is further confounded by high expression of P-glycoprotein.^1,17–19 P-gp, also referred to as MDR1, belongs to an adenosine 5’ triphosphate (ATP)-binding cassette transporter family and serves as the most prevalent multi-drug resistance transporter.^20–22 Expressed at many membrane “barriers”, but predominately at the BBB, P-gp actively pumps out xenobiotics, thereby preventing the accumulation of harmful compounds; thus, a compound must be permeable and also evade P-gp-mediated efflux to accumulate in the brain.^1,17–22 One approach to mitigate the effect of P-gp efflux is co-administration of a drug with a P-gp inhibitor.²³ Several decades of research have culminated in third-generation inhibitors, such as elacridar and tariquidar, that are selective and potent,^23,24 and these compounds indeed potentiate the brain accumulation of compounds of interest in preclinical models and in some preliminary human positron emission tomography imaging studies.^25,26 However, clinical studies indicate that use of these P-gp inhibitors erodes therapeutic indices, and thus, these compounds have had a limited translational impact.^23,26

Even though P-gp is the most extensively studied transporter, its broad specificity necessitates empirical determinations of the efflux status of compounds in cell-based models (such as Caco-II or MDR1-MDCK assays) in the CNS-targeting drug development process.^{17–19,27,28} P-gp has multiple binding sites in a large binding cavity, and there are no defined pharmacophores for substrate recognition. Predicting efflux of structurally related compounds in lead optimization efforts has improved with machine learning,¹⁸ but the transferability across different chemotypes may be limited. Pharmaceutical companies have published some of their proprietary in-house QSAR models of P-gp efflux^17,18 and brain penetrance^27–30 using the wealth of data from candidates across different CNS discovery campaigns. These retrospective analyses suggest that reducing hydrogen bond donor ability, polar surface area, and the most basic pK_a can decrease efflux propensities of P-gp substrates.^1,17,18,28

Despite these efforts, no systematic method for the direct transformation of P-gp substrates into P-gp non-substrates has been described. As such, conversion of a lead candidate to a BBB-penetrant version is laborious and ad hoc, typically requiring lengthy medicinal chemistry campaigns and the synthesis of a large number of compounds; this is especially true for kinase inhibitors, many of which are P-gp substrates and generally have physicochemical properties drastically different from typical CNS-targeting molecules. For example, a recently reported BBB-penetrant CDK4/6 inhibitor required the complete redesign of the core scaffold,²⁹ and the conversion of crizotinib (an ALK/ROS1 fusion inhibitor) to the CNS-penetrant lorlatinib required the synthesis and evaluation of over 1000 compounds (Figure 1A).^30,31 Herein, we describe our efforts to elucidate properties of P-gp recognition via prospective analysis of a diverse set of compounds. We demonstrate that reduction of molecular weight (MW) or installation of a carboxylic acid can, in many cases, facilitate P-gp efflux evasion in cell-based systems and in mouse models. These lessons were applied to redesign a BRAF inhibitor, leading to a potent version that has reduced efflux propensity, enhanced brain penetration, and activity in a challenging intracranial mouse model of melanoma (Figure 1B). The strategies described herein can serve as a guide to reducing P-gp-mediated efflux and, together with existing approaches to improve brain accumulation, should facilitate conversion of other non-CNS-penetrant drugs into ones that have enhanced accumulation and activity in the CNS.

Figure 1.

(A) Extensive medicinal chemistry campaigns are often required to reduce P-gp efflux and identify a CNS-penetrant analogue, with the conversion of crizotinib to lorlatinib as an example. (B) Identification and application of molecular features that evade efflux could facilitate conversion of P-gp substrates to non-substrates.

RESULTS AND DISCUSSION

Unbiased Assessment of P-gp Efflux Using Diverse Compounds.

To assess chemical features that influence P-gp efflux in a target-agnostic fashion, a collection of structurally diverse compounds was required. Commercially available drug-like compounds (including kinase inhibitors) were utilized, and these were complemented by a collection of natural product-like compounds. Produced using the “complexity-to-diversity” (CtD) strategy,^32–38 this collection includes scores of compounds with structures distinct from those in traditional screening sets and CNS drugs, and members can be systematically modified as needed during follow-up experiments.

The permeability of compounds and their recognition by the critical efflux pump was evaluated using the P-gp transwell assay, adapted from known protocols.^39–41 Compound transport across MDCK epithelial cells overexpressing human P-gp (MDR1) can be quantified via liquid chromatography–tandem mass spectrometry (LC–MS/MS) to calculate permeability (P_app) from apical (A)-to-basolateral (B) (A–B) and basolateral-to-apical (B–A) directions, providing an efflux ratio (ER) (Figure 2A). To ensure the monolayer integrity and to account for any compound cytotoxicity, a cell-impermeable fluorescent probe, Lucifer Yellow, was utilized in each experiment. The optimized assay was validated using a series of controls, both drug-like and natural product-like small molecules whose activity in this assay has been previously reported. These include periphery-limited compounds, such as ritonavir (anti-viral),^40–42 crizotinib (anti-cancer),^31,42 binimetinib (anti-cancer),^42,43 quinidine (anti-arrhythmic),^39,40,42 prazosin (anti-hypertensive),^39,40,42 vemurafenib (anti-cancer),^42,44 and terfenadine (non-drowsy anti-histamine),^39,41,42 all of which were found to have high ERs in this assay (Figure 2B). Compounds that have CNS indications and/or activity, such as yohimbine (reverse sedation),^39,42 PAC-1 (anti-cancer),^45–47 naltrexone (analgesic),^41,42 propranolol (anti-arrhythmic),^39,41,42 lorlatinib (anti-cancer),^30,31,42 scopolamine (anti-emetic),^41,42 trimipramine (anti-depressant),^41,42 and chlorpheniramine (drowsy anti-histamine),^39,41,42 all exhibit low ERs (less than 2.0) in this assay (Figure 2B). Notable examples include crizotinib and lorlatinib, shown in Figure 1A, ALK/ROS1 fusion inhibitors with high (ER = 10.02) and low efflux (ER = 1.22), respectively, consistent with their periphery-limited/CNS-penetrant phenotype.^30,31

Figure 2.

Assay validation and assessment of P-gp efflux in the MDR1-MDCK transwell assay. (A) Schematic of the MDR1-MDCK transwell assay. On a monolayer of MDCK cells, human P-glycoprotein (MDR1) is overexpressed on its apical side creating a polarized transport system. Compounds (10 μM) are added to the apical (A) or basolateral (B) side, and compound transport is measured by quantifying compound concentration in all four compartments via LC–MS/MS after 90 min incubation with shaking, allowing for the calculation of an ER. (B) Compounds with known CNS penetrance have lower ERs than those limited in the periphery; structures and permeability data are in Supporting Information Table S1. (C) Co-treatment with elacridar, a P-gp inhibitor, dramatically reduces the ERs of substrate controls; structures and permeability data are in Supporting Information Table S2. (D) Assessment of 93 compounds for P-gp efflux plotted against TPSA; structures, efflux, and permeability data are in Supporting Information Table S3. Physicochemical properties were calculated using Chemicalize Platform (ChemAxon) and are reported in Supporting Information Table S4. The graph excludes two data points with ER > 30. (E) Efflux of neutral compounds (replotted from Figure S2B) is significant at high MWs. Data in (B) and (C) are plotted as mean ± s.e.m., n ≥ 3 biological replicates. Statistical significance (relative to propranolol in (B) and relative to without elacridar in (C)) was determined by using a two-sample Student’s t-test (two-tailed test, assuming equal variance) *P < 0.05, **P < 0.01, ***P < 0.0001, ****P < 0.0001.

Based on the evaluation of these controls, the threshold for P-gp substrate versus non-substrate was set at an ER of 2.0, recognizing that there would be some ambiguity and a continuum of efflux events for compounds around the threshold. In a similar fashion a threshold for low permeability was set, as compounds must be permeable enough to reach the trans-membrane domain of P-gp for recognition.⁴⁸ Mannitol, often used as a low-permeability marker,³⁹ has a reported P_app between 8 and 11 nm/s. We found one of the substrate controls, terfenadine, to have a low permeability (12 nm/s) but reliably an ER of 3.1, consistent with previous reports.^39,41 As such, compounds with an average permeability in both directions (P_{app avg}) of less than 10 nm/s were considered as low-permeable compounds and not included in the analysis, an approach consistent with other reports.^18,39–41 Compounds with P_{app B–A} greater than 100 nm/s were conservatively considered as highly permeable, guided by literature thresholds.^28,40,45,49 All permeability and efflux data of control compounds are listed in Supporting Information Table S1. To further ensure that the ERs of the training set were not solely due to differences in their compound type and permeability, a subset of controls with various permeabilities was evaluated upon co-treatment of a P-gp inhibitor, elacridar. Reduced ERs for substrates were observed in the presence of the inhibitor in all cases, while the ERs of non-substrates remained the same (Figure 2C, data in Supporting Information Table S2), validating these classifications.

With the key P-gp assay validated, 117 compounds (including controls) were evaluated in this optimized transwell assay. Ninety-three compounds met the permeability, lack of cytotoxicity, and mass balance thresholds for the initial test set (representing both drug-like and CtD compounds, see Supporting Information Table S3 for all structures and Supporting Information Table S4 for their properties). Upon analysis of the results from this test set, no clear trend of efflux recognition was observed with total polar surface area (TPSA) (Figure 2D), lipophilicity (Figure S2A), MW (Figure S2B), or other in silico metrics used to predict CNS penetrance such as cLogBB (Figure S2C)¹¹ or MPO scores (Figure S2D)¹² that take into account multiple physicochemical properties; they were all unable to differentiate between substrates and non-substrates (Figures 2D and S2A–D). Stratifying these compounds according to their charge states, however, revealed that within neutral compounds, higher MWs (especially those above 500 g/mol) tended to correlate with higher propensity for efflux (Figure 2E). Our second and perhaps more striking observation was that many negatively charged compounds had low efflux; 17 of 19 carboxylic acids in this initial test set evaded efflux (yellow diamonds, Figure 2D). Although our observation of an MW dependence of efflux has not been observed before in this type of unbiased prospective analysis, it is consistent with the known dependence of MW on BBB penetrance;^49–51 as such, we sought to explore in depth the more surprising observation regarding carboxylic acid-containing compounds and their evasion of P-gp efflux.

Carboxylic Acid-Containing Compounds Evade Efflux in the Transwell Assay.

To follow up on this unexpected observation in the MDR1-MDCK system, a subset of compounds was evaluated in two secondary assays (see structures in Figure S3A). As some carboxylic acid-containing compounds cross the BBB through active uptake mechanisms,^1,52 a subset of carboxylic acid-containing compounds was evaluated in the parent MDCK cells without the overexpression of P-gp. The results from this experiment revealed that all compounds had similar ERs in the MDCK and MDR1-MDCK systems, suggesting that the observed lack of efflux is not due to other modes of uptake by endogenous transporters present in the MDCK cells (Figure S3B, permeability and efflux data in Supporting Information Table S5). Another consideration was exploring the potential of carboxylic acids as P-gp inhibitors, and this was assessed utilizing the calcein-AM fluorescent dye in MDR1-MDCK cells, adapted from known protocols.^39–41,53 The fluorescence of calcein-AM is released only upon activation (to calcein) by intracellular esterases, and calcein is subsequently expelled by P-gp. With P-gp activity intact, a low-fluorescent signal is measured in this assay, but P-gp inhibition results in higher fluorescence.⁵³ As expected, known inhibitors elacridar and quinidine give robust inhibition in this experiment, but none of the seven structurally diverse carboxylic acids evaluated showed significant inhibition in this assay (Figure S3C).

A larger set of structurally diverse carboxylic acid-containing compounds (an additional 70 compounds, together with the 19 above for a total of 89, see Supporting Information Table S6) was then evaluated in the MDR1-MDCK transwell assay. Most of these carboxylic acids show low efflux and low permeability (Figure 3A); however, a wide range of permeabilities was observed while maintaining P-gp evasion, illustrating that evasion of efflux is not simply due to low permeability of these compounds. With a negative charge at physiological pH, many carboxylic acids are expected to suffer from low permeability (as will be discussed further), with low P_{app avg} (<10 nm/s); indeed, of these 89 compounds, 19 of them had low permeability by this definition. However, of the carboxylic acid-containing compounds that do have appreciable permeability in this experiment (n = 70), over 85% are not recognized by P-gp (Figure 3B). Exceptions were observed, including cetirizine—often referred to as an example of P-gp’s affinity toward carboxylic acids¹⁷—which was indeed a P-gp substrate with an ER of 5.88 (Figure 3A), consistent with the literature.⁵⁴ A subset of carboxylic acids was further assessed in the transwell assay with co-administration of a P-gp inhibitor, and substrate carboxylic acids had their efflux significantly reduced, while this inhibition did not affect the efflux classification of non-substrate carboxylic acids (Figure S3D,E and permeability data in Supporting Information Table S7).

Figure 3.

(A) ERs and permeability of 89 carboxylic acids assessed in the MDR1-MDCK transwell assay. P_app is expressed as the average of A–B and B–A directions, in nm/s. Structures, permeability, and efflux data are in Supporting Information Table S6. (B) Summary of P-gp efflux of the 89 carboxylic acid-containing compounds in the MDR1-MDCK transwell assay. Compounds are categorized as substrates when ER ≥ 2.0 and non-substrates when ER < 2.0.

Carboxylic Acid-containing Compounds Evade P-gp Efflux in Vivo.

To investigate the actionability of this observation that most carboxylic acid-containing compounds evade P-gp efflux, several compounds recognized by P-gp were modified to contain carboxylic acids. Upon appending carboxylic acid moieties to these substrates, P-gp efflux decreased (albeit at the cost of lower permeability) across four examples from different scaffolds (Figure 4): quinine was recognized by P-gp with an ER of 5.77, and upon installation of carboxylic acid, quinine-CO₂H (1) evaded P-gp recognition and exhibited a low ER of 0.48. When Q8 (2), sinomenine, and A7 (5) were all functionalized with a carboxylic acid moiety, their ERs were also dramatically reduced. While the converse was not universally true, removal of the carboxylic acids (via creation of the methyl ester) did significantly increase the ER in many cases (9 out of 20, Figures S4 and S5), suggesting that the observed efflux evasion in Figure 4 is not merely due to masking of a hydrogen bond donor. The observation that some compounds where the carboxylic acid is converted to the methyl ester evade efflux is unsurprising as there are multiple compounds lacking carboxylic acids that evade P-gp efflux, including some control compounds included in Figure 2B. Overall, our data suggest a high likelihood (>85%) of efflux avoidance for compounds containing a carboxylic acid.

Figure 4.

P-gp substrates can be converted to non-substrates by appending a carboxylic acid moiety. Structures of P-gp substrates and their carboxylic acid-containing derivatives. Below each compound is the respective permeability (P_app of A–B and B–A directions, in nm/s) and ER, assessed in the MDR1-MDCK transwell assay. Compounds are categorized as substrates when ER ≥ 2.0 (in red) and non-substrates when ER < 2.0 (in green). Efflux data are reported as mean ± s.e.m. The s.e.m. value of permeability is in Supporting Information Table S8. n ≥ 3 biological replicates.

To investigate these observations and trends in animals, the P-gp assay was implemented and optimized in mice, adapted from prior reports.^44,55–57 Briefly, mice were pretreated with elacridar; then after 30 min, compounds of interest were administered (Figure 5A). Blood and brain were harvested after 5, 15, or 60 min, and compound concentrations in each compartment were quantified through LC–MS/MS analysis to determine the potentiation by P-gp inhibition. Using this system, the brain partitioning of quinidine, a P-gp substrate control, significantly increased (by 7.32-fold) at 60 min, while that of propranolol, a P-gp non-substrate control, was unchanged with elacridar pretreatment (Figure 5B,C). Bio-distribution data of controls (serum and brain concentration and brain-to-serum ratios of multiple time points) are found in Figure S6A–G.

Figure 5.

P-gp in vivo assay. (A) Schematic of the P-gp in vivo assay: mice were treated with 2.5 mg/kg elacridar or its vehicle intravenously; after 30 min, mice were administered with a single injection of 25 mg/kg compound (unless noted otherwise) via the lateral tail vein. At the specified time point, blood and brain samples were collected to measure the BBB penetrance. (B) Structures of controls (quinidine and propranolol) and levofloxacin and its methyl ester. Below each compound is the respective ER assessed in the MDR1-MDCK assay. (C) Validation of the P-gp in vivo assay. Brain-to-serum ratios at 60 min of quinidine (a P-gp substrate control) are potentiated by P-gp inhibition, while propranolol (a P-gp non-substrate control) is not. Propranolol was administered at 12.5 mg/kg due to observed neurotoxicity at a higher dose. All biodistribution data for controls for multiple time points in Figure S6A–G. (D) Brain-to-serum ratios of levofloxacin and its methyl ester with or without elacridar at 60 min (data for 15 min and all biodistribution data in Figure S7A–F). P-gp inhibition has no statistically significant effect on levofloxacin, a carboxylic acid-containing P-gp non-substrate, while it potentiates brain partitioning of its structurally similar substrate, levofloxacin-methyl ester. (E) Structures of carboxylic acid-containing compounds. Below each compound is the respective permeability (P_app of A–B and B–A directions, in nm/s) assessed in the MDR1-MDCK transwell assay. n ≥ 3 biological replicates. (F) Summary of brain-to-serum ratios of some carboxylic acid-containing compounds with reasonable permeability at 60 min (data for 15 min in Figure S8M and all biodistribution data in Figure S8A–L). (G) Fold changes in the mean brain-to-serum ratios measured in mice with or without elacridar in the P-gp in vivo assay (used as in vivo ERs), and correlated with ERs assessed in the cell-based MDR1-MDCK transwell assay. Efflux data are shown as mean ± s.e.m, n ≥ 3 biological replicates. Number of mice per cohort n ≥ 3. Statistical significance was determined by using a two-sample Student’s t-test (two-tailed test, assuming equal variance). *P < 0.05, **P < 0.01.

With these controls in place, carboxylic acid-containing levofloxacin, an antibiotic with clinical evidence for some CNS penetrance,⁵⁸ was evaluated along with its methyl ester. From our cell culture experiments, levofloxacin was shown to evade P-gp efflux (ER = 0.74) with appreciable permeability, while the methyl ester derivative of levofloxacin (7) was expelled by P-gp with an ER of 5.66 in the transwell assay (Figure 5B). In mice, P-gp inhibition had a negligible effect on the brain-to-serum ratio of levofloxacin with a 1.67-fold increase but significantly potentiated the methyl ester counterpart of levofloxacin by 4.66-fold at 60 min (Figure 5D). Comparing the brain exposure of the two compounds in the absence of the P-gp inhibitor, levofloxacin had statistically superior partitioning to the brain compared to P-gp-recognized levofloxacin-ME after 60 min, although they had comparable partitioning at 5 min (Figure S7A–F). In comparing another non-substrate acid [G8 (8), ER = 0.99] and a substrate non-acid [G8-ME (9), ER = 9.62] head-to-head, we observed a similar result as the levofloxacin pair, in which the partitioning of G8-ME was potentiated by elacridar, while G8 was not at 5 min (Figure S7G–L); we were unable to assess this relationship at 60 min as the serum level of G8-ME was undetectable due to its hydrolysis (Figure S7J).

Five additional structurally diverse carboxylic acid-containing compounds were evaluated in this in vivo assay to determine if the observations about efflux evasion in cell culture also hold true in mice. These compounds included approved CNS agents (tiagabine and tianeptine) and a non-CNS agent (cilomilast) with distinct biological targets as well as tool compounds without annotated biological activities (10 and G8). While all these compounds had ERs of less than 2.0 in the cell-based assay, they differ in permeability, from 30 to 200 nm/s (Figure 5E). G8, which had the lowest passive permeability, also had the lowest brain-to-serum ratio (Figure 5F). Four of five compounds with permeability above ~50 nm/s had brain-to-serum ratios in accord with their permeability (Figure 5F), and P-gp inhibition only minimally potentiated their brain partitioning (Figure 5F, tiagabine had statistically different brain partitioning at 60 min only). Only compound 10 exhibited a low brain partitioning despite its high permeability observed in the transwell assay (Figure 5F), suggesting rapid metabolism and clearance of this molecule both from the brain and circulation. This result highlights the multitude of processes that small molecules undergo beyond a simple traversal of the epithelial cell layer, mimicked in the transwell assay. In general, however, the observed in vivo ERs are concordant with the ERs measured in the transwell assays for most compounds, as shown in Figure 5G, suggesting that compounds with the carboxylic acid moiety can evade P-gp efflux in vivo. All biodistribution data are found in Figures S7A–L and S8A–M.

Design of ^V600EBRAF Inhibitors that Evade P-gp Efflux.

Our twin observations about MW and carboxylic acids should, in principle, aid in the conversion of drugs that are periphery-limited (due to high P-gp efflux) to those that evade efflux and can hence be active in the CNS. For addition of a carboxylic acid, ideal candidates for this conversion to a BBB-penetrant version would be compounds that are highly permeable yet limited by P-gp efflux. Anti-proliferative targeted kinase inhibitors, such as dabrafenib and imatinib (with limited CNS exposure due to P-gp efflux), are of particular interest, as BBB-penetrant versions are highly sought for treatment of metastatic lesions in the brain.^59–61 Considering that kinase inhibitors exist outside of the traditional CNS-targeted small-molecule chemotypes as discussed, the ability to rationally redesign such compounds would be a valuable and directly actionable feature of our findings. Dabrafenib (Figure 6A) has marked potency and selectivity toward melanoma cells harboring the ^V600EBRAF mutation and was approved to treat peripheral melanoma in 2013;^62,63 however, dabrafenib is strongly recognized by P-gp and thus unable to effectively accumulate in the brain.⁶⁴ When assessed in mice lacking P-gp, dabrafenib reaches therapeutically relevant concentrations in the brain.⁶⁴ Dabrafenib is efficacious in intracranial tumors in mice when co-treated with a membrane permeabilizer but has minimal activity on its own,⁶⁵ suggesting that a BBB-penetrant version could be highly efficacious.

Figure 6.

Conversion of dabrafenib to versions that evade P-gp efflux. (A) Structures of dabrafenib, everafenib, and everafenib-CO₂H, along with their permeabilities and ERs as assessed in the MDR1-MDCK transwell assay. (B) Everafenib displays similar potency to dabrafenib, while everafenib-CO₂H is superior to vemurafenib (Figure S11B,C), and both maintain cytotoxicity selective to cell lines harboring the ^V600EBRAF mutation across different tissue origins in a 72 h cell viability assay. Data are plotted as mean ± s.e.m., n = 3 biological replicates. (C) Everafenib and everafenib-CO₂H inhibit MAPK signaling, suggesting retention of the mode of action in a cell line harboring ^V600EBRAF mutation (A375) but not WT (CHL-1). Phospho-MEK1/2 and phospho-ERK1/2 inhibition by DMSO, 10 μM PLX4720, 1 μM dabrafenib, 1 μM encorafenib, 1 μM everafenib, and 10 μM everafenib-CO₂H for 1 h. Higher concentrations of PLX4720 and everafenib-CO₂H were chosen due to their higher IC₅₀ value. (D) Dose–response of phospho-ERK1/2 inhibition in A375 cells treated with DMSO; 0.01, 0.1, 1, and 10 μM of everafenib or everafenib-CO₂H; and 10 μM of dabrafenib or PLX4720 for 1 and 24 h.

The wealth of structure–activity relationship (SAR) data established in the development of BRAF inhibitors informed our design of derivatives incorporating reduced MW or a carboxylic acid. Disclosed SAR information on over 290 derivatives⁶⁶ and the co-crystal structure of dabrafenib bound to human ^V600EBRAF enzyme⁶⁷ reveals that the aminopyrimidine tail group (substitution of the 2 position of the pyrimidine ring, see Figure 6A) is amenable to modification, and polar functional groups can be incorporated without compromising selective anti-cancer activity.^62,66,67 These considerations guided the design of three derivatives containing carboxylic acid moieties on the aminopyrimidine tail (compound 11 in Figure 6A and others in Figure S9). Evaluation of these carboxylic acid-functionalized dabrafenib derivatives in the transwell assay gratifyingly revealed a significant reduction in ER; while dabrafenib was readily expelled with an ER of 17.9, the new compounds had ER values less than 2.0 (Figures 6A and S9). Evaluation of their ability to induce death of A375 melanoma cells (which harbor ^V600EBRAF) revealed compound 11 as the most potent with an IC₅₀ of 115 nM (Figure S9).

With this first success in hand and with the goal of creating more potent compounds that evade P-gp efflux through reduction in MW or addition of a carboxylic acid, examination of the three approved drugs targeting ^V600EBRAF inspired the design of hybrid compounds (see Figure S10). These novel compounds incorporate the propyl sulfonamide from vemurafenib (for reduction in MW), the 5-chloro-2-fluoro substitution pattern of the phenyl core from encorafenib (to increase lipophilicity), the tert-butyl thiazole from dabrafenib, and 2,4-pyrimidine from dabrafenib and encorafenib. This led to the hybrid compound 12, hereafter referred to as everafenib (Figure 6A), a neutral compound with an MW that has now been reduced below 500 g/mol (484.01 g/mol relative to 519.56 g/mol for dabrafenib), as well as its carboxylic acid-containing version 13, everafenib-CO₂H (Figure 6A). After the synthesis of these new compounds, their assessment in the transwell assay gratifyingly validated our design and revealed that both these compounds indeed have low ERs: everafenib has an ER of 1.40, and everafenib-CO₂H has an ER of 1.17 (Figure 6A). Encorafenib, a structurally similar approved BRAF inhibitor with a non-carboxylic acid-containing side chain, is strongly recognized by P-gp with an ER of 21.8 (Figure S10).

Evaluation against cancer cells in culture revealed that everafenib is a highly potent inducer of death as assessed against a panel of ^V600EBRAF melanoma cell lines with IC₅₀ values of 2–10 nM, comparable to or better than dabrafenib (Figures 6B and S11A,B). Everafenib-CO₂H is also potent against cancer cells with the ^V600EBRAF mutation, although not as potent as everafenib (Figure 6B) but still markedly more potent than vemurafenib (Figure S11B,C). Critically, both everafenib and everafenib-CO₂H maintain their selectivity for cancer cells harboring ^V600EBRAF mutation, with IC₅₀ values greater than 1 μM in cell lines with ^WTBRAF (Figures 6B and S11B,C).

To further assess if the cytotoxicity of everafenib and everafenib-CO₂H is derived from inhibition of the constitutively activated MAPK signaling found in ^V600EBRAF cells, MEK1/2 and ERK1/2 phosphorylation was assessed. The MAPK pro-growth signaling was strongly inhibited upon treatment with everafenib and everafenib-CO₂H after 1 h of treatment, comparable to dabrafenib, encorafenib, and PLX4720 (a progenitor compound to vemurafenib, Figure S11D) in A375 cells (Figure 6C) and AM-38 glioma cells with ^V600EBRAF (Figure S11E). Following the identical treatment in CHL-1, a ^WTBRAF melanoma cell line, inhibition of the MAPK signaling was not observed (Figure 6C). This is consistent with previous reports where vemurafenib does not inhibit the MAPK signaling but paradoxically activates ERK1/2 phosphorylation in a ^WTBRAF cell line.⁶⁸ After both 1 and 24 h of treatment in A375 cells, there was a clear dose-dependent inhibition imparted by everafenib and everafenib-CO₂H (Figure 6D, see Figure S11F for AM-38). Inhibition of ERK1/2 phosphorylation was sustained even after 24 h at the highest concentrations of everafenib and everafenib-CO₂H, while dabrafenib or PLX4720 had diminished inhibition at this time point (Figure 6D).

Everafenib is BBB-Penetrant and Has Efficacy in an Intracranial Model of Metastatic Melanoma.

Given their low efflux liabilities and potent anti-cancer efficacy in cell culture, we sought to evaluate the BBB penetrance of everafenib and everafenib-CO₂H in mice and compare them alongside approved agents targeting ^V600EBRAF. The cell culture permeabilities of everafenib (49–64 nm/s) and everafenib-CO₂H (22–25 nm/s) (Figure 6A) suggested that everafenib may be the more promising candidate in vivo. Indeed, assessment of brain and serum drug levels 60 min following a single intravenous injection of encorafenib, PLX4720, dabrafenib, or everafenib-CO₂H revealed that all these compounds have poor brain-to-serum ratios (0.0014–0.0069 ng/g:ng/mL, Figure 7A). Excitingly and significantly, everafenib achieved a markedly higher brain-to-serum ratio than all other inhibitors (0.286 ng/g:ng/mL, Figure 7A). Most importantly, the absolute brain concentration of everafenib was substantially higher than that of dabrafenib and other BRAF inhibitors (Figure 7B). Evaluating dabrafenib and everafenib-CO₂H in the P-gp in vivo assay revealed that P-gp inhibition greatly enhanced the BBB penetrance of dabrafenib 4.3- and 5.8-fold at 15 and 60 min (Figure S12A–C). Consistent with the results from the cell-based efflux assay, everafenib-CO₂H had slight but not statistically significant potentiation of 1.5- and 2.5-fold (Figure S12D–F). As shown in Figure 7B, everafenib achieves an absolute concentration in the brain markedly higher than that of dabrafenib, although it did display potentiation with elacridar (Figure S12G–I), suggesting that other efflux mechanisms (beyond P-gp) may be operational in vivo or that murine and human P-gp recognize this compound differently. All in vivo biodistribution data of BRAF inhibitors (serum and brain concentration and brain-to-serum ratios of multiple time points) are found in Figures S12A–I and S13A–C.

Figure 7.

Everafenib is BBB-penetrant and efficacious in the A375 intracranial model. (A,B) Everafenib is more BBB-penetrant than dabrafenib. Brain-to-serum ratios (A) and brain concentration (B) of BRAF inhibitors at 60 min, following a single 25 mg/kg IV injection. Everafenib achieves the highest brain partitioning. Number of mice per cohort n ≥ 3. Statistical significance of each compound compared to everafenib was determined by using a two-sample Student’s t-test (two-tailed test, assuming equal variance). (C) Evaluation of everafenib in an intracranial mouse model. 50,000 A375 melanoma cells were intracranially implanted into 7-week old female athymic nude mice. Five days after inoculation, mice were treated with the vehicle or 50 mg/kg of dabrafenib or everafenib intraperitoneal (IP) once a day for 5 days. After 2 days off, another five daily doses were administered, a total of 10 treatments in the model. Cells were tested to be free of pathogens prior to inoculation. Vehicle vs dabrafenib: P = 0.009, vehicle vs everafenib: P = 0.005, dabrafenib versus everafenib: P = 0.0018. Number of mice per cohort = 6. Statistical significance was determined by a two-tailed log-rank (Mantel–Cox) test. **P < 0.01 ***P < 0.001 ****P < 0.0001.

The increased BBB penetrance observed for everafenib relative to dabrafenib, in addition to its pharmacokinetic assessment in mice (Figure S14), its outstanding potency and selectivity in cell culture (Figure 6B), and sustained inhibition of ERK1/2 phosphorylation (Figure 6D), suggested this compound as an excellent candidate for evaluation in a mouse model of melanoma metastasis to the brain. Such a model has been developed using intracranial implantation of the A375 melanoma cell line,^69–71 and thus, this macro-metastatic model via injection of A375 cells to the mouse forebrain was used to assess everafenib and dabrafenib head-to-head. In the experiment, mice were intracranially implanted with A375 cells and then 5 days later were treated with 50 mg/kg of dabrafenib or everafenib (two cycles of once-per-day for 5 days via IP injection, a total of 10 doses). Mice treated with everafenib outperformed dabrafenib and increased median survival from 39 to 50.5 days (Figure 7C). This result suggests that the increased BBB penetrance of everafenib, attributed to reduced P-gp efflux, improved its efficacy compared to dabrafenib.

CONCLUSIONS

The major findings of this target-agnostic evaluation of P-gp efflux are that (1) reducing MW or (2) appending a carboxylic acid can often reduce P-gp-mediated efflux. Our observations of MW dependence are consistent with other studies reporting high-MW compounds as strong P-gp binders and efflux substrates, including paclitaxel,⁷² vincristine,⁷³ cyclic peptides,²⁰ and others.^41,49,50,74 While in silico predictive models for BBB penetrance have utility, they do not predict significant brain accumulation for the FDA-approved drug dabrafenib (MPO score = 1.81, cLogBB = −0.67) or our novel BBB-penetrant BRAF inhibitor everafenib (MPO = 2.07, cLogBB = −0.79); both are predicted to be excluded from the CNS by these common prediction algorithms. Our second observation that the majority of carboxylic acid-containing compounds evade P-gp efflux is even more surprising, and indeed, the use of a carboxylic acid to enhance CNS penetrance is counterintuitive. However, observations in the literature that carboxylic acid moieties are detrimental to overall CNS penetrance¹⁷ appear to predominately be due to reduction in permeability of anionic compounds and perhaps overly influenced by the cetirizine example, with only a small number of carboxylic acid-containing compounds themselves (such as cetirizine, evaluated herein) actually displaying high P-gp efflux (Figure 3A).

Of course, some carboxylic acid-containing compounds have specific transporters mediating uptake into the brain, but there are scattered lines of evidence suggesting that carboxylic acids can achieve non-transport-mediated brain accumulation. First, there are indeed some approved CNS drugs (including tiagabine and tianeptine, evaluated herein)^75–78 and preclinical candidates^79–82 with carboxylic acids that do not fit substrate scopes of common amino-acid transporters;⁵² our new data suggest that the carboxylic acid functional group on these drugs may help them evade efflux, leading to brain accumulation and biological activity. Second, in accord with our observations, fluroquinolones and methotrexate (each containing carboxylic acid moieties) have no ATPase activity measured in P-gp embedded in plasma membrane vesicles, indicating no interaction between these compounds and the efflux transporter,⁸³ and functionalizing paclitaxel with a carboxylic acid moiety improved BBB penetrance in rats.^84,85 Thus, while the general phenomena of carboxylic acid-containing compounds evading efflux has not been reported, through the lens of our new data, some supporting evidence can be found in the literature. P-gp’s binding cavity is composed of hydrophobic residues, and the translocation pathway of P-gp has been described as hydrophobic with some acidic patches;⁸⁶ as such, unfavorable interactions of anionic species with the protein may explain the observed evasion of P-gp efflux for carboxylic acid-containing compounds and suggests that anionic non-carboxylic acids may also evade P-gp recognition.

There are limitations to this study that are worth noting. First and foremost, the general utility of appending a carboxylic acid to reduce efflux will depend on being able to maintain reasonable permeability in the final compound, and the addition of a carboxylic acid may well push the permeability of a compound below a useful threshold (as is the case for everafenib-CO₂H) and could also deleteriously affect the volume of distribution. Second, the translatability of these findings to the human BBB remains to be determined. While it is heartening to observe that ERs from the cell-based assay expressing human P-gp were largely recapitulated in mice, the entire murine P-gp protein shares only 87% sequence identity with humans (although the drug binding pocket of murine and human P-gp do share 100% sequence homology).²⁰ Third, the transwell system utilized herein accounts only for P-gp, but other efflux transporters, such as breast cancer resistance protein (BCRP) and multidrug resistance-associated proteins (MRP), also play a gatekeeping role at the BBB interface.^19,87 The P-gp inhibitor used in this study, elacridar, also inhibits BCRP,^44,57 and thus, potentiation observed in vivo may also be due to BCRP inhibition.

These actionable twin observations from our unbiased analysis enabled the development of two potent and selective ^V600EBRAF inhibitors with reduced efflux liabilities. Following the discovery of the V600E mutation in BRAF as an oncogenic driver in 50% of melanomas, the development and approval of vemurafenib, followed by dabrafenib and encorafenib, led to dramatic improvement of the survival outcome of melanoma patients whose tumors harbored that mutation.^88,89 However, over 50% of metastatic melanoma patients eventually develop metastasis to the CNS.⁹⁰ Despite undetectable and/or low cerebrospinal fluid levels in patients^91,92 and limited penetrance to the brain in preclinical models,^44,64,65 there are several reports on the use of dabrafenib and vemurafenib in melanoma patients with brain metastatic lesions.^93–97 When combined with trametinib (a MEK inhibitor and also a P-gp substrate),⁴³ dabrafenib provides only a short duration (6.5 months) of intracranial response in clinical trials.⁹⁴ This modest activity is attributed to limited brain accumulation and highlights the need for novel ^V600EBRAF inhibitors that have significantly enhanced BBB penetrance. The brain accumulation of the top compound detailed herein, everafenib, supersedes those of all approved agents targeting ^V600EBRAF in our head-to-head experiments, suggesting its potential for superior efficacy against intracranial tumors. In the A375 intracranial mouse model, everafenib is indeed superior to dabrafenib, presumably due to its potent activity, sustained phospho-ERK1/2 inhibition, and the lack of P-gp efflux.

Beyond metastatic melanoma, dabrafenib and vemurafenib have some efficacy in primary brain cancer patients with the ^V600EBRAF mutation.⁹⁸ In a histology agnostic trial with vemurafenib, four out of seven pleomorphic xanthoastrocytoma patients responded to ^V600EBRAF inhibition,⁹⁹ and it also had meaningful clinical activity in pediatric low-grade glioma patients with ^V600EBRAF.^100,101 In our cell culture studies, everafenib exhibited comparable cytotoxicity to dabrafenib against AM-38, a glioblastoma cell line with ^V600EBRAF, while having no activity against glioma cell-lines with ^WTBRAF (U118MG, T98G, U87). These data suggest that a BRAF inhibitor with superior CNS penetrance may provide greater survival benefits for patients with primary brain cancers whose tumors harbor this mutation. More generally, the observations herein about MW and the presence of a carboxylic acid complement the existing tools to improve BBB penetrance of compounds and, in conjunction with carboxylic acid isosteres that maintain their anionic nature, should aid the design of BBB-penetrant versions of drugs from a variety of classes.

ABBREVIATIONS

BBB

blood–brain barrier

P-gp

P-glycoprotein

CNS

central nervous system

MPO

Multi-Parameter Optimization

pMPO

probabilistic MPO

TEMPO

Technically Enhanced Multiparameter Optimization

CtD

complexity-to-diversity

ER

efflux ratio

TPSA

total polar surface area

SAR

structure–activity relationship

BCRP

breast cancer resistance protein

MRP

multidrug resistance protein