Central Nervous System Distribution of Panobinostat in Preclinical Models to Guide Dosing for Pediatric Brain Tumors

Abstract

Achieving adequate exposure of the free therapeutic agent at the target is a critical determinant of efficacious chemotherapy. With this in mind, a major challenge in developing therapies for central nervous system (CNS) tumors is to overcome barriers to delivery, including the blood-brain barrier (BBB). Panobinostat is a nonselective pan–histone deacetylase inhibitor that is being tested in preclinical and clinical studies, including for the treatment of pediatric medulloblastoma, which has a propensity for leptomeningeal spread and diffuse midline glioma, which can infiltrate into supratentorial brain regions. In this study, we examined the rate, extent, and spatial heterogeneity of panobinostat CNS distribution in mice. Transporter-deficient mouse studies show that panobinostat is a dual substrate of P-glycoprotein (P-gp) and breast cancer resistant protein (Bcrp), which are major efflux transporters expressed at the BBB. The CNS delivery of panobinostat was moderately limited by P-gp and Bcrp, and the unbound tissue-to-plasma partition coefficient of panobinostat was 0.32 and 0.21 in the brain and spinal cord in wild-type mice. In addition, following intravenous administration, panobinostat demonstrated heterogeneous distribution among brain regions, indicating that its efficacy would be influenced by tumor location or the presence and extent of leptomeningeal spread. Simulation using a compartmental BBB model suggests inadequate exposure of free panobinostat in the brain following a recommended oral dosing regimen in patients. Therefore, alternative approaches to CNS delivery may be necessary to have adequate exposure of free panobinostat for the treatment of a broad range of pediatric brain tumors.

SIGNIFICANCE STATEMENT

This study shows that the central nervous system (CNS) penetration of panobinostat is limited by P-gp and Bcrp, and its efficacy may be limited by inadequate distribution to the tumor. Panobinostat has heterogeneous distribution into various brain regions, indicating that its efficacy might depend on the anatomical location of the tumors. These distributional parameters in the mouse CNS can inform both preclinical and clinical trial study design and may guide treatment for these devastating brain tumors in children.

Introduction

Brain tumors are the leading cause of childhood cancer-related death (Siegel et al., 2020). Despite recent advances in the development of efficacious chemotherapeutics for peripheral cancers, the prognosis for central nervous system (CNS) tumors remains poor. When used in support of surgery and radiation therapy, chemotherapy has the potential to be beneficial for patients with malignant or metastatic CNS tumors (Buckner et al., 2007; Tan et al., 2020). Although various compounds have been shown to kill cancer cells efficaciously in vitro, many of them have failed to exert robust efficacy in vivo following systemic administration. The lack of in vivo efficacy for systemically administered drugs is often due to limited CNS penetration of the active moieties, where the blood-brain barrier (BBB) plays a primary role. The BBB is a dynamic and highly selective barrier system characterized by tight junctions and transport systems that simultaneously protect the brain from blood-borne toxins and present a major challenge to effective CNS drug delivery. A crucial component of CNS drug development is examining the CNS exposure of the active compound. When using preclinical models, concentrations of chemotherapeutics in the brain and spinal cord can be directly measured for the evaluation of CNS penetration and aid in predicting the CNS delivery in humans (Zhou et al., 2007) and motivates our work here.

These studies focus on the epigenetic modifying drug panobinostat. Epigenetic deregulation, in addition to genetic mutations, is a common mechanism in cancer development (Jones and Baylin, 2002; Lund and van Lohuizen, 2004). Aberrant alterations of epigenetic regulators are prevalent in CNS tumors (Gräff et al., 2011). The involvement of histone deacetylases (HDACs) in tumorigenesis has been widely recognized and explored (Glozak and Seto, 2007). For instance, HDAC5 and HDAC9 have been found to be highly expressed in medulloblastoma patients with a poor therapeutic outcome (Milde et al., 2010). HDAC inhibition can regulate gene transcription, cause cell-cycle arrest, and/or induce apoptosis of transformed cells. HDAC inhibitors can change gene expression without altering the DNA sequence and play a crucial role by interfering with the cellular processes that underly tumor initiation and progression (Fig. 1A). Panobinostat is a potent pan-inhibitor of HDACs with IC₅₀ values for enzyme inhibition in the low nanomolar range (Atadja, 2009). The IC₅₀ values of panobinostat in medulloblastoma and diffuse midline glioma (DMG) cell lines have been reported to be in the low nanomolar range, and data are summarized in Table 1 (Milde et al., 2012; Grasso et al., 2015; Pei et al., 2016; Hennika et al., 2017; Phi et al., 2017; Lin et al., 2019; Pak et al., 2019).

Fig. 1.

(A) Mechanisms of action of HDAC inhibitors and (B) CNS delivery of panobinostat following systemic administration and the blood-brain barrier limiting its CNS penetration. Figure created with BioRender.com.

TABLE 1.

Reported IC₅₀ values of panobinostat in medulloblastoma and DMG cell lines

	Cell line	IC50 (nM)	Reference
Medulloblastoma	Group 3	6.5	Milde et al., 2012
	MYC-driven group 3	8–10	Pei et al., 2016
	SHH	54–67	Phi et al., 2017
	Group 3	46
	SHH	8.2	Pak et al., 2019
DMG	DIPG	50–100	Grasso et al., 2015
	H3.3-K27M DIPG	12–28	Hennika et al., 2017
	H3.3 wild-type DIPG	18
	Spinal cord glioma	29.4	Lin et al., 2020
	Thalamic glioma	40.7

DIPG, diffuse intrinsic pontine glioma.

Despite robust in vitro potency, in vivo efficacy of panobinostat has not been routinely observed following systemic administration. A preclinical study targeting diffuse intrinsic pontine glioma showed that panobinostat, although potently killing diffuse intrinsic pontine glioma cells in vitro, did not provide overall survival benefit in mice bearing orthotopic xenografts following intraperitoneal injections (Hennika et al., 2017). Oral panobinostat, in combination with bevacizumab, did not increase overall survival for patients with recurrent glioblastoma when compared with historical controls in a phase II study (Lee et al., 2015). This lack of an in vitro–in vivo correlation may suggest insufficient brain penetration, and currently, there is controversy surrounding the CNS penetration of panobinostat. Some studies have suggested that panobinostat is BBB penetrable and able to reach effective concentrations in the brain, whereas others propose that panobinostat has limited BBB penetration. Table 2 shows results from multiple studies, where total panobinostat concentrations were determined and compared with total IC₅₀ values in some studies. It is noteworthy that the broad conclusions regarding panobinostat CNS penetration seem to be species and/or biomarker related (Grasso et al., 2015; Rasmussen et al., 2015; Chopra et al., 2016; Hennika et al., 2017; Goldberg et al., 2020; Guntner et al., 2020; Rodgers et al., 2020; Homan et al., 2021). Panobinostat has been regarded as BBB penetrable to some extent using an indirect measure, i.e., changes of biomarkers such as histone H3 acetylation (Hennika et al., 2017), or directly by measuring total (bound and unbound) brain concentrations in the mouse brain (Chopra et al., 2016; Homan et al., 2021). In nonhuman primates and patients, cerebrospinal fluid (CSF) concentrations were used as a surrogate approach to evaluate the CNS delivery of panobinostat. The reliability of using CSF concentration as a surrogate for the prediction of BBB penetration can be compromised when the compound is a P-glycoprotein (P-gp) and/or breast cancer resistant protein (Bcrp) substrate (Kodaira et al., 2011). Therefore, investigating the influence of efflux transporters on the CNS delivery of panobinostat can help interpret these data. Also, a quantitative examination of panobinostat CNS penetration is necessary to predict its CNS distributional kinetics, and this information is critical in evaluating and optimizing its usage in a variety of brain tumors and in the future development of HDAC inhibitors with similar physicochemical properties.

TABLE 2.

Panobinostat CNS delivery data from literature

Reference	Species	Target Disease/Disease Status	CNS Delivery
			Dose/route	Measurement
Grasso et al., 2015	Mice	DIPG/healthy	20 mg/kg, i.p.	196 nM in pontine tissue at 30 min postdose (IC50 50–100 nM)
Chopra et al., 2016	Mice	Huntington disease/healthy	10 mg/kg, i.p.	110 nM in brain at 2 h postdose
Hennika et al., 2017	Mice	DIPG/genetic brainstem glioma model	20 mg/kg, i.p.	70 nM in cerebral cortex at 4 h post last dose (IC50 12–28 nM)
Homan et al., 2021	Mice	DIPG/healthy	15 mg/kg, i.v.	Kp,brain = 2.22
Rodgers et al., 2020	Nonhuman primate	DIPG/healthy	1–3 mg/kg, p.o.	CSF concentrations measured at 30 min to 4 h; below LOQ (LOQ 1.43 nM)
Rasmussen et al., 2015	Human	HIV	20 mg M.W.F every other week, p.o.	CSF concentrations measured 8 h postdose; below LOQ (LOQ 0.3 nM)
Guntner et al., 2020	Human	Ependymoma	0.35 mg/kg t.i.w., p.o.	CSF concentrations measured 9 days post first dose; below LOD (LOD 0.43 nM)
Goldberg et al., 2020	Human	Leukemia	24–34 mg/m2 per day t.i.w., p.o.	CSF concentrations measured on day 29; below LOQ (LOQ 0.3 nM)

DIPG, diffuse intrinsic pontine glioma; K_p,brain, brain-to-plasma partition coefficient; LOD, limit of detection; LOQ, limit of quantification; M.W.F., Monday, Wednesday, Friday; t.i.w., three times a week.

The objective of the current study was to evaluate the CNS delivery of panobinostat following systemic administration in the mouse and examine the influence that major efflux transporters at the BBB, P-gp, and Bcrp have on the CNS distribution of panobinostat (Fig. 1B). In addition, the evaluation of drug distribution into different anatomic regions of the brain and spinal cord is important for exploring the application of panobinostat and the future development of methods to target its delivery to tumors that can be found in different CNS locations. These data, taken as a whole, can inform both preclinical and clinical trial study design and may eventually help guide treatment of these devastating brain tumors in children.

Materials and Methods

Chemicals and Reagents

Panobinostat was purchased from Selleck Chemicals (Houston, TX). Panobinostat-d8 hydrochloride salt was purchased from Toronto Research Chemicals (Toronto, ON, Canada). PEG300 was obtained from MedChemExpress (Monmouth Junction, NJ). Sodium heparin was from Meitheal Pharmaceuticals (Chicago, IL). PBS (1×, pH 7.4) and Dulbecco’s modified Eagle’s medium (DMEM) were from Gibco (Fisher Scientific, Waltham, MA). Pooled human plasma was purchased from Innovative Research, Inc (Novi, MI). All other chemicals and reagents were high-performance liquid chromatography–grade or analytical grade and obtained from Thermo Fisher Scientific (Waltham, MA) and MilliporeSigma (Burlington, MA).

Animals

Friend leukemia virus strain B wild-type and transgenic mice lacking either or both efflux transporters P-gp and/or Bcrp, including Mdr1a/b^−/− [P-gp knockout (PKO)], Bcrp1^−/− [Bcrp knockout (BKO)], and Mdr1a/b^−/−Bcrp1^−/− [triple knockout of P-gp and Bcrp (TKO)] mice, were used for all in vivo and in vitro studies that involved animals. All mice were between 8 and 16 weeks old when used. An equal number of male and female mice were allocated to each time point for pharmacokinetic studies and the spatial distribution study. Animals were sourced from Taconic Biosciences, Inc. (Germantown, NY), and colonies have been maintained in the Research Animal Resources facility located at the University of Minnesota following an established breeding agreement. Animal genotypes were routinely verified by genotyping following tail snip (TransnetYX, Inc., Cordova, TN). Animals were maintained on a 12-hour light/dark cycle with ad lib access to food and water. Experiments have been approved by the University of Minnesota Institutional Animal Care and Use Committee and performed in accordance with the Guide for the Care and Use of Laboratory Animals established by the US National Institutes of Health (Bethesda, MD).

Free Fraction in Mouse Plasma, Brain, Spinal Cord, and Media

Free fractions of panobinostat were determined in mouse plasma, mouse brain, mouse spinal cord, media and human plasma by ultrafiltration. Centrifree ultrafiltration centrifugal filters (30 kDa molecular weight cutoff) were purchased from MilliporeSigma. Mouse plasma, brain, and spinal cord were collected in-house. DMEM was supplemented with 10% fetal bovine serum. Ultrafiltration was used as the separation technique to avoid the long incubations at 37°C that equilibrium dialysis would entail. It has been shown that panobinostat can form its carboxylic acid metabolite in mouse plasma when incubated at 37°C (https://www.accessdata.fda.gov/drugsatfda_docs/nda/2015/205353Orig1s000PharmR.pdf). Therefore, to limit this metabolic conversion, the binding experiments were conducted using ultrafiltration at room temperature. In vitro stability experiments indicated that less than 5% of panobinostat in the plasma was converted during the rapid ultrafiltration procedure. Furthermore, panobinostat in the ultrafiltrate was stable since no enzyme is present in the filtrate. Panobinostat was spiked into plasma at 1 μΜ. The final DMSO concentration was 0.035%. After incubation at 25°C with agitation for 5 minutes, 100 μl plasma was sampled from each replicate (n = 5) to determine total concentration. Then, 1 ml of spiked plasma was loaded into each Centrifree reservoir. The devices were centrifuged at 3500 rpm for 5 minutes at 25°C using a Clay Adams Triac centrifuge. Plasma ultrafiltrate was collected from the filtrate cup for the determination of free concentration. Plasma was replaced by PBS to measure nonspecific binding (NSB) and replaced by DMEM to determine the binding in media. For brain protein-binding measurement, brain homogenates were freshly prepared in three volumes of PBS using an Omni THb homogenizer on the day of the experiment. Brain homogenates were spiked with panobinostat and incubated at 25°C with agitation for 5 minutes. After 100 μl homogenate was sampled from each replicate to determine the total concentration, homogenates were centrifuged at 12,000 rpm for 15 minutes at 25°C. One ml of the supernatant was taken and loaded into each reservoir. The devices were centrifuged at 3500 rpm for 10 minutes at 25°C. Ultrafiltrate was collected from the filtrate cup for the analysis of free concentration. The measurement of spinal cord unbound fraction used the same assay procedure as that for the brain.

The NSB to the filtration device was calculated according to a reported equation (Lee et al., 2003):

where C_total,PBS and C_filtrate,PBS are panobinostat concentrations in PBS before centrifugation and in the ultrafiltrate, respectively. NSB was used for the correction of unbound fraction in plasma and CNS tissue homogenates using the following equation previously reported (Lee et al., 2003):

where F_u is the unbound fraction, and C_free is panobinostat concentration in the ultrafiltrate. C_total is panobinostat concentration before centrifugation for plasma and media and before the first centrifugation for brain and spinal cord homogenate. The unbound fractions of panobinostat in the brain and spinal cord were further adjusted following the equation previously reported (Kalvass and Maurer, 2002):

where F_u is the unbound fraction calculated using eq. 2 for homogenates, and D is the dilution factor.

Systemic Pharmacokinetics and CNS Distribution of Panobinostat following Intravenous Administration and the Influence of Efflux Transporters

In previous mouse studies, Imai et al. (2016) found body weight loss in mice treated with panobinostat and set 20 mg/kg as the maximum tolerated dose. However, Hennika et al. (2017) has observed significant toxicity following daily treatment of 20 mg/kg panobinostat and de-escalated to 10 mg/kg. Therefore, we dosed the mice with 10 mg/kg. Panobinostat was formulated with 0.8% DMSO, 0.8% Tween80, and 19.2% PEG300 in water for injection, yielding a final drug concentration of 2 mg/ml. Dosing solutions were freshly prepared on the day of experiment. A single dose of 10 mg/kg was administered as an intravenous bolus via tail vein. CNS distribution of panobinostat and the influence of efflux transporters were studied in wild-type, PKO, BKO, and TKO mice. Samples were collected at 10 minutes, 30 minutes, 1 hour, 4 hours, 8 hours, 16 hours, and 24 hours after dose administration. Blood, brain, and spinal cord were harvested at each time point (n = 4 at each time point). Blood collected via cardiac puncture was immediately centrifuged at 7500 rpm for 10 minutes at 4°C to separate plasma. Spinal cords were obtained via hydraulic extrusion with saline (Richner et al., 2017). The blood and meninges on the surface of the brain and spinal cord were carefully rinsed with saline and removed using Kimwipes. Plasma, brain, and spinal cord samples were snap frozen using dry ice upon collection and kept at −80°C until analysis using liquid chromatography–tandem mass spectrometry. Brain concentrations were adjusted by subtracting residual drug in the brain vasculature, assuming a residual plasma volume of 1.4% in the mouse brain (Dai et al., 2003). Spinal cord concentrations were not corrected because of the lack of information regarding the vascular volume.

Regional Distribution of Panobinostat following Intravenous Administration

Regional distribution of panobinostat in the mouse CNS tissues was evaluated following an intravenous bolus dose of 10 mg/kg in wild-type mice. The brain was removed and dissected into six regions, including cortex, thalamus and hypothalamus, midbrain, pons, cerebellum, and medulla. Different regions were distinguished based on their anatomic structure and location. The spinal cord was divided into three regions, the cervical, thoracic, and lumbosacral regions, indicated by the cervical and lumbosacral enlargements (Roberts et al., 2005). The locations of different brain and spinal cord regions are depicted in Supplemental Fig. 1. Plasma, brain, and spinal cord regions were collected at 30 minutes and 2 and 4 hours after intravenous administration (n = 4 at each time point).

Liquid Chromatography–Tandem Mass Spectrometry Analysis to Determine Panobinostat Concentrations

Brain and spinal cord were homogenized in three volumes of 5% bovine serum albumin (w/v) prior to extraction. Brain samples were prepared using the Omni THb homogenizer, and the spinal cord samples were homogenized using a Kimble Pellet Pestle cordless motor (DWK Life Sciences, LLC., Millville, NJ), and the homogenates were kept on dry ice while on the benchtop. Plasma samples were used as-is for extraction. Panobinostat and the internal standard, panobinostat-d8, were extracted from matrices via liquid-liquid extraction by adding one volume of pH 12 buffer (sodium bicarbonate–sodium hydroxide buffer solution) and five volumes of ethyl acetate. Samples were then vortexed for 5 minutes followed by centrifugation at 14,000 rpm for 5 minutes at 4°C. The organic phase supernatant was collected and dried under a gentle stream of nitrogen gas followed by reconstitution with mobile phase prior to injection. A reversed-phase liquid chromatography method on an Acquity ultraperformance liquid chromatography system (Waters, Milford, MA) coupled with the Quattro Ultima triple quadrupole mass spectrometer (Waters) was used for the analysis of panobinostat specimens from the pharmacokinetic studies in wild-type and PKO mice and regional distribution study. An isocratic method was developed using a YMC-PACK ODS-AM column (35$\times$2.0 mm, S-3 μm, 12 nm; YMC, Inc.). The mobile phase consisted of 80% distilled water with 0.1% formic acid and 20% acetonitrile with 0.1% formic acid, and the flow rate was 0.5 ml per minute. The mass to charge (m/z) ratios were 350.04 > 157.91 and 357.92 > 165.04 for panobinostat and panobinostat-d8, respectively (positive ionization mode). An UltiMate 3000 system (Thermo Fisher Scientific) paired with a TSQ Vantage triple quadrupole (Thermo Finnigan, San Jose, CA) was used for the analysis of samples from pharmacokinetic studies in BKO and TKO mice. The gradient method employed was at 0.5 ml per minute and started at 82% distilled water with 0.1% formic acid and 18% acetonitrile with 0.1% formic acid, increased to 90% organic phase over 3 minutes, maintained for 3 minutes, and brought back to 18% and equilibrate for 2 minutes. The m/z ratios were 350.27 > 158.08 and 358.32 > 164.1 for panobinostat and panobinostat-d8, respectively (positive ionization mode). The limit of quantification was 28.6 nM for all matrices. The standard calibration curve was linear within the range of 28.6–2862 nM, with the linearity assessed by weighted linear regression (1/Y²). Standards and quality controls were prepared on the same day of the analysis. Quality controls were in duplicates. The calibration curve and quality controls needed to have a coefficient of variation of less than 15% for the analysis to be accepted.

Pharmacokinetic Modeling and Simulation

An open two-compartment model with first-order absorption (Supplemental Fig. 2A) was fit to the panobinostat total plasma concentration-time profile following a single oral dose from the clinical study CLBH589B2102 (NCT00621244) (data digitized) using SAAM $\text{ΙΙ}$ (version 2.3; The Epsilon Group, Charlottesville, VA). Key parameters estimated by the two-compartment model, including the clearance of total panobinostat, volume of distribution of total panobinostat, and transfer rate constants between compartments, were then used as a forcing function in a compartmental BBB model, created using STELLA (iSEE systems, Lebanon, NH). This allowed the population prediction of typical patient plasma and brain panobinostat concentrations, both total and free, following recommended oral dosing regimens. The compartmental BBB model consists of a central compartment, brain compartment, and “other” tissues (peripheral) compartment. This model has the following assumptions: 1) that there is instantaneous equilibrium between bound and unbound panobinostat in both plasma and brain, 2) that only free panobinostat transfers between the central and the brain compartment, and 3) that the unbound tissue-to-plasma partition coefficient (K_p,uu) of panobinostat is the same between the human and the mouse.

Pharmacokinetic Data Analysis

Pharmacokinetic parameters from the concentration-time profiles in plasma, brain, and spinal cord were obtained by noncompartmental analysis performed using Phoenix WinNonlin version 8.3 (Certara USA, Inc., Princeton, NJ). Areas under the curve were calculated by the linear trapezoidal method. The equation for pharmacokinetic parameters including half-life (t_1/2), clearance (CL), and steady-state volume of distribution (V_ss) are as follows:

where ${\lambda }_z$ is the first-order rate constant associated with the terminal portion of the curve, and MRT_inf is the mean residence time.

The brain or spinal cord tissue-to-plasma partition coefficients (K_p) and K_p,uu were computed using area under the curve from time 0 to infinity (AUC_inf) and the following:

The free distribution advantage (DA_free) used to compare the relative magnitude of the panobinostat exposure in the brain and spinal cord between the transporter knockout (KO) mice (BKO, PKO, and TKO) and wild-type mice was determined as follows:

The brain-to-plasma (B/P) or spinal cord-to-plasma (SC/P) concentration ratios were the ratios of panobinostat concentration in CNS tissues to panobinostat plasma concentration at each time point.

Statistical Analysis

Data are presented as mean ± S.D. in figures generated using the GraphPad Prism software (Version 9; GraphPad software, La Jolla, CA). The AUC_inf is presented as mean ± S.D., determined by the Yuan method, an extension of the Bailer method (Bailer, 1988; Yuan, 1993). The standard errors of K_p values were calculated following propagation of error for division (You et al., 2013). Statistical tests were performed using GraphPad Prism software. Unbound fractions between matrices were compared by a two-tailed t test using the nonparametric Mann-Whitney U test. A one-way ANOVA followed by Tukey’s post hoc test was used for the comparisons between multiple groups. A significance level of $P<0.05$ was used for all statistical analyses.

Results

Free Fraction of Panobinostat in Plasma, Brain, and Spinal Cord

Free (unbound) fractions of panobinostat in plasma, brain, spinal cord, and media are listed in Table 3. The NSB to the ultrafiltration device was −1.4% ± 1.0%. A two-tailed one-sample t test comparing the NSB average with the hypothetical mean, zero, was conducted, and no significance was reported (P = 0.1411). Therefore, we regarded NSB as zero for the calculation of free fractions in matrices (eq. 1). Unbound fractions of panobinostat in the brain and spinal cord were not statistically different (P = 0.0556). Panobinostat has a higher fraction unbound in mouse plasma than CNS tissues, with approximately 4.2-fold and 3.6-fold differences when compared with the brain and spinal cord, respectively. The unbound fraction of panobinostat was significantly higher in the media compared with both human and mouse plasma (P = 0.0079 and 0.0079 for human and mouse plasma, respectively).

TABLE 3.

Unbound fractions measured by ultrafiltration (n = 5, mean ± S.D.)

	% Unbound
Human plasma	9.36 ± 0.45
Mouse plasma	14.68 ± 0.42
Mouse brain	3.47 ± 0.23
Mouse spinal cord	4.05 ± 0.38
DMEM media	69.49 ± 1.07

Systemic Pharmacokinetics and CNS Distribution of Panobinostat

The plasma, brain, and spinal cord concentration-time profiles of panobinostat in wild-type mice following an intravenous bolus dose of 10 mg/kg are depicted in Fig. 2. There was no sex difference observed in our pharmacokinetic studies. Panobinostat showed similar distributional kinetics in the brain and spinal cord. The B/P and SC/P concentration ratios were initially limited but increased with time and reached a plateau after 8 hours (Fig. 3). Table 4 summarizes the major pharmacokinetic parameters derived from the concentration-time data. The V_ss indicated that panobinostat was extensively distributed into extravascular compartments of the body upon systemic administration. The total tissue exposure in the brain and spinal cord was estimated by AUC_inf, and the resulting distributional partition coefficients (AUC_tissue-to-AUC_plasma; K_p) can be used to assess the CNS exposure. A brain-to-plasma partition coefficient of 1.36 could lead one to a conclusion that panobinostat had significant penetration across the BBB in wild-type mice. However, the high brain-to-plasma partition coefficient can be misleading and overestimate the active drug exposure in the brain. Since the free drug concentration is the driving force for distribution, and it is the concentration of free molecules at the target site that determines the in vivo efficacy (Smith et al., 2010), K_p,uu evaluates the extent of BBB penetration of active moieties and is a more appropriate metric to describe the extent of CNS delivery. The K_p,uu of panobinostat is 0.32 and 0.21 in the brain and spinal cord of wild-type mice, respectively. Usually, in a high-throughput application and interpretation of the exposure metric K_p,uu, a compound with K_p,uu larger than 0.3 to 0.5 is generally classified as CNS “penetrable” (Loryan et al., 2022). But this numerical cutoff of penetration, by itself, does not take into account the potency of the compound. Therefore, we compared the CNS concentrations with IC₅₀ values to predict potential efficacy.

Fig. 2.

Concentration-time profiles of panobinostat in the (A) plasma, (B) brain, and (C) spinal cord following a 10 mg/kg intravenous bolus dose in Friend leukemia virus strain B (FVB) wild-type, PKO (Mdr1a/b^−/−), BKO (Bcrp1^−/−), and TKO (Mdr1a/b^−/−Bcrp1^−/−) mice (n = 4, mean ± S.D.). WT, wild type.

Fig. 3.

Panobinostat (A) brain-to-plasma and (B) spinal cord-to-plasma concentration ratio-time profiles following a 10 mg/kg intravenous bolus dose in Friend leukemia virus strain B (FVB) wild-type, PKO (Mdr1a/b^−/−), BKO (Bcrp1^−/−), and TKO (Mdr1a/b^−/−Bcrp1^−/−) mice (n = 4, mean ± S.D.). WT, wild type.

Table 4.

The pharmacokinetic parameters of panobinostat in FVB wild-type, PKO (Mdr1a/b^−/−), BKO (Bcrp1^−/−), and TKO (Mdr1a/b^−/−Bcrp1^−/−) mice after the administration of a single intravenous dose of 10 mg/kg Standard error of K_p is calculated via propagation of error.

Parameter	Unit	WT			TKO			PKO			BKO
		Plasma	Brain	Spinal Cord	Plasma	Brain	Spinal Cord	Plasma	Brain	Spinal Cord	Plasma	Brain	Spinal Cord
Dose	mg/kg	10			10			10			10
t1/2	h	13.6	21.9	18.1	13.5	12.5	8.8	10.4	11.7	9.1	N.D.a	10.4	11.3
CL	L/h per kg	9	—	—	6.6	—	—	9.3	—	—	10.6	—	—
Vss	L/kg	57.5	—	—	42.8	—	—	50.3	—	—	17.6	—	—
AUCinf	ng*h/mL	1113 ± 87	1515 ± 45	848 ± 41	1519 ± 59	6562 ± 162	4765 ± 159	1073 ± 40	3770 ± 112	2286 ± 54	943 ± 44	892 ± 19	569 ± 20
AUCinf % extrapolated		8.6	48.6	38.0	9.0	24.6	13.3	6.5	23.1	15.1	5.3	33.5	59.6
Kp		—	1.36 ± 0.11	0.76 ± 0.07	—	4.32 ± 0.20	3.14 ± 0.16	—	3.51 ± 0.17	2.13 ± 0.09	—	0.95 ± 0.05	0.6 ± 0.04
Kp,uu		—	0.32	0.21	—	1.02	0.87	—	0.83	0.59	—	0.22	0.17
DAfree			1	1		3.17	4.12		2.58	2.80		0.69	0.79

CL, clearance; N.D., not determined; t_1/2, half-life.

^aPlasma half-life of panobinostat in BKO mice was not determined because of the lack of concentration information to characterize the terminal phase.

The reported IC₅₀ values for different targeted tumors range from 6.5 to 100 nM (Table 1). Therefore, 6.5 nM was used as a hypothetical minimum effective concentration required for panobinostat in vivo efficacy. Figure 4A shows that panobinostat total brain and spinal cord concentrations were above the minimum total IC₅₀ at all times for at least 24 hours after a 10 mg/kg i.v. dose in the mouse. These data might suggest a conclusion that adequate exposure of panobinostat was achieved to be efficacious. However, the situation changes when considering the nonspecific binding of the drug in the target tissue. When total concentrations were adjusted to free concentrations using average unbound fractions in tissues and media (Fig. 4B), panobinostat free brain concentrations exceeded the minimum free IC₅₀ until approximately 4 hours postdose. At this time point, plasma concentrations were quite high, whereas spinal cord concentrations barely crossed the threshold of the minimum free IC₅₀. This indicates that exposure of active panobinostat following this dose is not likely adequate to be efficacious against tumors that are in the spinal cord.

Fig. 4.

Panobinostat (A) total concentrations and (B) free concentrations in CNS tissues compared with the reported minimum IC₅₀ (n = 4, mean $\pm$ S.D.). The dash lines represent the expected minimum (A) total and (B) free IC₅₀. The minimum free IC₅₀ was converted from total IC₅₀ using the average unbound fraction measured in media.

Influence of P-gp and Bcrp on Panobinostat CNS Distribution

The plasma, brain and spinal cord concentration-time profiles and B/P and SC/P concentration ratio-time profiles of panobinostat in transporter deficient mice (TKO, PKO, BKO) following an intravenous bolus dose of 10 mg/kg are also depicted in Figs. 2 and 3. Studies in all four genotypes have been carried out up to 24 hours. Data are not shown on graphs when concentrations were below the detection limit. The B/P and SC/P concentration ratios reached a plateau after 4 to 8 hours. The systemic (plasma) exposure was not affected by the deficiency of efflux transporters as indicated by the superimposable plasma concentration-time profiles in wild-type, TKO, BKO, and PKO mice (Fig. 2A). Conversely, the brain and spinal cord concentrations vary among the four genotypes (Fig. 2, B and C). The partitioning of unbound panobinostat in the brain (K_p,uu,brain) and in the spinal cord was higher in PKO mice compared with wild-type mice (2.6-fold and 2.8-fold, respectively) (Table 4), whereas the plasma exposure was not statistically different between the two genotypes. These results indicate that P-gp has a moderate influence in limiting the brain penetration of panobinostat. The brain concentrations in the TKO mice were the highest among the four genotypes at all measured time points, and this trend was also observed in the spinal cord. The K_p,uu,brain in TKO mice was higher than that in PKO mice, with an approximately 1.2-fold difference between K_p,uu,brain values and a 1.5-fold difference between the unbound spinal cord-to-plasma partition coefficient values, suggesting that Bcrp may also be restricting panobinostat BBB penetration. And the higher-than-additive increase of DA_free suggests that P-gp and Bcrp have compensatory effects regarding efflux (Kodaira et al., 2010; Talele et al., 2022b). The brain and spinal cord free distribution advantages in BKO mice were less than those in PKO mice, suggesting that P-gp efflux is the primary mechanism limiting the CNS delivery of panobinostat and that P-gp has a significant compensatory effect when Bcrp is lacking. The distribution advantages caused by the deficiency of efflux transporters indicates that P-gp and Bcrp have a moderate effect in limiting the CNS penetration of panobinostat, with P-gp being the dominant restrictive factor.

Regional Distribution of Panobinostat following Intravenous Administration

Brain and spinal cord regions were dissected upon collection before snap freezing on dry ice. An illustration of different regions is depicted in Supplemental Fig. 1. Medulla and cervical regions were pooled from four mice at each time point due to limited weight of a specific specimen from each mouse, and therefore, these regions were not included in statistical tests. Figure 5A shows panobinostat concentrations in different brain regions at 30 minutes after a single bolus intravenous 10 mg/kg dose in wild-type mice. Panobinostat showed heterogeneous distribution among brain regions when concentrations were compared using one-way ANOVA followed by Tukey’s post hoc test (P < 0.05). When compared with whole brain and spinal cord concentrations obtained from the systemic pharmacokinetic study (Fig. 2, B and C), the rank order of panobinostat average concentrations in spatially distinct CNS regions is thalamus-hypothalamus > pons > cerebellum > whole brain > midbrain. Similar trends, i.e., heterogeneous distribution among brain regions, maintained at 2 and 4 hours postinjection and concentrations in all regions were higher than the minimum total IC₅₀ (Fig. 5, B and C). As shown in Fig. 6, concentrations in the thoracic to lumbosacral spinal cord regions were not statistically different from each other at all three time points when compared using a two-tailed t test (P > 0.05). Panobinostat free concentrations, however, showed that effective unbound concentrations were not reached in all brain and spinal cord regions (Figs. 7 and 8). The heterogeneous spatial distribution of panobinostat in the brain upon systemic administration informs that its in vivo efficacy can depend on the anatomic location of the tumor cells. In addition, free panobinostat concentrations in the thoracic and lumbosacral regions were below the minimum free IC₅₀, indicating that current dosing, 10 mg/kg i.v., is not likely to exert efficacy on tumors located in these spinal cord regions.

Fig. 5.

Spatial distribution of panobinostat into brain regions at (A) 30 minutes, (B) 2 hours, and (C) 4 hours following a 10 mg/kg intravenous dose in Friend leukemia virus strain B (FVB) wild-type mice (n = 4, mean ± S.D., except that the medulla regions were pooled from four mice). The dash lines represent the expected minimum total IC₅₀. *P ≤ 0.05; **P ≤ 0.01; ***P$\le$ 0.001; ****P ≤ 0.0001. Statistical analysis performed by one-way ANOVA followed by Tukey test. The heatmaps show the average total panobinostat delivery to each brain region.

Fig. 6.

Spatial distribution of panobinostat into spinal cord regions at (A) 30 minutes, (B) 2 hours, and (C) 4 hours following a 10 mg/kg intravenous dose in Friend leukemia virus strain B (FVB) wild-type mice (n = 4, mean ± S.D., except that the cervical regions were pooled from four mice and excluded from the statistical test). The dash lines represent the expected minimum total IC₅₀. Statistical analysis performed by two-tailed t test. No statistically significant difference observed. The heatmaps show the average total panobinostat delivery to each spinal cord region.

Fig. 7.

Spatial distribution of unbound panobinostat into brain regions at (A) 30 minutes, (B) 2 hours, and (C) 4 hours following a 10 mg/kg intravenous dose in Friend leukemia virus strain B (FVB) wild-type mice (n = 4, mean ± S.D., except that the medulla regions were pooled from four mice). The dash lines represent the expected minimum free IC₅₀. *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. Statistical analysis performed by one-way ANOVA followed by Tukey test.

Fig. 8.

Spatial distribution of unbound panobinostat into spinal cord regions at (A) 30 minutes, (B) 2 hours, and (C) 4 hours following a 10 mg/kg intravenous dose in Friend leukemia virus strain B (FVB) wild-type mice (n = 4, mean ± S.D., except that the cervical regions were pooled from four mice and excluded from the statistical test). The dash lines represent the expected minimum free IC₅₀. Statistical analysis performed by two-tailed t test. No statistically significant difference observed.

Pharmacokinetic Modeling and Simulation Predicting Human Brain Concentrations

The total panobinostat plasma concentration-time profile (0–48 hours) was digitized (PlotDigitizer) from a graph presented in the CLBH589B2102 clinical study following a single 20 mg oral dose in patients. An oral bioavailability of 21.4% and absorption rate constant of 0.42 hours⁻¹ (Savelieva et al., 2015) were adopted for application in our open two-compartment model (Supplemental Methods; Supplemental Fig. 2A). The open two-compartment model described the panobinostat plasma concentrations in humans following a single oral dose well as shown by the overlaying predictions and observations (the majority of weighted residuals were less than one) (Supplemental Fig. 2B), and the coefficients of variation for all the parameters were less than 20%. The parameters estimated by SAAM $\text{ΙΙ}$ used in the compartmental BBB model are described in Supplemental Table 1. The BBB model consists of three compartments and incorporates passive diffusion between the central and brain compartment as well as active efflux from the brain into the central compartment (Fig. 9A). The input parameters in the STELLA model for panobinostat are listed in Supplemental Table 1. The ability of the model to predict plasma concentrations was confirmed by comparing simulated and observed data (Supplemental Fig. 3A). The simulation used an oral dosing regimen of panobinostat given 20 mg once every other day for three doses per week (Monday, Wednesday, Friday) of weeks 1 and 2 of a 21-day cycle. The simulation showed that panobinostat total brain concentrations were above the minimum total IC₅₀ for about 1 day (Fig. 9B). However, as shown in Fig. 9C, panobinostat free brain concentrations were significantly below the minimum free IC₅₀. In addition, according to previous safety studies, 60 mg oral panobinostat has been determined to be the maximum tolerated dose in patients (DeAngelo et al., 2013). Assuming linear pharmacokinetics, the maximum free brain concentration following the maximum tolerated dose is predicted to be 1.2 nM, which is still lower than the minimum free IC₅₀ (4.5 nM) (Supplemental Fig. 3B). Therefore, using this exhaustive analysis, one can conclude that the current oral multiple dosing regimen is not expected to result in adequate exposure of active panobinostat in the brain for medulloblastoma and DMG.

Fig. 9.

(A) A schematic of the pharmacokinetic model used to simulate the unbound panobinostat brain concentration. (B) Simulated plasma and brain total concentrations for multiple 20 mg oral doses in a 21-day cycle of treatment in humans. (C) Simulated plasma and brain unbound concentrations for multiple 20 mg oral doses in a 21-day cycle of treatment in patients. The dash lines represent the expected minimum (B) total and (C) free IC₅₀.

Discussion

Chemotherapy is an essential part of effective treatment of CNS tumors. The efficacy of chemotherapeutics involves the requirement of potency and adequate exposure at the site of action. Many therapeutic agents are potent at killing CNS tumor cells in vitro, but the development of an effective in vivo chemotherapy is challenging. This is due in large part to the presence of the BBB, which impedes delivery of many therapeutic molecules into the CNS. It is known that tumor growth can compromise the integrity of the BBB, possibly leading to improved therapeutic agent delivery in some areas of the tumor (Arvanitis et al., 2020). However, this BBB disruption is almost always heterogeneous, and as such, the BBB proves to be an important hurdle to overcome during the optimization of the delivery and exposure of potential therapeutics (Sarkaria et al., 2018; Warren, 2018). For instance, among the four subgroups of medulloblastoma, WNT-medulloblastoma has been reported to have a relatively leaky BBB, whereas SHH-, and likely group 3- and group 4-medulloblastoma as well, maintain an intact BBB (Phoenix et al., 2016). DMG often has a largely intact BBB as indicated by limited contrast enhancement on magnetic resonance imaging (Poussaint et al., 2011; Welby et al., 2019). Therefore, given the known potency of panobinostat against medulloblastoma and DMG (Table 1), the ability of panobinostat to cross the BBB is a necessary component of any treatment that will result in efficacy when treating these pediatric CNS tumors.

As a “small” molecule (molecular mass 349.2 g/mol) with intermediate lipophilicity (LogP = 2.64), panobinostat possesses the physicochemical properties that could result in the potential to cross the BBB. In such a case, where passive diffusion could be significant, efflux transporter systems might be a major mechanism limiting its CNS penetration. Panobinostat has been reported to be a P-gp substrate (https://www.accessdata.fda.gov/drugsatfda_docs/label/2015/205353s000lbl.pdf). However, data quantifying the in vivo influence of P-gp is, to our knowledge, unavailable. Therefore, we quantitatively evaluated the CNS delivery and influence of efflux transporters on the CNS penetration of panobinostat in efflux transporter-intact and efflux transporter-deficient mice. Agarwal et al. (2010) have shown that the BBB in TKO mice with the absence of P-gp and Bcrp had comparably intact tight junctions with wild-type mice, indicated by similar sucrose and inulin brain spaces measured in the two genotypes. Therefore, given similar tight junction integrity, the increased panobinostat brain penetration in the TKO mice was due to the lack of P-gp and Bcrp. In addition, our data show that genetic deletion of P-gp and Bcrp lead to approximately three- and fourfold greater exposure of unbound panobinostat in the brain and spinal cord, respectively, indicating that panobinostat is not a particularly avid substrate of P-gp and Bcrp. Therefore, the pharmacological inhibition by efflux transporter inhibitors is not expected to significantly improve the CNS accumulation of panobinostat.

Although both P-gp and Bcrp are limiting the CNS delivery of panobinostat, it has K_p values of 1.36 and 0.76 in the brain and spinal cord, respectively. However, a K_p value close to or even larger than one is not always indicative of good penetration and CNS delivery. In the current study, the total concentration of panobinostat was shown to be higher than the expected minimum total effective concentration for at least 24 hours postdose. However, when examining the free concentrations in the brain and plasma (i.e., the K_p,uu,brain) and comparing with free IC₅₀, as pointed out above, the results indicate that intravenously administered (systemic) panobinostat at 10 mg/kg has sufficient penetration to engage its target in the brain for up to 4 hours postdose, and the duration for maintaining adequate “active” concentration was even shorter in the spinal cord. This discrepancy, illustrated in Fig. 4, underscores the importance of taking into account the relative binding to plasma and CNS components to describe the exposure of active moieties.

How translatable are these findings to patient care? By applying the estimated pharmacokinetic parameters from clinical plasma data and K_p,uu obtained in our mouse model to predict the brain concentration-time profile in patients, we were able to demonstrate that, following multiple 20 mg oral doses, there was inadequate exposure of active panobinostat in the brain to treat brain tumors like medulloblastoma and DMG. The low exposure of free panobinostat in the brain shown in our simulation can help explain the lack of in vitro–in vivo correlation for oral panobinostat in clinical studies for CNS tumors (Lee et al., 2015; Wood et al., 2018; Monje et al., 2022). This delivery issue can be supported by the undetectable level of panobinostat in patients’ CSF following systemic dosing. However, one must recognize that using CSF concentration as a surrogate could overestimate the unbound brain concentration for a P-gp substrate given the apical localization of P-gp and apical-to-basal transportation mediated by P-gp at the choroid plexus (Rao et al., 1999; Sun et al., 2003). Even with a possible overestimation, the panobinostat level is not quantifiable in the CSF, suggesting limited concentration of active panobinostat in the brain. Collectively, these results suggest that current dosing regimens in clinical studies are not enough to be efficacious for brain tumor treatment.

Therefore, given the predictions above and the assumptions involved, methods to increase panobinostat CNS exposure, such as targeted delivery, will be necessary to treat these pediatric CNS tumors with panobinostat. Attempts to improve the CNS delivery of cyclodextrin-solubilized panobinostat (MTX110) via convection-enhanced delivery have yielded efficacy with minimal toxicity in a rat glioma model (Singleton et al., 2017). The survival of glioma-bearing rats was improved by a single infusion of 5 μl of 1.7 μM panobinostat compared with vehicle and nontreatment controls, and safety was demonstrated by lack of neurotoxicity following the infusion of MTX110 into the pons in healthy rats at 1 μl/min for 5 minutes with concentrations up to 30 μM and in healthy pigs with 100 μl infusate at 30 μM Singleton et al., 2018). Our group has explored opportunities to encapsulate panobinostat within polymeric nanoparticle carriers that exhibit controlled release; early data suggest that this may be a useful approach for local delivery to treat orthotopic brain tumors (Chaudhuri et al., 2021), although this does not solve the problem of BBB penetration for the diffuse, infiltrative, or metastatic growth patterns that are often observed for pediatric brain tumors such as medulloblastoma or DMG. Another strategy is intraventricular administration into the fourth ventricle, which has been studied in healthy nonhuman primates (Sandberg et al., 2020). Although being undetectable in the serum post an infusion of 0.5 ml water-soluble panobinostat at 300 μM, panobinostat reached an average peak concentration in the fourth-ventricle CSF at 18 μM, which was significantly higher than that in the lumbar cistern CSF. Therefore, the authors proposed an application of the fourth-ventricular infusion of MTX110 to treat recurrent medulloblastoma. The combination of controlled release with direct-to-CSF administration approaches may offer a future opportunity to better optimize delivery of panobinostat to pediatric tumors exhibiting leptomeningeal spread.

Another essential factor that requires consideration is the location of tumors. CNS tumors can develop in any part of the brain and spinal cord. Some CNS tumors can be confined in certain regions at an early stage of growth; however, others can be widely disseminated. For instance, diffuse midline glioma, as the name suggests, has a midline location, especially in the pons (Louis et al., 2016). Medulloblastomas often arise in the cerebellum with frequent leptomeningeal metastases (Roussel and Hatten, 2011; Garzia et al., 2018). Since in vivo efficacy can be correlated with the concentrations of the therapeutic agent at the target site, we measured panobinostat concentrations in different brain and spinal cord regions in wild-type mice following systemic administration. Unbound concentrations in thoracic and lumbosacral regions of the spinal cord were not statistically different, and both were lower than the minimum free IC₅₀ at the three measured time points. On the contrary, our data showed that panobinostat delivery into various brain regions was heterogeneous. After the 10 mg/kg i.v. dose, panobinostat free concentrations in the thalamus-hypothalamus, pons, and cerebellum were higher than or close to the minimum free IC₅₀ until 4 hours postdose, whereas free concentrations in the cortex and midbrain were lower than the threshold. We previously studied the regional distribution of molecules in the CNS and observed similarly heterogeneous distribution (Cook et al., 2015). Taken together, these data suggest that the efficacy of panobinostat may be influenced by the location of the tumor, even in the context of systemic dosing, and that heterogeneous delivery of circulating molecules to the CNS will depend heavily on specific drug properties or use of a carrier.

A potential contributor to the lack of uniform distribution of panobinostat among brain regions is variable local binding. A difference in the extent of binding, if it exists, might be attributed to heterogeneous expression of proteins and lipids. Lipid composition in rat brain has been found to vary among regions, with the lowest total lipid level measured in the cortex (Chavko et al., 1993). In addition, tissue distribution of weakly basic drugs is correlated with tissue lipid concentrations with a binding preference to acidic lipids (Nishiura et al., 1986; Yata et al., 1990). Considering that panobinostat is a base with intermediate lipophilicity, binding to lipids could possibly contribute to its heterogeneous brain distribution. Besides lipids, protein expression in different mouse brain regions has been evaluated using proteomic analysis, and region-specific proteins have been found (Korovesi et al., 2020). Different affinities to proteins could potentially influence the extent of binding of panobinostat among regions. Alternatively, heterogeneous brain distribution of panobinostat may be a result of various vascular densities in different regions. Xiong et al. (2017) measured the microvascular densities in different mouse brain regions and found high vascular density in the thalamus, which could possibly explain the higher concentration of panobinostat observed in the thalamus in our study. However, the heterogeneous spatial distribution of panobinostat in the brain is not likely due to variable P-gp- and/or Bcrp-mediated efflux among different regions. A spatial distribution study of peposertib, a substrate of P-gp and Bcrp with P-gp being the dominant limiting factor, showed homogeneous distribution among six brain regions, with which our experiment shared the criteria for brain-region differentiation and dissection (Talele et al., 2022b). Also, the brain regional distribution of P-gp substrates AZD1390, an ataxia telangiectasia mutated kinase inhibitor, and alisertib, an aurora A kinase inhibitor, showed no statistically significant differences among brain regions (Talele et al., 2022a). These results suggest that P-gp- and/or Bcrp-mediated efflux should be homogeneous among different brain regions.

In conclusion, we have shown that the CNS delivery of panobinostat is limited by P-gp and, to a lesser extent, Bcrp. Its distribution is heterogeneous among various brain regions upon systemic intravenous administration, indicating that panobinostat efficacy can be influenced by the anatomic location of tumors. Additionally, oral administration of panobinostat following a tolerable dosing regimen in patients may not be sufficient to result in efficacy for brain tumors. Clinically feasible methods of targeted delivery, such as convection-enhanced delivery, intrathecal administration (Fowler et al., 2020), or the utilization of carrier-mediated strategies to bypass the BBB, are worth exploration for the application of panobinostat in pediatric patients with CNS tumors like DMG and medulloblastoma.

Abbreviations

AUC_inf

area under the curve from time zero to infinity

B/P

brain to plasma

BBB

blood-brain barrier

Bcrp

breast cancer resistance protein

BKO

Bcrp knockout

CNS

central nervous system

CSF

cerebrospinal fluid

C_{total, PBS}

concentrations in PBS before centrifugation

D

dilution factor

DA_free

free distribution advantage

DMEM

Dulbecco’s modified Eagle’s medium

DMG

diffuse midline glioma

F_u

unbound fraction

HDAC

histone deacetylase

K_p

tissue-to-plasma partition coefficient

K_{p, uu}

unbound tissue-to-plasma partition coefficient

K_{p, uu, brain}

unbound brain-to-plasma partition coefficient

NSB

nonspecific binding

P-gp

P-glycoprotein

PKO

P-glycoprotein knockout

SC/P

spinal cord-to-plasma

TKO

triple knockout of P-glycoprotein and Bcrp

V_ss

steady-state volume of distribution

Authorship Contributions

Participated in research design: Wenq. Zhang, Oh, Sirianni, Elmquist.

Conducted experiments: Wenq. Zhang, Wenj. Zhang, Rathi.

Performed data analysis: Wenq. Zhang, Oh, Elmquist.

Wrote or contributed to the writing of the manuscript: Wenq. Zhang, Oh, Larson, Wechsler-Reya, Sirianni, Elmquist.