Cerebrospinal fluid penetration of targeted therapeutics in pediatric brain tumor patients

Treatment with small-molecule inhibitors, guided by precision medicine has improved patient outcomes in multiple cancer types. However, these compounds are often not effective against central nervous system (CNS) tumors. The failure of precision medicine approaches for CNS tumors is frequently attributed to the inability of these compounds to cross the blood-brain barrier (BBB), which impedes intratumoral target engagement. This is complicated by the fact that information on CNS penetration in CNS-tumor patients is still very limited. Herein, we evaluated cerebrospinal fluid (CSF) drug penetration, a well-established surrogate for CNS-penetration, in pediatric brain tumor patients. We analyzed 7 different oral anti-cancer drugs and their metabolites by high performance liquid chromatography mass spectrometry (HPLC-MS) in 42 CSF samples obtained via Ommaya reservoirs of 9 different patients. Moreover, we related the resulting data to commonly applied predictors of BBB-penetration including ABCB1 substrate-character, physicochemical properties and in silico algorithms. First, the measured CSF drug concentrations depicted good intra- and interpatient precision. Interestingly, ribociclib, vorinostat and imatinib showed high (> 10 nM), regorafenib and dasatinib moderate (1–10 nM) penetrance. In contrast, panobinostat und nintedanib were not detected. In addition, we identified active metabolites of imatinib and ribociclib. Comparison to well-established BBB-penetrance predictors confirmed low molecular weight, high proportion of free-drug and low ABCB1-mediated efflux as central factors. However, evaluation of diverse in silico algorithms showed poor correlation within our dataset. In summary, our study proves the feasibility of measuring CSF concentration via Ommaya reservoirs thus setting the ground for utilization of this method in future clinical trials. Moreover, we demonstrate CNS presence of certain small-molecule inhibitors and even active metabolites in CSF of CNS-tumor patients and provide a potential guidance for physicochemical and biological factors favoring CNS-penetration.


Detailed Outiline on Clinical Parameters and Sample Collection
In the following Tables (S1 and S2) detailed information on patient therapy and sample collection are given.  were further diluted with methanol to obtain stock solutions containing analyte concentrations of 10 mg L -1 and were stored at 4 °C. The internal standard (ISTD) carbamazepine (> 98 %) was obtained from Sigma-Aldrich Handels GmbH (Vienna, Austria) and used as a methanolic stock solution with a concentration of 2 g L -1 , which was diluted as needed.
Succinanilic acid (p.a.) was purchased from Sigma-Aldrich Handels GmbH (Vienna, Austria) diluted to a 400 mg L -1 methanolic stock solution and stored at 4 °C.
All used solvents were HPLC grade and were purchased from VWR International GmbH (Darmstadt, Germany).
Water was used in Millipore quality obtained from a Millipore purification system (Molsheim, France).
CSF samples (< 0.5 mL) of pediatric patients receiving cancer therapy drugs were collected using an Ommaya reservoir, and were stored at -80 °C prior to further analysis.
Sample preparation included protein precipitation (100 µL serum or CSF + 300 µL methanol containing the internal standard carbamazepine). Although protein levels in CSF are significantly lower compared to serum, precipitation is needed in order to avoid precipitation of proteins within the HPLC system resulting in shortened lifetime of analytical columns and other instrument parts. Proper homogenization of sample material was achieved using a Thermoshaker Thermal Shake lite (1500 rpm, 10 min, VWR International GmbH, Darmstadt, Germany). Subsequently the samples were centrifuged using a Megafuge 1.0 (4000 rcf, 10 min, Heraeus Instruments, Hanau, Germany) and the supernatant was collected in 200 µL HPLC vial inserts.
The separation of analytes and CSF matrix compounds was achieved by the means of high performance liquid chromatography (HPLC) using a 1200 Series HPLC purchased from Agilent Technologies (Santa Clara, California) combined with a C18 Nucleoshell column (150 x 3 mm; 2.7 µm; endcapped silica) from Macherey-Nagel (Düren, Germany) coupled to a 6420 triple quadrupole mass spectrometer (QqQ

MS) or a 6560 Ion-Mobility Quadrupole Time-of-Flight MS (IMS-Q-TOF MS) both from Agilent
Technologies (Santa Clara, California).
High performance liquid chromatographic separation was accomplished using an octadecyl reversed phase column in combination with a gradient mobile phase of methanol and 1 vol.% formic acid. In this context, Figure S1 gives detailed information on the developed chromatographic gradient system.

Figure S1
Depiction of the used gradient system within the multi-method to separate all mentioned analytes as well as matrix components.
In all corresponding analyses, a flow rate of 0.6 mL min -1 at a column compartment temperature of 50 °C and an injection volume of 5 µL was used.
HPLC-QqQ MS hyphenation by the means of positive electrospray ionization was used for quantitational purposes. Hereby, a gas flow of 11 L min -1 at a temperature of 300 °C, a nebulizer pressure of 15 psi and a capillary voltage of 4 kV were used within the source of the Triple Quadrupole instrument for proper ionization of target analytes. Additionally, as part of method development, selected reaction monitoring transitions were investigated for each analyte to ensure sensitivity and selectivity. The most abundant transitions were used as quantifier and at least one lower abundant product ion was used as qualifier. Quantifier/qualifier ion ratios were then used to ensure selectivity.
Data processing was achieved using MassHunter Workstation Software B.05.02 from Agilent Technologies. In this context, Table S3 shows the used dynamic multiple reaction monitoring method for quantitation, including all transitions and corresponding collision energies as well as fragmentor voltages and retention times. Metabolite detection and determination of collision cross sections was performed using an IMS-Q-TOF MS within a mass range of 100 -1700 m/z in QTOF-only and IMS mode respectively. Ionization was again obtained by the means of positive electrospray ionization using 5 L min -1 of nitrogen stream with 300 °C for drying, a capillary voltage of 3.5 kV and a nebulizer pressure of 35 psig. Further parameters included a sheath gas flow of 11 L min -1 at 350 °C and a nozzle voltage of 1 kV. The fragmentor voltage was set to 400 V. Ion mobility experiments were performed with a maximal drift time of 60 ms at a frame rate of 0.9 frames per second resulting in a total of 600 TOF-transients within one frame. Data processing was achieved using IM-MS Browser B.08.00 from Agilent Technologies on the basis of ion mobility feature extraction and database-assisted Qualitative Workflows B.08.00 Software from Agilent Technologies.
The method validation of the HPLC-QqQ MS method included recovery analysis, linearity testing and determination of lower limits of detection. For that purpose, multiple calibrators were used to test the linearity of the method (including sample preparation) in a range of 800 ng L -1 up to 250 µg L -1 . The lower limit of detection was defined as the lowest concentration giving a signal with a signal-to-noise ratio of 3. The lower limit of quantification was defined at a signal-to-noise ratio of 10. Intra-day precision was tested on the basis of 5 consecutive measurements of model samples and inter-day precision was determined as the relative standard deviation on average between 2 quality control sets on individual days.
Overall process efficiency (taking into account recovery and matrix effects) was tested by the addition of spike solutions of pre-determined concentration to blank serum or CSF (10 µL + 90 µL), followed by protein precipitation and HPLC-QqQ MS analysis and comparison of resulting signal intensities with standard solutions of known concentrations. Matrix-matched calibration was eventually used to overcome matrix induced suppression/enhancement effects during electrospray ionization.
The developed HPLC-QqQ MS multi-method to quantitate dasatinib, imatinib, nintedanib, panobinostat, regorafenib, ribociclib and vorinostat in cerebrospinal fluid samples was proven robust, reproducible and reliable over a broad range of analyte concentration levels (800 ng L -1 to 250 µg L -1 ). (1)(2) (3) (4) (5) (6) (7) (8) Figure S2 shows in this context a representative chromatogram of a mixed standard solution. Intra-day and inter-day precision were determined to be ≤ 10 % for all analytes. Additionally the developed HPLC-QqQ MS method showed good linearity (R 2 ≥ 0.99) in a range between the lower limit of detection and 250 µg L -1 . The lower limit of detection (LOD) of the method was determined to be ≤ 0.8 µg L -1 for all analytes, as shown in detail in Table S4.
. In addition to the quantification of the mentioned pharmaceuticals in human sample materials, HPLC-IMS-Q-TOF MS measurements were used in combination with prediction software and in-house generated databases to test for metabolites. In this context, Table 2

Calculation of free drug levels
The free level of a drug may be calculated, as documented in the work of Avery et al. 38 . Accordingly, following mathematical derivation is feasible: reflects hereby the amount of protein within a matrix. , and describe unbound, bound and total drug concentrations, where: including an unbound fraction parameter , which can be calculated. This is possible, because protein concentrations of CSF and serum are available in the literature, but should be determined for individual samples to ensure validity of following predictions. Besides, also a drug's serum protein binding is typically available.
In addition, the therapeutic concentration of drugs is usually significantly lower than the present protein levels within the investigated compartments. Consequently becomes negligible in Equation 4, so can be calculated using serum data.
If is known, the unbound fraction parameter becomes predictable for CSF according to:

Protein levels in patient samples
In the following vorinostat and imatinib levels are shown as a function of the individual (see Figure S4 and S5). In addition, a possible correlation between imatinib levels and the amounts of CSF proteins was tested, as shown in further detail in Figure S6-S7.

Figure S6
Depiction of the correlation between imatinib levels (µg/L) and CSF protein amounts (g/L). 7. Color-coding in Figure 4 of the Manuscript Additional information on the color-coding used in Figure 4 of the manuscript is given in Table S5.