A validated LC-MS/MS assay for quantification of 24(S)-hydroxycholesterol in plasma and cerebrospinal fluid[S]

24(S)-hydroxycholesterol [24(S)-HC] is a cholesterol metabolite that is formed almost exclusively in the brain. The concentrations of 24(S)-HC in cerebrospinal fluid (CSF) and/or plasma might be a sensitive marker of altered cholesterol metabolism in the CNS. A highly sensitive 2D-LC-MS/MS assay was developed for the quantification of 24(S)-HC in human plasma and CSF. In the development of an assay for 24(S)-HC in CSF, significant nonspecific binding of 24(S)-HC was observed and resolved with the addition of 2.5% 2-hydroxypropyl-β-cyclodextrin (HP-β-CD) into CSF samples. The sample preparation consists of liquid-liquid extraction with methyl-tert-butyl ether and derivatization with nicotinic acid. Good linearity was observed in a range from 1 to 200 ng/ml and from 0.025 to 5 ng/ml, for plasma and CSF, respectively. Acceptable precision and accuracy were obtained for concentrations over the calibration curve ranges. Stability of 24(S)-HC was reported under a variety of storage conditions. This method has been successfully applied to support a National Institutes of Health-sponsored clinical trial of HP-β-CD in Niemann-Pick type C1 patients, in which 24(S)-HC is used as a pharmacodynamic biomarker.


INTRODUCTION
In central nervous system (CNS) cholesterol originates almost exclusively from in situ synthesis (1), while circulating cholesterol is normally prevented from entering the CNS by the blood-brain-barrier (2).
As cholesterol cannot be eliminated in CNS, and may be toxic to neurons when in excess, it is secreted from CNS into the circulation predominantly in the form of its polar metabolite 24(S)-hydroxycholesterol (24(S)-HC) (3). 24(S)-HC is formed almost exclusively in the brain. The enzymatic conversion of CNS cholesterol to 24(S)-HC, which readily crosses the blood-brain barrier, is the major pathway to eliminate cholesterol and maintain cholesterol homeostasis in brain tissue. The cholesterol 24-hydroxylase (CYP46A1) mediating this conversion is mainly located in neurons (4). The concentrations of 24(S)-HC in cerebrospinal fluid and/or plasma might be a sensitive marker of increased cholesterol metabolism in CNS. Plasma 24(S)-HC is decreased in Alzheimer disease, vascular dementia, multiple sclerosis, Parkinson's Disease, and Huntington Disease, reflecting disease burden, the loss of metabolically active neurons, and the degree of structural atrophy (5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17). Increased cholesterol turnover (i.e., myelin breakdown or neurodegeneration), which occurs an early stage in these diseases, appears to be associated with a transient increase of 24(S)-HC efflux and higher plasma or CSF 24(S)-HC concentration (18,19).
Here, we report a sensitive and robust LC-MS/MS method with total run time of 7.5 min for determination of free 24(S)-HC in human plasma and CSF involving a liquid-liquid extraction and derivatization into nicotinate. The lower limit of quantification (LLOQ) was found to be 1 and 0.025 ng/mL for plasma and CSF, respectively. The validated method has been successfully applied to support a NIH sponsored clinical trial of HP-β-CD in Niemann-Pick type C1 (NPC1) patients, in which 24(S)-HC was explored as a pharmacodynamic biomarker (38).

Stock solution preparation
All the stock solutions (1 mg/mL) were prepared in methanol. A working solution containing 10 µg/mL of 24(S)-HC was prepared by the dilution of the stock solution with methanol. The internal standard working solutions for plasma (50 ng/mL of D7-24-HC) and CSF (5 ng/mL of D7-24-HC) were prepared in methanol-water (1:1).

Standard curves
Because of the endogenous presence of 24(S)-HC in human plasma and CSF, aqueous solutions of 5% BSA and 2.5% HP-β-CD were used to prepare the calibration standards for plasma and CSF, respectively. Calibration curves were prepared by spiking the 24(S)-HC working solution into 5% BSA and 2.5% HPβ-CD solutions, and preparing serial dilutions that yielded eight calibration standards (1, 2, 5, 10, 20, 50, 100, 200 ng/mL for plasma assay; 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5 ng/mL for CSF assay). 5% BSA and 2.5% HP-β-CD solutions served as blanks. The same calibration standards in plasma and CSF were also prepared and used to assess responsiveness in different matrixes, which was evaluated by parallelism between standard curves prepared in biological matrix (plasma and CSF) and surrogate matrix (5% BSA and 2.5% HP-β-CD in water).

Quality control samples
The pooled-plasma and CSF samples were analyzed to establish the mean concentration of endogenous 24(S)-HC by the LC/MS/MS method. The low (LQC), middle (MQC), high (HQC), dilution (DQC) quality control samples (endogenous level + 0 ng/mL, endogenous level + 75 ng/mL, endogenous level + 150 ng/mL, and endogenous level + 300 ng/mL for human plasma assay; endogenous level + 0 ng/mL, endogenous level + 2 ng/mL, endogenous level + 4 ng/mL, and endogenous level + 8 ng/mL for human by guest, on September 23, 2016 www.jlr.org Downloaded from CSF assay) were prepared by serial dilution after 24(S)-HC working solution was spiked into blank biological matrix. The lower limit of quantification (LLOQ) sample for human plasma (1 ng/mL) and CSF (0.025 ng/mL) were prepared in 5% BSA and 2.5% HP-β-CD solutions, respectively. The 24(S)-HC in DQC sampler was higher than the upper limit of quantification (ULOQ: 200 ng/mL for human plasma; 5 ng/mL for human CSF). The human plasma and CSF DQC samples were diluted 1:4 with 5% BSA and 2.5% HP-β-CD solutions, respectively, prior to extraction.

Sample preparation
For plasma, standards, QCs, blank or study samples (50 µL) were aliquotted into 10 mL glass test tubes.

Linearity, precision and accuracy
The linearity response of analytes was assessed over their respective calibration range from three batches of analytical runs. The precision and accuracy of the assay were determined for each analyte at LLOQ, LQC, MQC and HQC concentration levels in human plasma and CSF over the three batch runs. The dilution QC was used to assess the dilution integration. These QC concentrations included the known fortified levels added to the plasma or CSF plus the endogenous concentration of analyte. For each QC concentration, analysis was performed in six replicates on each day except for dilution QCs for which three replicates were prepared. Precision and accuracy are denoted by percent coefficient of variance (%CV) and percent relative error (%RE), respectively. The accuracy and precision were required to be within ± 15%RE of the nominal concentration and ≤15%CV, respectively, for LQC, MQC, HQC, and DQC samples. The accuracy and precision were required to be within ± 20% RE of the nominal concentration and ≤20% CV for LLOQ samples in the intra-batch and inter-batch assays (39).

Downloaded from
For 24(S)-HC, long-term storage, freeze/thaw stabilities, and stabilities on the bench-top and in the autosampler were determined at the LQC and HQC concentration levels (n = 3). Long-term storage stability of analyte in human plasma and CSF was tested up to 48 and 34 days upon storage at -80 o C, respectively. Bench-top stability was evaluated from human plasma and CSF that were kept on lab bench at room temperature for 4 hours before sample extraction. Freeze/thaw stability was tested by freezing the samples overnight, followed by thawing to room temperature the next day. This process was repeated three times. In the autosampler, stability was tested over three days by injecting the first batch of the validation samples. Stock solution stability was established by quantification of samples from dilution of two stock solutions that have been stored at -20 o C for 48 days and at room temperature on the bench for 18 hours, respectively, to the final solution (200 ng/mL in methanol). A fresh standard curve was established each time.

Analysis of clinical samples
Samples consisted of calibration standards in duplicate, a blank, a blank with internal standard, QC standards (LQC, MQC and HQC), and unknown clinical samples were analyzed. The standard curve covered the expected unknown sample concentration range, and samples that exceeded the highest standard could be diluted and re-assayed. In the dilution sample re-assay, a diluted QC in triplicate would be also included in the analytical run. The results of the QC samples provided the basis of accepting or rejecting the run according to FDA guidelines (39).

Drug administration and sample collection in PD study
This clinical study was approved by the Institutional Review Board of the Eunice Kennedy Shriver

LC-MS/MS assay development
Although underivatized 24(S)-HC in plasma can be detected by ESI as [M+NH 4 ] + on 4000 QTRAP and by APCI as [M+H-H 2 O] + on Thermo TSQ triple-quadrupole mass spectrometers, the sensitivities are insufficient to detect unesterified 24(S)-HC in CSF. Derivatization of 24(S)-HC to its picolinyl ester significantly increased the detection sensitivity; however, a long LC run (>20 min) was necessary to separate the 24(S)-HC from other isomers (35). By contrast, we found that the 24(S)-HC nicotinate derivative was easily separated from other major isomers (7α-hydroxycholesterol, 7β-hydroxycholesterol, 25-hydroxycholesterol, 27-hydroxycholesterol, 4β-hydroxycholesterol) present in the human plasma using an Eclipse XBD column (3x100 mm, 3.5 µm) with 9 min LC run time including equilibration time Earlier methods to extract 24(S)-HC from plasma used a mixture of chloroform and methanol to simultaneously disrupt lipoproteins and partition 24(S)-HC into chloroform. We found that the acidification of plasma with ammonium acetate-formic acid buffer (pH 3) affected the protein conformation, which in turn led to reduced affinity between 24(S)-HC and plasma proteins. We further used a liquid-liquid extraction with methyl-tert-butyl ether to extract 24(S)-HC from plasma and CSF. An advantage of liquid-liquid extraction with methyl-tert-butyl ether is its high efficiency in removing the phospholipids that are the major source of matrix effect (ionization suppression or enhancement) (41).

Overcoming non-specific adsorption issues for 24(S)-HC in human CSF samples
The lack of significant amount of protein and lipids as well as the relatively high ionic strength in CSF samples can be associated with the loss of lipophilic and highly protein bound molecules via non-specific binding or adsorption to hydrophobic surface of polypropylene container in which they are collected, stored or processed. Failure to adequately address this issue would result in underestimated CSF analyte concentrations, as compounds with log D larger than 3.8 would be expected to experience ≥ 20% adsorption loss in untreated CSF (42). The log D of 24(S)-HC is 7.56 (43), suggesting that it is very likely to be lost by non-specific binding to polypropylene container. To confirm this prediction, we conducted a nonspecific binding diagnose experiment which consisted of five consecutive transfer and incubations steps of CSF samples in polypropylene tubes (44), and found that 55% of 24(S)-HC was lost after five consecutive transfer and incubations. To prevent non-specific binding of 24(S)-HC in CSF samples to the polypropylene container, HP-β-CD was added to a final concentration of 2.5% and the recovery of 24(S)-HC was greater than 93.1% after five consecutive transfer and incubations. Based on these findings, the clinical samples were collected in polypropylene tubes pre-loaded with HP-β-CD.

Selection of surrogate matrix for standard curves
As no 24(S)-HC-free human plasma and CSF are available, we prepared calibration standards by spiking the analyte in surrogate matrixes. The 5% BSA in water was used to simulate generic binding of the analyte to endogenous proteins, and served as surrogate standard curve matrix for plasma samples. 2.5% HP-β-CD in water was used as surrogate standard curve matrix for CSF samples, since 2.5% HP-β-CD was used to prevent non-specific binding of 24(S)-HC in CSF. Because surrogate matrixes were used, the impact of matrixes was investigated using another set of standard curves prepared in pooled human plasma and CSF. The standard curves prepared in human plasma and CSF were parallel to those prepared in 5% BSA and 2.5% HP-β-CD, respectively, and the differences in slopes of the standard curves in surrogate and authentic matrixes were 0 and 3.1% for plasma and CSF, respectively. The intercepts of the surrogate matrix standard curve were close to zero, while they were slightly greater than zero for the authentic matrixes due to the presence of the endogenous analytes (Supplemental Table S1). These results suggested that the same responsiveness of 24(S)-HC in different matrixes was observed and calibration curves prepared in surrogate matrixes were suitable for analysis of plasma and CSF samples.

Extraction efficiency and matrix effects
To evaluate the recoveries of the 24(S)-HC from human plasma, CSF and surrogate standard curve matrixes (5% BSA and 2.5% HP-β-CD), signals of D7-24-HC from pre-extraction spiked samples were

Selectivity
To ascertain the selectivity of the plasma and CSF methods, blank (5% BSA solution for plasma method and 2.5%) with and without internal standard and six independent human plasmas were analyzed. As shown in Figure 3A and 3D, no interfering peaks to analyte and internal standard from blanks for plasma and CSF were observed. There are no interfering peaks to analyte from blank with internal standard for plasma and CSF. In the highest calibrator (ULOQ, 200 ng/mL for human plasma; 5ng/mL for human CSF) without internal standards, there are no interfering peaks to internal standard.  Table S2).

Sensitivity
The LLOQ for plasma and CSF were prepared in BSA and 2.5% HP-β-CD solutions and at 1 and 0.025 ng/mL, respectively. The LLOQ samples were processed and analyzed with a calibration curve and QC samples. The intra-run precisions at LLOQ level were 2.7 -3.4% CV and 2.8 -3.4% CV for plasma and CSF, respectively. The intra-run accuracy levels were -3.2 -0.0% RE and -1.0 -4.1% RE for plasma and CSF, respectively. The inter-run precision was 3.2% CV and 3.6% CV for plasma and CSF, respectively. The inter-run accuracy was -1.6% RE and 1.3% RE for plasma and CSF, respectively ( Table 2). A typical MRM chromatogram at the LLOQ concentration is shown in Figure 3C and 3F.

Accuracy and precision
The accuracy and precision of the method were assessed by analyzing QC samples along with a calibration curve on three different days. The calibration curve consisted of eight standards of different concentrations, each in duplicate, ranging from 1 to 200 ng/mL for plasma and 0.025 to 5 ng/mL for CSF.
The calibration curve (24(S)-HC peak area/internal standard peak area for Y-axis and analyte concentration for X-axis) of 24(S)-HOC was obtained using the least square linear regression fit (y = ax + b) and a weighting factor of 1/x 2 . Excellent results were obtained for the calibration curves, as the deviations of the back-calculated concentrations from their nominal values were within 15% for all the calibration standards in the three days of validation. The coefficients of determination (r 2 ) greater than 0.99 were observed for the calibration curves. All the QC samples were prepared in human plasma or CSF, and the endogenous levels of 24(S)-HC were determined by mean of multiple replicates (n=12). The endogenous levels were used to calculate the nominal concentrations of the spiked (MQC and HQC) and dilution QC (DQC). The plasma and CSF DQC were diluted 5 times with 0.5% BSA and 2.5% HP-β-CD solutions before extraction, respectively, and followed the procedure for other samples. The results of the QC samples in the three validation runs and dilution integration are shown in Table 1. The analysis of the plasma-and CSF-based QC samples demonstrated acceptable precision and accuracy based on the preset validation criteria of ±15% CV and 15% RE.

Carryover
To evaluate carryover, a blank sample was immediately injected following the highest standard (200 ng/mL for plasma and 5 ng/mL for CSF). No carryover was observed in the regions of interest.

Evaluation of 24(S)-HC as a biomarker in Phase 1 trial of HP-β-CD in NPC1 patients
NPC1 is a fatal neurodegenerative lysosomal storage disorder characterized by abnormal accumulation of unesterified cholesterol and sphingolipids in late endosome/lysosomes of many cell types (49). HP-β-CD has been shown to prevent neurodegeneration and prolongs survival in NPC1 animal models (50)(51)(52), is currently being studied in a Phase 1 trial at NIH (38). Administration of HP-β-CD in the animal models

DISCUSSION
In the present study, the intended use of the assay was for measuring the 24(S)-HC as pharmacodynamic biomarker in clinical trials, and the data generated to be used for critical decision-making in the development of HP-β-CD as new drug candidate. Thus, an increase in the rigor of 24(S)-HC method validation is essential. To ensure high quality data, a rigorous full validation was conducted to extensively evaluate the performance of the assay according to FDA guidelines (39) and "fit-for-purpose" strategy (55,56). We developed a highly selective, sensitive, and high-throughput 2D-LC-MS/MS assay for quantification of 24(S)-HC in human plasma and CSF. This assay used a "surrogate matrix" strategy that reduced the issue of potential interference from the endogenous analyte. The assay demonstrated excellent accuracy, precision, linearity, selectivity for the intended purpose of using plasma and CSF 24(S)-HC as biomarkers in clinical trials. The lower limit of quantification was sufficient to capture the basal levels of 24(S)-HC in human plasma and CSF. Stability of the analytes was also thoroughly investigated in current study, and it was found that 24(S)-HC in plasma and CSF demonstrated good bench-top stability, freezethaw stability, and long-term storage stability. In CSF, 24(S)-HC is subjected to absorption loss due to non-specific binding to the polypropylene container, which is prevented by the presence of 2.5% HP-β-CD. Accordingly, appropriate CSF sample collection for clinical studies was established to prevent reported that 24(S)-HOC in sera from health subjects ranged from 4 to 21 ng/ml, indicating that 24(S)-HOC in some samples (e.g. 4 ng/mL) cannot be reliably quantified as they are below the LLOQ of the method. Thus, our method provides sufficient sensitivity to accurately measure these samples, and therefore may be more appropriate for pediatric studies, where sample volumes are limited.
Our method was developed on 4000QTRAP, in which 24(S)-HOC was derivatized with nicotinic acid to improve the detection sensitivity. Newer generations of mass spectrometers such as API5000, API5500, 5500QTRAP and 6500QTRAP offer 5 to 20 times higher sensitivity than 4000QTRAP. The underivatized method using non-selective MRM transition as in Haughey's method (28) on the newest 6500QTRAP may achieve same LLOQ as our method; however, the signal to noise ratio in plasma/serum sample is still inferior to our method. Nonetheless, coupling 6500QTRAP with nicotinic acid derivatization may provide enough sensitivity to detect 24(S)-HOC in dried blood spot samples (equivalent to 1.6 µL plasma/3 mm disk) that is ideal for pediatric studies in infants and children.
The 2D chromatography was employed in our assay. Although 2D chromatography involves more complex instrumentation and method, it has a number of advantages over separation with a single column.
Combination of different separation mechanisms offers a high peak capacity to resolve samples of great complexity. The improved separation capacities reduce/eliminate the interferences, background noises, and ion suppression, all of which improve sensitivity. The peaks eluting before and after the window of the peak transfer from the first to the second column are directed to waste, thus there is less potential for matrix buildup on the second analytical column. The analytical column has a longer life, and consequently the method is more robust. The gradient run time can be reduced by equilibration of a column during the elution on another column. In this study, we used 2D chromatography to achieve online clean up, prolong column life, increase method robustness, and reduce the run time. Although only one switching valve is required to perform 2D chromatography, and we used a second switching valve to keep ion source clean. Using this method, we have analyzed 1700 samples from the Phase 1 clinical trial on a single analytical column, without needing to clean the mass spectrometer ion source. Advances in the theory of 2D separations (57,58), instrument technology (59) and control software (60) in recent years have led to widespread applications in proteomics (60,61), metabolomics (62), and pharmaceutical analysis (63). To date, adoption of 2D chromatography for the separation of lipids has been slow (64), but it is anticipated that we will see more application of this powerful tool to lipid research in the near future. 24(S)-HC has been evaluated as a pharmacodynamic biomarker for HP-β-CD treatment of NPC1 animal models (53). Subcutaneous administration of HP-β-CD has been shown previously to delay neurodegeneration and to prolong lifespan in NPC1 mice (50)(51)(52). We observed significant increases of by guest, on September 23, 2016 www.jlr.org Downloaded from 24(S)-HC in NPC1 mouse plasma after subcutaneous administration of 4000 mg/kg or stereotactical injection of 6 mg/kg of HP-β-CD, and in NPC1 cat plasma and CSF after intracisternal administration of 30 and 120 mg (53). Although a subsequent report suggested that intraperitoneal injection of 4000 mg/kg of HP-β-CD to NPC1 mice had no effect on 24(S)-HC in plasma (65)