Highly automated nano-LC/MS-based approach for thousand cell-scale quantification of side chain-hydroxylated oxysterols.

Iso-octyl chain-hydroxylated oxysterols were determined in attomoles per 10,000 cells concentrations in 10,000-80,000 cultured pancreatic adenocarcinoma cells, using a sensitive, highly automated nano-LC-ESI-MS-based method. Identified oxysterols included 24S hydroxycholesterol (24S-OHC), 25 hydroxycholesterol (25-OHC), and 27 hydroxycholesterol (27-OHC), while 20S hydroxycholesterol and 22S hydroxycholesterol were not detected. Lower mass limit of quantification was 23 fg (65 amol) for 25-OHC and 27-OHC (100 times lower than our previous method) and 54 fg (135 amol) for 24S-OHC, after derivatization into Girard T hydrazones and online sample cleanup using simplified and robust automatic filtration and filter back flushing solid phase extraction LC/MS/MS. The instrument configuration was easily installed using a commercial nano-LC/MS system. Recoveries in spiked sample were 96, 97, and 77% for 24S-OHC, 25-OHC, and 27-OHC, with within- and between-day repeatabilities of 1-21% and 2-20% relative SD, respectively. The study demonstrates the potential of nano-LC in lipidomics/sterolomics.

Calibration solutions for quantifi cation. Calibration solutions were prepared from 1 nM working solutions and covered the concentration range 13-108 pM of each oxysterol. New calibration solutions were made for each assay.
Internal standard. Aliquots of 50 µl of 1.5 nM internal standard working solution were added to all standard, calibration, and validation solutions and cell samples before evaporation into dryness and derivatization with Girard T as described subsequently.

Charge tagging of oxysterol
Standard and calibration solutions, as well as cell and validation samples, were charge tagged with Girard T reagent as described previously ( 9 ) with small changes to adapt to the change in sample size; after evaporation into dryness, the residue was redissolved in 20 µl 2-propanol. Aliquots of 200 µl of 30 µg/ml cholesterol oxidase dissolved in 50 mM phosphate buffer pH 7 were added to all solutions and samples to convert the 3 ␤hydroxy-5-ene to a 3-oxo-4-ene ( Fig. 1B , step 1), performed at 37°C for 1 h using a Grant-Bio PHMT thermoshaker (Grant Instruments, Cambridge, UK) set to 300 rpm. To attach the Girard T reagent ( Fig. 1B , step 2), 500 µl of a mixture consisting of 15 mg Girard T reagent, 15 µl glacial acetic acid, and 485 µl methanol was added to the sample. This resulted in a fi nal volume of 720 µl of all solutions. The reaction was completed in the dark at room temperature overnight. Schematic view of the sample preparation is shown detection, and a 100-fold increased mass sensitivity compared with our previous efforts ( 9 ). The targeting method enabled identifi cation and quantifi cation of Hh active 24S-OHC, 25-OHC, and 27 hydroxycholesterol (27-OHC) in cultured pancreatic adenocarcinoma cells (cell line BxPC-3). The compounds 20S hydroxycholesterol (20S-OHC) and 22S hydroxycholesterol (22S-OHC) were not found to be present above the detection limit in our samples.

Chemicals and cell lines
All chemicals, reagents, and cells were from commercial suppliers and of analytical grade. See supplementary Section I for details of chemicals, suppliers, and cell culturing.

Preparation of standard solutions, calibration solutions, validation samples, and cell samples
Standard solutions and samples for method development and validation. Standard solutions containing 13, 27, 54, 81, and 108 pM of each oxysterol, with 30 pmol cholesterol-25,26,27-13 C as autoxidation-monitoring standard, were prepared from working solutions (1 nM).
For preparation of spiked validation samples, 10,000 cells were lysed in 300 µl absolute ethanol (Kemetyl, Vestby, Norway) containing 120 pmol cholesterol-25,26,27-13 C and spiked with standard solutions to give a concentration of 27, 54, and 108 pM of each oxysterol. Waltham, MA). However, other mass spectrometers (e.g., triple quadrupole) would also be suitable. Ionization was performed in positive mode using a capillary voltage of 2.0 kV. The monitored MS/MS transitions were based on the fragmentation of the Girard T group (see supplementary Section III for illustration of the MS/MS fragmentation) and were 514.44 → 455.36 (analytes), 520.40 → 461.36 (internal standard), and 517.40 → 458.36 (autoxidation monitoring). Collection of data and processing were performed using Xcalibur 2.2 software from Thermo Scientifi c. More MS instrument parameters (mostly based on recommended settings from the manufacturer) are found in supplementary Section IV.

RESULTS AND DISCUSSION
Our goal was 2-fold: both enabling sensitive, simple, highly automated analysis of small samples (10,000 cell scale) and quantifi cation of 27-carbon iso-octyl hydroxylated oxysterol analytes possibly present in cells. Candidates were 20S-, 22S-, 24S-, 25-, and 27-OHC, which are structural isomers with very similar LC/MS properties.
High sensitivity was achieved by using nano-LC and online sample preparation. Nano-LC columns (50-100 µm ID) allow for compounds to elute in substantially smaller volumes compared with that in more conventional columns (e.g., 1 mm ID or larger). Therefore, analytes enter the ESI-MS in larger concentrations, resulting in enhanced signal. A nanobore column packed with the same stationary phase and particles as used in our previously published microbore LC-based method ( 9 ) was used, as the material has good oxysterol separation properties and durability. When replacing our previous microbore LC-ion trap MS methodology with nano-LC coupled with a Q Exactive MS, the mass limit of detection (and quantifi cation) was reduced by a factor of 100 (from 2.5 pg to 23 fg). Separation properties (retention and selectivity) in the nano-and microbore format were similar, as expected.
The oxysterols were charge tagged with Girard T reagent ( Fig. 1B , step 2). This established approach was chosen as in supplementary Section II. Derivatized samples were stored at 4°C and analyzed within a week.

Automatic fi ltration and fi lter back fl ushing/solid phase extraction/nano-LC
Sample cleanup was performed online using automatic fi ltration and fi lter back fl ushing (AFFL) and online solid phase extraction (SPE), coupled upstream to nano-LC-ESI-Q Exactive MS. With the AFFL setup ( Fig. 2 and supplementary Animation I) ( 12,13 ), sample with excess reagent was injected (5 µl) by an Agilent G1377A micro well autosampler (Agilent Technologies, Santa Clara, CA) ( Fig. 2 , step 1). An Agilent 1100 series pump was used as loading pump and fi lter back-fl ush pump [15 µl/min of 0.1% formic acid ( aq )]. Cell debris and precipitates/particulate matter from cells were trapped online on a 1 µm Valco (Houston, TX) stainless steel fi lter (1/16 ″ , 1 µm screen) fi tted in a Valco union (1/16 ″ , 0.25 mm bore), while a reversed phase (RP) HotSep Tracy C 8 trap column [0.3 mm inner diameter (ID) × 5 mm] from G and T Septech (Ytre Enebakk, Norway) trapped derivatized oxysterols ( Fig. 2 , step 2). Excess reagent, not suffi ciently hydrophobic to be trapped by the RP trap column, was fl ushed to waste. The valve was subsequently switched after 3.5 min, and the loading pump automatically redirected to back fl ush the fi lter unit with 0.1% formic acid ( aq ) to clean the fi lter prior to the next injection ( Fig. 2 , step 3). Simultaneously, an Agilent 1200 series pump eluted the oxysterols from the trap column onto the ACE 3 C 18 (0.1 mm ID × 150 mm) RP analytical column (Advanced Chromatography Technologies, Aberdeen, UK) for isocratic separation using a mobile phase consisting of formic acid-water-methanol (0.1:5:95 v/v/v %) ( Fig. 2 , step 3) at a fl ow rate of 500 nl/min. Cholesterol residues in the column might accumulate and make a pseudostationary phase ( 14 ) and hence disturb the chromatographic separation. To remove cholesterol residues from column, the column was washed with formic acid-water-acetonitrile (ACN ) (0.1:5:95 v/v/v %) (40 column volumes) after approximately six injections of cell samples to avoid changing the separation properties of the column. The column switching was performed with a 10-port two-position switching valve from Valco (1/16 ″ , 0.25 mm bore) controlled by the LC pump's Chemstation software.

MS of oxysterols
MS detection was performed in MS2 mode using a Q Exactive Orbitrap with nanospray fl ex ion source (Thermo Scientifi c, performed to avoid autoxidation effects that can negatively affect method accuracy. Enzymatically formed side chain-hydroxylated oxysterols (our analytes) are not typically associated with autoxidation; on the other hand, little is known, for example, about the formation of 20S-OHC, as no known enzyme is associated with the formation of this compound. Nonetheless, heavy cholesterol (cholesterol-25,26,27-13 C) was added to all samples and calibration solutions. If autoxidation occurred during sample preparation (turning cholesterol into our analytes), heavy cholesterol would be converted to heavy oxysterols. This mass transition was monitored and controlled for each sample with MS. Autoxidation(-like) effects were in fact observed for ‫ف‬ 1 in every 500 samples (these samples were rejected).
The chromatographic resolution of our method revealed that, presumably due to generation of syn/anti forms during the derivatization process ( 7,17 ), several of the analytes appeared as two peaks. As these peaks were well resolved, their areas could be combined for quantification. By using methanol as the LC organic modifi er instead of ACN, we were able to quantitatively distinguish 24S-OHC and 27-OHC, which coeluted with ACN as modifi er. In spiked samples, 27-OHC and 20S-OHC coeluted with methanol as the organic modifi er, but it was possible to resolve 20S-OHC from the other oxysterols when using ACN (see supplementary Section VII for chromatograms). In accordance with other recent studies ( 9,18,19 ), 20S-OHC was not observed in the analyzed cell samples using ACN as mobile phase, even when the sample amount was increased 8-fold. Hence, methanol could be used as the organic modifi er for our method for reliable quantifi cation of 27-OHC.
In a pancreatic cell line (BxPC-3) and a derived clonal mutant cell line ( ⌬ ␤ cat BxPC-3), the sensitivity of the method allowed detection of 24S-OHC, 25-OHC, and 27-OHC in as little as 10,000 cultured cells ( Fig. 3 ). In contrast, 20S-OHC and 22S-OHC were not detected in the cells even when using larger samples. The compounds 24S-OHC, 25-OHC, and 27-OHC were quantifi ed (38, 19, and 35 fmol/10,000 cells, respectively). In support of these oxysterols being generated in the cells was the within-cell presence of corresponding enzymes responsible for formation of these oxysterols, both at the transcriptional level by RT-PCR and the protein level by Western blot and nano-LC/ MS/MS. Also, a consistent and signifi cant upregulation of 27-OHC and its corresponding enzyme (cholesterol 27-oxidase) as function of a cell perturbation (i.e., knockout of ␤ catenin) was observed (for more details, see supplementary Section IX).

Validation of the nano-AFFL-SPE-LC/MS/MS method
The concentration limit of detection (cLOD) was defi ned as the lowest concentration that repeatably produced a chromatographic peak with signal-to-noise ratio >3. The concentration limit of quantifi cation (cLOQ) was defi ned as the concentration that produced peak areas with a relative SD (RSD) ‫ف‬ 20%. The cLODs were 13 pM derivatized analyte for 24S-OHC, 25-OHC, and 27-OHC, corresponding the entire procedure can be performed in one tube, minimizing sample loss ( 9 ). In addition, it is compatible with automated SPE enrichment and cleanup ( 9 ).
The Girard T reagent reacts with the keto group as shown in Fig. 1B , step 2. A potential pitfall was that the LC/MS method would not distinguish between species with an already existing keto group and those with keto groups formed during sample preparation. However, analyte analogs with preexisting keto groups were not detected in the analyzed cell samples, investigated by performing Girard T tagging with and without step 1 in Fig. 1B (see supplementary Section V for chromatograms).
The 23 fg mass detection limit of this method corresponds to 13 pM in the vial (720 µl). It is important to notice that the in-vial concentration will be affected by the available number of cells because the resulting total volume is the same for all samples and standard solutions after derivatization. Using more cells per sample will therefore give higher analyte amounts and concentrations, and hence a lower limit of detection per cell. For calculation example, see supplementary Section VI.
A 0.3 mm ID SPE column was coupled online with the nano-LC column. This avoids the overloading effects associated with direct injection of microliter amounts onto nano-LC columns (which can usually only handle nanoliter injections). Such overloading effects may call for extensive column purifi cation routines; in the only previous nano-LC study of oxysterols (to the authors' knowledge), Karu et al. ( 10 ) had to inject a washing solvent between each sample injection to avoid carryover effects. In SPEnano-LC mode, however, 5 µl injections could easily be SPE trapped, and concentrated analyte bands were subsequently transferred to the LC column in appropriate volumes for satisfactory chromatographic resolution without carryover effects. In addition to allowing rather large injection volumes, online SPE greatly simplifi es analysis of our 10,000 cell-scale samples, as excess Girard reagent removal is performed online. This reduces diffi cult, manual handling of minute samples/fractions, which can be a source of analysis variance ( 15 ).
Conventional online SPE-nano-LC can be especially prone to pressure buildups due to the small, relatively easily clogged connections. To avoid this, an AFFL setup was installed upstream to the SPE step. This allows crude samples to be injected directly (here: unfi ltered lysates) as remaining cell debris and so forth is trapped on a stainless steel fi lter and washed out of the system after each sample injection. This reduces off-line sample preparation steps (and hence possible loss of analyte). The AFFL-SPE-nano-LC/MS methodology could handle thousand injections of relatively unprepared samples, representing an unprecedented robustness regarding nano-LC-based instrumentation. Our AFFL-SPE-nano-LC system can be assembled using a commercial SPE-LC system [see ( 16 ) for various systems], only requiring replacement of one valve (10 port instead of a 6 port) ( 13 ), compared with our original system, which required an additional pump ( 12 ).
An additional simplifying aspect of the method was the absence of a (off-line) cholesterol removal step, commonly Recovery [or more correctly, the apparent recovery ( 20 )] was examined by comparing samples spiked at three concentration levels with standard solutions of the same concentration levels, low (27 pM), medium (54 pM), and high (108 pM). The recovery was 96, 97, and 77% for 24S-OHC, 25-OHC, and 27-OHC, respectively. See supplementary Section VIII for more details (linearity curves, equations, etc.).

CONCLUSIONS
A highly sensitive, robust, and automated method for quantifi cation of the natural iso-octyl chain hydroxylated oxysterols 24S-OHC, 25-OHC, and 27-OHC was successfully developed and validated and applied to measure oxysterols in small samples ( у 10,000 cells). The study illustrates robust and successful use of nanoscale LC in targeted lipidomics/sterolomics, an approach that is usually "reserved" for proteomics. The identifi cation/presence of the endogenous oxysterols in BxPC-3 cells was supported by identifi cation of corresponding enzymes responsible for formation of these oxysterols within the cells. The method will be further used in studying endogenous side chain-hydroxylated oxysterols in the Hh pathway in the context of cancer.
Linearity of the method in spiked samples was examined in the range 27-108 pM, and the correlation coeffi cient ( r 2 ) for all linearity curves was >0.99 for all analytes (three concentration points, three spiked samples per concentration point). The slope of the curves corresponded to that in standard solutions (fi ve concentration points, three standard solutions per point). Standard solutions could therefore be used as calibration solutions. Ideally, calibration solutions should be based on cell samples, but this was not possible due to lack of cell matrix without oxysterols. The linear concentration range examined covered the concentration range observed in studied samples.
Repeatability was examined for low (13 pM), medium (54 pM), and high (108 pM) concentration levels. Three individually prepared standard solutions per level were injected over 4 days (total of 12 standard solutions for each concentration level) for all cell-detected analytes. Withinday repeatability at low concentration level was between 2% and 21% RSD for 25-OHC and 27-OHC. The betweenday repeatability was between 15% and 20% RSD for 25-OHC and 27-OHC (but somewhat higher for 24S-OHC, 27% RSD) at low concentration level and 2-6% RSD at medium and high levels for all analytes (4 days).