Validation and application of sub-2 μm core–shell UHPLC–UV–ESI–Orbitrap MS for identification and quantification of β-carotene and selected cleavage products with preceding solid-phase extraction

A validated ultrahigh-performance liquid chromatography method using 1.7 μm core–shell particles is presented for the identification and quantification of β-carotene (BC) and related cleavage products (CPs) in primary cell culture media. Besides BC, apo-4′-, apo-8′-, apo-10′-, and apo-12′-carotenals, as well as 5,6-epoxy-β-carotene, were selected as target analytes. Detection was performed via an 80-Hz diode array detector and an electrospray ionization–linear quadrupole ion trap–Orbitrap XL mass spectrometer, both hyphenated in series. Total analysis time was below 6 min with peak widths <12 s. Addition of trifluoroacetic acid and tetrahydrofuran to the mobile phase allowed for the mass spectrometric detection of BC and related CPs and reduced peak tailing due to improved solubility of hydrophobic analytes. Intra-day and inter-day precision for UV and mass spectrometric detection were ≤1.5 % for retention times and ≤5.1 % for peak areas. Instrumental linearity was confirmed by Mandel’s fitting test between 0.25 (or 1.00 μg/mL) and 5.00 μg/mL for UV detection. The higher sensitivity of mass spectrometric detection allowed for the coverage of three concentration domains between 0.025 and 5.00 μg/mL in linearity testing. Homoscedasticity was confirmed between 0.10 and 5.00 μg/mL for Orbitrap XL MS. The limits of quantification were between 52.6 and 889.4 ng/mL for UV detection and between 19.3 and 102.4 ng/L for mass spectrometric detection. Offline solid-phase extraction from culture media fortified with BC and CPs provided intra- and inter-day recoveries between 65.8 and 102.4 % with coefficients of variation ≤6.2 %. Primary rat hepatocyte cultures treated with BC and subjected to different oxidative stress conditions contained 5,6-epoxy-BC and apo-4′-carotenal besides residual BC. Apparently, 5,6-epoxy-BC was formed in the medium via autoxidation of BC by ambient oxygen.


Introduction
Carotenoids and particularly β-carotene (BC)-a precursor of vitamin A-have been assigned beneficial health effects [1] including a prominent role in the prevention of cataract, senile macular degeneration, cardiovascular diseases, neurodegenerative diseases, cancer, and in cancer therapy [2,3]. Several studies have provided evidence for anti-oxidative effects of BC by quenching singlet oxygen and scavenging peroxy radicals [4,5]. Thus, BC has been applied comprehensively in micro-supplementation [6]. Besides, BC and its degradation product apo-8′-carotenal were frequently applied as food and beverage colorants E 160 a and E 160 e [7].
Nonetheless, certain frame conditions trigger pro-oxidative and pro-carcinogenic effects of carotenoids [8,9]. High oxygen pressure, pronounced oxidative stress, and BC administration at elevated doses apparently promote adverse effects [4,10]. Two large-scale human intervention trials, i.e., the ATBC [11] and the Carotene and Retinol Efficacy (CARET) [12,13] studies, have addressed this paradox by oral longterm administration of 20-30 mg BC/day to human probands. Test persons comprised inter alia cigarette smokers and workers exposed to asbestos. These groups showed a 16-28 % increased lung cancer incidence with higher mortality risk when compared to control groups which requested the termination of the CARET trial ahead of schedule [12,13]. Other BC supplementation studies, such as the Physician Health Study (PHS) [14] and the Linxian Trial [15,16], revealed no effect on the health state or demonstrated indeed reduced cancer-related mortality [5,14]. Different dietary basis uptake of BC and pre-existing tumors might explain divergent study results [5,16].
Adverse effects in humans have been mimicked in animal models sharing physiological similarity in BC adsorption and tissue metabolism, i.e., ferrets [2,16]. Thereby, BC supplementation at pharmacological doses was combined with exposure to cigarette smoke [17]. The high number of free radicals prevailing in cigarette smoke promotes the formation of BC oxidation products (= cleavage products (CPs)) [18]. Moreover, CPs, such as apo-10′- [19] and apo-14′-carotenal, have also promoted the binding of benzo[a]pyrene to DNA [16]. Apparently, the described BC paradox was not evoked by BC per se but by its degradation products (CPs) generated by eccentric cleavage as stated elsewhere [16,18,20]. This has been corroborated by recent studies on primary rat hepatocytes treated with apo-8′-carotenal and CP mixtures [21,22]. Non-volatile CPs, e.g., apo-carotenals, epoxides, and carbonyls, are generated by hypochloric acid (HClO) that is formed by myeloperoxidase secreted from activated phagocytes [23]. The reaction mechanism between BC and HClO has been mimicked in vitro by Handelman et al. [24] and revealed a comprehensive set of CPs. The proposed eccentric degradation products [25] are also present to a minor extent in intestinal mucosa cells and hepatic cells due to the presence of β-carotene-9′,10′-monooxygenase (BCO 2) [26,27].
Linear quadrupole ion trap (LTQ)-Orbitrap MS with full scan option provides the possibility for a mid-term untargeted strategy for high sensitivity, mass accuracy in low parts per million range [44], and a mass resolution of >100,000 (FWHM) [45]. Consistently, the hyphenation of UHPLC to linear ion trap-Orbitrap MS represents a highly promising combination for UHPLC-MS analysis of carotenoids.
The current work aims to develop and validate an UHPLC-diode array detection (DAD)-electrospray ionization (ESI)-Orbitrap MS method for BC and related CPs selected from previously described degradation profiles [24,25] applying sub-2 μm core-shell particles with preceding offline solid-phase extraction (SPE). The combination with offline SPE intends to further reduce interferences [46]. Thereby, the SPE strategy recently developed for volatile CPs [47] should be adopted to allow for a simultaneous extraction of volatile and non-volatile CPs, and BC with optional splitting of the eluate for final GC-EI-MS [47] and UHPLC-DAD-ESI-Orbitrap MS analyses. This combined extraction step for CPs of highly variant properties provides an innovation in the analysis of cell culture media. Analyte identification refers to a combination of several decision parameters, i.e., t R , UV spectra, accurate masses, and related natural isotopic patterns.
Due to its high mass accuracy, mass stability, detection sensitivity, and analyte specificity, Orbitrap MS was preferred over time-of-flight MS or triple-quadrupole MS, allowing the highly confident identification of the analytes extracted from biological samples. Mass spectrometric measurements were performed by an LTQ-Orbitrap XL mass spectrometer which was hyphenated by the Ion Max ESI ion source (all from Thermo Fisher Scientific) to the UHPLC system. Tuning of the system was done in the automatic tune mode by means of a 100-μg/mL apo-8′ solution prepared in 50 % (v/v) THF-50 % (v/v) ACN for the monoisotopic mass of [apo-8′+H + ] at m/z 417.3, by adding the tuning solution to the column effluent via a T-splitter. The optimized MS parameters included a heated capillary temperature of 250°C; sheath gas and auxiliary gas flow of 30 and 15 arb. units, respectively; a source voltage of +3.0 kV; a capillary voltage of 46.00 V; and a tube lens voltage of 35.00 V. Calibration of the system was performed according to the protocol of the manufacturer. Data acquisition in single ion monitoring (SIM) mode with the LTQ-Orbitrap XL mass spectrometer addressed theoretical masses of analytes with a tolerance window of ±0.5m/z. By reason of the targeted approach in mass spectrometry, SIM mode is preferred over full scan. Due to the analysis of cell culture

Solid-phase extraction
Strata Phenyl 500 mg/3 mL SPE columns were obtained from Phenomenex and applied in the SPE optimization for BC and apo-carotenals. The SPE procedure is based on a recently published SPE method for volatile cleavage products of β-carotene [47]. Briefly, SPE columns were conditioned with 3 mL methanol followed by 3 mL ultrapure water before 1.0 mL of blanks, of the respective MEM working solution-spiked either with β-carotene and the internal standard or β-carotene, related cleavage products, and the internal standard-or of the cell culture medium after incubation was loaded. The column was then washed with 2 mL ultrapure water. Elution of β-carotene, nonvolatile cleavage products, and methylisoeugenol (IS) was done with 2.0 mL 10 % (v/v) THF in n-hexane at a flow rate ≥2 mL/min to comply with the SPE requirements of volatile cleavage products measurements by GC-EI-MS. The eluate was cooled to −20°C to facilitate the separation of the organic and aqueous phase. The hydrophobic fraction of the eluate was then collected [47]. The implementation of this SPE strategy offers a simultaneous extraction of volatile and non-volatile CPs for future analyses with subsequent equal splitting of the 2-mL eluate for GC-EI-MS and UHPLC-DAD-ESI-Orbitrap MS, respectively. When analyzing only BC and non-volatile CPs, the entire eluate can alternatively be transferred to a Kuderna Danish micro-evaporator system and evaporated to dryness under a gentle stream of nitrogen. Residues were reconstituted in 1.0 mL 50 % (v/v) ACN-50 % (v/v) THF and then injected directly into the UHPLC-DAD-ESI-Orbitrap MS system. An appropriate volume of this diluted BC treatment solution was then spiked to the cell culture medium to make up a further 1:10 dilution. Consistently, the final concentration in the cell culture was 10 μmol/L BC including also 9.5 μmol/L Tween 20. Prior to the addition to cell cultures, the spiked culture medium is mixed to assure a homogenous solution. Primary hepatocyte cultures Primary hepatocytes were chosen as a model system because the culture medium is not supplemented with serum (regularly 10-20 %) and contains only proteins excreted by the cells [48], thus facilitating the chemical analysis. Primary parenchymal hepatocytes were prepared from female Fischer 344 rats according to the protocols given elsewhere [48,49]. Isolated primary hepatocytes were cultivated in 5 mL serumfree MEM supplemented with non-essential amino acids, pyruvate (1 mmol/L), aspartate (0.20 mmol/L), serine (0.20 mmol/L), and penicillin (100 U)/streptomycin (100 μg/mL) in collagen-coated 60-mm-diameter plastic culture dishes for 21 h with change of the culture medium after the first 3 h. After these 21 h, incubation of cells with the respective treatment solutions was performed directly without any further change of medium for 3 h. Incubation under control conditions was done at 37°C, 5 % CO 2 , and 95 % relative humidity without oxidative stress treatment applying either culture medium alone (i.e., deficient of Tween 20 and BC) or fortified with appropriate volumes of DMNQ, H 2 O 2 , or Fe ++ stock solutions. Fortified culture media were prepared immediately prior to their application, which was done by replacing the medium after 21 h incubation. In case of treatment, primary hepatocytes were incubated with 10 μmol/L BC and concurrent oxidative stress.
Cell treatment with BC under oxidative stress Different approaches were applied to mimic the oxidative stress of the risk groups in the chemoprevention trials employing the validated SPE UHPLC-DAD-ESI-Orbitrap MS method to culture media. The tested treatment solutions for primary hepatocytes address in vitro degradation of BC to non-volatile CPs over the incubation interval, subsequent to BC spiking to the cell culture medium. In case of treatment under different prooxidant conditions, primary hepatocytes were incubated for 3 h with 10 μmol/L BC in the presence of either (i) 40 μmol/L DMNQ, (ii) 10 μmol/L H 2 O 2 , or (iii) a combination of 10 μmol/L H 2 O 2 and 10 μmol/L Fe(II)lactatethe latter subsequently termed "Fenton condition" [50]. Cultures exposed to these regimens without addition of BC served as references for oxidative conditions. For chemical analysis, the culture supernatants were collected immediately after the 3-h treatment period and subjected to SPE. For cultures treated with BC and subsequent oxidative stress, hepatocytes from two different animals were prepared on different days. Hepatocytes of each rat were cultivated in three separate Petri dishes, respectively. For either Petri dish, one SPE was performed and the final eluate analyzed in triplicate.

Statistical treatment
Data were statistically evaluated by means of the SPSS Statistical Software Package version 16 applying appropriate test procedures, e.g., one-way analysis of variance (ANOVA), Levene's test, and independent t test. Mandel's fitting test (MFT) is not included as standard SPSS operation and was thus established via the SPSS syntax function. Details of the calculation procedure have been outlined recently [47] and are briefly recalled in the respective sections. Results were considered significant at p<0.05.

Result and discussion
The selection of non-volatile target analytes, i.e., apocarotenals and 5,6-epoxy-BC, was based on the constituent profiles of CPs previously described for the in vitro cleavage of BC with HClO [24] and for activated neutrophils [10] and other cells, such as pneumocytes [34], when incubated with BC.
UHPLC with core-shell columns Currently, UHPLC separations have been merely published for carotenoids, retinol and tocopherol [33,51], but not for BC and related CPs, i.e., apo-carotenals. In comparison to previously published HPLC separations for apo-carotenals which required 30-40 min total analysis time [52], the developed UHPLC separation was performed in less than 6 min with peak widths at base of analyte peaks smaller than 12 s. The gradient composition and programming were essential to achieve narrow peak widths and avoid precipitation as well as trapping of highly hydrophobic analytes on the column. Therefore, a ternary gradient composed of ultrapure water, ACN, and THF was crucial for dissolving moderately and highly hydrophobic analytes in a moderate hydrophilic solution.
TFA has been added both to reduce secondary ionic interactions by shielding residual silanol groups [53] and reduce the pH in the mobile phase to promote ionization in positive mode. During the optimization of the UHPLC separation, severe tailing of the BC peak due to a restricted solubility [54] was an issue of concern (data not shown). Therefore, the THF content in the mobile phase was stepwise increased within the gradient up to final 45 % (v/v) immediately before BC was passing the detector. This improved peak shape and symmetry substantially and is in accordance with recent strategies [19]. Although a minor tailing of the BC peak persisted under the selected conditions (Fig. 2a), an increase beyond the selected THF content showed no further improvement (data not shown). UV spectra for BC and related CPs recorded under the selected UHPLC conditions are depicted in Fig. 2b together with related absorbance maxima (Table 1). These wavelengths were also selected for detection of analytes and validation of UHPLC-DAD. The maximum at 458 nm and the side maximum at 487 nm for BC correspond to previous data [36] despite a minor shift which is related to the different compositions of the mobile phase [54]. Apocarotenals showed an increase in their respective absorbance maximum in the order of apo-12′ to apo-4′ due to the increase in the polyene chain length [54], as described elsewhere [36,28,55]. As expected, the elution order of BC and related CPs is governed by their hydrophobicity (Figs. 1 and 2a).

LTQ-Orbitrap MS
The lack of protonation sites in BC and other carotenes makes their ionization in ESI-MS inefficient if not taking appropriate means to ameliorate the situation. However, this does not refer to the CPs which can be protonated (see structures in Fig. 1). Ionization of BC has been realized either by post-column addition of oxidizing agents, e.g., halogenated solvents [56], or of silver salts [32,57]. In the current case, ionization of BC is effected upon electrochemical oxidation [58] with formation of radical ions (M •+ ) [56,57]. Moreover, it has been shown that electrochemical oxidation is supported under acidic condition [56]. Contrary to BC, all target CPs are detected as [M+H] + under the selected conditions. EICCs of a standard solution are depicted in Fig. 3. In addition, detected masses and isotope patterns for BC and target apo-carotenals are included and compared with their theoretical masses and simulated isotope patterns (Fig. 4) which were used for formula confirmation.

Solid-phase extraction in sample preparation
The concept of the presented SPE approach pursues the strategy of a simultaneous extraction of physicochemically divergent analytes, i.e., BC, non-volatile, and volatile CPs. Thus, the UHPLC-DAD-ESI-Orbitrap MS method for non-volatile CPs should share the SPE procedure with GC-EI-MS for volatile CPs [47]. This strategy ensures identical adsorption and elution conditions for all target analytes and reduces variations in the SPE performance for individual analyte classes within the same sample. However, since GC-EI-MS and UHPLC-DAD-ESI-Orbitrap MS are applied as final analysis steps, a splitting of the eluate and change to solvents compatible with the final separation and detection step are required. As in GC-EI-MS, methylisoeugenol was applied as an IS to correct quantification in UHPLC-DAD-ESI-LTQ-Orbitrap MS.

Instrument validation of UHPLC-DAD-ESI-Orbitrap MS
The developed UHPLC-DAD-ESI-Orbitrap MS method was subject of a single-laboratory validation according to the IUPAC and the relevant ICH guideline Q2(R1) [59,60]. Instrumental basis validation addressed intra-day and interday precision of retention times (t R ) and peak areas, linearity testing for up to three concentration ranges, homoscedasticity testing, and determination of limit of detection (LOD) and limit of quantification (LOQ) for UV and mass spectrometric detection, respectively. Results are given in Tables 2, 3 Precision As part of the instrumental basis validation, intraand inter-day precision of t R and corrected peak areas were determined for UV and MS detection. Therefore, a 2.0-μg/L standard solution including the IS was injected repetitively. For calculation of the intra-day precision, this standard solution was injected five times, whereas the inter-day precision was based on three replicate injections on five consecutive days. Coefficients of variation (CV) for t R are given only for mass spectrometric data, since due to its higher spatial distance from the injection site, this detection system is expected to give higher CVs of t R . Both intra-and inter-day precision of t R were ≤1.5 % (corresponding to deviations of less than 2 s). Precision of corrected peak areas was ≤4.1 % for intra-day and ≤5.1 % for inter-day measurements for UV and mass spectrometric detection, respectively.
Linearity Response linearity of DAD and Orbitrap MS for BC and related CPs was tested with both detection systems hyphenated in series to UHPLC. Two ranges between 0.25 and 5.00 μg/mL were covered for UV. However, due to the higher sensitivity of Orbitrap MS in comparison to UV detection, another low concentration domain was included to extend quantification to lower concentrations of CPs. Consistently, three concentration ranges between 0.025 and 5.00 μg/mL were tested with Orbitrap MS. Each calibration region encompassed four to five different concentrations evenly distributed over the respectively addressed domain. Regression equations refer to the least square approach and consider peak areas corrected by IS. In the case non-linearity is revealed by MFT over the entire concentration range, a restriction to individual concentration ranges based on the respective signal intensity will offer improved quantification.
DAD The tested concentration ranges for UV covered 0.25-5.00 μg/mL. Analyte signals were acquired at their respective wavelengths as given in Table 1. The highest concentration tested, i.e., 5.00 μg/mL, refers to the BC concentration intended for subsequent application in primary cell cultures to reveal possible genotoxic effects with concomitant oxidative stress.
The significance of intercepts was tested by means of an ANOVA approach offered by SPSS. Additionally, ANOVA A apo-12′-carotenal, B apo-10′carotenal, C apo-8′-carotenal, D apo-4′-carotenal, E 5,6-epoxy-BC, and F β-carotene was also applied to evaluate the linear regression model, i.e., the significance of the slope, which was confirmed in either case (data not shown). Furthermore, linearity was tested by means of MFT [61]. The coefficient of determination R 2 which indicates the correlation between the applied concentration and the measured signal, i.e., peak area, but does not represent a measure for linearity [62]-as frequently stated erroneouslyis given as well. All data are surveyed in Table 2. However, due to the higher LOQ of apo-4′, 5,6-epoxy-BC, and BC that were >0.50 μg/mL (see Table 4), linearity was only considered between 1.00 and 5.00 μg/mL for these CPs (see Table 2). The MFT indicated no improved fitting for a second order regression model for either tested concentration domain. Even when both tested concentration regions were combined in case of apo-12′-, apo-10′-, and apo-8′, covering the entire domain of 0.25-5.00 μg/mL, calculated F values derived by MFT proved no significance (see Table 2). This implies linearity over the entire concentration range tested in UV.
Orbitrap MS Due to the enhanced sensitivity of Orbitrap MS in comparison to the previously addressed UV detection, the tested concentration range was extended to lower concentrations and divided into three consecutive domains, i.e., 0.025-0.10, 0.10-1.00, and 1.00-5.00 μg/mL which allows for a quantification of CPs and BC in considerably lower concentrations than with UV detection. Calculation of the significance of the intercept, as well as linearity testing with MFT and the evaluation of the linear regression model with ANOVA, was performed as outlined in the previous paragraph. Measured peak areas refer to extracted ions for the monoisotopic masses of BC and target CPs acquired in SIM mode (see Table 1) with a mass window of ±0.05m/z.
With increasing concentrations, results of Orbitrap MS indicated a continuous reduction in the slope (Table 3). However, the statistical evaluation by MFT indicated significant deviations from linearity only for apo-8′. In parallel, R 2 are also considerably reduced in comparison to corresponding UV data, which additionally points to reduced correlations. This is particularly true for the highest concentrations between 1.00 and 5.00 μg/mL, with the exception of apo-12′-and apo-10′-carotenal (Table 3). Thus, current results indicate an incipient saturation of detector response, and the selection of a regression model for individual concentration regimes in correspondence to experimentally measured peak areas seems most advantageous. Although this might be arguable from the statistical point of view, i.e., for MFT results which are not significant, this strategy will reduce the uncertainty calculated as confidence interval when quantifying analytes via regression.

Testing for homogeneity of variance
In addition, calibration data were tested for absence of heteroscedasticity, since this would change the uncertainty of analytical results in a concentration dependent manner. Due to its higher sensitivity, quantification of CPs will be done by Orbitrap MS with optional confirmation of concentrations by UV in case analytes are situated in appropriate concentration domains. Therefore, homogeneity of variance was tested for ESI-LTQ-Orbitrap MS by analyzing standard solutions of CPs (including IS) at concentration levels of 0.1 and 5.0 μg/mL by ten injection replicates, respectively. F calc was derived by dividing the relative variance of the higher by the relative variance of the lower concentration. In all tested cases, F calc was lower than the tabulated F tab (=5.34), confirming homogeneity of relative variances over the tested range at a 99 % confidence level (Table 3) [61].

Instrumental limit of detection and limit of quantification
LOD and LOQ of BC and CPs were calculated for UV and Orbitrap MS detection according to the ICH guideline Q2(R1) applying the so-called standard deviation approach. Thereby, the standard deviation of signals has to be calculated either from the noise of a blank or from a standard solution or sample containing the respective analytes at low concentrations [59]. Instrumental LOD and LOQ were determined from a combined standard solution containing BC and CPs at 0.025 μg/mL. LOD and LOQ both for the UV and the MS measurement step were calculated to address possible differences and derive applicable minimum concentrations for linearity testing. Baseline fluctuation, i.e. noise, was calculated differently for UV and MS detection. For UV, noise was calculated from the lowest standard concentration considered in linearity testing, i.e., 0.10 μg/mL. Thereby, a retention window considering twice the base peak width on either side of the respective analyte peaks was constructed for determination of the baseline noise. In case of MS detection, noise values given by the software via the signal-to-noise option in the SIM mode were derived for either analyte signal. LOD was calculated as the 3.3-fold of the determined noise divided by the slope of the peak height calibration curve over the low concentration domain for either CP. For LOQ, the 10-fold noise is considered otherwise duplicating the calculation approach for LOD [59]. Since both LOD and LOQ refer to signal heights, calibration curves for peak heights had to be calculated. Due to previous linearity data, considered concentration ranges were selected close to preliminarily estimated LOQs to prevent deviation from linearity. Selected concentration ranges for linearity testing of peak heights comprised 1.00 to 5.00 μg/mL for UV and 0.10 to 1.00 μg/mL for Orbitrap MS. For either concentration range linearity was proven by MFT as outlined in the previous section (data not shown). Calculated LODs and LOQs for UV detection and MS are provided in Table 4. Depending on the respective UV absorbance, LOQs for UV detection were between 52.6 and 889.4 ng/mL. Remarkably, LOQs for Orbitrap MS were between 19.3 and 102.4 ng/L and thus between 5.15×10 2 -and 2.60×10 4 -fold smaller than their UV counterparts (see Table 4). The low LOD and LOQ realized for BC with Orbitrap MS are related to electrochemical oxidation and formation of radical ions (M •+ ) as described previously [56,57].

Intra-day and inter-day precision of SPE recovery
The SPE protocol pursued for non-volatile CPs and BC refers to [47]. Briefly, the principle of interaction between the phenyl stationary phase employed in SPE and the analytes is mainly based on π-π interactions. In the current case, the entire 2-mL SPE eluate was applied to UHPLC subsequent to the aforementioned solvent change, but operation with a 1-mL aliquot is applicable as well when employing GC-EI-MS and UHPLC-DAD-ESI-Orbitrap MS in parallel. The intra-day precision of SPE was determined at two concentration levels within the single-laboratory validation, based on five SPE replicates within 1 day, respectively. Model samples for SPE were prepared by spiking MEM with a combined stock solution of CPs and BC to give final concentrations of 1.0 and 0.5 μg/mL for either analyte. Recoveries were between 65.8 and 102.4 % with corresponding CVs ≤4.0 %, respectively (Table 5; n=5). A comparison of the recovery between 1.0 and 0.5 μg/mL (intra-day) by means of a two-sided independent sample t test revealed no significant differences. Moving to target cells other than hepatocytes might require the inclusion of fetal calf serum in the cell culture medium. Inter-day precision was based on three extractions per day performed over five consecutive days. Spiked model samples were prepared independently on either day. Inter-day recoveries were between 66.6 and 91.1 % (CV ≤6.2 %; n=15). Onefactor ANOVA of inter-day recoveries revealed significant differences for apo-10′-, apo-4′, 5,6-epoxy-BC, and BC between individual days, whereas Levene's test for testing homogeneity of variance was only significant for 5,6-epoxy-BC and BC. The stationary phenyl phase previously applied in the SPE of volatile CPs [47] appears thus equally applicable for the extraction of highly hydrophobic BC and non-volatile target CPs.

Specificity
To confirm the method specificity, fresh MEM and MEM after 3 h incubation in primary hepatocyte cultures were subjected to SPE and subsequently analyzed by UHPLC-DAD-ESI-Orbitrap MS. No interfering peaks were observed in UV and MS chromatograms in either case (data not shown). However, with the inclusion of fetal calf serum, a slight blank was revealed that was ≤3 % when compared to peak area of the BC concentration applied in culture treatment (data not shown). This has to be considered when analyzing low BC concentrations in the presence of fetal calf serum.

SPE linearity for spiked MEM samples
Preliminary results of BC degradation in cell cultures revealed BC, 5,6-epoxy-BC, and apo-4′ in quantifiable amounts (data not shown). Therefore, SPE linearity was confirmed by spiking MEM with these analytes in appropriate concentrations. Apo-4′ and 5,6-epoxy-BC were quantified by Orbitrap MS. The 100-to 4,000-fold higher concentrations of BC in comparison to these CPs were due to its application in form of the treatment solution and required quantification by DAD. Five concentrations distributed equidistantly over the tested range were analyzed for either compound. In addition, a MEM blank was analyzed. Subsequent to SPE, each concentration was injected in triplicate. Derived calibration curves refer to the mean of the injection replicates, respectively. All calibration curves were linear as proven by MFT. Intercept was not significant for apo-4′and 5,6-epoxy-BC (see Table 6). The addition of 10 % FCS to cell culture media showed neither changes in t R of analytes nor in their recovery and recovery precision of SPE. Moreover, the influence on the selectivity was minute, obviously requiring consideration only at greatly reduced concentrations of BC. However, this does not necessarily guarantee the maintenance of regression slopes in UV and MS detection and SPE calibration nor the respective linear ranges confirmed for serum deficient culture media (see Tables 2, 3, and 6). Actually, matrix effects due to addition of fetal calf serum might alter these parameters and additionally change LOD and LOQ. In case of quantification in the presence of fetal calf serum, a comprehensive further validation that is beyond the current scope is mandatory to assure correct concentrations.

Identification of CPs in treated cell cultures
The validated SPE UHPLC-DAD-ESI-Orbitrap MS method was applied to cell cultures of primary rat hepatocytes. Subsequent to the addition of the BC treatment solution to primary cultures to give a final BC concentration of 10 μM (5360 ng/mL), primary rat hepatocyte cultures were incubated for 3 h and simultaneously subjected to oxidative stress induced by different means. This approach targets to elucidate possible differences in the generated profiles and concentrations of CPs that were produced either directly in the medium and/or in the cells with subsequent secretion. Oxidative stress was either induced by (i) 40 μmol/L DMNQ [21], (ii) 10 μmol/L hydrogen peroxide (H 2 O 2 ) [63,64], or (iii) 10 μmol/L H 2 O 2 in the presence of 10 μmol/L Fe(II)lactate (the latter simulating the Fenton reaction) [50]. The H 2 O 2 treatment concentration was selected 10-fold lower than published elsewhere [65], to assure integrity of hepatocytes. The H 2 O 2 and Fe(II)lactate concentrations were selected equimolar and exerted no cytotoxic effects over the incubation interval [66]. Application of DMNQ was designed as initial treatment since DMNQ induces oxidative stress after its uptake into the cell. This is considered to simulate the cell physiological environment of patients subjected to strong oxidative stress, such as smokers. CPs derived from this treatment and proven not to result from autoxidation in the culture medium can thus be assigned to enzymatic (and possibly non-enzymatic) formation taking place in the cell with subsequent secretion into the medium. The second approach applied H 2 O 2 as an oxidant agent to promote oxidative stress. In the last approach, hepatocytes were simultaneously incubated with H 2 O 2 and Fe(II)lactate. According to the Fenton reaction, H 2 O 2 reacts with Fe 2+ with consecutive generation of OH − and • OH and Fe 3+ [50]. In addition, either treatment procedure was also performed (i) by addition of BC in the absence of cells and (ii) in the presence of cells, but without addition of BC. As a further control, (iii) hepatocytes were treated with BC without induction of oxidative stress. All other experimental frame conditions were maintained. To prevent instrumental sample carry-over between analytical runs and thus false positive results, measurement series comprising samples of individual treatments were always completed with blank injections, which provided no signals.
In case cells were treated with BC and subjected to oxidative stress, 5,6-epoxy-BC and apo-4′ were detected in addition to BC, irrespective of the mode of oxidative stress (see Table 7). Remarkably, cell cultures treated with BC but without oxidative stress showed the same profile of CPs, i.e., 5,6-epoxy-BC and apo-4′. However, apo-4′ was only detected in quantifiable amounts in hepatocyte cultures of one rat, but not in cultures of the other in either case. If only BC was added to cell culture medium without cells, 5,6-epoxy-BC was detected as well but not apo-4′, irrespective of the mode of oxidative stress and even in its absence (Table 7). Based on these results, 5,6-epoxy-BC which constitutes the major CP under the selected experimental conditions seems to be formed irrespective of externally induced oxidative stress and the presence of cells. Since no CPs were detected after SPE of MEM solutions spiked with BC, the extraction step can be excluded as a causative source for BC degradation. Instead, the ambient oxygen pressure prevailing during the 3-h incubation period appears sufficient to induce oxidation of BC, forming 5,6-epoxy BC but none of the other target CPs. This provides essential information for the interpretation of adverse effects on cells, since even by exclusive application of BC encountered effects might be related to the formed epoxy variant of BC. In case of apo-4′, the situation is more intriguing. Since this CP was never found in the absence of cells, formation is most likely mediated by cells but depends apparently on their individual response, as not all test animals provided this CP. Altogether, the application of the validated SPE UHPLC-DAD-ESI-Orbitrap MS method allowed for a deeper insight into CP profiles and quantification of individual target CPs formed in the course of in vitro experiments. This is an essential prerequisite for the interpretation of adverse cell effects when trying to reveal possible mechanisms attempting to mimic the BC paradox in probationers suffering from oxidative stress, such as smokers and asbestos workers.

Conclusion
For the first time this work presents a fast UHPLC method employing sub-2 μm core-shell particles for separation of BC and related long-chain CPs. UHPLC hyphenation to DAD and-via electrospray ionization-to LTQ-Orbitrap MS allows the identification and quantification of these analytes in cell culture media. Preceding offline SPE as well as the applied internal standard are shared with a GC-EI-MS method for volatile CPs previously published by our group [47]. This permits a simultaneous SPE of volatile and non-volatile CPs and BC under identical extraction conditions and thus a more comprehensive investigation of BC degradation in in vitro systems subjected to oxidative stress. The method was validated for BC and selected CPs which are currently considered most relevant for adverse effects observed in risk groups of BC intervention studies. Therefore, intra-day and inter-day precision for peak areas and retention times, linearity of detector responses, LOD and LOQ, as well as intra-and inter-day SPE recoveries were determined. LOQs between 19.3 and 102.4 ng/L for Orbitrap MS allowed for a trace detection of CPs. The profiling and identification of CPs by LTQ-Orbitrap MS with high mass accuracy as well as their quantification in primary cell cultures is essential to relate observed genotoxic effects with prevailing ensembles of CPs. When culture media were incubated with BC, 5,6- epoxy-BC was quantified after 3 h irrespective of oxidative stress. Apparently the formation occurs during the incubation by ambient oxygen. The observed concentration of this CP was equivalent in the presence and absence of primary rat hepatocytes. However, within a primary hepatocyte population derived from one rat, an early long-chain CP, i.e., apo-4′carotenal, was generated under oxidative stress in all tested cultures, but not in a second biological replicate. This might indicate individual degradation kinetics over the investigated 3-h incubation. Since the method has proven its applicability for cell culture media, it will provide an essential input for an improved interpretation of in vitro models and contribute to a future deciphering of the BC paradox by testing various BC concentrations with increased biological replicates.