A sensitive and specific method for measurement of multiple retinoids in human serum with UHPLC-MS/MS.

Retinol (vitamin A) circulates at 1-4 μM concentration and is easily measured in serum. However, retinol is biologically inactive. Its metabolite, retinoic acid (RA), is believed to be responsible for biological effects of vitamin A, and hence the measurement of retinol concentrations is of limited value. A UHPLC-MS/MS method using isotope-labeled internal standards was developed and validated for quantitative analysis of endogenous RA isomers and metabolites. The method was used to measure retinoids in serum samples from 20 healthy men. In the fed state, the measured concentrations were 3.1 ± 0.2 nM for atRA, 0.1 ± 0.02 nM for 9-cisRA, 5.3 ± 1.3 nM for 13-cisRA, 0.4 ± 0.4 nM for 9,13-dicisRA, and 17.2 ± 6.8 nM for 4oxo-13-cisRA. The concentrations of the retinoids were not significantly different when measured after an overnight fast (3.0 ± 0.1 nM for atRA, 0.09 ± 0.01 nM for 9-cisRA, 3.9 ± 0.2 nM for 13-cisRA, 0.3 ± 0.1 nM for 9,13-dicisRA, and 11.9 ± 1.6 nM for 4oxo-13-cisRA). 11-cisRA and 4OH-RA were not detected in human serum. The high sensitivity of the MS/MS method combined with the UHPLC separation power allowed detection of endogenous 9-cisRA and 4oxo-atRA for the first time in human serum.

dichloromethane. The 4oxo-RA-methyl ester was hydrolyzed using 2 M KOH in methanol, and 4oxo-at RA was extracted using ethyl acetate and crystallized. 4OH-at RA was synthesized from 4oxo-RA using two equivalents of NaBH 4 in methanol. The product was extracted with 1:1 mixture of ethyl acetate and ether, evaporated to dryness, and crystallized. 9,13-dicis RA was synthesized as previously published ( 28 ). In brief, 100 mg of 9-cis -retinal was dissolved in 20 ml of methanol. Eight milliliters of Tollens Reagent (equal volumes of 10% AgNO 3 and 10% NaOH mixed and titrated with ammonium hydroxide until precipitate dissolved) was added to the solution, and the mixture was stirred at 37°C for 6 h. The reaction was quenched with 4 N HCl on ice and fi ltered, and the products were extracted with hexane and evaporated to dryness. The mixture of 9-cis RA and 9,13-dicis RA was purifi ed using silica gel chromatography and ethyl acetate:hexane mobile phase. The fraction containing 9,13-dicis RA was quantifi ed by NMR by calculating the ratio between the area of the 9-cis RA doublet at 6.09 ppm (H in C10) to the corresponding doublet from 9,13-dicis RA at 6.2 ppm. The identity of the 9,13-dicis RA was also confi rmed by the ratio of the 9,13-dicis RA doublet at 7.7 ppm (H at C12) to the corresponding doublet in 9-cis RA at 6.3 ppm. Based on NMR quantifi cation, the reaction yielded 20% 9,13-dicis RA and 80% 9-cis RA.

Extraction of retinoic acid isomers and 4oxo-retinoic acid
All sample processing, preparation, and extraction was conducted on ice under red light. 9-cis RA, 13-cis RA, at RA, 4oxo-13cis RA, 4oxo-at RA, 4OH-at RA, and 4OH-9-cis RA were spiked into blank serum. Into 500 l of serum, 10 l of a 60:40 ACN:MeOH mixture with 1 M 13-cis RA-d 5 and 2 M 4oxo-13-cis RA-d 3 was added as internal standards. All compounds were extracted according to previously described method for extraction of RA isomers ( 29 ). In brief, 1 ml ACN and 60 l 4 N HCl were added to each sample, the samples were vortexed, and retinoids were extracted twice with 5 ml of Hexanes. The organic phase was separated by centrifugation at 1000 rpm for 3 min and evaporated to dryness under nitrogen stream at 32°C. The dry residue was reconstituted in 50 l of 60:40 ACN:H 2 O and transferred to an amber autosampler vial.
Serum retinoids could also be measured with no extraction using a simple protein precipitation with ACN. In this method, 50 l of ACN is added to 50 l of serum followed by centrifugation at 2000 rpm for 10 min at 4°C to pellet serum proteins. The supernatant is transferred to a 96-well plate, and 20 l is injected into the UHPLC/MS/MS for analysis.

UHPLC MS/MS analysis
The retinoids were separated using an Agilent 1290 UHPLC (Santa Clara, CA) equipped with a Sigma (St. Louis, MO) Ascentis Express RP Amide column (2.7 m; 150 mm × 2.1 mm). Gradient elution with a fl ow rate of 0.5 ml/min using water (A) and acetonitrile (B) with 40% methanol and 0.1% formic acid in A and B was used. The gradient was from initial 60% (A) for 2 min to 45% (A) over 8 min and then to 10% (A) over 7 min. The column was then washed with 95% (B) for 3 min and returned to initial conditions. The column heater was set to 40°C. Samples were kept in the autosampler at 4 ° C with the light turned off. Ten microliters of sample was injected for analysis. Analytes were detected using an AB Sciex 5500 qTrap Q-LIT mass spectrometer (Foster City, CA) operated in positive ion APCI mode. The compound independent MS parameters were curtain gas: 20; collision gas: low; nebulizer current: 5; temperature: 350 ° C; ion source gas 1:80. The fi nal compound dependent parameters used for analysis are summarized in Table 1 . The specifi c MS/MS transitions for each analyte were optimized using the Analyst software (Applied Biosystems, Foster City, CA) using direct infusion with The major limitation in understanding the roles and importance of 4OH-RA, 4oxo-RA, and specifi c RA isomers in vivo has been the lack of knowledge on the concentrations of the specifi c isomers and metabolites in human tissues.
Analysis of RA isomers and their metabolites has been diffi cult due to their similar structures and low concentrations in human tissues. A lack of analytical separation power may have confounded individual isomer detection, whereas a lack of good MS/MS sensitivity has prevented selective quantifi cation of individual retinoids. Previous methods have been developed using LC-MS/MS ( 9,(16)(17)(18)(19)(20)(21), GC/MS ( 22 ), and LC/diode array detector-atmospheric pressure chemical ionization/MS/MS ( 23 ). However, none of these methods has incorporated measurement of hydroxylated RA metabolites at endogenous levels in serum. Endogenous at RA (4.6-5.8 nM), 13-cis RA (4.7-6 nM), and 4oxo-13-cis RA (8.1-9.8 nM) were measured in human serum using HPLC with UV detection ( 24,25 ). However, 9,13-dicis RA and 9-cis RA were not measured, nor was separation of the main four isomers of RA demonstrated. Separation and quantifi cation of at RA, 9-cis RA, 9,13-dicis RA, and 13-cis RA in mouse tissues was shown using LC-MS/MS analysis ( 9 ), but no measurement of the metabolites of RA was conducted.
The goal of this study was to develop a method for the simultaneous measurement of biologically active endogenous retinoids in human serum. Due to the challenges in separating the isomers of RA as well as the metabolites, chromatographic separation was optimized for UHPLC. The method was validated with spiked charcoal-treated blank human serum and used to determine retinoid concentrations in serum from healthy men. The developed assay is useful for characterization of retinoid disposition and endogenous retinoid homeostasis and for following therapeutic interventions with retinoids. The method can also be used in preclinical studies of compounds that target changing retinoid concentrations.

Chemicals and reagents
at RA, 9-cis RA, 13-cis RA, and 9-cis -retinal were purchased from Sigma (St. Louis, MO). 4oxo-at RA, 4OH-atRA , and 9,13-dicis RA were synthesized as described below. 4oxo-13-cis RA, 4OH-9-cis RA, 4oxo-13-cis RA-d 3 , 13-cis RA-d 5 , and 11-cis RA were purchased from Toronto Research Chemicals (North York, Ontario). All compounds were stored in amber vials in ethanol as 1 mM stocks at Ϫ 80°C. Acetonitrile, methanol, water, and formic acid used in the UHPLC-MS/MS method were from Fischer Scientifi c (Pittsburg, PA), and all were Optima LC/MS grade. Blank human serum (DC Mass Spect Gold MSG 4000) was purchased from Golden West Biologics (Temecula, CA). This serum has a normal range of triglycerides (30-200 mg/dl) and cholesterol (> 20 mg/dl) to mimic the extraction environment of clinical samples. This is important due to the matrix effects associated with lipids during extraction and analyte ionization.
4OH-RA and 4oxo-RA were synthesized according to previously published methods ( 26,27 ). In brief, the methyl ester of at RA was generated using trimethylsilyldiazomethane. Methyl 4oxo-retinoate was prepared using activated MnO 2 in anhydrous proteins with acetonitrile, centrifugation of precipitated proteins, and direct analysis of the supernatant as described above for retinoid extraction.
To determine whether matrix effects altered the ionization and detection of retinoids in serum, four serum samples were extracted and reconstituted in duplicate with 50 l of 60:40 ACN/H 2 O mixture containing 5 nM and 10 nM 13-cis RA-d 5 . In parallel, the same ACN/H 2 O mixture with 5 nM and 10 nM 13cis RA-d 5 was added into glass tubes with no extracted samples. After a quick vortex, all samples were transferred into amber vials and analyzed by LC-MS/MS. The response was quantifi ed and compared between the matrix containing samples and the clean standards.

Subjects
Serum samples were collected from 20 healthy men between the ages 18 and 65 who were enrolled in a study examining the relationship between retinoids and spermatogenesis. The study was approved by the Institutional Review Board at the University of Washington, and all subjects gave written informed consent before any study procedures. All subjects exhibited normal blood chemistry, liver function, hematology, and hormones. Subjects were excluded from the study if they used anabolic steroids or illicit drugs, ingested more than four alcoholic drinks a day, or were being treated with ketoconazole, fi nasteride, dutasteride, methadone, or lithium. The fi rst set of blood was collected between 8 AM and 12 PM during a fed state within 2-6 h of breakfast. The second blood draw occurred at least 7 days after the fi rst one, and blood was collected between 8 AM and 12 PM after an overnight fast of at least 8 h. Blood samples were collected from the subject's antecubital vein, and blood tubes were immediately wrapped in aluminum foil to minimize light exposure (light-protected samples). The blood was allowed to clot at 4 ° C and then spun for 20 min at 3,000 g . Serum was aliquoted into amber sample tubes and stored at -80 ° C until analysis. Samples were also collected without the aluminum foil wrapping and light-protected vials. Five of these samples were extracted and analyzed together with the light-protected samples.
Results are expressed as mean ± SD. All statistical analysis was done using GraphPad Prism (La Jolla, CA). Due to nonnormality, comparisons of retinoids between the fasting and fed states were performed using a Wilcoxon signed-rank test. Correlation between data sets was tested with linear regression. For all comparisons, a P value of 0.05 was considered signifi cant.

Analysis of the human serum samples and confi rmation of analyte identity
All the serum samples were analyzed using the described method. The identities of the quantifi ed retinoids were confi rmed by collecting MS/MS spectra of each analyte. For this analysis, four serum samples were extracted with hexanes as described above, and the hexane phases were combined. After drying under nitrogen fl ow, the sample was reconstituted in 50 l of 40:60 H 2 O:ACN, and 20 l was injected into the UHPLC-MS/ MS. The UHPLC conditions were identical to those described for quantitative analysis, and the same MS/MS parent-fragment pairs used for quantifi cation were recorded to detect the analytes and trigger MS/MS spectrum acquisition. Once the signal for the MS/ MS transition exceeded a threshold, a fragment ion scan for the same parent ion was triggered using positive ion APCI and with collision energy spread of 15 from a set value of 35. A dynamic fi ll time, which allows for the maximum amount of ions to be collected in the linear ion trap for best sensitivity, was used to collect the fragment ion spectra.
To confi rm that each quantifi ed peak for the detected RA isomers in serum represented only a single compound, two and without the addition of the LC solvents. Initially, MS/MS transitions were identifi ed for each analyte using the automatic optimization features of the Analyst software. These transitions were compared with those from a manual optimization, and the fi ve most abundant fragments for each analyte found in both the automatic and manual process were optimized independently for their declustering potential (DP), entrance potential (EP), collision energy (CE), and collision cell exit potential (CXP). All fi ve identifi ed potential MS/MS transitions for each analyte were included in a method that was used to analyze blank, spiked, and normal human serum to determine interference, matrix effects, and signal-to-noise ratios of each MS/MS transition. The MS/MS transitions that showed the least amount of matrix interference in the blank serum and the highest signal-to-noise ratio in spiked samples were chosen for the fi nal analysis.

Assay validation
The assay was validated according to the published guidelines for bioanalytical method validation ( 30,31 ). The on-column lower limit of detection (LLOD) determined as signal-to-noise ratio >3 and the on-column lower limit of quantifi cation (LLOQ) determined as signal-to-noise ratio >9 were measured for each compound as standard solution. The linearity of response was determined using standard curves generated over four orders of magnitude (0.0001-1.0 M). For quantifi cation, blank serum was spiked at six concentrations between 0.1 nM and 20 nM for RA isomers and between 0.5 nM and 20 nM for the metabolites to construct standard curves for each analyte. The peak area ratios of analyte/internal standard were plotted as a function of concentration. 13-cis RA-d 5 was used as the internal standard for quantifi cation of at RA, 9-cis RA, 13-cis RA, and 9,13-dicis RA, and 4oxo-13-cis RA-d 3 was used for metabolite quantifi cation. Each retinoid, with the exception of 9,13-dicis RA, was quantifi ed using standard curves of the same compound. 9,13-dicis RA concentrations were measured using the standard curve for 9-cis RA after confi rming that the MS/MS responses were identical for the two compounds using the NMR quantifi ed standard. Quality control (QC) samples were extracted along with each standard curve and run at the beginning, middle, and end of each run. All QC samples had accuracy and precision within published guidelines for bioanalytical method validation ( 30,31 ).
To determine the intraday and interday coeffi cients of variation (CVs), the samples were extracted on three separate days. The CV was measured at the LLOQ for serum of each compound and at 7.5 nM of RA isomers and 10 nM of 4oxo-RA metabolites spiked into blank serum. The LLOQ in serum was determined by spiking 20 ml of blank serum with RA isomers and RA metabolites at concentrations of 0.05 nM (RA isomers) or 6 nM (metabolites). Aliquots of the QC samples were frozen and analyzed on three separate days. Due to their poor extraction recovery, the LLOQ (5 nM spiked in blank serum) of 4OH-at RA and 4OH-9-cis RA was determined after precipitation of serum  )  301  205  100  80  10  17  10  4OH-RA  299  157  100  59  10  16  15  4oxo-RA  315  159  100  66  10  23  16  4oxo-RA-d 3  300  226  100  71  10  35  2  13-cis RA-d 5  306  116  100  97  10  97  6 CE, collision energy; CXP, collision cell exit potential; DP, declustering potential; EP, entrance potential.
analysis. This was due to the apparent loss of water ( Ϫ 18) from the 4OH-RA [M+H] + protonated molecule ( m/z 317) resulting in a base peak at m/z 299. In negative ion mode, an [M-H] Ϫ ion could be detected for the 4OH-RA compounds, confi rming that they were stable during chromatography and unstable in the mass spectrometer (data not shown).
To achieve separation of the fi ve RA isomers ( at RA, 9-cis RA, 9,13-dicis RA, 11-cis RA, and 13-cis RA), 4OH-RA isomers, and 4oxo-RA isomers, numerous C18 and C8 columns with UHPLC capability were evaluated as well as several chiral columns with mobile phases, including methanol, water, acetonitrile, acetic acid, and formic acid. Signifi cant differences in the separation capacity between the different columns was observed, and in most columns the 13-cis RA and 9,13-dicis RA were not separated, whereas separation between 13-cis RA and at RA was obtained in virtually all columns tested. Baseline separation of at RA, 9-cis RA, 9,13-dicis RA, and 13-cis RA was achieved only by using the amide column described, and separation of the oxidized RA metabolite isomers was satisfactory ( Fig. 2 ). The 11cis RA isomer could not be separated from the 9,13-dicis RA regardless of the analytical conditions (vide infra ). Optimal separation of the oxidized metabolite isomers was achieved using a C18 column, but the separation achieved using the amide column was suffi cient for quantifi cation.

Selectivity of the method for endogenous retinoids
Initially, the fi ve MS/MS parent-fragment pairs with the greatest signal magnitude were chosen for the assay, and each of these SRM transitions (MS/MS parent-fragment pairs monitored) was tested in blank human serum for the existence of interference from the matrix. For several of the mass transitions, endogenous interference was detected, including m/z 301 > 161, m/z 301 > 159, m/z 301 > 91, and m/z 301 > 105 for RA isomers; m/z 299 > 91, m/z 299 > 128, and m/z 299 > 115 for 4OH-RA; and m/z 315 > 120 for 4oxo-RA. The m/z 315 > 297 MS/MS transition was not considered for quantifi cation of 4oxo-RA due to the lack of specifi city of a loss of water fragment. An alternative fragment that reduced interference from the matrix was chosen for quantitative analysis for each analyte as summarized in Table 1 . Due to matrix interference, fragments with the highest abundance were generally not the MS/MS transitions used for fi nal analysis. Close to the retention time of at RA-d 5 , a signifi cant interfering peak was detected ( Fig. 2E ) ( Fig. 2C and E ), its abundance interfered with the quantifi cation of at RA-d 5 . Therefore, 13-cis RA-d 5 was chosen as the internal standard for RA isomers ( Fig. 2E ). No interference was detected at the retention time of 13-cis RA-d 5 .

Validation of the method for serum samples
The on-column LLOD and LLOQ for clean standards for each analyte are shown in Fig. 3 . Standard curves of at RA, 9-cis RA, 13-cis RA, 4oxo-at RA, and 4oxo-13-cis RA were independent MS/MS transitions from the extracted serum samples were monitored, and the response ratio across the peak was recorded. If a peak consists of two compounds, the ratio between the two transitions usually changes across the peak. Both transitions used parent ion m/z 301, and the fragment ions monitored were m/z 205 and m/z 123. These fragment ions were chosen based on their signal-to-noise ratios from spiked serum and retinoic acid standards. The MS parameters for the m/z 301 > 123 transition were DP:62, CE:23, EP:10, CXP:14. The m/z 301 > 123 MS/MS fragmentation of RA is most likely a result of a cleavage of the bond between carbons C6 and C7, resulting in the ␤ -ionone-ring fragment with an m/z 123 as shown previously ( 32 ). The structure of the m/z 301 > 205 fragment could not be assigned due to a likely rearrangement of the retinoid structure during mass spectrometry. However, a corresponding fragment at m/z 306 > 210 was detected from RA-d 5 , showing that this fragment also retained the ␤ -ionone-ring (data not shown). The corresponding fragments ( m/z 301 > 205 and m/z 305 > 209) were detected previously from at RA and 13 C 4 -at RA, respectively, providing additional confi dence for the reproducibility of this fragmentation ( 32 ).
Despite extensive efforts, 11-cis RA could not be separated chromatographically from the 9,13-dicis RA. To determine whether the retinoid detected in human serum was 9,13-dicis RA or 11-cis RA, palladium(II)nitrate was used to convert 11-cis RA to at RA according to a published method ( 33 ). Four light-protected serum samples were pooled to obtain a total volume of 2.5 ml of serum. 13-cis RA-d 5 was spiked into the sample to a fi nal concentration of 20 nM as an internal standard. A 1.1 ml aliquot was removed from the pooled sample, and 5 nM 11-cis RA was spiked into the sample. Two 500 l aliquots were collected from the 11-cis RA spiked serum and from the remaining pooled control serum. The aliquots were extracted as described for serum samples. All four of the samples were reconstituted in 500 l ACN. A solution containing a 1:4.8 ratio of palladium(II)nitrate to Triethylamine in ACN was added to one sample with 11-cis RA spiked in and to one sample without added 11-cis RA. The fi nal Palladium(II)nitrate to retinoic acid molar ratio was approximately 50,000:1. The reaction was allowed to proceed at 50 ° C for 5 h in the dark, after which the ACN was removed under a stream of dry nitrogen. The samples were reconstituted in ethyl acetate and washed twice with water to remove the Palladium reagent. Samples were then dried again under nitrogen and reconstituted in 50 l of a 60:40 ACN/H 2 O solvent.

Optimization of chromatographic separation and MS/MS detection of retinoids
The ionization and fragmentation of the retinoids were evaluated using electrospray and APCI in both the positive and negative ion modes. The negative ion mode provided approximately 10-fold lower sesitivity than the positive ion mode (data not shown); hence, the positive ion mode was chosen for further evaluation. The fragmentation and specifi c MS/MS ion transitions (SRM transition) for each RA isomer, for the isotope-labeled internal standards, and for the metabolites were optimized for maximum sensitivity. The MS/MS fragmentation patterns of the protonated molecules ([M+H] + s) for representative RA, 4oxo-RA, and 4OH-RA isoform are shown in Fig. 1 . The protonated molecules ([M+H] + s) of 4OH-at RA and 4OH-9-cis RA were not stable enough in the positive ion mode to isolate for MS/MS A lack of signifi cant ion suppression or other matrix effects on retinoid detection was determined by comparing the peak areas for 13-cis RA-d 5 standard without matrix and 13-cis RA-d 5 in the sample spiked after extraction (100% recovery) at three different concentrations. The matrixcontaining samples were not different in terms of peak area or retention time from the samples without matrix.
The effect of light exposure during sample collection to retinoid measurements was tested by comparing measurements from light-protected samples with measurements in samples that were not collected in light-protected conditions. All of the samples showed signifi cant decreases ( ≈ 70%) in the concentrations of RA isomers, demonstrating that protecting the samples from UV light already during sample collection is critical. No decrease in the concentrations of generated in blank human serum and extracted along with the study samples. The generated standard curves were linear, and the equations with r 2 values are shown in Table 2 . The validation data at the LLOQ for serum (0.05 nM for RA isomers and 6 nM for 4oxo-13-cis RA) and at the mid to high concentration (7.5 nM for RA isomers and 10 nM for 4oxo-13-cis RA) are shown in Table 2 . All of the measured concentrations were within 15% of the true value (accuracy) for all analytes at the mid-high QC and within 20% of the true value at LLOQ. The LLOD in serum for 4OH-RA isomers was 2 nM when the samples were analyzed after protein precipitation with acetonitrile. The retention times of the analytes were reproducible between days and within days with variability in the retention times, being <0.5% for RA isomers and <0.8% for the metabolites. The identitiy of each detected retinoid was confi rmed by obtaining an MS/MS spectrum of the compound in serum ( Fig. 1 ). There were no detectable differences between the MS/MS spectra of 13-cis RA, 9,13-dicis RA, at RA, and 9-cis RA or between the 4oxo-13-cis RA and 4oxo-at RA. Analysis of the human serum samples allowed collection of an MS/MS spectrum for each RA isomer, confi rming the identity of the detected analytes. A representative MS/MS spectrum from human serum is shown in Fig. 1B . Additionally, the fragmentation for the endogenous compound 4oxo-13-cis RA ( Fig. 1D ) was similar to the standard. Peak purity was confi rmed by monitoring two separate SRM transitions for the RA isomers. The difference in the ratio between the two transitions ( m/z 301 > 205 and m/z 301 > 123) remained constant across each peak in the standard and analyte in serum. 4oxo-13 cis RA was detected in response to light exposure (data not shown).

Detection and identifi cation of endogenous retinoids in human serum
The detection and separation of the endogenous RA isomers in serum is shown in Fig. 2C and 2D, with the identifi ed retinoids indicated. As shown in Fig. 4C with 11-cis RA spiked serum, 11-cis RA could not be separated from 9,13-dicis RA. However, treatment of serum with palladium(ii) nitrate converted the spiked 11-cis RA quantitatively to at RA ( Fig. 4D ). In the nonspiked sample ( Fig 4A, B ), no decrease in the 9,13-dicis RA peak area or increase in at RA peak area was detected, confi rming that 11-cis RA is not present in human serum at quantifi able concentrations. to determine the normal concentrations of endogenous RA isomers and metabolites in human serum. All four RA isomers and 4oxo-13-cis RA were detected in all samples. 4oxo-at RA could also be detected in 18 of the 40 samples, but the concentrations were below LLOQ. The concentration of 4oxo-13-cis RA could not be quantifi ed in one subject at the fed state and in another subject in the fasted state, 9-cis RA could not be quantifi ed in one sample from fed state, and 9,13-dicis RA could not be quantifi ed in samples from two subjects in the fasted state because concentrations were below the LLOQ. The mean concentrations of the four RA isomers and 4oxo-13-cis RA are shown in Table 3 , and a box and whiskers plot representing the measured concentrations for each retinoid in the 20 volunteers is shown in Fig. 6 . The most abundant RA isomer in human serum was 13-cis RA, although at RA concentrations were only about 30-40% lower. The concentrations of at RA and 13-cis RA were about 10-fold higher than The hydroxylated metabolites 4OH-9-cis RA and 4OH-at RA were not detected in the samples extracted and analyzed using the method optimized for RA isomers ( Fig. 2 ). To determine whether this was due to poor extraction recovery of 4OH-RA isomers, serum was analyzed after precipitation of proteins with ACN followed by centrifugation. The 10 serum samples analyzed had no quantifi able levels of 4OH-9-cisRA and 4OH-at RA ( Fig. 5 ) , showing that the concentrations of these metabolites, if they are present in human serum, are <2 nM. It is possible that the two peaks detected in the serum ( Fig. 5C ) are 4OH-RA isomers, but this could not be confi rmed because experiments to obtain MS/MS spectra on these peaks were unsuccessful.

Quantifi cation of retinoic acid isomers and metabolites in human serum
Serum samples collected after an overnight fast and 2-6 h after normal breakfast from 20 healthy men were analyzed  Table 1, and the corresponding SRM channels ( m/z 301 > 205 for RA isomers, m/z 315 > 159 for 4oxo-RA isomers, and m/z 299 > 157 for 4OH-RA isomers) are shown.

DISCUSSION
The described UHPLC MS/MS method is to date the most sensitive method to selectively monitor bioactive retinoids. The selectivity for RA isomers and metabolies is achieved using a combination of UHPLC, specfi c solvents, and an amide column that provides the specifi c separation capacity needed for each isomer. Surprisingly, positive ion APCI ionization produced the best signal-to-noise ratio for the RA isomers despite the readily ionizable carboxylic acid group in the molecule. However, the RA metabolites exhibit better sensitivity using electrospray ionization and negative ion detection. For the quantifi cation of the retinoids in serum, positive ion APCI was chosen because matrix interference was observed using electrospray 9,13-dicis RA concentrations and 30-to 50-fold higher than 9-cis RA concentrations. Two of the RA metabolites (4oxo-13-cis RA and 4oxo-at RA) were also detected in human serum, and the concentrations of 4oxo-13-cis RA exceeded those of 13-cis RA by 3-fold, making this compound the most prevalent retinoid measured. No statistically significant differences ( P > 0.05) were found between the fed and fasted states for any of the detected analytes ( Table 3 ).
Linear regression analysis was used to determine whether correlations exist between the serum concentrations of the detected analytes. There was a signifi cant correlation between the fed and fasted concentrations of 13-cis RA ( P < 0.001), 9,13-dicis RA ( P < 0.001), and 4oxo-13-cis RA ( P < 0.001), whereas the concentrations of at RA and 9-cis RA did not correlate between the fed and fasted states ( P > 0.05). a The LLOQs were determined at 0.05 nM RA isomers and 6 nM 4oxo-RA isomers.
b The LOQs were determined at 7.5 nM RA isomers and 10 nM 4oxo-RA isomers. No decrease in 9,13-dicis RA peak between panel A and B is detected, and no increase in at RA peak suggesting that 11-cis RA was not present in serum. In the control reaction, the spiked 11-cis RA was quantitatively converted to at RA (panel C vs. D).
13-cis RA, and 9,13-dicis RA in mouse tissues, and hence the sample preparation is expected to be applicable for tissue analysis ( 9 ). However, for any given tissue, especially the eye, a confi rmatory analysis should be conducted to determine the potential presence of 11-cis RA and 9,13-dicis RA in the samples. In addition, special attention should be paid to the potential interfering compounds in the matrices analyzed. As demonstrated here, signifi cant interference is present in many of the SRM transitions examined, and it is expected that abundance and identity of interfering matrix compounds varies between tissues. Because blank tissues are generally unavailable for retinoid analysis, simple transfer of a method validated for one tissue to another should be exercised with caution.
One advantage of the described method is the use of isotope-labeled internal standards for quantifi cation of endogenous RA and its metabolites. The use of deuterium-labeled internal standards is important because they are likely to mimic the extraction effi ciency of RA isomers and metabolites. The results of this study show that the extraction characteristics of retinoids, even within a closely chemically related group (e.g., RA, 4oxo-RA, and 4OH-RA), can be very different, and hence using a chemical analog as an internal standard may confound quantifi cation results. In addition, binding of the analytes to cellular retinoic acid-binding proteins and fatty acid-binding proteins may affect recovery and may be corrected by isotope labeled internal standards. Finally, a signifi cant advantage for the use of deuterated retinoids as internal standards is that it allows monitoring of isomerization during sample ionization with negative ion, and limited amount of biological samples may require a single analysis for all relevant bioactive retinoids. The best sensitivity for retinoid detection was obtained using a liquid-liquid extraction with hexanes, and the method was validated for serum using this extraction method. However, the recovery of the hydroxylated metabolites is poor using the extraction method, and a direct protein precipitation with acetonitrile is recommended for analysis of hydroxylated metabolites of RA. A similar extraction method as shown here for serum samples has been previously used for detection of at RA, Fig. 5. Detection of 4OH-RA isomers in human serum using an ACN protein precipitation and centrifugation for sample preparation. Blank serum (A), 5 nM 4OH-at RA and 4OH-9-cis RA spiked into blank serum (B), and a representative human serum sample (C) are shown. The peak at 5.75 min in panel C is likely 4OH-at RA, but the peak area is below LLOQ for this compound. increased sensitivity of the UHPLC/MS/MS method. The identity of the detected 9-cis RA was confi rmed by obtaining an MS/MS spectrum of the detected 9-cis RA. This detection of 9-cis RA is unlikely to be an artifact of the assay or a result of isomerization because no isomerization of the internal standards was detected and because the ratio between isomers did not change when samples were reanalyzed. In addition, 9-cisRA was detected in the samples after ACN precipitation as well as after extraction, providing additional support for 9-cis RA being present in human serum. The 9-cis RA has also been detected in mouse pancreas and has been shown to regulate glucose homeostasis ( 3 ). The ratio of RA isomers determined in human serum was different than what has been previously reported in mouse serum ( 9 ). In mouse serum, at RA was the most abundant retinoid, with 9,13-dicis RA being present at similar concentrations. The concentrations of 13-cis RA were much lower than the two other detected retinoids. In contrast, in human serum 9,13-dicis RA was a minor circulating species, and 13-cis RA was the most abundant retinoid. This demonstrates a signifi cant species difference between the isomers. It is unlikely that the observed difference is due to isomerization in the two studies because the extraction methods were similar and both studies used LC-MS/MS. In addition, analysis of the human serum samples using the ACN precipitation protocol decribed here resulted in an identical ratio of RA isomers as seen with the extraction method. The precipitation method eliminates the potential for isomerization, which may result from the extraction process and which requires multiple pieces of glassware and the addition of acid. It is likely that the species difference is due to differences in the age and/or diet of the mice versus humans. The biological importance of this interspecies variability is not known, and further studies in species-and age-dependent changes in retinoid profi les are warranted.
The results presented show no signifi cant differences in retinoid serum levels after at least 8 h of fasting despite the fact that a previous study discovered decreasing concentrations of at RA and 13-cis RA during fasting ( 25 ). However, the previous study was conducted after a 5-day fast, suggesting that the length of the fasting period has an effect on the retinoid concentrations. One of the subjects in this study showed extremely high levels of 13-cis RA and 4oxo-13-cis RA compared with the mean. It was confi rmed that this individual was not taking exogenous retinoids during the study. Although he was administered prostaglandin F2 ␣ eyedrops, it is unlikely that these eyedrops affected 13-cis RA concentration in serum. The possibility of 13-cis RA and 4oxo-13cis RA contamination in the samples is low because his serum concentration was measured multiple times on different days. In studies involving isotretinoin treatment (80 mg m Ϫ 2 bd), the average concentration of 4oxo-13-cis RA was calculated to be approximately 4.7 M ( 7 ), which is over 100-fold higher than the concentrations observed in this study. In a previous study, the effect of diet on at RA levels in serum was tested, and at RA levels were shown to increase 100% after the consumption of carrot juice ( 34 ). However, no large preparation and analysis. When serum samples were exposed to light in this study, isomerization of the internal standard could be detected (data not shown). Deuterated standards can also be added into the samples during sample collection to control for stability during storage and freeze-thaw. Despite the fact that only one isomer internal standard was used in this study, one may choose to add individual isomer standards to samples to confi rm the retention times of individual analytes.
The retinoid concentrations reported here are likely to represent average concentrations in healthy adult humans. Previous studies monitoring endogenous retinoid concentrations in the serum of men and women have not noted any differences between the two ( 24,25 ). The observed concentrations of at RA are lower and concentrations of 4oxo-13-cis RA are higher than what has been previously reported. The endogenous concentrations measured in this study are approximately 40% lower for at RA and about 80% higher for 4oxo-13-cis RA than previously reported ( 24,25 ). This may be due to improved assay selectivity in comparison to previous studies that used LC-UV, which may not have separated all RA and its metabolite isomers, and matrix interference. As shown in this study, a significant interference was detected eluting close to at RA, but it was mainly detected at the mass transitions relevant for at RA-d 5 . However, this interference would likely confound quantifi cation in LC-UV assays.
The fact that 9-cis RA can be detected in human serum is of interest because 9-cis RA activates RAR and RXR receptors and may have specifi c biological functions. In previous studies, endogenous 9-cis RA has not been detectable in human or mouse serum ( 8,9,29 ), and dosing with alitretinoin (9-cis RA) was required to detect 9-cis RA in human serum ( 8 ). However, 9-cis RA was quantifi able in all but one of the samples in this study, most likely due to the increases in 13-cis RA concentrations were reported. The lack of correlation between at RA concentrations and the concentrations of other RA isomers suggests that measurement of at least at RA and 13-cis RA separately is necessary to characterize an individual's RA status. Because the different isomers have different pharmacological activity, separation of at RA and 13-cis RA from the other isomers is important in biological assays.
The 4oxo-at RA was detectable in over half of the human serum samples, but the concentrations were too low to be quantifi ed. In agreement with published results ( 24,25 ), 4oxo-13-cis RA was the retinoid with highest concentration in human serum. 4oxo-at RA has properties that overlap with at RA, such as activation of RAR ␤ , but it also shows weak activation of RXR ␣ ( 35 ). In addition, treatment of MCF7 breast cancer cells with 4oxo-at RA results in inhibition of proliferation ( 12 ), demonstrating that 4oxo-RA isomers may contribute to the biological activity of RA. Due to the activity of the 4oxo-RA compounds, future studies, including accurate quantifi cation in tissues, are needed to determine their overall biological importance.
The signifi cantly higher concentrations of 4oxo-13-cis-RA in comparison to 4oxo-at RA are unexpected, based on the similar levels of at RA and 13-cis RA in serum. Pharmacokinetic studies to determine the clearance of at RA, 4oxo-at RA, 13-cis RA, and 4oxo-13-cis RA are needed to better understand the reason for the large difference in the ratios between the RA isomers and their corresponding 4oxo-metabolites. These studies would need to include monitoring glucuronidation of the retinoids in serum and urine and the determination of accurate clearance values. Although the 4OH-RA preferentially is glucuronidated at the hydroxy position, RA and 4oxo-RA are known to undergo glucuronidation at the carboxyl function ( 26 ).
In conclusion, a UHPLC-MS method of retinoid measurement in serum was developed and validated. The method enabled quantifi cation of four major RA isomers in serum as well as quantifi cation of 4oxo-13-cis RA. The use of isotope labeled internal standards and the careful evaluation of matrix interference provided increased confi dence for the quantifi cation of the important retinoids. The developed method can be used in future studies to correlate specifi c retinoid concentrations in human tissues to pharmacological effects and in evaluating the relationships between disease states and retinoid concentrations.