Human mitochondrial cytochrome P450 27C1 is localized in skin and preferentially desaturates trans-retinol to 3,4-dehydroretinol

Recently, zebrafish and human cytochrome P450 (P450) 27C1 enzymes have been shown to be retinoid 3,4-desaturases. The enzyme is unusual among mammalian P450s in that the predominant oxidation is a desaturation and in that hydroxylation represents only a minor pathway. We show by proteomic analysis that P450 27C1 is localized to human skin, with two proteins of different sizes present, one being a cleavage product of the full-length form. P450 27C1 oxidized all-trans-retinol to 3,4-dehydroretinol, 4-hydroxy (OH) retinol, and 3-OH retinol in a 100:3:2 ratio. Neither 3-OH nor 4-OH retinol was an intermediate in desaturation. No kinetic burst was observed in the steady state; neither the rate of substrate binding nor product release was rate-limiting. Ferric P450 27C1 reduction by adrenodoxin was 3-fold faster in the presence of the substrate and was ∼5-fold faster than the overall turnover. Kinetic isotope effects of 1.5–2.3 (on kcat/Km) were observed with 3,3-, 4,4-, and 3,3,4,4-deuterated retinol. Deuteration at C-4 produced a 4-fold increase in 3-hydroxylation due to metabolic switching, with no observable effect on 4-hydroxylation. Deuteration at C-3 produced a strong kinetic isotope effect for 3-hydroxylation but not 4-hydroxylation. Analysis of the products of deuterated retinol showed a lack of scrambling of a putative allylic radical at C-3 and C-4. We conclude that the most likely catalytic mechanism begins with abstraction of a hydrogen atom from C-4 (or possibly C-3) initiating the desaturation pathway, followed by a sequential abstraction of a hydrogen atom or proton-coupled electron transfer. Adrenodoxin reduction and hydrogen abstraction both contribute to rate limitation.

Cytochrome P450 (P450) 2 enzymes are rather ubiquitous in nature. The mammalian P450 enzymes are important in drug and carcinogen metabolism (2,3) and also have critical roles in steroid metabolism (4). In addition, P450s have important roles in the metabolism of vitamins A, D, E, and K (5).
Of the 57 human P450 genes, 50 code for proteins that are localized mainly in the endoplasmic reticulum, and seven code for P450s that are nucleus-encoded and transported to mitochondria (5). These seven P450s utilize adrenodoxin (Adx), a ferredoxin, as a source of electrons, instead of the flavoprotein NADPH-P450 reductase used by the P450s in the endoplasmic reticulum ( Fig. 1). In addition, a complement of some of the microsomal P450s is also found in mitochondria, and these P450s also utilize electrons from Adx (6,7). Of the seven mammalian mitochondrial P450s, X-ray crystal structures of three have been reported (11A1, 11B2, and 24A1) (8 -11). Only P450s 11A1 and 11B1 have been extensively characterized in terms of their interactions with Adx and their catalytic mechanisms (12)(13)(14)(15)(16)(17)(18)(19)(20)(21).
In this work, we established the localization of human P450 27C1 in skin. Both 3-and 4-hydroxyretinol were characterized as minor products of P450 27C1. Kinetic isotope effect (KIE) studies with 3-and 4-deuterated retinol showed contributions of C-H bond breaking as a rate-limiting step in catalysis of desaturation and the hydroxylations. Kinetic analysis of other steps in the reaction cycle indicated that P450 27C1 reduction by Adx also contributes in terms of rate limitation.

Identification of P450 27C1 in skin by mass spectrometry and immunochemical methods
MS profiling was done with human liver and skin samples, based on previous work showing P450 27C1 mRNA in the liver (23) and the demonstration of retinol desaturation activity in human skin epidermis (41). Because P450 27C1 is considered to be a mitochondrial protein (due to sequence similarity to P450s 27A1 and 27B1 and the demonstrated reduction by Adx (24)), we utilized tissue homogenates instead of microsomal fractions. Human P450 27C1 is 43% identical to P450 27A1 (a liver enzyme (5)) and 38% identical to P450 27B1 (a kidney enzyme (5)), based on Uniprot.org alignment (http://www.uniprot.org/ align). 3 Purified recombinant P450 27C1 (see supplemental Fig.  S1) was used to guide the selection of migration distances selected for analysis (the M r is less because 66 residues were deleted from the N terminus in the construct that we used (GenBank TM accession number AC027142; March 28, 2000) (and two new ones were added at the N terminus and His 6 at the C terminus) (23)). The P450 27C1 sequence listed in Uniprot Q4G0S4 (and also as NG007986, NM_001001165, and AC027142 elsewhere) begins with MRSVLRQ, is only 372 residues long, and is missing the N-terminal 106 residues of our expressed P450 27C1 (23), based on the original sequence of Nelson (http://drnelson.uthsc.edu/cytochromeP450.html) 3 (95), which denotes the shortened sequences in that website. Initial LC-MS/MS of tryptic peptides from a human skin sample showed the presence of P450 27C1, with 51% sequence coverage in this 372-residue region (supplemental Table S1) (Fig.  3). We hypothesized that the 372-amino acid entry in the databases is an error due to misassignment of the methionine start site and did extensive proteomic analysis with a second skin sample (i.e. different individual) to define the protein found in human skin.
Polyclonal antibodies were raised against P450 27C1 and purified by adsorption with immobilized P450 27C1. Immunoblotting consistently showed the presence of two bands in all five human skin samples examined (Fig. 4A) but no detectable bands in any of the human liver or kidney samples, only background staining (Fig. 4, B and C) (the limit of detection was 0.1 pmol of P450 27C1). (Note that the tissue samples were not from the same five individuals.) Because strong denaturing conditions were required to solubilize human skin samples, conventional methods for antibody pull-down experiments were not possible for validating the specificity of the antibody. Accordingly, we digested the proteins in the gel regions corresponding to antibody binding (Fig. 4)

P450 27C1 oxidation of retinol
protein) Ϫ1 and from 5 to 11 pmol P450 27C1 (mg of protein) Ϫ1 for the lower band. Further proteomic analysis of the two gel regions was done. Peptides corresponding to both bands were identified, validating the specificity of our antibody (Fig. 5). The coverage was 37% in the upper band and ϳ40% in the lower one (depending on exactly where the N terminus is). The identification of the peptide MQTSAMALLAR (in the higher M r band) (    Fig. 5 is 524 -537, and the same peptide is shown again in supplemental Fig. S3. Identified fragment ions are annotated and color-coordinated (blue, y-ions; purple, b-ions). Heavy lines, peptide fragments identified; dashed lines, missed fragment ions. Included is the precursor ϩ3 ion (inset, isotopic distribution) identified and selected for fragmentation. The spectrum was generated using SeeMS for MS 2 data integrated into IDPicker software and Xcalibur Qual Browser for MS 1 data. See also supplemental Table S1.     Fig. 4) was used, and the two immunoreactive bands were located by comparison with a part of the same gel in which the proteins were transferred to a sheet of nitrocellulose and located by immunostaining (Fig. 4). A, predicted full-length amino acid sequence of P450 27C1, with highlighted and underlined peptides identified in the higher-M r band of Fig. 4; see supplemental Figs. S2 and S3 for the N-terminal and most C-terminal peptides identified. B, predicted full-length amino acid sequence of P450 27C1, with highlighted and underlined peptides identified in the lower-M r band of Fig. 4. See Fig. 3 and supplemental Table S1 for more data on peptide analysis with a sample from a different individual.
is. We did identify the peptide SLAAMPGPR (amino acids 70 -78 in Fig. 5B) in an analysis. The C terminus is nearly intact, with the peptide THGLLTPGGPIHVR identified (in both bands; Fig. 5 and supplemental Fig. S3). Therefore the cleavage site is postulated to be near residue 70 (Fig. 5C), in that the M r of the lower band is very similar to that of recombinant P450 27C1 (Fig. 4A), which is 58 residues shorter than the full-length sequence shown (23). No P450 27C1 peptides were detected in a (single) human liver sample when a similar analysis was done under similar conditions (results not shown).

Identification of reaction products
Oxidation of all-trans-retinol to 3-and 4-OH products-We previously reported that human P450 27C1 oxidized retinol to a product that co-eluted with 4-OH retinol (Fig. 2), in addition to the desaturation product 3,4-dehydroretinol (25). However, we did not have any independent evidence for its identity. We synthesized both 3-and 4-OH retinol (supplemental Figs. S4 and S5) and found that both of these alcohols are minor products, based on co-chromatography by HPLC and UV and mass spectral comparisons. Because of the trace amounts of these products formed, we did not determine steady-state kinetic parameters for their production. At a substrate concentration of 7 M, the molar ratio of products formed was 100:3:2 for 3,4-dehydroretinol, 4-OH retinol, and 3-retinol, respectively (Fig. 6). Incubation of 3-or 4-OH retinol with P450 27C1 (in the presence or absence of NADPH-Adx reductase (ADR), Adx, and NADPH) did not show any formation of 3,4-dehydroretinol, indicating that neither of the hydroxy compounds is an intermediate in the formation of dehydroretinol.
Other reactions-In previous studies, we found that retinoic acid and retinaldehyde (oxidized forms of all-trans-retinol) were substrates for P450 27C1, with lower catalytic activities compared with retinol (23,24). We also tested two known biologically important retinyl esters as substrates with P450 27C1. Retinol acetate and retinol palmitate (20 M) were incubated in the presence of P450 27C1, ADR, Adx, and NADPH. After mild KOH hydrolysis of reaction mixtures, 3,4-dehydroretinol was detected in the reaction with retinol acetate but not with retinol palmitate (data not shown).

Individual reaction steps and catalytic mechanism
Substrate binding-We had previously shown that human P450 27C1 forms a tight complex with all-trans-retinol, shifting the heme Soret peak from the low-to high-spin iron form ("Type I" difference spectrum), with an estimated K d of ϳ5.6 nM (25). The rate of binding was measured using two different approaches. Equimolar concentrations of P450 27C1 and the substrate all-trans-retinol were mixed in a stopped-flow spectrophotometer (Fig. 7A), and the data (Fig. 7B) were fit to a second-order equation, in that the binding reaction is first-order in both P450 and substrate (42). The estimated k on value was 1.9 ϫ 10 5 M Ϫ1 s Ϫ1 (Fig. 7C). In an alternative approach, first-order rates were estimated with varying concentrations of substrate and plotted versus the substrate concentration (43) to obtain a k on value of 1.6 ϫ 10 5 M Ϫ1 s Ϫ1 and k off of 0.022 s Ϫ1 (Fig.  7D). This k off value is subject to considerable error because of the need to use Ն1 M P450 in these stopped-flow measurements. A trapping assay was used in which a 1:1 molar mixture of 2 M P450 27C1 and trans-retinol was mixed with 20 M ketoconazole, an inhibitor that was measured to have a K d of 7 Ϯ 4 M and gives a spectrally distinct complex with P450s ("Type II") (44). The observed rate was 0.017 s Ϫ1 , or 1 min Ϫ1 (compared with a rate of 1,200 Ϯ 240 min Ϫ1 for ketoconazole binding to P450 27C1 in the absence of dehydroretinol; results not shown).
Reduction by Adx-Adx forms a complex with ADR and also has been known to be involved in electron delivery to mitochondrial P450s (45). Tight complexes with Adx have also been demonstrated with mitochondrial P450s 11A1 (12) and 11B1 (13), involving changes of low-to high-spin iron. However, we did not observe any spectral perturbation of P450 27C1 by the addition of Adx (in the absence or presence of equimolar amounts of all-trans-retinol), although the P450 27C1 activity is clearly dependent upon Adx for electrons (data not shown).  Preliminary assays were done by mixing photoreduced Adx with ferric P450 27C1 under an anaerobic CO atmosphere (supplemental Fig. S7). Considerable evidence for Adx acting as a mobile electron carrier has been published with another mitochondrial P450, P450 11A1 (13). In our experiments, the estimated rate of reduction (at 23°C) was 1.3 Ϯ 0.3 min Ϫ1 in the absence of the substrate retinol and 3.6 Ϯ 0.3 min Ϫ1 in the presence of 5 M retinol. We concluded that the presence of the substrate enhanced the rate of reduction, and further measurements were all done in the presence of substrate.
In a system with a 1:10:1 molar ratio of ADR/Adx/P450 27C1, the rate of reduction (in the presence of retinol) was 21 Ϯ 1 min Ϫ1 at 37°C (Fig. 8). (Rates of reduction of ADR by NADPH and Adx by ADR have been measured previously (32 and 5.4 s Ϫ1 , respectively, at 15°C (21)). (In the steady-state kinetic assays, the enzymes were much less concentrated than in these anaerobic experiments, with 7.5-fold less ADR, 3-fold less Adx, and 75-fold less P450.) Product release-A trapping assay was used in which a 1:1 molar mixture of 2 M P450 27C1 and dehydroretinol was mixed with 20 M ketoconazole (see above). A first-order rate of 54 Ϯ 5 min Ϫ1 was estimated for the release of the product dehydroretinol at 23°C.
Another important approach to determining whether a step following product formation is rate-limiting is to establish whether or not there is a kinetic burst (43). When we extrapolated to time 0, no burst was seen (Fig. 9). We conclude that neither product release nor any step following product formation is rate-limiting.

KIEs
Deuterated retinols were synthesized (with Ͼ95% atomic excess; see supplemental material, including supplemental Fig.  S6) and used to test the hypothesis that C-H bond breaking is a rate-limiting step in the desaturation or hydroxylation of retinol. We do not have information about the stereospecificity of hydrogen abstraction from retinol (which is pro-chiral at C-3 and C-4), so the dideuterated retinol derivatives 3,3-d 2 and 4,4-d 2 retinol were prepared, as well as tetradeuterated 3,3,4,4-d 4 retinol. (The synthesized 4,4-d 2 and 3,3,4,4-d 4 retinols also had deuterium at the side chain carbinol position (C-15), but this is irrelevant in the assays in that no oxidation occurs there.)   The KIE experiment that is most informative about a ratelimiting step is a noncompetitive intermolecular design, in which k cat and K m are measured with varying concentrations of protiated and deuterated substrates and used to calculate D V and D (V/K), defined as H k cat / D k cat and ( H k cat / H K m )/( D k cat / D K m ), respectively, with H and D indicating parameters measured with protium and deuterium at a site (1,46,47). 4 Relatively small but reproducible changes were observed in the rates of desaturation when using deuterium-labeled substrates; low KIEs were observed for desaturation with 3,3-d 2 , 4,4-d 2 , and 3,3,4,4-d 4 retinol (Fig. 10). Most significantly, a KIE of ϳ2 ( D (V/K)) was observed when deuteriums were present at C-3 of the substrate (3,3-d 2 and 3,3,4,4-d 4 ), whereas the D (V/K) for 4,4-d 2 retinol was calculated to be only ϳ1.5.
We considered the possibility that deuterium scrambling might occur after abstraction of a hydrogen atom from C-3 via a radical intermediate (or, alternatively, from a carbocationic intermediate that might be formed from further electron transfer (37)) ( Fig. 2). Accordingly, we analyzed the desaturation and 4-hydroxylation products of 4,4-d 2 retinol by LC-HRMS. The masses detected at 100,000 resolution were consistent with those of the expected products, providing evidence against migration of deuterium from the C-4 to the adjacent C-3 site (Fig. 12).

Discussion
We have established that P450 27C1 is located in the skin, is involved in the desaturation of retinol, and also forms minor amounts of 3-OH and 4-OH retinol. Within the catalytic cycle, the reduction of ferric iron and C-H bond-breaking steps both contribute to limiting the reaction rates, but substrate binding and product release do not.
Skin is a major organ, in terms of collective body mass, and metabolism there is of interest with many drugs (applied cutaneously) and with a number of endogenous chemicals. Regarding P450s, there is some evidence that P450s 1A1, 1A2, 1B1, 2A6, 2B6, 2C18, 2C19, 2D6, 2E1, 2J2, 2R1 2S1, 2U1, 2W1, 3A4, 3A5, 4B1, 4F2, 4F3, 4F22, 11A1, 17A1, 24A1, 26A1, 26B1, 26C1, 27A1, and 51A1 are expressed in human skin samples or in derived cells (in culture) (48 -54). However, almost all of the data are at the mRNA level. Only in a few cases has protein expression been reported, and then only at the immunohistochemical or enzyme activity level (52,54). Previous studies by Törmä and Vahlquist (41) demonstrated the presence of retinol 3,4-desaturation activity in human skin epidermis, as well as in human keratinocytes, melanoma cells, and HeLa cells (55). In the present study, we identified P450 27C1 in human skin samples using proteomic and immunochemical methods (Figs. 3-5) but did not find P450 27C1 in the human liver and kidney samples that we analyzed (Fig. 4, B and C). Earlier studies with an oligonucleotide probe selective for P450 27C1 (compared with P450s 27A1 and 27B1) indicated the presence of P450 27C1 mRNA in several human tissues, including the liver, kidney, and pancreas (23). Skin mRNA had not been included in that study. One possibility for the apparent difference in tissue localization is that the oligonucleotide probe used previously (23) was not specific, although that seems unlikely in the context of the selectivity demonstrated versus P450 27A1 and 27B1 sequences. An alternative possibility is that mRNA is formed in several tissues but not translated into protein, which would not be surprising in light of the overall concordance between mRNA and protein expression (56). The possibility also exists that the mRNA could be alternatively spliced or transcribed in  other tissues. In reviewing the Human Protein Atlas (http:// www.proteinatlas.org), 3 the highest level of mRNA expression of CYP27C1 is in skin (57). Some other sites in which higher than baseline levels were seen include the uterine cervix, urinary bladder, and vagina (57). Trace but apparently real levels of CYP27C1 mRNA were found in many other tissues, including those that we reported previously (23). Our conclusion is that we were observing these low levels of expression in our earlier data (23), which was misleading in that skin was not included. Although expression in retinal epithelium is critical in the biology of dehydroretinol in fish (24), the literature does not indicate the presence of dehydroretinoids (vitamin A 2 ) in human eyes (57). 3,4-Dehydroretinoic acid is also formed by human P450 27C1 (25), but the biological activities of retinoic acid and 3,4-dehydroretinoic acid are similar in human keratinocytes (58). In human skin, dehydroretinol accounts for ϳ25% of the retinoid pool (59,60), but the function remains unknown. Levels of dehydroretinol are elevated in psoriasis (60) and squamous cell carcinoma and keratocanthoma (61), and the desaturation activity is induced by exposure to UVA/B light (62). In our analyses, the levels of the two forms of P450 27C1 only varied ϳ2-fold among the five human samples analyzed (Fig.  4A). This level of variation is similar to that commonly seen for P450s involved in normal physiological processes (63) and differs from the wide variation in levels of the P450s that are involved in the metabolism of xenobiotic chemicals (5, 63).
We utilized the original human CYP27C1 sequence (Gen-Bank TM AC027142; March 28, 2000) (Fig. 5) in consideration of what is actually expressed in human skin. The Uniprot Q4G0S4 372-residue sequence is clearly wrong. No P450 this short has ever been identified to our knowledge, even in bacteria, and it would not be expected to be catalytically active. The 482-amino acid version that we expressed (25) and used in catalytic assays (Figs. 6 -9) is similar in M r to the smaller protein found in the skin (Fig. 4A). The larger protein (Fig. 4) contained the peptides that correspond to residues 1-537 (Fig. 5A), in that the peptides for residues 1-11 and 524 -537 were characterized. The smaller protein contains residues 70 -78 (Fig. 5B), so this peptide must be near the N terminus, but we cannot determine exactly where P450 27C1 was cleaved. We presume that the smaller protein is catalytically active, in that its size is similar to that of the recombinant protein (Fig. 4A). Which of these proteins is present in mitochondria is currently unknown. The program MitoFATES (64) (mitf.cbrc.jp/MitoFates/cgi-bin/top.cgi) 3 predicts a TOM20 recognition motif at residues 9 -14 and an MPP cleavage site at residue 66, the latter of which is what we had serendipitously designed in our original Escherichia coli expression construct, leaving an N terminus similar to that observed for murine P450 27B1 (23). Further work is needed, however, to characterize the transport of a specific sequence into mito-

P450 27C1 oxidation of retinol
chondria. Our current results indicate the methionine at residue 1 as the start site for protein translation (Fig. 5A).
Desaturation is a relatively infrequent event in P450 oxidations, at least compared with hydroxylation, and we were interested in defining rate-limiting steps. Very little work has been published on rate-limiting steps in reactions by the seven mitochondrial P450s (11A1, 11B1, 11B2, 24A1, 27A1, 27B1, and 27C1), including KIE studies (5, 65). The literature indicates that the KIEs observed for some microsomal P450 desaturation reactions are relatively high (33,66), depending on the experimental design. We measured D V and D (V/K), the latter of which is considered the most useful KIE experiment to establish to what extent C-H bond breaking is rate-limiting in a reaction (1,46,47). The KIE values are relatively low, in terms of the values seen in some (but certainly not all) P450 oxidations (47), but nevertheless the KIE values are significantly greater than unity, indicating some contribution to rate limitation.
Several other steps in the catalytic cycle were also measured. The rate of substrate binding was measured to be 1.6 -1.9 ϫ 10 5 M Ϫ1 s Ϫ1 , using two approaches (Fig. 7). The reaction appeared to be monophasic but is slow in terms of what is expected for a diffusion-limited reaction (67). Further, the apparent rate constant is at least 1 order of magnitude less than those measured for substrate/ligand binding to some human microsomal P450s (68 -70) and bacterial P450 101A1 (P450 cam ) (71). The slower rate measured for P450 27C1 may reflect the existence of a slow step, following initial binding, that leads to the spin state change, as in the case of P450 3A4 (69,72). Regardless, substrate binding and product release are considerably faster than the overall oxidation, and the absence of burst kinetics already indicates that a step following product formation cannot be ratelimiting (Fig. 9). The rate of reduction of ferric P450 27C1 is somewhat faster than overall catalysis (Fig. 8), and we propose that it can contribute in part as a rate-limiting step. Preliminary analysis of a simplified KinTek Explorer mechanism (supplemental Fig. S8) indicated that the simulated rate of the overall reaction was increased when the rate constant for the reduction step was raised above the rate observed in our work, in support of the view that the reduction of ferric P450 27C1 is a partially rate-limiting step. The reduction rate was stimulated in the presence of the substrate retinol (ϳ3-fold) (supplemental Fig.  S7); sometimes P450 reduction rates are stimulated by substrates and sometimes not (73). It is of interest to note that Tuckey et al. (19) concluded that the reduction of mitochondrial P450 11A1 by Adx was rate-limiting in human placenta, as well as in reconstituted systems (74). To our knowledge, of the seven mitochondrial P450s, only with P450 11A1 has a ferrous-O 2 complex been characterized (and at low temperature) (17). We attempted to characterize such a P450 27C1 complex by mixing photoreduced P450 27C1 with O 2 but were not successful. It would be desirable to measure the rate of Adx reduction of this complex, but no such mitochondrial P450 reaction has been characterized in detail yet.
The KIE values for desaturation (Fig. 10) and 4-hydroxylation are relatively low. We have not estimated the intrinsic KIE, D k, in either case but suspect that it may be considerably higher. Thus, the values of 1.5-2.3 we measured for D (V/K) (Fig.  10) may be attenuated due to the tight binding of retinol (previously estimated K d of 5.6 nM (24) and k off of 1 min Ϫ1 (this work)). This represents a case of high forward commitment to catalysis (C f , using the conventions of Northrop (1) and as elaborated by Northrop (1) and elsewhere (47)), (when there is no commitment to reverse catalysis). Thus, although it might not be intuitive, raising C f (i.e. decreasing the substrate k off ) lowers the observed D (V/K) value for a reaction such as a P450-catalyzed oxidation, which is essentially irreversible. Therefore, the D (V/K) values measured here are probably lower than the D k and may not reflect a higher KIE. The KIE and other deuterium experiments provide some insight into steps associated with the oxidation of retinol (Figs. 10, 11, and 13). We propose that abstraction of a C-4 hydrogen atom is the first step in both 4-hydroxylation and desaturation, in that both processes had low KIEs (Fig. 10). 3-Hydroxylation is a separate reaction and begins with abstraction of a C-3 hydrogen atom. This is an unfavorable reaction, possibly because, unlike the C-4 abstraction, it does not yield an allylic radical. The measured KIE values were much higher, and hydrogen atom abstraction is clearly rate-limiting for the minor 3-hydroxylation reaction (Fig. 11F). We were unable to measure a K d for binding of the product dehydroretinol to P450 27C1, but the measured k off rate was 50 min Ϫ1 , and no burst kinetics were seen (Fig. 10). HRMS results indicate that the 3and 4-position radicals do not equilibrate (Fig. 13). The "metabolic switching" to yield more 3-OH retinol takes place with the substrate, not the C-4 allylic radical. In other words, slowing C-4 radical formation yields more retinol for the C-3 abstraction. C-3 abstraction is a minor pathway and, accordingly, not substantial enough to yield a noticeable shift to 4-hydroxylation or desaturation. Apparently, C-3 hydrogen abstraction is slower than Adx reduction. In the C-4 pathways, Adx reduction is slow enough to contribute to reaction rate limitation.

P450 27C1 oxidation of retinol
desaturases are important in oxidation of fatty acids and some other alkanes, and a few residues have been shown to alter the balance of desaturation and hydroxylation (27,77,78). Some mammalian P450s are also capable of (-)desaturation of fatty acids (34). The bifurcation of desaturation and hydroxylation has also been considered at a theoretical level (27,37,38). Kumar et al. (37) concluded that desaturation involved carbocationic intermediates, generated by further electron transfer from an initial carbon radical. Desaturation is proposed to occur because of stereoelectronic inhibition of the oxygen rebound process (37). The conclusion that desaturation occurs only through a carbocationic intermediate (37) may or may not be correct, and our results do not address this issue directly. It is of interest to note, however, that no shifts of the 3-or 4-deuterium atoms were observed (Fig. 12). This lack of allylic scrambling does not necessarily negate the hypothesis about carbocations. Ji et al. (38) did not consider carbocations but concluded that the difference in O-H versus C-O bonds formed is critical in the outcome, and Cooper et al. (27) also favored a radical mechanism instead of carbocationic. This is concluded to be important, and -conjugation, which occurs in the desaturation of retinol, lowers the excitation energy. In these theoretical papers (37,38), the difference between hydroxylation and desaturation can ultimately be rationalized in the context of the structure of the enzymesubstrate complex.
One alternative mechanism of desaturation that can be considered is shown in Fig. 14, in which the initial reaction is an epoxidation of the cyclohexenyl ring, followed by two basecatalyzed abstractions of the C-3 and C-4 hydrogens. This mechanism could be consistent with all of our data on isotope effects, and there is no single experiment that we have done that would rule this out. We are inclined against this mechanism in that the epoxide would be expected to be relatively stable, as in the case of cyclohexene oxide (79) and vitamin K epoxide (80). Additionally, we did not find any additional LC-MS peaks with m/z 285 (Fig. 6). If this were the mechanism, then the hydroxylation reactions would be separate events, not as shown in Fig.  13, and discussion of competition between hydroxylation and desaturation would be a moot point.
The effect of limited changes in amino acids on the partitioning between desaturation and hydroxylation has been appreciated in non-heme iron monooxygenases for some time (77,78). A similar balance has been observed in two bacterial P450s (102A1 and 199A4) by Whitehouse et al. (39) and Bell et al. (40), respectively, in which a 5-amino acid change in P450 102A1 or a single-amino acid change in P450 199A4 (F185V) could change the ratio of hydroxylation to desaturation. Although no three-dimensional structure of a P450 27C1-retinoid complex is available yet, we believe that it might be instructive as to the proclivity toward desaturation and suggest specific site-directed mutagenesis experiments.
In summary, we have characterized several aspects of human P450 27C1, a retinol desaturase found in skin. The enzyme also catalyzes 3-and 4-hydroxylation as minor events, and the balance of these reactions is probably due to the allylic nature of the reaction and to the juxtaposition of the substrate in the active site. Small but significant deuterium KIEs were calculated, and metabolic switching of minor products was observed, leading us to conclude that abstraction at C-4 may occur first. Rate limitation at other key steps in P450 oxidation of retinol was considered, including substrate binding and product release, as well as reduction. We conclude that at least two reaction steps, C-H bond breaking and Adx reduction of the ferric enzyme, contribute to limiting rates of the desaturation of retinol by P450 27C1.

Tissue samples
All work was done with the approval of the Vanderbilt institutional review board, which considers these particular studies exempt if sample donors are not identified. The human liver and kidney samples were obtained through the Nashville Regional Organ Procurement Agency. Two human skin samples, from the abdominal area of an adult male (sample W, code A16-6) and female (sample P, code A16-4), were obtained from Prof. Joseph Corbo (Washington University School of Medicine, St. Louis, MO) (these were used in Fig. 4). A third sample, from the breast area of an adult male, was also obtained from the same source and used in the work presented in Fig. 3. Three other skin samples (all from adult female breast) were obtained from the tissue procurement source at Vanderbilt University Medical Center and are coded 37, 51, and 57 (these were used in Fig. 4, along with samples W and P from Washington University). The identities of the individuals were not disclosed.
Tissue samples were homogenized using a two-step procedure. The first step was done with an IKA disperser homogenizer (on ice) (4 ϫ 10 s). The resulting mixture was then processed with a motorized glass-Teflon Potter-Elvehjem tissue homogenizer (on ice) (4 ϫ 15-s pulses, with ϳ4 up-and-down strokes during each). Approximately 400 mg of human liver and kidney was homogenized in 10 ml of 50 mM potassium phosphate (pH 7.4) buffer. Much more disruptive methods were required for human skin samples: ϳ400 mg of human skin was homogenized in 6 ml of lysis buffer (6 M urea, 2 M thiourea, 4% (w/v) sodium CHAPS (w/v), 0.1% SDS (w/v)). Unhomogenized skin tissue was filtered out through glass wool, and the homogenates were centrifuged at 2,000 ϫ g for 10 min, and the resulting supernatant was centrifuged at 16,000 ϫ g for 20 min (at 4°C). The supernatant was collected and precipitated with 5 volumes of cold acetone and chilled overnight at Ϫ20°C. The precipitated proteins were pelleted at 16,000 ϫ g for 20 min (at 4°C), washed once with chilled 50% acetone (aqueous) with vortex mixing, and centrifuged again at 16,000 ϫ g for 10 min. The resulting pellets were resuspended in 50 mM Tris buffer (pH 7.4) containing 2% SDS (w/v) and heated at 95°C for 5 min. The concentrations of protein in all tissue samples were estimated using a BCA assay (Bio-Rad) or Pierce 660-nm protein assay (with detergent compatibility reagent) (Thermo Fisher Scientific).

Antibodies
Rabbit antibodies were raised by Cocalico Biologicals (Stevens, PA). Three rabbits were immunized with a total of 6 mg of purified P450 27C1 (see above; further purified by by DEAE and hydroxylapatite chromatography to remove some impurities). The rabbits were injected with purified protein in PBS, followed by three boosts over a period of 49 days. A test bleed was collected, and antibody strength was tested using an immunoblot with purified P450 27C1 as an antigen. Unless further boosting was required for more antibody production, the rabbits were then boosted a final time before exsanguination. The sera were compared, and the best serum was used based on specificity and titer.
Rabbit anti-27C1 antibody was purified by affinity chromatography with purified recombinant protein (84). Purified 27C1 was coupled to CNBr-activated Sepharose (GE Healthcare) according to the manufacturer's instructions. The coupled beads were incubated with diluted antiserum from rabbit number 497 (1:10 dilution in PBS, v/v) at room temperature overnight with end-over-end rotation. The suspension was washed extensively with PBS until the A 280 was Ͻ0.01. Bound antibody was eluted with 0.20 M glycine-HCl buffer (pH 2.5) in 1-ml fractions and neutralized with 1.0 M Tris-HCl buffer (pH 8.0). Fractions containing antibody (monitored by A 280 measurements) were combined, and protein concentration was measured using a BCA assay (Pierce/Thermo Fisher Scientific).
Immunoblotting (Western blotting) of tissue homogenates was carried out using the purified 497 antibody and reagents purchased from LI-COR (Lincoln, NE). Proteins from skin (20 g), liver (50 g), and kidney (50 g) were separated using SDS-polyacrylamide gel electrophoresis and then transferred to nitrocellulose membranes (85) and stained with SyPro Ruby Blot Stain (Thermo Fisher Scientific) for total protein visualization using a PharosFX Plus molecular imager (Bio-Rad). Higher protein loads of the skin homogenates led to poor resolution from the keratins. The membranes were blocked with LI-COR Odyssey blocking buffer. Immunoblots were optimized with a 1:200 dilution (v/v) of the immunopurified rabbit antiserum (ϳ1 g of protein/ml) and a 1:10 4 dilution (v/v) of goat antirabbit 800CW near-infrared (IR) dye for the secondary antibody incubation (LI-COR protocol). Fluorescence signals were detected and quantified using a LI-COR Odyssey infrared imaging system (Thermo Fisher Scientific).

Proteomic analysis of skin homogenates
Proteins in the skin homogenate samples (prepared above) were separated by SDS-polyacrylamide gel electrophoresis (10% acrylamide, w/v), and bands were visualized with Coomassie SimplyBlue SafeStain (Thermo Fisher Scientific). Recombinant P450 27C1 was used to guide the selection of migration distances selected for analysis (the M r is less because it is 58 residues shorter (23)) as well as immunoblots done on portions of the same gel (Fig. 4). The molecular weight region corresponding to recombinant P450 27C1 and the higher region, which was suspected to contain a higher-M r P450 27C1 protein, was excised and processed for in-gel digestion. In-gel trypsin digestion was performed as described (86). Extracted peptides were analyzed on a nanoLC Ultra system (Eksigent Technologies, Dublin, CA) interfaced with an LTQ Orbitrap XL mass spectrometer (Thermo Fisher Scientific). Approximately 2 g of peptides were reconstituted in 0.1% HCO 2 H and pressure-loaded (1.5 l min Ϫ1 ) onto a 360-m outer diameter ϫ 100-m inner diameter microcapillary analytical column packed with Jupiter octadecylsilane (C18) resin (3 m, 300 Å; Phenomenex) and equipped with an integrated electrospray emitter tip. Peptides were then separated with a linear gradient composed of 0.1% HCO 2 H in H 2 O (solvent A) and 0.1% HCO 2 H in CH 3 CN (solvent B) as follows: 2% B held from 0 to 10 min and increased from 2 to 45% B from 10 to 55 min at 500 nl P450 27C1 oxidation of retinol min Ϫ1 . The spray voltage was set to 2.0 kV, and the heated capillary temperature was set to 200°C. HCD MS/MS spectra were recorded in the data-dependent mode using a Top 2 method with an inclusion list containing m/z values corresponding to P450 27C1 tryptic peptides (Fig. 5) determined in silico with Skyline software (87). MS1 spectra were measured with a resolution of 70,000, an AGC target of 1e6, and a mass range from m/z 300 to 1,500. HCD MS/MS spectra were acquired with a resolution of 7,500, an AGC target of 1e5, and normalized collision energy of 35. Peptide m/z values that triggered MS/MS scans were dynamically excluded from further MS/MS scans for 20 s with a repeat count of 1. In some cases, a Q Exactive mass spectrometer (Thermo Fisher Scientific) was used for parallel reaction monitoring analysis. Tryptic peptides from the recombinant P450 27C1 protein were used to determine retention time windows to monitor for the skin homogenate. Retention time windows were set at Ϯ1 min for each peptide in the C-terminal region. Peptides from the previously undetermined N-terminal region were monitored for the entirety of the LC gradient. MS1 spectra were measured with a resolution of 70,000, an AGC target of 3e6, and a mass range from m/z 300 to 1,500. HCD MS/MS spectra were acquired with a resolution of 17,500, an AGC target of 5e5, and NCE of 27. For C-terminal peptides, a maximum injection time of 200 ms was used with a loop count of 10. An isolation width of m/z 1.5 with an offset of m/z 0.1 was used. For N-terminal peptides, a maximum injection time of 120 ms was used with a loop count of 18. Skyline software (87) was used to monitor both precursors and products from the parallel reaction monitoring analysis. Raw data files were analyzed using MyriMatch (88) against a decoy protein database consisting of forward and reversed human Uniprot/Swissprot database (version 20160620). The precursor ion mass tolerance was 10 ppm, and the fragmentation tolerance was 20 ppm for the database search. Methionine oxidation (15.9949 Da, dynamic) and cysteine modifications by iodoacetamide (carbamidomethyl, 57.0214 Da, static) were searched as modifications. The maximum Q values of peptide spectrum matches were adjusted to achieve either a peptide or a protein false discovery rate Ͻ5% using IDPicker software (version 3.1.642.0) (89).

Spectroscopy
UV-visible absorbance spectra were recorded using OLIS/ Cary 14 and OLIS/Aminco DW-2 spectrophotometers (On-Line Instrument Systems, Bogart, GA). Stopped-flow measurements were made with an OLIS RSM-1000 instrument operating in the rapid-scanning mode (1,000 spectra s Ϫ1 , with signal averaging for runs Ͼ4 s). In general, the slit widths were 1.24 mm (nominal 8-nm bandwidth), the gratings were 600 lines mm Ϫ1 (spectral width 300 nm), and the wavelength center was 450 nm. The temperature was controlled with a jacketed water bath and an external Jubalo F-25 circulating water bath (temperatures are noted for individual experiments). Collected spectra were analyzed with single-variable decomposition (SVD) software (OLIS Global Works) to estimate rates, using appropriate models in the manufacturer's software. The number of replicate traces and the S.D. are indicated for each experiment.
Anerobic kinetic experiments were conducted by loading the stopped-flow instrument with samples from tonometers that had been degassed and placed under a CO atmosphere as described in detail elsewhere (46,90,91). The syringes of the stopped-flow apparatus had previously been scrubbed with an anaerobic solution of a mixture of protocatechuate dioxygenase and 3,4-dihydroxybenzoic acid (92,93).

Incubations with P450 27C1
Incubations were done at 37°C in a shaking water bath. Typically, incubations for steady-state reactions were based on work done with other mitochondrial P450s and included 0.02 M P450 27C1, 5 M Adx, and 0.2 M ADR in 50 mM potassium phosphate buffer (pH 7.4) (0.5-ml final volume) (25), along with varying concentrations of all-trans-retinol and an NADPHgenerating system (94). All procedures with retinoids were done in amber glass vials under dim light because of the photosensitivity. Retinol was dissolved in ethanol, and the final concentration of ethanol in the reactions was Յ1% (v/v). We found that the presence of 30 M L-␣-dilauroyl-sn-glycero-3-phosphocholine did not enhance the initial reaction rate but yielded a longer period of linearity of product formation. Typical reactions were done for 2 min and quenched by the addition of 2 volumes of tert-butyl methyl ether containing 20 M butylated hydroxytoluene (to prevent chemical oxidation). Samples were resuspended in 50% ethanol (aqueous, v/v) for chromatographic separation (100 l).

Chromatography
HPLC with UV-visible absorbance detection was routinely used for measurement rates of oxidation of retinol in steadystate kinetic assays. Products were generally separated using a Waters Acquity UPLC system and an Acquity UPLC-BEH octadesylsilane (C18) column (50 mm ϫ 2.1 mm, 1.7 m) with a solvent system consisting of 4.9% CH 3 CN, 0.1% HCO 2 H, and 95% H 2 O (solvent A) and 95% CH 3 CN, 0.1% HCO 2 H, and 4.9% H 2 O (solvent B) (all v/v). Samples were injected with a 20-l loop, and a linear gradient was used for analysis of 3,4-dehydroretinol: 60 -66% B (v/v) over 10 min at a flow rate of 0.3 ml min Ϫ1 . For detection of all P450-dependent oxidation products the following was used: 40 -45% B over 5 min followed by 60 -66% (all v/v) over 10 min at a flow rate of 0.3 ml min Ϫ1 . Similar UPLC systems were used for mass spectrometry measurements for enzyme incubations.

LC-MS experiments for detection of retinol oxidation products
LC-MS analyses of retinol oxidation products were done with samples introduced following UPLC as described above. For detection of oxidized products, a Thermo LTQ mass spectrometer was used; for high resolution mass measurement, a Thermo LTQ-Orbitrap mass spectrometer was required at 100,000 resolution. In both cases, the mass spectrometer was operated in the atmosphere pressure chemical ionization (APCI) positive ion mode and was tuned with authentic alltrans-retinol. The tune settings are as follows: sheath gas flow rate, 40; auxiliary gas flow rate, 10; sweep gas flow rate, 0; capillary temperature, 300°C; APCI vaporizer temperature, 350°C; source voltage set to 6 kV; source current, 5 A; capillary voltage, 10 V; tube lens, 55 V.
Author contributions-K. M. J. synthesized the retinoids and did all of the enzymatic assays. T. T. N. P. purified P450 27C1 and ADR, purified and characterized the antibody, and performed the immunoquantitation experiments. K. M. J. and M. E. A. did the proteomic analysis of P450 27C1 expression in tissues, with K. M. J. F. P. G. conceived the project, purified Adx, and did the anaerobic stoppedflow measurements with K. M. J. F. P. G. and K. M. J. wrote most of the manuscript.