Location and Functional Significance of Retinol-binding Sites on the Serine/Threonine Kinase, c-Raf*

Redox activations of serine/threonine kinases represent alternate pathways in which vitamin A plays a crucial co-factor role. Vitamin A binds the zinc finger domain of c-Raf with nanomolar affinity. The retinoid-binding site has been mapped within this structure by scanning mutagenesis. The deduced contact sites were found anchored on Phe-8, counting from the 1st conserved histidine of the zinc finger. These sites agreed with contact amino acids identified by computational docking. The boundaries of a related binding pocket were identified by mutagenesis and partially confirmed by docking trials in the protein kinase C-α C1A zinc finger. They comprised Phe-7, Phe-8, and Trp-22. This trio was absent from the αC1B domain, explaining why the latter did not bind retinol. Reconfiguring at a minimum the two corresponding amino acids of αC1B, Thr-7 and Tyr-22, to conform to αC1A converted this domain to a binder. Deletion of the predicted retinoid-binding site in the full-length molecule created a mutant c-Raf that was deficient in retinol-dependent redox activation but fully responsive to epidermal growth factor. Our findings indicate that ligation of retinol to a specific site embedded in the regulatory domain is an important feature of c-Raf regulation in the redox pathway.

The history of vitamin A research contains a medley of observations concerning widespread physiological roles of retinoids other than the well known functions of retinoic acid in transcription and retinaldehyde in vision (reviewed in Ref. 1). Most convincing for the non-nuclear functions of vitamin A are arguments pointing to the evolution of an elaborate retinoid biochemistry and biology in eukaryotic organisms (2), predating by far the advent of retinoid retinoic acid receptors and retinoid X receptors and the conservation of the vitamin A metabolites, along with the requisite enzymes, from insects to man. Furthermore, essentially all nucleated cells of higher vertebrates store vitamin A in the form of retinyl esters for ready retrieval and conversion to a variety of metabolites. Because these retinoid products, more often than not, exclude retinoic acid, the question arises as to their purpose.
The multitude of defects caused by nutritional vitamin A deficiency, not completely reversible by retinoic acid and ranging from multiple developmental abnormalities (3,4), to immune defects (5)(6)(7)(8), and to male sterility (9, 10), was not explainable by a single non-nuclear target. In fact, multiple molecular targets emerged when the serine/threonine kinases were found by us to harbor high affinity retinoidbinding sites (11). These were encoded within the cysteinerich domains of several PKC 1 isoforms and c-Raf and overlapped with known structures intimately involved with kinase regulation. This is where lipid second messengers bind and activate the conventional and novel PKC isoforms (12)(13)(14) and where, in c-Raf, a crucial half-site is located for recognition of the activating GTP/Ras protein (15)(16)(17). Nevertheless, the purpose of retinoid-binding sites remained elusive as the classical receptor tyrosine kinase pathways leading to PKC and c-Raf activation operated independently of vitamin A. With the discovery of the alternate pathway of serine/threonine kinase activation via reactive oxygen species (18,19), this situation changed. We could show that vitamin A itself served as an essential co-factor in redox activation of both PKCs and c-Raf (20,21). The hypothesis was developed that the binding of vitamin A to the cysteinerich domains was required for the controlled oxidation of defined cysteines (22). When absent because of nutritional deficiency or when experimentally displaced by non-functional retinoid antagonists, such as anhydroretinol, PKC, and c-Raf, activation by reactive oxygen species was severely compromised.
Jakob et al. (23) proposed that in bacteria, cysteine-rich domains are organized into a zinc finger fold that functions as a molecular hinge. The mammalian counterpart, although more complex, also forms a composite zinc finger (24,25). Upon oxidation, the latter has been shown to relinquish the central zinc ions, allowing potentially a conformational change (26,27). Bound retinol accelerated this oxidation process. Surprisingly, phorbol ester and diacylglycerol also caused the release of Zn 2ϩ from the PKC␣ zinc finger (27). Thus, the two chemically dissimilar activators, reactive oxygen species and phorbol ester, nevertheless produced the same outcome: disassembly of the zinc finger, and presumably, a conformational shift. Unfolding has long been postulated as a prelude to the release of the auto-inhibition that the regulatory domain imposes on the catalytic domain (28,29).
Although the role as co-factor in redox regulation presents a conceptual advance to explain vitamin A action, explorations of biological significance have not kept pace. The reason can be traced to the multitude of target molecules (i.e. the PKC and Raf families) and attendant vitamin A-dependent signal chains. To overcome the difficulties stemming from simultaneous engagement by reactive oxygen species of the diverse vitamin A-dependent signal pathways, we have devised a genetic approach that is predicated on the elimination of retinol-binding sites in select signaling molecules. This was accomplished by mutation of critical contact amino acids of the zinc finger domain. As reported here for the example of c-Raf, the impairment of binding for retinol was paralleled by the selective loss of redox regulation, whereas kinase activation by the classic hormone receptor signals was retained. Additionally, introducing three critical contact residues, copied from the PKC␣ C1A domain, into the natural non-binding C1B domain conferred retinol binding capacity.
Cell Culture-COS-7 cells were grown and maintained in Dulbecco's modified Eagle's medium high glucose supplemented with 10% fetal calf serum and L-glutamine without antibiotics.
Transfection and Cell Activation-COS-7 cells were transfected by the calcium phosphate method as described (20). To deplete endogenous retinol, cell cultures were first incubated for 30 min with 1 M anhydroretinol and then cultured with 2 ml of retinoid-free, serum-free, phenol red-free Dulbecco's modified Eagle's medium high glucose for 2 1 ⁄2 days (to deplete retinyl esters (30)). Where indicated, cultures were restored to vitamin A sufficiency by incubation for 20 min with 1 M retinol with 0.1% bovine serum albumin as carrier. Subsequently, cells were activated by UV-irradiation for 2 min at 400 milliwatt/cm 2 using 312-nm wavelength as described (20), or alternatively, by the classic receptor tyrosine kinase pathway with human EGF at 100 ng/ml. Cultures were incubated at 37°C for 10 min after activation and harvested.
c-Raf Immunoprecipitation/Kinase Assay-This was carried out as described (20). Briefly, COS-7 cells transfected by calcium phosphate were lysed with 100 l of lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 2 mM EDTA, 1 mM EGTA, 1% Triton X-100, 25 g/ml each leupeptin and aprotinin, 1 mM p-methanesulfonyl fluoride, 1 mM vanadate, 30 mM ␤-glycerophosphate) and precleared. FLAG-c-Raf protein was precipitated using 30 l of anti-FLAG M2 affinity gel (Sigma). The immunoprecipitates were washed four times with lysis buffer containing 0.5 M NaCl and twice with kinase buffer (35 mM Tris-HCl, pH 7.5, 10 mM MgCl 2 , 0.5 mM EGTA, and 1 mM vanadate). The kinase reaction was performed in 20 l of kinase buffer using 200 ng of kinase-disabled His-MEK(K97M) as substrate, 60 M ATP, and 10 Ci of [␥-32 P]ATP (6,000 Ci/mmol). The reaction was carried out for 20 min at 30°C and terminated by the addition of 10 l of 5ϫ Laemmli buffer. Quantitation was carried out by densitometry using Quantity-One software (Bio-Rad). Phosphotransferase signals were normalized on the amounts of c-Raf immunoprecipitated.
Bacteria Growth and Protein Purification-The c-Raf-cys domain WT and mutants were expressed as GST fusion protein in the BL21/DE3 strain of Escherichia coli (Novagen) (11). Bacteria were initially grown at 37°C to an optical density (OD) at 600 nm of 0.5, transferred to room temperature. At an OD 660 nm of 0.7-0.8, protein synthesis was induced by 0.5 mM isopropyl-␤-D-thiogalactopyranoside, and the cells harvested 2 h later. Bacteria were passed twice through a French press, and the GST fusion proteins were recovered from the lysates by affinity chromatography on the glutathione-agarose matrix. Purity by Coomassie Blue staining of SDS-PAGE was usually Ͼ90% by this protocol.
Retinoid Binding Assay by Quenching of the Endogenous Protein Fluorescence-Quantitative fluorescence measurement of 250 nM GST fusion protein with retinoid titration at 25 nM increments were performed as described (11) in phosphate-buffered saline, purged of oxygen by sparging with helium for 15 min, in a JASCO spectrofluorometer (model FP777). The protein solution was excited at 280 nm, and the protein emission was monitored at 330 nm. Binding constants were calculated by non-linear curve fitting according to the theorem by Norris et al. (31). Qualitative retinol binding assays based on vibronic fine structure determinations were performed as described (22).
Ras/c-Raf Binding Assay-Quantitative binding of GST⅐Raf cys WT and mutants to GTP␥S/Ras was measured by the ELISA described by Gosh et al. (15).
Computational Biology Methods-Docking of retinol into c-Raf-1 and PKC␣ was carried out using the software Autodock3.05 (32) to find potential binding sites of retinol. Autodock used a Monte Carlo simulated annealing technique for configurational exploration with a rapid energy evaluation using grid-based molecular affinity potentials. It thus combined the advantages of exploring a large search space and a robust energy evaluation. The method has proven to be a powerful approach to the problem of docking a flexible ligand into the binding site of a static protein. Van der Waals interactions were calculated using a Lennard-Jones 12-6 potential, whereas the hydrogen-bonding term was modeled by Lennard-Jones directional 12-10 potential. Electrostatic potential energies were calculated with a distance-dependent dielectric function. Chemical preference terms were also added in its scoring function. The program was tested on a number of protein-substrate complexes, which had been characterized by x-ray crystallography (33) as recommended for docking studies.
c-Raf structure was taken from Protein Data Bank accession number 1FAR (24), and PKC␣ C1A structure was modeled using 1PTQ (25) as template for the program Modeler (34) as part of the molecular modeling software package InsightII from Accelrys. NMR structure coordinates for ␣C1B were kindly provided by Dr. Marcel Luyten (35). Modeler is known to produce reliable models when high homologous high resolution crystal structures are available as template, which is the case in our study. Quality of the model was checked with modules available in InsightII package.

RESULTS
The GST fusion protein comprising c-Raf amino acids 139 -184, spanning the zinc finger domain, encodes one high affinity retinol-binding site (11) and a half-site for GTP/Ras recognition (16,36). PKC␣ C1A and C1B zinc fingers comprise amino acids 37-86, and 102-151, respectively. To maintain a unified numbering system between the c-Raf and PKC zinc fingers, the conserved cysteines and histidines were aligned with each other, requiring a gap for c-Raf since this domain lacks the 4-amino-acid phorbol-binding loop. Numbering began at the 1st histidine. (Fig. 1).
To map the retinol-binding site of c-Raf, we converted consecutively all amino acids (except Cys and His of the zinc finger core) to Trp and tested the bacterially expressed mutant peptides for retinol binding by the fluorescence quench assay. This assay registers the amount of bound retinol by the relative decrease of intrinsic protein fluorescence emission caused by fluorescence resonance energy transfer between a tryptophan residue and proximal retinol. Titrations of retinol yielded, after correction for inner filtering, Scatchard plots that allowed us to compute the apparent binding affinities by a theorem developed by Norris et al. (31). Examples are shown for WT and the mutants F8W, T33W, and K37W. The first two mutants were unable to bind retinol, as indicated by the absence of quenching of protein fluorescence (Fig. 2), whereas the control K37W mutant yielded binding comparable with WT.
Because point mutations can cause broad structural disruptions, it was desirable to ascertain the integrity of each mutated protein. Otherwise, the loss of retinol binding might merely reflect a general structural collapse. The zinc finger domain encodes a face where GTP/Ras docks (17). The mutated peptides were tested for retention/loss of GTP/Ras binding capacity by an ELISA devised by Gosh et al. (15). Briefly, GTP/ Ras was adsorbed to ELISA plates and incubated with WT and mutant GST fusion proteins. Anti-GST antibody-conjugated phosphatase was used to detect the amount of bound fusion proteins. As shown in Fig. 3 the two retinol-binding loss mutants, F8W and T33W, retained Ras binding capacity comparable with WT, whereas that of the non-permissible framework mutant C34S was at least 4-fold weaker (25). Table I summarizes the results of a systematic study of all mutants. Of 41 mutations, 12 resulted in substantial loss of retinol binding affinity. However, within this category several, including C34W, S43W, and K45W, were disregarded as they showed weakened Ras binding capacity.
To narrow the search for presumptive contact residues among the remaining 9, a computational docking study was undertaken, using the coordinates of three known PKC and c-Raf zinc finger structures (24,25,35). To account for the mutual competition of retinoids for the same site on zinc fingers (21,37,38), retinol was assumed to bind by the ␤-ionone ring and polyene structure. Therefore the computer was instructed to search for a hydrophobic pocket that would accommodate retinol in headfirst orientation at an energy minimum. A groove was identified in the c-Raf zinc finger with Phe-8 at its head that permitted retinol to dock optimally. Additional con-tact residues that consistently showed up in the two top ranking docking models were Leu-9, Ala-12, and Arg-30 (Fig. 4, models B and C). These bounded the presumptive pocket that held the head group. In the first model, the residues contacting the tail end were Gly-35, Thr-33, and framework Cys-34 (Fig. 4,  model B). Mutations of Thr-33 and Gly-35 both led to impaired retinol binding (Table I). In the alternate model, the orientation and contacts with amino acids for holding the tail end were less well defined. Thus, residues Phe-8 and Thr-33 were identified as retinol contact residues by both docking modeling and site-directed mutagenesis.
PKC␣ zinc fingers are highly homologous in their structures to c-Raf, except for the presence of the loop formed by amino acids 23-26 that defines the phorbol ester-binding site of PKC but is missing in c-Raf. To determine whether the site mapped to a similar topographical region in the C1A domain of PKC␣ (which is known to bind retinol), comparative docking trials were undertaken. The results identified a phenylalanine at position 8 that furnished the principal contact site for the ␤-ionone ring. This residue corresponded precisely to Phe-8 in c-Raf. The surrounding amino acids were, however, quite different from those of the c-Raf-binding site, ␣C1A having Phe-7, instead of Thr-7 found in c-Raf. The PKC␣ C1B domain does not bind retinol (21). Consistent with this, a docking trial yielded no firm assignment of a binding pocket in C1B (Fig. 4, model E).
To experimentally confirm predictions from the docking trials, mutations were introduced into the ␣C1A domain. As shown in Table II, mutating any of the presumptive contact amino acids (Phe-7, Phe-8, or Trp-22) alone was ineffective, unlike in c-Raf, where single point mutations impaired retinol binding significantly. Double mutations of the presumptive hydrophobic amino acids Phe-7 and Phe-8 believed to contact the head group slightly lowered the binding affinity. However, dual substitutions involving either Phe-7 or Phe-8 in combination with Trp-22 drastically diminished retinol binding. Amino acids Thr-7, Tyr-8, and Tyr-22 are characteristic of the nonbinding ␣C1B domain. When copying this motif into the ␣C1A domain, retinol binding was abolished, whereas phorbol ester binding was preserved (data not shown), indicating that this triple mutation was structurally permissible.
To test whether ␣C1B can be converted to a binder, we introduced the reverse set of mutations, T7F/Y22W. Retinol binding assays of the corresponding GST fusion protein by quench of intrinsic fluorescence (Fig. 5A), fluorescence resonance energy transfer, enhancement of retinol fluorescence emission, and the red-shifted fluorescence excitation spectrum with vibronic fine structure of retinol (Fig. 5B) all indicated strong retinol binding. By titrating retinol and applying nonlinear fitting to the quench curve of Fig. 5A, an apparent binding affinity of 15.7 nM was obtained, comparable with the value reported for the ␣C1A domain (20.7 nM (22)). Docking trials with this in silico mutant yielded a well defined site (Fig.  4, model F).
As reported previously, c-Raf activation by UV was retinoldependent (20). This was deduced from experiments showing a marked decrease in the endogenous c-Raf kinase activity visa-vis MEK, when retinol-deprived cells were activated by UV irradiation, as compared with cells grown in vitamin A-sufficient medium or with cells reconstituted with retinol shortly prior to activation. We tested whether c-Raf lacking a func-tional retinol-binding site would yield a similarly reduced response as wild-type c-Raf nutritionally deprived of its retinol co-factor. The FLAG-tagged, full-length c-Raf WT, single mutants F146W (F8W), T167 (T33W), or double mutant F146W/ T167W (F8W/T33W) were expressed in COS cells. After serum starvation, cultures were reconstituted with retinol 0.5 h prior to activation by UV irradiation. c-Raf phosphotransferase activity was determined in immunoprecipitates using kinasedead MEK as substrate. As shown in Fig. 6, retinol permitted expression of high kinase activity in response to UV irradiation. The ratio of activity with 1 M retinol over reference value without retinol was 3.5. The important result was that the loss of either contact site, Phe-8 or Thr-33 of the zinc finger domain, that prevented retinol-binding also diminished kinase activation by roughly 1 ⁄3 or 1 ⁄2, respectively, indicating that the retinol TABLE I Scanning mutagenesis of c-Raf zinc-finger to identify contact residues mediating retinol binding Point mutations were systematically introduced in the 45 amino acid stretch from His-139 to Cys-184 comprising the zinc-finger domain (* indicates numbering beginning with first histidine of zinc-finger; a 4-amino-acid gap was allowed for alignment with PKC zinc fingers; ** indicates numbering according to full-length cRaf molecule). Six cysteines and 2 histidines defining the two zinc coordination centers were left unchanged. Mutated peptides were expressed as GST fusion proteins and tested for retinol binding by the quench method (31) (see also Fig. 2 for examples). The dissociation constants and standard deviations of the mean are listed in column 4. Each mutated peptide was also tested for capacity to bind GTP/ras by the ELISA assay of Gosh et al (15), and results are reported in column 5 as ϩ for binding capacity within 90th percentile of wild type, Ϫ for absence of binding, and ϩ/Ϫ for intermediate binding (see Fig. 3 for examples).
co-factor function normally required for c-Raf activation in the redox pathway decreased markedly (from 3.5-to 2.3-fold (p Ͻ 0.05) and 1.8-fold (p Ͻ 0.01), respectively. The double mutant also became insensitive to retinol (relative enhancement 1.2fold with p Ͻ 0.001), but the EGF responsiveness declined as well.

DISCUSSION
Systematic scanning mutagenesis is a valid, albeit not conclusive, method of mapping receptor sites. Although pertinent contact sites can be reliably identified, the major drawback is the frequency of false negatives. To the extent that the retinolbinding site likely depends on proper protein folding, amino acid substitutions violating the overall structure may lead to loss of binding. This would be expected for the 6 Cys and 2 His conserved residues needed for zinc coordination (Fig. 1). The respective point mutations were therefore omitted. Other amino acids may be structurally essential in unexpected ways. As a criterion for structural integrity, we measured the interaction with GTP/Ras. A Ras-binding half-site is localized within the stretch of amino acids 4 -11 close to the N terminus (36). Four mutants with complete Ras binding loss and 17 mutants with partial Ras binding loss were deemed to result from possible structural deficit. These were excluded from further consideration. Among the remaining 20 mutants with intact Ras binding were five, K6W, F8W, Q18W, F29W, and T33W, that selectively abolished retinol binding. The respective WT residues were considered candidate contact sites. Ad-

FIG. 4. Presumptive retinol-binding grooves on c-Raf and PKC␣ zinc finger domains.
Retinol is illustrated in a green ball-and-stick representation. Presumptive residues in contact with retinol are labeled and space-filled. Model A, nuclear magnetic resonance structure of c-Raf zinc finger domain based on the Protein Data Bank code 1FAR (24). According to the top scoring docking results, Phe-8, Leu-9, and Ala-12 are common residues to hold the ␤-ionone ring in c-Raf (models B and C) with the lowest energy retinol conformations. The retinol tail is oriented slightly different around residue Cys-34 in model B as compared with model C. A cluster of retinols is shown to bind the same PKC␣ C1A region as that of the c-Raf zinc finger (model D). Phe-7 and Phe-8 contact the retinol head group. Retinols scattered around ␣C1B (model E) indicate no preferred binding site, whereas introducing T7F and Y22W mutations into ␣C1B permits docking of a cluster of retinol in the same region as identified in ␣C1A and c-Raf (model F).

TABLE II
Effect of mutagenesis of predicted contact amino acids of PKC␣ C1A mediating retinol binding Mutated peptides were expressed as GST fusion proteins and tested for retinol binding by the quench method (31) (legend for Fig. 2). Listed are the dissociation constants and standard deviations of the means. Mutated peptides preserved capacity to bind phorbol ester (data not shown).

␣C1A
Retinol ditionally, the mutations, S43W, T44W, K45W, V46W, T48W, and M49W, led to retinol binding loss without impacting Ras binding. These amino acids are located on a contiguous stretch on the ␣-helical peptide near the C terminus, form a mildly hydrophilic topology, and represent a potential second retinolbinding region. Sequence comparisons have not yielded any obvious motif for a retinol-binding site. To narrow the choices among candidate binding sites empirically, ligand-docking experiments were performed, using the published coordinates of the PKC␣ and -␦ and c-Raf zinc fingers from x-ray crystallographic (25) and NMR (24,35) studies. We searched for a hydrophobic groove accommodating the retinol molecule, with the stipulation that retinol bound headfirst. This choice recommended itself because of our previous findings that different retinoid species bound zinc finger domains with similar affinity (11) and that any one of these could displace the other (21). The most likely explanation was that the conserved hydrophobic ␤-ionone ring and polyene furnished the contact surfaces, whereas the hydrophilic groups at carbon 15, where most of the chemical differences resided, were less likely to contribute binding affinity. The inferred orientation also accounted for the biological antagonism between the hydroxylated family of retinoids (retinol and 14-hydroxy-retro-retinol) and carboxylated ones (retinoic acids) or non-substituted retinoids (anhydroretinol) (38) since pharmacological mutual reversible inhibition is best explained by competition of a conserved structure for binding to a single receptor.
Based on the highest number of hits in the docking trials with c-Raf, the most likely binding groove was defined by amino acid Phe-8, anchoring the retinol ionone ring, whereas Thr-33 bound the tail end (Fig. 4, model B). These contact sites also stood out in the mutational analysis. Importantly, Phe-8, or a Tyr residue at that position, is conserved among the zinc FIG. 5. A, determination of the binding constant of the reconstructed ␣C1B domain. GST fusion proteins were tested for retinol binding by the quench methods. Shown are quench curves obtained as detailed in the legend for Fig. 2. B, fluorescence excitation spectra of retinol bound to WT or reconstructed ␣C1B domains. Excitation is from 300 to 380 nm; emission is at 466 nm. A red-shifted excitation spectrum and shoulders at 325, 348, and 368 nm, characteristic for vibronic fine structure, indicate that retinol was bound to the reconstructed C1B domain (solid line). Retinol in buffer (broken line) or retinol in a solution with GST⅐␣C1B does not display fine structure and hence did not bind.  Table I) were expressed in COS cells and activated by UV light alone, by the combination of UV light and 1 M retinol, or by 100 ng/ml EGF (positive control) as described (20). FLAG-c-Raf was immunoprecipitated, and phosphotransferase activity was determined with kinase-inactive MEK as substrate. Shown are autoradiographs representative of independent experiments (n Ͼ 5). B, quantitation of autoradiograph by densitometry using Quantity One software (Bio-Rad) of at least five independent experiments. Cumulative data of each set were normalized with respect to UV response. *, probability values (Student's t test) as compared with wild-type UV-irradiation ϩ retinol (UVϩRol). finger structures and represented the best candidate for a major anchoring site. However, the presence of an aromatic amino acid at that position is insufficient for binding, and it is unclear what makes for strong binding. The amino acids surrounding Phe-8 differ greatly between the c-Raf and PKC␣ C1A structures, although both bind retinol with similar affinity. We conclude that although the general region harboring retinolbinding sites is conserved, the details determining the orientation of retinol may differ.
The comparison of ␣C1A, ␣C1B, and c-Raf with one another showed little homology in the retinol-binding region, making it hard to explain why two bound retinol, whereas one (␣C1B) did not. Yet empirically replacing the Thr-7 residue by Phe in the latter, together with the Y22W mutation, created a high affinity binding site (Fig. 4, model F). That Thr-7 of ␣C1B appeared incompatible with binding, whereas the same amino acid was tolerated in c-Raf, implies a fundamentally different orientation of the ligand in the two molecules, although ostensibly retinol serves the same function as co-factor for redox control. The importance of the PKC␣ C1A retinol-binding motif characterized by Phe-7, Phe-8, and Trp-22 was underscored by the finding that substitutions of either one, or both, of the phenylanines, concurrently with the tryptophan, ablated retinol binding. Our findings establish the region surrounding Phe-8 as the likely binding site. Whether a second binding site is centered on the stretch of the C-terminal residues, mentioned above, remains an open question.
Upon mutation of the Phe-8 contact site in the intact c-Raf molecule, the retinol-dependent function was diminished, illuminating the functional importance of this site. Retinol is believed to regulate the oxidative activation of c-Raf kinase. Because it is a vitamin, retinol can be removed from cells by nutritional deprivation. Alternately, displacement by anhydroretinol mimics a retinol-deficient condition. Using both tactics, we showed previously that the absence of bound retinol markedly inhibited the redox-mediated Raf kinase activation (20). This finding has now been verified by molecular association between retinol and c-Raf since removing the intrinsic binding site was as effective as retinol deprivation. As shown in Fig. 6, the UV-induced kinase activity of the two retinol-binding site mutants, F8W and T33W, was significantly decreased as compared with that of WT, although retaining a small advantage with retinol. Perhaps in the single point mutations, retinolbinding capacity was diminished, but not abolished altogether, allowing some residual retinol-mediated enhancement. When both contact sites were mutated, the retinol enhancement decreased further, although at the expense of overall responsiveness to EGF.
Our work has highlighted the direct initiation of the c-Raf kinase activation cycle by a redox mechanism. The likely chemical targets of oxidation are cysteine residues of the zinc finger domain. We have presented in vitro evidence elsewhere that oxidation leads to thiol modification and release of coordinated zinc (20). Zinc release in response to oxidation as well as to phorbol ester was also the defining event in PKC activation (27). Thus, redox activation of c-Raf obeys the same paradigm that governs PKC activation, although in distinction to PKC, it requires interaction with GTP/Ras. Requirement of intact retinol-binding sites, modification of thiol, and release of zinc as preludes for catalytic activation add up to the notion that serine/threonine kinases can sense and respond directly to changes in the redox microenvironment.