The Structure of a Retinal-Forming Carotenoid Oxygenase

See allHide authors and affiliations

Science  08 Apr 2005:
Vol. 308, Issue 5719, pp. 267-269
DOI: 10.1126/science.1108965


Enzymes that produce retinal and related apocarotenoids constitute a sequence- and thus structure-related family, a member of which was analyzed by x-ray diffraction. This member is an oxygenase and contains an Fe2+-4-His arrangement at the axis of a seven-bladed β-propeller chain fold covered by a dome formed by six large loops. The Fe2+ is accessible through a long nonpolar tunnel that holds a carotenoid derivative in one of the crystals. On binding, three consecutive double bonds of this carotenoid changed from a straight all-trans to a cranked cis-trans-cis conformation. The remaining trans bond is located at the dioxygen-ligated Fe2+ and cleaved by oxygen.

Retinal and its derivatives participate in numerous cellular activities; they are crucial for vision and the immune system (1, 2) and are therefore of nutritional importance (3). Retinal-forming carotenoid oxygenases constitute a sequence-related family of more than 100 currently known members. The family was discovered through a 9′-cis-epoxycarotenoid oxygenase that participates in the biosynthesis of the important plant hormone abscisic acid (4). A prominent family member is β-carotene-15,15′-oxygenase from animals, which cleaves β-carotene symmetrically to two molecules of retinal (5-7). Another member is β-carotene-9′,10′-oxygenase, cleaving β-carotene asymmetrically to form apo-10′-β-carotenal (8), which is thought to be converted to retinoic acid (9), a key actor in developmental processes (10). The family includes the retinal pigment epithelial protein RPE65 (11), mutations of which cause Leber's congenital amaurosis, a severe blinding disease (12). In plants, the genome of Arabidopsis thaliana codes for as many as nine family members (2, 13), several of which have been established as carotenoid oxygenases (14, 15). Some of these genes yield products regulating growth and development (16, 17). Other plant members catalyze the biosynthesis of pigments (18, 19). Cyanobacterial retinal-producing members have been proposed (20) and recently identified in Synechocystis (21). Here, we report the crystal structure of the Synechocystis enzyme at 2.4 Å resolution, revealing the reaction geometry and establishing a solid base for modeling all other family members.

The apocarotenoid-15,15′-oxygenase (ACO) from Synechocystis sp. PCC 6803 was expressed in Escherichia coli inclusion bodies and (re)natured with yields of about 2 mg purified enzyme per liter culture (22). ACO was soluble without detergent. It became active after adding Fe2+ ions and retained its activity over several days. The catalyzed reaction requires dioxygen (Fig. 1A), but the oxygen atoms in the two resulting aldehydes most likely originate from both dioxygen and water in a 1:1 ratio (23). On purification, the elution pattern of the final gel filtration column showed a dominant peak at the monomer mass and a small peak at the dimer mass (fig. S1). After adding the detergent octylpolyoxyethylene (C8E4-8), the dominant peak ran at the trimer mass, presumably because ACO had been recruited to a detergent micelle. Because numerous attempts to crystallize the soluble detergent-free protein failed, detergent was added in all further experiments.

Fig. 1.

Enzyme data. (A) The reaction catalyzed by ACO (21). ACO accepts the all-trans conformations of the homologs a, b, c,and d as alcohols or aldehydes with and without the 3-hydroxy group, but it does not accept β-carotene (21). (B) ACO crystallized in the absence of Fe2+ and in the presence of the b-type substrate all-trans-(3R)-3-hydroxy-8′-apo-β-carotenol. The shaded part of the substrate was used to interpret the electron density in a native crystal produced by soaking with Fe2+ and subsequent freezing to 100 K. Surprisingly, the all-trans substrate had changed to the 13,14-13′,14′-di-cis conformation.

First, inactive Fe-free ACO was crystallized in the presence of its substrate (Fig. 1) and the crystal structure was solved by x-ray diffraction (table S1). The phases were determined by two heavy-atom derivatives and solvent flattening. A model was built and refined to good-quality indices (table S2). In a second experiment, we exposed the Fe-free crystals to Fe2+ ions for 30 min and flash froze to 100 K. Visually, the crystals remained the same, but an x-ray analysis showed a lower symmetry space group with four independent ACO molecules instead of two in Fe-free ACO crystals (table S1). The crystal structure was solved by molecular replacement and refined (table S2). On soaking, the ACO molecules had rotated by 1° to 3° and shifted by 1 to 4 Å, improving the crystal quality. In both crystal forms, the crystallographically independent molecules associated asymmetrically, indicating that ACO is monomeric as shown by gel filtration.

During refinement, the chain folds of the four individual ACO molecules remained virtually identical to each other and also to those of the Fe-free crystals. Obviously, catalysis does not involve extended main chain displacements, because these should have been detectable in at least one of the six independently packed molecules. Fe2+ was bound at four histidines as derived from an 8σ electron density peak. Such an Fe2+ coordination is known for only four other proteins (24-27). The presence of Fe2+ was confirmed by an anomalous difference Fourier map showing only one significant peak.

ACO consists of a seven-bladed β propeller with four histidines at the propeller axis that hold the Fe2+ ion and thus mark the active center (Fig. 2). Such a chain fold was initially observed in a neuraminidase (28). The closest structural relative of ACO is presently a muconate-lactonizing enzyme (29) showing a z score of 17.3 in a chain-fold comparison with program DALI (30). Four other known structures have z scores above 12. The top end of the β-propeller axis as defined by the blade connections (31) usually accommodates the active center. This also applies for ACO. The four histidines holding Fe2+ are all at the beginning of the innermost β strands of the propeller (Fig. 2). Each blade consists of four locally connected antiparallel strands, except for the first and seventh blade, which are five-stranded as a result of an N-terminal addition (fig. S3). The loops at the top side of the propeller are very long, forming a large dome over the active center, whereas those at the bottom are generally short.

Fig. 2.

Stereoview ribbon plot of the ACO chain fold consisting of 490 residues with a mass of 54,286 daltons. The 11 N-terminal residues are disordered. The seven-bladed β propeller has the usual topology (31). The four histidines holding the Fe2+ ion in an octahedral arrangement are shown together with two water molecules and the substrate as ball-and-stick models. The view is along the active center tunnel, the rear exit of which is marked by a red line. Strands β1 and β2 (yellow) are additions to the common propeller.

The surface of ACO contains a nonpolar patch that consists mostly of protruding leucines and phenylalanines (Fig. 3A). We propose that ACO uses this patch to dip into the membrane and extract its nonpolar substrate from there. In both crystal forms, two ACO molecules associate asymmetrically to form a combined nonpolar patch opposite a similar patch formed by another couple about 15 Å away. Most likely, the detergent forms micelles between these patches as known from monotopic membrane protein crystals (32). The high 66% solvent content of ACO crystals and the exceptional behavior during gel filtration (fig. S1) also point to this group of proteins.

Fig. 3.

Surface representation of ACO showing the general features of this putative monotopic membrane protein. The propeller axis is nearly vertical. (A) View into the active center tunnel, as in Fig. 2. The nonpolar patch at the top contains Trp121, Ile125, Phe126, and Phe263 as well as leucines 122, 128, 259, 262, and 265 with a total nonpolar surface of about 800 Å2. The depicted patch surface (yellow) follows fake glycines replacing the protruding residues, which in turn are shown as ball-and-stick models under the transparent true surface. In situ, this patch most likely dips into the membrane, facilitating substrate uptake and release. (B) Side view of the molecule as cut through the center, outlining the bent tunnel lined by the Fe2+ (red circle) at the active center and showing the added substrate. The small “exit” hole is indicated by the red line, as in Fig. 2. Because the deep pocket at the propeller axis at the bottom had to be cut obliquely, it shows a spurious constriction that is absent in a central cut.

ACO contains a tunnel which enters the protein near the nonpolar patch and extends to the active center. After passing the Fe2+ ion, the tunnel turns upward and exits the protein on the far side of the nonpolar patch. This tunnel runs perpendicular to the propeller axis and is lined with numerous nonpolar residues, mainly with aromatic sidechains (Figs. 3B and 4). In addition, ACO contains a deep and narrow pocket entering from the bottom end of the propeller axis (Fig. 3B), which ends at the Fe2+-ligating imidazoles and does not reach the Fe2+. It results from the propeller architecture and is presumably present in all family members. In ACO, it may serve as an auxiliary pathway for dioxygen.

Fig. 4.

Geometry of the reaction. The view corresponds to Fig. 3B. Fe2+ has six ligand sites arranged in an octahedron. Four sites are assumed by histidines, among which those at positions 238, 304, and 484 are fixed by glutamates. The remaining two sites most likely accept the required dioxygen. The reaction starts by one of these oxygen atoms attacking the 15-15′ double bond and continues by a further attack of a second oxygen or oxygen derivative. Because one of the produced aldehyde oxygens comes from water (23), it is most likely that the Fe2+ ligands exchange during intermediate states of catalysis. The electron densities of the substrate and water molecules represent the original (Fo-Fc)-map averaged over the four ACO molecules in the asymmetric unit at a contour level of 2σ. The double bonds of the substrate are darkened. E, Glu; F, Phe; H, His; N, Asn; T, Thr; W, Trp; Y, Tyr.

The inactive Fe-free crystals contained some low additional electron density in the active center that was modeled as a C8E4 detergent molecule. On soaking with Fe2+ in the presence of the substrate, however, the reconstituted native enzyme showed an electron density at the active center that was far too strong for a C8E4 molecule. The actual presence of a ligand was confirmed by a large rotation of the Phe303 sidechain providing the required space (Fig. 4). This rotation was one of the very few differences between the Fe-free and the native structure.

Because no other suitable compound had been added during crystal growth and handling (22), we filled this density with a substrate molecule (Fig. 4). The density had the form of a cranked rod that could only be fitted by changing the 13-14 and 13′-14′ double bonds from trans to cis (Fig. 1B). This fit implies that the β-ionone ring sits at the tunnel entrance, which acts as a bottleneck, arresting the substrate in the correct position for retinal formation (Fig. 3A). Unfortunately, because this ring is invisible, our interpretation cannot be considered proven in all details. As expected for full occupancy, the B-factors in the middle of the fitted substrate moiety met those of the surrounding polypeptide. They increased toward both ends, indicating that the β-ionone ring is invisible because it is mobile. At the other end of the substrate, there is space for longer isoprenoid tails, which agrees with the observed activity for 4′-apocarotenoids (21). β-carotene itself is not a substrate given that neither of its ionone groups can enter the tunnel. Likewise, a kinked isoprenoid tail caused by a cis double bond is not able to enter the tunnel, explaining the restriction to all-trans substrates.

When entering the tunnel, the straight isoprenoid tail collides with the oxygen ligands of the Fe2+ ion that most likely sterically enforce the observed two conversions from trans to cis. The isomerizations occur at methyl-substituted double bonds in an extended conjugated double-bond system with an activation energy of about 100 kJ/mol (33), which equals that of nonproline trans-cis peptide flips known from protein folding. The 15-15′ bond remains trans in agreement with the higher barrier of an unsubstituted double bond.

In the native structure we find a near-perfect octahedral coordination to Fe2+ with distances of 2.1 Å to four histidines and one water molecule. We suggest that dioxygen is bound with one atom each to the fifth and sixth Fe2+ coordination sites. The fifth site accommodates a water molecule in our crystal. The sixth site is large enough for one of the dioxygen atoms but unfavorable for a water because it is lined by the methyl group of Thr136, which is fixed by Asn135 (Fig. 4). Most likely, the oxygen at the fifth site attacks the C15 atom. The distance of the respective water to the C15 atom is 3.2 Å. In our structure, we found a second water bound to the Fe2+-ligated water in front of the C15′ atom. A derivative of this water may attack the C15′ atom during the reaction, explaining the observed 1:1 ratio of oxygens from dioxygen and from water in the products of a related enzyme (23). After double-bond cleavage, the resulting 13′,14′-cis and 13,14-cis-aldehydes readily convert to trans. Our interpretation is corroborated by the observed small amounts of 13-cis-retinal observed in cleavage assays (21).

Comparisons within the carotenoid oxygenase family (fig. S4) show that the four active center histidines are strictly conserved and that their environment is well conserved. Presumably, all members of this family share a common chain fold, possess similar active centers, and follow a similar reaction mechanism. The β propeller is a solid structural base allowing the various family members to define appropriate specificity-determining loops covering the active center. The suggested trans-to-cis isomerizations of methyl-substituted double bonds upon binding may also occur in other family members, so that some of them may be just isomerases, as is vividly discussed for RPE65 (11).

Supporting Online Material

Materials and Methods

Figs. S1 to S4

Tables S1 to S3


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article