Báo cáo khoa học: Analysis of the NADH-dependent retinaldehyde reductase activity of amphioxus retinol dehydrogenase enzymes enhances our understanding of the evolution of the retinol dehydrogenase family pot

Members of the SDR-RDH group such as RDH2, RDH5 and 17b hydroxysteroid dehydrogenase type 9 17bHSD9 [10–12], non-RDH SDR enzymes, including mouse retinal reductase RRD, retinal short-cha

activity of amphioxus retinol dehydrogenase enzymes

enhances our understanding of the evolution of the retinol dehydrogenase family

Diana Dalfo´, Neus Marque´s and Ricard Albalat

Departament de Gene`tica, Facultat de Biologia, Universitat de Barcelona, Spain

Retinoic acid (RA) regulates critical physiologic

pro-cesses in vertebrates, such as anterior–posterior pattern

formation, cell proliferation, tissue differentiation,

morphogenesis, and embryonic development [1] The

main source of retinoids stems from the enzymatic

cleavage of dietary b-carotenes, which produces

retin-aldehyde This, in turn, is reduced to retinol, which is

subsequently esterified to retinyl esters and stored in

the liver [2] Upon demand, these esters can be

hydro-lyzed to retinol, which is released into the circulation

to be used in target tissues, to undergo oxidation into retinaldehyde, and to be further transformed into RA Retinol dehydrogenase and retinaldehyde reductase activities are therefore major players in retinoid meta-bolism, making essential contributions to the physio-logic control of RA availability In vertebrates, retinol dehydrogenase activity has been classically associated with two distinct protein families, the short-chain

Keywords

cephalochordates; chordate evolution;

retinaldehyde reductases; retinoic acid

metabolism; retinol dehydrogenases

Correspondence

R Albalat, Departament de Gene`tica,

Facultat de Biologia, Universitat de

Barcelona, Av Diagonal, 645, 08028

Barcelona, Spain

Fax: +34 934034420

Tel: +34 934029009

E-mail: ralbalat@ub.edu

Website: http://www.ub.edu/genetica/

indexen.htm

(Received 9 March 2007, revised 4 May

2007, accepted 30 May 2007)

doi:10.1111/j.1742-4658.2007.05904.x

In vertebrates, multiple microsomal retinol dehydrogenases are involved in reversible retinol⁄ retinal interconversion, thereby controlling retinoid meta-bolism and retinoic acid availability The physiologic functions of these enzymes are not, however, fully understood, as each vertebrate form has several, usually overlapping, biochemical roles Within this context, amphi-oxus, a group of chordates that are simpler, at both the functional and genomic levels, than vertebrates, provides a suitable evolutionary model for comparative studies of retinol dehydrogenase enzymes In a previous study,

we identified two amphioxus enzymes, Branchiostoma floridae retinol dehy-drogenase 1 and retinol dehydehy-drogenase 2, both candidates to be the cephalochordate orthologs of the vertebrate retinol dehydrogenase enzymes We have now proceeded to characterize these amphioxus enzymes Kinetic studies have revealed that retinol dehydrogenase 1 and retinol dehydrogenase 2 are microsomal proteins that catalyze the reduc-tion of all-trans-retinaldehyde using NADH as cofactor, a remarkable com-bination of substrate and cofactor preferences Moreover, evolutionary analysis, including the amphioxus sequences, indicates that Rdh genes were extensively duplicated after cephalochordate divergence, leading to the gene cluster organization found in several mammalian species Overall, our data provide an evolutionary reference with which to better understand the origin, activity and evolution of retinol dehydrogenase enzymes

Abbreviations

AKR, aldo-keto reductase; AR, aldose reductase; CRAD, cis-retinol/androgen dehydrogenase; ER, endoplasmic reticulum; GFP, green fluorescent protein; HAR, human aldose reductase; HSD, hydroxysteroid dehydrogenase; HSI-AR, human small intestine aldose reductase; MDR, medium-chain dehydrogenase ⁄ reductase; NLS, nuclear localization sequence; PAN2, pancreas protein 2; RA, retinoic acid; RRD, mouse retinal reductase; SDR, short-chain dehydrogenase ⁄ reductase.

retinol dehydrogenases [short-chain dehydrogenase⁄

reductase (SDR)-RDHs] and the medium-chain

alcohol dehydrogenases [medium-chain dehydrogenase⁄

reductase (MDR)-ADHs] [3,4] Despite the many

bio-chemical studies on these two protein families, the major

enzyme(s) responsible for the in vivo oxidation of retinol

remains uncertain In previous studies, we analyzed

the functionality and evolution of the MDR-ADH

family [5–8,9] Here, we focus on the contribution of

SDR-RDH enzymes to retinol⁄ retinal metabolism

Classically, RDH enzymes have been regarded as a

complex vertebrate group of microsomal proteins

that catalyze the conversion of retinol to retinaldehyde

in vitro using NADH as cofactor [4] However, the

RDH family contains many enzymes with diverse

substrate specificities towards cis and trans isomeric

forms, and, mostly, toward steroids Hence, attempts to

determine the physiologic contribution of each RDH

enzyme to RA metabolism have been impaired by the

variety of enzymes as well as by the overlaps in substrate

recognition

Substantial multiplicity and redundancy is also

pre-sent in the reductive direction of the pathway Four

vertebrate protein families have been associated with

retinaldehyde reduction Members of the SDR-RDH

group such as RDH2, RDH5 and 17b hydroxysteroid

dehydrogenase type 9 (17bHSD9) [10–12], non-RDH

SDR enzymes, including mouse retinal reductase

(RRD), retinal short-chain dehydrogenase/reductase 1

(retSDR1), photoreceptor outer segment all-trans

retinol dehydrogenase (prRDH), retinal reductase 1

(RalR1) and pancreas protein 2 (PAN2) [13–17],

MDR-ADH forms such as ADH1, ADH4 [18] and

amphibian ADH8 [19], and members of the aldo-keto

reductase superfamily, including human aldose

reduc-tase (AR), human small intestine aldose reducreduc-tase

(HSI-AR) and chicken aldo-keto reductase (AKR)

[20,21], all catalyze retinaldehyde reduction in vitro

To shed light on the evolutionary origin and

physio-logic basis of the RDH and retinaldehyde reductase

multiplicity of vertebrates, analysis of the

cephalochor-date amphioxus is invaluable Cephalochorcephalochor-dates are

useful organisms for comparative analyses, as their low

gene complexity and archetypical body plan

organiza-tion suggest that they retain many ancient

characteris-tics Amphioxus did not undergo the extensive gene

duplications that occurred during early vertebrate

evo-lution [22], but rather exhibits an RA-signaling system

and a retinoid content comparable to that of

verte-brates [23,24] In a previous study, we identified two

enzymes, RDH1 and RDH2, that belong to the

SDR-RDH group in the species Branchiostoma floridae [25]

Here we present experimental data showing that these

two enzymes are endoplasmic reticulum (ER)-associ-ated proteins that may participate in retinoid metabo-lism, by catalyzing retinaldehyde reduction Moreover, phylogenetic analysis indicates that most vertebrate RDHs derive from lineage-specific tandem duplications

of an ancestral form that may resemble the current amphioxus enzymes The novel vertebrate RDH enzymes would have evolved new biochemical activities

in retinoid and steroid metabolism after cephalochor-date divergence, thereby contributing to the increased physiologic complexity of the vertebrate subphylum

Results

Enzymatic properties of recombinant RDH1 and RDH2

Amphioxus RDH1 and RDH2 proteins tagged at the N-terminus with the hemagglutinin (HA) epitope were produced in COS-7 cells and purified in the

microsom-al fraction The enzymatic activity of this fraction was assayed against retinoids Given that most vertebrate RDHs can catalyze cis-retinol and⁄ or trans-retinol oxi-dation, these were the substrates initially evaluated Indeed, mouse RDH1 (kindly provided by J L Napoli, University of California, Berkeley, CA, USA) was used

to monitor the retinol oxidation assay Unexpectedly, the oxidative activity observed for the amphioxus enzymes was below the detection capacity of the assay,

< 0.002 nmol (Fig 1A–C), although a wide range of conditions were used: from pH 6 to 9, 5–12.5 lm all-trans-retinol, 0.5–2 mm NAD+ and NADP+, and 10–100 lg of microsomes obtained from independent assays Negligible activity was also observed when 9-cis-retinol was assayed (data not shown) Next, we analyzed whether RDH1 and RDH2 exhibited reduc-tase activity (Fig 1D,E) Retinol production in vitro increased 2.5-fold and 25-fold for RDH1 and RDH2, respectively, in the presence of NADH, as compared

to controls However, these differences were not detec-ted with NADPH, even though COS cells showed intrinsic NADPH-dependent retinal reductase activity [12] The specific activity of each amphioxus enzyme was 0.25 nmol of retinolÆmin)1Æ(mg of microsomes))1 for RDH1 incubated with 15 lm all-trans-retinal, and 1.4 nmol of retinolÆmin)1Æ(mg of microsomes))1 for RDH2 with 12.5 lm all-trans-retinal (Table 1) Fur-thermore, to examine whether RDH forms had isomer specificity, we also assayed the RDH1 and RDH2 activities toward 9-cis-retinal However, only residual 9-cis-retinal reductase activity, less than 0.03 nmolÆ min)1Æmg)1, was detected for these two enzymes (Fig 1F,G) The reaction products were extracted and

analyzed by RP-HPLC, and the kinetic constants of

the two RDHs for all-trans-retinal were determined

(Fig 1H,I; Table 1) The apparent Km values of

RDH1 and RDH2 (8.7 lm and 8.9 lm, respectively)

were similar, whereas the maximum specific activities

(0.3 nmolÆmin)1Æmg)1and 2.3 nmolÆmin)1Æmg)1,

respect-ively) and the maximum specific activities⁄ Km ratios

(0.03 and 0.26, respectively) were > 7.5-fold higher for

RDH2 than for RDH1 Finally, the apparent cofactor

Km values of amphioxus enzymes (Fig 1J,K; Table 1)

were 224 lm and 98 lm NADH for RDH1 and

RDH2, respectively, whereas no significant activity

was detected with NADPH This cofactor preference is

consistent with the presence and absence of specific

amino acids at certain positions in the amphioxus enzymes (Fig 2A): both enzymes contain the D and T residues at the equivalent positions of cow RDH5 for NADH specificity, and lack any positively charged amino acid at the corresponding position of the rat RDH2 K64, which may be essential for NADPH preference [26]

Activity of recombinant RDH1 and RDH2

in intact cells Amphioxus RDH1 and RDH2 and mouse RDH1 were expressed in COS-7 cells to evaluate their activities with retinoids in an intact cell system In agreement

Fig 1 Enzymatic activity of amphioxus

RDH1 and RDH2 enzymes The biochemical

activity of the microsomal fraction of COS-7

cells transfected with amphioxus

HA-RDH1-expressing (A, D, F), HA-RDH2-HA-RDH1-expressing

(B, E, G) and mouse Rdh1-expressing (C)

constructs was analyzed For oxidative

reac-tions (A–C), the microsomal fraction (15 lg)

was incubated with all-trans-retinol (10 l M )

and NAD + (1 m M ) at pH 8.0 for 15 min at

37 C For retinal reduction, the microsomal

fraction (15 lg) was incubated with 10 l M

all-trans-retinal (D, E) or 9-cis-retinal (F, G)

and NADH (1 m M ) at pH 6.0 for 15 min at

37 C Elution was monitored at 380 nm for

retinal detection (A–C) and 325 nm for

ret-inol (D–G) detection The values for

all-trans-retinaldehyde reduction of amphioxus RDH1

(H) and RDH2 (I) were determined at 1 m M

NADH using eight concentrations of

sub-strate, from 0.5 to 20 l M and from 0.5 to

15 l M for RDH1 and RDH2, respectively.

The apparent Kmvalues for cofactor NADH

were determined at 15 l M and 12.5 l M

all-trans-retinaldehyde for RDH1 (J) and RDH2

(K), respectively, using six concentrations of

cofactor, from 0.005 to 1.5 m M Assays

were performed with 15 lg of microsomes

for 15 min at 37 C Each point represents

the average of three replicates.

with the biochemical analysis of the

microsomal-puri-fied enzymes, amphioxus RDH1 and RDH2 catalyzed

the reduction of retinaldehyde into

all-trans-retinol in intact cells (Fig 3A), whereas RDH1- and

RDH2-transfected cells showed no differences from

mock-transfected cells in the generation of

all-trans-retinaldehyde from all-trans-retinol (Fig 3B)

Mock-transfected COS-7 cells reduced all-trans-retinaldehyde,

indicating that the cells harbor reductases

Transfec-tion with amphioxus RDH1 cDNA produced a net

47 pmol and 99 pmol of retinol per mg of total protein

after 1 h of incubation with 10 lm and 20 lm of

retinal, respectively RDH2 enzymes were more effi-cient than RDH1 enzymes, and generated 64 and

134 pmol of retinol in the same assay conditions Overall, transfection with RDH1 and RDH2 cDNA increased the level of retinaldehyde reduction by

 35% and  50%, respectively, as compared to the mock-transfected cells Mouse RDH1, in contrast, oxidized retinol to retinaldehyde (Fig 3B) but did not support retinaldehyde reduction Indeed, mouse RDH1 decreased the amount of retinol generated in the assays with retinaldehyde incubation (data not shown), sug-gesting that the substrate used by this enzyme was the retinol generated from retinaldehyde by endogenous COS-7 cell reductase activity

Intracellular localization COS-7 cells were transiently transfected with con-structs expressing the amphioxus RDH1 and RDH2 enzymes fused to several peptides: HA epitope, green fluorescent protein (GFP) and b-galactosidase In agreement with the RDH1 and RDH2 purification in the microsomal fraction (Fig 2D), immunostaining of cells expressing HA-RDH1 and HA-RDH2 proteins revealed a typical pattern of ER-associated proteins, with no nuclear staining or plasma membrane localiza-tion observed (Fig 2B,C) GFP fusion was used to visualize the intracellular localization in living cells, thereby avoiding any artefacts caused by the cell fixation process RDH2 fused to GFP either at the C-terminus (RDH21)335-GFP) or at the N-terminus (GFP-RDH21)335) of the enzyme (Fig 2E,I) exhibited

a pattern that overlapped with the ER-Tracker Blue White DPX, which was used as a specific ER marker

in living cells (Fig 2F,J) The subcellular localization

of the RDH2 enzyme (RDH21)335) fused to the b-ga-lactosidase protein was also consistent with a typical pattern of ER-associated proteins (Fig 2M)

Table 1 All-trans-retinal activity and kinetic constants of B floridae

RDH enzymes compared with those of known vertebrate retinal

reductases Values are from this work, B floridae RDH1 (BfRDH1)

and BfRDH2, or from the literature [6-10,13,16,17,25] ND, not

determined HAR, human aldose reductase.

All-trans-retinal NADH

Specific

activity

(nmolÆ

min)1Æmg)1)

Maximum specific activity (nmolÆ min)1Æmg)1)

Km (l M )

Km (l M ) BfRDH1 0.25 0.3 ± 0.06 8.7 ± 4.1 224 ± 81

Chicken

AKR

a

k cat values in min)1.

Fig 2 ER subcellular localization of amphioxus RDH1 and RDH2 proteins (A) Sequence alignment of amphioxus RDH1 and RDH2 enzymes Amino acid substitutions are shown and identities are represented by dashes The active site (YTVAK) and the cofactor-binding motifs are marked in bold The D43, T67 and A69 residues, involved in cofactor specificity, are indicated by asterisks Flanking the N-terminal hydropho-bic segment, the LERGR motif is underlined Arrows indicate the truncated RDH2 forms fused to GFP or to b-galactosidase proteins (B, C) Immunostaining with an antibody to HA of COS-7 cells transfected with constructs encoding HA-RDH1 and HA-RDH2, respectively, and examined using confocal microscopy (D) Western blot of the pellets after 13 000 g (lanes 2 and 4) and 100 000 g centrifugations (lanes 1 and 3, microsomal fractions) of homogenates of COS cells transfected with HA-RDH1 (lanes 1 and 2) and HA-RDH2 (lanes 3 and 4) (E–L)

In vivo localization of RDH2 in the ER of the cells COS-7 cells were transfected with constructs encoding RDH21)335-GFP (E), RDH21)28 -GFP (G), RDH21)58-GFP (H), GFP-RDH21)335(I), GFP-RDH2295)335(K) and GFP (L) ER-tracker Blue White DPX marker was used to specific-ally visualize the ER in living cells (F, J) (M–R) Localization of RDH2-b-galactosidase chimeric proteins COS-7 cells were transfected with constructs encoding the full-length (RDH21)335) (M) and four C-terminal truncated RDH2 forms (RDH21)229, RDH21)165, RDH21)137 and RDH21)58) (N–Q, respectively) fused to the NLS-b-galactosidase protein, and with the pb-galactosidase-N2 empty vector (R) The pb-galac-tosidase-N2 vector contains a nuclear localization sequence (NLS) 5¢ to the LacZ gene The SV40 NLS localizes the b-galactosidase codified

by the empty vector to the nucleus Cells were immunostained with an antibody to b-galactosidase and examined using a Zeiss Axiophot fluorescence microscope.

To analyze the contribution of protein domains to

the ER anchorage, the localization of the full-length

enzyme was compared with those of five truncated

forms The pattern of the full-length construct was

essentially identical to that of the C-terminal truncated

RDH2 forms (RDH21)229, RDH21)165, RDH21)137, and RDH21)58 RDH21)28) fused either to b-galactosi-dase (Fig 2N–Q) or to GFP (Fig 2G,H), whereas the b-galactosidase and GFP controls showed a signal, mainly in the nucleus (Fig 2L,R) The observation

that nuclear or cytosolic staining was not increased for

any construct suggested that the N-terminal segment

would be sufficient to target and anchor the protein to

the ER membranes Furthermore, we analyzed the

contribution of the C-terminal end to ER localization

We fused the last 41 amino acids of the amphioxus

RDH2 enzyme (a region equivalent to the reported

C-terminal segment of mouse RDH1 [27]) to GFP:

GFP-RDH2295)335 The protein did not localize to the

ER of transfected COS cells, but showed a diffuse signal similar to that of the GFP control (compare Fig 2K,L)

Evolution of the RDH group tblastn comparisons showed that the sequences most similar to the amphioxus enzymes were those of the vertebrate RDHs (E-value¼ 2e-75 and 1e-71 with Pan troglodytes, similar to sterol⁄ retinol dehydrogenase, for RDH1 and RDH2, respectively) In the phylogenetic analysis, amphioxus RDH branched outside a clade comprising the ‘classic’ SDR-RDH1⁄ 2 ⁄ 3 ⁄ 4 ⁄ 5 ⁄ 6 ⁄ 7 ⁄ 9 (RDH1–7⁄ 9) members, which includes six human enzymes [similar-RDH2, RDH4, orphan short-chain dehydrogenase/reductase (SDR-O), RDH, RDH5 and dehydrogenase/reductase member 9 (DHRS9)], eight rat forms (similar-RDH1, RDH2, similar-RDH2, RDH3, SDR-O, 17b-HSD19, RDH5 and DHRS9) and

11 mouse proteins [RDH1, RDH9, RDH6, truncated-RDH, similar-truncated-RDH, cis-retinol/androgen dehydrogen-ase (CRAD)-L, RDH7, SDR-O, 17b-HSD19, RDH5 and DHRS9] (Fig 4A) Except for DHRS9, the genes encoding these enzymes were not spread over several chromosomes, but rather clustered in the human gen-ome at 12q13–14 and in the syntenic regions of rat and mouse chromosomes 7 and 10, respectively [28] (Fig 4B) DHRS9 genes are located in human chromo-some 2, rat chromochromo-some 3 and mouse chromochromo-some 2, which would be paralogous to chromosomes 12, 7 and

10, respectively [29] The overall analysis, examining the topology of the phylogenetic tree and the position

of each gene inside the cluster, was informative regard-ing the orthology relationships of the distinct enzymes, and allowed us to define five RDH classes (Fig 4A) It

is of note that other vertebrate SDRs (some reported

as RDH enzymes), such as RDH8, RDH10, RDH11, RDH12, RDH13, RDH14, similar to epidermal retinal

Fig 4 (A) Phylogenetic relationship of the RDH1–7 ⁄ 9 and other retinoid ⁄ steroid active SDR forms from human (Hs), rat (Rn) and mouse (Mm) genomes with the amphioxus RDH enzymes A neighbor-joining tree was generated with the CLUSTALX program, and confidence in each node was assessed by 1000 bootstrap replicates The RDH1–7 ⁄ 9 cluster comprises all the vertebrate sequences that group with the amphioxus enzymes Additional enzymes involved in retinoid ⁄ steroid metabolism appear to be distantly related (less than 55% of sequence identity in the region used for the tree reconstruction, data not shown), and were therefore considered to be members of distinct SDR groups The bootstrap values defining each group are shown (black numbers) Internally, the RDH1–7 ⁄ 9 enzymes grouped into five classes, I–V The bootstrap values defining each class are shown (red numbers) (B) Structural organization of the human, rat and mouse RDH clus-ters using the Map Viewer website from NCBI The name of the each Rdh gene (black boxes) is indicated Alternative names for each gene are listed in supplementary Table S2 Orthology relationships among genes of several species are indicated (continuous lines) Notice that Rdh5 genes are in the same chromosome but outside the RDH clusters (dotted line), and that Dhrs9 genes are located in distinct chromo-somes Genes flanking the RDH sequences are also depicted (green boxes) TAC3, tachykinin 3; KIAA0352 (ZBTB29), zinc finger and BTB domain containing 39; ADMR, adrenomedullin receptor; PRIM1, primase polypeptide 1; NACA, nascent-polypeptide-associated complex alpha-polypeptide; CD63, CD63 antigen; BLOC1S1, biogenesis of lysosome-related organelles complex-1, subunit 1; ABCB11, ATP-binding cassette, subfamily B, member 11; LRP2, low-density lipoprotein receptor-related protein 2.

Fig 3 Synthesis of retinol and retinaldehyde in transfected COS-7

cells expressing amphioxus RDH enzymes (A) Different

concentra-tions (5, 10 and 20 l M ) of all-trans-retinaldehyde were added to

COS-7 cells expressing RDH1 (black bars) and RDH2 (white bars).

Retinol was extracted from the cells and analyzed by HPLC The

bars represent the net retinol production per mg of total protein

after 1 h of incubation with the three substrate concentrations (B)

Retinol oxidation in transfected cells was also evaluated for RDH1,

RDH2 and mouse RDH1 (gray bars), which was used as a positive

control of the reaction The bars represent the net retinal

produc-tion per mg of total protein after 1 h of incubaproduc-tion with 5, 10 and

20 l M all-trans-retinol.

B

dehydrogenase 2 (RaDH9), epidermal retinal

dehydro-genase 2 (eRaDH2), retSDR1, DHRS4, 17b-HSD11,

17b-HSD12, 11b-HSD11, 11b-HSD12 and short-chain

dehydrogenase/reductase 10 isoform B (SCDR10B),

branched outside the vertebrate–cephalochordate clade

(Fig 4A) and were located in diverse nonparalogous

mammalian chromosomes These enzymes would be

therefore distantly related to the RDH1–7⁄ 9 forms and

should be considered members of separate enzyme

families Indeed, the position of the amphioxus

enzy-mes in the phylogenetic tree implied that these families

are ancient, pre-dating the emergence of the chordate

phylum

Discussion

The biochemical characterization revealed that the

amphioxus RDH enzymes catalyzed all-trans-retinal

reduction (Table 1, Figs 1 and 3) Unfortunately,

com-parison with other RDHs was limited to rat RDH2,

mouse 17bHSD9 and RDH5, as the reductase capacity

of most vertebrate RDHs has not been assayed Thus,

we also compared amphioxus data with those from

other non-RDH vertebrate retinal reductases, such as

retSDR1, RalR1, PAN2, RRD, human AR, HSI, and

chicken AR (Table 1), although the comparison was

also hindered by the variety of assay conditions used

We have shown that amphioxus enzymes show

iso-mer preference, trans versus cis retinal forms, as occurs

with the vertebrate RalR1, PAN2, RRD, retSDR1,

prRDH, HSI and chicken AKR enzymes Moreover,

although retinol⁄ retinal interconversion is a reversible

reaction, neither amphioxus RDH1 or RDH2 showed

significant activity towards retinol in the in vitro assays

with microsomal proteins or in the intact cell systems

This strict preference towards the reductive direction

has been reported for other retinoid-active enzymes

(e.g the vertebrate retinal reductases RRD [13], HAR

[20,21] and prRDH [15]), whereas other enzymes

(RalR1 [30], PAN2 [17], HSI and chicken AKR

[20,21]) also catalyze retinol oxidation, albeit with

considerably lower efficiency The specific activities

towards all-trans-retinal of amphioxus enzymes were

0.25 nmolÆmin)1Æmg)1 for RDH1 and 1.4 nmolÆmin)1Æ

mg)1 for RDH2; these were 6.3-fold and 23-fold

higher, respectively, than that reported for retSDR1

(0.04 nmolÆmin)1Æmg)1), a photoreceptor enzyme that

reduces all-trans-retinal in the visual cycle [14] The

specific activity of RDH2 was 5.6-fold and 10.8-fold

higher than that of rat RDH2 (0.25 nmolÆmin)1Æmg)1)

[10] and mouse 17bHSD9 (0.13 nmolÆmin)1Æmg)1) [12],

respectively, whereas the activity of amphioxus RDH1

was comparable to those of these enzymes In contrast,

the amphioxus enzymes showed lower retinaldehyde reductase efficiency than some vertebrate enzymes The specific activity of mouse RDH5 with all-trans-retinal (16 nmolÆmin)1Æmg)1 [11]) was higher than that of either RDH1 or RDH2; RalR1 [30], PAN2 [17] and RRD [13] showed lower Kmand higher maximum spe-cific activity values for all-trans-retinal (Table 1) and therefore higher maximum specific activity⁄ Km ratios, which are a measure of the catalytic effectiveness of the enzymes; the AKR members (human AR, HSI and chicken AKR) [20,21] showed similar Km values but higher maximum specific activities (Table 1), which also implied higher maximum specific activity⁄ Km

ratios, i.e greater effectiveness, for the vertebrate AKR than for the cephalochordate forms Overall, these data support our finding that amphioxus RDH shows retinal reductase activity within the range repor-ted for diverse vertebrate enzymes

The most significant difference between the amphi-oxus and the other retinal reductases was, nevertheless, their preference for the NADH cofactor To our knowledge, these are the first SDR retinaldehyde reductases reported to use NADH instead of NADPH Conventionally, cofactor preference had been directly related to the oxidative or reductive direction of the reaction Therefore, it was assumed that oxidative RDHs would be NADH-dependent, whereas NADPH enzymes would catalyze the reductive reaction This hypothesis was based on the ratios between the oxid-ized and reduced forms of the coenzymes [31] It appears, however, that cofactor ratios vary greatly among organs and cell types, and that the redox status can be greatly influenced by external factors [32] Noticeably, amphioxus enzymes have the capacity to reduce retinaldehyde to retinol in intact cells (Fig 3), suggesting that the endogenous NADH level in COS-7 cells is enough to allow this reduction Our data sup-port the contention that coenzyme preference does not necessarily constrain the direction of the reaction In fact, several RDHs (e.g human, mouse and bovine RDH10 [33] and human RDH-E2 [32]) prefer NADP

to NAD as a cofactor

Structurally, most of the cytosolic SDR enzymes are composed of 250–280 amino acid residues [34,35], whereas the membrane-associated SDR enzymes are extended at both the N-terminal and C-terminal ends

by up to about 350 amino acids [36,37] Amphioxus RDH1 and RDH2 were 332 and 335 amino acids long, respectively, and their subcellular localization in trans-fected COS-7 cells concurred with that of ER-associ-ated proteins The observation that nuclear or cytosolic staining was not increased for any C-terminal truncated constructs indicated that the N-terminal

segment was sufficient to target and anchor the protein

to the ER membranes As the shortest segment was

only 28 amino acids long (from the initial methionine

to the LERGR motif), it can be assumed that the

sign-aling sequence for ER localization falls in this region

of the protein In addition to the targeting function,

signal sequences are also crucial in protein topogenesis,

as they participate in the final cytosolic⁄ lumenal

orien-tation The most prominent determinant of signal

ori-entation is the distribution of charged amino acids at

either end of the hydrophobic sequence According to

the ‘positive-inside’ rule, the most positively charged

flanking transmembrane segment is usually found on

the cytosolic side of the membrane [38,39] Amphioxus

RDH did not show positively charged amino acids at

the N-terminus of the signaling sequence, but rather

contained the L24ERGR motif at the C-terminus

of the hydrophobic sequence, thereby resembling

R19DRQ(S⁄ C) sequences of a number of vertebrate

RDHs [40] This motif would predict, therefore, a

cytosolic orientation of the amphioxus enzymes

For several RDH enzymes, the C-terminal

trans-membrane segment forms a hydrophobic

helix-turn-helix that is sufficient to retain them in the ER, e.g

CRAD1 [41] We fused the last 41 amino acids of the

amphioxus RDH2 enzyme (a region equivalent to the

reported C-terminal segment of mouse RDH1 [27]) to

GFP This protein did not localize to the ER of

trans-fected COS cells (Fig 2K), and therefore the

C-ter-minal end of amphioxus RDH2 was not sufficient for

ER targeting Amphioxus RDH2 was structurally

more similar to the enzymes that rely exclusively on

the N-terminal hydrophobic segment as membrane

anchor (e.g mouse RDH1 [27], human 11bHSD1 [42],

human 11bHSD2 [43] and human RalR1 [16]) than to

the other RDHs such as bovine RDH5 [44], mouse

RDH4 [45] and mouse CRAD1 [41,46], which would

be anchored to both the N-terminal and C-terminal

hydrophobic segments

Finally, evolutionary analysis including the

amphi-oxus enzymes highlighted the relevance of using

evolu-tionary criteria rather than biochemical classifications

for gene nomenclature and family description The

phylogenetic tree and the genomic organization now

permit a proper definition of the vertebrate RDH1–7⁄ 9

group and reveal an internal classification of

mamma-lian RDH1–7⁄ 9 enzymes into five classes, pointing to

recurrent gene tandem duplications as the most likely

mechanism for the cluster organization of the Rdh

genes In a recent study [47], Belyaeva & Kedishvili

proposed a model for the evolution of the vertebrate

RDH1–7⁄ 9 group (referred to as the RDOH-like SDR

group in their article) On the basis of a comparative genomic and phylogenetic analysis that included sev-eral vertebrate species, these authors suggested that early in vertebrate evolution, an initial tandem duplica-tion of the Rdh ancestor gave rise to the ‘Dhrs9⁄ Rdhl– 11-cis-RDH-homolog’ cluster The 11-cis-RDH-homolog gene was afterwards duplicated by a mechanism that implied translocation of the new copy to another region of the genome to generate the 11-cis-RDH⁄ Rdh5 gene Later on, the 11-cis-RDH⁄ Rdh5 gene underwent several tandem duplication events in its new chromoso-mal location, which led to the appearance of the cur-rent RDH cluster in tetrapods However, an alternative evolutionary model is possible (Fig 5) We hypothesize that an initial tandem duplication of an Rdh ancestor gave rise to a two-gene cluster, which was further duplicated, probably as a result of the genome duplica-tion events that took place during early vertebrate evo-lution [22] During fish evoevo-lution, one gene was lost, leading to the ‘Rdh5 (AAH97151) + Dhrs9 (Rdhl-like) + Rdhl’ combination currently found in zebrafish [47] In amphibians and mammals, extra tandem dupli-cations produced the RDH clusters found in Xenopus, human, mouse, rat, dog and cow [47] (Fig 4B) Even-tually, the mammalian Rdhl⁄ CX410306 ortholog was lost The closer phylogenetic relationship of RDH5⁄ 11-cis-RDH to Rdhl⁄ CX410306 enzymes than to RDH4-SDR-O-RDH forms [47] is consistent with this model Furthermore, the observation that the Ddrs9 genes and the Rdh5-RDH cluster are located in paralogous chro-mosomes also supports our hypothesis

In conclusion, the analysis of amphioxus enzymes contributes to improving our understanding of the functional complexity of vertebrate gene families regarding retinoid metabolism However, to date, no convincing enzymes for retinol oxidation have been found among cephalochordate RDH members The full genome sequence of amphioxus, currently being released, will allow comprehensive searches for novel candidates, which may also have relevant physiologic roles in the retinoid pathway of vertebrates

Experimental procedures

Expression of HA-RDH, GFP-RDH, and b-galactosidase-RDH proteins

To produce RDH1 and RDH2 proteins tagged at the N-terminus with the HA epitope, the full-length coding sequences of the Rdh1 and Rdh2 genes were PCR-amplified (oligonucleotides 1–2, and 3–4, respectively; the oligonucleo-tide sequences used in this study are provided in supple-mentary Table S1) from plasmids containing the Rdh1 and

Rdh2 cDNAs and cloned in the pACT2 vector (Clontech,

Mountain View, CA, USA) The HA-tagged Rdh1 and

Rdh2coding fragments were released from the pACT2

vec-tor and cloned into the pCDNA3 vecvec-tor (Invitrogen,

Carls-bad, CA, USA) To fuse amphioxus RDH2 either at the

N-terminus or C-terminus of the GFP, the full-length

coding region, two C-terminal truncated RDH2 forms and

one N-terminal truncated RDH2 form were PCR-amplified

(RDH21)335, oligonucleotides 5 and 6, amino acids 1–335,

full-length; RDH21)58, oligonucleotides 5 and 7, amino

acids 1–58, truncated just after the cofactor-binding

sequence GXXXGXG; RDH21)28, oligonucleotides 5 and

8, amino acids 1–28, truncated after the LERGR motif;

RDH2295)335, oligonucleotides 9 and 6, amino acids

295–335) and cloned into the pEGFP-N2 and pEGFP-C2

vectors (Clontech) To produce both the full-length and the

four C-terminal truncated RDH2 enzymes fused at the

N-terminus of b-galactosidase, the coding regions of Rdh2

were generated by PCR amplification: RDH21)335

(oligonu-cleotides 10–11; amino acids 1–335, full-length), RDH21)229

(oligonucleotides 10–12; amino acids 1–229, lacking the

C-terminal end), RDH21)165(oligonucleotides 10–13; amino

acids 1–165, truncated just before the active site, YXXXK),

RDH21)137 (oligonucleotides 10–14; amino acids 1–137,

truncated after the GLVNNAG region), and RDH21)58

(oligonucleotides 10–15; amino acids 1–58, truncated just

after the cofactor-binding sequence GXXXGXG) The design of these constructs was based on the predicted trans-membrane segments of the RDH2 enzyme given by the tmpred[48], das [49] and hmmtop [50] programs (data not shown) The five PCR fragments were cloned in the pbGal-N2 vector, in frame at the 5¢ end of the coding region of LacZ This vector expresses b-galactosidase protein fused

to a nuclear localization sequence (NLS) driven by the strong human cytomegalovirus immediate early promoter, and was created by cloning the NLS-LacZ gene from the PSP-1.72b-galactosidase plasmid [51] into a pEGFP-N2 vector from which the Gfp coding sequence had been removed The SV40 NLS localizes b-galactosidase to the nucleus All the constructs were verified by sequencing For subcellular localization experiments, COS-7 cells (African green monkey kidney cells; ECACC, Porton Down, Wiltshire, UK) were grown in DMEM with Gluta-MAX II (Invitrogen) and 4500 mgÆL)1 d-glucose, supple-mented with 10% fetal bovine serum, 100 UÆmL)1penicillin

G and 100 lgÆmL)1 streptomycin in a 5% CO2 humidified atmosphere at 37C Cells were seeded on glass coverslips into 12-well plates (5· 104cells per well) and transfected

24 h later with 0.5 lg of purified plasmid DNA per well, using 2.3 lL of FuGene6 (Roche, Basel, Switzerland) Cells were transfected with constructs encoding HA-RDH1, HA-RDH2, and the full-length and the four C-terminal

Fig 5 Hypothetical model of RDH1–7 ⁄ 9 group evolution leading to the current ver-tebrate multiplicity Fish and amphibian arrangements are those described for Danio rerio and Xenopus tropicalis in [47].