Substrate Recognition by a Dual-Function P450 Monooxygenase GfsF Involved in FD-891 Biosynthesis

Article abstract of DOI:10.1002/cbic.201700429

GfsF is a multifunctional P450 monooxygenase that catalyzes epoxidation and subsequent hydroxylation in the biosynthesis of macrolide polyketide FD-891. Here, we describe the biochemical and structural analysis of GfsF. To obtain the structural basis of a

Full text of DOI:10.1002/cbic.201700429

Accepted Article
Title: Substrate recognition by a dual functional P450 monooxygenase
GfsF involved in FD-891 biosynthesis
Authors: Tadashi Eguchi, Akimasa Miyanaga, Ryuichi Takayanagi,
Takashi Furuya, Ayano Kawamata, Tomohiro Itagaki,
Yoshiharu Iwabuchi, Naoki Kanoh, and Fumitaka Kudo
This manuscript has been accepted after peer review and appears as an
Accepted Article online prior to editing, proofing, and formal publication
of the final Version of Record (VoR). This work is currently citable by
using the Digital Object Identifier (DOI) given below. The VoR will be
published online in Early View as soon as possible and may be different
to this Accepted Article as a result of editing. Readers should obtain
the VoR from the journal website shown below when it is published
to ensure accuracy of information. The authors are responsible for the
content of this Accepted Article.
To be cited as: ChemBioChem 10.1002/cbic.201700429
Link to VoR: http://dx.doi.org/10.1002/cbic.201700429
A Journal of
www.chembiochem.org

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
Substrate recognition by a dual functional P450 monooxygenase
GfsF involved in FD-891 biosynthesis
Akimasa Miyanaga,[a] Ryuichi Takayanagi,[a] Takashi Furuya,[a] Ayano Kawamata,[b] Tomohiro Itagaki,[b]
Yoshiharu Iwabuchi,[b] Naoki Kanoh,[b] Fumitaka Kudo[a] and Tadashi Eguchi*[a]
Abstract: GfsF is a multifunctional P450 monooxygenase that
catalyzes the epoxidation and subsequent hydroxylation in the
biosynthesis of macrolide polyketide FD-891. Here, we describe the
biochemical and structural analysis of GfsF. To obtain the structural
basis of a dual functional reaction, we determined the crystal
structure of ligand-free GfsF, which revealed GfsF to have a
predominantly hydrophobic substrate binding pocket. The docking
models in conjunction with the results of the enzymatic assay with
substrate analogs as well as site-directed mutagenesis suggested
two distinct substrate binding modes for epoxidation and
hydroxylation reactions, which explained how GfsF regulates the
order of two oxidative reactions. These findings provide new insights
catalytic steps. Each oxidation step is a prerequisite for the next
oxidation step. Although some of the multifunctional P450s have
been characterized in terms of function and structure, the
underlying mechanisms for their diverse reactivity and ordered
reaction sequence are not well understood.
FD-891 (1a) is a 16-membered macrolide antibiotic isolated
from Streptomyces graminofaciens A-8890.[7] FD-891 (1a)
shows toxicity against human leukemia (HL-60) cells. The FD-
891 (1a) biosynthetic gene cluster has been identified and found
to contain five type I polyketide synthase (PKS) genes (gfsA,
gfsB, gfsC, gfsD and gfsE) and two post-PKS modification
enzyme genes (gfsF and gfsG).[8] The five PKSs are responsible
for the biosynthesis of FD-892 (2c). Cytochrome P450 GfsF and
methyltransferase GfsG catalyze the post-PKS modification in
parallel.[9] GfsF catalyzes the epoxidation at the C8–C9 olefin of
FD-892 (2c) and 25-O-methyl-FD-892 (1c) and subsequent
hydroxylation at the C10 position of the resulting intermediate to
produce 25-O-demethyl-FD-891 (2a) and FD-891 (1a),
respectively (Scheme 1). GfsG catalyzes the O-methylation at
the C25 hydroxy group of FD-892 (2c) and 25-O-demethyl-FD-
891 (2a) to produce 25-O-methyl-FD-892 (1c) and FD-891 (1a),
respectively. Among these post-PKS reactions, GfsF catalyzes
epoxidation and hydroxylation in a stepwise manner. The order
of the two oxidative reactions is strictly regulated. It is unclear
how GfsF regulates the order of dual reactions.
into
the
reaction
mechanism
of
multifunctional
P450
monooxygenases.
Introduction
Cytochrome P450 monooxygenases are a large superfamily of
heme-containing enzymes. They catalyze various regio- and
stereo-specific oxidative reactions such as hydroxylation,
epoxidation, dehydrogenation, demethylation, aryl–aryl coupling
and C–C bond cleavage in primary and secondary metabolism.[1]
P450 enzymes are often involved in the biosynthesis of natural
products in bacteria, fungi and plants, and provide the structural
diversification of natural products.[2] In general, P450s catalyze a
single oxidative reaction, but some P450s catalyze multistep
oxidative reactions at separate sites of the substrate.[3] For
example, MycG catalyzes sequential hydroxylation and
epoxidation reactions in mycinamicin biosynthesis (Scheme
S1A).[4] AurH catalyzes the hydroxylation and the
tetrahydrofuran ring formation in aureothin biosynthesis
(Scheme S1B).[5] TamI catalyzes two hydroxylations and one
epoxidation in tirandamycin biosynthesis.[6] These multifunctional
P450s seem to have an apparent hierarchy in the order of
Here, we describe the biochemical characterization and
structural determination of GfsF. The structural data combined
with
a computational docking study and mutational study
provides insights into the mechanism of substrate recognition
important for tandem oxygenation processes.
Results
Substrate Specificity of GfsF. Previously, we showed that
GfsF can accept 25-O-methyl-FD-892 (1c) and FD-892 (2c) as
substrates to produce FD-891 (1a) and 25-O-demethyl-FD-891
(2a), respectively (Scheme 1).[9] To further understand the
substrate specificity of GfsF, we conducted the reaction of GfsF
with 1c and 2c to compare the velocity of the GfsF reaction. As
a result, GfsF efficiently converted 2c to 2a, whereas GfsF
converted 1c to 1a with lower efficiency (Figure 1A, B). This
result suggested that GfsF prefers the hydroxy group at the C25
position rather than the methoxy group. We also found that GfsF
catalyzed the epoxidation step faster than the hydroxylation step.
After 1c or 2c was almost converted to 1b or 2b in the GfsF
reaction, the formation of 1a or 2a started to be observed,
respectively. We estimated both
[a]
Dr. A. Miyanaga, R. Takayanagi, T. Furuya, Professor Dr. F.
Kudo, Professor Dr. T. Eguchi
Department of Chemistry
Tokyo Institute of Technology
2-12-1 O-okayama, Meguro-ku, Tokyo 152-8551 (Japan)
E-mail: eguchi@chem.titech.ac.jp
A. Kawamata, T. Itagaki, Professor Dr. Y. Iwabuchi, Professor
Dr. N. Kanoh
Graduate School of Pharmaceutical Sciences
Tohoku University
[b]
6-3 Aza-aoba, Aramaki, Aoba-ku, Sendai 980-8578, (Japan)
Supporting information for this article is given via a link at the end
of the document.
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
O
O
O
HO
HO
HO
7
OH OH OH
OH OH OH
OH OH OH
O
O
O
O
O
GfsF
GfsF
25
10
OH
FD-892 (2c)
GfsG
8,9-epoxy-FD-892 (2b)
25-O-demethyl-FD-891 (2a)
GfsG
O
O
O
HO
HO
HO
OH OH OCH3
OH OH OCH3
O
O
OH OH OCH3
O
O
O
GfsF
GfsF
OH
25-O-methyl-FD-892 (1c)
10-deoxy-FD-891 (1b)
FD-891 (1a)
Scheme 1. Parallel biosynthetic pathway from FD-892 (2c) to FD-891 (1a)
Figure 1. Time-course of the reactions of GfsF wild-type with 1c (A) and 2c (B), and GfsF T300V mutant with 2c (C). HPLC traces (detection at 275 nm) are
shown.
initial velocities of epoxidation (0.21 units/mg for 1c and 1.5
units/mg for 2c) and hydroxylation (0.045 units/mg for 1b and
0.23 units/mg for 2b) reactions (Figure S1).
Next, we investigated the GfsF reaction with substrate
analogs (Scheme 2) to obtain an insight into the substrate
formation rate of the epoxy product was significantly lower
(0.96% of relative activity towards 2c). Conversely, GfsF showed
no detectable hydroxylation activity (<0.05% of relative activity
towards 2b), suggesting that GfsF strictly recognizes the
presence of a trihydroxyalkyl side-chain in the hydroxylation
specificity in each oxidation step of the GfsF reaction. Previously, reaction. Thus, GfsF seems to recognize the alkyl side-chain
we synthesized FD-892 analog 3c that has a shorter C16–C18
alkyl side-chain,[10] so we first carried out the reaction with 3c to
investigate whether the alkyl side-chain moiety is important for
the GfsF reaction. We analyzed the reaction product by high-
performance liquid chromatography (HPLC) and liquid
chromatography-mass spectrometry (LC-MS) and then found
that GfsF catalyzed only a single oxidation to produce 3b
(Figure 2A). We isolated 3b and analyzed its chemical structure
by nuclear magnetic resonance (NMR) spectroscopy, which
showed that 3b has an epoxide at the C8–C9 positions and the
methylene at the C10 position (Table S1 and Figures S2–S3).
This finding suggested that GfsF catalyzed only the epoxidation
of the double bond at the C8–C9 position of 3c (Fig. 2B). Thus,
the recognition of a trihydroxyalkyl side-chain moiety does not
seem to be critical for the epoxidation step, although the
moiety in
hydroxylation reactions.
a different manner between epoxidation and
O
HO
OH OH OH
OH OH OH
O
R
FD-892 (2c)
OH
OH OH
OH OH OH
R
OH
R
R
R
3c
4c
5c
6c
O
HO
O
OH
OH
3d
Scheme 2. Synthetic analogs that have a truncated alkyl side-chain
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
Figure 2. Reactions of GfsF wild-type with 3c (A and B) and 7c (C and D). (A) Time-course of the reaction with 3c. (B) Formation of 3b from 3c. (C) Time-
course of the reaction with 7c. (B) Formation of 1a from 7c. (A and C) HPLC traces (detection at 275 nm) are shown.
To further investigate the recognition of an alkyl side-chain
moiety in the GfsF reaction, we synthesized additional FD-892
analogs possessing a truncated alkyl side-chain using the same
strategy as for the synthesis of FD-891 side-chain truncated
analogs[11] and examined the reaction (Figures S1 and S4). In
the reaction with FD-892 analog 4c that has a monohydroxyalkyl
side-chain (i.e., C16–C21), GfsF showed moderate epoxidation
activity (35% relative activity towards 2c) and significantly low
hydroxylation activity (2.7% relative activity towards 2b). These
findings suggested that the C21 hydroxy group of substrate is
recognized by GfsF in the epoxidation step because the
epoxidation activity against 4c was 36-fold higher than that
against 3c. In the reaction with FD-892 analog 5c that has a
dihydroxyalkyl side-chain (i.e., C16–C23), GfsF also showed
moderate epoxidation activity (25% relative activity towards 2c)
and significantly low hydroxylation activity (1.4% relative activity
towards 2b), both of which were at the same level as those with
4c. The presence of a C23 hydroxy group did not improve
epoxidation or hydroxylation activity, suggesting that GfsF might
not recognize the C23 hydroxy group. In the reaction with FD-
892 analog 6c that has a one-carbon truncated alkyl side-chain
(i.e., C16–C25), GfsF efficiently catalyzed epoxidation (190%
relative activity towards 2c) and hydroxylation (260% relative
activity towards 2b). GfsF seems to strictly recognize the C24–
C25 region in the hydroxylation step because the hydroxylation
activity against 6b was 190-fold higher than that against 5b.
GfsF showed even higher epoxidation and hydroxylation
activities against 6c than those against FD-892 (2c). The flexible
C25 hydroxy group of 6c might facilitate the interaction with
GfsF.
To investigate whether GfsF recognizes the C7 hydroxy
group of the substrate, we synthesized 7,25-O-dimethyl-FD-892
(7c) from 25-O-methyl-FD-892 (1c). Unexpectedly, GfsF
converted 7c to FD-891 (1a) in the reaction (Figure 2C). This
implied that GfsF accepted 7c as a substrate and catalyzed the
demethylation at the C7 hydroxy group during the reaction. To
clarify the timing of demethylation, we investigated the time-
course of this conversion (Figure 2C). We found that the
reaction with 7c gave two reaction intermediates, one of which
was 10-deoxy-FD-891 (1b). We isolated another reaction
intermediate, 7b, and structurally identified it as 7-O-methyl-10-
deoxy-FD-891 by LC-MS and NMR (Table S2 and Figures S5–
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
S6). We did not detect the formation of 7-O-methyl-FD-891 or
25-O-methyl-FD-892 (1c) from 7c. We also carried out the
reaction with 7b, which was converted to FD-891 (1a) via 10-
deoxy-FD-891 (1b) (Figure S7). These results clearly suggested
that the C7 methoxy group was retained in the epoxidation step
and that demethylation occurred before the hydroxylation step
(Figure 2D). We detected the formation of formaldehyde (Figure
S8), which was concomitant with the conversion of 7c to 1a.
This observation confirmed that GfsF catalyzed the
hydroxylation of the methyl group at the C7 methoxy moiety of
7b to produce 1b via a hemiacetal form.
We also carried out the reaction with FD-892 analog 3d that
has a hydroxy group at the C10 position[10] to obtain an insight
into the order of dual oxidative reactions. As a result, GfsF did
not catalyze the epoxidation of 3d. Thus, the hydroxylation at the
C10 position prior to the epoxidation at the C8–C9 positions
completely blocked the further oxidative reaction of GfsF.
invariant Cys363 as an axial thiolate (Figures 3 and S9). A water
molecule is coordinated to the heme iron as a distal pocket
ligand, as observed in many ligand-free P450 structures.[2a] The
overall structure of GfsFΔN15 is most similar to that of P450nor
(PDB code 2ROM; Z-score = 52.2, rmsd of 1.3 Å, sequence
identity of 38%), which catalyzes the reduction of nitric oxide in
Fusarium oxysporum.[13] GfsFΔN15 also shows structural
similarity with post-PKS modification P450s such as CYP105P1
(PDB code 3E5L; Z-score = 49.0, rmsd of 1.7 Å, sequence
identity of 42%) and CYP105D6 (PDB code 3ABB; Z-score =
48.9, rmsd of 1.6 Å, sequence identity of 44%), both of which
are involved in filipin biosynthesis in Streptomyces avermitilis.[14]
In addition, GfsFΔN15 shows similarity with multifunctional
P450s such as MycG[4] (PDB code 2Y5N; Z-score = 47.6, rmsd
of 1.9 Å, sequence identity of 36%) and AurH[5] (PDB code
3P3X; Z-score = 41.5, rmsd of 2.2 Å, sequence identity of 28%).
The FG helices, which are known to undergo closing motion
on ligand binding in P450s,[2,14b] adopt an open conformation in
the GfsFΔN15 structure. The BC loop located at the entrance of
the substrate binding pocket shows a relatively high B-factor and
contains a disordered region (Met94–Pro97), suggesting the
flexibility of the BC loop. These structural features are, in
general, observed in ligand-free P450 structures.[2,14b] GfsFΔN15
has a predominantly hydrophobic pocket constructed by the BC
loop (His75, Tyr82–Ile85, Phe89 and Phe99–Gly101), the I helix
(Met244, Gly247, Ala248 and Thr252), the loop after the K helix
(Thr300), and the C-terminal loop (Ile403 and Ala404). We
observed the electron density for di(hydroxyethyl)ether molecule,
which likely originated from polyethylene glycol (PEG) in the
crystallization solution, in the substrate binding pocket (Figure
Crystal Structure of GfsF. To understand the structural basis
of the dual oxidation reactions, we attempted to crystallize GfsF,
but we failed. Computational analysis using Protein DisOrder
prediction System (PrDOS)[12] suggested that the first 15 amino
acid residues at the N-terminus of GfsF are flexible. To minimize
the conformational flexibility, we constructed the heterologous
expression system of GfsFΔN15 protein that lacks the N-
terminal 15 amino acid residues. After we confirmed the activity
of GfsFΔN15 protein, we succeeded in the crystallization of
GfsFΔN15 and determined the ligand-free structure at 2.0 Å
resolution (Table S3).
GfsFΔN15 has a typical P450 fold and contains a heme with
4A).
A PEG-type molecule was reported to bind in the
hydrophobic tunnel or pocket in some enzymes.[15] The
di(hydroxyethyl)ether molecule is positioned almost parallel to
the heme molecule (3.8–4.5 Å). The di(hydroxyethyl)ether
molecule forms a hydrogen bond (3.3 Å) with Thr300 and makes
hydrophobic contacts (3.6–4.1 Å) with Ala248 and Thr252. The
Thr252 residue of GfsFΔN15 is positioned at the same position
as the corresponding Thr residues in many P450s, such as
P450cam, CYP105P1 and CYP105D6. This conserved Thr
residue is proposed to play an important role in protonation for
oxygen activation in the P450 reaction.[16]
G helix
BC loop
F helix
heme
N
I helix
Cys363
C
Figure 3. Overall structure of GfsFΔN15. The N- and C-terminals of
GfsFΔN15 are denoted as N and C, respectively. The BC loop region
(Asp70–Asp104), FG helix region (Tyr171–Leu213),
I helix region
(Gln235–Glu266) and heme are shown in blue, orange, cyan and yellow,
respectively.
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
A
D
His84
Tyr82
His84
Tyr82
Ile85
Ile85
Ile403
Ile403
His75
His75
Ala100
Phe89
Phe89
Ala100
Thr300
Thr300
10
7
Heme
Heme
Thr252
Thr252
I helix
I helix
B
2c
2c
most stable
conformer
2b
most stable
conformer
matastable
conformer
(2.3 kcal/mol
higher in
free energy)
10
10
10
7
7
7
C
E
His84
His84
Tyr82
Tyr82
Ile85
Ile85
Ile403
His75
Ala100
His75
Ala100
Ile403
Phe89
Phe89
Thr300
Thr300
10
10
Heme
7
7
Heme
Thr252
Thr252
I helix
I helix
Figure 4. GfsF docking models. The GfsFΔN15 and docked FD-892-related molecules are shown in green and cyan, respectively. The C7 and C10
positions of docked molecules are denoted as 7 and 10, respectively. The hydrogen bond with Thr300 is shown as a blue broken line. The olefin
bonds of docked molecules are shown as white dotted lines. (A) The crystal structure of GfsFΔN15. The bound di(hydroxyethyl)ether molecule is
shown as gray sticks. (B) A comparison of stable conformers for FD-892 (2c) and 8,9-epoxy-FD-892 (2b) macrolactones. 2c and 2b macrolactone
fragments that have a truncated C16–C17 alkyl side-chain were used for MacroModel calculation. (C) The docking model with FD-892 (2c). (D) The
docking model with 8,9-epoxy-FD-892 (2b). (E) The docking model with 7-O-methyl-8,9-epoxy-FD-892 (7b).
Docking Analysis and Mutational Study. We attempted to
cocrystallize GfsFΔN15 with a substrate or substrate analog.
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
However, these attempts were unsuccessful. Therefore, we
conducted computational docking analysis with FD-892 (2c) and
(Figure S1). The presence of an alkyl side-chain moiety might be
important for fixing the position of the macrolide ring for the
hydroxylation reaction. As mentioned above, Phe89 seems to be
important for fixing the macrolide ring in the epoxidation reaction
and for binding the alkyl side-chain in the hydroxylation reaction.
To confirm the role of Phe89 in both reactions, we constructed a
F89A mutant. The F89A mutant showed significantly low
epoxidation activity (2.3% of relative activity of the wild-type) and
no detectable hydroxylation activity in the reaction with 2c.
The position of the C7 hydroxy group is different between
FD-892 (2c) and 8,9-epoxy-FD-892 (2b) in the docking models.
The C7 hydroxy group of 2c is directed toward the space in the
substrate binding pocket and shows no direct interaction with
any GfsF residue (Figure 4C). Thus, the methoxy group at the
C7 position could be accommodated with no steric hindrance in
this binding mode, which is consistent with the result that GfsF
catalyzed the epoxidation of 7c. In contrast, the C7 hydroxy
group of 2b forms a hydrogen bond with the side-chain hydroxy
group of Thr300 (Figure 4D). This finding might explain why the
hydroxylation reaction occurred only after the demethylation at
the C7 hydroxy group in the reaction with 7c. To evaluate the
importance of Thr300 in the hydroxylation reaction, we
constructed T300V and T300A mutants. The T300V mutant
retained the epoxidation activity towards 2c (250% relative
activity of the wild-type), suggesting that Thr300 is not involved
in the epoxidation step. The substitution of Thr300 with Val
might even be preferable in the epoxidation step because the
C3–C5-conjugated olefin region was docked close to Thr300 in
the 2c docking model (Figure 4C). Conversely, the T300V
mutant completely lost the hydroxylation activity (Figure 1C).
The substitution of Thr300 with Val likely caused repulsion with
the C7 hydroxy group of 2b. Similarly, the T300A mutant
retained the epoxidation activity (93% relative activity of the wild-
type) and reduced the hydroxylation activity (13% relative
activity of the wild-type), probably because of the loss of
interaction with the C7 hydroxy group in the hydroxylation step.
Thus, the result of the mutational experiment supported the
docking models.
8,9-epoxy-FD-892 (2b) models whose macrolide rings are
®
energy-minimized by MacroModel
software (Schrödinger,
New York, NY, USA). Based on MacroModel calculation, the
macrolide ring of 2c seems to adopt multiple conformations
(Figures 4B and S10). The conformation of the most stable
conformer closely resembles the X-ray structure of the FD-892
derivative.[10] However, the most stable 2c conformer did not
correctly dock near the heme cofactor. Therefore, we conducted
the docking analysis with several other 2c conformers. We found
that the metastable 2c conformer (2.3 kcal/mol higher in free
energy) showing a different conformation in the C1–C9 polyene
region was docked near the heme cofactor (Figure 4C). The C6
methyl and C7 hydroxy groups of the most stable conformer are
oriented horizontally along the macrolide ring, whereas those of
the metastable conformer are oriented almost vertically. The
metastable conformation of 2c might be necessary to fit into the
substrate binding pocket of GfsF. In particular, the vertically
oriented conformations of C6 methyl and C7 hydroxy groups
seem to be necessary for C8–C9 olefin to access near the heme
group without steric hindrance. The C8–C9 olefin is located
within 4.7 Å of the heme iron and positioned to give an epoxide
with the correct stereochemistry in FD-891 (1a). In the case of
the 2b docking model, the most stable conformer was well
docked (Figure 4D). The orientations of C6 and C7 substituent
groups of the 2b conformer are similar to those of the most
stable 2c conformer (Figures 4B, S10 and S11). The C10
hydroxylation site of 2b is located near the heme iron (4.0 Å)
and positioned to give
a hydroxy group with the correct
stereochemistry in the docking model.
We found that the position of macrolide ring is significantly
different between FD-892 (2c) and 8,9-epoxy-FD-892 (2b)
binding modes (Figure 4C, D). In the 2c docking model, the
macrolide ring is sandwiched between the heme and Phe89,
which seems to be important for fixing the position of the C8–C9
olefin region for the epoxidation reaction (Figure 4C). The
macrolide ring moiety forms hydrophobic interactions with Ile85,
Phe89, Ala100 and Ile403. The alkyl side-chain moiety is
positioned around the flexible BC loop region. The C21 and C25
hydroxy groups are located close to the main-chain carbonyl
oxygen atoms of Phe89 and His75, respectively. In the docking
model with 2b, the macrolide ring is bound near the I helix,
which is on the opposite side from Phe89 (Figure 4D). The
introduced C8–C9 epoxide does not interact with the GfsF
residue. The region between the C7 hydroxy and C12 methylene
groups of the docked 2b molecule almost overlaps with the
To obtain mechanistic insights into the demethylation
process, we conducted docking analysis with 7b. In the docking
model with 7b, the positions of the macrolide ring and alkyl side-
chain moieties of 7b occupies relatively similar positions to those
of 8,9-epoxy-FD-892 (2b) rather than those of FD-892 (2c)
(Figures 4D, E). However, the C7 position of 7b is moved away
from Thr300 by ≈3 Å compared with that of 2b. The methyl
group of the C7 methoxy moiety of 7b is placed close to the
heme iron (3.1 Å) in the docking model so that GfsF could
di(hydroxyethyl)ether molecule in the crystal structure (Figure 4A, catalyze the hydroxylation of the methyl group to produce 10-
D). The alkyl side-chain moiety is oriented perpendicular to the
macrolide ring, and the C21–C25 region reaches into the BC
loop region. The C21–C25 region is anchored in the
hydrophobic groove constructed by the side-chains of Tyr82,
Ile85 and Phe89. The C24 methyl group is located close to the
side-chains of Ile85 and Phe89, and the C25 hydroxy group is
close to the main-chain nitrogen atom of Ile85. This observation
seems to be consistent with the result that the hydroxylation
activity against 6b was 190-fold higher than that against 5b
deoxy-FD-891 (1b) via a hemiacetal form. Because the methoxy
group is not bulky, GfsF might be able to accommodate 7b in
the substrate binding pocket for the demethylation reaction.
Discussion
Several multifunctional P450s have been structurally
analyzed and their substrate recognition mechanisms proposed.
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
For example, the complex structure of MycG with its substrate
was determined by Li and colleagues.[4] The C14 hydroxylation
site and the C12–C13 epoxidation site of the substrate are 8.9 Å
and 10.0 Å away from the heme iron in the structure of the
MycG complex, respectively. Li and colleagues proposed that
this substrate binding mode is the initial recognition mode and
that the substrate translocates from the initial recognition site to
the active site of the enzyme. Translocation of the substrates in
fit well at the active site of GfsF toward the efficient
hydroxylation at C10.
A comparison of GfsF reactions with FD-892 substrates (1c
and 2c) and analogs (3c, 4c, 5c and 6c) suggested that the
presence of an alkyl side-chain moiety affects the reaction
efficiencies of epoxidation and hydroxylation steps (Figures 1,
2A, S1 and S3). The presence of C21 and C25 hydroxy groups
greatly improved the reaction efficiency, suggesting that these
two directions might account for the bifunctional activity of MycG. hydroxy groups are recognized by GfsF. In addition, the C24
In the case of AurH, docking studies have suggested that Gln91
forms a hydrogen bond with the introduced hydroxy group of the
hydroxylated intermediate.[5] The hydroxylated intermediate was
proposed to be pushed deeper into the substrate binding pocket
so that the C9 methyl group could be placed closer to the heme
iron for the second oxygenation–heterocylization step. The
relocation also likely happened during the GfsF reaction based
on the docking models with FD-892 (2c) and 8,9-epoxy-FD-892
(2b). The docking models suggested two substrate binding
modes for epoxidation and hydroxylation reactions.
In the GfsF reaction, only epoxidized compounds were
initially generated from all of the accepted FD-892-related
compounds (1c–7c). The conformation of macrolide ring could
influence the order of catalytic steps. GfsF seems to select the
metastable conformer of FD-892 (2c) rather than the most stable
conformer with the hydrophobic substrate binding pocket. The
conformation of the metastable conformer of 2c is important for
fixing the C8–9 site near the heme iron in the first epoxidation
reaction. GfsF hydrophobic residues such as Phe89 contribute
greatly to the correct placement of 2c in the substrate binding
pocket (Figure 4C). The substitution of olefin at C8–C9 with
epoxide causes a conformational change in the macrolide ring,
which allows 8,9-epoxy-FD-892 (2b) to relocate and adopt the
orientation in which the C10 site is placed close to the heme iron
for the second hydroxylation step. GfsF recognizes the
macrolide ring moiety of the substrate mainly through
hydrophobic interactions, although the hydrogen bond between
Thr300 and the C7 hydroxy group is important for the second
hydroxylation step. These predominant hydrophobic interactions
might allow two distinct substrate binding modes.
It is known that several P450 enzymes catalyze both double
bond epoxidation and allylic hydroxylation of cyclohexene,[17]
although the double oxidation of cyclohexene has never been
reported. The reaction energy barriers of the double bond
epoxidation and the allylic hydroxylation of cyclohexene were
calculated to be similar.[18] However, actually, the reactivity of
alkenes largely depends on the structures of substrate and
enzyme.[17] In the GfsF reaction, the reaction order seems to be
completely controlled by the substrate structure, presumably due
to the preferable conformation of FD-892 macrolactone as
described above. The rate of the second hydroxylation was 15%
of that of the first epoxidation in the reaction with 2c, even
though the available substrate in the second hydroxylation was
limited in this highly ordered oxidation. Thus, the rate of the
second hydroxylation at the sp3 carbon adjacent to the oxirane
ring was not so significantly slower than the first epoxidation.
The preferable conformation of the epoxidized compounds could
methyl group seems to be important for the hydroxylation step.
Previously, we proposed the parallel post-PKS modification by
GfsF and GfsG in FD-891 (1a) biosynthesis (Scheme 1),[9] but
the conversion from 2c to 1a via 25-O-demethyl-FD-891 (2a)
might be the preferred pathway in the producer strain, as judged
from the substrate preference of GfsF. GfsF catalyzed the
epoxidation of 3c with low efficiency, whereas GfsF showed no
hydroxylation activity (Figure 2A). Thus, the macrolide ring itself
is not sufficient for the productive binding of the substrate in the
hydroxylation step. The presence of an alkyl side-chain moiety
seems to be necessary for the position of the macrolide ring of
the substrate in the hydroxylation reaction.
Conclusions
We conducted biochemical and structural analyses on the
multifunctional P450 monooxygenase GfsF, which catalyzes the
epoxidation and hydroxylation reactions in the biosynthesis of
the macrolide antibiotic FD-891. We determined the crystal
structure of ligand-free GfsF, which enabled the computational
docking analysis with FD-892 and 8,9-epoxy-FD-892. The
docking models in conjunction with the results of the enzymatic
assay with FD-892 analogs and site-directed mutagenesis
suggested two distinct substrate binding modes for epoxidation
and hydroxylation reactions. The substitution of C8–C9 olefin of
FD-892 with epoxide causes a conformational change in the
macrolide ring, which allows 8,9-epoxy-FD-892 to relocate for
the second hydroxylation reaction. These results provide an
insight into how GfsF regulates the order of epoxidation and
hydroxylation reactions.
Experimental Section
Synthesis of FD-892 Side-chain Truncated Analogs (4c, 5c and 6c).
See details in the Supporting Information.
Synthesis of 7,25-O-dimethyl-FD-892 (7c). See details in the
Supporting Information.
Preparation of Recombinant GfsF Protein and Related Proteins.
Escherichia coli BL21(DE3) cells harboring pET28b-gfsF plasmids[9] were
grown at 28 °C in Luria–Bertani broth containing kanamycin (50 µg mL−1).
After the optical density at 600 nm reached 0.6, protein expression was
induced by the addition of isopropyl β-D-1-thiogalactopyranoside (0.05
mM), FeCl3 (0.15 mM) and 5-aminolevulinic acid (40 µg mL−1), and the
cells were then cultured for an additional 16 h at 28 °C. The recombinant
protein with an N-terminal His-tag was collected from cell-free extracts
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
prepared by sonication, and was purified on a His60 Ni-Superflow affinity
column (Clontech, Mountain View, CA, USA). The protein was then
desalted and concentrated using a PD-10 column (GE Healthcare,
Buckinghamshire, UK) and an Amicon Ultra centrifugal filter (Merck
Millipore, Billerica, MA, USA), respectively. For preparation of GfsFΔN15
protein, the gfsFΔN15 fragment was first amplified from pET28b-gfsF
with the primers 5'-AAAAAAACATATGGCCCCCGAGTGGCCC-3' and 5'-
ATGCTAGTTATTGCTCAGCGG-3' and then inserted between the NdeI
and XhoI sites of pET28b. The resulting pET28b-gfsFΔN15 was
transformed into E. coli BL21(DE3) cells. The GfsFΔN15 protein was
expressed and purified by a His60 Ni-Superflow affinity column as
described above. The purified GfsFΔN15 protein was further treated with
thrombin to remove the His-tag for crystallization. After subjection to
another round of a His60 Ni-Superflow affinity chromatography, the His-
tag free GfsFΔN15 protein was purified by Resource Q (GE Healthcare)
anion-exchange chromatography with a linear gradient from 0.15 to 0.45
M NaCl in 10 mM HEPES-Na buffer (pH 7.7) containing 10% (v/v)
glycerol. The purified GfsFΔN15 protein was concentrated to 12.5 mg/mL
in 10 mM HEPES-Na (pH 7.7). The sodium dithionite-reduced GfsF
solution was bubbled with carbon monoxide and then analyzed with a
spectrophotometer UV-2450 (Shimadzu, Tokyo, Japan) to determine the
spectroscopy signals are shown in the Supporting information (Table S1
and Figures S2–S3).
Isolation and Structural Determination of 7b. 0.5 mM 7c (20 mg) were
mixed with 3 µM GfsF, 10 µM CamA, 20 µM CamB and 1 mM NADH in
20 mM HEPES-Na (pH 7.7) buffer containing 10% glycerol and 1%
DMSO in a total volume of 71.5 mL. The reaction was carried out at
28 °C with agitation (600 rpm) for 20 min. After the reaction, the reaction
product was extracted with EtOAc. The organic layers were dried on
Na2SO4 and the solvent was removed by evaporation. The crude residue
was purified by HPLC (90% CH3OH) to give 7b (3.4 mg). HR-FAB-MS
(positive mode): m/z calculated for C34H57O7: 577.4104 ([M+H]+); found:
577.4106. The assignments of the 1H and 13C NMR spectroscopy signals
are shown in the Supporting information (Table S2, Figures S5–S6).
Crystallization, Data Collection and Structural Determination.
GfsFΔN15 crystals were grown from a 1:1 mixture of a protein solution
(12.5 mg mL−1 in 10 mM HEPES-Na (pH 7.5)) and a reservoir solution
containing 0.2 M KCl, 0.1 M Tris-HCl (pH 8.5) and 27.5% PEG3350 using
the sitting-drop vapor diffusion method at 26 °C. Prior to collection of the
X-ray data, the crystals were flash-frozen in a stream of liquid nitrogen.
The X-ray diffraction data were collected on a beamline BL-5A at the
Photon Factory (Tsukuba, Japan) and were subsequently indexed,
integrated, and scaled using the HKL2000 program.[20] The initial phase
was determined by molecular replacement using the Molrep program[21]
with the MoxA structure (PDB code: 2Z36)[22] as a search model. Model
building of GfsF was carried out automatically with the ARP/wARP
program[23] and subsequently inspected by Coot.[24] Refmac[25] was used
for refinement of the structures. The structural representations were
prepared with PyMOL (DeLano Scientific, Palo Alto, CA, USA). The
geometries of the final structure were evaluated using the program
Rampage.[26] The resulting coordinates and structure factors have been
deposited in the Protein Data Bank (PDB code: 5Y1I).
functional P450 concentration using an extinction coefficient of 91,000
[19]
M−1cm−1
.
CamA and CamB proteins were prepared as described
previously.[9]
GfsF Reaction. The GfsF assay mixture (100 µL each) consisted of 0.1
mM 1c, 2c or FD-892 analogs, 1 mM NADH, 30 nM or 3 µM GfsF, 8 µM
CamA, and 20 µM CamB in 50 mM HEPES-Na (pH 7.7) containing 10%
glycerol. The enzymatic reaction was carried out at 28°C for 1 min–3 h.
The product of the enzymatic reaction was extracted twice using 150 µL
of EtOAc, and the solvents from the combined organic layers were
removed using a centrifugal evaporator. The residue was dissolved in 10
µL of CH3OH, and a 5-µl aliquot was analyzed by HPLC using an ELITE
LaChrom L-2455 DAD Detector and L-2130 Pump (Hitachi, Tokyo,
Japan) equipped with a PEGASIL ODS column (100 Å, 250 × 4.6 mm;
Senshu, Tokyo, Japan). Except for the reaction with 7c, CH3OH (80%) in
water was used as an eluent at a flow rate of 0.7 mL min−1. In the
reaction with 7c, CH3OH (90%) in water was used as an eluent at a flow
rate of 0.7 mL min−1. One unit was defined as the amount of enzyme that
released 1 µmol of product per minute.
Docking Analysis. The docking study was carried out using AutoDock
v4.2.[27] 2c, 2b and 7b molecules were generated using MacroModel
(Schrödinger) and the PRODRG2 Server.[28] See details for MacroModel
calculation in the Supporting Information. Chain A of the crystal structure
of GfsF was used for the docking study. Using AutoDockTools, polar
hydrogen atoms were added to amino acid residues, and Gasteiger
charges were assigned to all atoms of the protein. All rotatable bonds of
the alkyl side-chain moiety of the ligand molecule were set to be flexible
for the calculation, whereas all of the protein residues and the macrolide
ring of the ligand molecule were kept rigid. The cubic energy grid was
centered at the substrate binding pocket and had an extension of 50 Å in
each direction.
Detection of Formaldehyde. The GfsF assay mixture (50 µL each)
consisted of 0.1 mM 7c, 1 mM NADH, 3 µM GfsF, 8 µM CamA, and 20
µM CamB in 50 mM HEPES-Na (pH 7.7) containing 10% glycerol. The
enzymatic reaction was carried out at 28°C for 10 min–2 h. Then, 150 µL
of CH3CN was added to quench the reaction. After the addition of 4 µL of
20% phosphoric acid and 10 µL of 1 mg mL−1 2,4-dinitrophenylhydrazine,
the reaction mixture was incubated at 28°C for 20 min. The resulting
mixture was analyzed by HPLC using an ELITE LaChrom L-2455 DAD
Detector and L-2130 Pump equipped with a PEGASIL ODS column (100
Å, 250 × 4.6 mm). CH3CN (50%) in water was used as an eluent at a flow
rate of 1.0 mL min−1. The derivatized formaldehyde (formaldehyde 2,4-
dinitrophenylhydrazone) was detected at 360 nm.
Site-directed Mutagenesis. pET28-gfsF was used for construction of
GfsF mutants. Site-directed mutagenesis was carried out with the
following oligonucleotides and their complementary oligonucleotides:
F89A, 5'-CACATCAGCGCGGACGCCAAGTTCCTCAGC-3'; T300A, 5'-
GATCGCTCAGGACGCCGTGCGCCGGATCG-3';
T300V,
5'-
GATCGCTCAGGACGTCGTGCGCCGGATCG-3'. The mutations were
confirmed by determining the nucleotide sequences. The plasmids were
transformed into E. coli BL21(DE3) cells, and the mutated enzymes were
prepared as described above.
Isolation and Structural Determination of 3b. GfsF-containing reactant
was prepared by using E. coli BL21(DE3) cells harboring pgfsF-camAB
plasmid as described previously.[8] 3c (6.2 mg) were then mixed with the
GfsF-containing reactant. The reaction was carried out at 28 °C with
agitation (200 rpm) for 36 h. After the reaction, the reaction product was
extracted with EtOAc. The organic layers were dried on Na2SO4 and the
solvent was removed by evaporation. The crude residue was purified by
silica gel chromatography with hexane/EtOAc (1:1) to give 3b (1.9 mg).
HR-FAB-MS (positive mode): m/z calculated for C22H35O5: 379.2484
([M+H]+); found: 379.2512. The assignments of the 1H and 13C NMR
Acknowledgements
This work was performed with the approval of the Photon
Factory Program Advisory Committee (Proposal No. 2012G508).
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
[11]
T. Itagaki, A. Kawamata, M. Takeuchi, K. Hamada, Y. Iwabuchi, T.
Eguchi, F. Kudo, T. Usui, N. Kanoh, J. Antibiot. 2016, 69, 287–293.
T. Ishida, K. Kinoshita, Nucleic Acids Res. 2007, 35, W460–W464.
S. Y. Park, H. Shimizu, S. Adachi, A. Nakagawa, I. Tanaka, K.
Nakahara, H. Shoun, E. Obayashi, H. Nakamura, T. Iizuka, Y.
Shiro, Nat. Struct. Biol. 1997, 4, 827–832.
a) L. H. Xu, S. Fushinobu, H. Ikeda, T. Wakagi, H. Shoun, J.
Bacteriol. 2009, 191, 1211–1219; b) L. H. Xu, S. Fushinobu, S.
Takamatsu, T. Wakagi, H. Ikeda, H. Shoun, J. Biol. Chem. 2010,
285, 16844–16853.
a) L. H. Xu, H. Ikeda, L. Liu, T. Arakawa, T. Wakagi, H. Shoun, S.
Fushinobu, Appl. Microbiol. Biotechnol. 2015, 99, 3081–3091; b) R.
Satou, A. Miyanaga, H. Ozawa, N. Funa, Y. Katsuyama, K.
Miyazono, M. Tanokura, Y. Ohnishi, S. Horinouchi, J. Biol. Chem.
2013, 288, 34146–34157.
a) I. Schlichting, J. Berendzen, K. Chu, A. M. Stock, S. A. Maves,
D. E. Benson, R. M. Sweet, D. Ringe, G. A. Petsko, S. G. Sligar,
Science 2000, 287, 1615–1622; b) Nagano S, Poulos TL. J. Biol.
Chem. 2005, 280, 31659–31663.
a) A. D. Vaz, D. F. McGinnity, M. J. Coon, Proc. Natl. Acad. Sci.
USA 1998, 95, 3555–3560; b) E. T. Farinas, M. Alcalde, F. Arnold,
Tetrahedron 2004, 60, 525–528.
R. Gupta, X. X. Li, K. B. Cho, M. Guo, Y. M. Lee, Y. Wang, S.
Fukuzumi, W. Nam, J. Phys. Chem. Lett. 2017, 8, 1557–1561.
T. Omura, R. Sato, J. Biol. Chem. 1964, 239, 2379–2385.
Z. Otwinowski, W. Minor, Method Enzymol. 1997, 276, 307–326.
A. Vagin, A. Teplyakov, Acta Crystallogr. D Biol. Crystallogr. 2010,
66, 22–25.
Y. Yasutake, N. Imoto, Y. Fujii, T. Fujii, A. Arisawa, T. Tamura T,
Biochem. Biophys. Res. Commun. 2007, 361, 876–882.
R. J. Morris, A. Perrakis, V. S. Lamzin, Acta Crystallogr. D Biol.
Crystallogr. 2002, 58, 968–975.
P. Emsley, K. Cowtan, Acta Crystallogr. D Biol. Crystallogr. 2004,
60, 2126–2132.
This work was supported in part by Grants-in-Aid for Scientific
Research on Innovative Areas from the Ministry of Education,
Culture, Sports, Science and Technology (16H06451 to TE,
17H05434 to FK and 23102013 to NK) and JSPS A3 Foresight
Program (16822333), The Agricultural Chemical Research
Foundation to AM and The Naito Foundation to FK in Japan.
[12]
[13]
[14]
[15]
Keywords: Biosynthesis, Crystal structure, Cytochromes,
Multifunctional oxygenase, Polyketides
[1]
a) D. C. Lamb, M. R. Waterman, S. L. Kelly, F. P. Guengerich,
Curr. Opin. Biotechnol. 2007, 18, 504–512; b) F. P. Guengerich, J.
Biochem. Mol. Toxicol. 2007, 21, 163–168; c) S. C. Moody, E. J.
Loveridge, J. Appl. Microbiol. 2014, 117, 1549–1563.
a) L. M. Podust, D. H. Sherman, Nat. Prod. Rep. 2012, 29, 1251–
1266; b) J. D. Rudolf, C. Y. Chang, M. Ma, B. Shen, Nat. Prod.
Rep. 2017, doi: 10.1039/c7np00034k.
a) R. V. Cochrane, J. C. Vederas, Acc. Chem. Res. 2014, 47,
3148–3161; b) L. Hang, N. Liu, Y. Tang, ACS Catal. 2016, 6,
5935–5945.
[16]
[2]
[3]
[4]
[17]
[18]
[19]
[20]
[21]
S. Li, D. R. Tietz, F. U. Rutaganira, P. M. Kells, Y. Anzai, F. Kato,
T. C. Pochapsky, D. H. Sherman, L. M. Podust, J. Biol. Chem.
2012, 287, 37880–37890.
[5]
[6]
[7]
G. Zocher, M. E. Richter, U. Mueller, C. Hertweck, J. Am. Chem.
Soc. 2011, 133, 2292–2302.
[22]
[23]
[24]
[25]
[26]
J. C. Carlson, S. Li, S. S. Gunatilleke, Y. Anzai, D. A. Burr, L. M.
Podust, D. H. Sherman, Nat. Chem. 2011, 3, 628–633.
a) M. Seki-Asano, T. Okazaki, M. Yamagishi, N. Sakai, K. Hanada,
K. Mizoue, J. Antibiot. 1994, 47, 1226–1233; b) T. Eguchi, K.
Yamamoto, K. Mizoue, K. Kakinuma, J. Antibiot. 2004, 57, 156–
157; c) S. Inaba, T. Eguchi, A. Motegi, K. Mizoue, T. Usui, K.
Nagai, T. Kataoka, J. Antibiot. 2009, 62, 507–512.
G. N. Murshudov, A. A. Vagin, E. J. Dodson, Acta Crystallogr. D
Biol. Crystallogr. 1997, 53, 240–255.
S. C. Lovell, I. W. Davis, W. B. Arendall, 3rd, P. I. de Bakker, J. M.
Word, M. G. Prisant, J. S. Richardson, D. C. Richardson, Proteins
2003, 50, 437–450.
G. M. Morris, R. Huey, W. Lindstrom, M. F. Sanner, R. K. Belew, D.
S. Goodsell, A. J. Olson, J. Comput. Chem. 2009, 30, 2785–2791.
A. W. Schüttelkopf, D. M. F. van Aalten, Acta Crystallogr. D Biol.
Crystallogr. 2004, 60, 1355–1363.
[8]
[9]
F. Kudo, A. Motegi, K. Mizoue, T. Eguchi, Chembiochem 2010, 11,
1574–1582.
F. Kudo, K. Kawamura, T. Furuya, H. Yamanishi, A. Motegi, A.
Komatsubara, M. Numakura, A. Miyanaga, T. Eguchi,
Chembiochem 2016, 17, 233–238.
[27]
[28]
[10]
N. Kanoh, A. Kawamata, T. Itagaki, Y. Miyazaki, K. Yahata, E.
Kwon, Y. Iwabuchi, Org. Lett. 2014, 16, 5216–5219.
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.

10.1002/cbic.201700429
ChemBioChem
FULL PAPER
Table of Contents
FULL PAPER
Akimasa Miyanaga, Ryuichi Takayanagi,
Takashi Furuya, Ayano Kawamata,
Tomohiro Itagaki, Yoshiharu Iwabuchi,
Naoki Kanoh, Fumitaka Kudo and
Tadashi Eguchi*
GfsF
O
O
HO
HO
OH OH OH
OH OH OH
O
O
O
OH
FD-892
25-O-demethyl-FD-891
Page No. – Page No.
GfsF is a multifunctional P450 monooxygenase that catalyzes the epoxidation and
subsequent hydroxylation in the biosynthesis of FD-891. X-ray structural and
biochemical analyses of GfsF revealed two distinct substrate binding modes for
epoxidation and hydroxylation reactions, which explained how GfsF regulates the
order of two oxidative reactions.
Substrate recognition by a dual
functional P450 monooxygenase
GfsF involved in FD-891 biosynthesis
For internal use, please do not delete. Submitted_Manuscript
This article is protected by copyright. All rights reserved.