Biosynthesis of plant tetrahydroisoquinoline alkaloids through an imine reductase route

Article abstract of DOI:10.1039/c9sc03773j

Herein, we report a biocatalytic approach to synthesize plant tetrahydroisoquinoline alkaloids (THIQAs) from dihydroisoquinoline (DHIQ) precursors using imine reductases and N-methyltransferase (NMT). The imine reductase IR45 was engineered to significant

Full text of DOI:10.1039/c9sc03773j

Chemical
Science
View Article Online
View Journal | View Issue
EDGE ARTICLE
Biosynthesis of plant tetrahydroisoquinoline
alkaloids through an imine reductase route†
Cite this: Chem. Sci., 2020, 11, 364
ab
Lu Yang,ab Jinmei Zhu,b Chenghai Sun,b Zixin Denga and Xudong Qu
*
All publication charges for this article
have been paid for by the Royal Society
of Chemistry
Herein, we report a biocatalytic approach to synthesize plant tetrahydroisoquinoline alkaloids (THIQAs)
from dihydroisoquinoline (DHIQ) precursors using imine reductases and N-methyltransferase (NMT). The
imine reductase IR45 was engineered to significantly expand its substrate specificity, enabling efficient
and stereoselective conversion of 1-phenyl and 1-benzyl 6,7-dimethoxy-DHIQs into the corresponding
(S)-tetrahydroisoquinolines (S-THIQs). Coclaurine N-methyltransferase (CNMT) was able to further
efficiently convert these (S)-THIQ intermediates into (S)-THIQAs. By assembling IRED, CNMT, and
glucose dehydrogenase (GDH) in one reaction, we effectively constituted two artificial biosynthetic
pathways in Escherichia coli and successfully applied them to the production of five (S)-THIQAs. This
highly efficient (100% yield from DHIQs) and easily tailorable (adding other genes) biosynthetic approach
will be useful for producing a variety of plant THIQAs.
Received 30th July 2019
Accepted 17th November 2019
DOI: 10.1039/c9sc03773j
rsc.li/chemical-science
BIA biosynthesis, reconstruction of plant biosynthetic pathways
in microbial hosts has become a feasible strategy for producing
Introduction
BIAs. By incorporation of partial or complete biosynthetic path-
ways in microbial hosts, a number of BIAs (i.e. opiates, sangui-
narine and noscapine) have been successfully synthesized from
simple precursors or glucose.6 Such biosynthetic approaches can
become more economical and easy-to-operate than chemical
synthesis, making biosynthesis a highly attractive alternative.
Although great success has been achieved, there still remain
many challenges for microbial production of plant THIQAs:7 (i)
the biosynthesis of PIAs and many BIAs remain elusive, and
Tetrahydroisoquinoline alkaloids (THIQAs) are a class of
important bioactive natural products.1 Molecules in this family
typically contain benzyl (benzylisoquinoline alkaloids, BIAs) or
phenyl (phenylisoquinoline alkaloids, PIAs) groups at the C-1
position and show a remarkable spectrum of biological activity
(Scheme 1).1,2 Currently, more than a dozen of naturally occur-
ring or semisynthetic THIQAs have been used in the clinic for
treating a diverse range of diseases.1–3 As most THIQAs are
produced by plants, signicant efforts have been devoted in the
past few decades to develop synthetic approaches for their
preparation.1 However, the unsatisfactory enantioselectivity and
high cost of asymmetric catalysis make chemical approaches
still commercially unfeasible.1
In natural products, THIQ scaffolds are biosynthesized
through a Pictet–Spengler reaction.4 The major enzymes which
catalyze these reactions are norcoclaurine synthases (NCSs).
NCSs condense dopamine with 4-hydroxyphenylacetaldehyde (4-
HPAA) or 3,4-hydroxyphenylacetaldehyde (3,4-HPAA) to give rise
to (S)-norcoclaurine or (S)-norlaudanosoline.5 Following decora-
tions by methyltransferases (MTs) or P450s, 4-HPAA and 3,4-
HPAA can be converted into (S)-reticuline (Scheme 2), a key
intermediate of ꢀ2500 BIAs.5 With the growing understanding of
aState Key Laboratory of Microbial Metabolism, School of Life Sciences and
Biotechnology, Shanghai Jiao Tong University, Shanghai 200240, China. E-mail:
quxd@whu.edu.cn
bKey Laboratory of Combinatorial Biosynthesis and Drug Discovery Ministry of
Education, School of Pharmaceutical Sciences, Wuhan University, Wuhan 430071,
China
† Electronic supplementary information (ESI) available. See DOI:
10.1039/c9sc03773j
Scheme 1 Representative THIQAs.
364 | Chem. Sci., 2020, 11, 364–371
This journal is © The Royal Society of Chemistry 2020

View Article Online
Edge Article
Chemical Science
Scheme 2 Approaches for plant THIQA biosynthesis.
thus, knowledge of essential genes for reconstitution of their
biosynthetic pathways is lacking. (ii) Most valuable THIQAs are
highly modied, i.e. bearing methoxy groups at the C6, C7, C30
and C40 positions. (iii) Reconstitution of the biosynthetic
pathways in microbes requires the introduction of a large
numbers of genes, which brings a heavy metabolic burden on
the microbe. Moreover, many plant enzymes such as OMTs and
P450s are poorly expressed or show low reactivity (NCSs) in
microbial systems.6e,k,8 Therefore, production of plant THIQAs
in microbes oen results in very low yields.3 To avoid these
challenges, appropriately substituted precursors can be fed into
simplied biosynthetic pathways. However, such an approach is
restricted either by the narrow substrate scope of NCSs, whose
activity is strictly dependent on the C6-hydroxyl and C8a-meth-
ylene groups (Scheme 2)9 or the stereochemistry at C1, which is
challenging to prepare via practical chemical synthesis.1
Recently, our lab and another group identied imine
reductases (IREDs) that are capable of reducing sterically
hindered 1-phenyl dihydroisoquinolines (DHIQs) to yield the
corresponding chiral 1-phenyl-THIQs.10 In contrast to NCSs,
IREDs have a distinct catalytic mechanism, which does not rely
on the C6-hydroxyl or C8a-methylene group, and therefore can
accept highly modied DHIQ precursors. Moreover, IREDs from
bacterial sources can be well overexpressed and show good
activity in Escherichia coli. Thus, by combining IREDs with other
modication enzymes, it is potentially feasible to produce
THIQAs through feeding DHIQ precursors. Herein, we demon-
strate this concept by using IREDs, N-methyltransferases
(NMTs), and glucose dehydrogenase (GDH) to construct two
minimal biosynthetic pathways. By incorporating these arti-
cial pathways into E. coli, we successfully synthesize a group of
pharmaceutically valuable PIAs and BIAs. Our results demon-
strate that this strategy is an efficient approach to the biosyn-
thesis of plant THIQAs.
Results
Engineering the substrate specicity of IR45 for the synthesis
of 1-phenyl 6,7-dimethoxyl-THIQs
Cryptostyline I, II and III (1–3, Scheme 1) are three PIAs isolated
from the plant Cryptostylis fulva.11 They are used as pharmaco-
logical probes for understanding the pathophysiological roles
of peptides in the nervous system.12 Unlike BIAs, the biosyn-
thesis of PIAs is still poorly understood, and currently the
biosynthetic genes of 1–3 are not known. Given their valuable
pharmaceutical utility and representative roles for the PIAs,
these three products were chosen as the targets for this study.
Three 1-phenyl-6,7-dimethoxy-DHIQ precursors (1a–3a,
Scheme 3) corresponding to cryptostyline I–III (1–3) were ob-
tained in good yields (90–91%) by chemical synthesis (see the
ESI†). In a previous study, we identied an S-selective IRED,
IR45, which showed good tolerance toward 1-phenyl-DHIQs.10a
Therefore, this enzyme was selected for activity assays with the
new substrates. Encouragingly, IR45 effectively reduced 1a to
the
(S)-1-methylenedioxyphenyl-6,7-dimethoxy-THIQ
(1b,
Scheme 2) with 100% conversion and >99% ee (Table 1),
although it showed no activity toward the two bulkier substrates
(2a and 3a). To gain a more detailed understanding of its
substrate specicity, the structure of IR45 was modelled based
on the IRED Q1EQE0 (64% identity, PDB: 3ZHB).13 Docking of
1a into the binding pocket of IR45 indicated that the substrate
is closely surrounded by the conserved aspartic acid (D183) and
NADPH (Fig. 1). The imine bond is positioned close to the si-
face of NADPH, consistent with the observed S-selectivity in the
product. The methylenedioxyphenyl group points to the right
cle, which consists of residues from two subunits of the
enzyme (V81, S107, E2530 and H2570). On the other side of the
active site, the 6,7-dimethoxy aryl group is surrounded by two
residues, W191 and F190. These two bulky residues form the
This journal is © The Royal Society of Chemistry 2020
Chem. Sci., 2020, 11, 364–371 | 365

View Article Online
Chemical Science
Edge Article
Scheme 3 DHIQ substrates prepared and used in this study.
Table 1 Conversion and enantioselectivity of IR45 and its mutants
toward 1a–5a
Conversion (%) and enantiomeric excess (ee)a (%)
Subs.
IR45
W191F
F190L–W191F
F190M–W191F
1a
2a
3a
4a
5a
100; >99S
0
—
100; >99S
75; >99S
100; >99S
0
3; —
2; —
—
0
100; >99S
100; >99S
100; >99S
0
100; >99S
100; >99S
100; >99S
100; >99S
—
a
First value shown represents conversion and the second value
represents the ee value. Stereospecicity is indicated by the
superscripts. Dash indicates data not tested. All reactions were
performed at 30 ꢁC for 24 h.
bottom of the binding pocket, with F190 being the major
contributor to the pocket's overall shape.
From the model, it can be seen that the right cle forms the
major constraint for the accommodation of the methyl-
enedioxyphenyl group (Fig. 1). Engineering this area, however,
may signicantly impair the enzyme's activity and stability, as
residues in this region are crucial for inter-subunit interactions,
i.e. H2570 forms a stabilizing salt bridge with R112. Thus, we
envisioned enlarging the binding pocket by engineering the
side of the pocket comprising F190 and W191. By mutating
these sterically demanding residues, we thought that the
substrate could be moved slightly deeper into the le cavity,
thus potentially relieving the steric hindrance between the right
cle and substrates 2a and 3a.
In our previous studies, we found that W191 plays important
roles in substrate binding and maintaining the enzyme's cata-
lytically active conformation. Mutation of W191 to alanine or
leucine reduces the kcat and Km for the substrate 1-phenyl-DHIQ
(6a, Scheme 3).10a To identify optimal mutations, W191 was
further mutated to the remaining 17 amino acids. These vari-
ants together with the W191A and W191L were assayed for the
reduction of 2a and 3a. Although most mutants were inactive,
one mutant in particular, W191F, showed marginal but clear
activity toward these substrates (3% and 2% for 2a and 3a
respectively, Table 1 and Fig. S1A†). To unveil the detailed
effects of these mutations on the IRED's activity, we used the
more active substrate 6a for activity analysis. Similarly, W191F
shows a clear improvement in the conversion of 1-phenyl-DHIQ
(1.5-fold), while the others all drastically reduce or even abolish
the catalytic activity (Fig. S2†).
Fig. 1 In silico model of IR45 with substrates 1a and 4a. The structure
of IR45 is modelled based on Q1EQE0 (PDB: 3zhb) and docked with 1a
or 4a (grey) and NADPH (pink); critical residues in the binding cavity are
labelled. Protein backbones and residues from different subunits are
indicated by yellowish brown and green colours. (A) IR45 with 1a; (B)
IR45 with 4a.
Kinetic analysis of the most active mutants, including
W191F, W191C, W191M, W191H, W191N, and W191S as well as
previously obtained W191A and W191L revealed that their kcat
and Km values both have been reduced (Table S1 and Fig. S3A†).
However, unlike the other enzymes whose kcat is signicantly
reduced by one to two orders of magnitude, the kcat value of
W191F is decreased by only 1.8-fold (Table S1†), suggesting that
W191F has the least negative impact on the enzyme's catalytic
366 | Chem. Sci., 2020, 11, 364–371
This journal is © The Royal Society of Chemistry 2020

View Article Online
Edge Article
Chemical Science
conformation. As W191 is highly conserved, we assume that this activities of O-MTs result in low yields of (R,S)-laudanosine, and
mutation should be benecial for many other IREDs. the accumulation of many methylated intermediates (e.g. (R,S)-
The encouraging results with W191F inspired us to further reticuline, (R,S)-codamine and (R,S)-laudanine) complicates its
engineer the enzyme by targeting the adjacent residue F190. We isolation (Fig. S4†). We chose to test our IRED approach in the
mutated all 19 other residues on top of W191F and then assayed synthesis of 5 given its pharmaceutical value and (S)-1-benzyl-
these double mutants for activity on the bulkier DHIQ 6,7-dimethoxyl-N-methyl-THIQ (BzTHIQ, 4 in Scheme 1) given
substrates. To our gratication, one mutant F190L–W191F its prototypical structure.
shows drastically improved activity toward both 2a and 3a (75%
Two corresponding DHIQ substrates 4a and 5a were chem-
and 100% conversion within 24 hours respectively) with very ically synthesized (yields 95 and 93%) following a procedure
high stereoselectivity for the S-products 2b and 3b (Scheme 2 similar to that used in the synthesis of 1a–3a. As W191 and F190
and Table 1). Kinetic analysis conrmed that both 2a and 3a are are critical for substrate binding, wild-type IR45 along with the
good substrates for F190L–W191F (kcat/Km values are 0.011 and library of F190X–W191F (X denotes other 19 amino acids)
0.087 sꢂ1 mMꢂ1, respectively, Table 2). The identical Km values variants was tested for activity. To our gratication, wild-type
(0.307 mM for 2a and 0.372 mM for 3a) yet different kcat values IR45 effectively converted 4a and 5a into the corresponding (S)-
(0.003 sꢂ1 for 2a and 0.032 sꢂ1 for 3a) suggest that binding THIQ products (100% conversion and >99% ee). The best
conformation accounts for the difference in their activity. This mutant F190M–W191F showed a further 6.8- and 1.8-fold
indicates that other residues inuencing the binding confor- improvement in catalytic activity for 4a and 5a, respectively (kcat
mation in the cavity could be engineered to further improve the Km values for 4a and 5a are 0.989 and 0.306 sꢂ1 mMꢂ1, Table 2).
enzyme's catalytic activity. It is intriguing that the wild-type IR45 can accept 4a and 5a
/
Through protein engineering, we have successfully broad- whose planar frameworks are larger than those of the inactive
ened the substrate specicity and enabled IR45 to effectively substrates 2a and 3a. To explore how differences in binding in
convert very sterically hindered DHIQ precursors into their the enzyme cavity may account for this disparity in activity, we
corresponding THIQs. In addition, the successful expansion of docked 4a into IR45 (Fig. 1B). In silico analysis revealed that the
the substrate specicity of IR45 demonstrates that alteration of DHIQ plane of 4a takes an identical conformation to that of 1a.
the sites that do not directly repel the substrate is indeed an On the other hand, the benzyl group of 4a is bent toward the
effective way to optimize an enzyme's substrate specicity. This large cavity underneath the right cle unlike the 1-phenyl group
strategy will be useful for engineering other proteins, in which of 1a which points directly towards the right cle. This differ-
certain substrate–protein interactions are critical for main- ence in positioning may result in less steric hindrance for the
taining catalytic activity or active conformation.
otherwise ‘larger substrate’ 4a. This indicates that the rigid
frameworks of 1-phenyl DHIQs are indeed the most challenging
substrates for IREDs. Fortunately, our engineered mutant
F190L–W191F is able to effectively convert the very sterically
hindered substrates 2a and 3a. In addition to the elevated
activity on 4a and 5a, F190M–W191F also shows a 2.23- and
3.79-fold improvement (kcat/Km) on 1a compared to wild-type
IR45 and F190L–W191F, respectively (Table 2). This
outstanding activity makes the imine reduction step highly
efficient, ensuring the success of this strategy.
Employing IR45 mutants to synthesize 1-benzyl 6,7-
dimethoxyl-THIQs
Because BIAs make up the remainder of the THIQA family, we
were interested in applying IR45 to their synthesis as well. (S)-
Laudanosine (5, Scheme 1) is a pharmaceutically important BIA,
naturally found in opium in minute amounts (0.1%).14 Its
semisynthetic derivative atracurium is medically used for skel-
etal muscle relaxation during surgery and mechanical ventila-
tion.3 By mild chemical oxidation, (S)-laudanosine can be
readily converted into the antitussive drug (S)-glaucine.15
Conversion of THIQs into THIQAs by CNMT
Production of racemic (R,S)-laudanosine in E. coli has been Aer overcoming the bottlenecks associated with imine reduc-
previously achieved through feeding (R,S)-norlaudanosoline to tion, we turned our attention to the N-methylation step.
a strain expressing O-MTs and NMTs.16 However, the poor Coclaurine N-methyltransferase (CNMT) is a key enzyme in the
Table 2 Kinetics parameters of IR45 and its mutants for the conversion of 1a–5aa
IR45
F190L–W191F
Km [mM]
F190M–W191F
Km [mM]
kcat/Km
kcat/Km
kcat/Km
Subs.
Km [mM]
kcat [sꢂ1
]
[sꢂ1 mMꢂ1
]
kcat [sꢂ1
]
[sꢂ1 mMꢂ1
]
kcat [sꢂ1
]
[sꢂ1 mMꢂ1
]
1a
2a
3a
4a
5a
0.154
NA
NA
0.148
0.518
0.020
NA
NA
0.021
0.090
0.129
NA
NA
0.146
0.174
0.123
0.307
0.372
0.252
0.353
0.009
0.003
0.032
0.049
0.041
0.076
0.011
0.087
0.196
0.117
0.108
NA
NA
0.251
0.178
0.031
NA
NA
0.248
0.054
0.288
NA
NA
0.989
0.306
a
NA indicates no activity.
This journal is © The Royal Society of Chemistry 2020
Chem. Sci., 2020, 11, 364–371 | 367

View Article Online
Chemical Science
Edge Article
pathway of (S)-reticulene, introducing the N-methyl substituent respectively, aer 3 days. We found that temperature also had
into coclaurine.17 CNMT from Coptis japonica is known to have a signicant impact on the conversion. The biotransformation
ꢁ
ꢁ
a relaxed substrate specicity and is able to methylate THIQ efficiency at 30 C is 5 and 10 times higher than that at 25 C
products.17,18 Therefore, it was assayed for N-methylation and 37 C, respectively (aer 3 days) (Fig. S8B†). The optimal
ꢁ
activity on THIQ substrates 1b–5b. To improve expression, the conditions for the IRED–GDH combination appear to be 20 mM
CNMT gene was codon optimized for E. coli. As expected, IPTG and 30 ꢁC. Next the conversion of the other four substrates
following optimization, CNMT was robustly expressed in E. coli. was evaluated under these optimized conditions (F190M–
To our gratication, CNMT efficiently converted all ve of the W191F was replaced by F190L–W191F for 2a and 3a). Although
S-THIQ substrates into their corresponding N-methylated 4a and 5a can be efficiently converted into their reduced prod-
products (100% conversion, Fig. S5A†). Kinetic analysis revealed ucts (100%), 1a–3a exhibit poor-to-medium conversion (6.5% to
that CNMT showed similar activity to all 1-phenyl and 1-benzyl 50%) (Fig. S1C†). These ratios are much lower than the
substrates except for the most active substrate 4b (Table S2†). As conversion ratios of the in vitro enzymatic transformation,
these substrates are much bulkier than previously known suggesting that the 1-phenyl substrates 1a–3a face difficulty in
substrates of CNMT, we were interested in understanding how permeating the cell membrane.
they bind to the active site. Two representative substrates 1b
Next, the N-methyl transfer step was evaluated. Gratifyingly, all
and 5b were individually docked into the crystal structure of the amine substrates, 1b–5b, are completely converted into their
CNMT (Fig. S6†).18 Similar to the conformation of 4a in IR45, 5b corresponding nal products under the same optimal conditions
is bent, allowing it to be well accommodated in the binding as for IRED–GDH by E. coli containing CNMT (Fig. S5B†). As
cavity. The substrates are bound in an identical orientation with methylation by CNMT appeared to improve the activity of imine
their methylenedioxy and methoxy groups in close proximity to reduction, we co-expressed all three enzymes in a single cell (NMT
residues F322 and W329.18 Previous work showed that two and GDH were cloned in pACYCDuet and IR45 in pET28a),
mutants, F322A and W329A, had signicantly reduced catalytic forming a minimal biosynthetic pathway for the synthesis of
activity, with W329A having the most negative impact.18 In order THIQAs. As expected 1a, 4a and 5a (50 mg Lꢂ1) can be completely
to maintain these hydrophobic interactions while enlarging the converted into their corresponding alkaloids within one to three
binding cavity, we mutated W329 into phenylalanine. However, days (Table 3 and Fig. 2). By extraction with ethyl acetate, pure 1, 4
like W329A, W329F showed drastically reduced methylation and 5 can be readily obtained in yields of 95%, 98% and 95%,
activity toward 1–5 (Fig. S7A†), conrming that W329 is indeed respectively. Conversion of 2a and 3a also signicantly improved
critical for maintaining the enzyme's activity.
(89% and 51% respectively, Fig. S7C†), compared to the conver-
sion by IRED–GDH (6.5% and 9.3%, Fig. S1C†).
In order to overcome the barrier of membrane per-
meabilization and further improve the conversion of 2a and 3a,
we combined enzymatic and whole-cell biotransformation into
Production of THIQAs from DHIQ precursors through the
biosynthetic pathway of IRED–NMT–GDH
Armed with efficient imine reduction and N-methylation a single pot reaction. F190L–W191F and GDH were rst over-
enzymes, we proceeded to combine these two steps into one expressed in a single E. coli strain and then lysed by sonication.
biocatalytic reaction. The puried F190L–W191F and F190M– The crude enzymes were directly added to the E. coli–CNMT
W191F were coupled with CNMT and glucose dehydrogenase culture when the OD600 reached 0.7. At this time, IPTG, DHIQ
(GDH, for recycling NADPH from glucose) for the synthesis of 2– substrates and NADP (1.1& w/v) were added simultaneously to
3 and 1, 4–5 respectively. As expected, these enzyme cascades initiate the CNMT expression and cascade biotransformation.
can efficiently convert 1a–5a into THIQAs with 100% conversion To our delight, both 2a and 3a (50 mg Lꢂ1) were completely
(within 24 h, Fig. S7B†). Moreover, we observed that N-methyl- converted into their corresponding THIQAs (Table 3 and Fig. 2).
ation improved the conversion of substrates 2a and 3a into their The isolated yields of 2 and 3 are 93% and 96%, respectively.
corresponding THIQAs. We hypothesize that the continuous Therefore, by systematically improving enzyme activity and
removal of THIQ intermediates by their irreversible methylation optimizing biocatalytic conditions, we successfully imple-
drives the equilibrium of imine reduction from DHIQs to mented the IRED-based route to biosynthesize two types of
THIQs.
With the encouraging results of N-methylation, we turned to
plant THIQAs.
constitution of the articial biosynthetic pathway in E. coli. To
achieve the highest transformation efficiency, each step was
evaluated individually in the E. coli system. We rst evaluated
the effect of temperature and IPTG concentration on the IRED
reaction. E. coli stains containing F190M–W191F and GDH
(cloned into pET28a and pACYCDuet, respectively) were incu-
bated at 25, 30 or 37 ꢁC and supplemented with 1a and different
concentrations of IPTG when their OD600 reached 0.7. The
biotransformation results show that lower IPTG dosage (20 mM)
is better than a high IPTG dosage (Fig. S8A†) with conversion
reaching only 70% and 40% with 50 mM and 100 mM IPTG,
Table 3 Biosynthesis of 1–5 from precursors 1a–5a through the
IRED-based pathways
Subs.
Enzymes
Time (days)
Conv. (%)
1a
2a
3a
4a
5a
F190M–W191F + GDH + CNMT
F190L–W191F + GDH + CNMT
F190L–W191F + GDH + CNMT
F190M–W191F + GDH + CNMT
F190M–W191F + GDH + CNMT
3
3
3
1
3
100a
100b
100b
100a
100a
a
b
Whole cell biosynthesis. Enzyme-cell biosynthesis.
368 | Chem. Sci., 2020, 11, 364–371
This journal is © The Royal Society of Chemistry 2020

View Article Online
Edge Article
Chemical Science
our previous study.10a Due to their more rigid conformations, 1-
phenyl 6,7-dimethoxyl-DHIQ substrates are more sterically
hindered than the 1-benzyl 6,7-dimethoxyl-DHIQs, and wild-
type IR45 showed no activity on the most sterically hindered
substrates 2a and 3a. To address this issue, we engineered IR45
to signicantly expand its substrate scope. Instead of engi-
neering the site of direct steric repulsion, we chose to alter the
other side of the binding pocket. This strategy was very effective
for IR45 and may prove to be useful for engineering other
enzymes where certain protein interactions nearby the
substrate are critical for maintaining catalytic activity or
conformation.
The resultant mutant W191F–F190 efficiently and stereo-
selectively (ee > 99%) converted the very sterically hindered
substrates (2a and 3a) and the other three substrates (1a, 4a and
5a) into THIQs. The other mutant F190M–W191F showed
signicantly improved activity toward 1a, 4a and 5a. As these
substrates are more sterically hindered than most other natural
PIAs and BIAs,1,2 these IR45 mutants are potentially useful for
the synthesis of many other THIQAs. Practical application of
these enzymes can be backed up by the recently designed
strategy for scaling up IRED transformation.19 In addition to the
IRED, we also found that the CNMT has a sufficiently relaxed
substrate specicity to accept very bulky THIQ substrates.
Therefore, the broad substrate specicity of both the IR45
mutants and CNMT has provided a solid basis for the applica-
tion of this IRED-based biosynthetic strategy.
Although most natural THIQAs contain S-conguration at
the C1 position, it is still interesting to generate R-type THIQAs
using this strategy. However, our attempt to produce R-1b–5b
using the most broad-selective R-type imine reductase IR2 (ref.
10a) was not successful. Like IR45, IR2 shows S-specicity to
convert 1a, 4a and 5a into the corresponding S-THIQs (2a and 3a
are inactive, data not shown). The reverse stereo-specicity
might be due to the occurrence of the 6,7- and 30,40-substituent
groups in the THIQ scaffold. Currently, protein engineering of
IR2 is in progress to reverse its stereo-specicity.
Finally, to accommodate the different physical properties of
PIAs and BIAs, we developed two types of biocatalytic
approaches for their synthesis. BIAs can be biosynthesized
(100% yield) from their DHIQ precursors using the recombinant
E. coli strain expressing the articial biosynthetic pathway. Due
to the poor membrane-permeability of their DHIQ precursors,
PIAs 2 and 3 were not produced by a traditional whole-cell
system in sufficiently high yields. To overcome this challenge
and avoid using the expensive S-adenosylmethionine (SAM)
cofactor in the N-methylation step, we conceived a novel bio-
catalytic approach combining the enzymatic transformation of
IRED–GDH and the whole-cell transformation of CNMT in one
pot. This chimeric system is as efficient as the cell-based system
(produces PIAs in 100% yield) and will be useful for biosyn-
thesis of other products with similar problems.
Fig. 2 Production of THIQAs 1–5 in the IRED-based biosynthetic
systems. The retention time of the precursors 1a–5a is indicated by the
dashed lines. Samples from the biosynthetic systems are indicated by
the solid lines in dark blue. (I) Biosynthesis of 1 from 1a by E. coli with
F190M–W191F + GDH + CNMT; (II) biosynthesis of 2 from 2a by the
crude enzymes F190L–W191F + GDH and E. coli with CNMT; (III)
biosynthesis of 3 from 3a by the crude enzymes F190L–W191F + GDH
and E. coli with CNMT; (IV) biosynthesis of 4 from 4a by E. coli with
F190M–W191F + GDH + CNMT; (V) biosynthesis of 5 from 5a by E. coli
with F190M–W191F + GDH + CNMT.
Discussion
Lack of knowledge of key biosynthetic genes and poor expres-
sion/low reactivity of plant biosynthetic enzymes make recon-
stitution of THIQA biosynthetic pathways in microbes very
challenging. Currently, the biosynthetic enzymes, including the
key Pictet–Spengler synthase, that assemble PIAs remain
unknown. As such, reconstitution of PIA biosynthesis in
microbes is yet to be realized. Although the biosynthesis of
many BIAs is well understood and elegant examples of their
production in microbes have been reported, the poor perfor-
mance of NCS O-MT and P450 enzymes results in overall low
yields.3 Efforts to bypass such bottleneck enzymes by employing
modied precursors are also limited by the narrow substrate
scope of NCSs.9 IREDs, owing to their distinct catalytic mecha-
nism, are not restricted by the same limitations as NCSs.
Therefore, bottleneck enzymes such as O-MTs and P450
hydroxylases can be omitted and replaced by readily available
synthetic DHIQ precursors. In conjunction with NMT and/or
additional other modication enzymes, this IRED-based
approach has been shown to be very powerful in the biosyn-
thesis of plant THIQAs.
Conclusions
The major obstacle for application of this approach is the
generally poor tolerance of IREDs to bulky substituents at C1 of
DHIQs. IR45 is a steric-hindrance tolerating IRED identied in
We have developed an IRED-based strategy to successfully bio-
synthesize ve pharmaceutically valuable PIAs and BIAs.
This journal is © The Royal Society of Chemistry 2020
Chem. Sci., 2020, 11, 364–371 | 369

View Article Online
Chemical Science
Edge Article
Notably, the three PIAs' natural biosynthetic pathways have not
been elucidated. Through protein engineering, we signicantly
expanded the substrate specicity of the IRED IR45. The two
resultant mutants F190L–W191F and F190M–W191F can effi-
ciently convert bulky 1-aryl-6,7-dimethoxy-DHIQ substrates into
(S)-THIQs. These two IREDs, highly tolerant of steric hindrance,
will also be useful for the synthesis of many natural and
synthetic THIQs. The N-methylation enzyme (CNMT) was also
able to convert highly sterically hindered substrates. We also
developed an efficient and cost-effective enzyme/whole-cell
chimeric biosynthetic system to overcome the barrier for
biotransformation of membrane-impermeable substrates. Our
work demonstrates that this IRED-based biosynthetic approach
is efficient (100% yield) and exible for the production of plant
THIQAs. By the addition of other modication enzymes, e.g.
P450 enzymes, this minimal biosynthetic pathway can be
further extended to the biosynthesis of other complex THIQAs.
3283; (f) S. Galanie, K. Thodey, I. J. Trenchard, M. Filsinger
Interrante and C. D. Smolke, Science, 2015, 349, 1095–1100;
(g) W. C. DeLoache, Z. N. Russ, L. Narcross,
A. M. Gonzales, V. J. J. Martin and J. E. Dueber, Nat. Chem.
Biol., 2015, 11, 465; (h) E. Fossati, L. Narcross, A. Ekins,
J.-P. Falgueyret and V. J. J. Martin, PLoS One, 2015, 10,
e0124459; (i) Y. Li and C. D. Smolke, Nat. Commun., 2016,
7, 12137; (j) A. Nakagawa, E. Matsumura, T. Koyanagi,
T. Katayama, N. Kawano, K. Yoshimatsu, K. Yamamoto,
H. Kumagai, F. Sato and H. Minami, Nat. Commun., 2016,
7, 10390; (k) L. Narcross, L. Bourgeois, E. Fossati, E. Burton
and V. J. J. Martin, ACS Synth. Biol., 2016, 5, 1505–1518; (l)
Y. Li, S. Li, K. Thodey, I. Trenchard, A. Cravens and
C. D. Smolke, Proc. Natl. Acad. Sci. U. S. A., 2018, 115,
E3922–E3931; (m) C. J. Vavricka, T. Yoshida, Y. Kuriya,
S. Takahashi, T. Ogawa, F. Ono, K. Agari, H. Kiyota, J. Li,
J. Ishii, K. Tsuge, H. Minami, M. Araki, T. Hasunuma and
A. Kondo, Nat. Commun., 2019, 10, 2015.
7 S. Li, Y. Li and C. D. Smolke, Nat. Chem., 2018, 10, 395–404.
8 (a) F. Yesilirmak and Z. Sayers, Int. J. Plant Genomics, 2009,
2009, 296482; (b) I. J. Trenchard, M. S. Siddiqui, K. Thodey
and C. D. Smolke, Metab. Eng., 2015, 31, 74–83.
Conflicts of interest
The authors declare no competing nancial interest.
¨
9 (a) B. M. Ruff, S. Brase and S. E. O'Connor, Tetrahedron Lett.,
Acknowledgements
2012, 53, 1071–1074; (b) T. Pesnot, M. C. Gershater,
J. M. Ward and H. C. Hailes, Adv. Synth. Catal., 2012, 354,
2997–3008; (c) M. Nishihachijo, Y. Hirai, S. Kawano,
A. Nishiyama, H. Minami, T. Katayama, Y. Yasohara,
F. Sato and H. Kumagai, Biosci., Biotechnol., Biochem., 2014,
78, 701–707; (d) B. R. Lichman, M. C. Gershater,
E. D. Lamming, T. Pesnot, A. Sula, N. H. Keep, H. C. Hailes
and J. M. Ward, FEBS J., 2015, 282, 1137–1151; (e)
A. Bonamore, L. Calisti, A. Calcaterra, O. H. Ismail,
M. Gargano, I. D'Acquarica, B. Botta, A. Boffi and
A. Macone, ChemistrySelect, 2016, 1, 1525–1528; (f)
B. R. Lichman, J. Zhao, H. C. Hailes and J. M. Ward, Nat.
Commun., 2017, 8, 14883; (g) V. Erdmann, B. R. Lichman,
J. Zhao, R. C. Simon, W. Kroutil, J. M. Ward, H. C. Hailes
and D. Rother, Angew. Chem., Int. Ed., 2017, 56, 12503–
12507; (h) J. Zhao, B. R. Lichman, J. M. Ward and
H. C. Hailes, Chem. Commun., 2018, 54, 1323–1326; (i)
L. Y. P. Luk, S. Bunn, D. K. Liscombe, P. J. Facchini and
M. E. Tanner, Biochemistry, 2007, 46, 10153–10161; (j)
A. Ilari, S. Franceschini, A. Bonamore, F. Arenghi, B. Botta,
A. Macone, A. Pasquo, L. Bellucci and A. Boffi, J. Biol.
Chem., 2009, 284, 897–904; (k) Y. Wang, N. Tappertzhofen,
D. Mendez-Sanchez, M. Bawn, B. Lyu, J. M. Ward and
H. C. Hailes, Angew. Chem., Int. Ed., 2019, 58, 10120–10125.
The authors are grateful to Prof. Wenbo Liu for the help in
chiral analysis. This work was supported by the National Key
R&D Program of China (2018YFC1706200) and NSFC grant
(31800048, 31570057, and 31770063).
Notes and references
1 (a) W. Liu, S. Liu, R. Jin, H. Guo and J. Zhao, Org. Chem.
Front., 2015, 2, 288–299; (b) M. Chrzanowska, A. Grajewska
and M. D. Rozwadowska, Chem. Rev., 2016, 116, 12369–
12465.
2 (a) M. Perez, Z. Wu, M. Scalone, T. Ayad and
V. Ratovelomanana-Vidal, Eur. J. Org. Chem., 2015, 6503–
6514; (b) I. P. Singh and P. Shah, Expert Opin. Ther. Pat.,
2017, 27, 17–36.
3 L. Narcross, E. Fossati, L. Bourgeois, J. E. Dueber and
V. J. J. Martin, Trends Biotechnol., 2016, 34, 228–241.
4 (a) R. Stadler, T. M. Kutchan and M. H. Zenk, Phytochemistry,
1989, 28, 1083–1086; (b) W. De-Eknamkul, N. Suttipanta and
T. M. Kutchan, Phytochemistry, 2000, 55, 177–181.
5 (a) N. Samanani, D. K. Liscombe and P. J. Facchini, Plant J.,
2004, 40, 302–313; (b) H. Minami, E. Dubouzet, K. Iwasa and
F. Sato, J. Biol. Chem., 2007, 282, 6274–6282.
6 (a) H. Minami, J.-S. Kim, N. Ikezawa, T. Takemura, 10 (a) J. Zhu, H. Tan, L. Yang, Z. Dai, L. Zhu, H. Ma, Z. Deng,
T. Katayama, H. Kumagai and F. Sato, Proc. Natl. Acad. Sci.
U. S. A., 2008, 105, 7393–7398; (b) K. M. Hawkins and
C. D. Smolke, Nat. Chem. Biol., 2008, 4, 564; (c)
Z. Tian and X. Qu, ACS Catal., 2017, 7, 7003–7007; (b)
H. Li, P. Tian, J.-H. Xu and G.-W. Zheng, Org. Lett., 2017,
19, 3151–3154.
A. Nakagawa, H. Minami, J.-S. Kim, T. Koyanagi, 11 (a) A. Brossi and S. Teitel, Helv. Chim. Acta, 1971, 54, 1564–
¨
T. Katayama, F. Sato and H. Kumagai, Nat. Commun., 2011,
2, 326; (d) K. Thodey, S. Galanie and C. D. Smolke, Nat.
1571; (b) K. Leander, B. Luning and E. Ruusa, Acta Chem.
Scand., 1969, 23, 244–248.
Chem. Biol., 2014, 10, 837; (e) E. Fossati, A. Ekins, 12 D. L. Minor, S. D. Wyrick, P. S. Charifson, V. J. Watts,
L. Narcross, Y. Zhu, J.-P. Falgueyret, G. A. W. Beaudoin,
P. J. Facchini and V. J. J. Martin, Nat. Commun., 2014, 5,
D. E. Nichols and R. B. Mailman, J. Med. Chem., 1994, 37,
4317–4328.
370 | Chem. Sci., 2020, 11, 364–371
This journal is © The Royal Society of Chemistry 2020

View Article Online
Edge Article
Chemical Science
´
13 M. Rodrıguez-Mata, A. Frank, E. Wells, F. Leipold, 17 (a) K.-B. Choi, T. Morishige, N. Shitan, K. Yazaki and F. Sato,
N. J. Turner, S. Hart, J. P. Turkenburg and G. Grogan,
ChemBioChem, 2013, 14, 1372–1379.
J. Biol. Chem., 2002, 277, 830–835; (b) K.-B. Choi,
T. Morishige and F. Sato, Phytochemistry, 2001, 56, 649–655.
14 D. Mujahidin and S. Doye, Eur. J. Org. Chem., 2005, 2005, 18 M. R. Bennett, M. L. Thompson, S. A. Shepherd,
2689–2693.
M. S. Dunstan, A. J. Herbert, D. R. M. Smith, V. A. Cronin,
B. R. K. Menon, C. Levy and J. Mickleeld, Angew. Chem.,
Int. Ed., 2018, 57, 10600–10604.
´
15 E. Anakabe, L. Carrillo, D. Badıa, J. L. Vicario and
M. Villegas, Synthesis, 2004, 2004, 1093–1101.
16 L. Chang, J. M. Hagel and P. J. Facchini, Plant Physiol., 2015, 19 A. Bornadel, S. Bisagni, A. Pushpanath, S. L. Montgomery,
169, 1127–1140.
N. J. Turner and B. Dominguez, Org. Process Res. Dev.,
2019, 236, 1262–1268.
This journal is © The Royal Society of Chemistry 2020
Chem. Sci., 2020, 11, 364–371 | 371