of 17
A general tool for engineering the NAD/NADP cofactor
preference of oxidoreductases
Jackson KB Cahn
†,§
,
Caroline A Werlang
†,
,
Armin Baumschlager
†,
,
Sabine Brinkmann­
Chen
,
Stephen L Mayo
,
Frances H Arnold
Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena,
CA, 91125.
Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA,
91125.
§
Current Address: Institute for Microbiology, ETH Zürich, CH-8093 Zürich. Switzerland.
Current Address: Institute of Bioengineering, EPFL, CH-1015 Lausanne, Switzerland.
Current Address: Department of Biosystems Science and Engineering, ETH Zürich, CH-4058
Basel, Switzerland.
Abstract
The ability to control enzymatic nicotinamide cofactor utilization is critical for engineering
efficient metabolic pathways. However, the complex interactions that determine cofactor-binding
preference render this engineering particularly challenging. Physics-based models have been
insufficiently accurate and blind directed evolution methods too inefficient to be widely adopted.
Building on a comprehensive survey of previous studies and our own prior engineering successes,
we present a structure-guided, semi-rational strategy for reversing enzymatic nicotinamide
cofactor specificity. This heuristic-based approach leverages the diversity and sensitivity of
catalytically productive cofactor binding geometries to limit the problem to an experimentally
tractable scale. We demonstrate the efficacy of this strategy by inverting the cofactor specificity
of four structurally diverse NADP-dependent enzymes: glyoxylate reductase, cinnamyl alcohol
dehydrogenase, xylose reductase, and iron-containing alcohol dehydrogenase. The analytical
components of this approach have been fully automated and are available in the form of an
easy-to-use web tool: Cofactor Specificity Reversal – Structural Analysis and LibrAry Design
(CSR-SALAD).
AUTHOR CONTRIBUTIONS
This project was conceived by J.K.B.C., S.B-C., S.L.M., and F.H.A. J.K.B.C. designed and programmed CSR-SALAD, which was
implemented online by C.A.W. J.K.B.C. performed the validation experiments with assistance from A.B., and produced the figures
and tables. J.K.B.C., S.B-C., and F.H.A. wrote the manuscript with input from all authors.
SUPPORTING INFORMATION
The Supporting Information is available free of charge on the ACS Publications website at DOI [TBD].
• Previous cofactor specificity reversals in the literature; residue distributions of natural NAD-and NADP-bound structures; examples
of structural classifications; CSR-SALAD interface examples; detailed kinetic parameters; wild-type enzyme structures and modeled
mutants; detailed results of literature mutant recapitulation
• CSR-SALAD Users’ Manual Version 1.1; additional references; designed libraries for proteins described in this work.
HHS Public Access
Author manuscript
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Published in final edited form as:
ACS Synth Biol
. 2017 February 17; 6(2): 326–333. doi:10.1021/acssynbio.6b00188.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Keywords
cofactor specificity; oxidoreductases; protein engineering; library design; semi-rational
engineering
The vast majority of oxidoreductases, which make up the largest group of enzymes in the
Enzyme Commission nomenclature, use the functionally-equivalent cofactors nicotinamide
adenine dinucleotide (NAD) or nicotinamide adenine dinucleotide phosphate (NADP)
for the transport and storage of electrons in the form of hydride groups (Figure 1).
The phosphate group that distinguishes these cofactors is sufficiently distant spatially
and covalently from the chemically-active nicotinamide moiety that it plays no role in
the chemistry. Nonetheless, nearly all enzymes that use these cofactors exhibit a strong
preference for one or the other of them. This specificity enables cells to regulate different
classes of enzymes and pathways separately, prevent futile reaction cycles, and maintain
chemical driving forces by controlling the availability of the oxidized and reduced forms of
NAD(P).
Interest in nicotinamide cofactor specificity has been driven not only by scientific curiosity
but also by its importance to the engineering of cellular metabolism
1
. Several studies have
shown how balancing cofactor availability can increase pathway yields by removing carbon
inefficiencies and side products, eliminating oxygen requirements, or improving steady-state
metabolite levels
2
8
. For this reason, switching enzyme cofactor preference has been a
frequent target of protein engineering efforts ever since the first report of engineered
specificity reversal in 1990 by Scrutton and co-workers
9
. Supporting Tables 1 and 2 list
more than 25 examples of cofactor switching in each direction (NADP-to-NAD and
vice
versa
), updated from that of Khoury
et al
10
. However, a closer examination of these studies
shows that cofactor specificity reversal remains an unsolved problem: many of these efforts
were only marginally successful in achieving reversed specificity without compromising
catalytic activity.
A number of factors combine to make reversing enzyme cofactor specificity a challenging
task. Although the protein features and interactions which make up (phospho-)adenosine
binding pockets are distal from the catalytic sites of the enzymes, they have been shown
to have an outsize influence on enzyme activity. Subtle chemical changes to the cofactor
can have a dramatic effect on activity
11
, and mutations to the adenosine-interacting part of
the cofactor binding pocket can affect reaction kinetics
12
and even substrate specificity
13
.
Combined with the dynamic nature of cofactor binding
14
17
, this sensitivity to structural
perturbation has proven a major obstacle to rational and computational design approaches
10
.
These same factors have also been a hurdle to homology-guided approaches
16
,
18
, as has
the structural diversity of cofactor binding and specificity motifs
19
. Despite the sensitivity
of catalytic activity to cofactor binding pose, evolution has produced a diverse array
of structural motifs for binding NAD(P) and for conferring specificity for NAD or
NADP
15
,
19
23
. NAD(P) utilization has been documented in proteins with the canonical
Rossmann fold as well as many others, including TIM-barrel, dihydroquinoate synthase-like,
and FAD/NAD-binding folds. Considerable structural diversity may also exist within a given
Cahn et al.
Page 2
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
fold and even within an enzyme family
19
. Furthermore, many enzyme families contain
representatives with both specificities
24
27
, suggesting that specificity reversal is relatively
facile in natural evolution. In a previous study, we identified at least seven independent
cases where cofactor specificity had switched to NAD-utilization in the evolutionary history
of the otherwise exclusively NADP-preferring ketol-acid reductoisomerase enzyme (KARI)
family. Notably, each of these switches was achieved through unique combinations of amino
acid substitutions, insertions, and deletions
24
. Within the structural diversity of cofactor
binding pockets, specificity for NAD or NADP appears to be dictated largely by the
charge and polarity of the binding pocket. The negatively-charged phosphate of NADP is
often coordinated by positively-charged residues, particularly arginine, and hydrogen-bond
donating residues (Supporting Figures 1 and 2). In NAD-specific proteins, by contrast,
negatively-charged residues often serve to repel the NADP phosphate and accept hydrogen
bonds from the 2’ and 3’ ribose hydroxyls. More discrete recognition elements have
been noted, such as an arginine in NADP-binding Rossmann folds that forms a cation-pi
interaction with the adenine ring system, but none are either universal or deterministic
20
.
Finally, random mutagenesis and screening has also proven of limited utility: several amino
acids are nearly always involved in controlling specificity, and reversing specificity has
required multiple simultaneous mutations
5
, leading to an intractably large combinatorial
space of mutations to explore. This problem is compounded by strong non-additivity in
the effects of mutations
5
,
28
,
29
, which renders uphill-walk modes of optimization ineffective.
As a result, no single engineering approach has proven universally applicable, and the
full promise of manipulating the NAD(P) cofactor utilization of enzymes and metabolic
pathways has not been attained.
RESULTS AND DISCUSSION
Approach
We set out to develop a comprehensive and user-friendly approach to reversing the
nicotinamide cofactor specificity of any NAD(P)-utilizing enzyme. Though we had
previously developed a simple recipe for such a reversal for the enzymes of the KARI
enzyme family
18
, early tests showed that this recipe was not sufficient for the more general
problem. The more general strategy, which was developed with a focus on ease-of-use for
non-experts, is laid out in detail in Supporting Material 1. As with the recipe we developed
for KARIs, it comprises three steps (Figure 2): enzyme structural analysis, design and
screening of focused mutant libraries for reversing cofactor preference, and, finally, recovery
of catalytic efficiency. The development of each step proceeded from a careful study of
available literature on the structural biology, biophysics, and engineering of nicotinamide
cofactor specificity followed by development of rational engineering heuristics. We then
formalized these heuristics and implemented them in a computational framework: Cofactor
Specificity Reversal: Structural Analysis and LibrAry Design (CSR-SALAD), which is
freely available online at
http://www.che.caltech.edu/groups/fha/CSRSALAD/index.html
.
Supporting Figure 4 shows the web interface of CSR-SALAD and an example of its output.
As might be intuitively expected, nearly all of the mutations previously required for cofactor
specificity reversal are in the immediate vicinity of the 2’ moiety of the NAD/NADP
Cahn et al.
Page 3
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
cofactor (Supporting Tables 1 and 2, column 5). While distal residues are occasionally
mutated in cofactor-switched proteins, only rarely do they actually contribute to the reversal
of specificity. Based on these past studies, we hypothesized that targeting a limited set
of residues would be sufficient for cofactor switching. In CSR-SALAD, the specificity­
determining residues are therefore defined as those that contact the 2’ moiety directly,
those in position to contact it through water-mediated interactions, or – specifically for
NAD-to-NADP switching – those that can be mutated to contact the expanded 2’ moiety of
the NADP cofactor.
To assist with engineering, it is useful to obtain more information about a residue than
simply whether it contacts the cofactor. Therefore, CSR-SALAD utilizes a system of
classifications (described in Supporting Material 1 and illustrated in Supporting Figure
3) that we developed to describe a residue’s role in forming the cofactor-binding pocket.
This classification system is informed by that introduced in 1997 by Carugo and Argos
20
and is used in the library design process to discriminate among different sets of potential
mutations. Examples of these classes include residues interacting with the face of the
adenine ring system (what Carugo and Argos called S10), the edge of the rings (S8) or
interacting with both the 2’-moiety and the 3’-hydroxyl (S9).
The next step is to design a library of mutations directed at the identified specificity­
determining residues. To keep library sizes small and thus keep the screening for higher
activity with the new cofactor experimentally tractable, CSR-SALAD predicts sub-saturation
degenerate codon libraries
30
33
, wherein specified mixtures of nucleotides are used to
generate combinations of amino acids at each targeted position. So that library sizes can
be tailored to the experimental capabilities of the user, CSR-SALAD possesses, for each
residue in each structural class, a range of degenerate codons coding for different numbers of
amino acids. The selection of these degenerate codons is guided primarily by the inclusion
of mutations to structurally similar residues that have already been shown to be useful
for cofactor specificity reversal in the desired direction. This selection is subjective and
based on the accumulated lessons of prior studies by ourselves and others. We hope that
application of the approach and of the CSR-SALAD software will result in feedback that
will improve the algorithm as more data become available.
The final step is recovery of enzymatic activity. Cofactor-switched enzymes – in fact any
enzyme having multiple mutations – often suffer a significant loss of activity
18
,
34
,
35
, and
compensatory mutations must be identified to recover it. These mutations re-stabilize or
re-activate the protein for catalysis with the new cofactor, reversing whatever detrimental
side effects the specificity-reversing mutations produced. Because these mutations are
often remote from the cofactor-switching mutations, in our previous work on KARIs
we recommended using random mutagenesis and screening to discover them
18
. While
highly effective, the large library sizes for random mutagenesis render this approach time­
consuming and equipment- and labor-intensive. Here, we have taken the unprecedented
step of using structural information to predict positions in the amino acid sequence
with dramatically increased probabilities of harboring compensatory mutations. This
allows the production of highly active enzymes from the screening of just a handful
of single-site saturation libraries and combination of the most beneficial mutations. We
Cahn et al.
Page 4
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
have identified several types of activity recovery positions based on common features of
previous engineering efforts, as outlined in Supporting Material 1. The most effective have
consistently been mutations around the adenine ring, and all of the mutations used in
our experimental validation fall into this category. In a previous study we demonstrated
the power of these mutations to boost the
in vitro
and
in vivo
activities of a number
of NAD(P)-dependent enzymes
12
, and these effects are even more dramatic in cofactor­
switched enzymes.
Experimental Validation
To demonstrate the general utility of CSR-SALAD for finding cofactor-switching mutations,
we selected a test set of four oxidoreductases drawn from those previously studied in
our lab and enzymes of industrial interest:
Arabidopsis thaliana
glyoxylate reductase
(GR),
Saccharomyces cerevisae
cinnamyl alcohol dehydrogenase (CinADH),
Talaromyces
emersonii
xylose reductase (XR), and
Thermotoga maritima
iron-containing alcohol
dehydrogenase (FeADH) (Table 1). Although CSR-SALAD is capable of processing NAD­
and NADP-bound structures, we focused only on NADP-preferring enzymes because
switching in this direction is more relevant for applications
36
38
and has also proved to
be more difficult to achieve in the past (Supporting Tables 1 and 2). The selected enzymes
include the most common NAD(P)-binding fold, the Rossmann fold
39
, as well as other, less
common cofactor binding folds (TIM barrel and dihydroquinoate synthase-like folds). To
demonstrate the robustness of CSR-SALAD analysis, we also included one protein (XR)
without a crystal structure, but for which the structure of a homologous enzyme with 55%
sequence identity was known, and one enzyme (GR) crystallized in the absence of cofactor,
for which the cofactor-binding pose could be modeled. We also included proteins with strict
(XR) and loose (FeADH) natural cofactor specificity. For each enzyme, we generated the
mutant library recommended using CSR-SALAD default parameters and screened it for
cofactor-switched variants. The best variants – those which had the greatest activity toward
NADH provided that activity was greater than that toward NADPH – were then subjected to
between zero and two rounds of single-site saturation mutagenesis for activity recovery, and
the best final variants were purified and characterized for Michaelis-Menten kinetics (Table
1 and Supporting Table 3).This procedure led to successful cofactor specificity reversal
for all four enzymes, and for three of the four test enzymes the catalytic efficiency was
recovered back to at least 10% of wild type. Indeed, by this metric these represent three of
the most successful NADP-to-NAD cofactor switches ever reported (Supporting Table 1).
For XR, which we were also able to switch, activity recovery was less successful, but this
enzyme showed the most dramatic reversal of cofactor specificity, nearly 5,000-fold, due
to the extremely low initial activity on NADH. We suspect that, should greater activity be
required, further compensatory mutations could be found by directed evolution
18
.
Cofactor specificity was reversed in all four enzymes by a concerted set of three binding­
pocket mutations, although for several enzymes more than three positions were probed in
the libraries designed by CSR-SALAD (Supporting Material 3). Further catalytic activity
was recovered in three of the four enzymes by single mutations in the adenine-binding
pocket; FeADH had sufficient activity after specificity reversal that no additional mutations
were warranted. Notably, the final sets of cofactor-switching mutations identified for each
Cahn et al.
Page 5
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
enzyme show no significant overlap. Indeed, the only mutation that appears twice, serine-to­
glycine in XR and FeADH, occurs at residues that are in strikingly different regions of
their respective binding pockets. While some simple patterns emerge, broader engineering
principles are not readily apparent. For instance, while each set of mutations contains one or
more mutations to a negatively-charged residue, these mutations come from residues which
play different structural roles in the wild-type binding of the cofactors (Supporting Figure
5). The rest of the mutations introduce amino acids whose role in cofactor specificity is less
immediately evident. Even though the precise determinants of cofactor specificity are subtle
and often difficult to predict, the successful reversal of cofactor specificity by CSR-SALAD
demonstrates that active proteins can be selected from small libraries containing residues
that are primed to alter specificity.
Comparison to Previous Studies
Because the components of the approach described here were developed based on a
synthesis of the previous literature, we sought to determine whether modestly-sized CSR­
SALAD libraries could also recapitulate cofactor-switching mutations identified in previous
experiments. Indeed, for the majority of the previously published enzymes with a reversal
of specificity and final activity at least 10% of initial activity, the CSR-SALAD libraries
contained the precise combination of mutations used (not including distal compensatory
mutations) (Table 2). Given that there may be multiple sets of mutations that reverse
cofactor specificity for a given enzyme
5
,
40
42
, it seems reasonable to hypothesize that CSR­
SALAD libraries would have been able to successfully switch enzymes where the precise
combination of mutations was not recapitulated, particularly because the same sites are often
targeted for mutagenesis (full data on mutant recapitulation can be found in Supporting
Tables 4 and 5). Even without these, the ability to recapitulate known mutation combinations
means that this approach has been validated not merely on the four structurally diverse
proteins described above, but has been shown capable of producing switched and active
enzymes in 22 different proteins.
As mentioned above, we did not experimentally target NAD-preferring proteins in this study
because these enzymes tend to be easier to switch. This tendency is likely a result of the
significant potential binding energy that the phosphate provides. While the recapitulation of
the mutations used in nearly a dozen of the best prior engineered switches in this direction
provides strong evidence that the approach described here is capable of providing guidance
for both classes of switching efforts, there may exist structures of NAD-binding pockets
with unforeseen challenges that require modifications to the library design algorithm.
Another potential challenge comes from enzymes that use NAD(P) in the context of multi­
step electron transfer pathways, such as mono- and dioxygenases. Previous attempts to
switch these enzymes’ cofactor preference have resulted in nicotinamide cofactor oxidation
uncoupled from substrate turnover
43
,
44
, and preliminary CSR-SALAD-based engineering
of a Baeyer-Villiger monooxygenase encountered this same hurdle (data not shown). We
propose that the more complex electron transfers of these enzymes are more susceptible to
lethal perturbation, and that reversing specificity in these enzymes may require simultaneous
monitoring of uncoupled activity and likely directed evolution to recover coupling of
cofactor utilization to product formation.
Cahn et al.
Page 6
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
It is also worth noting that the libraries designed by CSR-SALAD are inspired by
previous laboratory results more than by natural patterns of specificity. Hence, whereas
it recapitulates protein engineering studies well, CSR-SALAD does not necessarily replicate
the natural patterns of cofactor specificity. For example, the two known naturally NAD­
preferring xylose reductases
45
,
46
possess 0 and 1 of the amino acids used here to
alter the specificity of
T. emersonii
XR. Although protein engineers typically restrict
themselves to amino acid substitutions, natural evolution often involves insertions and
deletions
47
, as observed in the KARI enzyme family
19
. No simple methods exist for making
random in-frame insertion/deletion mutants, and the results of insertions and deletions
can be challenging to predict, as loop geometries may change significantly. However, CSR­
SALAD’s ability to find novel engineering solutions based solely on substitutions supports
the basic hypothesis that multiple cofactor-switching solutions exist for a given protein.
Conclusion and Outlook
Protein engineers generally approach engineering tasks in a detail-oriented fashion and
on case-by-case basis. This is an ‘artisanal’ approach that requires considerable expertise.
Here, we have taken on an important and challenging engineering problem and developed a
general approach that reliably produces the desired enzyme properties with a minimum of
expertise and labor. As protein engineering continues to mature as a discipline, we foresee
the development of a number of similar tools to make enzyme optimization more accessible
to users in the broader fields of metabolic and biological engineering.
In developing this method, the diversity of cofactor binding motifs in nature inspired
us to hypothesize that multiple mutational solutions exist for each specificity reversal
problem. Combined with the ability to recover activity without affecting specificity, we
believed that this would allow for at least one satisfactory enzyme to be found in a
focused pool of mutants. Careful consideration of the previous literature on nicotinamide
cofactor specificity allowed us to develop heuristics that limit the mutational space in
this semi-rational approach. Formalization into the CSR-SALAD computational framework
makes this approach broadly accessible. Nicotinamide cofactor specificity reversal was
a particularly inviting problem for this approach, as we were able to take advantage of
the extensive literature and a facile high-throughput screen based on the unique spectral
signature of reduced nicotinamide. It remains to be seen whether similar structure-based and
knowledge-guided semi-rational approaches can be developed for other protein engineering
problems.
We foresee that CSR-SALAD will be useful in the fields of synthetic biology, metabolic
engineering, and biocatalysis and will make the reversal of cofactor specificity a routine task
rather than a formidable engineering endeavor.
METHODS
CSR-SALAD development
CSR-SALAD was built in Python, relying heavily on the PDB module of the
Biopython package
48
. The analysis component was tested on a representative set of
Cahn et al.
Page 7
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
499 NAD-bound structures and 463 NADP-bound structures selected based on sequence
identity and resolution. Degenerate codon selection was assisted by the LibDesign
31
and AA-Calculator
49
tools and optimized using the Ambiguous Nucleotide Tool
(ANT) framework
32
. CSR-SALAD has been implemented as a PHP-based web-server
that is available at
http://www.che.caltech.edu/groups/fha/CSRSALAD/index.html
. Further
documentation of CSR-SALAD’s internal workings and guidelines for use can be found on
the website or in the Supporting Material 1.
Structure analysis
Homology models were created using the SWISS-MODEL server
50
. For proteins
crystallized without bound cofactor, cofactor was manually placed by alignment with a
homologous protein, followed by manual side-chain rotamer adjustments in PyMol
51
.
Cloning and library creation
Genes were either obtained from Integrated DNA Technologies (IDT) as gBlock linear
fragments or cloned from pre-existing vectors and were cloned into pET22b(+) in frame
with the C-terminal His
6
-tag for expression in
Escherichia coli
BL21(DE3) ‘E. Cloni’
(Lucigen) using Gibson cloning
52
with overlap at the T7 promoter and terminator sequences.
Mutagenic primers for site-saturation mutagenesis were obtained from IDT and treated
as suggested by IDT protocols. Libraries were generated using a modified version of the
QuikChange method (Stratagene) as described previously
53
. Some of the libraries screened
were based on early versions of the CSR-SALAD algorithm, but all reported mutants are
found in the CSR-SALAD libraries generated using default parameters.
Following transformation with library DNA, single colonies were picked with sterile
toothpicks and inoculated into 300 μL of Luria broth supplemented with 100 μg/mL
ampicillin (LB-Amp) in shallow-well 96-well plates. Following overnight growth at 37 °C
with shaking at 225 rpm and 80% humidity, 50 μL of the pre-cultures were added to 600 μL
of fresh LB-Amp media in deep-well 96-well plates and grown for 3 h at 37 °C. Then, 50
μL of additional LB-Amp containing 0.25 mM isopropyl thiogalactopyranoside (IPTG) were
added and expression was continued at a reduced temperature. For expression temperatures
and times, see Table 3. The expression cultures were harvested through centrifugation, and
the plates containing cell pellets were stored at −20 °C until screening.
Library screening
Assay procedures varied depending on the protein. The following is the general protocol, but
deviations for specific proteins are listed in Table 3.
E. coli
cells were resuspended in the
appropriate lysis buffer (Table 3) containing 750 mg/L lysozyme, 10 mg/L DNase I, and 2
mM MgCl
2
. Lysis was accomplished at 37 °C for 1 h. Enzyme activities were then assayed
by monitoring NAD(P)H consumption in the presence of the substrate molecule at 340 nm
in a plate reader.
Enzyme expression, purification, and kinetic measurements
For larger-scale expression, 5–50 mL pre-cultures were grown overnight and diluted 1:250
into fresh medium for expression. After expression cultures reached an OD
600
of 0.6–0.9,
Cahn et al.
Page 8
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
they were induced with IPTG to a final concentration of 0.5 mM. Expression then proceeded
for the same times and temperatures used for library expression. Expression cultures were
harvested by centrifugation, the supernatant was discarded, and the pellets were frozen at
−20 °C.
For purification, cell pellets were resuspended in 10–20 mL buffer A (25 mM Tris, 100
mM NaCl, 20 mM imidazole, pH 7.4) and lysed by sonication or using BugBuster protein
extraction reagent (EMD Millipore). The lysate was clarified by centrifugation, and the
enzymes were purified via their C-terminal His
6
-tag using High Performance (HP) Ni-NTA
Sepharose columns (GE Healthcare, Waukesha, WI, USA) on an Äkta Xpress FPLC (GE
Healthcare). The concentration of purified protein was determined using the Bradford assay
(Bio-Rad, Hercules, CA, USA).
For rate measurements,
k
cat
values were determined using the same assay conditions as
above with saturating cofactor and substrate. Michaelis–Menten constants were determined
by varying cofactor concentration, and activity was monitored using fluorescence (excitation
340 nm, emission 440 nm) for improved sensitivity. At least six cofactor concentrations
were used for these determinations, and all measurements were performed at least three
times. MATLAB (MathWorks, Natwick, MA, USA) was used for parameter fitting.
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
ACKNOWLEDGMENTS
This work was supported by the Gordon and Betty Moore Foundation through grant number GBMF2809 to
the Caltech Programmable Molecular Technology Initiative and by American Recovery and Reinvestment Act
(ARRA) funds through the National Institutes of Health Shared Instrumentation Grant Program, grant number
S10RR027203, to F.H.A. J.K.B.C. acknowledges the support of the Resnick Sustainability Institute (Caltech). The
authors thank Ruchi Jahagirdar and Lisa Mears for experimental assistance and Tilman Flock for providing the list
of nonredundant PDBs used for CSR-SALAD testing and validation. They also thank numerous former and current
members of the Arnold and Mayo labs for invaluable suggestions and discussions.
REFERENCES
1. Wang Y, San K-Y and Bennett GN (2013) Cofactor engineering for advancing chemical
biotechnology, Curr. Opin. Biotechnol. 24, 994–999. [PubMed: 23611567]
2. Matsushika A, Watanabe S, Kodaki T, Makino K, Inoue H, Murakami K, Takimura O, and
Sawayama S (2008) Expression of protein engineered NADP
+
-dependent xylitol dehydrogenase
increases ethanol production from xylose in recombinant
Saccharomyces cerevisiae
, Appl.
Microbiol. Biotechnol. 81, 243–255. [PubMed: 18751695]
3. Bengtsson O, Hahn-Hagerdal B and Gorwa-Grauslund MF (2009) Xylose reductase from
Pichia
stipitis
with altered coenzyme preference improves ethanolic xylose fermentation by recombinant
Saccharomyces cerevisiae
, Biotechnol. Biofuels 2.
4. Hasegawa S, Uematsu K, Natsuma Y, Suda M, Hiraga K, Jojima T, Inui M, and Yukawa H (2012)
Improvement of the redox balance increases L-valine production by
Corynebacterium glutamicum
under oxygen deprivation conditions, Appl. Environ. Microbiol. 78, 865–875. [PubMed: 22138982]
5. Bastian S, Liu X, Meyerowitz JT, Snow CD, Chen MMY, and Arnold FH (2011) Engineered ketol­
acid reductoisomerase and alcohol dehydrogenase enable anaerobic 2-methylpropan-1-ol production
at theoretical yield in
Escherichia coli
, Metab. Eng. 13, 345–352. [PubMed: 21515217]
Cahn et al.
Page 9
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
6. Tamakawa H, Ikushima S and Yoshida S (2011) Ethanol production from xylose by a recombinant
Candida utilis
strain expressing protein-engineered xylose reductase and xylitol dehydrogenase
Biosci., Biotechnol., Biochem 75, 1994–2000. [PubMed: 21979076]
7. Shen CR, Lan EI, Dekishima Y, Baez A, Cho KM, and Liao JC (2011) Driving forces enable
high-titer anaerobic 1-butanol synthesis in
Escherichia coli
, Appl. Environ. Microbiol. 77, 2905–
2915. [PubMed: 21398484]
8. Xiao Z Lv C, Gao C, Qin J, Ma C, Liu Z, Liu P, Li L, and Xu P (2010) A novel whole-cell
biocatalyst with NAD
+
regeneration for production of chiral chemicals, PLoS ONE 5, e8860.
[PubMed: 20126645]
9. Scrutton NS, Berry A and Perham RN (1990) Redesign of the coenzyme specificity of a
dehydrogenase by protein engineering Nature 343, 38–43. [PubMed: 2296288]
10. Khoury GA Fazelinia H, Chin JW, Pantazes RJ, Cirino PC, and Maranas CD (2009) Computational
design of
Candida boidinii
xylose reductase for altered cofactor specificity, Prot. Sci. 18, 2125–
2138.
11. Mesecar AD, Stoddard BL and Koshland DE Jr. (1997) Orbital steering in the catalytic power
of enzymes: Small structural changes with large catalytic consequences, Science 277, 202–206.
[PubMed: 9211842]
12. Cahn JKB, Baumschlager A, Brinkmann-Chen S and Arnold FH (2016) Mutations in adenine­
binding pockets enhance catalytic properties of NAD(P)H-dependent enzymes, Prot. Eng., Des.
Sel 29, 31–38.
13. Maddock DJ, Patrick WM and Gerth ML (2015) Substitutions at the cofactor phosphate-binding
site of a clostridial alcohol dehydrogenase lead to unexpected changes in substrate specificity, Prot.
Eng., Des. Sel 28, 251–258.
14. Naylor CE, Gover S, Basak AK, Cosgrove MS, Levy HR, and Adams MJ (2001) NADP(+)
and NAD(+) binding to the dual coenzyme specific enzyme
Leuconostoc mesenteroides
glucose
6-phosphate dehydrogenase: Different interdomain hinge angles are seen in different binary and
ternary complexes, Acta Cryst. D 57, 635–648. [PubMed: 11320304]
15. Kavanagh KL, Klimacek M, Nidetzky B and Wilson DK (2003) Structure of xylose reductase
bound to NAD(+) and the basis for single and dual co-substrate specificity in family 2 aldo-keto
reductases, Biochem. J. 373, 319–326. [PubMed: 12733986]
16. Campbell E, Chuang S & Banta S (2013) Modular exchange of substrate-binding loops alters both
substrate and cofactor specificity in a member of the aldo-keto reductase superfamily, Prot. Eng.,
Des. Sel. 26, 181–186.
17. Plapp BV (2010) Conformational changes and catalysis by alcohol dehydrogenase, Arch. Biochem.
Biophys. 493, 3–12. [PubMed: 19583966]
18. Brinkmann-Chen S Flock T, Cahn JKB, Snow CD, Brustad EM, McIntosh JA, Meinhold P, Zhang
L, and Arnold FH (2013) General approach to reversing ketol-acid reductoisomerase cofactor
dependence from NADPH to NADH, Proc. Nat. Acad. Sci. USA 110, 10946–10951. [PubMed:
23776225]
19. Cahn JKB, Brinkmann-Chen S, Spatzal T, Wiig JA, Buller AR, Einsle O, Hu Y, Ribbe MW, and
Arnold FH (2015) Cofactor specificity motifs and the induced fit mechanism in class I ketol-acid
reductoisomerases, Biochem. J. 468, 475–484. [PubMed: 25849365]
20. Carugo O and Argos P (1997) NADP-dependent enzymes. I: Conserved stereochemistry of
cofactor binding, Proteins: Struct., Funct., Bioinf 28, 10–28.
21. Moodie SL, Mitchell JB and Thornton JM (1996) Protein recognition of adenylate: An example of
a fuzzy recognition template, J. Mol. Biol. 263, 486–500. [PubMed: 8918603]
22. Denessiouk KA, Rantanen VV and Johnson MS (2001) Adenine recognition: a motif present
in ATP-, CoA-, NAD-, NADP-, and FAD-dependent proteins, Proteins 44, 282–291. [PubMed:
11455601]
23. Wallen JR, Paige C, Mallett TC, Karplus PA and Claiborne A (2008) Pyridine nucleotide
complexes with
Bacillus anthracis
coenzyme A-disulfide reductase: A structural analysis of dual
NAD(P)H specificity, Biochemistry 47, 5182–5193. [PubMed: 18399646]
24. Brinkmann-Chen S, Cahn JKB and Arnold FH (2014) Uncovering rare NADH-preferring ketol­
acid reductoisomerases, Metab. Eng. 26, 17–22. [PubMed: 25172159]
Cahn et al.
Page 10
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
25. Di Luccio E, Elling RA and Wilson DK (2006) Identification of a novel NADH-specific aldo­
keto reductase using sequence and structural homologies, Biochem. J. 400, 105–114. [PubMed:
16813561]
26. Hatrongjit R and Packdibamrung K (2010) A novel NADP(+)-dependent formate dehydrogenase
from
Burkholderia stabilis
15516: Screening, purification and characterization, Enzyme Microb.
Technol. 46, 557–561.
27. Jensen CN, Ali ST, Allen MJ and Grogan G (2014) Exploring nicotinamide cofactor promiscuity in
NAD(P)H-dependent flavin containing monooxygenases (FMOs) using natural variation within the
phosphate binding loop. Structure and activity of FMOs from
Cellvibrio
sp. BR and
Pseudomonas
stutzeri
NF13 J. Mol. Cat. B 109, 191–198.
28. Rodriguez-Arnedo A, Camacho M, Llorca F and Bonete MJ (2005) Complete reversal of
coenzyme specificity of isocitrate dehydrogenase from
Haloferax volcanii
, Protein J. 24, 259–266.
[PubMed: 16284723]
29. Rosell A, Valencia E, Ochoa WF, Fita I, Pares X, and Farres J (2003) Complete reversal
of coenzyme specificity by concerted mutation of three consecutive residues in alcohol
dehydrogenase, J. Biol. Chem. 278, 40573–40580. [PubMed: 12902331]
30. Reetz MT and Wu S (2008) Greatly reduced amino acid alphabets in directed evolution: Making
the right choice for saturation mutagenesis at homologous enzyme positions, Chem. Comm 5499–
5501. [PubMed: 18997931]
31. Mena MA and Daugherty PS (2005) Automated design of degenerate codon libraries, Prot. Eng.,
Des. Sel. 18, 559–561.
32. Engqvist MKM and Nielsen J (2015) ANT: Software for generating and evaluating degenerate
codons for natural and expanded genetic codes, ACS Syn. Biol. 4, 935–938.
33. Cornish-Bowden A (1985) Nomenclature for incompletely specified bases in nucleic acid
sequences: Reccomendations 1984, Nucleic Acids Res. 13, 3021–3030. [PubMed: 2582368]
34. Fasan R, Meharenna YT, Snow CD, Poulos TL and Arnold FH (2008) Evolutionary history of a
specialized P450 propane monooxygenase, J. Mol. Biol. 383, 1069–1080. [PubMed: 18619466]
35. Bloom JD, Arnold FH and Wilke CO (2007) Breaking proteins with mutations: Threads and
thresholds in evolution, Mol. Syst. Biol. 3, 76. [PubMed: 17262035]
36. Woodyer R, van der Donk WA & Zhao HM (2003) Relaxing the nicotinamide cofactor specificity
of phosphite dehydrogenase by rational design, Biochemistry 42, 11604–11614. [PubMed:
14529270]
37. Verho R, Londesborough J, Penttila M and Richard P (2003) Engineering redox cofactor
regeneration for improved pentose fermentation in
Saccharomyces cerevisiae
, Appl. Environ.
Microbiol. 69, 5892–5897. [PubMed: 14532041]
38. Huffer S, Roche CM, Blanch HW and Clark DS (2012)
Escherichia coli
for biofuel production:
Bridging the gap from promise to practice, Trends Biotechnol. 30, 538–545. [PubMed: 22921756]
39. Rossmann MG, Moras D and Olsen KW (1974) Chemical and biological evolution of nucleotide­
binding protein Nature 250, 194–199. [PubMed: 4368490]
40. Rane MJ and Calvo KC (1997) Reversal of the nucleotide specificity of ketol acid
reductoisomerase by site-directed mutagenesis identifies the NADPH binding site, Arch. Biochem.
Biophys. 338, 83–89. [PubMed: 9015391]
41. Liang L, Zhang JQ and Lin ZL (2007) Altering coenzyme specificity of
Pichia stipitis
xylose
reductase by the semi-rational approach CASTing, Microb. Cell Fact. 6, 36. [PubMed: 18028553]
42. Zeng Q-K, Du H-L, Wang J-F, Wei D-Q, Wang X-N, Li Y-X, and Lin Y (2009) Reversal of
coenzyme specificity and improvement of catalytic efficiency of Pichia stipitis xylose reductase by
rational site-directed mutagenesis, Biotechnol. Lett. 31, 1025–1029. [PubMed: 19330484]
43. Maurer SC, Kühnel K, Kaysser LA, Eiben S, Schmid RD, and Urlacher VB (2005) Catalytic
hydroxylation in biphasic systems using CYP102A1 mutants, Adv. Synth. Catal. 347, 1090–1098.
44. Fasan R, Crook NC, Peters MW, Meinhold P, Buelter T, Landwehr M, Cirino PC, and Arnold FH
(2011) Improved product-per-glucose yields in P450-dependent propane biotransformations using
engineered
Escherichia coli
, Biotechnol. Bioeng. 108, 500–510. [PubMed: 21246504]
Cahn et al.
Page 11
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
45. Lee JK, Koo BS and Kim SY (2003) Cloning and characterization of the xyl1 gene, encoding
an NADH-preferring xylose reductase from
Candida parapsilosis
, and its functional expression in
Candida tropicalis
, Appl. Environ. Microbiol. 69, 6179–6188. [PubMed: 14532079]
46. Cahn JKB (2016) Discovery of an NAD-preferring xylose reductase. In Engineering,
Predicting, and Understanding Nicotinamide Cofactor Specificity pp 45–55, California Institute
of Technology - Ph.D. Thesis.
47. Tóth-Petróczy Á and Tawfik DS (2013) Protein insertions and deletions enabled by neutral
roaming in sequence space, Mol. Biol. Evol. 30, 761–771. [PubMed: 23315956]
48. Hamelryck T and Manderick B (2003) PDB file parser and structure class implemented in Python,
Bioinformatics 19, 2308–2310. [PubMed: 14630660]
49. Firth AE and Patrick WM (2008) GLUE-IT and PEDEL-AA: New programmes for analyzing
protein diversity in randomized libraries, Nucleic Acids Res. 36, W281–285. [PubMed: 18442989]
50. Arnold K, Bordoli L, Kopp J & Schwede T (2006) The SWISS-MODEL workspace: A web-based
environment for protein structure homology modelling, Bioinformatics 22, 195–201. [PubMed:
16301204]
51. Schrodinger LLC. (2010) The PyMOL Molecular Graphics System, Version 1.3r1.
52. Gibson DG (2011) Enzymatic assembly of overlapping DNA fragments, In Methods in
Enzymology (Christopher V, Ed.), pp 349–361, Academic Press.
53. Engqvist MKM, McIsaac RS, Dollinger P, Flytzanis NC, Abrams M, Schor S, and Arnold FH
(2015) Directed evolution of
Gloeobacter violaceus
rhodopsin spectral properties, J. Mol. Biol.
427, 205–220. [PubMed: 24979679]
Cahn et al.
Page 12
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Figure 1.
Cofactors nicotinamide adenine dinucleotide (NAD, top) and nicotinamide adenine
dinucleotide phosphate (NADP, bottom) in representative Rossmann fold binding pockets
(PDBs 4XDY and 4TSK, respectively). The highlighted 2’ recognition element (the
phosphate of NADP or hydroxyl of NAD) and the chemically relevant hydride-bearing
nicotinamide are separated in space and by multiple covalent bonds. In this paper, we
use NAD and NADP, collectively NAD(P), when speaking generally about these cofactors,
independent of their oxidation state, and only indicate the state, i.e. NADH / NAD+, when
referring to those compounds specifically, such as for experimental details.
Cahn et al.
Page 13
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Figure 2.
CSR-SALAD performs three tasks: structure analysis, design of cofactor-switching libraries,
and identification of positions for activity recovery.
Cahn et al.
Page 14
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Cahn et al.
Page 15
Table 1.
NADP-preferring enzymes used for experimental validation of the CSR-SALAD method for switching cofactor preference to NAD while retaining
catalytic activity. UniProt IDs, PDB accession codes, and Structural Classification of Proteins (SCOP) classification of their cofactor binding folds
are listed for each. Mutations in bold are those used for switching cofactor specificity, those from recovery libraries are italicized. Fold changes from
wild-type kinetics are provided, as is the log of the mutant’s NADH
k
cat
/
K
M
divided by the wild-type NADPH
k
cat
/
K
M
. Specific values for wild-type and
mutant enzymes are provided in
Supporting Table 3
.
Enzyme
UniProt
RCSB PDB
Fold (SCOP)
Mutations
Fold change from wild type
Specificity
(NADH/
NADPH)
c
log (NADH
mut
/
NADPH
wt
)
NADH
NADPH
k
cat
K
M
k
cat
/
K
M
k
cat
K
M
k
cat
/
K
M
Arabidopsis thaliana
glyoxylate reductase
Q9LSV0
3DOJ
a
Rossmann
(c.2.1.6)
R31L, T32K,
K35D,
C68R
2.4
0.9
2.7
0.4
4.6
0.1
33
−0.7
Saccharomyces cerevisae
cinnamyl alcohol
dehydrogenase
Q04894
1PIW
Rossmann
(c.2.1.1)
S210D, R211P,
K215E,
S253P
110
0.5
190
1.5
0.5
2.9
65
0.8
Talaromyces emersonii
xylose reductase
C5J3R6
1K8C
b
TIM Barrel
(c.1.7.1)
S272G, N273D,
R277Y,
Q280E
14
0.4
31
0.02
2.5
0.01
4900
−1.5
Thermotoga maritima
iron-containing alcohol
dehydrogenase
Q9X022
1VHD
DHQS-like
(e.22.1.2)
G36E, S38N, S39G
0.9
0.2
5.9
0.2
2.3
0.1
84
0.5
a
Structure 3DOJ has no cofactor, so the cofactor was introduced from structure 3PEF (
G. metallireducens
γ
-hydroxybutyrate dehydrogenase) on the basis of a backbone alignment.
b
T. emersonii
xylose reductase (XR) has not been crystallized, so a homology model was generated from 1K8C, the structure of an XR from
C. tenuis
.
c
Specificity is the ratio of NADH
k
cat
/
K
M
divided by NADPH
k
cat
/
K
M
for each enzyme. Presented here is fold change.
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Cahn et al.
Page 16
Table 2.
Recapitulation of mutations from 13 successful NADP-to-NAD and 17 successful NAD-to-NADP cofactor
specificity reversals reported in the literature. Includes only mutants that had
k
cat
/
K
M
for the switched cofactor
greater than the wild type’s preferred cofactor and where mutant
k
cat
/
K
M
was at least 10% of wild type. This
count excludes residues distal to the 2’ binding pocket. For a full breakdown of mutant recapitulation, see
Supporting Tables 4
and
5
.
Percent mutations recapitulated
0%
1–49%
50–99%
100%
NADP-to-NAD
0
1
5
7
NAD-to-NADP
1
4
1
11
ACS Synth Biol
. Author manuscript; available in PMC 2021 November 24.