Structural and mutational analysis of the nonribosomalpeptide synthetase heterocyclization domain providesinsight into catalysisKristjan Bloudoffa, Christopher D. Fageb, Mohamed A. Marahielb, and T. Martin Schmeinga,1

aDepartment of Biochemistry, McGill University, Montreal, QC H3G 0B1, Canada; and bDepartment of Chemistry/Biochemistry, LOEWE Center for SyntheticMicrobiology, Philipps-Universität Marburg, 35032 Marburg, Germany

Edited by Peter B. Moore, Yale University, New Haven, CT, and approved November 30, 2016 (received for review August 31, 2016)

Nonribosomal peptide synthetases (NRPSs) are a family of multido-main, multimodule enzymes that synthesize structurally and func-tionally diverse peptides, many of which are of great therapeutic orcommercial value. The central chemical step of peptide synthesis isamide bond formation, which is typically catalyzed by the conden-sation (C) domain. In many NRPS modules, the C domain is replacedby the heterocyclization (Cy) domain, a homologous domain thatperforms two consecutive reactions by using hitherto unknowncatalytic mechanisms. It first catalyzes amide bond formation, andthen the intramolecular cyclodehydration between a Cys, Ser, or Thrside chain and the backbone carbonyl carbon to form a thiazoline,oxazoline, or methyloxazoline ring. The rings are important for theform and function of the peptide product. We present the crystalstructure of an NRPS Cy domain, Cy2 of bacillamide synthetase, at aresolution of 2.3 Å. Despite sharing the same fold, the active sitesof C and Cy domains have important differences. The structureallowed us to probe the roles of active-site residues by using muta-tional analyses in a peptide synthesis assay with intact bacillamidesynthetase. The drastically different effects of these mutants, inter-preted by using our structural and bioinformatic results, provideinsight into the catalytic mechanisms of the Cy domain and impli-cate a previously unexamined Asp-Thr dyad in catalysis of thecyclodehydration reaction.

nonribosomal peptide synthetase | heterocyclization domain | X-raycrystallography | bacillamide | mutational analysis

Nonribosomal peptide synthetases (NRPSs) are a family oflarge, multimodular enzymes that produce a range of im-

portant bioactive secondary metabolites (1). NRPS products havegreat diversity because they can use more than 500 different acylmonomer substrates, including L-amino acids, D-amino acids, arylacids, fatty acids, hydroxy acids, and keto acids, and they cansubsequently modify these moieties during peptide synthesis.NRPSs function in a modular, assembly-line fashion. A typical

elongation module consists of a condensation (C), an adenylation(A), and a peptidyl-carrier protein (PCP) domain. The A domainspecifically recognizes and activates a monomer acyl substratethrough adenylation, then transfers it onto the prosthetic phos-phopantetheine arm of the PCP domain. This acyl-PCP then travelsto the C domain acceptor site for condensation with the upstreammodule’s acyl-PCP at the C domain donor site (2–5). The PCPdomain brings the elongated peptide chain to the downstreammodule, where it is passed off and further elongated in the nextcondensation reaction. This process is repeated in each moduleuntil the growing peptide reaches the termination module, where itis elongated and then released from the NRPS, often by a thio-esterase domain. However, most NRPSs, along with their C, A, andPCP domains, also include tailoring and/or alternative domains,which cosynthetically modify the nonribosomal peptide.One important modification that can occur during peptide

synthesis is cyclodehydration of Cys, Ser, or Thr residues intothiazoline, oxazoline, or methyloxazoline rings, respectively, by theNRPS heterocyclization (Cy) domain (6–12). These heterocyclic

rings are found in many peptides with important clinical and re-search utility, such as the antibiotics bacitracin A (6) and zelko-vamycin (13), the antitumor agents bleomycin (14) and epothilone(8), the immunosuppressant argyrin (15), the siderophores myco-bactin (16) and yersiniabactin (17), and the microbiome genotoxincolibactin (18). Introduction of the five-membered heterocyclicring makes the peptide resistant to proteolytic cleavage and caninduce conformations in the peptide that favor interaction withbiological targets (19).In NRPSs that synthesize these heterocycle-containing products,

the module specific for the Cys, Ser, or Thr monomer substratecontains a Cy domain in place of the C domain. Cy domains areevolutionarily and structurally related to C domains (20). The Cydomain first catalyzes nucleophilic attack on the thioester of a PCP-linked donor substrate by the α-amino group of a Cys-, Ser-, or Thr-PCP substrate, presumably in a manner similar to C domains (6, 7,10–12, 21–23) (Fig. 1). In the two-step cyclodehydration reactionthat follows, the thiol of the Cys side chain or hydroxyl of the Ser orThr side chain first attacks the carbonyl carbon of the newly formedamide to create the heterocycle (10, 11), and then the former car-bonyl oxygen is removed in a dehydration reaction, which intro-duces the carbon-nitrogen double bond of the thiazoline or (methyl)oxazoline ring. The nascent heterocyclic peptidyl-PCP can beused as the donor substrate by the next module’s C domain, or isfirst oxidized or reduced by discrete oxidase or reductase domains(10, 24).

Significance

Nonribosomal peptide synthetases produce peptides with widevarieties of therapeutic and biological activities. Monomer sub-strates are typically linked by a condensation domain. However,in many modules, a heterocyclization (Cy) domain takes its placeand performs both condensation and cyclodehydration of a cys-teine, serine, or threonine to form a five-membered ring in thepeptide backbone. Although studied for decades, the mecha-nisms of condensation and cyclodehydration by Cy domains werepreviously unknown. The crystal structure of a Cy domain, andaccompanying mutagenic and bioinformatics analyses, uncoverthe importance of an aspartate and a threonine for the cyclo-dehydration reaction. This study provides insight into the catal-ysis of condensation by the Cy domain and enables the proposalof a reaction mechanism for cyclodehydration.

Author contributions: K.B. and T.M.S. designed research; K.B. and T.M.S. performed re-search; K.B., C.D.F., M.A.M., and T.M.S. analyzed data; and K.B. and T.M.S. wrotethe paper with comments from C.D.F. and M.A.M.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The atomic coordinates and structure factors have been deposited in theProtein Data Bank, www.pdb.org (PDB ID code 5T3E).1To whom correspondence should be addressed. Email: martin.schmeing@mcgill.ca.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1614191114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1614191114 PNAS | January 3, 2017 | vol. 114 | no. 1 | 95–100

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

May

15,

202

0

All C domain superfamily domains share the same protein fold,so the overall form of the Cy domain is not in doubt (25, 26).However, the features that allow the Cy domain to catalyze twoseparate and different reactions are not known. Cy domains con-tain a conserved Asp motif, DxxxxD, which directly replaces thecatalytic His motif, HHxxxD, of C domains (6). Mutation of theaspartate residues of the Asp motif diminishes or abolishes cata-lytic activity of the protein (12, 23, 27, 28). Furthermore, othermutations have differential effects on the condensation reactionand the cyclodehydration reaction, suggesting that the reactions arenot catalyzed by a completely overlapping set of residues (10, 22).Because Cy domains are not larger than C domains, the addedfunction in Cy domains must occur in a confined sequence space.Here, we present the crystal structure of Cy2 of bacillamide

synthetase at 2.3 Å resolution. Bacillamide synthetase (Fig. 1) isa trimodular NRPS that produces bacillamide E, whose deriva-tives bacillamide A–D exhibit algicidal activity against dinofla-gellates, raphidophytes, and cyanobacteria (SI Appendix, Fig. S1)(29, 30). Structural determination, along with mutagenic analysisof this Cy domain active site in the context of the full, intact, andactive bacillamide synthetase, and bioinformatic investigationallowed us to identify D1226 and T1196 as important residuesfor cyclodehydration, providing a better understanding of thestructure and mechanism of the Cy domain.

Results and DiscussionThe Crystal Structure of an NRPS Heterocyclization Domain. We havesolved the crystal structure of an NRPS heterocyclization domainby X-ray crystallography to a resolution of 2.3 Å. Like all known Cdomain superfamily proteins, BmdB-Cy2 adopts two chloram-phenicol acetyltransferase folds (2), with the N- and C-terminallobes forming a pseudodimer (Fig. 2). In C domains, these twolobes assume a range of relative conformations (more “open” or“closed”; ref. 5). Superimposition of BmdB-Cy2 with each C do-main shows that it fits within the observed range and is quitesimilar to AB3403 (31) (SI Appendix, Fig. S2). The latch formed bya crossover between N- and C-terminal subdomains in C domainsis present, meaning that the Cy domain active site is also situatedin the center of a tunnel, between donor and acceptor PCPbinding sites (see SI Appendix, SI Results and Discussion for moredetailed description of the overall Cy domain structure).

BmdB-Cy2 Active Site. The main established signature that differ-entiates Cy domains from C domains is an active site motif: Cdomains have a catalytic His motif of HHxxxD, and Cy domainshave an Asp motif of DxxxxD (6). The first aspartate of the Aspmotif is essential for catalytic activity in only some Cy domains.When the first aspartate is mutated to alanine in HMWP2-Cy1,AngN-Cy1, and AngN-Cy2, condensation and cyclodehydrationare completely eliminated (23, 27), but in VibF-Cy1, VibF-Cy2,and EpoB-Cy1, both condensation and cyclodehydration occur,although they are significantly diminished (12, 23, 28) (SI Ap-pendix, Table S1 lists current and previous Cy domain mutations).The second aspartate is critical for activity (12, 23, 27, 28), butwhether its role is catalytic or structural (like the aspartate at thatposition in C domains) (32) was not definitively determined.Furthermore, a triple mutant of the Asp motif of HMWP2-Cy1 tointroduce a C domain His motif results in a catalytically inactiveprotein (23), demonstrating that a straight swap of the motifs isnot sufficient to interconvert catalytic activities.In the structure of BmdB-Cy2, both aspartate residues in

DxxxxD (D980 and D985) are oriented away from the active site(Fig. 2B). The side chain of D980 makes a bifurcated hydrogenbond with the hydroxyl of S873 and a hydrogen bond with thebackbone amide of L982. D985 hydrogen bonds with the sidechainof S988 and the amides of A987 and F1134. The interaction thisaspartate makes with a nearby arginine in C domains is present,but is water-mediated for D985-R1120 of BmdB-Cy2. Thus, theseaspartate residues directly replace the first histidine and the as-partate in the C domain HHxxxD motif, occupying the same po-sition in their respective domains and making the same or similarinteractions with the rest of the protein (SI Appendix, Fig. S2B).Indeed, the whole DxxxxD motif is essentially in the same con-formation as the HHxxxD motif. The position of the second his-tidine of the HHxxxD motif, which is the most important residuefor condensation in C domains (2–5), is occupied by A981 inBmdB-Cy2 and, thus, unable to contribute to catalysis. TheHHxxxD motif in C domains does not reorient upon substrate

A1 PCP1 Cy2 A2 PCP2 C3 2 ATP +Ala + Cys

2 AMP +2 PPi

A1&A2

adenylation +

thiolation

Cy2

condensationcondensation

cyclodehydration

C3 Cy2

H2O

TpmBacillamide E

SO

H2N SH

SO

NH2

SO

HN SHO

NH2

SO

N S

NH2

SH SH

SH SH

HN

HN

O

N

SNH2

Fig. 1. Schematic representation of BmdB and the bacillamide E bio-synthesis cycle. The A domains (orange) adenylate alanine and cysteine andtransfer them onto the phosphopantetheine arm of the PCP domains (blue).The Cy2 domain (dark green) first catalyzes amide bond formation betweenthe PCP-linked Ala and Cys residues, then catalyzes the intramolecularcyclodehydration reaction. The C3 domain (light green) catalyzes amidebond formation between the PCP-linked Ala-thiazoline moiety and freetryptamine, which releases the bacillamide E from the NRPS.

A

C-termsubdomain

N-termsubdomain

BDonor PCP binding site

D980D985

R1120

S873

Active SiteChannel

L982

A987

F1134

S988

Donor sideAcceptor side

Acceptor PCP binding site

Fig. 2. The crystal structure of BmdB-Cy2. (A) Ribbon representation ofBmdB-Cy2. The cyclization domain adopts two chloramphenicol acetyl-transferase folds, similar to C domains. (B) Close-up view of the BmdB-Cy2DxxxxD motif (orange). D980 makes a bifurcated hydrogen bond with thehydroxyl of S873 and a hydrogen bond with the backbone nitrogen of L982.D985 makes hydrogen bonds with S988 and the amides of A987 and F1134,and water-mediated interaction with R1120.

96 | www.pnas.org/cgi/doi/10.1073/pnas.1614191114 Bloudoff et al.

Dow

nloa

ded

by g

uest

on

May

15,

202

0

binding (5), and there is no indication that the Cy domain DxxxxDmotif would do so. Overall, the structure strongly suggests thatboth of these aspartate residues play structural rather than cata-lytic roles, as has been established for the corresponding residuesin C domains (2–5).

Bacillamide Synthesis Assay and Mutational Analysis. To determine theimportance of other active site residues in condensation and cycli-zation, we established a bacillamide synthesis assay. Bacillamidesynthetase is a one-protein, six-domain, 265-kDa NRPS (Fig. 1) (33).It adenylates and thiolates its Ala and Cys substrates, after which theCy2 domain performs condensation and cyclodehydration. The finaldomain in this NRPS is not a thioesterase domain, but a specializedC domain that condenses the thiazoline-containing intermediatewith free tryptamine (Tpm) to release bacillamide E. Therefore, thisterminal C domain plays all of the roles typically associated with afour-domain termination module—substrate selection, peptide bondformation, and release of the peptide product—which is rare, butnot unprecedented in NRPSs (12, 27).The Tpm in the product makes the reaction convenient to

follow by HPLC at 280 nm (Fig. 3 and SI Appendix, Fig. S3).Wild-type bacillamide synthetase showed a large peak at 15 minthat corresponded by high resolution mass spectrometry, frag-mentation mass spectrometry, and NMR to bacillamide E (1) (SIAppendix, Fig. S3C). A minor peak at ∼15.5 min, with 10% of theintensity of the main peak, corresponded to the uncyclized tri-peptide Ala-Cys-Tpm (2). This product shows that, at least in vitro,BmdB-Cy2 is not completely efficient at cyclizing all of the inter-mediates it condenses, and that the terminal C3 domain is notcompletely selective for the cyclized substrate. Furthermore, a smallpeak for the dipeptide Cys-Tpm is also evident (3; SI Appendix, Fig.S3), indicating that the terminal C3 domain also uses Cys-PCP2 asa donor substrate to some extent. Note that bacillamide E containsa thiazoline ring, and not a thiazole ring as in bacillamide A–D

extracted from natural sources (29, 34, 35). The adjacent oxidaseBmdC likely performs the oxidation postsynthetically.We next produced a series of bacillamide synthetase constructs

with mutations in the Cy domain. We targeted residues in spatialproximity to the DxxxxD motif in the active site tunnel of BmdB-Cy2, with side chains that could play an important role in con-densation or cyclization, and which are largely conserved inalignments of characterized Cy domains (SI Appendix, Fig. S4).Therefore, bacillamide synthetase with mutations Y859F, T1116A,F1118G, T1135A, T1196A, D1226G, and D1226N, as well asN1114A and S1197A (two residues shown previously to selectivelyeffect cyclodehydration; ref. 10) were purified and assayed.The bacillamide synthetase mutants had radically different effects

on condensation and cyclodehydration (Fig. 3 and SI Appendix, Fig.S3 and SI Results and Discussion). Mutant S1197A had almost noeffect on production of linear or heterocyclized tripeptide. WithY859F, bacillamide E production decreased to 79% of wild type,but linear Ala-Cys-Tpm production doubled. Although the hydroxylof Y859 points directly into the heart of the active site, the relativelyminor effect of its removal suggests it is unlikely to act chemically.Mutants N1114A and T1116A each drastically reduced synthesis ofboth the linear and heterocyclic product. T1116A maintained ap-proximately the same ratio of heterocyclic to linear product as wildtype, whereas the little product made by N1114A was successfullyheterocyclized, consistent with previously reported decoupling ofcondensation and heterocyclization by this residue (10). BothN1114 and T1116 thus appear important for condensation, andN1114 is also important for cyclodehydration.For insight into the cyclization reaction, the most interesting

mutations are those that selectively affect cyclodehydration. Mu-tation of residues F1118, T1135, T1196, and D1226 resulted insubstantially more linear Ala-Cys-Tpm and less bacillamide E thanthe wild-type enzyme. F1118 is positioned along the tunnel thatthe phosphopantetheine arm of donor PCP1 occupies when pre-senting the Ala substrate to the active site (SI Appendix, Fig. S5).In the structure of BmdB-Cy2, the F1118 side chain completelyblocks that tunnel. This residue must move to allow alanyl-PCP1to bind and participate in the condensation reaction. However, forcyclization, PCP1 likely departs, leaving the cyclization substrateAla-Cys-PCP2 bound at the acceptor site. It is possible that F1118aids cyclization by blocking the donor side of the tunnel to create asingle-entry active site more optimal for cyclodehydration. Thisfeature is reminiscent of the (permanent) blocking of the acceptorsite by a tryptophan side chain to form a single-entry active site inrelated epimerization domains, observed in a recently publishedcrystal structure (25). Even more striking are the effects of theT1196A mutation, which allowed only minimal cyclization, andmutations of D1226 to glycine or asparagine, which obliteratedcyclization while retaining robust condensation. These two resi-dues are adjacent to one another, oriented toward the active sitecavity, and hydrogen bond to one another. Their position andimportance for cyclization activity make them the most likely toplay a direct role in catalysis (see below).

Bioinformatic Analysis of Cy Domains.We undertook a bioinformaticanalysis of Cy domains. We retrieved 36,853 C domain super-family sequences that had a maximum pairwise sequence identityof 90% and sorted them by using the Natural Products DomainSeeker web server (36). Multiple sequence alignment of 1,790 Cydomains showed two motifs in the C-terminal region that standout as highly conserved in Cy domains but absent from the other Cdomain superfamily proteins (SI Appendix, Fig. S6). These motifsbear the consensus sequences PVVFTS and SQTPQVxLD (Fig.4A) and are part of what had been recognized by Konz et al. (6) asconserved signature sequences 6 and 7. They exist in BmdB-Cy2 asPIVFTS (1192–1197) and ARTPQVYLD (1218–1226). Thesemotifs are as close as 9 Å in space to the DxxxxD, but they areseparated from it in sequence by more than 200 amino acids. The

0

200

400

600

800

1000

12 13 14 15 16 17

A28

0 (m

Au

)

Time (min)

1

23

A B

C

-ATP

Wild

type

S1197A

Y859F

T1116A

N1114A

T1135A

F1118G

T1196A

D1226G

D1226N0

5

10

15

2 Ala-Cys-Tpm

1 Bacillamide E

Pe

ak

Are

a (

Au

x m

in)

317.121+

335.121+0.5

1.0

1.5

Inte

ns.

x1

08

300 400 500200 m/z

1

HN

HN

OSH

O

NH2

HN

HN

NH

O

N S

NH2

335.011+

372.971+

0.5

1.0

1.5

2.0

200 300 400 500m/z

2

Inte

ns.

x1

07

Fig. 3. Effects of structure-guided mutations in BmdB-Cy2 on bacillamide Esynthesis. (A) Representative HPLC trace of a BmdB activity assay. Compound1 is bacillamide E, compound 2 is linear Ala-Cys-Tpm, and compound 3 is Cys-Tpm. (B) Representative mass spectra of 1 (blue) and 2 (red). A representa-tive mass spectrum of 3 is shown in SI Appendix, Fig. S3A. (C) Quantificationof the relative production of 1 (blue) and 2 (red) in reactions with mutantBmdB. All reactions were measured in triplicate, except F1118G, which wasmeasured in duplicate. Error bars represent SD.

Bloudoff et al. PNAS | January 3, 2017 | vol. 114 | no. 1 | 97

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

May

15,

202

0

motifs form a surface on the acceptor substrate side of the activesite (Fig. 4B and SI Appendix, Fig. S6A). Interestingly, this region isthe putative location of the Ala-Cys-phosphopantetheinyl substratein the cyclization reaction and includes the residues that displayedthe most drastic effect on cyclization when mutated, namely T1196of the PVVFTS motif and D1226 of the SQTPQVxLD motif.D1226 is one of the most conserved residues in Cy domains,present in 96% of the 1,790 sequences. T1196 is somewhat morevariable: It is a threonine or serine in 88% of the sequences.Putative functions had not been assigned to these motifs. S1197

was the only residue in either motif previously targeted by muta-tion, and it had differential effects on Cy domain reactivity inEpoB, BacA2, and BmdB, consistent with its moderate conser-vation (Figs. 3 and 4A and SI Appendix, Table S1) (10, 12). No-tably, a portion of Cy domain sequence that largely overlaps withthe two new motifs is assigned PFam08415. It is annotated only asbeing found in NRPSs with C and PCP domains, and ends partwaythrough the SQTPQVxLD motif, before the critical D1226. Wepropose that this motif be extended to residue 1241 (to incorporate

the whole SQTPQVxLD and a conserved WD; SI Appendix, Fig.S6A) and annotated as a signature for Cy domains.

A Trend for Tandem Cy Domains in (Methyl)oxazoline-Forming Modules.Cy domains can be subdivided by their use of thiol (Cys) or hy-droxyl (Ser or Thr) groups as the nucleophile in the cyclizationreaction. Each subset contains the highly conserved PVVFTS andSQTPQVxLD motifs, and there are no discernable differencesbetween Cy domain sequences in each subset. However, one cleartrend did emerge: In general, modules that incorporate Cys sub-strates contain only one Cy domain per module, whereas thoseusing Thr or Ser have tandem Cy-Cy (or Cy-C) domains. Of a set of505 full-length Cy domain-containing proteins, 454 are predicted touse Cys (440 with single Cy domains, 14 with Cy-Cy domains),whereas 51 are predicted to use Ser or Thr (2 with a single Cydomain, 49 with Cy-Cy domains). This trend is maintained after theproteins have been filtered for unique architecture and/or uniquepredicted product: Of 200 remaining proteins, 187 use Cys (177with a single Cy domain, 10 with Cy-Cy domains) and 13 use Ser orThr (1 with single a Cy domain, 12 with Cy-Cy domains).The trend of tandem Cy domains for hydroxyl-containing sub-

strates and single Cy domains for thiol-containing substrates holdsfor most characterized NRPS systems that feature Cy domains (6–8, 10, 12, 22, 23, 37–41). The tandem Cy domain arrangement isexemplified by the well-characterized VibF protein. In VibF, Cy2is primarily responsible for condensation and Cy1 for cyclo-dehydration of the Thr substrate (22). Furthermore, the tandemCy arrangement can be maintained even when the module is splitbetween two proteins, as in serratiochelin synthetase SchF1 andSchF2, which also cyclodehydrates Thr (40, 41). Exceptions to thetrend are mycobactin synthetase (42) (Thr-specific module with asingle Cy domain in MbtB) and anguibactin synthetase (Cys-spe-cific module with Cy-Cy domains in AngN) (27). The tandem Cydomains in anguibactin synthetase could be a remnant of evolu-tion from a Thr-using ancestor, because, other than containing aCys in place of a Thr, actinomycin is identical to acinetobactin,and the two synthetases share identical domain configurations(43). Furthermore, the Cy domains in AngN can each performboth condensation and cyclodehydration reactions and are largelyredundant with one another (27).The Cys thiol is a better nucleophile than the Ser or Thr hydroxyl

(44). This difference in reactivity may be why thiazoline-formingNRPS modules are ∼10× more prevalent than (methyl)oxazoline-forming modules and may explain the trend to dedicate two do-mains in tandem in (methyl)oxazoline-forming modules. Tandemdomains may increase the probability of cyclodehydration beforethe peptide is donated in the downstream module’s condensationreaction. The downstream C domains are likely not completelyselective for the heterocycle-containing peptide, as shown withBmdB (Fig. 3) and VibF (28), so efficiency of cyclodehydrationwould be important for cognate heterocycle production.

Insight into Catalysis and Model of the Cyclodehydration Intermediate.The first reaction performed by Cy domains is condensation. TheN1114A and T1116A mutants of BmdB-Cy2 exhibited the greatesteffects on condensation without fully abolishing it (Fig. 3), andneither residue is highly conserved (SI Appendix, Fig. S6A). Like-wise, all mutations that were reported to decrease condensation inCy domains are shown by the BmdB-Cy2 structure to be too distalto (residues 900, 988, 1089, 1114, 1120) or turned away from (res-idues 980, 985 of the DxxxxD motif) the atoms participating incondensation (SI Appendix, Table S1) (10, 12, 23, 27, 28). Unless theDxxxxD motif radically reorients (which is possible, although thereis no supporting data), there do not appear to be any critical,conserved residues that could act to abstract or donate a proton inthe condensation reaction in Cy domains. We have argued thatsubstrate orientation is the principal source of catalytic power forcondensation in C domains (21), and this hypothesis appears to be

* *

- BmdB-Cy2

- Cy domains

D1226

T1196

V12283Å

3Å

3Å 3Å

Fig. 4. Model of the cyclodehydration intermediate and critical residues forcyclodehydration reaction. (A) WebLogo3 (57) of Cy domain motifs PVVFTS(core Cy6) and SQTPQVxLD (part of core Cy7) compared with the sequence ofBmdB-Cy2. Putative catalytic residues T1196 and D1226 are labeled withyellow asterisks. (B) BmdB-Cy2 (mostly green, with DxxxxD in orange,PVVFTS in purple, and SQTPQVxLD in brown) with a model of the cyclo-dehydration intermediate (blue). Putative catalytic residues T1196 andD1226 are shown in sticks, as is V1228, a position occupied by glutamine inmost Cy domains. (C) Possible mechanism of the dehydration step. Wesuggest that deprotonation of the amino hydrogen occurs after dehydrationand double bond formation. The full putative reaction pathway is dia-gramed in SI Appendix, Fig. S7.

98 | www.pnas.org/cgi/doi/10.1073/pnas.1614191114 Bloudoff et al.

Dow

nloa

ded

by g

uest

on

May

15,

202

0

even more likely in Cy domains. Comparison of the kinetic pa-rameters of uncatalyzed peptide bond formation and ribosome-catalyzed peptide bond formation shows that the rate enhancementprovided by the ribosome comes predominantly from the loweringof activation entropy via positioning of ester substrates (and perhapswater molecules) for reaction (45). Thus, Cy domains should beable to rely principally on substrate positioning to catalyze con-densation of their more reactive thioester substrates.The second reaction performed by Cy domains, cyclodehydration,

is a nonsimplistic, two-step reaction. Cy domains and at least twoother types of enzymes perform this reaction to form five-memberedrings in independent natural product biosynthesis systems. YcaO-and TruD-type enzymes processively modify ribosomally synthe-sized and posttranslationally modified peptides by using cyclo-dehydration reactions to introduce thiazoline and (methyl)oxazolinerings in the synthesis of peptides such as microcin and the trun-kamides (46). YcaO and TruD, which are structurally unrelatedto Cy domains, covalently attach a phosphate or adenylate fromATP to the carbonyl oxygen to promote both cyclization and de-hydration (46). In Cy domains, which do not use ATP, the ener-getics of cyclodehydration is presumably linked to the high-energythioester bond that is broken in the subsequent step of bacillamidesynthesis. Cy domains share this characteristic with the newly de-scribed specialty C domain, NocB-C5, that catalyzes β-lactam ringformation (47). NocB-C5 is proposed to dehydrate a serine toβ-alanine via base catalysis by a histidine directly upstream of itsHHxxxD motif, prior to cyclization. Dehydration before cycliza-tion is the reaction order shared with lantibiotic cyclodehydrata-ses, but not with Cy domains (48). The position of the upstreamhistidine is occupied by a hydrophobic residue in Cy domains, andthis V979 in BmdB-Cy2 is somewhat recessed from the active site.Thus, Cy domains rely on a completely different mechanismfor catalysis than YcaO, TruD, NcoB-C5, or the lantibioticcyclodehydratases.To help integrate our structural, mutagenic, and conservation

data, we created a model of the intermediate of the BmdB-cata-lyzed cyclodehydration reaction (Fig. 4B). The configuration ofthe model and the present data are consistent with the absolutelycritical catalytic action of D1226 in the first step of the cyclo-dehydration reaction, orienting and abstracting a proton from thethiol (or hydroxyl in Ser- or Thr-specific Cy domains) (SI Ap-pendix, Fig. S7). T1196, 3.0 Å from D1226, appears well positionedto aid in catalysis by interacting with D1226, and may donate aproton to the former carbonyl oxygen to form the hydroxyl-thia-zolidine. D1226 may protonate the same oxygen again to allow itto leave as water in the dehydration reaction (Fig. 4 B and C andSI Appendix, Fig. S7). Upon dehydration, the pKa of the aminoproton is lowered such that it can be facilely lost to solvent toproduce the final thiazoline moiety.Our mutational results indicate that D1226 is critical for cyclo-

dehydration, and T1196 is nearly as important (Fig. 3A). However,T1196 is only a threonine or serine in 88% of Cy sequences, andappears as an alanine in 7% of cases, including in the functionallycharacterized PchE-Cy1 (39). How can its apparent importance inthe reaction and its relative lack of conservation be reconciled?Homology modeling of PchE-Cy1 onto BmdB-Cy2 provides apotential answer: The residue at the nearby position equivalent to1228 of PchE-Cy1 is a glutamine, which could compensate for themissing T1196. A double mutant of T1196A/V1228Q did not re-store cyclization activity to BmdB (SI Appendix, Fig. S8), but is notsurprising given the BmdB-Cy has evolved to perform cyclo-dehydration with T1196 and not Q1228. In the analyzed Cy domainsequences that have an alanine at position 1196, all but two have aglutamine at 1228 (with the remaining two having glutamate orasparagine). In the majority of Cy domains, all three residues arepresent to form a T-D-Q triad. A similar S/T-D-Q catalytic triad inEscherichia coli thioesterase II activates water for nucleophilic at-tack (49). We suggest that mutation of the threonine or glutamine

in the Cy domain triad may be tolerated in some instances, becausechemistry is not rate-limiting in the overall slow synthetic cycle ofNRPSs, and at least in thiazoline-forming Cy domains, it would notbe difficult to deprotonate the Cys side chain (pKa of ∼8.2 inaqueous solution). Deprotonation of Ser and Thr side chains ismore difficult and could require the full catalytic triad. Variation incatalytic residues is not abnormal in nature, as exemplified byserine protease active sites (50). Analogous to the observed vari-ation of T1196, thioesterase II enzymes feature hydrophobic resi-dues in place of the catalytic serine/threonine in more than 5%of proteins.Finally, parallels can be drawn between the role of D1226 in

deprotonating the substrate side chain for nucleophilic attackwith the mechanisms of microbial transglutaminase, cytosolicphospholipase A2, patatin, and TEM-1 β-lactamase, in which theaspartate of a C-D or S-D dyad deprotonates the cysteine orserine for nucleophilic attack on a carbonyl carbon (51–53).During the review process for this study, a structure and muta-

tional analysis of the heterocyclization domain of epothilone syn-thase (EpoB-Cy) was reported by Dowling et al. (54). The structuresof EpoB-Cy and BmdB-Cy2 are similar. Their mutational data areconsistent with those presented here, in particular the identificationof D1226 (D449 in EpoB-Cy) as important for catalysis, althoughwithout differentiation between defects in condensation and het-erocyclization (SI Appendix, SI Results and Discussion and TableS1). Their study and ours complement one another and provide agreater understanding of NRPS heterocyclization domains.

ConclusionIn summary, we have presented the crystal structure of an NRPSheterocyclization domain, solved to a resolution of 2.3 Å.We cloned,expressed, and purified the entire bacillamide synthetase containingthis Cy domain, and used it to assay effects of Cy domain mutationson peptide production. We were able to identify two residues, T1196and D1226, which are important for the cyclodehydration reaction.Finally, we presented a putative mechanism for cyclodehydration inwhich D1226 acts as general acid/base catalyst.

MethodsBacillamide Synthetase Cy2 Crystallography. The Thermoactinomyces vulgarisBmdB-Cy2 construct, with an N-terminal octahistidine tag, tobacco etch virus(TEV) protease cleavage site, and BmdB residues 844–1287, was synthesizedby DNA 2.0. BmdB-Cy2 was heterologously expressed in E. coli and purifiedto homogeneity (SI Appendix, SI Methods). BmdB-Cy2 was crystallized byusing 0.88% Tween 20, 1.62 M ammonium sulfate, 0.1 M Hepes pH 7.5,2.67% (wt/vol) PEG 400, and 3% (wt/vol) 6-aminohexanoic acid, with 20%(vol/vol) ethylene glycol, added before flash-cooling. Diffraction data werecollected at CLS beamline 08ID-1 (SI Appendix, Table S2) and the structure ofBmdB-Cy2 was determined by molecular replacement.

Peptide Synthesis Assay for Full-Length BmdB. bmdB was cloned from T. vulgarisF-5595 genomic DNA into a pET21-derived vector containing an N-terminal TEV-cleavable calmodulin binding peptide tag and a C-terminal TEV-cleavable octa-histidine tag. Mutations were introduced via site-directed mutagenesis (SI Ap-pendix, Table S3). Wild-type and mutant protein was expressed in E. coli cellsand purified to homogeneity. In a 1-mL reaction, 50 mM Tris·HCl pH 7.5,100 mM NaCl, and 10 mM MgCl2, 1 mM Ala, 1 mM Cys, 10 mM tryptamine,2 mM ATP, and 100 nM BmdB were incubated for 2 h at 37 °C. Reactions wereanalyzed by HPLC using C18 media and a gradient of water/0.1% TFA toacetonitrile/0.1% TFA. High-resolution MS and fragmentation MS analysis wasperformed at the Proteomics Platform at the Research Institute of the McGillUniversity Health Centre, and NMR analysis at Quebec/Eastern Canada HighField NMR Facility (QANUC), McGill University.

Bioinformatic Analysis. Sequences of known C domains were queried againstthe nr90 database (55), using low threshold search. Matched sequences werethemselves filtered for maximum pairwise sequence identity of 90%, whichgave 36,853 C domain superfamily sequences. These sequences were classi-fied into subtypes by using NaPDoS (36), and included 1,790 Cy domains. Tolook for trends in Cy domains by substrates, all full-length protein sequencesfrom which the 1,790 Cy domains originate were retrieved. Sequences were

Bloudoff et al. PNAS | January 3, 2017 | vol. 114 | no. 1 | 99

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

May

15,

202

0

discarded if there was not an A domain adjacent to the Cy domain, or if theystarted as a single Cy domain leaving a set of 505 proteins. The Cy domainacceptor/cyclodehydration substrates were inferred from substrate of the Adomains in the same module, as predicted by the program ANTISMASH (56).

ACKNOWLEDGMENTS. We thank Chaitan Khosla and James Kuo for thekind gift of BAP1 cells; Vikram Alva and Johannes Soeding for providing

the current nr90 database; Christopher Boddy and Adrian Keatinge-Clayfor conversations; Asfarul Haque for performing the ITC experiments;Janice Reimer for schematic figure design; the Proteomics Platform atthe RI-MUHC (Montreal, Canada) for LC-MS analysis; members of theT.M.S. laboratory for helpful discussions; Alexander Wahba for MS analy-sis; Varoujan Yaylayan for fragmentation MS assignment; and Robin Steinfor NMR analysis.

1. Weissman KJ (2015) The structural biology of biosynthetic megaenzymes. Nat ChemBiol 11(9):660–670.

2. Keating TA, Marshall CG, Walsh CT, Keating AE (2002) The structure of VibH repre-sents nonribosomal peptide synthetase condensation, cyclization and epimerizationdomains. Nat Struct Biol 9(7):522–526.

3. Samel SA, Schoenafinger G, Knappe TA, Marahiel MA, Essen LO (2007) Structural andfunctional insights into a peptide bond-forming bidomain from a nonribosomalpeptide synthetase. Structure 15(7):781–792.

4. Bergendahl V, Linne U, Marahiel MA (2002) Mutational analysis of the C-domain innonribosomal peptide synthesis. Eur J Biochem 269(2):620–629.

5. Bloudoff K, Rodionov D, Schmeing TM (2013) Crystal structures of the first conden-sation domain of CDA synthetase suggest conformational changes during the syn-thetic cycle of nonribosomal peptide synthetases. J Mol Biol 425(17):3137–3150.

6. Konz D, Klens A, Schörgendorfer K, Marahiel MA (1997) The bacitracin biosynthesisoperon of Bacillus licheniformis ATCC 10716: Molecular characterization of threemulti-modular peptide synthetases. Chem Biol 4(12):927–937.

7. Marshall CG, Burkart MD, Keating TA, Walsh CT (2001) Heterocycle formation in vi-briobactin biosynthesis: Alternative substrate utilization and identification of a con-densed intermediate. Biochemistry 40(35):10655–10663.

8. Chen H, O’Connor S, Cane DE, Walsh CT (2001) Epothilone biosynthesis: Assembly ofthe methylthiazolylcarboxy starter unit on the EpoB subunit. Chem Biol 8(9):899–912.

9. Keating TA, Marshall CG, Walsh CT (2000) Reconstitution and characterization of theVibrio cholerae vibriobactin synthetase from VibB, VibE, VibF, and VibH. Biochemistry39(50):15522–15530.

10. Duerfahrt T, Eppelmann K, Müller R, Marahiel MA (2004) Rational design of a bi-modular model system for the investigation of heterocyclization in nonribosomalpeptide biosynthesis. Chem Biol 11(2):261–271.

11. Gehring AM, Mori I, Perry RD, Walsh CT (1998) The nonribosomal peptide synthetaseHMWP2 forms a thiazoline ring during biogenesis of yersiniabactin, an iron-chelatingvirulence factor of yersinia pestis. Biochemistry 37(48):17104.

12. Kelly WL, Hillson NJ, Walsh CT (2005) Excision of the epothilone synthetase B cycli-zation domain and demonstration of in trans condensation/cyclodehydration activity.Biochemistry 44(40):13385–13393.

13. Tabata N, Tomoda H, Zhang H, Uchida R, Omura S (1999) Zelkovamycin, a new cyclicpeptide antibiotic from Streptomyces sp. K96-0670. II. Structure elucidation. J Antibiot(Tokyo) 52(1):34–39.

14. Shen B, et al. (2002) Cloning and characterization of the bleomycin biosynthetic genecluster from Streptomyces verticillus ATCC15003. J Nat Prod 65(3):422–431.

15. Vollbrecht L, et al. (2002) Argyrins, immunosuppressive cyclic peptides from myxobac-teria. II. Structure elucidation and stereochemistry. J Antibiot (Tokyo) 55(8):715–721.

16. Snow GA (1970) Mycobactins: Iron-chelating growth factors from mycobacteria.Bacteriol Rev 34(2):99–125.

17. Drechsel H, et al. (1995) Structure elucidation of Yersiniabactin, a siderophore fromhighly virulent yersinia strains. Liebigs Ann 1995(10):1727–1733.

18. Vizcaino MI, Crawford JM (2015) The colibactin warhead crosslinks DNA. Nat Chem7(5):411–417.

19. Roy RS, Gehring AM, Milne JC, Belshaw PJ, Walsh CT (1999) Thiazole and oxazolepeptides: Biosynthesis and molecular machinery. Nat Prod Rep 16(2):249–263.

20. Rausch C, Hoof I, Weber T, Wohlleben W, Huson DH (2007) Phylogenetic analysis ofcondensation domains in NRPS sheds light on their functional evolution. BMC EvolBiol 7:78.

21. Bloudoff K, Alonzo DA, Schmeing TM (2016) Chemical probes allow structural insightinto the condensation reaction of nonribosomal peptide synthetases. Cell Chem Biol23(3):331–339.

22. Marshall CG, Hillson NJ, Walsh CT (2002) Catalytic mapping of the vibriobactin bio-synthetic enzyme VibF. Biochemistry 41(1):244–250.

23. Keating TA, Miller DA, Walsh CT (2000) Expression, purification, and characterizationof HMWP2, a 229 kDa, six domain protein subunit of Yersiniabactin synthetase.Biochemistry 39(16):4729–4739.

24. Patel HM, Walsh CT (2001) In vitro reconstitution of the Pseudomonas aeruginosanonribosomal peptide synthesis of pyochelin: Characterization of backbone tai-loring thiazoline reductase and N-methyltransferase activities. Biochemistry 40(30):9023–9031.

25. Chen WH, Li K, Guntaka NS, Bruner SD (2016) Interdomain and intermodule orga-nization in epimerization domain containing nonribosomal peptide synthetases. ACSChem Biol 11(8):2293–2303.

26. Samel SA, Czodrowski P, Essen LO (2014) Structure of the epimerization domain oftyrocidine synthetase A. Acta Crystallogr D Biol Crystallogr 70(Pt 5):1442–1452.

27. Di Lorenzo M, Stork M, Naka H, Tolmasky ME, Crosa JH (2008) Tandem hetero-cyclization domains in a nonribosomal peptide synthetase essential for siderophorebiosynthesis in Vibrio anguillarum. Biometals 21(6):635–648.

28. Marshall CG, Burkart MD, Meray RK, Walsh CT (2002) Carrier protein recognition insiderophore-producing nonribosomal peptide synthetases. Biochemistry 41(26):8429–8437.

29. Jeong SY, Ishida K, Ito Y, Okada S, Murakami M (2003) Bacillamide, a novel algicidefrom the marine bacterium, Bacillus sp SY-1, against the harmful dinoflagellate,Cochlodinium polykrikoides. Tetrahedron Lett 44(43):8005–8007.

30. Churro C, et al. (2009) Effects of bacillamide and newly synthesized derivatives on thegrowth of cyanobacteria and microalgae cultures. J Appl Phycol 21(4):429–442.

31. Drake EJ, et al. (2016) Structures of two distinct conformations of holo-non-ribosomalpeptide synthetases. Nature 529(7585):235–238.

32. Roche ED, Walsh CT (2003) Dissection of the EntF condensation domain boundary andactive site residues in nonribosomal peptide synthesis. Biochemistry 42(5):1334–1344.

33. Yuwen L, et al. (2013) The role of aromatic L-amino acid decarboxylase in bacillamideC biosynthesis by Bacillus atrophaeus C89. Sci Rep 3:1753.

34. Ivanova V, et al. (2007) Microbiaeratin, a new natural indole alkaloid from a Micro-bispora aerata strain, isolated from Livingston Island, Antarctica. Prep BiochemBiotechnol 37(2):161–168.

35. Socha AM, Long RA, Rowley DC (2007) Bacillamides from a hypersaline microbial matbacterium. J Nat Prod 70(11):1793–1795.

36. Ziemert N, et al. (2012) The natural product domain seeker NaPDoS: A phylogenybased bioinformatic tool to classify secondary metabolite gene diversity. PLoS One7(3):e34064.

37. Du L, Chen M, Zhang Y, Shen B (2003) BlmIII and BlmIV nonribosomal peptide syn-thetase-catalyzed biosynthesis of the bleomycin bithiazole moiety involving both incis and in trans aminoacylation. Biochemistry 42(32):9731–9740.

38. Silakowski B, et al. (1999) New lessons for combinatorial biosynthesis frommyxobacteria.The myxothiazol biosynthetic gene cluster of Stigmatella aurantiaca DW4/3-1. J BiolChem 274(52):37391–37399.

39. Quadri LE, Keating TA, Patel HM, Walsh CT (1999) Assembly of the Pseudomonasaeruginosa nonribosomal peptide siderophore pyochelin: In vitro reconstitution ofaryl-4, 2-bisthiazoline synthetase activity from PchD, PchE, and PchF. Biochemistry38(45):14941–14954.

40. Seyedsayamdost MR, et al. (2012) Mixing and matching siderophore clusters: Struc-ture and biosynthesis of serratiochelins from Serratia sp. V4. J Am Chem Soc 134(33):13550–13553.

41. Wang H, Fewer DP, Holm L, Rouhiainen L, Sivonen K (2014) Atlas of nonribosomalpeptide and polyketide biosynthetic pathways reveals common occurrence of non-modular enzymes. Proc Natl Acad Sci USA 111(25):9259–9264.

42. McMahon MD, Rush JS, Thomas MG (2012) Analyses of MbtB, MbtE, and MbtF sug-gest revisions to the mycobactin biosynthesis pathway in Mycobacterium tuberculosis.J Bacteriol 194(11):2809–2818.

43. Dorsey CW, et al. (2004) The siderophore-mediated iron acquisition systems of Aci-netobacter baumannii ATCC 19606 and Vibrio anguillarum 775 are structurally andfunctionally related. Microbiology 150(Pt 11):3657–3667.

44. Belshaw PJ, Roy RS, Kelleher NL, Walsh CT (1998) Kinetics and regioselectivity ofpeptide-to-heterocycle conversions by microcin B17 synthetase. Chem Biol 5(7):373–384.

45. Beringer M, et al. (2005) Essential mechanisms in the catalysis of peptide bond for-mation on the ribosome. J Biol Chem 280(43):36065–36072.

46. Dunbar KL, Melby JO, Mitchell DA (2012) YcaO domains use ATP to activate amidebackbones during peptide cyclodehydrations. Nat Chem Biol 8(6):569–575.

47. Gaudelli NM, Long DH, Townsend CA (2015) β-Lactam formation by a non-ribosomalpeptide synthetase during antibiotic biosynthesis. Nature 520(7547):383–387.

48. Zhang Q, Yu Y, Vélasquez JE, van der Donk WA (2012) Evolution of lanthipeptidesynthetases. Proc Natl Acad Sci USA 109(45):18361–18366.

49. Li J, Derewenda U, Dauter Z, Smith S, Derewenda ZS (2000) Crystal structure of theEscherichia coli thioesterase II, a homolog of the human Nef binding enzyme. NatStruct Biol 7(7):555–559.

50. Ekici OD, Paetzel M, Dalbey RE (2008) Unconventional serine proteases: Variations onthe catalytic Ser/His/Asp triad configuration. Protein Sci 17(12):2023–2037.

51. Dessen A, et al. (1999) Crystal structure of human cytosolic phospholipase A2 reveals anovel topology and catalytic mechanism. Cell 97(3):349–360.

52. Rydel TJ, et al. (2003) The crystal structure, mutagenesis, and activity studies revealthat patatin is a lipid acyl hydrolase with a Ser-Asp catalytic dyad. Biochemistry 42(22):6696–6708.

53. Minasov G, Wang X, Shoichet BK (2002) An ultrahigh resolution structure of TEM-1beta-lactamase suggests a role for Glu166 as the general base in acylation. J Am ChemSoc 124(19):5333–5340.

54. Dowling DP, et al. (2016) Structural elements of an NRPS cyclization domain and itsintermodule docking domain. Proc Natl Acad Sci USA 113(44):12432–12437.

55. Biegert A, Söding J (2009) Sequence context-specific profiles for homology searching.Proc Natl Acad Sci USA 106(10):3770–3775.

56. Weber T, et al. (2015) antiSMASH 3.0-a comprehensive resource for the genomemining of biosynthetic gene clusters. Nucleic Acids Res 43(W1):W237-43.

57. Crooks GE, Hon G, Chandonia JM, Brenner SE (2004) WebLogo: A sequence logogenerator. Genome Res 14(6):1188–1190.

100 | www.pnas.org/cgi/doi/10.1073/pnas.1614191114 Bloudoff et al.

Dow

nloa

ded

by g

uest

on

May

15,

202

0