Correction
25 Jul 2014: The PLOS ONE Staff (2014) Correction: Improved Lanthipeptide Detection and Prediction for antiSMASH. PLOS ONE 9(7): e103665. https://doi.org/10.1371/journal.pone.0103665 View correction
Figures
Abstract
Lanthipeptides are a class of ribosomally synthesised and post-translationally modified peptide (RiPP) natural products from the bacterial secondary metabolism. Their name is derived from the characteristic lanthionine or methyl-lanthionine residues contained in the processed peptide. Lanthipeptides that possess an antibacterial activity are called lantibiotics. Whereas multiple tools exist to identify lanthipeptide gene clusters from genomic data, no programs are available to predict the post-translational modifications of lanthipeptides, such as the proteolytic cleavage of the leader peptide part or tailoring modifications based on the analysis of the gene cluster sequence. antiSMASH is a software pipeline for the identification of secondary metabolite biosynthetic clusters from genomic input and the prediction of products produced by the identified clusters. Here we present a novel antiSMASH module using a rule-based approach to combine signature motifs for biosynthetic enzymes and lanthipeptide-specific cleavage site motifs to identify lanthipeptide clusters in genomic data, assign the specific lanthipeptide class, predict prepeptide cleavage, tailoring reactions, and the processed molecular weight of the mature peptide products.
Citation: Blin K, Kazempour D, Wohlleben W, Weber T (2014) Improved Lanthipeptide Detection and Prediction for antiSMASH. PLoS ONE 9(2): e89420. https://doi.org/10.1371/journal.pone.0089420
Editor: Eugene A. Permyakov, Russian Academy of Sciences, Institute for Biological Instrumentation, Russian Federation
Received: October 2, 2013; Accepted: January 21, 2014; Published: February 20, 2014
Copyright: © 2014 Blin et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding was provided by the German Ministry of Education and Research (BMBF) (Grant 0315585A to TW), German Centre for Infection Research (DZIF) (8000-402-2 to TW and WW), LAPTOP (Grant 245066 to WW). Funding for open access charge: Deutsche Forschungsgemeinschaft (DFG) and Open Access Publishing Fund of Tübingen University. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Lanthipeptides
Lanthipeptides are polycyclic peptides named after the thioether-linked amino acids lanthionine and (2 S,3 S,6 R)-3-methyllanthionine contained in the mature peptide. Formerly called lantibiotics from “lanthionine-containing antibiotics”, the new name lanthipeptide was proposed to also include non-antibiotic peptides of the same biosynthetic origin [1]. The best-known lanthipeptide is Nisin, which has been used as a food preservative since the 1940 s. Still a hot spot of natural products research, other lanthipeptides, for example NVB302, entered phase 1 clinical trials in 2011 as a treatment against Clostridium difficile infections [2]. Lanthipeptides are ribosomally synthesised and post-translationally modified peptides (RiPPs). The extensive post-translational modifications enhance the stability of the mature peptide against proteolysis and temperature stress. Lanthipeptides are encoded on the genome as a precursor peptide containing a leader and a core peptide part. The leader peptide, which is removed at a very late stage of the biosynthesis by a cluster-encoded or host-encoded protease, serves as a docking site for the modifying enzymes acting on the core peptide [3]. The lanthionine (Lan) and methyllanthionine (MeLan) residues are introduced in a two-step reaction. First, serine (Ser) and threonine (Thr) residues are dehydrated to dehydroalanine (Dha) and dehydrobutyrine (Dhb), respectively, usually with an intermediate phosphorylation step. In the second step, a Michael-type addition by cysteine (Cys) residues onto the dehydro amino acids then yields the thioether cross-links.
Depending on the biosynthetic enzymes installing the thioether cross-links, lanthipeptides are divided into different classes [4]. Currently, four lanthipeptide classes are known. In class I lanthipeptides, the dehydration is catalysed by a dedicated dehydratase commonly called LanB. Cyclisation is carried out by a cyclase called LanC. In specific gene clusters, the generic enzyme names might be replaced by a more specific name: for example in the nisin gene cluster, the LanB-type dehydratase is called NisB and the LanC-type cyclase is called NisC. For the remaining class II, III and IV lanthipeptides, both dehydration and cyclisation are catalysed by a single bi-functional enzyme. A class II LanM enzyme carries an N-terminal dehydratase domain with little sequence similarity to other characterised enzymes. The C-terminal cyclisation domain is similar to the LanC enzymes from class I lanthipeptide cyclases. The bi-functional enzymes for class III (LanKC) and IV (LanL) have a common N-terminal phospho- serine/phosphothreonine lyase domain and a central kinase domain. The C-terminal cyclisation domain in class III enzymes, while similar to the cyclisation domains from the other classes, lacks three zinc- bindinding residues that are conserved in the other classes. In class IV, those residues are present. In addition to the introduction of Lan and MeLan, a number of further post-translational modifications may occur if the appropriate tailoring enzymes are present in the gene cluster. Among the modifications found in lanthipeptides is the formation of S-[(Z)-2-aminovinyl]-D-cysteine (AviCys) or S-[(Z)-2-aminovinyl]-(3 S)-3-methyl-D-cysteine (AviMeCys) [5]. The formation of AviCys and AviMeCys is catalysed by an enzyme of the family of homo-oligomeric flavin-containing cysteine decarboxylases.The enzyme with the generic designation LanD catalyses the oxidative decarboxylation of a C-terminal cysteine residue to a reactive thio–enol intermediate, which then cyclises with a Dha or Dhb residue, respectively, yielding AviCys or AviMeCys. An example would be the AviCys residue in epidermin introduced by EpiD [6]. Another post-translational modification is the chlorination of tryptophan residues catalysed by a flavin-dependent tryptophan halogenase designated LanH. This kind of reaction has been observed in the chlorination of tryptophan by MibH in microbisporicin biosynthesis [7]. If the cluster contains a cytochrome P450 oxygenase designated LanO, amino acids in the modified precursor peptide can be hydroxylated, as observed in the hydroxylation of proline in microbisporicin biosynthesis [7]. If the N-terminal amino acid is Dha and an oxidoreductase is present in the cluster, the N-terminal amino acid can be converted to lactate, as observed in the epicidin 280 cluster [8].
antiSMASH
antiSMASH, the antibiotics and secondary metabolite analysis shell, is a software pipeline for the automated identification of secondary metabolite biosynthesis clusters in whole genome sequenes of bacteria and fungi [9], [10]. In antiSMASH, the user, who does not need specialized bioinformatics training, can upload microbial genome sequences which then are mined for secondary metabolite biosynthetic pathways in a fully automated manner. In the initial release of antiSMASH, the prediction of the products of the biosynthetic pathways was only possible for non-ribosomal peptide synthase (NRPS) and polyketide synthase (PKS) gene clusters [9]. In 2013, we released antibiotics and secondary metabolite analysis shell (antiSMASH) 2.0 [10]. In the new release, the architecture of the software was redesigned, now making it possible to add new predictors as self-contained plug-ins. antiSMASH is available as a web service at http://antismash.secondarymetabolites.org and can also be downloaded to run standalone. It is released under the GNU Affero Public License version 3, an OSI-approved Open Source license. Here we present the implementation of a lanthipeptide-specific analysis module for antiSMASH 2. With this module, lanthipeptide biosynthetic pathways can be automatically detected, classified and their products predicted. In addition, the module can identify putative protease cleavage sites for the most abundant class I and class II lanthipeptides and predict the molecular masses of the final products. The module is shipped with the antiSMASH 2.2 release and is also running on the public web server.
Design, Implementation and Validation
Secondary metabolite clusters in antiSMASH are identified using Hidden Markov Models (HMMs) of protein motifs for key biosynthetic enzymes. Which profiles are required to be identified for a specific secondary metabolite type is described by a rules file containing one rule-set per cluster type. Rule-sets can be simple hits against a single profile, AND and OR combinations of multiple profiles, or a selection of more complex rules, e.g. requiring a match against a minimum of n hits of a set of profiles. New secondary metabolite types can be included by adding new profile HMMs and extending the rules file. Once the cluster detection has identified a secondary metabolite cluster of a certain type, specific analysis modules can be run to generate a more detailed analysis of the pathway and the prediction of the product of a given cluster. Specific analysis modules are written as self-contained plug-ins that are loaded from the user’s PYTHONPATH at run-time.
Identification of Lanthipeptide Biosynthetic Gene Clusters
To make more detailed cluster information available to the downstream specific analysis module, the cluster detection rules have been extended to include domain-specific Pfam [11] HMMs for the N-terminal domain (PFAM: PF13575) of class II LanM enzymes, the central kinase domain (PFAM: PF00069) of class III and IV enzymes, LanD-type flavin-dependent decarboxylases (PFAM: PF02441), LanH-type flavin-dependent halogenases (PFAM: PF04820), LanO-type cytochrome P450 oxygenases (PFAM: PF00067) and EciO-type short chain dehydrogenases (PFAM: PF00106, PF13561) (see Table 1 for details). Cutoff-values were determined empirically by analysing known sequences with the HMM profiles and manually adjusted to gain optimal sensitivity while keeping false positive hits low.
Prediction of the Lanthipeptide Class
Lanthipeptide classes are assigned by determining the domains present in the biosynthetic enzymes. Characteristic for class I lanthipeptides is the separate LanB enzyme containing the dehydratase domain, so the class prediction checks for a hit against the Lant_dehyd_N (PFAM: PF04737) or Lant_dehyd_C (PFAM: PF04738) domains. The dehydratase domain of class II LanM-type enzymes is characteristic as well, so if the cluster contains this dehydratase domain (PFAM: PF13575), the lanthipeptide will be considered class II. Class III LanKC-type and class IV LanL-type enzymes are identified via the central kinase domain (PFAM: PF00069). To differentiate between class III and IV enzymes, the algorithm checks if the conserved zinc binding sites in the C-terminal cyclase domain are absent (class III) or present (class IV).
Cleavage Site prediction
In the final step in lanthipeptide biosynthesis, a protease cleaves the leader peptide part off the modified precursor peptide to yield the mature peptide. Depending on the class of the lanthipeptide, the cleavage site motives vary widely. In order to predict the cleavage site, we have created a manually curated set of HMMs, one for lanthipeptide classes I and II each (Tables 2, 3). Profiles for the HMMer 2.3.2 software [12] were generated using the command hmmbuild profile.hmm alignment.fa; hmmcalibrate profile.hmm. As the method depends on the size of the seed sequence data set, we decided not to include cleavage site predictions for class III (only six seed sequences available) and class IV (no experimentally verified sequences available) lanthipeptides. Once more seed sequences become available for these two classes, adding cleavage site predictions using the same method will be straightforward.
Monoisotopic Mass, Molecular Weight and Alternative Weights
Once the cleavage site is predicted, both the monoisotopic mass and the average molecular weight are calculated. For the calculation of these numbers it is assumed that all Ser and Thr residues are dehydrated to Dha and Dhb respectively. As a lack of dehydration is frequently observed but the mechanism behind this has not been elucidated, we also calculate alternative weights under the assumption that one up to n Ser or Thr are not dehydrated, where n is the number of Ser and Thr residues in the core peptide subtracted by the number of Cys residues in the core peptide. This upper bound is set to account for the observation that all Cys residues tend to participate in Lan or MeLan bridges with Dha or Dhb residues.
Predicting Tailoring Reactions
Tailoring reactions are not performed by the core biosynthetic enzymes that perform the dehydration and cyclisation but instead by additional enzymes also encoded on the cluster.
AviCys and AviMeCys formation.
The unusual amino acids AviCys and AviMeCys are formed by oxidative decarboxylation of the C-terminal Cys residue. The resulting thio–enol intermediate cyclises with a Dha or Dhb side-chain respectively. This reaction is catalysed by a LanD-type flavin-dependent decarboxylase, identified by a hit against the PFAM PF02441 profile with a score ≥20. The formation of AviCys or AviMeCys reduces the predicted peptide weight by 46 Da.
Halogenation.
LanH-type halogenases chlorinate an amino acid side chain, increasing the predicted peptide weight by 34 Da. These enzymes are identified by a hit against the PFAM PF04820 profile with a score ≥20.
Hydroxylation.
LanO-type cytochrome P450 oxygenases catalyse the regiospecific oxidation of non-activated hydrocarbons. The enzymes are identified by a hit against PFAM PF00067 with a score ≥60. The hydroxylation increases the predicted peptide weight by 16 Da.
Validation and Benchmarking
To validate the robustness of the cleavage site HMM-profiles, we used n-fold cross validation. For a seed alignment of size n, we built n different profiles by including n − 1 sequences, and then checked if a cleavage site was predictable and correct for the left out sequence. A cleavage site was predictable if the profile produced a hit with a score above the threshold. A cleavage site was considered correct if the prediction matched the ungapped seed sequence not used for building the profile. A cleavage site was predicted in all test sequences for both class I and II lanthipeptides. Cleavage sites were correct in all class I test inputs and in 81% of the class II test inputs (Table 4). To benchmark the overall performance of the prediction, we ran a number of lanthipeptide biosynthetic gene clusters through antiSMASH. We checked if the gene cluster was identified, the precursor peptide was detected, and finally the peptide mass was predicted correctly.
Among the clusters run for benchmarking, we included the planosporicin and epilancin 15X clusters. The cleavage sites of both of these lanthipeptides were not part of the seed alignments. For planosporicin, the algorithm predicted a mass of 2193.3 Da (actual mass 2194 Da [13]) and correctly predicted 5 lanthionine bridges. For epilancin 15X, the algorithm predicted a mass of 3171.8 Da (actual mass 3171.7 Da [14]) and correctly predicted 3 lanthionine bridges and the N-terminal lactate.
Results and Discussion
Only few tools are currently available that allow the automated identification of RiPPs. Apart from antiSMASH, there is BAGEL, recently released in version 3 [15]. BAGEL targets a large number of different ribosomally synthesised peptides. For lanthipeptides, BAGEL only predicts the leader peptide and the class, but does not attempt to predict tailoring reactions, number of Lan and MeLan bridges or the molecular weight.
The detection of enzymes responsible for tailoring reactions is central in the prediction of the mature peptide mass. antiSMASH can predict the presence of AviCys residues which are found, for example, in epidermin and microbisporicin. Halogenations and hydroxylations, which occur in microbisporicin biosynthesis are also detected. A remaining issue in mass prediction is that not all Ser and Thr residues are dehydrated in all the peptides, resulting in mass predictions that are 18 Da too low per undehydrated amino acid. antiSMASH assists in detecting the presence of undehydrated residues by providing alternative mass predictions for lanthipeptides that carry more Ser and Thr than Cys residues. Once the tailoring reactions have been predicted, the final step is the prediction of the number of Lan and MeLan bridges. The naïve heuristic of counting the Cys and Ser/Thr residues and then using the smaller number fails if the mature peptide contains an AviCys or AviMeCys residue and needs to be adjusted accordingly.
The prediction details of the lanthipeptide module are included in the standard antiSMASH output. For the HTML output (Figure 1), lanthipeptide class and leader/core peptide split predicted are shown in the “detailed annotation” section of the cluster page. The score of the class prediction, predicted monoisotopic mass and molecular weights, the number of bridges and the identified additional modifications are shown in the “prediction details” sidebar.
To test the accuracy of the antiSMASH lanthipeptide predictions, a benchmark dataset was analyzed and the results evaluated:
The determination of the lanthipeptide class based on the biosynthetic enzymes in the cluster is straightforward, and antiSMASH performed this tasks flawlessly on the benchmark data set (Table 5). Using the advanced heuristic, antiSMASH was able to correctly predict the number or bridges in almost all lanthipeptides of the benchmark data-set. The heuristic only fails if two Cys residues form a disulphide bridge, a rare occurrence observed in e.g. thermophilin 1277 [16]. Unfortunately, the enzyme catalysing the formation of the disulphide bridge is not present in the gene cluster and thus cannot be used to predict a disulphide bridge formation [16]. Predicting the core peptide sequence is more difficult. The cleavage site motif of class II lanthipeptides (Table 3) is relatively uniform, largely consisting in two amino acids with small side chains that are preceded by alternating hydrophobic and hydrophilic residues. In fact, the four class II cleavage sites incorrectly predicted during validation (Table 4) all differed from this pattern and contained a site that more closely matches the motif upstream of the actual cleavage site. Some class II core peptides like mersacidin [17] or lichenicidin A2 [18] lose an additional six amino acids at the N-terminus after the proteolytic cleavage [18], so it seems likely that the predicted cleavage sites may be accurate and some additional enzyme catalyses the N-terminal modifications. Class I leader peptides also carry a short motif of alternating hydrophobic and hydrophilic amino acids, usually called the FNLD motif. The spacer between this motif and the actual cleavage site varies. At position −2 in front of the cleavage site, many leader peptides carry a proline residue (Table 2). During validation, all cleavage sites were predicted correctly (Table 4). Due to the strong signal of the FNLD motif, class I prediction (stability 100%) is even more robust than the class II prediction (stability 81%) with the shorter motif. As a proof of concept, we used the gene clusters of the recently published planosporicin [13] and the lactate-containing epilancin 15X [14]. Both precursor peptides contain cleavage sites that are distinct from all the sequences included in the class I seed alignment. For both lanthipeptides the algorithm is able to correctly predict the mass, number of Lan/MeLan bridges and tailoring modifications.
Conclusions
With the algorithm described in this paper, antiSMASH gains extensive lanthipeptide-specific predictive capabilities. antiSMASH is the only software currently available that will predict lanthipeptide class, core peptide cleavage, tailoring reactions, number of Lan and MeLan bridges, and the molecular weight of the mature peptide product. These informations obtained by analyzing the genome sequence of the producers can give crucial hints for the identification of the lanthipeptide compounds in chemical analytics.
Author Contributions
Conceived and designed the experiments: KB TW. Performed the experiments: KB DK. Analyzed the data: KB DK. Wrote the paper: KB WW TW.
References
- 1. Knerr PJ, van der Donk WA (2012) Discovery, Biosynthesis, and Engineering of Lantipeptides. Annual Review of Biochemistry 81: 1–27.
- 2. Crowther GS, Baines SD, Todhunter SL, Freeman J, Chilton CH, et al. (2013) Evaluation of NVB302 versus vancomycin activity in an in vitro human gut model of Clostridium difficile infection. Journal of Antimicrobial Chemotherapy 68: 168–176.
- 3. Plat A, Kluskens LD, Kuipers A, Rink R, Moll GN (2011) Requirements of the engineered leader peptide of nisin for inducing modification, export, and cleavage. Applied and Environmental Microbiology 77: 604–611.
- 4. Arnison PG, Bibb MJ, Bierbaum G, Bowers AA, Bugni TS, et al. (2013) Ribosomally synthesized and post-translationally modified peptide natural products: overview and recommendations for a universal nomenclature. Natural Product Reports 30: 108–160.
- 5. Sit CS, Yoganathan S, Vederas JC (2011) Biosynthesis of Aminovinyl-Cysteine-Containing Peptides and Its Application in the Production of Potential Drug Candidates. Accounts of Chemical Research 44: 261–268.
- 6. Kellner R, Jung G, Hörner T, Zähner H, Schnell N, et al. (1988) Gallidermin: a new lanthionine-containing polypeptide antibiotic. European Journal of Biochemistry 177: 53–59.
- 7. Foulston LC, Bibb MJ (2010) Microbisporicin gene cluster reveals unusual features of lantibiotic biosynthesis in actinomycetes. Proceedings of the National Academy of Sciences 107: 13461–13466.
- 8. Heidrich C, Pag U, Josten M, Metzger J, Jack RW, et al. (1998) Isolation, characterization, and heterologous expression of the novel lantibiotic epicidin 280 and analysis of its biosynthetic gene cluster. Applied and Environmental Microbiology 64: 3140–3146.
- 9. Medema MH, Blin K, Cimermancic P, de Jager V, Zakrzewski P, et al. (2011) antiSMASH: rapid identification, annotation and analysis of secondary metabolite biosynthesis gene clusters in bacterial and fungal genome sequences. Nucleic Acids Research 39: W339–W346.
- 10. Blin K, Medema MH, Kazempour D, Fischbach MA, Breitling R, et al. (2013) antiSMASH 2.0– a versatile platform for genome mining of secondary metabolite producers. Nucleic Acids Research 41: W204–W212.
- 11. Punta M, Coggill PC, Eberhardt RY, Mistry J, Tate J, et al. (2012) The Pfam protein families database. Nucleic Acids Research 40: D290–D301.
- 12.
Durbin R, Eddy SR, Krogh A, Mitchison G (1998) Biological sequence analysis: probabilistic models of proteins and nucleic acids: Cambridge University Press.
- 13. Sherwood EJ, Bibb MJ (2013) The antibiotic planosporicin coordinates its own production in the actinomycete Planomonospora alba. Proceedings of the National Academy of Sciences 110: E2500–E2509.
- 14. Velasquez JE, Zhang X, van der Donk WA (2011) Biosynthesis of the antimicrobial peptide epilancin 15X and its N-terminal lactate. Chemistry & Biology 18: 857–867.
- 15. van Heel AJ, de Jong A, Montalban-Lopez M, Kok J, Kuipers OP (2013) BAGEL3: automated identification of genes encoding bacteriocins and (non-) bactericidal posttranslationally modified peptides. Nucleic Acids Research 41: W448–W453.
- 16. Kabuki T, Uenishi H, Seto Y, Yoshioka T, Nakajima H (2009) A unique lantibiotic, thermophilin 1277, containing a disulfide bridge and two thioether bridges. Journal of Applied Microbiology 106: 853–862.
- 17. Altena K, Guder A, Cramer C, Bierbaum G (2000) Biosynthesis of the lantibiotic mersacidin: organization of a type B lantibiotic gene cluster. Applied and Environmental Microbiology 66: 2565–2571.
- 18. Dischinger J, Josten M, Szekat C, Sahl H-G, Bierbaum G (2009) Production of the novel two-peptide lantibiotic lichenicidin by Bacillus licheniformis DSM 13. PLoS ONE 4: e6788.
- 19. Birri DJ, Brede DA, Nes IF (2012) Salivaricin D, a novel intrinsically trypsin-resistant lantibiotic from Streptococcus salivarius 5M6c isolated from a healthy infant. Applied and Environmental Microbiology 78: 402–410.
- 20. Wirawan RE, Klesse NA, Jack RW, Tagg JR (2006) Molecular and Genetic Characterization of a Novel Nisin Variant Produced by Streptococcus uberis. Applied and Environmental Microbiology 72: 1148–1156.
- 21. Zendo T, Fukao M, Ueda K, Higuchi T, Nakayama J, et al. (2003) Identification of the lantibiotic nisin Q, a new natural nisin variant produced by Lactococcus lactis 61–14 isolated from a river in Japan. Bioscience, Biotechnology, and Biochemistry 67: 1616–1619.
- 22. Fuchs SW, Jaskolla TW, Bochmann S, Kötter P, Wichelhaus T, et al. (2011) Entianin, a novel subtilin-like lantibiotic from Bacillus subtilis subsp. spizizenii DSM 15029T with high antimicrobial activity. Applied and Environmental Microbiology 77: 1698–1707.
- 23. Garg N, Tang W, Goto Y, Nair SK, van der Donk WA (2012) Lantibiotics from Geobacillus thermodenitrificans. Proceedings of the National Academy of Sciences 109: 5241–5246.
- 24. Wescombe PA, Tagg JR (2003) Purification and characterization of streptin, a type A1 lantibiotic produced by Streptococcus pyogenes. Applied and Environmental Microbiology 69: 2737–2747.
- 25. Castiglione F, Lazzarini A, Carrano L, Corti E, Ciciliato I, et al. (2008) Determining the Structure and Mode of Action of Microbisporicin, a Potent Lantibiotic Active Against Multiresistant Pathogens. Chemistry & Biology 15: 22–31.
- 26. Novak J, Caufield PW, Miller EJ (1994) Isolation and biochemical characterization of a novel lantibiotic mutacin from Streptococcus mutans. Journal of Bacteriology 176: 4316–4320.
- 27. Wescombe PA, Upton M, Dierksen KP, Ragland NL, Sivabalan S, et al. (2006) Production of the lantibiotic salivaricin A and its variants by oral streptococci and use of a specific induction assay to detect their presence in human saliva. Applied and Environmental Microbiology 72: 1459–1466.
- 28. Hynes WL, Ferretti J, Tagg J (1993) Cloning of the gene encoding Streptococcin A-FF22, a novel lantibiotic produced by Streptococcus pyogenes, and determination of its nucleotide sequence. Applied and Environmental Microbiology 59: 1969–1971.
- 29. Ryan MP, Jack RW, Josten M, Sahl H-G, Jung G, et al. (1999) Extensive Post-translational Modification, Including Serine to D-Alanine Conversion, in the Two-component Lantibiotic, Lacticin 3147. Journal of Biological Chemistry 274: 37544–37550.
- 30. Hyink O, Wescombe PA, Upton M, Ragland N, Burton JP, et al. (2007) Salivaricin A2 and the novel lantibiotic salivaricin B are encoded at adjacent loci on a 190-kilobase transmissible megaplasmid in the oral probiotic strain Streptococcus salivarius K12. Applied and Environmental Microbiology 73: 1107–1113.
- 31. Widdick D, Dodd H, Barraille P, White J, Stein T, et al. (2003) Cloning and engineering of the cinnamycin biosynthetic gene cluster from Streptomyces cinnamoneus cinnamoneus DSM 40005. Proceedings of the National Academy of Sciences 100: 4316–4321.
- 32. Herzner AM, Dischinger J, Szekat C, Josten M, Schmitz S, et al. (2011) Expression of the Lantibiotic Mersacidin in Bacillus amyloliquefaciens FZB42. PLoS ONE 6: e22389.
- 33. Boakes S, Cortes J, Appleyard AN, Rudd BA, Dawson MJ (2009) Organization of the genes encoding the biosynthesis of actagardine and engineering of a variant generation system. Molecular Microbiology 72: 1126–1136.
- 34. Holtsmark I, Mantzilas D, Eijsink V, Brurberg M (2006) Purification, characterization, and gene sequence of michiganin A, an actagardine-like lantibiotic produced by the tomato pathogen Clavibacter michiganensis subsp. michiganensis. Applied and Environmental Microbiology 72: 5814–5821.