Nonribosomally and ribosomally synthesized bioactive peptides constitute a source of molecules of great biomedical importance, including antibiotics such as penicillin, immunosuppressants such as cyclosporine, and cytostatics such as bleomycin. Recently, an innovative mass-spectrometry-based strategy, peptidogenomics, has been pioneered to effectively mine microbial strains for novel peptidic metabolites. Even though mass-spectrometric peptide detection can be performed quite fast, true high-throughput natural product discovery approaches have still been limited by the inability to rapidly match the identified tandem mass spectra to the gene clusters responsible for the biosynthesis of the corresponding compounds. With Pep2Path, we introduce a software package to fully automate the peptidogenomics approach through the rapid Bayesian probabilistic matching of mass spectra to their corresponding biosynthetic gene clusters. Detailed benchmarking of the method shows that the approach is powerful enough to correctly identify gene clusters even in data sets that consist of hundreds of genomes, which also makes it possible to match compounds from unsequenced organisms to closely related biosynthetic gene clusters in other genomes. Applying Pep2Path to a data set of compounds without known biosynthesis routes, we were able to identify candidate gene clusters for the biosynthesis of five important compounds. Notably, one of these clusters was detected in a genome from a different subphylum of Proteobacteria than that in which the molecule had first been identified. All in all, our approach paves the way towards high-throughput discovery of novel peptidic natural products. Pep2Path is freely available from http://pep2path.sourceforge.net/, implemented in Python, licensed under the GNU General Public License v3 and supported on MS Windows, Linux and Mac OS X.
Citation: Medema MH, Paalvast Y, Nguyen DD, Melnik A, Dorrestein PC, Takano E, et al. (2014) Pep2Path: Automated Mass Spectrometry-Guided Genome Mining of Peptidic Natural Products. PLoS Comput Biol 10(9): e1003822. doi:10.1371/journal.pcbi.1003822
Editor: Paul P. Gardner, University of Canterbury, New Zealand
Received: June 2, 2014; Accepted: July 22, 2014; Published: September 4, 2014
Copyright: © 2014 Medema 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.
Data Availability: The authors confirm that all data underlying the findings are fully available without restriction. The Pep2Path software, documentation and test data are available from http://pep2path.sourceforge.net/. All other relevant data are within the paper and its Supporting Information files.
Funding: This work was supported by the Dutch Technology Foundation STW, which is the applied science division of NWO and the Technology Programme of the Ministry of Economic Affairs [STW 10463 to ET]. Peptidogenomics is supported by the U.S. National Institutes of Health GM097509 (PCD) and GMS10RR029121 (PCD). STW approved the decision to publish, but otherwise had no role in study design, data collection and analysis, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
After a steady decline at the end of the last millennium, natural products are back in the centre of attention as leads for drug discovery –. The secondary metabolites from which they are derived can be categorized into various classes, according to their chemical structures and the different ways in which they are synthesized enzymatically. Two of the most abundant classes are the NonRibosomally synthesized Peptides (NRPs) and Ribosomally-synthesized and Post-translationally-modified Peptides (RiPPs). Many NRPs and RiPPs are of great clinical and societal importance, having applications as antibiotics, immunosuppressants or cytostatics .
Unlike RiPPs, NRPs are synthesized not by the ribosome but by large enzymatic complexes called nonribosomal peptide synthetases (NRPSs). These NRPSs form assembly lines of modules, which each add one specific amino acid to the growing peptide chain . Each NRPS module normally consists of at least three domains: a condensation (C) domain that catalyses the condensation to the next amino acid, the adenylation (A) domain that selects the amino acid to be incorporated, and the thiolation (T) domain to which the growing peptide chain is attached. Because they function independently of the ribosome, NRPSs can introduce not only proteinogenic but also nonproteinogenic amino acids into the peptides they produce. After its synthesis by NRPSs, the core peptide scaffold of an NRP can be further modified by a wide variety of tailoring enzymes.
For a long time, drug discovery from bioactive peptides depended on laborious identification and characterization of one peptide at a time. Moreover, re-discovery of already known peptides occurred with increasing frequency, which made the entire process slow, inefficient and costly . Conversely, the genomic identification of biosynthetic gene clusters (BGCs) that could potentially encode the enzymatic machinery to make novel bioactive peptides is rapidly becoming easier, as thousands of genomes are being sequenced each year and various algorithms have been developed to automatically detect the BGCs that encode NRP and RiPP biosynthesis –.
Recently, a new technology, peptidogenomics, has been introduced, which uses the potential of high-throughput mass spectrometry (MS)  to dramatically speed up the discovery of novel bioactive peptides . Using this technology, short amino acid sequence tags (which represent a part of the complete peptide) can be reconstructed from the MS peak patterns by looking at the mass differences between peaks of various peptidic fragments. In turn, these ‘mass shift sequences’ can be assessed for their potential to represent novel peptides using dereplication tools such as iSNAP  and matched to BGCs predicted by methods such as antiSMASH , . So far, the matching of mass shift sequences to BGCs has remained a tedious and complicated procedure, especially for NRPs. In this procedure, possible amino acid sequences are manually compared to substrate specificity predictions of NRPSs by algorithms such as NRPSPredictor2 , after identification of NRPS gene clusters by antiSMASH. This lack of automation has severely impeded high-throughput peptidogenomic experimentation, and has precluded the use of peptidogenomics on microbial communities with large metagenome datasets. Moreover, the effective use of peptidogenomics on unsequenced strains  also depends on the development of computational approaches, in order to be able to compare identified sequence tags with dozens or even hundreds of genomes of related genome-sequenced strains to identify orthologous BGCs. Here we fill this gap by introducing Pep2Path, a set of algorithms that facilitate the rapid and automatic identification of candidate BGCs that correspond to NRP- and RiPP-derived mass shift sequence tags detected by mass spectrometry. Moreover, we show how Pep2Path can be used to detect BGCs for previously characterized NRPs for which the biosynthetic enzymes had not been identified yet.
Design and Implementation
Matching nonribosomal peptide BGCs to tandem mass spectra
The Pep2Path suite consists of two algorithms, NRP2Path for NRPs and RiPP2Path for RiPPs. The NRP2Path algorithm has to solve by far the more challenging problem. In order to assess how likely it is that a given MS-derived NRP mass shift sequence originates from a certain BGC, NRP2Path uses a Bayesian algorithm to estimate the probability for each amino acid in the tag to originate from each of the NRPS modules of a BGC that encodes a certain NRPS assembly-line (Figure 1, see paragraph below for details). Of course, sequence tags can be aligned to a BGC-encoded NRPS assembly line in multiple ways, and BGCs that encode multiple NRPSs may have various possible assembly-line configurations. Hence, the final score for a match between a BGC and a sequence tag will be the maximum score obtained for any of the possible alignments of a sequence tag with each of the possible assembly line orders of the NRPSs encoded by a BGC. To help interpret cases in which multiple NRPS BGCs receive similarly high scores, NRP2Path also calculates a colinearity index, which measures the proportion of contiguous NRPS module pairs that are colinear with the order of amino acids in the identified sequence tag.
The input for NRP2Path consists of mass shift sequences (or amino acid search tags) on the one hand, and genome sequences on the other hand. The latter are processed into databases by makedb, using antiSMASH and NRPSPredictor2. When a database is queried with a mass shift sequence or amino acid search tag, Pep2Path scores all possible matches between search tags and all possible assembly line configurations of each of the NRPS BGCs in the database.
Using Bayes' Theorem, the probability P(M|A) that an NRPS module M is responsible for the biosynthesis an observed amino acid A in a peptide sequence tag T is:For most applications, the prior probability P(M) of a module M to synthesize any observed amino acid will be the same for all modules and can be neglected. We do note that there are scenarios where the value of this prior probability would become relevant. For example, if one studies peptidic compounds from a microbial community, the abundance of the various species in the community would largely determine P(M). In such a case, one might estimate P(M) using the metagenomic coverage of each of the observed species from which genome information is available.
The score S for an entire gene cluster C given a sequence tag T is then the sum of the log likelihoods P(M|A):To calculate P(M|A), the probability of an amino acid to be part of an observed peptide P(A) is estimated using the frequencies of amino acids in known NRPs from the NORINE database. At those positions in a search tag that represent multiple possible amino acids, a baseline value of (1/AA alphabet size) is used for P(A). This is done in order not to introduce an inappropriate bias towards matches to rare amino acids, which would otherwise lead to high scores because the possible prior observation of such an amino acid would erroneously be treated as an actual prior observation. We use a pseudocount k of 1 to correct for the limited sample size. The formula to calculate P(A), in which represents the NRP ‘alphabet’ considered, is:P(A|M) is estimated as a function of the baseline probability P(A) and a variable IA,M, which is determined by the two types of NRPSPredictor2 substrate specificity predictions for the A domain in the module. Hence, the IA,M variable is an average value of two I values: one for the support vector machine matches and one for the Stachelhaus code matches for an NRPS module M to an amino acid A. For the Stachelhaus code predictions, the value of I corresponds to the degree of identity of the Stachelhaus code of an NRPS module (10 amino acids long) to that of the most closely related known NRPS module within the NRPSPredictor2 search space. For the SVM predictions, I will be 1.0 if the amino acid in the sequence tag matches the single amino acid prediction for the NRPS module, 0.75 if the amino acid matches the small class prediction, 0.5 if the amino acid matches the large class prediction, 0.25 if the amino acid matches the three-class prediction, and 0.0 if the amino acid matches none of these. P(A|M) is calculated with the following formula, in which c is a confidence factor (default value = 1) that accounts for how much of the final probability is determined by the substrate specificity predictions, as compared to the contribution by the baseline probability P(A), and x is again a pseudo-count variable to correct for finite sample size, this time considerably smaller than 1 (the default settings of the Pep2Path software use x = 0.01):
Combining the equations, we obtainfrom which the score S(C|T) can be calculated directly:Of course, sequence tags can be aligned to a BGC-encoded NRPS assembly line in multiple ways, and BGCs that encode multiple NRPSs may have various assembly-line configurations. Hence, the final score for a match between a BGC and a sequence tag will be the maximum S score obtained for any of the possible alignments of a sequence tag with each of the possible assembly lines encoded by a BGC.
Constructing and merging NRP2Path databases
The input of genomic data to which NRP sequence tags can be matched proceeds through the construction of NRP2Path databases. For this purpose, two accessory programs, makedb and mergedb, are provided with Pep2Path. The makedb tool uses antiSMASH2  to search user-provided input sequences for BGCs that encode NRPSs and integrates information on these BGCs into a database. Each entry in this database consists of the accession number or name of the nucleotide entry from which the BGC originates, a list of genes that constitute the BGC, taxonomic information on the species whose genome encodes it, the modular architecture of the NRPSs within the BGC, and substrate specificity predictions as given by the two NRPSPredictor2 algorithms . A database with all NRPS-containing BGCs within the GenBank database is already provided with Pep2Path. The mergedb tool can be used to merge this database with custom-made databases created from locally available sequence data, or to combine multiple custom-made databases with one another.
Input data for NRP2Path
The input for NRP2Path is either a list of amino acids or a list of mass shifts. In the latter case, Pep2Path converts the mass shifts into amino acid sequence tags using the conversion table provided by Kersten et al. . Because some amino acids or amino acid derivatives have identical masses, NRP2path will generate a list of all possible short peptide sequences. Then, NRP2path assesses for each of the BGCs within the selected NRP database how likely it is that this BGC encodes the peptide from which the sequence tag derives (Figure 1). Depending on how much is known about the source of the biological material analyzed by MS, the user can select a taxonomic range (strain, species, genus, etc.) within which to search.
RiPP2Path: a simple accompanying tool to match RiPPs to prepeptide-encoding genes
Besides matching of NRPs to their BGCs, the Pep2Path suite also offers the ability to match RiPPs to the prepeptide-encoding genes that encode their primary sequences, similar to the matching module from the recently launched RiPPquest . This is done by the RiPP2Path algorithm. Like NRP2Path, RiPP2Path uses a dedicated translation table to convert an identified mass shift sequence into all possible amino acid sequences that it could represent. To match these possible RiPP amino acid sequences to genomically encoded precursor peptides, it then reads in a set of (meta)genomic sequences and retrieves their six translation frames. Finally, it performs a simple matching of each possible RiPP sequence to the translation frames, using a sliding window with a word size equal to the tag length. The key difference between the RiPPquest and RiPP2Path algorithms is that RiPPquest (in its current version) is specifically aimed at lanthipeptides, while RiPP2Path can identify prepeptides of any type: hence, RiPPquest will be better suited to search for lanthipeptides in a large number of genomes, while RiPP2Path will be the method of choice for discovering other (and novel) types of RiPPs. Also, unlike RiPPquest, which uses molecular networking to locate matches to a database, RiPP2Path directly uses mass shift or amino acid sequences as input.
Benchmarking shows that NRP2Path is powerful and robust
As a first assessment of how NRP2Path would confront the difficult challenge to match NRP-derived mass shift sequences to BGCs, we tested whether it would be successful in matching the V-V-T(S)-T(S)-A-I(L)-V-G sequence tag that was used to identify the BGC of the nonribosomal peptide stendomycin by Kersten et al. . Pep2Path appeared to do this very effectively: even when the search space consisted of all 12,470 NRPS BGCs in the GenBank database, the stendomycin BGC emerged as the top hit. By varying the sequence tag size between 2 and 8 and introducing 0–8 simulated mispredicted substrate specificities (by artificially setting the score for a query amino acid to zero) for the corresponding stendomycin NRPS modules, we show that also BGCs with less accurate NRPSPredictor2 predictions (i.e., with up to one or two completely wrong predictions) can be identified robustly within a genome or species, even with relatively small sequence tags (Figure 2).
The heat map shows the average number of correct BGC predictions for Pep2Path searches with the stendomycin sequence tag V-V-T(S)-T(S)-A-I(L)-V-G across the Streptomyces hygroscopicus ATCC 53653 genome (20 NRPS BGCs) or across all Streptomyces nucleotide entries (342 NRPS BGCs). The searches were done for all possible search subtags of 2–8 amino acids long, and for all combinations of 0–8 simulated mispredictions for the corresponding NRPS modules. Mispredictions are simulated with zero scores given by Pep2Path for sequence tags matching to these domains.
NRP2Path benchmarking on 18 NRPs of known structure
To investigate how well NRP2Path would work in practice on novel compounds, we mined 18 NRPs from the recent scientific literature, together with their corresponding NRPS BGCs (Table S1). None of these had yet been incorporated into the NRPSPredictor2 training sets. In order to assess how well Pep2Path would be able to match tags from these NRPs to the correct BGCs under varying conditions, we varied the sequence tag size from 2 to 8 and tested all possible search tags of these sizes on databases with sizes ranging from 5 NRPS BGCs to 100 NRPS BGCs (an average bacterial genome contains ∼5 NRPS BGCs). For each database size, 50 randomly permuted BGC databases were created from BGCs originating from genomes within the same (sub)phylum, and the results were averaged across all of these permutations. The results (Table 1) confirm that the minimum sequence tag size to confidently match an NRP to a BGC is around 2–4 when the genome sequence is known. When the search space is larger, a situation that represents the mining of unsequenced strains for NRPs and attempting to match them to orthologous BGCs within the same genus , larger sequence tags (e.g., 5–8 amino acids long) are often still sufficient to identify the correct BGC.
New gene clusters for old molecules: an application
Finally, we used NRP2Path in an effort to find BGCs for NRPs within the NORINE database  for which no BGC had been discovered so far. Intriguingly, we discovered novel candidate BGCs for four molecules by searching all NRPS BGCs from the species from which the compound had originally been isolated (Table 2). When we expanded the search space to screen the entire database, we discovered another very good match, for tripropeptin A (Table 2). Surprisingly, although this eight-amino-acids-long peptide was originally discovered in the gamma-proteobacterium Lysobacter sp. , we found a match with a BGC in the genome of the beta-proteobacterium Collimonas fungivorans Ter33 (Figure S1). The tripropeptins are highly important molecules from a pharmaceutical point of view, as tripropeptin B and C display potent antibacterial activity against methicillin-resistant Staphylococcus aureus (MRSA), vancomycin-resistant enterococci (VRE), and penicillin-resistant Streptococcus pneumoniae , . All in all, the results of the NORINE searches show how even in the absence of new peptidogenomic data Pep2Path can identify putative genomic loci that encode the biosynthesis of important molecules, opening up new possibilities to further engineer these pathways and/or produce the compounds in high titers through metabolic engineering or heterologous expression.
RiPP2Path is able to specifically detect RiPP prepeptides in the majority of cases
When testing RiPP2Path on the nine Streptomyces RiPPs identified by Kersten et al. , seven of these produced unique matches of their search tags to the corresponding genome sequences (Table 3). The remaining two search tags could still be manually matched to gene clusters based on the antiSMASH-based identification of RiPP BGCs containing RiPP2Path matches.
Availability and Future Directions
Pep2Path is freely available from http://pep2path.sourceforge.net/, implemented in Python, licensed under the GNU General Public License v3 and supported on MS Windows, Linux and Mac OS X. To maximize the convenience for the majority of experimental scientists in using a new computational method, it is often helpful to provide a user-friendly web-server with accompanying step-by-step protocols for standard analytical scenarios. Providing these for the Pep2Path tools is one of the remaining challenges for future development of the software.
All in all, the automated peptidogenomics technology offered by Pep2Path constitutes a radical departure from the one-molecule-per-study approach to drug discovery from natural products that has dominated the field for long. The combination of the rapid sampling of chemical space by tandem mass spectrometry with effective computational matching of the chemistry to BGCs will allow quick identification and dereplication of large numbers of novel peptides. While detailed structural characterization of the hits for now remains a bottleneck, new methods are appearing that will accelerate this process as well . For now, identified peptides can be prioritized for further detailed characterization based on the biosynthetic novelty inferred from the BGC architecture, combined with phenotypic assays.
With nanoDESI mass-spectrometry , automated peptidogenomics also allows genome mining directly from environmental samples, without strain isolation or genome sequencing: the identified peptides can be matched directly to gene clusters in the rapidly growing database of already sequenced genomes . When combined with large-scale metagenomics, automated peptidogenomics will also make it possible to sample NRPs in environmental samples at large scales, by integrating mass spectral molecular networks ,  from MS data with BGC similarity networks , and cross-linking these two network types with a third type of networks that represent Pep2Path matching scores between molecule families and gene cluster families.
The key limitation of Pep2Path is that the quality of its results depends on the accuracy of the substrate specificity predictions by NRPSPredictor2. For some organisms, particularly fungi, the training data for these predictions is currently still limited, which leads to a decrease in predictive power. Hence, we will strive to continuously update the dataset of experimentally verified NRPS adenylation domain substrate specificities according to rigorous data standards, in order to keep improving the power of approaches such as Pep2Path as knowledge progresses.
The approach described here can potentially be extended to other compound classes, such as saccharides , once an automated system for the detection of sugar biosynthesis genes and prediction of glycosyltransferase substrate specificities has been set up. Hence, Pep2Path paves the way for the high-throughput characterization of the vast universe of BGCs discovered in genomic data of organisms throughout the tree of life.
Source code and test data. Full source code of Pep2Path version 1.1.0 and associated test data that allow users to test the functionality of the software. Updated versions of the Pep2Path code will be made available on http://pep2path.sourceforge.net in the future.
Gene cluster and predicted possible biosynthetic scheme for the production of tripropeptin-like molecules by Collimonas fungivorans Ter331. The gene cluster was identified by Pep2Path based on the raw amino acid sequence of compounds originally purified from Lysobacter sp. BMK333-48F3. The outline of the gene cluster and its encoded NRPS assembly-line above shows that the architecture of the gene cluster is consistent with the chemical structure of molecules highly similar to the tripropeptins that have been identified in Lysobacter. This BGC has eight NRPS modules, of which seven gave NRPSPredictor2 predictions exactly matching the tripropeptin A sequence in the right order, while the eighth prediction was only a near miss (ornithine predicted instead of arginine). This indicates that such a gene cluster might have undergone horizontal gene transfer at least once, from one subphylum to another. The peptides actually produced by Collimonas might have small differences in chemistry, compared to the Lysobacter tripropeptins, due to slight variations and/or promiscuity in the tailoring reactions and substrate acceptance.
NRP2Path benchmarking dataset. The table displays 18 recently experimentally characterized NRPs and their biosynthetic gene clusters used for benchmarking NRP2Path. Rare amino acids that are not covered by the Pep2Path translation table are marked as ‘Xxx’.
Pep2Path documentation. Full documentation for the Pep2Path software, including installation instructions, information on general use, instructions on command-line usage and the software license.
Pep2Path test data documentation. Description of the comprehensive set of test data that showcases the Pep2Path software. This includes instructions for testing NRP2Path using the stendomycin sequence tag and the set of 18 recently discovered NRPs mentioned in the manuscript text, as well as instructions for testing RiPP2Path using the set of nine recently discovered RiPPs.
Analyzed the data: MHM. Contributed to the writing of the manuscript: MHM. First conceived of the study: MHM PCD. Designed the algorithm: MHM RB. Wrote early prototype of the software: YP. Wrote the final software: MHM. Tested and validated the software: MHM DDN AM. Provided feedback on manuscript and figures content: YP DDN AM PCD ET RB.
- 1. Baltz RH (2008) Renaissance in antibacterial discovery from actinomycetes. Curr Opin Pharmacol 8: 557–563.
- 2. Walsh CT, Fischbach MA (2010) Natural products version 2.0: connecting genes to molecules. J Am Chem Soc 132: 2469–2493.
- 3. Winter JM, Behnken S, Hertweck C (2011) Genomics-inspired discovery of natural products. Curr Opin Chem Biol 15: 22–31.
- 4. Zotchev SB, Sekurova ON, Katz L (2012) Genome-based bioprospecting of microbes for new therapeutics. Curr Opin Biotechnol 23: 941–947.
- 5. Marinelli F (2009) Chapter 2. From microbial products to novel drugs that target a multitude of disease indications. Methods Enzymol 458: 29–58.
- 6. Fischbach MA, Walsh CT (2006) Assembly-line enzymology for polyketide and nonribosomal Peptide antibiotics: logic, machinery, and mechanisms. Chem Rev 106: 3468–3496.
- 7. Li JW, Vederas JC (2009) Drug discovery and natural products: end of an era or an endless frontier? Science 325: 161–165.
- 8. Khaldi N, Seifuddin FT, Turner G, Haft D, Nierman WC, et al. (2010) SMURF: Genomic mapping of fungal secondary metabolite clusters. Fungal Genet Biol 47: 736–741.
- 9. Li MH, Ung PM, Zajkowski J, Garneau-Tsodikova S, Sherman DH (2009) Automated genome mining for natural products. BMC Bioinformatics 10: 185.
- 10. 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 Res 39: W339–W346.
- 11. 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 Res 41: W204–W212.
- 12. De Jong A, van Hijum SA, Bijlsma JJ, Kok J, Kuipers OP (2006) BAGEL: a web-based bacteriocin genome mining tool. Nucleic Acids Res 34: W273–W279.
- 13. De Jong A, van Heel AJ, Kok J, Kuipers OP (2010) BAGEL2: mining for bacteriocins in genomic data. Nucleic Acids Res 38: W647–W651.
- 14. Bouslimani A, Sanchez LM, Garg N, Dorrestein PC (2014) Mass spectrometry of natural products: current, emerging and future technologies. Nat Prod Rep 31: 718–729.
- 15. Kersten RD, Yang YL, Xu Y, Cimermancic P, Nam SJ, et al. (2011) A mass spectrometry-guided genome mining approach for natural product peptidogenomics. Nat Chem Biol 7: 794–802.
- 16. Ibrahim A, Yang L, Johnston C, Liu X, Ma B, et al. (2012) Dereplicating nonribosomal peptides using an informatic search algorithm for natural products (iSNAP) discovery. Proc Natl Acad Sci U S A 109: 19196–19201.
- 17. Röttig M, Medema MH, Blin K, Weber T, Rausch C, et al. (2011) NRPSpredictor2–a web server for predicting NRPS adenylation domain specificity. Nucleic Acids Res 39: W362–W367.
- 18. Nguyen DD, Wu C-H, Moree WJ, Lamsa A, Medema MH, et al. (2013) MS/MS networking guided analysis of molecule and gene cluster families. Proc Natl Acad Sci U S A 110: E2611–E2620.
- 19. Mohimani H, Kersten RD, Liu W-T, Wang M, Purvine SO, et al. (2014) Automated genome mining of ribosomal peptide natural products. ACS Chem Biol 9: 1545–1551.
- 20. Caboche S, Pupin M, Leclere V, Fontaine A, Jacques P, et al. (2008) NORINE: a database of nonribosomal peptides. Nucleic Acids Res 36: D326–D331.
- 21. Hashizume H, Igarashi M, Hattori S, Hori M, Hamada M, et al. (2001) Tripropeptins, novel antimicrobial agents produced by Lysobacter sp. I. Taxonomy, isolation and biological activities. J Antibiot (Tokyo) 54: 1054–1059.
- 22. Hashizume H, Sawa R, Harada S, Igarashi M, Adachi H, et al. (2011) Tripropeptin C blocks the lipid cycle of cell wall biosynthesis by complex formation with undecaprenyl pyrophosphate. Antimicrob Agents Chemother 55: 3821–3828.
- 23. Hirosawa S, Takahashi Y, Hashizume H, Miyake T, Akamatsu Y (2013) Synthesis and antibacterial activity of tripropeptin C derivatives modified at the carboxyl groups. J Antibiot (Tokyo) 67: 265–268.
- 24. Inokuma Y, Yoshioka S, Ariyoshi J, Arai T, Hitora Y, et al. (2013) X-ray analysis on the nanogram to microgram scale using porous complexes. Nature 495: 461–466.
- 25. Roach PJ, Laskin J, Laskin A (2010) Nanospray desorption electrospray ionization: an ambient method for liquid-extraction surface sampling in mass spectrometry. Analyst 135: 2233–2236.
- 26. Guthals A, Watrous JD, Dorrestein PC, Bandeira N (2012) The spectral networks paradigm in high throughput mass spectrometry. Mol Biosyst 8: 2535–2544.
- 27. Watrous J, Roach P, Alexandrov T, Heath BS, Yang JY, et al. (2012) Mass spectral molecular networking of living microbial colonies. Proc Natl Acad Sci U S A 109: E1743–E1752.
- 28. Cimermancic P, Medema MH, Claesen J, Kurita K, Wieland Brown LC, et al. (2014) Insights into secondary metabolism from a global analysis of prokaryotic biosynthetic gene clusters. Cell 158: 412–421.
- 29. Kersten RD, Ziemert N, Gonzalez DJ, Duggan BM, Nizet V, et al. (2013) Glycogenomics as a mass spectrometry-guided genome-mining method for microbial glycosylated molecules. Proc Natl Acad Sci U S A 110: E4407–E4416.