Mass spectral molecular networking of living microbial colonies

Results and Discussion

In the past decade, it has become possible to ionize molecules from native surfaces with MS at ambient pressures through the development of novel ionization sources (8⇓⇓⇓⇓⇓⇓⇓⇓⇓–18). Each ionization technique, because of its unique ionization mechanisms, has specific applications and uses in diagnostics, forensics, histopathology, and high-throughput screening. Although many of these ionization sources provide “gentle” desorption of surface ions, none of the most widely used ambient ionization methods are delicate enough to not significantly disturb microbial colonies grown on the relatively soft surface of agar medium in a Petri dish. MS analysis of such samples typically results in the destruction of the agar and of the microbial colony (19).

Analysis of solid agar as opposed to liquid culture-based growth conditions is essential, as some microbial behavior including many developmental processes and production of certain secreted factors, such as cannibalistic factors in Bacillus subtilis, have only been observed when the microbes were grown on solid media (20). Furthermore, soft-agar culturing is very common in microbiology laboratories. As the ionization source usually dictates the types of samples that can be characterized with MS, we surveyed available ambient pressure ionization sources and discovered that the recently invented nanoDESI source was able to ionize dyes, peptides, and proteins from glass slides and Teflon surfaces without needing a destructive vector, such as a carrier gas (21). Importantly, the nanoDESI ionization source is of fairly simple and straightforward design that can be built at low cost on most common mass spectrometers such as ion traps, Orbitraps, Q-TOFs, and others.

NanoDESI uses solvent delivered via a syringe pump or HPLC through a primary line of fused silica capillary (Fig. 1A) in which the terminal end is positioned slightly above the sample surface. A second self-aspirating nanospray capillary is then positioned to form a liquid bridge between the terminal end of the primary fused silica capillary and the anterior end of the nanospray capillary. The analyte is desorbed from the sample surface at the liquid bridge and transferred into the nanospray capillary by capillary action and by vacuum from the heated mass spectrometer inlet. Solvent flow through the nanospray capillary and subsequent electrospray into the mass spectrometer is maintained by the effective vacuum from the mass spectrometer inlet as well as the application of high voltage to the solvent allowing for a constantly flowing yet constant volume microliter sized droplet at the liquid bridge. The size of the droplet formed at the junction between the two capillaries is controlled by adjusting the flow rate into and out of the liquid bridge with the latter being accomplished by adjusting the position of the terminal end of the nanospray capillary relative to the mass spectrometer inlet. When the desired droplet size is achieved at the liquid bridge, surface sampling is achieved by simply making contact with the droplet at the liquid bridge with the sample surface allowing molecules to desorb into the solvent stream followed by immediate analysis by MS. This simple instrument design allows for good interlaboratory reproducibility, as this work was performed on an LTQ-Orbitrap and an LTQ-FT-ICR located in two different laboratories using two independently built custom nanoDESI ionization sources constructed from two different starting platforms and yet yielded very similar results.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

An overview of nanoDESI analysis of microbial colonies from a Petri dish. (A) A schematic overview of nanoDESI. HV, high voltage; SS, stainless steel. (B) A photograph of the nanoDESI setup with a microbial colony grown on an agar surface in a Petri dish. (C) A mass spectrum obtained from a B. subtilis 3610 colony with nanoDESI. Asterisks mark agar sugar signals, “I” is diglycosyl diglyceride, “II” is lysyl-phosphatidylglycerol, “III” is surfactin, “IV” is sublancin (3⁺), “V” is sporulation killing factor (2⁺), “VI” is plipastatin, and “VII” is subtilosin (2⁺). (D) Time-dependent analysis at a single location within a 3-d-old B. subtilis 3610 colony to indicate the changes in signal intensity of specific molecules over time. DGDG, diglycosyl diglyceride; LPG, lysyl-phosphatidylglycerol; SKF, sporulation killing factor. Solvent used was methanol:acetonitrile:toluene (50:35:15). Movie S1 shows how nanoDESI works in real time.

Initial nanoDESI experiments were carried out on individual biofilm forming colonies of B. subtilis 3610, which produces a wide range of well characterized compounds, and therefore provides an ideal test system (20, 22). NanoDESI analysis was shown to be very sensitive in its ability to detect metabolic exchange factors compared with our previous studies of the same system using MALDI thin-layer agar analysis as well as desorption electrospray ionization imprinting (19, 20, 23). As the analysis requires absolutely no sample preparation, the Petri dish containing the sample can simply be placed under the liquid bridge and raised into position until the nanoDESI liquid probe makes contact with the living colony, allowing for instant chemical characterization of the sample surface. Movie S1 provides a live demonstration of this technique where a Petri dish containing colonies of B. subtilis 3610 (Movie S1, Left) can be seen moving into position while the time-synced computer screen capture of the mass spectra (Movie S1, Right) shows the signal response in real time (Fig. 1B and Movie S1). As with any surface liquid extraction method, the types of the molecules observed in the spectrum depend on the solvent used. Specifically, compounds soluble in a particular solvent are preferentially desorbed and ionized by nanospray. This flexibility allows for careful tuning of the detection of certain compounds, which can greatly aid in data collection and analysis. The use of a nanospray capillary to direct the postdesorption solvent stream toward the mass spectrometer inlet also results in great signal intensity, as little to no solvent is lost and this allows for fast and easy cleaning of the ionization source because this is the only piece of the source that gets dirty.

Although this technique has proved very versatile within our laboratories across many sample types, no method is without its limitations. First, MS has inherent limitations whereby certain molecules ionize preferentially to other molecules present in the sample. Second, as nanoDESI is a surface sampling method, the chemical composition of the surface will influence ionization efficiency. For instance, although many agar media perform well using nanoDESI, such as Luria–Bertani (LB), International Streptomyces Project Media 2 (ISP2), bovine heart infusion (BHI), potato dextrose agar (PDA), M9 minimal media, Kings medium B (KB), Bennett's medium, and Spider media that we have tested, certain media containing extreme amounts of salt, such as M1 or A1 salt water marine media, and high amounts of glycerol, such as minimal salts glycerol glutamate (MSgg) media, are more challenging because they may cause ion suppression, resulting in decreased analyte signal. This becomes less of an issue as the microbial colony becomes larger and thicker, as the liquid droplet will have minimal contact with the media directly. Also, analysis of viscous samples can lead to clogging of the secondary capillary; however, this can usually be compensated for by changing the polarity of the solvent or by increasing the droplet size. Finally, very conductive sample surfaces, such as conductive indium-tin-oxidemicroscope slides, usually resulted in poor ionization and can be partially corrected for by changing the sample surface or by repeatedly “touching” the sample surface with the droplet as analyte signal is regained when it has been lifted from the conductive surface and will persist for a short time.

As no enrichment steps are performed before analysis, data collected via nanoDESI often result in numerous different classes of compounds represented within a single data set (Fig. 1C). In addition, signal for most analytes will persist from a single sampling location for extended periods of time (Fig. 1D), allowing for automated collection of tandem MS data. Characterization, annotation, and/or identification of the observed signals thus present a significant challenge. In metabolomics, identification of a molecule is usually carried out by targeted tandem mass fragmentation. The resulting fragmentation pattern can then be compared with metabolomics databases such as METLIN, MassBank, LIPID MAPS, or similar types of databases (24⇓–26). However, as most molecules involved in metabolic exchange are unique to one or a few organisms, there is currently no database that has a searchable mass fragmentation library accessible to the public of sufficient size to cover the unique molecules produced by most known microbes. Therefore, one has to usually resort to manual interpretation of fragmentation data, which unfortunately is currently the state of the art in the natural products field, with the basic experimental paradigm for discovery remaining essentially unchanged since the discovery of penicillin in the late 1920s. Manual analysis of a single MS/MS spectrum typically takes between 10 min to several hours depending on its nature and complexity and is therefore incredibly time-consuming and becomes impractical when the data consists of thousands to millions of MS/MS spectra. Furthermore, even with more evolved “omics”-based analysis methods, obtaining a global visualization of collected data for quality assessment and analysis is a key bottleneck within the workflow (27, 28). In the context of natural products, such tools for data organization and navigation, let alone compound identification, are simply nonexistent. Hence, alternative ways to look at fragmentation data need to be developed.

The visualization of networks by using freeware such as Cytoscape (www.cytoscape.org) is often used for the global display of large data sets such as protein interactions, biochemical pathways, population networks, and even airplane travel (29⇓⇓–32). In biology, such networks enable the direct observation of similarities as well as differences between two or more conditions in which similar entities within the network are clustered together while disparate or unique entities are grouped separately. As MS fragmentation of each individual molecule results in a unique MS/MS fingerprint, we decided to develop network-based workflows to organize large data sets of tandem mass spectra based on the similarity between fragmentation patterns of different, but related, precursor ions. By using a variation of spectral networks designed for proteomic applications, the data are initially simplified by forming consensus spectra whereby identical spectra exhibiting identical precursor ion mass-to-charge ratios (m/z) and fragmentation patterns are merged (33). The simplified MS/MS data are then used for generation of the molecular networks analysis (Fig. 2A). Vector similarities are calculated for every possible pair of spectra with a minimum of six matching fragment ions (i.e., peaks) with similarity determined by using a modified cosine calculation that takes into account the relative intensities of the fragment ions as well as the precursor m/z difference between the paired spectra. This extends the concept of spectral alignment as applied in proteomics, with the key exception that certain peptide-specific parameters, such as the use of peak likelihood scores, are generalized to apply this approach to all classes of molecules including lipids, polysaccharides, peptides, metabolites, and proteins. This is important because it is not known ahead of time what class of molecules will be ionized during analysis. When this has been completed, the significantly matched spectrum pairs are reported as a molecular network by using MATLAB scripts whereby each spectrum is allowed to connect to its top K scoring matches (we typically allow a maximum of 10 connections per node). Edges between spectra are retained only if in the top K for both paired spectra and the vector similarity score, represented as a cosine value, of the match is greater than the user-defined threshold. Cosine threshold values are usually set between 0.5 to 0.7 whereby a cosine value of 1.0 indicates identical spectra. Furthermore, the created MS/MS network is processed in MATLAB by removing control spectra representing agar, solvent, or control colonies and assigning colors and m/z-values to nodes. These data are then imported into the free visualization program Cytoscape to visualize the MS/MS networks (29). Cytoscape produces a visual representation of the molecular network where each node (i.e., circle) represents a single consensus MS/MS spectrum for a given parent mass, with the thickness of an edge between connecting nodes being indicative of the similarity score for that spectral pair, with higher scoring matches resulting in thicker connecting edges and, when possible, in closer distances. Depending on Cytoscape’s nondeterministic network rendering algorithms, the distance between nodes also depends on the direction and number of connections.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

The generation of molecular networks via spectral alignment. (A) A schematic representation of how the molecular networks are generated. The values are representative of cosine scores from 0 to 1, where 1 indicates identical spectra and 0 means no similarity whatsoever. In our data, we found that a cosine cutoff of 0.5 resulted in molecular networks that could be interpreted. The thickness of the edges (blue lines connecting nodes) indicates the level of similarity. (B) A Cytoscape visualization of the surfactin single adduct cluster from B. subtilis 3610. The full MS/MS network is shown in Fig. 3A. Nodes with red border are represented in C. (C) An example of four spectra from the molecular network shown in B that show a strong cosine score.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Molecular networks of nanoDESI fragmentation data obtained from single microbial colonies. (A) The annotated molecular network from B. subtilis 3610. (B) The annotated molecular network of S. coelicolor A3(2), M. smegmatis MC2, P. aeruginosa PAO1, and S. marcescens ES129. Insets: Images of samples were probed with nanoDESI. The structures of each of the annotated clusters are shown in SI Appendix, Figs. S1, S4, and S5. The color scale shows the mass range of the parent ions: green nodes represent the smallest masses; red nodes represent the largest masses fragmented. (Scale bar: 1 mm.)

The benefit of such an approach is that, as spectra are organized based on fragmentation similarity, identification of analogues and related compounds becomes much easier. A subset of a molecular network generated for B. subtilis 3610 from approximately 25,000 fragmentation spectra is shown in Fig. 2B. It shows that analogues of the cyclic lipopeptide surfactin are localized in one region within the MS/MS network. One can see analogues of surfactin separated by 14 or 28 Da largely as a result of differences in lipid side chains and exchange of amino acids (e.g., Gly and Ala) consistent with fragmentation data (Fig. 2C). This is a common observation with lipopeptides made via the nonribosomal peptide synthetase paradigm (34). Furthermore, the cluster shows numerous differences of 16 Da between nodes, which is usually attributed to loss or gain of oxygen as well as between Na and K adduct forms of the molecule, and differences by loss of 113 Da consistent with the amino acids Leu and Ile. Although the mass differences caused by oxidation and varying lipid chain length were expected, the loss of Leu/Ile was not. Comparison of the neighboring surfactin MS/MS spectra with the −113 Da MS/MS spectra (SI Appendix, Fig. S1) indicated that the parent compound exhibiting the loss of Leu/Ile was still a cyclic lipopeptide. The data are consistent with the biosynthetic pathway “skipping” one of the N-terminal leucine residues during the biosynthesis (Fig. 3A and SI Appendix, Fig. S2). It should be noted that the location of the nodes within the planar representation of the MS/MS network is not related to the nature of the molecule, as the spatial orientation of the MS/MS network is randomly generated when the network is rendered by Cytoscape. To further aid in identification, MS/MS of known molecules can be included within the MS/MS network and tracked for comparison and for propagation of annotations from known to unknown metabolites. In addition, data visualization using molecular networks allows one to discover molecules that are still unclassified but may be biologically relevant especially when comparing samples from two states, such as different time points or mutants.

To generate a more complete MS/MS network for B. subtilis, we combined data sets collected by using a variety of solvents. Specifically, we acquired nanoDESI data by using two solvents: (i) a mixture of acetonitrile and 0.05% formic acid in water (ratio of 65:35) for extraction of polar compounds and (ii) a methanol:acetonitrile:toluene (50:35:15) mixture for extraction of nonpolar compounds. These experiments enabled the detection of a wide range of molecules within a single mass spectral MS/MS data set including lipids, peptides, small metabolites, and polysaccharides (SI Appendix, Fig. S3). Molecular networking of these data revealed metabolic exchange factors such as the lipopeptides surfactin and plipastatin, the ribosomally encoded peptides SKF and subtilosin, fatty acids and lipids such as diglycosyl glycerides and lysyl-phosphatidylglycerols, and various polysaccharides from the agar medium (Fig. 3A). For each identified compound, the fragmentation pattern is consistent with the MS/MS reported in the literature (when available), with additional confidence through detection of multiple variations for almost all compounds including H⁺, Na⁺, and K⁺ adducts as well as comparison with purified standards (SI Appendix, Fig. S4) (20, 35⇓–37). By using molecular networking, another unique surfactin analogue was found: two nodes with masses of m/z 685 (because of variances in spectral quality) were tightly clustered with known surfactins (m/z 990–1,100). Further inspection of the spectrum revealed that the mass at m/z 685 is a linear analogue of surfactin with the sequence (I/L)(I/L)VD(I/L)(I/L) (Fig. 3A and SI Appendix, Fig. S5). As the molecular weight of this compound is significantly lower than for any reported surfactin, it would have been much more difficult to recognize the relationship between the MS/MS spectrum of this compound and other surfactins without molecular network visualization.

To highlight the broad applicability of nanoDESI-based molecular networking in microbiology, molecular networks were also constructed for the soil bacteria Streptomyces coelicolor, S. marcescens ES129 (a common soil organism as well as an opportunistic pathogen that is also associated with eczema), M. smegmatis MC2 (a model organism for Mycobacterium tuberculosis), and P. aeruginosa PAO1 (an opportunistic pathogen of the lung that is also associated with other infections; Fig. 3B). A variety of structurally and functionally diverse molecules were identified by using these MS/MS networks, which included quorum-sensing molecules like quinolones and peptides as well as glycolipids, antibiotics, pigments, and oligosaccharides (SI Appendix, Fig. S4). Furthermore, it is intriguing to note that a large number of tightly clustered nodes were not annotated, thus highlighting the capability of molecular networks to reveal molecular diversity and potentially novel unidentified compounds.

By using a liquid sampling protocol such as nanoDESI, in conjunction with MS-based molecular networking, the molecular profiles from intact microbial colonies of ∼50 different microbes have already been collected, with the rate-limiting step being the time needed to culture the organisms themselves. Most MS laboratories are not set up for culturing, and therefore tools such as this provide unique opportunities for collaborations with microbiologists and other life scientists. Also, by not selecting for a single class of compounds, these data sets give microbiologists the unique opportunity to gain insight into the general molecular content of their living systems at a given point in time. The small footprint of the probe, ranging from 10 μm to 500 μm depending on capillary size, also allows for discrete profiling across the sample surface to observe differences in metabolic output within a single colony or within a complex microbial population while at the same time providing minimal destructive impact on the sample being analyzed.

Molecular Networking Over Time.

Although current tools for temporally monitoring chemical changes of microbial colonies from solid surfaces are time-consuming, nanoDESI analysis combined with data visualization using molecular networking improves the efficiency of such experiments as data are acquired in real time (Movie S1). To monitor live colonies, single colonies of B. subtilis 3610 were subjected to repeated nanoDESI measurements over the course of 60 h (Fig. 4A), whereby, at each time point, single colonies, grown on ISP2 nutrient agar, were removed from the incubator, photographed, analyzed by nanoDESI for 10 to 20 s, and then placed back into the incubator until the following measurement. Upon visual inspection after each measurement, there appeared to be no significant physical damage and no contamination to the bacterial colony. However, decreased growth of the biomass was observed in the small area (<300 µm²) sampled during the analysis (Fig. 4B, arrow). This resulted in only local effects and did not appear to impact the overall growth of the entire colony compared with control colonies that were not analyzed by nanoDESI (Fig. 4B). This local damage to the colony can likely be minimized by using different solvents or smaller capillaries, which can bring the size of the nanoDESI droplet down to less than 10 μm in diameter.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

NanoDESI and molecular networking in a time-dependent manner. (A) NanoDESI analysis over time of a single microbial colony of B. subtilis 3610. (B) An optical photograph of a 72-h colony that was probed eight times and the effect nanoDESI has on the colony phenotype. “I” is the sample (arrow points to one of sampling locations) and “II” is the control that was not subjected to nanoDESI analysis. (C) The molecular network and annotation of specific clusters from tandem MS/MS data taken from B. subtilis 3610 over time. (Scale bar: 1 mm.)

An additional time-course experiment was performed by using separate colonies of B. subtilis 3610 whereby, at each time point, approximately 4,000 MS/MS spectra were collected from the colony surface, enabling the creation of molecular networks in a time dependent manner (Fig. 4C). The resulting MS/MS network showed a clear increase in metabolic output within B. subtilis across the 60-h time period, with increased production of structural variants of the cyclic peptides surfactin, plipastatin, and subtilosin steadily increasing over time. Surfactin can be seen immediately produced after inoculation followed by plipastatin at 12 h and finally subtilosin at 24 h. This latter result mirrors what we have seen by using MALDI-based imaging MS, in which there is a metabolic switch observed between 16 and 22 h resulting in the production of plipastatin and subtilosin. In addition to what we have annotated using MALDI-based imaging MS, we observed the production of the glycopeptide sublancin (38), whose signal only appeared between 12 and 36 h. There was also a major shift in lipid production at the 36-h mark indicating possible temporal, nutritional or quorum-sensing–based regulation of sublancin production. This observation of transient sublancin production is consistent with the RNA transcript analysis of the biosynthetic enzymes involved in the production of sublancin, which steeply declined after 27 h of culturing (SI Appendix, Fig. S6) (39). Lastly, a closer look at the surfactin and plipastatin node clusters show changes in production for specific compound variants over time. Plipastatin is initially produced in its A and B forms, with a gradual shift to only production of plipastatin B (SI Appendix, Fig. S7), whereas surfactin produces its longer lipid length variants across the whole time course but produces shorter lipid chain length variants only between 24 and 48 h (SI Appendix, Fig. S8). This again shows how the combination of nanoDESI analysis on live microbial colonies and molecular networking is giving unprecedented access to molecular information of microbial systems.

Molecular Networking of Microbial Interactions.

Microbial populations in nature almost always exist in assemblages interacting with their neighbors. When B. subtilis PY79 (a laboratory domesticated strain of B. subtilis 3610 in which most of the polyketide synthase and nonribosomal peptide synthetase metabolic machinery is nonfunctional) interacts with S. coelicolor A3(2), B. subtilis PY79 elicits pigment production and aerial hyphae formation in S. coelicolor, whereas S. coelicolor increases production of the cannibalistic factors SKF and SDP in PY79 in the region of interaction (23, 40). Here we show that such interactions can be easily examined by using nanoDESI in combination with molecular networking (Fig. 5).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

The molecular network of S. coelicolor A3(2) interacting with B. subtilis PY79. (A) The comparison of the molecular data from the S. coelicolor colony adjacent to B. subtilis vs. the S. coelicolor colony further away. (B) The comparison of the molecular data from the interacting and noninteracting sides of the B. subtilis PY79 colony. It should be noted that, although PY79 has a frame shift in sfp, the phosphopantetheinyl transferase required for surfactin and plipastatin biosynthesis, surfactin is still produced in small amounts (41). This has been observed before by MALDI imaging, as well as imprint desorption electrospray ionization, and can be attributed to promiscuity of another phosphopantetheinyl transferase or a ribosome slippage providing a low amount of in-frame translation of the frame-shifted sfp gene (19, 23).

Samples were prepared by inoculating two individual colonies of S. coelicolor 5 mm apart and allowing growth for 18 h at 30 °C, at which time a single 0.5-μL inoculum of B. subtilis PY79 was placed 1 mm from one of the S. coelicolor colonies. After 36 h, the Petri plate was removed from the incubator and immediately analyzed by nanoDESI whereby tandem mass spectra were automatically collected from each of four sampling locations: the S. coelicolor colony farthest from the B. subtilis PY79 colony, the S. coelicolor colony closest to the B. subtilis PY79 colony, the side of the B. subtilis PY79 colony closest to S. coelicolor, and the side of the B. subtilis colony furthest from S. coelicolor. Two MS/MS networks were created (one from both S. coelicolor colonies combined and one for both B. subtilis sampling locations combined as shown in Figure 5A and 5B, respectively) where background nodes from solvent and agar were removed. The resulting MS/MS network confirms previous reports with findings that actinorhodin and SapB are present at the S. coelicolor colony nearest B. subtilis whereas the signals for SKF from B. subtilis appear more prominently on the interacting side of the B. subtilis colony. In addition, the analyses showed that calcium-dependent antibiotic in S. coelicolor is also present at the colony interaction interface, as well as many other signals that are currently unknown. Also, although the red pigment prodiginine is present in the near and far S. coelicolor colonies, increased production of the compound in the near colony resulted in the formation of prodiginine oligomers and adducts, which range from 400 Da to 1,800 Da as determined by the MS/MS fragmentation data (SI Appendix, Fig. S9). The data indicate that multispecies comparative molecular networking is a powerful strategy to characterize molecular differences in live, interacting microbial colonies.

Detection and Partial Characterization of Thanamycin by Comparative Molecular Networks.

A recent metagenomic analysis of the rhizosphere microbiome showed that specific members of Pseudomonadaceae were identified as key players in the natural protection of sugar beet plants against specific fungal root pathogens (6). Bioassays followed by genetic analyses revealed that the antifungal activity of Pseudomonas sp. strain SH-C52 was mediated by a putative 9-aa lipopeptide designated thanamycin (7). Despite significant time and efforts, the investigators were not able to detect this novel compound by traditional biochemical assays. To demonstrate the capacity of the tools described here, live colony analysis by nanoDESI MS coupled with molecular networking were used to detect thanamycin by comparing the WT strain SH-C52 vs. two mutants; one disrupted in the nonribosomal peptide synthetase gene thaB and the other in the halogenase gene thaC2 (Fig. 6A). As there were a large number of signals that were fragmented (Fig. 6B), the clusters within the MS/MS networks were compared and only the signals unique to specific samples are displayed (Fig. 6A). The signals that were only found in the WT strain were subjected to peptidogenomics, an approach that matches peptidic natural products to their genetic signatures, including molecules that are made via the nonribosomal peptide synthetase paradigm (7).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Partial characterization of thanamycin from live colony analysis: molecular networking and peptidogenomics of Pseudomonas sp. strain SH-C52. (A) Comparative molecular network of WT strain SH-C52 and two mutants, disrupted in thaB or thaC2. Red indicates only found in WT, green indicates only found in the thaC2 mutant, and purple indicates only found in the thaB mutant. (B) Overlay of the MS spectra of all three strains with the same colors as in A. (C) MS/MS spectrum of the 2⁺ thanamycin ion and annotated sequence tag for thanamycin. Signals with asterisks are the chlorination isotopic signature, and signals with a number sign do not display this feature. (D) MS/MS/MS confirmation of the thanamycin signal and its sequence tag. (E) Comparison of the syringomycin gene cluster to the thanamycin gene cluster and the corresponding “best” predictions of the adenylation domain specificity according to the Stachelhaus rules (45, 46). The domains described are as follows: T, thiolation domain; A, adenylation domain; Cl, halogenase; TE, thioesterase domain. (F) Comparison of the structure of syringomycin and the partial sequence obtained for thanamycin that is consistent with the observed tandem MS data.

Peptidogenomic analysis revealed that the clusters containing m/z 646 [M+2H]²⁺ and m/z 1,291 [M+H]⁺ were monochlorinated (7). In particular, the monochlorination is evident by the 24% contribution of intensity in the +2 Da isotope of ³⁷Cl that is consistent with monochlorinated molecules (Fig. 6B, Inset), which was also observed in the isotopic patterns in the fragmentation data (SI Appendix, Fig.S10). The chlorination is consistent with the non–heme-dependent halogenase found in the thanamycin cluster, which itself exhibits high sequence similarity to the syringomycin biosynthetic pathway from the plant pathogen P. syringae pv. Syringae (42, 43), suggesting that thanamycin belongs to the syringomycin family of antifungal agents. Annotation of the MS/MS and MS/MS/MS data for thanamycin showed that it contained a chlorinated threonine followed by a hydroxyaspartate and a dehydrobutyrate (Fig. 6 C and D), which is also consistent with the structure of syringomycin (Fig. 6 E and F) (42⇓–44). Further annotation of the mass spectra extended this sequence tag to include a threonine and histidine, which is consistent with the predictions for the adenylation domain specificities of the gene cluster (45, 46). Full structural and bioactivity characterization of thanamycin is ongoing.

Comparison of the molecular networks of the WT SH-C52 strain and its mutants followed by peptidogenomic analysis allowed us to confirm that we had detected the thanamycin molecule that, despite application of several different cultivation and extraction procedures, eluded detection by traditional approaches. A possible reason why thanamycin was initially missed by conventional approaches was because thanamycin is produced transiently and in low quantities compared with the rest of the molecules observed in the sample. The discovery of thanamycin was made possible by using nanoDESI techniques, and this exemplifies that our methodology allows for the detection of the proverbial needle in the haystack.