Metagenomic study of lake microbial mats reveals protease-inhibiting antiviral peptides from a core microbiome member

Significance

Bacterial natural products are one of the main sources of today’s therapeutics and mediate countless microbial interactions. However, biosynthetic insights at a whole-microbiome level that go beyond in silico analyses are relatively rare. Here, we combined large-scale environmental genome reconstruction, microbial cultivation, prevalence-guided BGC targeting, and synthetic biology to reveal rich biosynthetic potential in complex aquatic microbial mats, a chemically poorly explored ecosystem. Compounds obtained by this workflow were active against several viruses of human relevance. The data suggest that a (meta)genome-based discovery focus on posttranslationally modified peptides, which are abundantly predicted in bacterial genomes but often not observed in pure bacterial cultures, might expand our currently limited small-molecule repertoire of antiviral drugs.

Abstract

In contrast to the large body of work on bioactive natural products from individually cultivated bacteria, the chemistry of environmental microbial communities remains largely elusive. Here, we present a comprehensive bioinformatic and functional study on a complex and interaction-rich ecosystem, algal-bacterial (microbial) mats of Lake Chilika in India, Asia’s largest brackish water body. We report the bacterial compositional dynamics over the mat life cycle, >1,300 reconstructed environmental genomes harboring >2,200 biosynthetic gene clusters (BGCs), the successful cultivation of a widespread core microbiome member belonging to the genus Rheinheimera, heterologous reconstitution of two silent Rheinheimera biosynthetic pathways, and new compounds with potent protease inhibitory and antiviral activities. The identified substances, posttranslationally modified peptides from the graspetide and spliceotide families, were targeted among the large BGC diversity by applying a strategy focusing on recurring multi-BGC loci identified in diverse samples, suggesting their presence in successful colonizers. In addition to providing broad insights into the biosynthetic potential of a poorly studied community from sampling to bioactive substances, the study highlights the potential of ribosomally synthesized and posttranslationally modified peptides as a large, underexplored resource for antiviral drug discovery.

Bacterial natural products (NPs) and NP-based synthetic compounds are a major source of therapeutics (1, 2). Over millions of years, these substances have evolved to serve a vast array of ecological functions (3, 4), and their activities aid humans in combating various health disorders, such as infectious diseases and cancer (5, 6). However, traditional bioactivity-guided screens, which mainly focus on extractions of individually cultured members of metabolically talented bacterial taxa, yield already-known NPs at increasingly high frequencies (7). Complementing these drug discovery approaches, advances in high-throughput DNA sequencing and computational technologies (8–10) now allow researchers to assess and predict the chemical potential of entire ecosystems comprising numerous chemically inconspicuous lineages and a vast diversity of as-yet uncultivated microbes. The latter is often referred to as “microbial dark matter” and constitute the great majority of bacteria in many habitats (11, 12).

Microbial mats formed by cyanobacteria have long been a focus of NP discovery (13, 14). These communities constitute highly dynamic, interaction-rich ecosystems harboring diverse microbes that are in a continuous interplay with other microbiome members and the surrounding environment. This creates a highly competitive environment that may contribute to the production of NPs conferring an evolutionary advantage (14–16). Additionally, the mats are exposed to predators, which may foster the evolution of metabolites that deter grazing (14). While diverse NPs were reported from cyanobacterial mats (13, 17), other mat communities remain poorly explored.

In this work, we used an ecosystem approach to study microbially complex algal-bacterial mats of Chilika Lake, Asia’s largest brackish water body located on the east coast of India (19° 28’–19° 54’ N; 85° 05’–85° 38’ E) in the state of Odisha (Fig. 1A). Multiple studies have been conducted on Chilika Lake’s macroorganisms including fish (18, 19), birds (20), macrophytes (21), and (macro)algae (22, 23). However, only recent studies started to focus on the microbial composition and dynamics in the lake (24, 25). According to these studies conducted by some of us, the mats follow an annual cycle of formation in winter, degradation in summer, and dispersion in the rainy season. Upon degradation, the mats release a pungent odor characteristic of sulfur-containing metabolites.

Fig. 1.

Geography and composition of Chilika mats showing their bacterial population dynamics. (A) The location of Chilika Lake on the east coast of India marked with a square. (B) Zoomed-in satellite view of Chilika Lake pinpointing the mat sampling sites. Maps were constructed with the R package leaflet (http://rstudio.github.io/leaflet/), used under Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/) from Sentinel-2 cloudless (https://s2maps.eu) by EOX IT Services GmbH (https://eox.at/) and contains modified Copernicus Sentinel data 2016 & 2017. (C) Morphology of various fresh mats. (D) Major clade-level distribution of eukaryotic organisms based on metagenomic 18S rRNA gene sequencing, grouped at >80% identity cut-off and annotated using the SILVA SSU Ref NR 138.1 August 2020 database (26). Seven most abundant clades are mentioned in the panel key, supported by the heatmap in SI Appendix, Fig. S2. Gray boxes (most prominent in MkB fresh) depict the OTUs that could not be assigned to any major-clade at >80% identity cut-off, hence remain unassigned. Black boxes represent the remaining OTUs and were significantly low in abundance (<1%). (E) NMDS plot of Chilika mats in fresh or degrading stages and the surrounding sediment based on the 16S metagenome sequencing. The stress value for the analysis was calculated at 0.054. The NMDS plot displays the 2D-representation of the Bray–Curtis distance matrix (bc_dm) between samples. Both PERMANOVA and ANOSIM tests were performed on the bc_dm. (F) Phylum-level distribution (top 10 colored) of the bacterial population (>80% identity cut-off) in mat samples and the surrounding sediment. At 90% identity cut-off, a notable proportion of OTUs in CP degrading (49%), MkB degrading (28%), OP degrading (18%) and HI degrading (42%) remain taxonomically unassigned based on the SILVA SSU Ref NR 138.1 August 2020 database. However, at 80% identity cut-off, taxonomically unknown OTUs accounted for less than 0.001% for each sample. Gray and black boxes represent the remaining OTUs and were significantly low in abundance (less than 2% and 1%, respectively). Fresh HI mats were not formed in December 2017 and could therefore not be collected.

Here, we investigated the microbiomes and NP potential for different stages of the mat lifecycle using metagenomic and cultivation-based strategies. Hundreds of taxonomically diverse bacterial metagenome-assembled genomes (MAGs) were obtained from fresh and degrading mats and sediment samples combined, revealing a plethora of biosynthetic gene clusters (BGCs) and providing insights into the compositional dynamics during mat lifecycles. Focusing on prevalent biosynthetic genes detected in distinct mat samples, we studied ribosomally synthesized and posttranslationally modified peptides (RiPPs) predicted from Rheinheimera MAGs. A combination of metagenomic, cultivation, synthetic biology, and chemical synthesis provided access to compounds from one graspetide and two spliceotide pathways. These showed activity in protease and viral inhibition assays, the latter against influenza A virus (IAV), herpes simplex virus (HSV), and lymphocytic choriomeningitis virus (LCMV). This study highlights that the metagenome-based bottom–up NP discovery from complex, interaction-rich microbiomes focusing on widespread BGC types in the community is a feasible strategy to access new bioactive compounds.

Results

Selection of Mat Samples.

Based on the salinity and depth, Chilika Lake is divided into 4 ecological sectors: southern, northern, central, and outer channel (27). Along the coastlines of the southern channels, microbial mats occur commonly (Fig. 1 A-C). Mats were collected from four regions in this area (Fig. 1B), abbreviated as CP, OP, MK, and HI. SI Appendix, Fig. S1 visualizes the morphology of each mat before and after degradation. CP mats upon degradation change from light green to a darker green color. Therefore, we named the samples “CP fresh” and “CP degrading” (for fresh and degrading mats, respectively). “OP fresh” mats maintain a dark green color toward the water. Their degradation mostly occurs close to the shore when the mats change into a black-colored layer (termed OP degrading). The MK site featured two different types of mats (MkP and MkB) located on opposite sides of an island. One of them (MkP fresh) contained red algae attached to a rocky substratum in a shallow part of the lake close to the shore. After degradation (MkP degrading), the algae are washed off to the shore and form layers of a purple mat. MkB fresh mats on the opposite coast of the island consist of a brownish green biomass, while MkB degrading mats form a brown surface layer with patches of black material around and underneath it. Sediment samples were also collected from the CP, OP, and MK sampling sites and were annotated as CP sediment, OP sediment, and MK sediment, respectively.

Identification of eu- and prokaryotic Mat Members.

To obtain information on eukaryotic organisms present in the mats, we sequenced the variable region 9 of the 18S rRNA gene (28). A heatmap of the most abundant eukaryotes suggests that the mats vary in eukaryotic complexity (SI Appendix, Fig. S2) with MkP fresh mats displaying the lowest diversity, while OP fresh, MkB fresh, and CP fresh mats were composed of a highly complex consortia of macroalgae, microalgae, and some metazoans (see SI Appendix for details).

Full-length 16S rRNA gene amplicon sequencing revealed a high bacterial diversity of the microbial mats. A nonmetric multidimensional scaling (NMDS) analysis (29) provided comparative data on the taxonomic compositions of fresh mats, degrading mats, and the sediment samples using the Bray–Curtis distance matrix (bc_dm) between the samples (Fig. 1E) applying PERMANOVA (multivariate) and ANOSIM (single factor i.e., sample type or location) tests as shown in SI Appendix, Table S1. For 10,000 permutations, the bc_dm analyzed an ANOSIM significance of 0.00019998 for sample type (fresh, degrading, or sediment) with a high R-value of 0.7671 suggesting significant dissimilarity between the types relative to within samples of the same type. However, a lower R-value of –0.2888 (significance: 0.90811) was inferred between locations (CP, OP, MK, and HI) suggesting similarity overlaps in the OTUs observed in samples from different locations. As reflected in the NMDS plot representing the PERMANOVA analysis (Fig. 1E), samples of the same type cluster together. Except for the distinct MkP degrading sample, all datasets of similar origin (i.e., fresh, degrading, or sediment) are compositionally more similar to each other than to those from other sources.

Phylum-level compositional data on mat and sediment bacteria (Fig. 1F) revealed a higher overall bacterial diversity at the degrading stage of the mat annual cycle. Most bacterial OTUs in the CP fresh and OP fresh mats were assigned to Firmicutes (approximately 54% and 80%, respectively) while Proteobacteria (approximately 71%) represented the majority of OTUs in the MkP fresh sample. On the other hand, the MkB fresh mat harbored similar proportions of Proteobacteria and Firmicutes (approximately 39 and 40%, respectively). Degrading mats displayed a different bacterial composition in comparison to the fresh mats. Bacteroidota was the major representative phylum in CP degrading, MkB degrading, and MkP degrading mats (ca. 40%, 36%, and 40%, respectively), closely followed by Campylobacterota in each case. However, OP degrading, and HI degrading mats comprised mainly Proteobacteria (approximately 33%, each). Additionally, in the degrading mats, we observed a sharp decline in the proportion of Firmicutes and an increase in that of Campylobacterota and Bacteroidota with respect to the fresh mats, showing a shift in the bacterial diversity at different stages of the mat lifecycle. The mean diversity indices of fresh, degrading, and sediment samples were inferred through alpha measures (SI Appendix, Fig. S3). The richness chart suggested that the number of different taxa was higher in the degrading mats and sediment samples relative to the fresh mats (SI Appendix, Fig. S3A). The Shannon diversity index demonstrated the taxonomic diversity of the degrading mats and sediments remained higher than the fresh mats (SI Appendix, Fig. S3B). However, the species evenness plot showed the fresh and degrading mats as well as the sediment communities exhibited comparable counts for the OTUs (SI Appendix, Fig. S3C).

Previous studies have conducted extensive culture-dependent/-independent and functional studies from 24 sampling stations over 3 seasons (72 samples) spread around the lake and have also discussed their correlations with various physicochemical parameters of the sampling points such as pH, salinity, turbidity, conductivity, temperature, ions (phosphate, ammonia, nitrate, nitrite), and total organic carbon (24, 30). However, based on the metadata collected in this study, it is difficult to infer conclusions on how pH and salinity affect the community structure of the microbial mats.

Metagenomic Analysis of Natural Product BGCs in Mat Microbiomes.

We next assessed the natural product biosynthetic potential of mat microbiomes at the DNA level. After enriching for unicellular organisms by filtration, we conducted DNA isolation, shotgun sequencing, and binning steps that resulted in 1366 MAGs encompassing 708 fresh-mat MAGs, 455 degrading-mat MAGs, and 376 MAGs from the sediment (SI Appendix, Table S2). Most MAGs belonged to the phyla Proteobacteria, Bacteroidota, and Firmicutes, with Proteobacteria being dominant in metagenomes from fresh mats and sediment metagenomes (65% and 48%, respectively). Genomes from degrading mats exhibited a more even distribution among the three phyla (SI Appendix, Fig. S4A). According to the CheckM single subunit (SSU) marker analysis (31) used to assess the quality of genomes, 422 of all MAGs combined were more than 70% complete with less than 10% contamination (SI Appendix, Table S2). This includes 174 MAGs assembled from the degrading mats, 143 MAGs from fresh mats, and the remaining 105 from the benthic population surrounding the mats. Comparing the taxonomic distribution of the MAGs with that of the 16S rRNA OTUs for the respective samples, the 10 most abundant phyla remained the same for each method (SI Appendix, Fig. S4B) with a few differences as mentioned in the figure legend.

AntiSMASH analysis of all MAGs obtained from the fresh and degrading mat metagenomes along with their sediment counterparts suggested the presence of 2,291 complete or partial BGCs. This constituted 587 BGCs predicted from the degrading mats, 1,060 BGCs from the fresh mats and 644 BGCs from the sediment metagenomes. The most abundant BGCs were terpene clusters, followed by those affiliated to ribosomally synthesized and posttranslationally modified peptides (RiPPs), and less frequently polyketide synthase (PKS) and nonribosomal peptide synthetase (NRPS) BGCs, among others (SI Appendix, Fig. S5). 283 of the 708 fresh-mat MAGs and 259 of the 455 degrading-mat MAGs contained at least one type of BGC. Through the genome taxonomy database toolkit (GTDB-Tk), 399 fresh-mat MAGs and 364 degrading-mat MAGs could be assigned to a taxon. Based on the SSU marker analysis, the remaining MAGs were less than 50% complete, leading to a low confidence in their taxonomic assignment. Using the GTDB-Tk taxonomic assignment, a phylogenetic tree was inferred to represent the fresh- and degrading-mat MAGs and a cladogram was created using the iTOL tool to match the MAGs with their predicted BGC types (Fig. 2). The analysis showed a large number of BGC types across the phylogenetic diversity of MAGs from fresh and degrading mats. This, however, excludes the unbinned fraction of the metagenome that could not be taxonomically assigned. In total, 137 BGCs were detected in the unbinned fractions of all metagenomes combined (SI Appendix, Fig. S5). Overall, we detected a higher number of BGCs in the fresh mat metagenomes compared to the degrading mats and the sediment samples (SI Appendix, Fig. S5). Further, we created a sequence similarity network (SSN) of the predicted BGCs using the Big-SCAPE toolkit with incorporation of characterized clusters available in the MiBiG database (SI Appendix, Fig. S6A) to better observe their habitat specificity. A similar SSN was also created for the RiPP families predicted from the metagenomes highlighting their sample distribution (SI Appendix, Fig. S6B). Overall, the vast majority of the mat and sediment BGCs clustered separately from the MiBiG entries, suggesting that the mat community harbors unstudied NPs.

Fig. 2.

Cladogram displaying the phylogenetic distribution of the MAGs and the detected BGC class. The cladogram shows a binary representation, i.e., the presence or absence of BGC types, irrespective of their count, in the MAGs identified from fresh (green branch labels) and degrading (black branch labels) mats that could be assigned to a taxon using GTDB-Tk (32). This includes 399 out of 708 fresh-mat MAGs and 364 out of the 455 degrading-mat MAGs spanning across all levels of completeness. The central part of the cladogram represents the phylogenetic tree of all MAGs created using GTDB-Tk. The inner branches are color-coded based on the bacterial phylogeny. The concentric circles represent, from inner to outer, i) the stage of mat life cycle, i.e., either green (fresh mat) or black (degrading mat), and ii) presence of absence of BGCs from various NP families. Abbreviations LAP, Linear azol(in)e-containing peptides; Hserlactone, homoserine lactone; RaS-RiPP, radical S-adenosylmethionine (RaS) enzyme modified-RiPP. The strain Rheinheimera pleomorphica CP1, identified in the metagenome (bin designation CP Degrading MAG 120) and isolated in this study, is shown in the chart with the curved arrows indicating the RaS-RiPP and graspetide that were further studied in this work. The tree was visualized using the iTOL annotation tool (33).

To obtain insights into the function of biosynthetic genes, we focused on RiPPs as the second-most abundant class of NPs in the mat and sediment metagenomes based on the sequence data. Collectively, 541 RiPP BGCs were detected, of which 391 BGCs belonged to the poorly defined “RiPP-like” category. These RiPP-like BGCs represent clusters annotated previously as bacteriocins according to antiSMASH v3 (under relaxed parameters) and constitute other unspecified RiPPs and understudied enzyme families. We focused on the remaining 150 RiPP BGCs clearly assignable to substance families, especially thiopeptides, lanthipeptides, lassopeptides, and sactipeptides. These BGCs were manually curated to remove errors in the automated BGC annotations (SI Appendix, Table S4). These analyses also showed that some RiPP classes were only found among related bacteria. For instance, sactipeptide BGCs were predominantly detected in Clostridiales, while the majority of lassopeptide and thiopeptide BGCs belonged to Proteobacteria.

Detection of Widespread and Conserved RiPP Multi-BGCs.

In the metagenomic data, the presence of several versions of a multi-BGC, i.e., a genomic region combining several BGCs, caught our attention. These occurred in the CP degrading MAG 120, the CP sediment MAG 146, the OP sediment MAG 40, and the MK sediment MAG 23 (Fig. 3A). All MAGs were affiliated to the genus Rheinheimera (SI Appendix, Fig. S7). The multi-BGC, named multi-BGC1, constitutes genes for a group 1 graspetide (termed rcm cluster), a group 2 graspetide (rcp cluster), and a spliceotide (rcs cluster). Metagenome-wide search for similar genome loci revealed a related but architecturally distinct gene cassette in an unbinned fraction of the OP fresh metagenome, and in the OP sediment MAG 33 and the CP sediment MAG 114, the latter two likewise assigned to Rheinheimera. This locus, termed multi-BGC2, contained BGCs for a group 2 graspetide (rap cluster) and a spliceotide (ras cluster) as shown in Fig. 3A. Additionally, multi-BGC1 and 2 were detected in published genomes of Rheinheimera pleomorphica KCTC 42365 and Rheinheimera aquimaris B26, respectively (34, 35).

Fig. 3.

Comparison of Rheinheimera sp. multi-BGCs detected across various mat and sediment samples. Schematic representation of the Rheinheimera RiPP multi-BGCs that were functionally studied. (A) Architecture of the multi-RiPP biosynthetic loci. Multi-BGC1 is present in one degrading mat MAG and three sediment metagenome MAGs. It was also found in Rheinheimera pleomorphica CP1 (red font) that was cultivated from the CP degrading mat in this work. A further locus with 100% sequence identity was identified in the genome of Rheinheimera pleomorphica KCTC 42365, previously isolated from Chilika Lake (34). Multi-BGC2 is present in two sediment metagenome MAGs and the unbinned fraction of one fresh mat. Complete or portions of the multi-BGCs were detected in these MAGs. Similarly, multi-BGC2 was also observed in OP fresh, OP sediment, and CP sediment metagenomes. An orthologous locus was detected in the genome of Rheinheimera aquimaris B26 from deep-sea sediments (35). (B) Alignment of the spliceotide precursors RcsA and RasA and (C) of the graspetide precursors RcpA and RapA showing similarity among the same class of RiPPs found in different types of Rheinheimera multi-BGCs. Black triangles and stars in the aligned sequences represent the putative leader-core boundaries and the predicted modification sites in the precursor peptides, respectively.

Cultivation of Rheinheimera pleomorphica CP1, a Mat Microbiome Core Member.

Rheinheimera are members of the phylum Proteobacteria (Pseudomonadota). We analyzed the distribution of other proteobacterial OTUs detected in the 16S rRNA gene metagenomic dataset at the class and genus level (SI Appendix, Fig. S8). Gammaproteobacteria were the majorly represented class of Proteobacteria across all mat and sediment samples, except for CP fresh mats where Alphaproteobacteria were more abundant (SI Appendix, Fig. S8A). According to our 16S rRNA data, bacteria belonging to Rheinheimera were present in MkB fresh, MkP fresh, and CP degrading mats as well as in sediment metagenomes of the CP, OP, and MK sites (SI Appendix, Fig. S8B black arrows, SI Appendix, Table S3). Further analysis of the shotgun metagenome sequencing data revealed 16 MAGs assigned to Rheinheimera (SI Appendix, Fig. S7A). This suggests that Rheinheimera members are widespread across mat and sediment samples at sufficient abundance to be represented in MAG datasets. All Rheinheimera MAGs contained BGCs of various NP types, such as arylpolyenes, type III polyketide synthase products, nonribosomal peptides, and RiPPs (SI Appendix, Fig. S7B).

For functional studies on the conserved RiPP genes, we performed extensive cultivation attempts to isolate a mat microbiome member carrying the BGCs, which ultimately yielded Rheinheimera pleomorphica CP1 from the CP degrading mat. R. pleomorphica CP1 was able to grow on marine broth (approx. 3.1% total salt content) and PH103 media (approx. 1.75% salt). For salt-tolerance tests, a modified-marine broth was reconstituted with 1, 2, 4, 6, 8, and 10% w/v NaCl as described previously (34). At 16 h time point, optimal growth in the reconstituted marine media was observed at 2-8% NaCl range. A wide range of salinity, pH, and temperature conditions have been tested for another member of this genera discussed in this study, R. aquimaris (35). According to the previous reports, R. pleomorphica KCTC 42365 was able to grow at NaCl concentrations ranging from 0 to 10%, while R. aquimaris B26 growth was impaired above 8% NaCl (34, 35). Genome sequencing confirmed that the strain contained multi-BGC1 (Fig. 3A) and shared 99.9% and 80.8% average nucleotide identity (ANI) with R. pleomorphica KCTC 42365 and R. aquimaris B26, respectively (SI Appendix, Fig. S7A). To isolate the corresponding RiPPs from the native producer, we tested an array of growth conditions at variable temperature, pH, and media composition, none of which, however, yielded the predicted RiPP products based on mass spectrometric (MS) analyses. Likewise unsuccessful were experiments conducted on R. pleomorphica KCTC 42365 and R. aquimaris B26 after obtaining these strains from culture collections. Additionally, no relevant products from multi-BGC1 and 2 were detected in the extracts of the mat samples containing Rheinheimera OTUs. Therefore, a synthetic biology approach was tested next.

Characterization of the rcs and ras Spliceotides from Rheinheimera Multi-BGCs.

To obtain insights into the structure and function of Rheinheimera multi-BGC products, we reconstituted several individual RiPP pathways in E. coli. Genes encoding the precursors and the tailoring enzymes (maturases) predicted to catalyze posttranslational modifications (PTMs) shared 100% amino acid identity within the respective multi-BGC group. Thus, the genomes of R. pleomorphica and R. aquimaris were utilized to PCR-amplify the desired genes. The N-terminally 6× His-tagged (NHis₆) precursors were produced with or without the maturases. Subsequently, peptides were isolated by Ni-affinity chromatography, proteolytically digested, and analyzed by high-performance liquid chromatography coupled to mass spectrometry (HPLC-MS) to characterize and localize PTMs.

Based on precursor motifs and maturase homologies, the rcs and ras BGC from multi-BGCs 1 and 2, respectively (Fig. 3 A and B), were suspected to belong to the recently discovered spliceotide family of RiPPs. These peptides are processed by maturase enzymes of the radical S-adenosyl methionine (rSAM) superfamily that install β-amino acids by splicing a formal tyramine (Tyn) unit out of a Tyr residue (36, 37). As features typical for spliceotides, the Rheinheimera BGCs encode RcsX and RasX as close splicease homologs and the precursors RcsA and RasA with conserved YG motifs (MYG and DYG in RcsA, LYG in RasA) that are normally the target of the splicing reaction (Fig. 4A).

Fig. 4.

HR-MS² analysis of the rcs and ras spliceotides and tyramine splicing reactions catalyzed by RcsX and RasX, respectively, in E. coli. (A) rSAM splicease-catalyzed excision of a tyramine (Tyn) equivalent [C₈H₉NO] at an XYG motif forming an α-keto-β-amino acid, where X = Met for RcsA (Left) and Leu for RasA (Right). RcsX and RasX are the spliceases of RcsA and RasA, respectively. (B) Representation of the core peptides of the RcsA spliceotide (Left) and the RasA spliceotide (Right). Protease digestions were performed in the presence of tris(2-carboxyethyl)phosphine (TCEP) as a reducing agent. In the absence of TCEP, we observed a 2 Da mass loss in the unmodified and modified precursors of NHis₆-RcsA and NHis₆-RasA, suggesting an additional disulfide linkage between the two Cys residues in both core peptides (SI Appendix, Tables S6–12). (C) HRMS analysis of GluC-digested NHis₆-RcsA (Left) and trypsin-digested NHis₆-RasA (Right) showing a mass loss of 135.06 Da owing to C₈H₉NO loss. Near-complete conversion is observed. In both cases, a ketone hydrate is also formed with a mass shift of +18 Da relative to the ketoamide. (D) MS² fragmentation patterns to localize the Tyr where the α-keto-β-amino acid residue is formed.

Individual coproductions of NHis₆-RcsA with RcsX and or NHis₆-RasA with RasX in E. coli resulted in precursor modifications based on HPLC-MS analysis (Fig. 4C). For each precursor, the data showed a characteristic 135.06 Da mass loss corresponding to a Tyn equivalent, C₈H₉NO, which was localized by tandem MS (MS²) to the MYG and LYG motif of the NHis₆-RcsA and NHis₆-RasA precursor, respectively (Fig. 4 A and D, SI Appendix, Fig. S9 A and C). The DYG site of NHis₆-RcsA was not modified. Unlike previously analyzed spliceases that show moderate conversion, product formation by the Rheinheimera splicease was almost quantitative (~99%), as evident from the endoproteinase GluC- and trypsin-digested core peptide fragments where ions of the unmodified precursor were detected at only very low abundance (SI Appendix, Fig. S9 A and C, respectively).

Characterization of the rcp Graspetide from the Rheinheimera Multi-BGC1.

Graspetide-type RiPPs contain multiple macrolactone or -lactam moieties formed by cross-links of nucleophilic and acidic side chains (38, 39). This reaction is catalyzed by maturases of the ATP-grasp ligase superfamily. Recent studies on graspetides have identified diverse BGCs classified into over 24 groups, based on core consensus sequence, putative cyclization patterns, and maturase homology (40–44). The rcp cluster encodes the predicted group 2 (plesiocin-type) graspetide precursor RcpA and the ATP-grasp enzyme homolog RcpB (Fig. 3A). As candidate sites for modification, RcpA contains a conserved TTxxxxEE motif that is also present in plesiocin, in addition to two further, less common motifs (TTxxxED and VTxxxxED) (39, 44).

Coproduction of NHis₆-RcpA with RcpB in E. coli resulted in the loss of up to 5 water equivalents in the precursor, consistent with dehydration or esterification (Fig. 5A, SI Appendix, Fig. S10B). The 4× modified species (RcpA-4H₂O) was the most abundant product, followed by the minor 5× modified product (RcpA-5H₂O). Coproduction of NHis₆-RcpA with RcmB, an ATP-grasp ligase homolog that we suspected to belong to another graspetide BGC in multi-BGC1 (Fig. 3A), did not modify RcpA (SI Appendix, Fig. S10C). Further, coproduction of NHis₆-RcpA, RcpB, and RcmB in E. coli did not reveal any additional modification of the precursor beyond the 5 water losses (SI Appendix, Fig. S10D), suggesting that RcpB is the sole enzyme that posttranslationally modifies RcpA and that rcmAB belong to a distinct pathway.

Fig. 5.

Maturation of the graspetide precursor RcpA. (A) HR-MS² analysis of trypsin-digested NHis₆-RcpA fragments with (orange spectra) or without (red spectra) treatment with NaOCH₃ (methanolysis). The data support loss of up to 5 water equivalents and gain of up to 5 methyl groups upon methanolysis with the fourfold dehydrated species as the main product. Digestion with trypsin generated two fragments from the predicted RcpA core, one containing the VTxxxxED (Left) and the other encompassing the TTxxxED and TTxxxxEE motifs. (B) HR-MS² fragmentation of trypsinized fragments showing dehydration in Thr at positions 10 (Left) and 25, 26, 38, and 39 (Right). Residues in red blocks are modified in the major product and those in purple blocks are modified only in the fivefold dehydrated product. Previously reported (39, 42) spontaneous ring opening of ester linkages in the mass analyzer is observed, leading to the dehydrated Thr side chains (shown on the modified Thr). (C) HR-MS² fragmentation of the trypsin-digested and methanolyzed major product showing methylations (orange blocks) of Glu at positions 15 (Left) and 30, 44, and 45 (Right). In the fivefold dehydrated product, the Asp at position 31 (purple block) is also methylated (Right). (D) Proposed structure of the fourfold modified major product (Left) and the fivefold, maximally modified minor one (Right), based on HR-MS, HR-MS², and methanolysis experiments. The starting residue of core peptides is marked as 1.

To characterize modifications, we combined MS² fragmentation with methanolysis. For graspetides, breakage of ω-ester-linkages due to spontaneous rearrangements in the mass analyzer has been previously reported, resulting in the formation of dehydro amino acids with a mass loss of 18 Da (42). This was exploited to detect modified Thr residues at positions 10, 26, 38, and 39 in RcpA-4H₂O (Fig. 5B, red blocks and SI Appendix, Table S13, Top and Middle). Additionally, RcpA-5H₂O showed dehydration at Thr 25 (Fig. 5 B, Right, purple block and SI Appendix, Table S13, Top and Bottom). The existence of ester bonds was supported by methanolysis, which resulted in a 14 Da mass gain of a participating acidic residue due to methylation. In this way, ester-forming Glu residues were localized to positions 15, 30, 44, and 45 in RcpA-4H₂O (Fig. 5C, orange blocks; SI Appendix, Table S14, Top and Middle). RcpA-5H₂O showed an additional modification at Asp 31 (Fig. 5 C, Right, purple block and SI Appendix, Table S14, Top and Bottom). The combined analyses support the structures shown in Fig. 5D.

Production of Leaderless RiPPs.

With information about RiPP maturation in hand, we next aimed to generate leaderless compounds for bioactivity assays. The rcp graspetide cluster encodes an ABC transporter (RcpC) containing an N-terminal C39 peptidase domain of ca. 120 amino acids. Secondary structure and transmembrane domain (TMM) prediction by the PSIPRED (45) tool suggested that the membrane-integrated helices start at position 140 from the N terminus of RcpC (SI Appendix, Fig. S11A). We cloned and expressed the rcpC region encoding the N-terminal 133 amino acids in E. coli. The protease did not cleave the unmodified or modified spliceotide precursor NHis₆-RcsA in vitro but showed site-specific proteolytic activity for the graspetide NHis₆-RcpA (SI Appendix, Fig. S11B). Unmodified NHis₆-RcpA as well as modified RcpA-4H₂O were cleaved C-terminal to the GG site, a common cleavage motif in bacterial RiPPs (SI Appendix, Fig. S11 and Tables S19 and S21) (46, 47). However, in vitro cleavage of RcpA by the native protease was inefficient, requiring equimolar amounts of protease for ca. 50% conversion. We therefore introduced a Tobacco Etch Virus (TEV) protease recognition site between the leader and core parts of RcpA and also RcsA. In addition, we improved the solubility of RcsA by fusing its N terminus with a SUMO tag (48) to generate NHis₆-SUMO-RcsA^TEV. This precursor variant was modified at similar efficiency with near-complete conversion (SI Appendix, Fig. S9B and Table S11 and S12). Similarly, upon coproduction of NHis₆-RcpA^TEV and RcpB in E. coli, the 4× dehydrated species was still the major product (SI Appendix, Fig. S10 E and F and Tables S15–S18). TEV digestion of NHis₆-RcpA^TEV and NHis₆-SUMO-RcsA^TEV releases the core with an additional Gly residue at the N terminus. TEV-digested NHis₆-RcpA^TEV and NHis₆-SUMO-RcsA^TEV cores were purified by HPLC to give G-RcpA*-4H₂O (1), G-RcpA*-5H₂O (2), and spliced G-RcsA*-Tyn (3), (G denoting Gly, asterisk refers to leaderless cores, “-Tyn” to Tyramine loss due to splicing), for use in bioassays.

In a previous study, a synthetic route for spliceotides was reported (37). A similar approach was used here to synthesize the unmodified and the ketoamide-containing (spliced) form of RasA. Synthesis of RcsA via this route was not possible, since the procedure was incompatible with the presence of a methionine-derived ketoamide. Hence, all our bioactivity tests were conducted with the chemically synthesized RasA*-Tyn (4) and the enzymatically obtained, purified G-RcsA*-Tyn (3).

Bioassays Reveal Antiviral and Protease-Inhibiting Activities.

Flow cytometry-based analysis of MC57G fibroblast cells treated with graspetide 1 and spliceotides 3 and 4 suggested no significant cytotoxic effects (SI Appendix, Fig. S13). This is supported by over 90% survival rate of cells in the presence of up to 200 µM concentration of the peptides. Spliceotide 4 exhibited low levels of cell detachment from the assay plates at concentrations above 50 µM. However, this was not detected in HeLa and Huh7 cell lines.

We tested the inhibitory activity of the Rheinheimera peptides against a range of serine and cysteine proteases, such as trypsin (Ser), chymotrypsin (Ser), human neutrophil elastase (Ser), and cathepsin B (Cys). Graspetides 1 and 2 specifically inhibited human neutrophil elastase with an IC₅₀ of 0.31 µM and 12.90 µM, respectively (Fig. 6A and SI Appendix, Table S22). Spliceotides 3 and 4 were active against a broad array of proteases. IC₅₀ values for the recombinant spliceotide 3 were 5.76 µM against human neutrophil elastase, 3.74 µM against cathepsin B, and 9.20 µM against chymotrypsin. Similarly, the chemically synthesized spliceotide 4 also inhibited the same proteases as 3. with IC₅₀ values of 0.03 µM, 0.54 µM, and 1.34 µM against human neutrophil elastase, chymotrypsin, and cathepsin B, respectively.

Fig. 6.

Protease inhibitory and antiviral activity assays of compounds 1 to 6. (A) Nonlinear regression plot showing Rheinheimera peptides-mediated inhibition of chymotrypsin, human cathepsin B, and human neutrophil elastase (HNE). The graspetide 1 inhibited chymotrypsin and HNE while 2 showed activity against HNE only. The spliceotide 3 exhibited broad-spectrum protease inhibition activity against chymotrypsin, HNE, and cathepsin B. The spliceotide 4 exhibited the highest potency among all Rheinheimera RiPPs against chymotrypsin, HNE, and cathepsin B. (B) Nonlinear regression plot showing inhibition of influenza A virus (IAV), herpes simplex virus (HSV), and lymphocytic choriomeningitis virus (LCMV) mediated by the Rheinheimera peptides 1, 3, and 4. The unmodified peptides 5 and 6 did not display any antiviral effect up to 200 µM concentrations. Viral growth in sterile water and 50% methanol was used as solvent control and for normalization. The number of plaque-forming units were normalized against cells that were not treated with any test peptide.

Graspetide 1 inhibited the growth of IAV at an IC₅₀ of 51.73 µM (Fig. 6B and SI Appendix, Table S22). IAV was also moderately susceptible to the spliceotide 4 at an IC₅₀ of 85.91 µM. Furthermore, the spliceotides 3 and 4 inhibited replication of HSV and lymphocytic choriomeningitis virus (LCMV) in their respective propagation cell lines. Spliceotide 3 displayed an IC₅₀ of 69.66 µM and 36.98 µM against HSV and LCMV, respectively, while 4 was more potent at an IC₅₀ of 28.84 µM for HSV and of 15.17 µM for LCMV. Notably, G-RcsA* (5) and RasA* (6) (i.e., the nonspliced versions of 3 and 4, respectively) did not exhibit any protease inhibitory and antiviral activity. The ketoamide moiety produced by Tyn excision in these peptides is therefore crucial for the inhibitory response.

Discussion

Specialized metabolites mediate a wide range of microbial interactions, including defense, mutualism, and exploitative relationships (3). Therefore, interaction-rich microbial habitats like soil have been favored hunting grounds for drug discovery and are the source of most of today’s natural product-based drugs (4). In contrast to studies on the chemistry of individually cultivated microbes, functionally validated insights into the chemical potential of whole microbiomes are still limited (7, 49, 50). Here we explored aquatic microbial mats, an organism- and interaction-rich environment that has, to our knowledge, not been chemically investigated at the community level. Our study encompassing compositional dynamics, environmental genome reconstruction, BGC analysis, cultivation of a microbiome member, heterologous production of compounds from two pathways, and bioactivity data provides bioinformatic and functional metabolic insights and represents a useful resource for further discovery. From the bacterial microbiome associated with the mostly eukaryotic algal substrate, over 1600 BGC-containing regions were identified in diverse MAGs, with ca. 600 further BGCs obtained from the sediment MAGs. In the mats, BGCs for terpenes and RiPPs were particularly abundant, followed by multimodular PKS and NRPS clusters. These results roughly correspond to studies from other habitats (51, 52); however, comparing numbers is challenging because computational detection methods for RiPP loci have greatly improved in recent years and PKS and NRPS counts are unreliable as their large BGCs are often split among several contigs. A clear trend was observed for type II PKS BGCs, which are widespread in actinomycetes of soil communities (53, 54) but largely absent in our data. Talented soil producer taxa like Actinobacteria and Myxobacteria were not represented in our dataset and were also reported at low abundance in previous studies on algal-bacterial mats (55–57).

A challenge in genome-based natural product discovery is that sequencing data provide a wealth of BGCs but only limited clues on bioactivities or other important features that may guide selection and downstream studies of relevant pathways. Studies have addressed this bottleneck by phylogeny-based strategies that target new members of known drug families (58, 59), by identifying BGCs that encode resistant antibiotic targets (“resistance gene mining”) (60–62), or by higher-throughput synthesis or heterologous production of BGC-predicted peptides (59, 63). In this work, we used a strategy focusing on multi-BGC loci that reoccur in MAGs of distinct samples as indicators of colonization success. The selected RiPP multi-BGCs were present in datasets from three of the four sites and detected in fresh and degrading mats as well as sediment samples. Their bioinformatic assignment to members of the genus Rheinheimera was confirmed by cultivating R. pleomorphica CP1 with a multi-BGC from a degrading-mat sample. However, multiple attempts to detect the RiPPs in laboratory cultures failed. This phenomenon seems to be particularly common for RiPPs (36, 63, 64) and is one of the reasons why large peptide families with diverse activities had remained undiscovered until recently (65–67), thus highlighting the value of synthetic biology.

The graspetides and spliceotides obtained from pathway reconstitution in E. coli belong to such recent RiPP families. RiPPs of the graspetide family utilize ATP-grasp ligase enzymes to install side chain-linked lactone or lactam rings of unusual and varied topology (40, 41). Based on the SSN created in a past study against 3,923 high-confidence graspetide BGCs, the Rheinheimera graspetides identified here were classified as either group 1 (RcmA) or group 2 (RcpA and RapA) (41). However, they were not characterized in that study. To our knowledge, core release as the final step in graspetide biosynthesis has not been experimentally demonstrated before in vitro, although indirect evidence points toward the involvement of peptidase-containing ATP-binding transporters (68). Studies have shown that the GG-motif is present in about 50% of both characterized and new graspetide precursors, sometimes with one glycine replaced by another small residue like Ala or Ser (40, 41). A subset of these BGCs containing a GG-motif precursor also seem to encode an exporter fused to a peptidase. Hence, the RcpC peptidase domain that catalyzed this reaction might be a useful tool for graspetide leader removal at GG sites in particular (Supplementary text), as the protease LahT (46) that is commonly used for RiPP core release was not suitable for the Rheinheimera graspetide (SI Appendix, Fig. S11E). The common feature of spliceotides is one or more β-amino acids installed by a remarkable enzymatic radical reaction that excises a tyramine unit out of the peptide backbone (36, 37). Homologs of the Rheinheimera splicease from our study have been categorized in a previous report as a type II splicease homolog but remained uncharacterized (37). Other members of the type II spliceases have only been observed in one other genus (Pseudoalteromonas) outside the phylum Cyanobacteriota (37). The Rheinheimera splicease identified here is noteworthy for its high efficiency resulting in near-quantitative precursor conversion, which has not been observed before. As tyramine splicing can also be applied to proteins (69, 70) and few other methods exist to generate ribosomal products with β-amino acids in vivo (71–73), this enzyme is an excellent candidate for protein and peptide synthetic biology. Few reports exist on other NPs from the gammaproteobacterial genus Rheinheimera. To our knowledge, the only characterized compounds are diketopiperazines from R. aquimaris QSI02 and Rheinheimera japonica KMM 9513, which inhibited quorum sensing and growth of bacteria (74, 75), and the blue pigment glaukothalin, a crustacean growth inhibitor (76). In addition, studies reported antimicrobial activities that remain uncharacterized (77, 78). However, data for our Rheinheimera bins showed various unassigned BGCs, suggesting a wider range of NPs.

The existence of multi-BGCs containing multiple biosynthetic loci is known for several other NPs (79), of which many also show related bioactivities. One important example is the streptogramin polyketide and nonribosomal peptide antibiotics that are used as two-component combination therapeutics (80), another multicluster of genes encoding the biosynthesis of the β-lactam antibiotic cephamycin C and the β-lactamase inhibitor clavulanic acid (81). While multiclustering can in some cases be the result of genomic BGC insertion hotspots rather than synergism (82), it has been proposed that a targeted bioinformatic search for such loci is a promising strategy to discover synergistically acting pharmaceuticals (79). Supporting this hypothesis, the graspetides and spliceotides products of the Rheinheimera locus likewise showed potent inhibition of serine-cysteine proteases and to some degree also shared antiviral activities (graspetide 1 for IAV, spliceotides 3 and 4 against a broader range of viruses). The antiviral activity of the Rheinheimera peptides is noteworthy because no cytotoxic effect on eukaryotic cells was observed in fibroblast cell lines. Protection against environmental proteases or bacteriophages might also be the natural function of the Rheinheimera RiPPs that could explain the common presence of the multi-BGCs in the mat communities, but such an ecological role remains to be tested.

Given the importance of proteases for the lifecycle of many viruses (83, 84), the observed protease inhibitory and antiviral activities might be mechanistically connected. The HSV genome encodes a Ser protease involved in capsid protein maturation (85). LCMV utilizes the host cell site 1 protease (S1P), which is also a Ser protease (84). Similarly, IAV recruits the host protease plasmin for proteolytic processing of its hemagglutinin component (86). Targeting proteolytic processing of viral protein precursors is an attractive strategy in antiviral drug development (84). In particular, the α-keto-β-amide moiety installed by the spliceotide maturase is also known as a privileged motif in medicinal chemistry for the design of serine-cysteine protease inhibitors and antiviral compounds (87) and has recently received attention as a promising design principle for potential SARS-CoV-2 drugs (88). RiPPs could thus represent a rich discovery resource for antiviral drugs that remains largely unexplored (64). Previous studies detected in bacterial and archaeal genomes thousands of BGCs for spliceotides (37) and graspetides (40, 89) alone, with several of the identified compounds being potent protease inhibitors. As many RiPPs, including those of the current study, are not detected in laboratory cultures of the native producers, genomic prediction and heterologous BGC expression methods are crucial to efficiently tap their therapeutic potential.

Materials and Methods

Sampling of Mats.

Samples belonged to five different locations as shown in Fig. 1, namely, Honeymoon Island “HI,” Chandrapur “CP,” Odialpur “OP,” Maladeikuda “MkP” and “MkB.” Metagenomic DNA and bacterial isolation experiments as well as chemical extraction of the isolate cultures were conducted at the Environmental Biology lab at KIIT University in India. The metagenomic DNA isolated from the mats was sequenced at the Functional Genomics Center Zurich (FGCZ), University of Zurich, Switzerland (for shotgun metagenomics), and Novogene, UK (for 16S and 18S rRNA sequencing). Additional information on sampling locations, metagenomic DNA isolation, 16S and 18S rRNA amplicon sequencing, shotgun metagenome sequencing and downstream analysis like distance matrix calculations, metagenome assembly, binning, and BGC analysis are provided in SI Appendix, Materials and Methods.

Bacterial RiPP Production.

Genes encoding the precursors and maturases from rcs cluster (multi-BGC1) were PCR-amplified from R. pleomorphica KCTC 42365. Likewise, the genes from the multi-BGC2 (ras cluster) were PCR-amplified from R. aquimaris B26. PCR-recovered genes were cloned into expression vectors as described in SI Appendix, Table S5 and expressed in E. coli BL21(DE3) Tuner. Additional information on cloning strategies, protein production, and isolation of matured peptide cores are available in SI Appendix, Materials and Methods. For the spliceotide RasA, a chemical synthesis strategy was employed, details of which are also provided in SI Appendix.

Protease Inhibition, Antiviral, and Cytotoxicity Assays.

Protease inhibition by the Rheinheimera peptides 1 to 6 was tested against trypsin, chymotrypsin, human neutrophil elastase, and cathepsin B. Antiviral activity was tested against lymphocytic choriomeningitis virus, IAV, and HSV. Cytotoxicity studies were conducted on MC57G cells. Virus production, cell line maintenance, and experimental procedures are provided in details in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Sequencing data have been deposited in European Nucleotide Archive under the accession ID (90). Sequencing data were submitted to the European Nucleotide Archive (ENA) global database. The accession codes for (meta)genome and rRNA sequencing are tabulated in a separate Excel file. The codes used for the bioinformatic analysis are provided in a GitHub repository archived via Zenodo under the DOI (91).