Predatory myxobacteria lay at the cross-roads of metallo-β-lactamase evolution

Abstract

Understanding the biology of B1 metallo-β-lactamases (MβLs) in their native hosts and the molecular bases of the adaptive mechanisms during the transfer process to human pathogens, may provide novel therapeutic insights against these important resistance factors. Through large scale phylogenomic analyses, we identify Bacteroidota and Myxococcota as the primary hosts of the enzymes in environmental habitats. The monophyletic lineage of MβL homologues from the predatory members of the latter phylum shares common ancestry with clinically relevant families found in γ-Proteobacteria, highlighting their evolutionary significance. Resistance phenotypes and biochemical properties of native MβLs from three myxobacterial genera expressed in the model γ-proteobacterium, Escherichia coli, and comparisons with the New Delhi Metallo-β-lactamase reveal that evolution of the enzymes concerned mainly adaptations enhancing their production in the new hosts rather than functional differentiation. Periplasmic localization is the main evolved trait, with E. coli not recognizing optimally the signal peptides of the environmental MβLs which are secreted as soluble proteins in the cell envelope despite predicted to be lipoproteins. The unique features of N-terminal secretory signals of myxobacterial MβLs likely reflect the enzymes integration into the physiology of their hosts and point to naturally occurring molecules potentially controlling their expression.

Phylogenomic and functional analyses suggest that predatory myxobacteria had a key role in the early evolution of B1 metallo-β-lactamases and revealed the adaptations required for expression of resistance in enterobacterial pathogens.

Introduction

Enzymatic inactivation is the primary resistance mechanism against β-lactams in Gram-negative bacteria. Multidrug-resistant clinical strains, causing life-threatening infections in hospitalized patients, often express multiple β-lactamase genes, reducing the efficacy of last-resort β-lactams like carbapenems and newer cephalosporins. Through mobilization by promiscuous genetic structures, genes encoding class A, C, and D (serine-reactive α/β hydrolases) and class B (zinc metalloenzymes) β-lactamases have spread among clinically relevant γ-Proteobacteria, including Enterobacterales, Pseudomonas aeruginosa, and Acinetobacter baumannii, complicating antimicrobial chemotherapy and leading to treatment failures^1–3.

The rise of metallo-β-lactamases (MβLs), particularly subclass B1 (IMP, VIM, and NDM groups⁴), presents a critical challenge. These enzymes efficiently hydrolyze all β-lactams except monobactams (e.g., aztreonam) and evade inhibition by clavulanic acid as well as other class A inhibitors including the recently developed diazabicyclooctanes (e.g., avibactam) due to their distinct reaction mechanism⁵. Subsequently, MβL-producing pathogens resist most β-lactam-based therapies, including initial empiric treatments (e.g., piperacillin/tazobactam, carbapenems) and newer combinations like ceftazidime/avibactam. Alarmingly, strains co-producing MβLs with serine-reactive carbapenemases are increasingly prevalent^6–9.

On top of these, it has been proven difficult to design inhibitory compounds that would show specific activity for the whole class with limited toxicity for the host. MβLs share similar fold and active sites with various metallo-enzymes spanning all life’s kingdoms (MβL superfamily¹⁰). Increased toxicity due to off-target inhibition of human members of the superfamily usually results in failures of promising molecules during clinical evaluation^11,12. Nonetheless, recently developed molecules having a cyclic-boronate nucleus can inhibit the NDM, VIM and IMP subclass B1 MβLs along with serine reactive enzymes¹³. Two of them, taniborbactam and xeruborbactam, will most likely be utilized in the clinical practice^14–18. Yet, there are already enzyme variants exhibiting resistance to inhibition while subclass B2 and B3 MβLs appeared not to be affected by the compounds^16,17,19. Thus, alternative strategies are urgently needed.

Assessing the evolutionary history of the MβL genes may provide such therapeutic breakthroughs. Comparing environmental and clinical MβLs could reveal key adaptations enhancing catalytic efficiency, periplasmic localization, and stability²⁰ that would facilitate the research on novel inhibitors. Furthermore, identification of ancestral MβL hosts might aid in discovering natural inhibitory molecules.

MβL fold proteins with β-lactam hydrolytic activities are present in the genomes of divergent bacterial phyla with the identified families exhibiting extended “radiation”^21,22. Phylogenetic analyses have showed that subclass B1 and B2 enzymes are highly related and that the B3 family has emerged by an independent event from a different MβL fold ancestral protein^{23, 24}. Yet, the origins of the clinically relevant B1 subclass cannot be clearly deduced. Most likely this is due to the fact that some of them are expressed as class I integron gene cassettes (e.g. IMP and VIM) indicating that their mobilization from an ancestral chromosome is ancient. The NDM group of enzymes is transposon mobilized but again this gene transfer event is obscured with the gene or its homologues being sporadically identified in the chromosomes of distantly related environmental taxa^25,26.

During the ongoing monitoring of β-lactamase gene spreading in prokaryotic genomes, taking place at the Hellenic Pasteur Institute, we identified specific families exhibiting high carriage rates (>90%) for subclass B1 enzymes. The families belonged to the order Cytophagales²⁷ of the Bacteroidota phylum and the Cystobacterineae suborder^28–31 of the Myxococcota phylum. The phylogeny of the identified enzymes presented herein, followed that of their hosts, indicating that they are part of their core genome. Most of the identified B1 MβL families were located in Bacteroidota forming distinct lineages while the myxobacterial enzymes were located in the clade of VIM and NDM. The above phylogenomic analyses pointed to predatory myxobacteria as likely intermediate organisms during the transfer of ΜβLs form soil Bacteroidetes to environmental Pseudomonadota early in the genes’ evolution. In order to assess the adaptive mechanisms that accompanied this hypothetical gene transfer event we studied the expression of myxobacterial MβLs in E. coli by assessing their phenotypic and biochemical properties. Our data indicated that the evolution of subclass B1 MβLs concerns mainly adaptations that permitted their efficient expression and in vivo activity in γ-Proteobacteria and limited functional differentiation. The likely ancient acquisition of the genes by myxobacteria and the organisms’ rich secondary metabolism^32,33 indicated that naturally occurring molecules able to control the function of the enzymes may have been co-evolved with them. Preliminary assays testing the above hypothesis revealed that a known myxococcal antibiotic may weaken β-lactam resistance in β-lactamase producing E. coli.

Results

Horizontal gene transfer events explaining the spreading of B1 MβLs in the Prokaryotes

The current “snapshot” of the B1 ΜβL spreading in the Prokaryotes (Supplemental Material, Supplementary Tables 9 and 10) suggested the following: (i) Archaea are unlikely to produce subclass B1 MβLs, (ii) the enzymes exhibit a moderate propagation in Bacteria (6.16% of the published genomes and 6.26% of the sequenced species carry an orthologous gene), (iii) there is an active selection of these proteins as β-lactam resistance determinants as the frequency of pseudogenes and “zombie” enzymes is low (5.88% of the genomes and 5.82% of the bacterial species expressing for secreted B1 MβLs with intact zinc binding sites; data given throughout the following sections correspond solely to such resistance capable enzymes), (iv) Actinomycetota and most members of the Terrabacteria super-phylum are not producing the enzymes (with Bacillota being an exception), (v) Myxococcota and Bacteroidota are the most frequent carriers of true B1 MβL homologues (62.9% and 33.1% of the different species carry a true B1 homologue), (vi) families belonging to the above phyla and more specifically to the suborder Cystobacterineae and the order Cytophagales are exhibiting the highest observed carriage rates (Fig. 1A) and vii) the α, β and γ subclasses of Proteobacteria are not usually expressing such enzymes with only two families (Kordiimonadaceae; α-Proteobacteria and Shewanellaceae; γ-Proteobacteria) exhibiting frequencies above 40%.

Fig. 1. The spreading of B1 MβLs in bacterial genomes and ML phylogeny of the identified enzyme families.

A Bacterial families exhibiting carriage rates >40% in respect to the number of genomes or different species. Each point represents the observed % carriage rates. Error bars depict the 95% Wilson/Brown confidence intervals obtained through fraction of total analysis. The Cytophagia class contained most of the families with high carriage frequency both in respect to genomes and species (Cyclobacteriaceae, Cytophagaceae, Flammeovirgaceae, Flectobacillaceae, Fulvivirgaceae, Leadbetterellaceae, Reichenbachiellaceae, Roseivirgaceae, and Spirosomataceae). Among the other bacterial classes, Flavobacteriia (Crocinitomicaceae and Weeksellaceae) and Myxococcia (Archangiaceae and Myxococcaceae) exhibited frequent carriage of the gene, with the latter being characterized by the highest observed rate. B ML tree of the identified B1 MβL orthologous families (n = 1196). Nodes and branches are coloured based on bootstrap values. The colored strip indicates the bacterial classes that contained the respective protein cluster. Red bars indicate the number of members of each cluster, expressed as log(#members)+1, and green bars denote the number of different genera again transformed according to log(#genera)+1. Red triangles show the families having intact catalytic machineries (with empty triangles denoting that the family, apart from intact binuclear centers, also contained members with disrupted ones). Blue circles mark the families with predicted signal peptides and yellow stars mark the clusters that have been experimentally verified previously. The lineage of B2 ΜβLs (n = 15) is highlighted in purple and the NDM-MYX-VIM clade in ecru, with the latter one being also magnified in a pruned tree (C). Scale bars denote substitutions per site, as inferred by RAxML. Tree annotation was carried out in iTOL (https://itol.embl.de/shared/1GnaKLqwXTsA6).

When the identified sequences were clustered at 80% sequence identity, it became evident that each enzyme family was largely confined to a specific bacterial class, with the exception of the clinically relevant mobile MβLs. Therefore, most identified B1 MβL homologues appear to be stably maintained within populations of related bacterial groups. The majority of enzyme families were hosted by Bacteroidota, with Flavobacteriia and Cytophagia accounting for most of them (313 and 254 families, respectively; Supplemental Material, Supplementary Table 10). Pseudomonadota also exhibited considerable enzyme diversity, predominantly within γ-Proteobacteria (152 families; Supplemental Material, Supplementary Table 10).

To investigate potential horizontal gene transfer (HGT) events underlying the emergence of B1 MβLs in clinical bacteria, we constructed phylogenies using both maximum likelihood (ML) and Bayesian inference methods (Fig. 1B; Supplemental Material, Supplementary Fig. 11). We also included B2 MβL protein clusters, obtained using the same approach as for B1 enzymes. Both phylogenetic methods consistently positioned the B2 families within a lineage of B1 homologues lacking histidine at position 116, previously identified in some Spirochaetia and Clostridia³⁴. The ML phylogeny appeared more robust, placing this branch adjacent to the main lineage of bona fide B1 homologues from these bacterial classes (Fig. 1B; Supplemental Material, Supplementary Fig. 12). However, bootstrap analysis indicated low statistical support for most clades in the ML tree (Fig. 1B), and the Bayesian phylogeny similarly showed low confidence (average standard deviation of split frequencies = 5.9% across three independent runs of 6 ∙ 10⁷ generations with three Markov Chain Monte Carlo chains; Supplemental Material, Supplementary Fig. 11). Therefore, the current bacterial genomic data are insufficient to enable a well-resolved reconstruction of the B1 MβL phylogeny.

Nonetheless, some meaningful insights could be drawn, as both phylogenies shared several common topological features (Fig. 1; Supplemental Material, Supplementary Figs. 11, 12). Almost all B1 orthologous groups from Bacteroidota were distributed into two distant lineages (L1 and L2; Fig. 1). Enzyme families from other phyla were located between these lineages, again forming distinct clusters. The first cluster included enzymes from γ-Proteobacteria—mainly from the order Alteromonadales—as well as clinically relevant enzymes such as IMP and other integron-borne families spread in Pseudomonas, Acinetobacter and Enterobacterales (e.g., DIM, GIM, SIM^35–37). This cluster encompassed several families present across multiple genera, suggesting a high degree of mobility (Fig. 1B). The second cluster included the clinically significant resistance determinants NDM and VIM, which grouped into a distinct branch along with enzymes from α-, β-, and γ-Proteobacteria, as well as from Myxococcota (Fig. 1B). Although positioned differently in the ML and Bayesian trees, the Bacilli lineage was closely related to the Bacteroidota L2 lineage in both analyses.

The inferred phylogenies, together with the high diversity and carriage rates of B1 MβLs in Bacteroidota classes, underscore the importance of this phylum in the gene’s evolution. The validity of our approach was confirmed as it was able to reproduce what is known regarding the origins of class A β-lactamases. The above enzymes are thought to have originated in soil-dwelling Actinobacteria³⁸, known for producing a wide array of β-lactams (e.g., carbapenems, cephamycins, and clavams), with many corresponding biosynthetic clusters encoding class A β-lactamase homologues^39–42. Our results showed that Actinomycetes not only exhibited high carriage rates of these genes but also hosted the greatest diversity of true class A homologues (1,996 families out of 5,212 in total; Supplementary Data 5) compared to other bacterial classes. Phylogenetic analyses using ML revealed that class A enzymes from Actinomycetes form a monophyletic lineage, which gave rise to various mobile gene families (Supplemental Material, Supplementary Fig. 13). In contrast, B1 MβLs from Bacteroidota classes exhibit a paraphyletic origin, with most representatives distributed across both the L1 and L2 lineages (Fig. 1). Therefore, the origins of B1 enzymes remain unclear and may trace back to bacterial phyla that lack sufficient sequencing data. Nonetheless, our phylogenomic data suggest a well-adapted mobilome within Bacteroidota, particularly among soil-associated members such as Cytophagia, Chitinophagia, and Flavobacteriia. These groups likely served as donors of B1 enzymes to Proteobacteria.

These HGT events likely involved intermediate steps, as Bacteroidota and Proteobacteria are distantly related, and newly transferred enzymes would have faced challenges achieving functional expression in the new host. Indeed, heterologous expression of some Bacteroidota-derived B1 MβLs in E. coli demonstrated poor expression and weak resistance phenotypes^43,44. Among neighbouring soil bacteria, the Cystobacterineae suborder of Myxococcota (which save from Anaeromyxobactereaceae are predatory organisms) emerged as the most likely intermediate host. This group showed the highest carriage rates, and its B1 orthologs formed a monophyletic lineage closely related to the VIM and NDM enzymes. Notably, expression of MβLs from Myxococcus xanthus (MYX-1), Stigmatella aurantiaca (STA-1), and Anaeromyxobacter dehalogenans (ANA-1) in E. coli demonstrated functional production in γ-Proteobacteria²¹.

Using RNA polymerase β-subunit (RpoB) as a phylogenetic marker, we estimated the evolutionary distances between bacterial classes carrying B1 MβLs and Klebsiella pneumoniae. Myxobacteria occupied an intermediate phylogenetic position between Bacteroidota and Proteobacteria, as did Bacillota (Fig. 2). However, Bacillota showed lower carriage rates and gene diversity, suggesting more recent acquisition of B1 enzymes in this group. The most likely entry point of B1 enzymes into Pseudomonadota was the α-Proteobacteria class. Although rare carriers, α-Proteobacteria exhibited high gene diversity, likely reflecting the influx of novel families—a pattern not observed in β-Proteobacteria (Fig. 2). Furthermore, B1 orthologs from α-Proteobacteria clustered within the MYX-NDM-VIM branch (Fig. 1C), reinforcing this hypothesis.

Fig. 2. Hypothetical horizontal transfer events of MβL genes between bacterial classes.

Three-dimensional graph combining carriage rates (B1 orthologous proteins with intact zinc sites and signal peptides per species), enzyme richness and phylogenetic distance to K. pneumoniae of bacterial classes that have participated in the spreading of Β1 ΜβLs. Dotted lines denote likely HGT events between the various bacterial classes. Based on these data, a plausible route for the spreading of the enzymes in environmental microbiota is Cytophagia → Myxococcia → α-Proteobacteria → γ-Proteobacteria and then to clinical strains of the latter class. Due to their low carriage rates and gene richness Bacilli and β-Proteobacteria most likely had a lesser role in the above HGT events.

B1 MβLs belong to the core genome of Myxococcaceae

The central role of predatory myxobacteria in the proposed HGT route is further supported by the observation that MβL genes are part of the core genome of at least one family within this group. Specifically, 96.3% (95% Confidence Interval: 99.0–89.8%) of sequenced species in the Myxococcaceae family harbour a B1 orthologue. Other Cystobacterineae families also commonly carry B1 MβL genes (e.g., Archangiaceae 34.4%; Anaeromyxobacteraceae 28.5%).

Phylogenetic analyses indicate that B1 MβLs from Myxococcaceae are closely related to enzymes from Archangiaceae, unlike those from the other Cystobacterineae families. Nonetheless, all these proteins cluster within the same branch, supported by a bootstrap value of 14% (Fig. 3A). Most identified myxobacterial B1 MβLs are predicted to be covalently anchored to the periplasmic side of the outer membrane secreted via a type II signal peptide, a feature also shared with NDM enzymes⁴⁵. Interestingly, the MYX-1 enzyme (cluster B1_264; Fig. 3A) was also found in a genome initially classified as Comamonas (a β-Proteobacterium). However, PGAP’s Average Nucleotide Identity (ANI) analysis of the corresponding assembly (GCF_016735485) confirmed it to be a misclassified Myxococcus strain, not a Comamonas, highlighting annotation errors that can occur even in reference databases.

Fig. 3. Subclass B1 ΜβLs are part of Myxococcaceae core genome.

A Sub-tree extracted from the large scale ML phylogeny of B1 ΜβLs (Fig. 1) focusing on the lineage of MYX. The enzymes exhibit a monophyletic spreading in Myxococcia. Enzymes from Myxococcaceae are clustered into two branches with the first containing B1 ΜβLs from Myxococcus and Pyxidicoccus and the second from Corallococcus and Citreicoccus. The Myxococcaceae lineage contained also most of the enzymes identified in Archangiaceae (Hyalangium, Archangium, Cystobacter, Vitiosangium, Mellitangium) with Stigmatella forming a separate branch. Five families found in the chromosomes of α-Proteobacteria appeared to be related with the above enzymes and in fact they were nearer compared to members from Anaeromyxobacteraceae and Vulgatibactereceae. Most of the enzymes found in Myxococcia are predicted to be secreted via a type II signal peptide (SPII). B1 ΜβLs from Anaeromyxobacter are secreted through type I signal peptides (SPI) while in the related α-proteobacterial enzymes posttranslational translocation is most likely facilitated by the twin arginine system (TAT). B Comparison of RpoB and B1 ΜβL ML phylogenies of reference myxobacterial species and the strains isolated in the present study. The phylogenetic marker from Labilithrix luteola (Polyangia) and the clinical ΜβLs – NDM-1, VIM-1, IMP-1, SPM-1 – have been included as outliers. Enzymes from Myxococcceae followed the phylogenies of their hosts. Imipenem hydrolysis yielded by the cell-free extracts of Myxococcus sp. 1LA and Pyxidicoccus sp. 3LFB2 is depicted in the lower left of the figure. The extract of the first strain contained 9.9 units per mg specific imipenemase activity and of the second 177 units per mg. In both cases, EDTA inhibited hydrolysis of imipenem. C Mauve generated alignment of the 50 Kb genomic regions surrounding the B1 ΜβL gene (highlighted in red) from reference Myxococcus species. Though a degree of conservation is observed, several recombination events have shaped the respective loci, especially in the downstream region.

To account for such misclassifications and mitigate potential biases due to the overrepresentation of certain myxobacterial families in public databases, we also analyzed a dataset of 69 unpublished Cystobacterineae genomes provided by the Helmholtz Institute of Biological Sciences (Saarland, Germany; prof. Rolf Müller personal communication). Most of these genomes were not assigned to specific genera. Our analysis revealed that 68.1% of these isolates carry B1 MβL genes, which clustered closely with those identified in public datasets. To further validate these findings, we isolated myxobacterial strains from Greek soil samples and obtained four axenic cultures identified as members of Cystobacterineae via Sanger sequencing of 16S rRNA genes (Supplemental Material, Supplementary Fig. 14). Whole genome sequencing and ANI analyses classified the isolates into Myxococcus sp. (isolate 1LA), Corallococcus sp. (isolate 4LFB) and Pyxidicoccus sp. (isolates 3LFB2 and 3LG). Based on RpoB phylogeny (Fig. 3B) Myxococcus sp. 1LA was closely related to M. macrosporus DSM 14697 (deposited as Corallococcus macrosporus DSM 14697 in NCBI⁴⁶), Corallococcus sp. 4LFB to the C. exiguus complex⁴⁷, Pyxidicoccus sp. 3LFB2 to P. trucidator CA060A⁴⁸ and, Pyxidicoccus sp. 3LG to P. xiebeiensis QG1ED-7-1⁴⁹. Using HMM-based genome scanning, we confirmed that all four isolates carried B1 MβL homologues (Fig. 3B). In addition, Corallococcus sp. 4LFB also carried a B3 MβL gene, while both Pyxidicoccus isolates harboured class D β-lactamase genes, which, unlike B1 enzymes, they do not form monophyletic groups (Supplementary Data 1). Imipenem hydrolysis assays using cell-free lysates from these strains demonstrated enzymatic activity, which was inhibited by EDTA, confirming the production of active MβLs (Fig. 3B).

Comparative phylogenetic analyses of RpoB and B1 MβLs among Myxococcaceae reference strains and our newly isolated genomes showed a strong correlation (Fig. 3B; Supplemental Material, Supplementary Fig. 15). In contrast, although B1 MβLs from Archangiaceae generally followed RpoB-based phylogeny, Hyalangium MβLs clustered with those from Myxococcus and Pyxidicoccus in the large-scale tree, rather than within the family of the genus, suggesting HGT from Myxococcaceae to Archangiaceae.

Genomic context analysis of regions flanking the MβL genes (25 Kb upstream and downstream) revealed no known mobile genetic elements that could facilitate such transfers. Furthermore, analysis using IntegronFinder 2.0 showed no typical attC sites in the 3′ intergenic regions of myxobacterial MβL genes, ruling out integron-mediated mobility. Although the MβL loci in Myxococcaceae showed a degree of conservation (Fig. 3C), the downstream regions were highly mosaic, with insertions and deletions indicating recombination events. While we cannot exclude the possibility of recent HGT from related donors, the overall data suggest an ancestral acquisition of B1 MβL genes within this lineage. This notion is further supported by the presence of these enzymes in Anaeromyxobacteraceae. The above family is physiologically and genomically distinct from the myxospore-forming and predatory Cystobacterineae^50,51 and yet, still harbours B1 enzymes that fall within the myxobacterial clade.

Expression of myxobacterial MβLs into a γ-proteobacterium

Assuming myxobacteria had a central role in the spreading of the genes in Pseudomonadota, heterologous expression into a γ-proteobacterium would reveal the adaptations that were associated with the underlying HGT events. The open reading frames of the B1 MβL genes from Myxococcus xanthus DK1622 (bla_MYX-1), Myxococcus sp. 1LA (bla_MYX-2), Corallococcus sp. 4LFB (bla_COR-1), Pyxidicoccus sp. 3LFB (bla_PYX-1) and Pyxidicoccus sp. 3LG (bla_PYX-2) as well as the clinical NDM-1 were isogenically produced in E. coli DH5α using the pZE21-MCS vector.

The β-lactam susceptibility profiles of the respective clones showed that the myxobacterial enzymes exhibited a typical MβL phenotype when expressed in a γ-proteobacterium. This phenotype was relatively weak, especially against oxyimino cephalosporins (cefotaxime, ceftazidime, and cefepime), cephamycins (cefoxitin), and carbapenems (meropenem and imipenem). They showed reduced susceptibility to meropenem, which was reversed by metal chelators, yielding positive results for MβL production in disk synergy tests. Minimum Inhibitory Concentration (MIC) estimations for the main β-lactam classes indicated that the environmental enzymes conferred much weaker resistance to all tested drugs compared to clinical NDM-1, with values more than two serial dilutions lower (Table 1). Phenotypic differences among myxobacterial MβLs primarily concerned cephalosporins, with the Corallococcus enzyme (COR-1) having marginal effects even against the first-generation drug cephalothin (Table 1). Enzymes from the Pyxidicoccus lineage (PYX-1 and PYX-2) were able to confer resistance to the oxyimino cephalosporin ceftazidime when expressed in E. coli, unlike MβLs from other myxobacterial genera (Table 1). Additional deviations included carbapenem susceptibility, with PYX-1 yielding the highest MIC for both imipenem and meropenem, followed by MYX-2 (Table 1).

Table 1.

β-Lactam susceptibility profile of E. coli DH5α clones expressing myxobacterial metallo-β-lactamases and NDM-1 under isogenic conditions^a

β-lactam	Minimum Inhibitory Concentration (μg/mL)
	pZE21-blaNDM-1	pZE21-blaMYX-1	pZE21-blaMYX-2	pZE21-blaCOR-1	pZE21-blaPYX-1	pZE21-blaPYX-2	pZE21-blarecNDM-1	pZE21-blarecMYX-1	pZE21-MCS
	NDM-1	MYX-1	MYX-2	COR-1	PYX-1	PYX-2	recNDM-1	recMYX-1	none
Ampicillin	>1024	512	512	1024	1024	256	1024	1024	4
Ampicillin + Clavulanateb	>1024	512	512	1024	1024	256	512	1024	4
Piperacillin	256	8	32	16	32	8	32	16	≤1
Cephalothin	>1024	64	128	8	128	32	512	256	≤2
Cefoxitin	512	8	4	≤2	16	4	64	8	≤2
Cefotaxime	128	2	4	0.125	8	0.5	32	4	≤0.0625
Ceftazidime	>128	2	1	0.5	16	16	>128	4	≤0.25
Cefepime	>32	≤0.0625	≤0.0625	≤0.0625	0.125	≤0.0625	4	≤0.0625	≤0.0625
Aztreonam	≤0.03125	≤0.03125	≤0.03125	≤0.03125	≤0.03125	≤0.03125	≤0.03125	≤0.03125	≤0.03125
Imipenem	2	0.125	0.25	0.25	1	0.5	0.125	0.125	≤0.03125
Meropenem	8	≤0.03125	0.5	≤0.03125	1	0.0625	0.0625	≤0.03125	≤0.03125

aDeterminations were performed in independent experiments starting from different cultures (biological replicates, n = 3) with the obtained MICs differing less than two serial dilutions from the reported values.

^bampicillin MIC with clavulanate at a fixed concentration of 4 μg/mL;

In order to assess if the weak β-lactam resistance conferred by myxobacterial enzymes when expressed in E. coli stem from the different optimum growth temperatures (30–32 ^oC for myxobacteria and 37 ^oC for E. coli) we examined their production at 30 and 37 ^oC under inducible expression in E. coli C600Z1. Our data indicated that myxobacterial MβL production was comparable at both conditions, with the higher temperature generally associated with more enzyme in the extracts for most of the induction time course (Fig. 4A).

Fig. 4. Expression of myxobacterial MβLs in E. coli.

A Effects of temperature on production of native myxobacterial MβLs in E. coli C600Z1 under the pZE21 expression module during an induction time-course at 30 and 37 ^oC. The enzymes could be readily produced at both temperatures with equivalent yields. B Production of native NDM-1 and MYX-1 under the E. coli C600Z1/pZE21 system at 37 ^oC and comparison with isogenic clones of their recombinant forms where their signal peptides have been replaced with that from PelB (recNDM-1 and recMYX-1). The clinical enzyme yielded extracts with weak meropenemase activity, especially when expressed with its native amino terminus. In A and B, data from three technical replicates are shown. C, D Production of recNDM-1 and recMYX-1 under the E. coli C600Z1/pZE21 and E. coli BL21(DE3)/pET26b(+) expression systems. C Total meropenem hydrolysis units of cell-free extracts and culture supernatants of plain LB broth and supplemented with 1% w/v glycine or 1% w/v glycine plus 0.1 M NaCl. Error bars represent means ± standard deviations of results from three biological replicates (n = 3). Total enzyme production (cell-free extract + culture supernatant) between each condition and clone was compared through one-way analysis of variance, implementing Holm-Sidak’s multiple comparisons test. Results from pairwise comparisons between recMYX-1 and recNDM-1 clones at the same conditions, as well as of the pET26b(+)-bla_recNDM-1 in the absence and presence of glycine, are shown (***: p < 0.001, ns: not significant). D β-Lactam resistance levels of the respective clones induced with anhydrotetracycline or IPTG in MH broth with and without 0.25 mM ZnSO₄. The doubling serial dilution differences between MICs measured for the pZE21 and pET26b(+) clones are given for each enzyme type. Error bars correspond to means ± standard deviations of doubling serial dilution differences observed across three biological replicates (n = 3) performed with different starting cultures. E SDS-PAGE analysis of purified enzymes. 3 – 5 μg of each enzyme has been loaded on the gel. Native myxobacterial ΜβLs could be readily purified from cell-free extracts after four chromatography steps. The recombinant recMYX-1 and recNDM-1 have been purified from culture supernatants.

Localization in the cell envelope of E. coli may also deviate between myxobacterial MβLs and NDM-1 due to different N-terminal signal peptides. Indeed, while SignalP predicted that the myxobacterial enzymes are secreted via a type II signal peptide (SP), reanalysis with LipoP, while confirming the presence of a lipobox, also yielded several alternative cleavage sites for signal peptidase I, some with equal or higher probability scores (Supplemental Material, Supplementary Fig. 16). This feature was common in most myxobacterial MβLs identified in our genome database analyses. In contrast, LipoP reanalysis of the NDM-1 sequence revealed only a single type II SP.

To account for the above SP differences, we replaced the respective sequences in MYX-1 and NDM-1 with those of pectate lyase B (PelB) from Erwinia carotovora, which promotes soluble periplasmic expression in E. coli. Isogenic expression of the recombinant enzymes (recMYX-1 and recNDM-1) alongside the native ones in E. coli DH5α showed that signal peptide replacement in NDM-1 significantly lowered resistance levels, while having marginal effects on MYX-1 (Table 1). Even with a common type I SP, the clinical enzyme still yielded higher MICs, though the differences were less pronounced, except for newer cephalosporins and cefoxitin (Table 1).

Signal peptide replacement increased enzyme recovery in cell lysates at 37 ^oC at the final time point for both MYX-1 and NDM-1 (Fig. 4B). Notably, both forms of NDM-1 yielded much less active extracts compared to MYX-1. This trend reversed when we switched to the pET26b(+) - E. coli BL21(DE3)pLysS expression system. Here, the soluble recNDM-1 was produced in much higher quantities than the recMYX-1, especially under conditions favouring extracellular secretion of periplasmic proteins (i.e., glycine and NaCl presence in the culture broth; Fig. 4C). These differences were also reflected in the resistance levels of the respective clones (Fig. 4D).

As myxobacterial MβLs were enriched in E. coli cellular extracts and appeared to be in a soluble form, even when expressed with their native sequences, we proceeded to purify them for further characterization. Relatively homogeneous final preparations (>90% purity; Fig. 4E) after four steps of liquid chromatography were obtained for all enzymes, with yields around 1.5 mg per liter of culture.

Purified preparations of MYX-1 and MYX-2 were analyzed via tandem mass spectrometry (MS/MS). In both cases, neither the N-terminal peptide of the unprocessed nascent protein nor the fragment following the type II cleavage site (containing type I peptidase recognition motifs) was detected (Supplemental Material, Supplementary Fig. 17A). The most N-proximal peptide we detected was AQSQGPDEYVLADDVSVRKLAP, strongly suggesting that the main species of purified enzymes corresponded to signal peptidase I-processed mature forms. Several detected peptides, beyond their native state, were also found with oxidized hydrophobic residues. These included Val63, Trp64, Val69, Trp109, Trp192, Val221, and Trp224, all positioned in the enzyme’s interior as indicated by a three-dimensional homology model of MYX-1 (Supplemental Material, Supplementary Fig. 17B). Since no protein denaturation was observed during the purification process, oxidation most likely occurred in vivo.

This in cell oxidation would imply the presence of unfolded forms in the periplasm that could induce inclusion body formation through aggregation. Indeed, extended membrane solubilization of pellets obtained after cell lysis revealed insoluble material that, when analyzed by SDS-PAGE, was enriched in a protein with an apparent molecular mass similar to the studied enzymes (Supplemental Material, Supplementary Fig. 18). While inclusion body formation might explain the weak resistance phenotypes, most likely this did not account for the phenotypic differences. NDM-1 may also form inclusion bodies when overexpressed in a soluble form as growth in the presence of factors known to improve the expression of inclusion body-prone proteins in E. coli (e.g., glycine and NaCl^52–54;) increased the total produced amount of both recMYX-1 and recNDM-1 enzymes (Fig. 4C). The recombinant form of MYX-1, purified from the culture supernatant under these growth conditions, was recovered at a higher yield (~3 mg per L of culture) compared to the native enzyme isolated from cell free lysates using the same chromatographic procedure (Fig. 4E). Soluble NDM-1 was also readily purified from culture supernatants (>10 mg per L of culture; Fig. 4E). The purified native and recombinant MYX-1 exhibited the same electrophoretic mobility in SDS-PAGE, yielding nearly identical apparent molecular masses (Fig. 4E).

Cellular localization of myxobacterial MβLs in E. coli

To better understand how myxobacterial MβLs are secreted in E. coli, we conducted localization studies using cellular fractionation. Partitioning of the carbapenemase activity in the various cellular compartments obtained at each step was utilized for enzyme detection. Three sequential fractionation steps were carried out concerning: (i) spheroplast formation, through a standard sucrose/lysozyme protocol^55,56, (ii) total membrane isolation after spheroplast disruption and (iii) inner and outer membrane separation using isopycnic ultracentrifugation in a sucrose gradient (Fig. 5A).

Fig. 5. E. coli fractionation and ΜβL localization.

A The partitioning of carbapenemase activity between the fractions obtained at each step was used to assess where the majority of each enzyme is localized. Data are presented as log ratios of the total meropenem hydrolysis units (nmol of substrate hydrolyzed per minute) measured for the two fractions of each step. Error bars denote means ± standard deviations of log ratios obtained during independent fractionation experiments using different starting cultures (biological replicates, n = 3). Statistical analysis was performed through multiple comparisons via one-way analysis of variance using Holm-Sidak’s test. Results from comparisons between the native NDM-1 and each of the remaining enzymes are shown. For the IM/OM fractionation, comparisons between pairs with meaningful meropenemase activity in membrane fractions are solely given (***p < 0.001, *p < 0.05). The table (B) lists the actual units of a typical experiment, which are analogous to the quantity of functional enzyme in each preparation during fractionation. Myxobacterial ΜβLs exhibited the same partitioning profile as the recombinant recMYX-1 and recNDM-1, which are secreted via a type I SP, in contrast to NDM-1, which was typically enriched in membrane preparations and displayed elevated levels in the OM fraction of the respective clone. C Though myxobacterial ΜβLs are predicted to be lipoproteins by SignalP, their type II SP (pink rectangles) differ significantly from the respective NDM-1 lipobox. Our data indicated that the LipoP predicted type I SP (purple rectangles) of myxobacterial enzymes are functional in E. coli while outer membrane localization may be obstructed, most likely during cleavage of type II SP by the Lsp peptidase. The pathways for post-translational Sec-dependent soluble periplasmic secretion and outer membrane localization of E. coli proteins are also shown. Created in BioRender. Kotsakis, S. (2026) https://BioRender.com/pdw63hr.

At each step, the carbapenemase activity of myxobacterial MβL clones was higher in the fractions lacking membranes (i.e., periplasm at fractionation step 1 and cytoplasm at fractionation step 2), similar to the SPI secreted recMYX-1 and recNDM-1. In contrast the lipobox containing NDM-1 was always enriched in the membrane containing fractions (Fig. 5A). Although a notable amount of enzyme was detected in the membrane fraction for some myxobacterial MβLs (e.g., MYX-2 and COR-1) this corresponded to unprocessed enzyme species still embedded in the inner membrane via their signal peptides, as indicated by further separation of cytoplasmic and outer membranes (Fig. 5B). Meaningful amounts of carbapenemase activity in the OM fractions were only detected for the NDM-1 enzyme (Fig. 2B).

Our cellular fractionation data indicate that the N-terminal sequences of myxobacterial MβLs favour soluble periplasmic expression in E. coli rather than outer membrane localization. The predicted type II signal peptides of myxobacterial MβLs differ significantly from that of NDM-1 (Fig. 5C). They are considerably shorter, and the residue three positions upstream of the N-terminal cysteine of the processed mature enzymes (which are covalently attached to membrane lipids) is not a leucine.

The shorter length suggests that the cleavage site of signal peptidase II (Lsp) might be buried well within the inner membrane during enzyme topogenesis. However, this might not hinder recognition, as structural studies have shown that the enzyme catalyzes intramembrane proteolysis⁵⁷. The absence of a hydrophobic amino acid at position -3 of myxococcal lipoboxes, however, may be detrimental for recognition and catalysis by signal peptidase II, as mechanistic studies have highlighted the importance of this residue⁵⁷. Nonetheless, since enzymes from the remaining genera contain a hydrophobic amino acid at this position, the failure for membrane localization may result from the impairment of the first step of cysteine lipidation concerning the addition of diacylglycerol (DAG) by lipoprotein diacylglyceryl transferase (Lgt; Fig. 5C). This initial lipidation is considered essential for recognition by Lsp^58,59. Assuming membrane anchoring is carried out unobstructed, then a possible explanation for our data would be that the type I SP peptide in IM-bound MβLs is cleaved by LepB, before the transfer of the enzymes to the OM by the Lol system (ref. ⁶⁰; Fig. 5C), ultimately resulting in soluble periplasmic localization.

Myxobacterial MβLs are potent β-lactam hydrolases

In order to assess the degree of correlation of our observations from expression studies with the functional properties of the studied MβLs, we measured their steady state kinetic constants during hydrolysis of the main β-lactam classes using purified preparations (Table 2; Supplemental material, Supplementary Fig. 19).

Table 2.

Steady state kinetic constants of the studied β-lactamases during hydrolysis of various β-lactams^a

β-lactamase	Kinetic Constant	Benzylpenicillin	Piperacillin	Cephalothin	Cefoxitin	Cefotaxime	Ceftazidime	Cefepime	Imipenem	Meropenem
MYX-1	kcat (s-1)	44	137	120	6	41	>6	>0.6	47	92
	Km (μM)	<5	31	13	68	7	>100	>140	13	42
	kcat/Km (M-1∙s-1)	>8.8 ∙ 106	4.4 ∙ 106	9.2 ∙ 106	8.8 ∙ 104	5.9 ∙ 106	5.9 ∙ 104	5.1 ∙ 103	3.6 ∙ 106	2.2 ∙ 106
MYX-2	kcat (s-1)	40	79	110	9	9	>3	>0.5	105	268
	Km (μM)	<5	10	5	233	4	>100	>140	41	73
	kcat/Km (M-1∙s-1)	>8.0 ∙ 106	7.9 ∙ 106	2.2 ∙ 107	3.8 ∙ 104	2.3 ∙ 106	2.9 ∙ 104	3.2 ∙ 103	2.6 ∙ 106	3.7 ∙ 106
COR-1	kcat (s-1)	257	>105	1.5	0.04	0.4	>0.6	>0.03	175	362
	Km (μM)	65	>100	2	19	4	>100	>140	112	180
	kcat/Km (M-1∙s-1)	4.0 ∙ 106	1.2 ∙ 106	8 ∙ 105	2 ∙ 103	1.0 ∙ 105	7.0 ∙ 103	2.0 ∙ 102	1.2 ∙ 106	2.0 ∙ 106
PYX-1	kcat (s-1)	56	130	15	1	25	14	>0.2	184	30
	Km (μM)	6.1	64	1.5	2	9	91	>140	42	7
	kcat/Km (M-1∙s-1)	9.2 ∙ 106	2.0 ∙ 106	1.0 ∙ 107	5.0 ∙ 105	2.8 ∙ 106	1.5 ∙ 105	1.4 ∙ 103	4.4 ∙ 106	4.3 ∙ 106
PYX-2	kcat (s-1)	44	45	8	0.9	4	14	>0.07	42	28
	Km (μM)	21	54	8	73	10	93	>140	9	14
	kcat/Km (M-1∙s-1)	2.1 ∙ 106	8.0 ∙ 105	1.0 ∙ 106	1.2 ∙ 104	4.0 ∙ 105	1.5 ∙ 105	4.0 ∙ 102	4.7 ∙ 106	2.0 ∙ 106
recMYX-1	kcat (s-1)	39	134	122	5	38	>6	>0.7	40	98
	Km (μM)	<5	30	8	67	4	>100	>140	16	36
	kcat/Km (M-1∙s-1)	>8.2 ∙ 106	4.5 ∙ 106	1.5 ∙ 107	7.4 ∙ 104	9.5 ∙ 106	6.0 ∙ 104	5.7 ∙ 103	2.5 ∙ 106	2.7 ∙ 106
recNDM-1	kcat (s-1)	467	190	36	7	27	26	14	130	61
	Km (μM)	118	38	2	12	2	6	98	81	26
	kcat/Km (M-1∙s-1)	4.0 ∙ 106	5.0 ∙ 106	1.8 ∙ 107	5.8 ∙ 105	1.4 ∙ 107	4.3 ∙ 106	1.4 ∙ 105	1.6 ∙ 106	2.3 ∙ 106

a: Data are the means of two technical replicates differing by less than 15%.

Data indicated that myxobacterial MβLs are potent β-lactam hydrolases, especially against substrates with structurally similar natural counterparts (e.g., penicillins, older cephalosporins, and carbapenems). Although myxobacterial MβLs share limited sequence identity with NDM-1 (33–37% identities; Fig. 6A), the majority of them exhibited equal or even higher catalytic efficiency compared to the clinical enzyme during hydrolysis of benzylpenicillin, piperacillin, cephalothin and carbapenems (imipenem and meropenem). For most reactions, the measured k_cat/K_m were above 10⁶ M^-1·s^-1 - i.e., near the diffusion limit (10⁷–10⁸ M^-1·s^-1) - approaching thus catalytic perfection towards these substrates. The 7α-methoxy substitution in the cephalosporin nucleus appeared to obstruct recognition by myxobacterial enzymes, as the majority of them were characterized by high K_m values during their interactions with cefoxitin, in contrast to what was observed for NDM-1. PYX-1, while exhibiting a higher affinity for this substrate, was slowly turning over it. A low k_cat was also evident for PYX-2 and COR-1.

Fig. 6. Structural comparisons between environmental ΜβLs and NDM-1.

A Primary structure alignment of myxobacterial MβLs (the sequence of the mature proteins is shown) and NDM-1 (portion of the enzyme contained in the crystal structure 6KXI). The secondary structure of myxobacterial enzymes was based on the AlphaFold structure prediction. The position of conserved residues forming the two zinc binding sites (i.e., Zn1: His116, His118 and His196; Zn2: Asp120, Cys221 and His263) is indicated on the alignment. The discussed secondary structural elements are also highlighted following the nomenclature used for the B. cereus BII MβL crystal structure. B Superimposition of the three-dimensional models of myxobacterial ΜβLs and the crystal structure of NDM-1. The backbone atoms are shown as ribbons and active site residues as sticks, focusing either on the active site (left) or on the L8 loop and downstream elements (right). In NDM-1 the latter segment was less structured, indicating a higher flexibility. C Molecular surface of the studied enzymes colored based on Poison-Boltzmann electrostatic potential. The clinical enzyme was characterized by a wide and negatively charged region in the vicinity of Zn(II) centers. In myxobacterial enzymes, the negative electrostatic potential in the respective area was weaker.

While the oxyimino cephalosporin cefotaxime was efficiently hydrolyzed by most environmental enzymes, with COR-1 and PYX-2 showing low turnover rates, the bulkier ceftazidime and cefepime (which have positively charged R2 side chains) were poor substrates. The binding of these molecules in the active site appeared to be impaired, as Michaelis constants were elevated and could not be accurately measured. Enzymes from the pyxidicocci lineage appeared to recognize ceftazidime more productively, exhibiting a two- to five-fold higher k_cat/K_m compared to myxococcal MβLs. The clinical NDM-1 exhibited one to two logarithms higher efficiency during ceftazidime and cefepime hydrolysis (e.g., 72 and 25 times more efficient than MYX-1).

In an initial effort to assess the structural basis of these differences, we compared AlphaFold3-generated three-dimensional models of myxobacterial MβLs (co-folded with two zinc ions) with a crystal structure of NDM-1. The residues forming the two zinc centers exhibited relatively similar positioning in the compared structures (Fig. 6B, left panel). However, the backbone of loops containing second-sphere residues, known to affect β-lactam binding and hydrolysis by B1 MβLs, showed deviations. The differences were less pronounced in L10, which participates in interactions with the C3/C4 carboxylate of β-lactams⁶¹. Differences were sounder in L3 and L8, which also influence substrate binding (refs. ^62,63; Fig. 6B). In Pyxidicoccus and Corallococcus enzymes, L3 adopts a helical conformation and is positioned away from the active site, while the backbone of L8 is also located further from the zinc ions in all environmental MβLs (Fig. 6B). The dynamics of L8 will likely be altered, as the subsequent segment is shorter in and includes an additional β-chain (B9; also present in Bacillus cereus BcII MβL⁶⁴) that forms an antiparallel sheet with B10. This feature is absent in NDM-1, suggesting that the enzyme would be more flexible in this region (Fig. 6A, 6B).

Myxobacterial enzymes exhibited a different composition of charged amino acids, particularly in the regions preceding the His116-His118-Asp120 and His263 catalytic residues, which were rich in arginine (Fig. 6A). Indeed, estimation of the Poisson-Boltzmann surface electrostatic potentials indicated that NDM-1 has a concentration of negative charges around the zinc ions in the region beneath L8 (Fig. 6C). In myxobacterial enzymes, the negative potential in that area is weaker (Fig. 6C), which may contribute to the impaired binding of oxyimino cephalosporins with a positively charged R2. Another effect of electrostatic potential differences in the active site’s vicinity could be the alteration of protonation equilibria of amino acids coordinating Zn(II), with implications on the ions’ binding strength and electronic structure. Myxobacterial MβL models were characterized by narrower active sites, especially regarding the accessibility of zinc centers, with COR-1 exhibiting the most obstructed one (Fig. 6C), which could account for the high K_m against bulky substrates.

In vitro – in vivo gap in myxobacterial MβLs

The catalytic properties of the studied enzymes didn’t fully explain the resistance phenotypes observed when they were expressed in various E. coli genetic backgrounds. Such in vitro–in cell discrepancies are common in MβL studies, often attributed to downstream protein processing and periplasmic homoeostasis rather than coding sequence features (e.g., codon usage and GC levels^20,65–68). For instance, NDM-1’s membrane localization has been shown to enhance its resistance to the Prc periplasmic protease under zinc-deprived conditions⁵⁶. Indeed, replacing NDM-1’s native SP with PelB’s resulted in weaker resistance phenotypes. Assuming myxobacterial MβLs are similarly susceptible to periplasmic degradation, their inability to localize in the outer membrane—despite possessing a lipobox—appears to be the primary cause of the observed deviation.

Still, our results showed that the type I SP of the environmental enzymes is operational in E. coli, typically yielding high quantities of soluble, functional molecules in the periplasm (Fig. 5B; PYX-2 is an exception due to reduced growth rates of its respective clone). Therefore, resistance levels shouldn’t be so low for certain antibiotics. This was also clearly demonstrated by comparing NDM-1 and MYX-1 when expressed under the same type I SP. Although their catalytic efficiency during cefotaxime and carbapenem hydrolysis was similar, the environmental enzyme exhibited a more than two serial dilution lower MIC. This was evident even with an expression system that doesn’t favour NDM-1 production (i.e., pZE21/E. coli C600Z1; Table 1 and Fig. 4).

Previous studies have shown that considering the relative kinetic stability of MβLs in the periplasm can improve the correlation between functional properties and resistance levels⁶⁹. Here, we assessed the kinetic stability of the studied enzymes in the periplasm using preparations obtained during cell fractionation. The apparent melting temperatures (T_m), determined by measuring the remaining meropenemase activity after denaturation by increasing temperatures, showed that the environmental MβLs are more stable in the periplasm than recNDM-1 (Fig. 7A; Supplemental Material, Supplementary Table 11). The same trend was observed when assaying pure enzyme preparations in both lysis buffer (used for cell fractionation) and plain salt buffer (Fig. 7A).

Fig. 7. Stability of myxobacterial MβLs and zinc requirements for their in vivo function.

A Kinetic stability of the studied enzymes. Residual activity of meropenem hydrolysis yielded by enzyme preparations subjected to thermal denaturation was plotted in relation to temperature (T). The apparent melting temperatures (T_m) were determined through interpolation. Periplasmic fractions and pure enzyme preparations in the lysis solution used for spheroplast formation and in plain salt buffer were assayed. Myxobacterial enzymes were more stable compared to the clinical ΜβL in every preparation tested. Although the periplasm content had a stabilizing effect on NDM-1, the environmental enzymes exhibited a lower apparent T_m there compared to their equivalent pure enzyme preparations. Data from two technical replicates are shown. B Thermal stability of MβLs as determined by differential scanning fluorimetry (DSF). The differential of fluorescence intensity (dF/dT) versus temperature plots were used for T_m estimation. Thermal stability correlated with kinetic stability, with the ranking being MYX-2 > PYX-1 = PYX-2 > MYX-1 = COR-1 > NDM-1. C Comparison of resistance levels conferred by recMYX-1 and recNDM-1 under the pZE21 and pET26b(+) expression systems in the absence and presence of 0.25 mM ZnSO₄. Error bars denote means ± standard deviations of doubling serial dilution differences observed across three biological replicates (n = 3) using different starting cultures. Under the pZE21/E. coli C600Z1 genetic background, zinc supplementation resulted in less than one serial dilution difference between recMYX-1 and recNDM-1 cefotaxime and carbapenem MICs. The pET26b(+)/E. coli BL21(DE3)pLysS clone expressing the recMYX-1 enzyme remains more sensitive to the above β-lactams compared to recNDM-1, irrespective of growth in the presence of the enzymes’ cofactor.

Interestingly, recNDM-1 exhibited somewhat higher kinetic stability in the periplasm compared to its purified form under equivalent conditions. This is opposite to what was observed for most myxobacterial MβLs (e.g., recNDM-1: ΔT_m^{Periplasm-Purifed/LysisBuffer} =+2.91 ^oC and recMYX-1: ΔT_m^{Periplasm-Purifed/LysisBuffer} =−2.84 ^oC; Supplemental Material, Supplementary Table 11), with the differences though being within experimental error. Nonetheless, while periplasmic molecules might differentially interfere with NDM-1 and myxobacterial enzymes’ folding and function, this isn’t the primary reason for the observed gap, as the latter molecules remain more stable. The higher kinetic stability of myxobacterial enzymes correlated with their thermal stability, as determined by differential scanning fluorimetry (DSF), with MYX-2 being the most stable enzyme, followed by pyxidicoccal MβLs (Fig. 7B; Supplemental Material, Supplementary Table 11).

Given the differences in charged amino acid composition in the vicinity of catalytic residues, environmental MβLs likely have different zinc requirements for optimal in-cell function and folding. Indeed, in MIC determinations using Mueller-Hinton broth supplemented with zinc, we observed an increase of resistance levels for both recNDM-1 and recMYX-1, but the differences were more pronounced for the latter enzyme. This resulted in the amelioration of the in vitro–in vivo gap for cefotaxime and carbapenems under the pZE21/E. coli C600Z1 expression system (Fig. 7C). However, resistance levels for recMYX-1 remained low in the pET26B(+)/E. coli BL21(DE3)pLysS background, even with zinc supplementation (Fig. 7C), underscoring the importance of the host in MβL expression.

Myxobacterial natural compounds active against β-lactamases

The apparent ancient acquisition of B1 MβLs by myxobacteria, in conjunction with their rich secondary metabolism—which has produced promising bioactive compounds (e.g., epothilones⁷⁰)—suggests a potential for the discovery of β-lactamase inhibitors. Analogous to Actinomycetes, which produce class A β-lactamase inhibitors, such as clavulanic acid and the β-lactamase inhibitor protein (BLIP)^71,72, myxobacteria may have evolved similar mechanisms, albeit for different purposes. While myxobacteria are not known to synthesize β-lactams, previous studies have shown that β-lactamase production in M. xanthus is associated with its complex developmental cycle^73,74, indicating a possible need for self-regulation of enzyme activity.

To assess this possibility, we examined the potential effects of secondary metabolites produced by Myxococcus sp. 1LA on β-lactamase-mediated resistance in E. coli DH5α clones producing various enzymes (NDM-1, MYX-1, VIM-1, and OXA-48). An organic extract containing the majority of secondary metabolites produced by the investigated myxococcal strain was obtained after extraction with organic solvents of the cells and absorbent resin added in the culture broth. Its antibacterial effect was then examined against the selected clones in the absence and presence of cephalothin (Fig. 8).

Fig. 8. Antibiotic activity of Myxococcus sp. 1LA crude organic extract.

The extract from the seven-day culture was used against E. coli DH5α clones expressing various β-lactamases in the absence and presence of cephalothin. The clone carrying the empty vector pZE21-MCS was included as a control. The crude extract was assayed undiluted as well as in four doubling serial dilutions, including also a control disc containing solely absolute ethanol (EtOH). The solvent was completely evaporated before application of the discs on the inoculated agar. Cephalothin concentrations corresponded to ¼MIC of the respective clone and they were 256 μg/mL for NDM-1, 8 μg/mL for MYX-1, 8 μg/mL for VIM-1 and 2 μg/mL for the OXA-48-producing clone. In the plates lacking cephalothin, discs containing 30 μg ampicillin, supplemented either with the four times diluted extract or with plain ethanol, were also examined. For all the tested β-lactamase producing clones, cephalothin exhibited a synergy with the extract, enabling the formation of inhibition zones as low as the 2⁻³ to 2⁻⁴ dilutions. Inhibition zones yielded by the undiluted extract and the 2⁻¹ dilution were considerably wider in the presence of cephalothin. The length of the scale bar represents 1 cm.

Our data indicated that the extract possesses antibacterial activity, which was further enhanced by sub-lethal concentrations of cephalothin in all β-lactamase-producing clones (Fig. 8). This observed synergy was not specific to MβLs, as it was also evident with the serine-reactive OXA-48. Based on WGS of Myxococcus 1LA, the strain carries the biosynthetic cluster for myxovirescins (contig JBKGHI010000019.1). Myxovirescins are antibiotics produced by the Myxococcus genus^75,76 which exert their bactericidal action by inhibiting signal peptidase II⁷⁷. The other known signal II peptidase inhibitor, globomycin⁵⁷, potentiated the action of cephalothin against the above strains as indicated by checkerboard assays (at sub-lethal concentrations the compound reduced cephalothin MIC > 2 serial dilutions; Supplemental Material, Supplementary Fig. 20). The phenomenon was again non-specific for the clone producing the membrane localized NDM-1. The effects of signal II peptidase inhibitors on β-lactam resistance are most likely due to the inhibition of Lpp (Braun’s lipoprotein) topogenesis, which covalently links the outer membrane with the peptidoglycan and could affect β-lactam susceptibility as it controls the mechanical properties of the E. coli cell wall⁷⁸. Although our preliminary assays with a Myxococcus strain did not reveal an MβL inhibitor, they pointed to a secondary metabolite able to interfere with β-lactamase production or the action of β-lactams against resistant enterobacteria.

Discussion

Horizontal gene transfer between organisms from different phyla is a rather rare event⁷⁹ and, in general, acquired antibiotic resistance genes circulating clinical γ-Proteobacteria have been transferred there in the recent “antibiotic era” from their kin environmental species⁸⁰. The phylogenomic analyses presented herein indicated that MβLs are not that common in Pseudomonodota, pointing to other bacterial phyla as the original hosts. Among the key organisms, predatory myxobacteria most likely had a pivotal role in these early HGT events, acquiring resistance determinants from neighbouring soil-dwelling Cytophagia, Flavobacteriia, and Chitinophagia, and subsequently transferring them to environmental proteobacteria. Although this linear model oversimplifies inter-phylum HGT, which likely involved numerous independent events across diverse organisms, the presence of the gene in the core genome of certain families within these bacterial groups underscores their importance in the deep evolutionary history of MβLs.

The early acquisition of B1 MβLs by these organisms was most likely accompanied by adaptations integrating the enzymes’ expression into their unique physiology. For instance, in myxobacteria, the secretion of MβLs is facilitated by signal peptides that exhibit both type I and type II likely cleavage sites. When expressed in a γ-proteobacterium, these enzymes localized as soluble proteins in the periplasm, suggesting that either lipidation or cleavage by Lsp cannot occur there, or that while the E. coli lipoprotein topogenesis system recognizes their N-terminal sequences, the presence of a downstream type I cleavage site results in further processing by the LepB peptidase. In the case of Myxococcus enzymes, the first hypothesis is more plausible, as the third residue upstream of the N-terminal cysteine is a serine. A hydrophilic residue at position -3 of myxococcal type II signal peptides is not as rare as in E. coli (e.g., 9.4% of predicted lipoproteins from M. xanthus versus 0.02% for E. coli; Supplemental Material, Supplementary Table 12). The concurrent presence of predicted type I and type II cleavage sites is also not limited to β-lactamases but also found in other secreted myxobacterial proteins, such as a penicillin acylase domain enzyme from M. xanthus DK1622 (GenBank: NC_008095; locus MXAN_2906). These features may represent an adaptation to the production of myxovirescin, an antibiotic that inhibits signal peptidase II. Notably, the Myxococcus genus encodes four variants of Lsp, one of which is located within the biosynthetic gene cluster of myxovirescin⁸¹. This suggests that their signal peptides may have evolved to be recognized specifically by Lsp isoforms resistant to inhibition by this compound.

However, other members of the Myxococcaceae family analyzed in this study—namely Corallococcus and Pyxidicoccus—do not produce myxovirescin but still possess dual cleavage sites in their signal peptides. In E. coli, only the type I site appears to be functional. Both genera carry two Lsp homologues in their genomes, which may have driven the adaptation of their signal peptides to be recognized by either enzyme. Given the complex life cycles and predatory nature of myxobacteria, it is plausible that their specialized cell envelope structures and protein secretion systems differ significantly from those of proteobacteria^82,83. Prior to the availability of complete myxobacterial genomes, it had been reported that M. xanthus produces a β-lactamase⁸⁴, and that its expression is developmentally regulated and induced during sporulation^73,74. Our current findings showed that the sole true β-lactamase homologue produced by M. xanthus is a B1 MβL. The dual signal peptide of this enzyme may therefore facilitate specialized localization during different stages of the organism’s life cycle. Investigating the expression and localization of this enzyme not only in Myxococcus, but also in other genera of Myxococcaceae, would provide valuable insight regarding the biology of B1 MβLs.

Differences in signal peptides between myxobacterial MβLs and the clinically significant, outer membrane-anchored NDM-1 likely explain the relatively weak β-lactam resistance phenotypes observed for environmental enzymes expressed in E. coli. These enzymes also exhibited a tendency to aggregate in the E. coli periplasm, likely due to a requirement for elevated zinc concentrations for proper folding and protection against oxidation. Recent studies have shown that zinc affinity is a major factor driving the evolution of NDM-type MβLs in clinical settings⁸⁵. This suggests that even clinically relevant B1 MβLs may not yet be fully optimized for expression in enterobacterial pathogens^66,86.

Myxobacterial MβLs efficiently hydrolyze most major β-lactam classes, indicating that the evolution of clinical enzymes mainly involved adaptations for improved expression in new hosts, rather than major functional changes. However, myxobacterial enzymes showed poor activity against ceftazidime and cefepime, a trait shared with other environmental B1 MβLs, such as those from Chryseobacterium⁸⁷. This likely reflects the absence of structurally similar β-lactams in soil environments. Nonetheless, elucidating the molecular basis of clinical MβLs’ high activity against the above semisynthetic β-lactams may aid in designing novel compounds that evade hydrolysis. Toward this goal, we have initiated a directed evolution study aiming to introduce NDM-1-like properties into the environmental MYX-1 enzyme.

The integration of MβL production into the physiology of myxobacteria may also involve the co-production of secondary metabolites that regulate enzyme activity. We observed a synergistic effect between crude organic extracts from a Myxococcus isolate and cephalothin against β-lactamase-producing enterobacteria. While this effect was likely due to myxovirescin-mediated inhibition of Lpp topogenesis, the possibility of specific MβL-targeting molecules cannot be excluded. Given the wealth of biosynthetic potential in myxobacteria, many uncharacterized metabolites may still await discovery. Furthermore, our findings underscore the importance of Cytophagia and other Bacteroidota in the evolutionary history of MβLs, highlighting them as additional promising targets in the search for natural MβL inhibitors.

Materials and Methods

Development of Hidden Markov Models specific for the known molecular classes of β-lactamases

HMM profiles for class A, C, D, B1, B2 and B3 were built using the reference sequences of the so far experimentally verified β-lactamases contained in NCBI’s Bacterial Antimicrobial Resistance Reference Gene Database (Bioproject: PRJNA313047; assessed on 25-3-2019). The LRA-13 bi-functional β-lactamase, containing distinct class C and D domains, was excluded (protein id: WP_063839877). The VarG MBL-fold protein identified in a Vibrio cholera isolate (protein id: WP_000778180.1) was also excluded as it doesn’t contain the zinc binding motifs of the B1, B2 and B3 subclasses of metallo-β-lactamases and therefore it represents a separate protein family within the MβL superfamily. Sequences of each β-lactamase class were clustered using CD-HIT v4.8.1⁸⁸ with an identity threshold of 80% (-c 0.8) and a length cut-off of 100% (-s 1.0). These cut-offs were optimal in grouping the known families of clinically relevant β-lactamases.

The six data sets were used for training the respective HMMs. CD-HIT extracted representative sequences for each cluster, which were aligned using Clustal Omega with default settings. The above alignments were used for the HMM training using the program hmmbuild of the HMMER v3.1b2 package⁸⁹. Cut-off scores for considering a positive HMMER hit as a likely β-lactamase were estimated using a one-out-cross-validation approach that was performed as follows. Each protein from each data-set was excluded and after alignment of the remaining sequences, an HMM was built. Then the excluded protein was checked against the generated model using the program hmmsearch of HMMER with default inclusion criteria. The above process was repeated for each protein in the data set and the lowest observed HMMER bit score was considered as the cut-off (Supplemental Material, Supplementary Tables 1, 2). Proteins subjected to the models generated with the whole data set using hmmsearch with default settings and having bit scores above the cut-off were considered positive.

The generated models were further validated regarding their sensitivity and specificity in detecting the correct β-lactamase class while excluding the related proteins not having β-lactamase activity. The latter proteins for serine reactive β-lactamases (classes A, C and D) concerned the Penicillin Binding Proteins as described by Massova and Mobashery⁹⁰ and for metallo-β-lactamases (B1, B2 and B3) the 17 families of the MβL superfamily¹⁰. Sequences of the above protein families retrieved from UniProt database were subjected to hmmsearch against the models and results were evaluated as described in Supplemental Material (Supplementary Figs. 1–10 and Supplementary Tables 3–8). The class D model misidentified the highly related BlaR1 proteins not having β-lactamase activity⁹¹ and to increase specificity, a specific HMM model was developed for them as above. The same was done for the class B3 homologues, having a HARLDQ motif⁹². The above analyses revealed that between the experimentally verified β-lactamases and homologous proteins not having hydrolytic activity, there are a limited number of sequences with intermediate HMM scores. This range corresponded to the “grey region” and although the respective hits were stored during the subsequent genome scanning were not included in analyses of this study and only positive hits were considered (Supplementary Data 1). Of note, B2 MβLs were located in the grey region of the B1 HMM. Hence, grey B1 hits were also subjected to analysis with the HMM specific for B2 enzymes.

Detection of β-lactamases in published genomes

Bacterial and archaeal genome assemblies were obtained from NCBI Refseq database (assessed on 10-18-2024). In total 378,459 bacterial and 2312 archaeal high-quality genomes were retrieved, representing 54,387 and 1192 different species, respectively, based on NCBI assigned taxid. For each genome, the nucleic acid sequence was obtained as well as the ORF annotation when available. Furthermore, for each taxid the respective lineage was extracted from NCBIs taxonomy dump file. In cases where there was no annotation, Prodigal v2.6.3 (as implemented in PROKKA v1.1.2⁹³) was utilized in order to obtain one. A six-frame translation was carried out for every genome and each frame was scanned with the developed β-lactamase-specific HMMs. Based on HMM scores, the hits were classified as positive or grey and the respective coding sequences (CDS) were obtained from the genome annotation using blastp (Blast+ v2.9.0) on the proteins encoded by the extended region, translating the detected polypeptide.

Positive hits expressed by intact open reading frames were further analyzed as for the presence of the catalytic residues known to be essential for hydrolysis of β-lactams using the alignment of the respective sequence with the specific HMM profile. The secretion potential was assessed by analyzing the identified proteins with SignalP v5⁹⁴ in both Gram-negative and Gram-positive modes. Proteins expressed by intact ORFs (no pseudogenes and >60% coverage of the HMM profile) and the respective CDS were clustered using CD-HIT and the cut-offs indicated above. Sequences of experimentally verified β-lactamases were also included in the clustering in order to assess membership in a known family.

Based on the above data, carriage rates and enzyme richness were calculated for each operating taxonomic unit (OTU) of the respective lineages. Frequency of carriage was expressed as the % percentage of the number of genomes or the number of different species carrying at least one hit of the specific β-lactamase class in respect to the total number of genomes or species belonging to the respective OTU. Frequencies concerning total positive hits, positive hits expressed by intact ORFs, proteins having operational catalytic machineries and, true enzymes with a secretion potential were determined for each taxonomic group. Confidence intervals (95%) for the measured frequencies were estimated using fraction of total analysis by the Wilson/Brown method as implemented in Prism v8 (GraphPad, USA). Enzyme richness was determined as the ratio of the total number of CDS or protein clusters identified in each OTU divided by the total number of hits of the respective type. Only data regarding class A, B1 and B2 enzymes are reported in the manuscript (Supplementary Data 2–5).

Large-scale phylogenetic analyses

The relationships of the positive B1 MβL homologues were assessed using Maximum Likelihood, and Bayesian inferred phylogenies. The analyses were performed using the representative sequence of each protein cluster, including those expressing proteins with non-canonical active sites. Representative sequences of the identified clusters (n = 1212) as well as the identified B2 ΜβL families (n = 18) were aligned using Clustal Omega v1.2.4⁹⁵. Low-quality regions in the alignments were removed using trimAl v1.2 as well as through manual inspection in order to include only the portion of the proteins corresponding to the mature ΜβL fold. After the removal of clusters corresponding to truncated forms of known β-lactamases that had passed our inclusion clustering criteria, 1196 B1 and 15 B2 orthologous families were subjected to phylogenetic analyses.

ML trees were estimated through 100 runs of RaxML v8.2.12 software⁹⁶ using the LG amino acid substitution model with fixed frequencies as determined by the program during an initial run with -PROTGAMMAAUTO. The confidence for each branch of the best ML tree was assessed through bootstrapping using the extended majority rule convergence criterion (convergence after 350 bootstrap replicates).

The tree space was also explored through three runs of Metropolis-coupling Markov chain Monte-Carlo sampling with three chains per run for 6 ∙ 10⁷ generations using the multi-threaded MPI variant of Exabayes v1.5⁹⁷. Chains were moved by implementing the default proposals and the LG model was solely used. Convergence was assessed through the average standard deviation of the log-likelihood split frequencies (asdf) between each run. Extended majority rule consensus Bayesian trees were computed using data from the three runs, with the initial 10% of trees being discarded (Effective Sample Size-ESS > 200 for the following parameters: LnPr, TL{0}, LnL and alpha{0}).

In order to assess the validity of our approach in unveiling the evolutionary history of B1 ΜβLs we estimated the phylogeny of enzyme families of class A β-lactamases for which there is a well-attested theory regarding their origins. The identified families of positive class A enzymes (n = 6120) were aligned with Clustal Omega using the pileup method and 40 ML trees were calculated as above.

Isolation of myxobacterial strains

Soil samples were collected from four sites in the Attica region of Greece (site 1: 37.6975^oN, 24.063611^oE; site 2: 37.701944^oN, 24.058056^oE; site 3: 37.681111^oN, 24.0675^oE; site 4: 37.726111^oN, 24.053056^oE) on October 2018. From each site, five samples were collected from a square with dimensions of 1 m, taking the upper 15 cm of soil. Pooled samples from each site were moisturized with sterile H₂O, containing 100 μg/mL voriconazole as an antifungal, until water holding capacity. A quantity of the pooled sample was placed on a 60 × 60 mm plate, and five sterile dung pellets collected fresh from wild red deer (Mnt. Parnitha, Attica, Greece) were embedded symmetrically in the soil. Lentil-sized quantities of each moisturized sample were also placed adjacent to fresh lawns of E. coli DH5α cells (washed thrice with 0.9% w/v NaCl) on WAT agar [0.1% w/v CaCl₂.2H₂O, 20 mM HEPES pH 7.2, 1.5% w/v agar bacteriological (Thermo Scientific-OXOID) containing 100 μg/mL voriconazole. The plates containing deer dung and E. coli baits were incubated at 30 ^oC and after 72 h were inspected daily for growth of myxobacterial fruiting bodies or swarms feeding on bacterial lawns (WAT agar) under a stereoscope.

The detected fruiting bodies and swarm edges were then placed aseptically in the center of VY/2 agar (0.5% w/v baker’s yeast, 0.1 w/v CaCl₂.2H₂O, 20 mM HEPES pH 7.2, 0.5 μg/mL methylcobalamin, 1.5% w/v agar bacteriological) containing 100 μg/mL voriconazole and the cultures were incubated at 30 ^oC for six days. The isolates yielding typical myxobacterial swarming growth with fruiting body formation were sub-cultured on fresh VY/2 agar. The cultures were further purified by preparing suspensions in sterile H₂O and plating serial dilutions on VY/2 agar. Myxobacterial “colonies” were apparent after the fourth to fifth day of incubation at 30 ^oC and were sub-cultured again on VY/2 agar. Purity of the final cultures was assessed by verifying the absence of microbial growth after 48 h incubation in Tryptic Soy Broth (TSB; Thermo Scientific-OXOID, UK) as well as PCR amplification of the 16S rRNA gene (applied on total genomic DNA extracted from cells grown on VY/2 agar) and Sanger sequencing (CeMIA, Greece) using the 27 F and 907 R primers.

We were able to isolate the following strains: Site 1: strain 1LA isolated from fruiting bodies on dung baits and classified as Myxococcus sp. based on the sequenced 16S rDNA; Site 2: strain 2LB1 isolated using dung baits and classified as Chondromyces sp.; Site 3: strain 3LFB2 isolated from WAT agar and strain 3LG isolated using dung baits. Both strains though, were placed in the lineage of Myxococcaceae could not be classified to a specific genus based on the sequenced portion of 16S rDNA; Site 4: strain 4LFB isolated from WAT agar belonging to genus Corallococcus. The Chondromyces sp. 2LB1 was excluded from further analyses as it belongs to the cellulolytic Polyangia which, based on our computational analyses, do not appear to contribute in the spreading of B1 MβLs. All isolates were able to grow on VY/2 agar containing 100 μg/mL ampicillin.

Whole genome sequencing

Strains 1LA, 3LFB2, 3LG and 4LFB were grown on Neopeptone-Yeast extract broth [NYE; 0.3% w/v Bacto Neopeptone (Thermo Scientific, USA), 0.2% w/v Bacto yeast extract, 25 mM HEPES pH 7,2] for three days at 30 ^oC. Growth in the liquid medium was typical for a wild-type myxobacterial strain, and total DNA was extracted from bacterial “flakes” using the gDNA kit (Zymo Research) by extending incubation in the lysis buffer for 1 h with gentle mixing. Whole genome sequencing was performed in-house using the Ion Torrent S5 platform. Libraries were prepared using the IonXpress Plus Fragment Library Kit, with 1 μg of genomic DNA used per library. The Ion Shear fragmentation method was applied under conditions optimized for generating 400 base-read libraries. Library quantification was performed using the Ion Library TaqMan Quantitation Kit, and all libraries were subsequently diluted to a final concentration of 100 pM. Template amplification onto Ion Sphere Particles (ISPs) was carried out before sequencing on the Ion S5 platform using an Ion 520 Chip with the Ion S5 Calibration Standard being included in each template preparation reaction.

Genome analyses

Raw reads extracted as bam files were used for genome assembling through SPADES v3.15⁹⁸ by applying the optimal settings for Ion Torrent data (i.e., single cell mode with 21, 33, and 55 k-mers). Assembly quality was assessed with the BBTools software (sourceforge.net/projects/bbmap/). Annotation was performed using PROKKA as well as the NCBI’s PGAP pipeline⁹⁹ which was run locally. Based on ANI analysis, strain 1LA was classified as Myxococcus sp., strains 3LFB2 and 3LG as Pyxidicoccus sp., and strain 4LFB as Corallococcus sp.

The genomes were scanned with our β-lactamase-specific HMMs, and the sequence of loci found to express for B1 MβL homologues was verified using PCR and Sanger sequencing (Supplemental Material, Supplementary Table 13). Reactions were catalyzed by the Q5 DNA polymerase (New England Biolabs, USA), including the High GC buffer provided by the manufacturer. Myxococcus sp. 1LA was carrying a B1 gene that was homologous to the already described family bla_MYX and was assigned the name bla_MYX-2 by the NCBI's Pathogens database, Pyxidicoccus sp. 3LFB2 and Pyxidicoccus sp. 3LG carries B1 homologues of a related family (bla_PYX), with assigned variants being bla_PYX-1 and bla_PYX-2, as was the gene detected in Corallococcus sp. 4LFB (bla_COR-1).

In order to better assess the taxonomic placement of the isolates, we performed phylogenetic analyses of the RNA polymerase II as well as the gene of the 16S rRNA from reference myxobacterial genomes. Both ML and Bayesian phylogenies were inferred. The RaxML software was used for computing 100 ML trees using the LG model with empirical base frequencies for the RpoB and the GTR gamma model for the 16S rDNA. Confidence of each branch was assessed with bootstrapping (RpoB converged after 300 replicates, while 16S rDNA did not converge after 1000 replicates). Bayesian analyses were performed through Exabayes using an equal prior probability for all the available amino acid models during RpoB simulations. Four runs with two MCMC chains were performed for 10⁶ generations, yielding acceptable convergence (RpoB asdf: 0.66% and 16S rDNA asdf: 1.44%). A consensus tree was computed from the four runs using a 25% burn-in (ESS > 200 for every parameter). The B1 MβL found in the respective strains, as well as clinically important enzymes, were also analyzed as above using the procedures followed for RpoB, with ML bootstrapping being converged after 700 replicates and Bayesian runs exhibiting a 0.71% asdf after 10⁶ generations. Phylogenetic distances between representatives of bacterial classes exhibiting meaningful carriage of a Β1 MβL homologue and Klebsiella pneumoniae were estimated using alignments of the respective RpoB proteins and the RaxML package, performing a calculation with the flag –f x.

Regions 25 Kb in both directions from the B1 MβL gene of reference Myxococcus species were aligned using the progressive alignment algorithm of the Mauve package¹⁰⁰. The extended loci of the genes in the above genomes, as well as in those of the strains isolated herein, were analyzed using IS Finder (https://isfinder.biotoul.fr/) with no transposase homologous genes being identified. Analyses using the Integron Finder v2 software¹⁰¹ of myxobacterial genomes obtained in this work (parameters used: --local-max --calin-threshold 1 --max-attc-size 300) failed to identify any integron-recognized gene cassette and the integrase homologs detected were not located near the B1 ΜβL genes, while they belonged to phages.

Secondary metabolite biosynthetic gene clusters were identified in our genomes using the bacterial version of the antiSMASH server (https://antismash.secondarymetabolites.org/#!/start¹⁰²;) with default parameters.

E. coli strains and plasmids

E.coli DH5α (fhuA2 lac(del)U169 phoA glnV44 Φ80’ lacZ(del)M15 gyrA96 recA1 relA1 endA1 thi-1 hsdR17), E. coli C600Z1 (laciq, PN25-tetR, SpR, lacY1, leuB6, mcrB + , supE44, thi-1, thr-1, tonA21) and E. coli BL21(DE3)pLysS (F(-) ompT gal dcm lon hsdSB(rB(-) mB(-)) λ(DE3 [lacI lacUV5-T7 gene 1 ind1 sam7 nin5]) were utilized as hosts for β-lactamase expression. Isogenic expression of β-lactamases in the above hosts was facilitated by the pZE21-MCS (ColE1, Kan^R, pTetOL1¹⁰³) and the pET26b(+) (pBR322, Kan^R, lacO; MilliporeSigma-Novagen). The first expression vector offers constitutive expression under a strong promoter of the genes cloned downstream of the provided ribosome binding site (RBS) in strains not producing the TetR transcriptional regulator (e.g., E. coli DH5α) and inducible production in strains carrying this protein (e.g., E. coli C600Z1) under anhydrotetracycline exposure. The second vector contains the open reading frame of the pectate lyase (PelB) signal peptide after its RBS and offers inducible expression using IPTG in strains producing the T7 RNA polymerase (e.g., E. coli BL21 DE3). In this work, we have used a host for expression of the pET26b(+) cloned genes that, in addition, carries the T7 lysozyme gene (plasmid pLysS), providing tighter regulation.

Cloning of β-lactamase genes

The open reading frames of MYX-2, PYX-1 and COR-1 MβLs were cloned in the KpnI and BamHI sites of the pZE21-MCS vector. The genes were amplified using gDNA extracts from the respective isolates as a template using the Q5 polymerase and the High GC buffer (Supplemental Material, Supplementary Table 13). As the bla_PYX-2 gene contained an internal KpnI site, we mutated the respective sequence of the pZE21-MCS vector to that recognized by the BsiWI endonuclease using the Pfu Turbo DNA (Supplemental Material, Supplementary Table 13). Then the PYX-2 ORF was amplified and cloned into the BsiWI and BamHI sites of the altered vector (Supplemental Material, Supplementary Table 13). As a source of MYX-1 and NDM-1 MβLs we used the previously described pZE1-bla_MYX-1 and pZE21-bla_NDM-1 plasmids^21,104.

In order to express the mature forms of NDM-1 and MYX-1 β-lactamases under a common signal peptide yielding soluble periplasmic expression, we cloned the respective gene fragments downstream of the PelB SP ORF. Nucleotides 83-813 of the bla_NDM-1 and 92-786 of the bla_MYX-1, corresponding to the mature proteins (for MYX-1, SP I signal peptide), were synthesized and cloned into the NcoI and XhoI sites of pET26b(+) (BioCat, USA). In the bla_MYX-1 gene, the silent mutation C429A was introduced in order to overcome an internal NcoI site. The N-termini of the mature enzymes were corrected in the above constructs using site-directed mutagenesis (Supplemental Material, Supplementary Table 13). The resulting ORFs of recombinant enzymes (recMYX-1 and recNDM-1) were sub-cloned in the KpnI and BamHI sites of the pZE21-MCS vector (Supplemental Material, Supplementary Table 13).

Sequence integrity of the cloned genes as well as of the expression module of each vector was assessed through Sanger sequencing of the generated plasmids (Supplemental Material, Supplementary Table 13).

β-Lactam susceptibility testing

The β-lactam susceptibility phenotypes of the generated clones was routinely assessed using the Kirby-Bower method, including a disc combining meropenem and EDTA¹⁰⁵. Quantitation of resistance levels was performed through minimum inhibitory concentration (MIC) determination using micro-dilution in Mueller-Hinton (MH) broth (Ca²⁺ and Mg²⁺ adjusted; EcoBio MH II broth, Biolab, Hungary). Antibiotics and concentrations ranges used were as follows: ampicillin (4 - 2048 μg/mL), ampicillin/clavulanate (4–2048 μg/mL with the inhibitor fixed at 4 μg/mL), piperacillin (1–512 μg/mL), cephalothin (2–1024 μg/mL), cefoxitin (2–1024 μg/mL), cefotaxime (1–512 μg/mL and 0.0625–32 μg/mL), ceftazidime (0.25–128 μg/mL), cefepime (0.0625–32 μg/mL), aztreonam (0.03125–16 μg/mL), imipenem (0.03125–16 μg/mL), meropenem (0.03125–16 μg/mL).

During β-lactam MIC determinations in E. coli C600Z1 and E. coli BL21(DE3)pLysS clones, β-lactamase expression was induced using 200 ng/mL anhydrotetracycline and 0.2 mM IPTG, respectively. In the above experiments, MICs were measured in the absence and presence of 0.25 mM ZnSO₄.

The effects of the signal II peptidase inhibitor globomycin from Streptomyces hagronensis (Sigma-Aldrich, USA) on cephalothin MIC of E. coli DH5α clones expressing the NDM-1, MYX-1, VIM-1 and OXA-48 β-lactamases were assessed in checkerboard assays using broth micro-dilution. Working concentrations of globomycin were prepared in dimethyl sulfoxide (DMSO) and added to MH broth containing doubling serial dilutions of cephalothin (4–2048 μg/mL for NDM-1, 1–512 μg/mL for MYX-1 and VIM-1 and, 0.125–64 μg/mL for the OXA-48 clone) with the final concentration of the organic solvent being 2% v/v. A series of cephalothin dilutions was also prepared, containing 2% DMSO without globomycin. The concentrations of globomycin assayed were in the range 1–10 μg/mL. Cephalothin MICs determined at the presence of the compound were compared to those obtained at its absence in parallel experiments.

β-Lactamase production studies

In order to assess the production of a B1 MβL by Myxococcus sp. 1LA, Pyxidicoccus sp. 3LFB2 and Pyxidicoccus 3LG we prepared cell-free extracts of the respective strains after a four-day growth in NYE broth at 30 ^oC using ultra-sonication with proteins being released in 20 mM HEPES, pH 7.5, containing 50 μΜ ZnSO₄ (buffer A), and cell debris, unbroken cells, and myxospores being removed by centrifugation at 14000xg for 90 min at 4 ^oC. The final preparations were sterilized with a 0.22 μm filter. We didn’t include Corallococcus sp. 4LFB in our experiments, as the strain also carried a B3 MβL homologue. Protein content in the above extracts was quantified using the Bradford method and the carbapenemase activity was measured using imipenem (100 μM) as substrate by monitoring the absorbance at 300 nm. The units of imipenem hydrolytic activity were obtained by the rate of absorbance fall and expressed as nmol of substrate hydrolyzed per minute using the Beer-Lambert law and a Δε of −9000 M^-1∙cm^-1 for the opening of the β-lactam ring at the above wavelength. The specific imipenemase activity of each extract was expressed as units per mg of total protein. The experiments were performed at 25 ^oC in 50 mM HEPES, pH 7.5, containing 50 μΜ ZnSO₄ using a quartz cuvette with 1 cm optical path in a Hitachi U-2001 UV/Vis spectrophotometer. Chelating agent-mediated inhibition of the imipenemase activity in the extracts was examined in parallel experiments using 400 μM EDTA with 1 min pre-incubation.

The optimal concentration of inducers for the production of the cloned MβLs by the pZE21/E. coli C600Z1 and the pET26b(+)/E. coli BLS1(DE3)pLysS systems was examined as follows: 5 mL Luria-Bertani Broth (LB; casein peptone 10 g/L, yeast extract 5 g/L; Sarlau, Spain) containing 30 μg/mL neomycin and 0.25 mM ZnSO₄ were inoculated with 0.5 mL of overnight cultures and after growth at 37 ^oC until OD_600nm = 0.8 various concentrations of anhydrotetracycline (40, 80 and 200 ng/mL) and IPTG (0.2, 0.5 and 1.0 mM) were added and incubation continued for 20 h. Cell-free extracts were obtained by disruption of bacteria (washed once) by ultra-sonication in 0.4 mL of buffer A. MβL levels in the above extracts were determined by monitoring hydrolysis of 80 μM meropenem at 298 nm using a Δε of –7200 M⁻¹∙cm⁻¹ as above, with consumption of substrate being between 5% to 10% of its initial concentration in order to maintain steady state conditions.

The effect of growth temperature on production of the studied β-lactamases in E. coli C600Z1 was estimated by inoculating 80 mL of LB broth containing 30 μg/mL neomycin and 0.25 mM ZnSO₄ with 3 mL of overnight cultures and after un-induced growth at 37 ^oC (OD_600nm = 0.8) anhydrotetracycline was added at a final concentration of 200 ng/mL. The cultures were then split in half, incubated at 30 and 37 ^oC and 5 mL were removed at t = 0, 3, 6 and 24 h post-induction. Cell-free extracts were prepared as above and hydrolytic activity against meropenem was determined by spectrophotometry. The MβL concentration in each sample would be proportional to meropenem hydrolysis units at the steady state per volume (μL) of extract.

In order to assess the presence of insoluble proteinaceous material in induced E. coli cells, the pellet obtained during clarification of the cell-free extract was subjected to extended membrane solubilization as described previously with some modifications¹⁰⁶. The pellets were suspended in 2% Triton X-100, 2 M urea, 50 mM HEPES pH 7.5, 100 μΜ ZnSO₄, 5 mM Tris(2-carboxyethyl)phosphine hydrochloride (TCEP-HCl) with the aid of a cell homogenizer and then centrifugation was carried out at 14000xg for 30 min at 4 ^oC and the supernatant was discarded. The process was repeated five additional times, eventually resulting in clear and colourless supernatant. The pellet was finally washed with the above buffer lacking Triton X-100 and urea. The sediment remaining in the final step was suspended in 8 M urea, 50 mM HEPES pH 7.5, 100 μΜ ZnSO₄, 5 mM TCEP-HCl and incubated overnight at 4 ^oC and after a spin-down, the supernatant was analyzed by SDS polyacrylamide gel electrophoresis (SDS-PAGE). Removal of urea by dialysis using 50 mM HEPES pH 7.5, 100 μΜ ZnSO₄, 5 mM TCEP-HCl did not lead to re-folding of solubilized proteins, as extended sedimentation was evident with marginal β-lactamase activity being detected.

The effects of agents favouring extracellular release of periplasmic proteins and improved production of inclusion bodies prone to molecules were examined using the recMYX-1 and recNDM-1 clones. Glycine (1% w/v), sorbitol (0.5 M) or sodium chloride (0.1 M) were added along with 200 ng/mL anhydrotetracycline or 0.2 mM IPTG to late log-phase cultures in LB broth containing neomycin and ZnSO₄ and induction was carried out for 20 h. At the end point, culture supernatants were stored and cell-free extracts were prepared as above. The quantity of the β-lactamase in each sample was expressed as the total meropenem hydrolysis units (units∙mL⁻¹ x V of preparation in mL) obtained during enzymatic assays performed as above. Glycine alone or combined with NaCl yielded the highest quantities of β-lactamases.

β-Lactamase purification

Homogenous preparations of myxobacterial MβLs having their native sequence were obtained using four steps of liquid chromatography. Overnight liquid cultures of E. coli C600Z1 clones were used to inoculate 1 L of LB broth containing neomycin and ZnSO₄ and after OD_600nm = 0.8 we added 200 ng/mL anhydrotetracycline and induction was carried out for 20 h at 37 ^oC. Cell-free extracts were prepared with sonication in buffer A. The crude extracts were diluted 1:3 using 20 mM Bis-tris, pH 7.2 (or 20 mM Tris, pH 8 for PYX-1) containing 50 μM ZnSO₄ (buffer B) and loaded on a 5 mL DEAE-Sephacel (Cytiva, USA) anion exchange column previously equilibrated with the same buffer. The majority of MYX-1, MYX-2, PYX-1 and PYX-2 were retained on the column during loading and washing with buffer B, while COR-1 was eluted in the effluent. The bound enzymes were obtained through a linear gradient of 1 M NaCl in buffer B and the fractions exhibiting β-lactamase activity were pooled and 0.5 M HEPES pH 7.5, was added, targeting a 50 mM final concentration. The COR-1 containing effluent was loaded on a 5 mL Unosphere S (Bio-Rad Laboratories, USA) cation exchanger previously equilibrated with 20 mM HEPES pH 7.2, 50 μM ZnSO₄. The total amount of enzyme was retained on the column and eluted with a linear gradient of 1 M NaCl in the equilibration buffer. The pooled fractions of COR-1 were again supplemented with 50 mM HEPES, pH 7.5, as above.

The preparations obtained during ion exchange chromatography were further purified with hydrophobic interaction chromatography. In the pooled fractions from the previous step, ammonium sulfate [(NH₄)₂SO₄] was added at a final concentration of 1.5 M using a 4 M stock solution and centrifugation was carried out at 14000xg for 20 min at 4 ^oC. The supernatants were then loaded on a 5 mL MacroPrep Methyl (Bio-Rad Laboratories) column previously equilibrated with 1.5 M (NH₄)₂SO₄ in buffer A. The myxococcal enzymes MYX-1 and MYX-2 were captured by the resin and eluted using a linear gradient of buffer A at 0.9–1 M (NH₄)₂SO₄ as determined by an electronic refractometer. The fractions exhibiting β-lactamase activity were pooled and loaded on 5 mL Toyopearl Phenyl 650S column (Tosoh Bioscience, Japan) equilibrated with 1 M (NH₄)₂SO₄ in buffer A and the bound enzymes were eluted with a linear gradient using buffer A. PYX-1 and PYX-2 did not bind to the MacroPrep Methyl. The effluent from this column was loaded on the Toyopearl Phenyl 650S equilibrated with 1.5 M (NH₄)₂SO₄ in buffer A and the bound β-lactamases were eluted as above. COR-1, while initially captured on MacroPrep Methyl, was eluted as a single peak during the washing step with 1.5 M (NH₄)₂SO₄ in buffer A.

The enzymes were further purified using size exclusion chromatography. The pooled fractions from the previous step were subjected to buffer exchange and concentration using Amicon 3 KDa cut-off ultrafiltration tubes and 150 mM NaCl in buffer A. The concentrated samples were then loaded on a 120 mL Sephacryl S100 HR (Cytiva) column equilibrated with 150 mM NaCl in buffer A and the enzymes were eluted as a single peak at 80 mL.

The recombinant forms of MYX-1 and NDM-1 were purified from culture supernatants of strains pZE21-bla_recMYX-1/E. coli C600Z1 and pET26b(+)-bla_recNDM-1/E. coli BL21(DE3)pLysS induced with 200 ng/mL anhydrotetracycline and 0.2 mM IPTG in LB broth which contained 0.7% w/v glycine and 50 mM NaCl. Cells were removed with 30 min centrifugation at 10,000 x g and 125 mL of the supernatant was diluted with 1.5xV H₂O and loaded on a 30 mL DEAE-Sephacel column previously equilibrated with buffer B. The bound proteins were eluted using a linear gradient of 1 M NaCl in buffer B. This process was repeated three additional times in order to obtain β-lactamases contained in 500 mL of culture supernatant. Fractions containing β-lactamase activity were pooled, passed through a 0.22 μm filter and further purified using the downstream procedure followed for MYX-1 and MYX-2 enzymes.

Detection of β-lactamase activity in the preparations obtained during the above chromatographic procedures was carried out using a nitrocefin assay, except in the COR-1 and PYX-2 separations, where the direct colorimetry of the imipenem decomposition method¹⁰⁷ was utilized, as both enzymes hydrolyzed the chromogenic cephalosporin poorly.

Purity of the final preparations was examined with SDS-PAGE using in-house prepared Bis-tris gels and Tris-MES running buffer (50 mM Tris, 50 mM 2-morpholin-4-ylethanesulfonic acid, 1 mM EDTA, 0.1% SDS) containing 5 mM sodium bisulfite. Enzyme concentration in the pure fractions was determined through spectrophotometry at 280 nm using the following absorption coefficients (ε): MYX-1: 28990 M⁻¹∙cm⁻¹; MYX-2: 28990 M⁻¹∙cm^-1; PYX-1: 31970 M⁻¹∙cm⁻¹; PYX-2: 26470 M⁻¹∙cm⁻¹; COR-1: 28990 M^-1∙cm⁻¹; recMYX-1: 28990 M⁻¹∙cm⁻¹; recNDM-1: 27960 M⁻¹∙cm⁻¹, calculated for the mature proteins using ProtParam with cysteines reduced. Fractions used for enzyme kinetics were diluted in 1 mg/mL bovine serum albumin and aliquoted before being stored at −80 ^oC.

MS/MS analyses of myxococcal enzymes

Purified MYX-1and MYX-2 were suspended in a buffer containing 4% SDS and 0.1 M dithiothreitol (DTT) in 0.1 M TEAB and processed according to the single-pot, solid-phase-enhanced sample preparation Sp3-mediated protein digestion protocol¹⁰⁸ including an alkylation step in 10 mg/ml iodoacetamide. Proteins were captured on magnetic beads (1:1 mixture of hydrophilic and hydrophobic SpeedBead Magnetic Carboxylate, Cytiva) at the presence of 50% ethanol. The beads were washed two times with 80% ethanol and once with 100% acetonitrile. Digestion was carried out using 0.5 μg Trypsin/LysC (MS grade, Promega) at 37 ^oC.

The isolated peptides were analyzed with a liquid chromatography tandem mass spectrometry (LC-MS/MS) setup consisting of a Dionex Ultimate 3000 nano RSLC online with a Thermo Q Exactive HF-X Orbitrap mass spectrometer. Samples were directly injected and separated on a 25 cm-long analytical C18 column (PepSep, 1.9 μm beads, 75 µm ID). Chromatographic separation was performed using a gradient of Solvent B (0.1% formic acid in 80% acetonitrile) as follows: 7% to 35% B over 40 min, 35% to 45% B over 5 min, and 45% to 99% B over 0.5 min, followed by equilibration for 14.5 min. MS data were acquired in a data-dependent (DDA) strategy selecting up to top 12 precursors based on precursor abundance in the survey scan (m/z 350–1500). The resolution of the survey scan was 120,000 (at m/z 200) with a target value of 3 ∙ 10⁶ions and a maximum injection time of 100 ms. HCD MS/MS spectra were acquired with a target value of 10⁵ and resolution of 15,000 (at m/z 200) using an NCE of 28. The maximum injection time for MS/MS was 22 ms. Dynamic exclusion was enabled for 30 s after one MS/MS spectra acquisition. Peptide match was set as preferred. The isolation window for MS/MS fragmentation was set to 1.2 m/z. A lock mass of m/z 445.12003 was used throughout the analysis.

The obtained data were searched using the MaxQuant Software v2.1.4.0¹⁰⁹ using Andromeda, against the MYX-1 and MYX-2 sequences with the MaxQuant common contaminant database. To be accepted for the identification, an error of less than 20 ppm (first recalibration search) and 4.5 ppm tolerance in the main search of peptide mass tolerance was accepted. Up to 2 missed cleavages were allowed and the modifications taken into account were: Oxidation (M); Acetyl (Protein N-term); Deamidation (NQ) as variable modifications and Carbamidomethylation (Cys) as fixed modifications. Tryptic semi-specific search was used. Matching between runs and second peptide options was activated. Protein, peptide and “site” identifications were validated at an FDR of 1% using a reversed database. The label minimum ratio count was set to 2. In order to find unknown modifications, the dependent peptide search was activated.

β-Lactamase localization studies

The E. coli C600Z1 clones carrying the pZE21-bla_MβL plasmids were induced for six hours in 250 mL LB broth and the cells were washed with 20 mM Tris pH 8 (buffer C). The pellet was re-suspended in 12 mL buffer C and then 3 mL 10 mg/mL lysozyme (in buffer C) and 15 mL 40% sucrose (in buffer C) were added slowly with continuous stirring. The suspension (30 mL; 20 mM Tris pH 8, 1 mg/mL lysozyme, 20% sucrose) was incubated at 4 ^oC for 30 min with periodic gentle mixing and then centrifugation was carried out at 4000 x g for 30 min at 4 ^oC. The supernatant would correspond to the periplasmic fraction and the pellet to spheroplasts.

The spheroplasts were washed with buffer C and then re-suspended in 5 mL buffer C containing 1 mM PMSF. A portion (200 μL) from this suspension was mixed 1:1 with 40% sucrose in buffer C and stored for β-lactamase quantitation while the remaining was subjected to ultra-sonication (4×15 sec). The lysate was clarified at 14000xg for 30 min at 4 ^oC. The supernatant, which would contain cytoplasmic proteins as well as membrane particles, was further diluted with buffer C and was ultra-centrifuged at 180000xg (RCFmax) using the SW-41Ti rotor (Beckman Coulter, USA) on a HITACHI CP100NX instrument for two hours at 4 ^oC. The supernatant would contain the cytoplasmic proteins and other floating particles while the pellet would be enriched in membranes. The pellet was washed with buffer C and 30 min ultracentrifugation as above and re-suspended in 3.2 mL 20% sucrose (in buffer C). A portion of the total membrane preparations (200 μL) was stored at −80 ^oC for enzymatic assays.

Inner and outer membrane fractions were obtained using isopycnic ultracentrifugation of the total membranes as described previously¹¹⁰ with some modifications. A sucrose gradient was formed using layers having the following densities (from bottom to top): 73% w/v (55% w/w), 61% w/v (50% w/w), 53% w/v (45% w/w) and, 47% w/v (40% w/w) with all solutions being prepared in buffer C. On top of the sucrose gradient, 80-100 mg (wet weight) of total membranes were loaded and ultracentrifugation was carried out at 120000xg (RCFmax) for 18 h at 4 ^oC using the SW-28.1 rotor. Fractions of the visible zones were collected and the presence of inner membrane particles was assessed by detecting NADH oxidase activity using UV spectrophotometry. Reactions were performed in 50 mM 3-(Morpholin-4-yl)propane-1-sulfonic acid (MOPS), pH 6.9, at 25 ^oC and the rate of absorbance fall of 120 μM NADH at 340 nm was transformed to NADH oxidase units (nmol of substrate oxidized per minute) using a Δε of −6300 M⁻¹∙cm⁻¹¹¹¹.

The levels of functional ΜβLs in the above preparations were quantified using meropenem hydrolysis at steady state conditions. All reactions were performed at 25 ^oC in 50 mM HEPES pH 7.5, 50 μΜ ZnSO₄, except in assays entailing spheroplasts, where the above buffer was supplemented with 20% sucrose in order to maintain the outer membrane pores. In all measurements, control reactions were performed in order to reproduce the conditions of the actual measurements and the marginal baseline meropenem hydrolysis observed was used to correct the obtained rates. The total meropenem hydrolysis units of each preparation were compared with its pair in the respective fractionation step and partitioning was expressed as logarithms of the obtained ratios.

Enzyme kinetics

Purified preparations of the studied β-lactamases were used in measurements of Michaelis-Menten kinetic parameters. Initial velocities of β-lactam hydrolysis at the steady state were determined through UV spectrophotometry and expressed as μM of substrate hydrolyzed per second. Hydrolysis rates were measured for various substrate concentrations and then k_cat (catalysis constant; s⁻¹) and K_m (Michaelis constant; μM) values were obtained from V/E versus S plots (where V is the initial velocity, E the enzyme concentration in the reaction in μM and S the substrate concentration in μM) through non-linear regression to the Michaelis-Menten equation using the Prism v. 8 software. In cases where the K_m exceeded the substrate concentration that could be measured through UV spectrophotometry, the k_cat/K_m was estimated through linear regression as previously described¹¹².

β-Lactams assayed, the respective wavelengths and Δε values were as follows: benyzlpenicillin: λ = 233 nm, Δε = −1140 M⁻¹∙cm⁻¹; piperacillin: λ = 235 nm, Δε = −820 M⁻¹∙cm⁻¹; cephalothin: λ = 262 nm, Δε = −7660 M⁻¹∙cm⁻¹; cefoxitin: λ = 260 nm, Δε = −7600 M⁻¹∙cm⁻¹; cefotaxime: λ = 266 nm, Δε = −6700 M⁻¹∙cm⁻¹; ceftazidime: λ = 260 nm, Δε = −8660 M⁻¹∙cm⁻¹; cefepime: λ = 260 nm, Δε = −10000 M⁻¹∙cm⁻¹; imipenem: λ = 300 nm, Δε = −9000 M^-1∙cm⁻¹; meropenem: λ = 298 nm, Δε = −7200 M⁻¹∙cm⁻¹. Reactions were performed at 25 ^oC in 50 mM HEPES pH 7.5, 50 μM ZnSO₄, except penicillin assays. There, 2-{[1,3-Dihydroxy-2-(hydroxymethyl)propan-2-yl]amino}ethane-1-sulfonic acid (TES) was used as a buffering agent, as HEPES exhibited high absorbance at the respective wavelengths.

Structural analyses

Signal peptide predictions were carried out using the SignalP v5 and LipoP v1.1¹¹³ software. Three-dimensional models of the mature forms (considering SP I processed peptides with the highest LipoP-likehood) of MYX-1, MYX-2, PYX-1, PYX-2 and COR-1 were built using the AlphaFold3 server (https://alphafoldserver.com/) co-folded with two zinc ions. The crystal structure of NDM-1 (Protein Data Bank: 6KXI; 1.38 Å resolution¹¹⁴;) was used for comparisons. The structures were superimposed through “Stamp structural alignment” using the MultiSeq tool of VMD v1.9.3 software (http://www.ks.uiuc.edu/Research/vmd/¹¹⁵;). Enzyme protonation was studied using the pH titration protocol of Discovery Studio with partial atomic charges obtained from the CHARMM force field (Supplemental Material, Supplementary Fig. 21). Poison-Boltzmann surface electrostatic potentials were calculated by the ABPS-PDB2PQR software suite and visualized using PyMol v3.1.5.1.

Enzyme stability assays

The kinetic stability of the studied enzymes was examined through measurements of the residual meropenem hydrolytic activity of enzyme preparations incubated for five minutes at increasing temperatures (35–90 ^oC), compared to the units of an equal volume that had been incubated in parallel on ice. Hydrolysis rates were measured in reactions containing 80 μM meropenem at 25 ^oC in 50 mM HEPES, pH 7.5, 50 μΜ ZnSO₄. Residual activity, expressed as ratios of hydrolysis units, was plotted against temperature and the apparent melting temperatures (T_m) were obtained through non-linear regression to a sigmoidal curve in Prism v. 8. Kinetic stability was studied in the periplasmic preparations obtained during cell fractionation as well as in pure enzyme preparations in the lysis buffer (i.e. 20 mM Tris pH 8, 1 mg/mL lysozyme, 20% sucrose) and in plain salt buffer (i.e. 150 mM NaCl in buffer A).

The enzymes’ thermal stability was studied using differential scanning fluorimetry (DSF). Denaturation was performed in 150 mM NaCl in buffer A containing 3 μM of enzyme and 5x Sypro Orange. Reactions were carried out in the StepOne Plus instrument (Applied Biosystems, Singapore) with the temperature being increased from 25 to 75 ^oC at a step of 0.1 ^oC using the settings for the TAMRA dye. The first order differential of fluorescence intensity reads over temperature (dF/dT) was obtained using Prism v. 8. The T_m corresponded to the temperature (T) at which the first derivative acquired its maximum, identified through peak detection in dF/dT versus T plots using the area under the curve (AUC) analysis of Prism v. 8.

Antibiotic activity of Myxococcus sp. 1LA extract

Starting cultures of Myxococcus sp. 1LA in NYE broth were used to inoculate 100 mL of 0.3% w/v neopeptone, 0.2% w/v MgSO₄.7H₂O, 0.05% CaCl₂.2H₂O, 0.5 μg/mL methylcobalamin, 25 mM HEPES pH 7.2 containing 1% w/v amberlite XAD-16 (Sigma-Aldrich, USA) as an absorbent resin. The cultures were incubated at 30 ^oC for three, seven and ten days. For each culture of the respective time point, the cells and resin beads obtained through centrifugation for 15 min at 10,000xg were extracted twice with 100 mL of dichloromethane/methanol (1:1, v/v) for 6 h under agitation in an orbital shaker (125 rpm). Filtration of the extract and removal of the solvent under vacuum at 40 ^oC afforded a solid residue that was freeze-dried to remove any traces of water. Subsequently, the residue was re-dissolved in 10 mL of ethanol, filtered and the filtrate was concentrated under vacuum at 40 ^oC to yield the crude organic extract. The preparations were reconstituted in 0.5 mL ethanol and used for antibiotic activity assays.

Antibiotic activity against β-lactamase-producing clones was examined in LB agar (casein peptone 10 g/L, yeast extract 5 g/L, NaCl 10 g/L, agar 1.5%; Sharlau) in the presence and absence of cephalothin. The β-lactam concentrations used were at one-fourth of the measured MIC of the respective clones. The β-lactamases examined, apart from the myxococcal MYX-1, included ΜβLs known to be secreted via a type I (VIM-1) or a type II (NDM-1) signal peptide, as well as a class D carbapenemase (OXA-48) acting via a catalytic serine. The source of the enzymes was the previously described pZE21 derivatives¹⁰⁴ expressed in E. coli DH5α. For each crude organic extract, serial doubling dilutions were prepared in ethanol and then 10 μL of the above solutions, as well as of the undiluted preparation, were placed on 9 mm antibiotic test discs (Macherey-Nagel, Germany). A control disc containing ethanol was also included. The discs were left to dry for 30 minutes and then placed on LB plates overlaid with soft LB agar (0.7% agar in LB broth) which has been inoculated with 10⁶ cfu of each E. coli clone (except the VIM-1 clone, where a 3 times higher inoculum was utilized). Results were recorded after 18 h incubation at 37 ^oC. The highest potency was observed in the organic extract obtained from the culture grown for seven days.

Statistics and Reproducibility

Confidence of the observed carriage rates of ΜβL genes in the various taxa was estimated through the fraction of total analysis using the Wilson/Brown method for interval calculations. The effects of temperature on the production of MβLs were assessed using three technical replicates. In experiments comparing the expression of MβLs under various E. coli genetic backgrounds through quantitation of functional enzyme levels and MIC determinations, three biological replicates were utilized. Statistical significance was assessed through one-way analysis of variance (ANOVA) using Holm-Sidak’s multiple comparisons test. No statistical test was performed on MIC data. According to the established interpretation of such data, a difference higher than 2 doubling serial dilutions is considered significant¹¹⁶. Antimicrobial susceptibility testing of E. coli DH5α clones through MIC determination was performed in three biological replicates, with results differing by less than two doubling serial dilutions. Localization of the enzymes was studied using three biological replicates, with results being compared again with one-way ANOVA as above. Micahelis-Menten kinetics constants measurements, as well as melting temperature estimation in kinetic stability assays, were performed using two technical replicates, with the results differing less than 15%. Statistical analyses were performed in Prism v8. The results supporting our main conclusions were reproducible during the above repetitions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

A.L. and S.D.K. conceived and designed the experiments. A.L., A.P. and S.D.K. performed the microbiological, molecular biology and biochemical experiments. A.L. and S.D.K. carried out the bioinformatics analyses. M.S. performed the MS/MS experiments. V.M. sequenced the bacterial genomes. M.A.T., E.I. and V.R. carried out the isolation of secondary metabolites. L.S.T. analyzed the data. A.L. and S.D.K. wrote the manuscript with input from all authors.

Peer review

Peer review information

Communications Biology thanks D. Cole Stevens and the other anonymous reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Haichun Gao and Rupali Sathe.

Funding

A.L. received support for this work from the Hellenic Pasteur Institute Excellence Scholarship –“Nostos Foundation” (7000/19-06-2020). S.D.K. and V.M. disclose support for the research of this work from the National Recovery and Resilience Plan Greece 2.0, funded by the European Union – NextGenerationEU, under the call RESEARCH-CREATE-INNOVATE (Implementation body: MIA RI; project: TAEDK-06179). M.S. discloses support for the mass-spectrometry instrumentation by the project “The Greek Research Infrastructure for Personalised Medicine (pMED-GR)” (MIS 5002802) which is implemented under the Action “Reinforcement of the Research and Innovation Infrastructure”, funded by the Operational Programme “Competitiveness, Entrepreneurship and Innovation” (NSRF 2014-2020) and co-financed by Greece and the European Union (European Regional Development Fund). M.A.T., E.I., V.R. and L.S.T. disclose no relevant funding.

Data availability

Whole genome sequencing data generated in this study have been deposited in the short reads archive (SRA) and assembly databases of NCBI under the bioproject accession code PRJNA1190413. Sequences of 16S rRNA genes and B1 MβL loci have been deposited in NCBI’s Genebank database under accession numbers PQ818161, PQ818162, PQ818163, PQ818164, PQ855794, PQ855796, PQ855797 and, PQ855795. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE¹¹⁷ partner repository with the dataset identifier PXD077803. Plasmids generated in the study have been deposited in Addgene under the following identification numbers: 256719, 256721, 256722, 256753, 256754, 256755, 256756 and 256757. Supplementary data 1–5 have been deposited in Figshare¹¹⁸. Source data for Figs. 1–5 and 7 can be found in supplementary data 6.

Code availability

All code developed for this manuscript is available on GitHub and Zenodo¹¹⁹. The software used, including the versions as well as the specific variables, is given in the respective sections of Materials and Methods.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-026-10466-8.