Lactococcal mobile genetic elements harbour a diverse phage defensome rich in restriction-modification systems

Abstract

The genomes of 43 distinct lactococcal strains were reconstructed by a combination of long- and short-read sequencing, resolving the plasmid complement and methylome of these strains. The genomes comprised 43 chromosomes of approximately 2.5 Mb each and 269 plasmids ranging from 2 to 211 kb (at an average occurrence of 6 per strain). A total of 953 antiphage genes representing 538 phage defence systems were identified in the 43 strains and were catalogued and cross-correlated with co-occurrent mobile elements, which indicated that almost 60% of these systems are predicted to be mobile. Detailed analysis established that restriction-modification (R-M) systems form a significant portion of this mobile phage defensome. As such, all detected Type I, II, and III-associated methylated motifs (46 of which were unique to this study) were matched to their corresponding methylating enzymes by homology detection or molecular cloning. The cumulative antiphage activity of selected systems and the ability of truncated R-M genes to contribute to methylation were demonstrated. This study reveals, for the first time, the dairy lactococcal plasmidome to be a rich reservoir of orphan HsdS-encoding genes, in a comprehensive survey of (mobile) phage defence systems in lactic acid bacteria.

Graphical Abstract

Graphical Abstract.

Introduction

The interplay between bacteria and their infecting, parasitic bacteriophages is of significant interest to microbiologists from clinical [1], ecological [2, 3], and industrial [4] perspectives. Central to this interaction is the enormous variety of mechanistically diverse phage countermeasures encoded by the hosts. In recent years, the discovery of the clustered regularly interspaced short palindromic repeats (CRISPR) systems [5] had somewhat diverted attention from the ever-present non-CRISPR anti-phage repertoire of bacteria, whose activity may be less readily detectable. Recently, however, the field has seen a bloom of newly identified phage defence candidates as well as characterised systems [6–9], re-emphasising the importance of identifying and characterising the diverse non-CRISPR anti-phage landscape.

The means and mechanisms by which beneficial genetic information, such as that encoding antiphage system components, is shared horizontally between bacterial strains, species, and genera, have been a mainstay of microbiological research for decades. This area is of significant interest to those working with lactic acid bacteria (LAB) in a commercial capacity due to the potential for favourable genetic elements to be naturally transferred from one strain to another. This phenomenon can be studied in remarkable detail using dairy starter cultures such as Lactococcus lactis, Lactococcus cremoris, and Streptococcus thermophilus, due to significant selective pressures in this environment leading to rapid evolution on the part of the fermenting cultures and their infecting phages. This has cemented the status of these (and other) species as archetypes of phage–host interactions. Interestingly, nonetheless, a significant and remarkable divergence occurs between mesophilic lactococci and their main counterparts in industrial fermentations, thermophilic streptococci, in their efforts to recognise and inactivate invading DNA. S. thermophilus encodes multiple, chromosomally encoded CRISPR-Cas systems to counter invading DNA [5, 10–12] and—perhaps consequently—does not contain many plasmids. Dairy lactococci, on the other hand, appear to have taken an evolutionary path without CRISPR-Cas systems, with currently just a single known, plasmid-encoded exception [13].

The latter bacteria have been shown to employ a vast array of abortive infection-like (henceforth termed ‘Abi-like’ based on previous nomenclature or demonstrated characteristics of the systems [14]) and restriction-modification (R-M) systems, many of which are plasmid-encoded [14–16]. This difference in defence strategy does not appear to be due to incompatibility of respective antiphage measures—indeed, many seem to act in a co-operative manner [17, 18]. The inverse correlation between the number of plasmids and CRISPR activity may be explained by the apparent preferential CRISPR-mediated targeting of fast-replicating genetic elements such as plasmids [19].

This near-absence of CRISPR-Cas systems has apparently allowed dairy lactococcal strains to accumulate and share an abundance of beneficial plasmids [16, 20, 21]; CRISPR-Cas loci may even be deleted to improve retention of beneficial plasmids [22]. Given the typical features of lactococcal plasmids, the preference for their maintenance is likely in response to nutrient [15, 21] and phage [23–25] pressures. It has been well established that plasmids are highly mobilisable between strains of many different bacterial genera [26, 27]. As well as plasmids, other mobile genetic elements (MGEs) may exist in the genomes of LAB, such as integrative conjugative/mobile elements known as ICEs [28], and prophages [29, 30], both also capable of transferring favourable traits between strains and species [31, 32]. Despite research into these elements separately, a comprehensive assessment of the lactococcal mobile defensome has to the best of our knowledge not been conducted.

R-M systems are widely recognised as crucial DNA defence systems in dairy lactococci, and in basic terms consist of a methyltransferase (MTase) gene, modifying “self” DNA in detectable patterns, and a restriction endonuclease (REase), inactivating “non-self” or invading DNA. Prokaryotic DNA MTases give rise to 6-methyladenine (6mA), 4-methylcytosine (4mC) or 5-methylcytosine (5mC) [33], all of which are detectable by single molecule real-time (SMRT) and Nanopore DNA sequencing methods [34, 35] (though satisfactory detection of 5mC modifications using SMRT sequencing usually requires pre-treatment of template DNA with, e.g. bisulfite or other compounds [36–38]). MTases have diverse roles in the cell [39–43], but their function in cellular defence involves their association with an REase. R-M systems comprise four types based on subunit content, ATP requirements and modes of DNA degradation [33]. Type I systems are multi-subunit protein complexes composed of one or two REase (HsdR) and MTase (HsdM) subunits and one specificity (HsdS) subunit [44]. Recognition of two 3–5 base pair DNA sites separated by 6–8 non-specific bases occurs through unique target recognition domains (TRDs) in the HsdS [45, 46], and digestion of DNA occurs at a site variably distanced from the recognition sequence [33]. Interestingly, more than one distinct HsdS subunit may be encoded in a given bacterial cell, yet each may be able to combine with separately encoded HsdM and HsdR subunits to execute DNA restriction [47, 48]. Furthermore, their encoding genes may recombine within a given strain, giving rise to novel specificities [49, 50]. Type II R-M systems have been intensively studied for decades due to the wide application of restriction enzymes to molecular biology [45, 51], largely due to their digestion of DNA at known sites. These target DNA motifs are usually short palindromic sequences which are protected from digestion through methylation by the accompanying MTase. Type III R-Ms have a similar genetic organisation [52], yet differ in that DNA digestion occurs downstream of one of two unmethylated, asymmetric recognition sites [53]. In contrast, Type IV systems target methylated DNA and have distinct genetic structures [33, 54, 55]. Because of these factors, they have received relatively little research attention.

The relevance of R-M systems to the dairy industry is underlined when one considers that they have been shown to target both incoming phage [56, 57] and plasmid [58] DNA, with the former considered potentially detrimental and the latter beneficial from a biotechnological perspective [27]. The somewhat indiscriminate nature of these systems is reflective of the overall dilemma faced by lactococcal strains regarding the fitness cost of maintaining plasmids [59–61] in the dynamic milk environment [62]. On balance, however, it appears that lactococci prefer to maintain plasmids, as evidenced by their prevalence in the genus [16]. This strategy confers remarkable adaptability to the species by transfer of traits necessary for survival through mobile, chromosome-independent DNA elements.

In the current study we performed complete genome sequencing of 43 strains of L. lactis and L. cremoris, including comprehensive determination and analysis of the MGEs, defensome, and methylation landscape, in one of the largest studies of its kind in LAB. Mapping of each detected methylated motif to its cognate methyltransferase (M) and/or specificity subunit (S)-encoding gene was performed, and the antiphage activity of selected R-M systems, including that of truncated S subunits, was proven.

Materials and methods

Bacterial strains and culture conditions

Bacterial strains applied to this study are listed in Supplementary Table S1. Lactococcal strains were routinely cultivated at 30°C in M17 broth (Oxoid, UK) supplemented with 0.5% lactose (Merck, Germany; LM17) or glucose (Merck, Germany; GM17), and 10 g/L technical agar (Neogen, USA) and 5 μg/mL tetracycline (Merck) where appropriate. For cases in which plasmid loss was suspected, 10% reconstituted skimmed milk (RSM) was used as medium prior to LM17 (as above) ahead of DNA extraction. E. coli strains were maintained at 37°C in Luria–Bertani (LB) broth (Merck, Germany) with the addition of technical agar (Neogen), 100 μg/mL ampicillin and glucose (Merck; 0.5%) where appropriate. All strains were stored in 20% glycerol (Fisher Bioreagents, USA) or 10% RSM at −20°C.

DNA isolation and genome sequencing

Genomic DNA was isolated from wild-type lactococcal strains or cloning intermediates using a Monarch Genome Purification kit (T3010; NEB, MA, USA) as per the manufacturer's instructions or the Macherey-Nagel NucleoBond system, utilising Buffer Set III and AXG or AX 100 columns (Macherey-Nagel, Germany). This was performed according to the manufacturer’s instructions except for the following: initial cell growth proceeded at 30°C until the OD_600nm reached at least 0.8. At the cell lysis step, the volume of lysozyme (100 mg/mL; Sigma-Aldrich, Germany) added was increased to 40 μL, and 50 μL mutanolysin (5 000 units/mL; Sigma–Aldrich) was also added prior to incubation, which proceeded for an increased time period of 1 h. In addition, where necessary, insoluble components were removed from the supernatant by centrifugation up to 10 000 x g for 10 min prior to loading on the column. DNA quantity was measured by Qubit 2.0 (Life Technologies, USA) and quality assessed by visual inspection using agarose (1%; Merck, Germany) gel electrophoresis.

SMRT sequencing was performed by first shearing genomic DNA to an average size of ∼10 kb using the G-tube protocol (Covaris, USA). DNA libraries were prepared using a SMRTbell express template prep kit 2.0 (100–938–900, Pacific Bioscience, CA, USA) and ligated with hairpin barcoded lbc adapters. Incompletely formed SMRTbell templates were removed by digestion with a combination of exonuclease III and exonuclease VII (NEB). The qualification and quantification of the SMRTbell libraries were made on a Qubit fluorimeter (Invitrogen, USA) and a 2100 Bioanalyzer (Agilent Technologies, USA). SMRT sequencing was performed using an RSII or SQ1 (Pacific Biosciences, CA, USA) instrument based on the multiplex protocol for 10 kb SMRTbell library inserts. Sequencing reads were collected and de novo assembled using the Microbial Assembly v10.1.0.1119588 [63] program with default quality and read length parameters. In addition to genome assembly, the SMRT Analysis pipeline from Pacific Biosciences (http://www.pacbiodevnet.com/SMRT-Analysis/Software/SMRT-Pipe) enables the determination of the epigenetic status of sequenced DNA by identifying the m6A and m4C modified motifs [34, 37, 40]. Standard SMRT sequencing cannot directly detect 5mC modifications, therefore we applied a modified SMRT Tet2 assisted protocol that includes an additional step of 5mC oxidation by the Tet2 module from the EM-seq kit (E7120, NEB) before PacBio sequencing of samples where 5mC methylation was predicted [64]. The identified methylation motifs were refined based on three criteria: (i) a mean modification QV cut-off of 40%, equivalent to a P-value of < 0.0005 was applied; (ii) secondly, motifs of unknown type were removed; (iii) motifs methylated at less than 50% of possible positions were removed.

Short read genome sequencing was performed using Illumina MiSeq platform run by GenProbio S.r.l. (Parma, Italy). Bacterial genome libraries were prepared using an Illumina Nextera XT DNA Library Preparation Kit (Illumina, CA, USA). Libraries were quantified using a fluorometric Qubit quantification system (Life Technologies, CA, USA), and additional quality control was performed using a 2200 TapeStation instrument (Agilent Technologies, CA, USA). Sequencing then proceeded using the Illumina MiSeq platform with a 600-cycle flow cell v3 (Illumina, CA, USA).

Where required, Nanopore sequencing (Oxford Nanopore Technologies, U.K.) was performed by Plasmidsaurus (OR, USA). Libraries were first prepared using a Rapid Barcoding Kit 96 v14 (Oxford Nanopore Technologies). Sequencing was carried out using a PromethION P24 instrument with R10.4.1 flow cell, bases called using the ont-doradod-for-promethion v7.1.4 algorithm on super-accurate mode and adapters trimmed using the MinKnow software. Reads that passed a Q score threshold of 10 were incorporated into the below described assembly protocol(s).

Genome assembly

Multiple genome assemblies were conducted for each strain in order to evaluate genome completeness when different datasets were used as inputs. Raw data in the form of .fastq files were assembled using either Unicycler [65] v0.5.0 (in bold mode) or Trycyler [66] v0.5.4, whereupon assessment of each assembly was conducted based on (i) the number of contigs generated, (ii) the number of circular contigs generated, (iii) the overall comparison to previously published data, (iv) the general agreement with methylation data generated for that strain, and (v) the consistency of output between assemblies. Individual assemblies were then manually curated to remove duplicate or contaminant (e.g. bacteriophage) contigs. The assembly approach for each strain is detailed in each relevant Genome assembly page on the NCBI database (https://ncbi.nlm.nih.gov/datasets).

Circular plasmid contigs were then rotated to the RepA or B (as applicable)-encoding genes using Circlator [67] v1.5.5. employing a manually curated database of Rep-encoding genes from interrogation of the NCBI ‘protein’ repository using custom search parameters. Correct rotation of each individual contig was verified using Blastx [68]. All sequences generated were deposited to the NCBI database (GenBank [69] and assigned the accession numbers listed in Supplementary Table S1.

Molecular cloning, protein expression and bacteriophage assays

Plasmids used and bacterial strains applied as cloning intermediates are listed in Supplementary Table S4. An L. cremoris NZ9000 mutant carrying deletion of a 5,945 bp DNA fragment encoding a complete Type I R-M system and termed VES7862 was constructed as recently described [70]. Primers used in this study are listed in Supplementary Table S4. Molecular cloning was performed using standard techniques [71] or using NEBuilder HiFi DNA assembly Master mix (NEB) using enzymes provided by either NEB or Thermo Fisher (MA, USA). All primers were synthesised by IDT (Leuven, Belgium or IA, USA) or Eurofins Genomics (Ebersberg, Germany). Plasmid DNA was isolated from cells using the GeneJET plasmid kit (Thermo Scientific, USA) or the Monarch Plasmid Miniprep kit (T1010, NEB, USA). Where applicable, plasmid DNA sequence was verified using enzymatic digestion, Sanger sequencing (performed by Eurofins Genomics), and/or Oxford nanopore sequencing (performed by Plasmidsaurus).

Cloning intermediate strains (VES7862, JM101 or ER2796; Supplementary Table S4) harbouring the plasmid of interest (or empty plasmid; pPTPi, pNZ8048 or pWSK29) were then subjected to protein induction using 1–10 ng/mL nisin (Merck, Darmstadt, Germany) or 0.5 mM IPTG (Fisher Scientific). Incubation proceeded at 30°C or 37°C until the OD_600nm reached at least 0.8, whereupon the cells were either employed in phage assays (as below) or harvested by centrifugation and stored at −70°C prior to DNA isolation (as described above).

Validation of R-M activity was performed by inducing the constructs and the empty plasmids (as a negative control) as above and (i) subjecting the genomes to SMRT sequencing (as above) to determine the methylation status, and/or (ii) testing the strains for phage-resistance activity against lactococcal Skunavirus sk1, p2, jj50, 712, and Ceduovirus c2, using the double agar plaque assay method as described previously [72]. EOP data were transformed, statistically analysed and visualised using GraphPad Prism v10.3.1 for Windows (Graphpad software, CA, USA).

In silico analyses

Annotation was largely performed by NCBI using the PGAP annotation pipeline [73–75]. Artemis [76] (v18) genome browser and annotation tool was used to inspect and (where necessary) manually curate ORFs and to calculate G + C percentages of individual DNA molecules. ORF annotations were refined where necessary using alternative databases; Interpro [77], HHpred [78] and Uniprot/EMBL [79]. All sequence comparisons at the protein level were performed via all-against-all, bi-directional BLAST alignments. The alignment cut-off criterion was an E-value < 0.0001, with >30% amino acid identity across 80% of the sequence length.

The primary scanning of the assembled genome for putative methyltransferase genes was performed using the Seqware program [80]. Additional searches for MTase genes were performed using HMMer (HMMER 3.3.2; http://hmmer.org/) to annotate each predicted protein coding sequence feature in the genomes. Genes matching MTase sequence profiles were further examined by structure prediction using the ColabFold implementation of AlphaFold2 [81, 82], followed by structure similarity search using predicted MTase models as query inputs to DALI [83].

Prophage content was determined by PHASTEST v3.0 [84, 85]. The output of this and analyses thereof is given in full in Supplementary Dataset S1.

Antiphage system detection was initially performed by the online Prokaryotic Antiviral Defence Locator (PADLOC) v2.0.0 with PADLOCDB v2.0.0 [7] and DefenseFinder [86]. These two sets of results were combined and redundancies were removed. Considering that neither PADLOC nor DefenseFinder currently identify superinfection exclusion (Sie) systems [87], the Tab [88] system, newly identified Abi-like systems [9], or some orphan R-M genes (i.e. genes not located adjacent to other R-M system genes), gene candidates in these categories were manually searched for. In the case of the Abi-like systems, this was performed by querying the nucleotide sequences of all genomes with one or more amino acid sequences representative of the above defence systems using the tblastn tool [89]. In the case of Sie systems, the protein sequences used in a similar way are listed in Supplementary Table S2. In the case of orphan R-M system components, genes not recorded by either PADLOC or DefenseFinder but listed in the REBASE [45] entry for the strain were used similarly. Redundancies (which had already been accounted for) were removed and the (appropriate) remainder added to the catalogue of lactococcal antiphage genes.

Enumeration of complete phage defence systems was in the first instance performed by combining the ‘systems’ outputs generated by PADLOC and DefenseFinder and removing redundancies in a similar manner to that described above. For those systems not detected by either tool, a determination of completeness was made based on previously published predictions and/or published experimental results. R-M systems were considered complete if they consisted of an R, M and (at least one) S-encoding gene (Type I), a single gene (Type I S), one M and one R (and/or one RM)-encoding gene (Type II), one M and one R (and/or one RM)-encoding gene (Type III) and either one or two Mcr-encoding genes (Type IV). Orphan Type I specificity subunits, whether full length or truncated, were considered to constitute distinct R-M systems in themselves, due to their apparent ability to combine with distantly encoded R and M subunits to methylate and subsequently target a distinct DNA motif. Additional S-encoding genes adjacent to already complete Type I systems were considered as part of the adjacent system. Orphan genes outside of these categories and frameshifted single-gene systems were considered to be incomplete. A detailed breakdown of all antiphage system-associated genes is given in Supplementary Dataset S2. The co-ordinates of the chromosomal complements of antiphage genes were normalised for a chromosome size of 2 595 288 bp and visualised by mapping to 43 separate tracks in DNAPlotter [90] which was implemented in Artemis [76]. pUC147G was visualised in part using Benchling (https://benchling.com/).

Putative integrated elements were identified using ICEfinder [91] using the complete nucleotide sequences of each genome as inputs. The output nucleotide files were subjected to dereplication using Cd-hit v4.8.1 (cd-hit-est) [92] to determine the number of unique elements, and the output amino acid files interrogated using the Pfam database v37.0 [93]. Where different Pfam families were assigned to proteins with similar function, groupings were manually curated to give the categories represented in Fig. 2B.

Figure 2.

(A) Normalised genomic positional distribution of detected ICEs/IMEs (light blue) in 43 lactococcal strains. The approximate GC skew is also shown (inner circle). (B) The number of deduced proteins encoded on ICE/IME, by predicted function. (C) Distance map showing a Hadamard matrix of all ANI values of predicted ICE/IMEs in 43 L. lactis / L. cremoris strains.

Pan-plasmidome analysis was performed by retrieving all (at time of analysis) complete, non-redundant plasmid sequences associated with L. cremoris (n = 168) and L. lactis (n = 286) from the RefSeq database (NCBI) using a custom search term. Plasmids sequenced as part of this study (n = 269) and not already captured in this search (n = 98) were then added to this database, giving a total of 552 sequences. These sequences were then annotated using Prokka v1.11 [94]. The .gff3 output from Prokka was used as input to Roary [95] v3.13.0 to analyse the pan-plasmidome. The resulting ‘genes-in-pan-plasmidome’ values were then fitted to a curve using a non-linear regression (power series) using GraphPad Prism (USA).

Data transformations and ANOVA calculations were performed (and all graphs visualised) using GraphPad Prism. ANI calculations (ANIm [96] of plasmids and ICE/IME regions were performed using Pyani [97] v0.2.12 implemented in Python v3.8.18. Distance calculations were performed and distance trees visualised using Orange [98] v3.37.0.

The core genome phylogenetic tree (Supplementary Fig. S1) was generated by first annotating all chromosome sequences using PROKKA [94], the outputs of which were used as input for Roary [95] for pan-genome analysis. A core genome alignment was then performed and a phylogenetic tree constructed using FastTree [99].The resultant tree was visualised using ITOL [100]. Annotations were added manually.

Structural predictions were performed with AlphaFold3 on google servers at golgi.sandbox.google.com [101]. The pLDDT values of predicted structures were stored in the PDB files as B-factors. Visual representations of the structures were prepared with ChimeraX [102]. The coordinates of predicted structures are accessible on Zenodo (zenodo.org).

Results & discussion

Genome sequencing, assembly, and plasmidome retrieval

To establish an overall view of the dairy lactococcal mobilome, complete sequencing of the genomes of 43 distinct lactococcal strains was performed. An overview of the genomes and accession numbers are given in Supplementary Table S1. Previously, our group suggested that a combined SMRT and Illumina sequencing and assembly approach was useful for resolving plasmids which were not returned by SMRT sequencing and assembly alone [16]. Despite this strategy being applied in part to the present strain cohort, upon more detailed analysis, genome-methylome discrepancies were encountered—i.e. some methylated motifs could not be ‘matched’ to a plausible specifying enzyme encoded on the corresponding retrieved genome. It was hypothesised that one or more plasmids may be ‘missing’ from the sequencing dataset, due to the documented issue of size-dependent plasmid retrieval [103] and due to the presence of so-called ‘orphan’, Type I R-M system-associated, specificity (S) subunits encoded on small plasmids. If such a plasmid is not retrieved during assembly, this creates a discrepancy between the deduced methylation profile and the genes encoding the enzymes responsible for this profile.

The importance of recovering the full plasmidome of target strains cannot be overstated since the proteins encoded on small plasmids can have a global effect on the phenotype of a given strain. This will be shown here in the context of methylation, an industrially relevant trait, but extends to clinically relevant phenotypes such as antibiotic resistance [104, 105]. In our specific case, satisfactory assemblies were achieved using a combination of Illumina MiSeq short reads, and SMRT or ONT long reads using either short reads first (SRF; Unicycler [65]) or long reads first (LRF; Trycyler [66]) assembly approaches. Using this process, 40 additional plasmids were retrieved compared to previously employed sequencing and assembly methods applied to these strains, such as SMRT sequencing and assembly using the SMRT portal platform [16]. While this represents a significant improvement on the initial retrieval effort, it should be noted that this does not preclude additional plasmids being present in these strains. Discrepancies, however, are presumed to be minimal due to the complete delineation of the deduced methylomes.

In all cases, genome completeness was evaluated by manual assessment of the (i) number, (ii) size, (iii) circularity, (iv) multimeric state, and (iv) content of contigs returned from each approach, as well as the consistency of the same between assembly methods. To illustrate the diversity of the collection, a core genome phylogenetic tree is provided in Supplementary Fig. S1, including the CWPS (sub)grouping, number of plasmids, and presence of the major groups of antiphage systems in each strain is indicated. In the first instance, to establish the mobilome, three types of mobile elements were explored: plasmids, integrative conjugative (or mobile) elements (ICEs/IMEs), and prophages, which are discussed in further detail below.

Mobilome I: the dairy lactococcal plasmidome

The abundance, size variation and genetic content of plasmids is a truly remarkable feature of dairy lactococcal strains [16, 20–21] and some of the salient features of the 269 plasmids identified in this study are presented in Table 1, including the largest plasmid (pUC073A) and plasmidome (UC073; by bp) currently in the NCBI database for L. cremoris and L. lactis species. Of note, none of the characteristics reported in Table 1 can clearly differentiate L. cremoris from L. lactis.

Table 1.

Properties of plasmids sequenced in this study

Characteristic	Value	Range among strains sequenced in this study
Total number of plasmids	269	3–11 plasmids per strain
Average plasmid size	31,139 bp	1,983 (pUC147J) – 211,414 bp (pUC073A)
Average plasmidome	194,800 bp	58,329 (UC047) – 434,793 bp (UC073)
Average % of genome	7.4%	2.2 (UC047) – 14.9% (UC147 & UC073)
Average plasmids/strain	6.3	3 (NCDO702 & UC047) – 11 (UC147)
Average plasmid G + C %	33.9%	29.5% (pHPE; pUC029E) – 40% (pHPB)

The biotechnological importance of plasmids to dairy lactococci cannot be overstated. The plasmid-encoded nature of the gene clusters for lactose utilisation, casein proteolysis and peptide uptake, for example, makes certain plasmids essential for their survival in the dairy environment [15, 21]. Previously, it has been shown that ten of the twelve plasmids present in the genome of L. lactis FM03P encode functions critical to adaptation to milk [21], and that plasmid-encoded genes can influence the global transcriptomic landscape of the species [59].

To assess the diversity of the dairy lactococcal plasmidome, a database of all known and complete plasmids of L. cremoris and L. lactis was constructed, yielding a total of 552 plasmid sequences (which included sequences from the RefSeq database and those sequenced in the context of the current study). These were annotated (using PROKKA [94]) and used as inputs for Roary [95] to calculate the number of genes in the pan-plasmidome. Visualisation of this dataset is provided in Fig. 1A. The general asymptotic trend of the deduced curve, as well as the exponent value generated (0.5488) both indicate that the pan-plasmidome remains open (i.e. that the genetic diversity of the plasmidome will be expanded by the addition of further sequences), despite the inclusion of 552 plasmids encompassing a total of 3759 genes. Interestingly, and as a further indication of the diversity of the plasmid complements of dairy lactococci, an empty core genome was returned following pan-plasmidome analysis, i.e. no single gene family was present in all plasmids analysed. A general overview of plasmid diversity (using a dereplicated dataset) was therefore generated by average nucleotide identity (ANI) analysis [97] and is visualised in Fig. 1B.

Figure 1.

(A) Nonlinear regression analysis of dairy lactococcal pan-plasmidome genes, indicating that the pan-plasmidome remains open in L. lactis and L. cremoris. The deduced mathematical equation is given (inset). (B) Distance map showing a Hadamard matrix of all ANI values of the entire currently available plasmid cohorts of L. lactis and L. cremoris (n = 552).

Mobilome II: ICE/IMEs

In their capacity as transferable genetic elements encoding genes beneficial to the host [28], putative ICE/IMEs were detected using the ICEfinder 2.0 web tool [91] and their corresponding gene complements interrogated for potential antiphage systems. Fig. 2 summarises the findings thereof.

A total of 75 ICE elements were detected in the 43 strains analysed herein, notably more than were detected in a recent study [28], in which a total of 36 ICEs were detected in 69 strains. This discrepancy is likely due to different methods of ICE identification—i.e. the use of (in this case) ICE finder or manual curation. Indeed, the majority (54 of the 75) of ICE/IMEs detected here were described as ‘without identified DR [direct repeat]’ by ICEfinder, in reference to the absence of a detectable integration site which represents the junction between the chromosome and formerly extra-chromosomal element [106], though it was not determined if the absence of this affects transferability. Integrative elements appeared to be well distributed throughout individual genomes (Fig. 2A), with the exception of the 2.25–2.5 Mb region, which may encode core cellular functions not prone to disruption by horizontal gene transfer. The identified ICE/IME elements were shown to harbour genes encoding proteins with a diverse range of predicted functions (Fig. 2B), but included putative integrases, replication proteins, oriT sites, relaxases and components of Type IV secretion systems, consistent with previous studies [106]. A snapshot of the diversity of integrated elements (at the nucleotide level) is presented in Fig. 2C. Additionally, dereplicating the identified ICEs and IMEs resulted in 32 distinct clusters (being less than 90% identical at the nucleotide level), encoding functionally similar genes (Fig. 2B). Although ICEfinder 2.0 attempts to identify accessory modules such as virulence factors within these elements, this feature does not extend to antiphage systems.

Mobilome III: prophages

All complete genomes generated as part of this study were subjected to prophage detection by PHASTEST, followed by manual inspection and verification of the gene content of selected prophages. An overview of this analysis is presented in Table 2, with an expanded version of this table provided in Supplementary Dataset S1.

Table 2.

Summary of prophage content of 43 L. lactis and L. cremoris genomes. Average total, intact, and incomplete prophages per strain are given in parentheses

	Total	L. cremoris (n = 23)	L. lactis (n = 20)	Range
Total prophages detected	234	87 (3.8/strain)	147 (7.3/strain)	0 – 9
Intact	91	28 (1.2/strain)	63 (3.2/strain)	0 - 6
Incomplete	143	59 (2.6/strain)	84 (4.2/strain)	0 - 6

234 prophage regions were predicted across the 43 genomes by PHASTEST and were found to have an average length of 34.8 kbp and G + C contents in the range of 32.0–40.9%. Manual adjustment of the predicted prophage region lengths provided by PHASTEST to include only the regions encoding phage genes (as opposed to between predicted attL and attR sites), as well as inspection of the gene content of those regions was performed to more accurately assess prophage completeness and to reassign ‘questionable’ regions returned by this bioinformatic tool as either intact or incomplete, thereby simplifying the analysis (Table 2). Interestingly, a stark difference in the number of predicted prophages was observed between L. lactis and L. cremoris (approximate ratios of 2:1 for total prophages and 3:1 for intact prophages) and in line with this finding, the chromosomes of the analysed L. lactis strains had an average length approximately 40 kbp greater than their L. cremoris counterparts. These characteristics may serve as informal differentiators of these species in addition to the original parameters of average nucleotide identity (ANI) values, digital DNA-DNA hybridisation (dDDH) values and the sequences of certain hallmark genes [107].

Nearly all currently known lactococcal prophages are members of the P335 phage group [108], and the lifestyles, genetic organisation, and structures of these phages have been studied in detail [29, 108–111]. The vast majority of the prophages identified in the analysed genomes were similar to previously characterised members of the P335 phage group, including suspected ‘satellite’ phages bIL310, bIL311, and bIL312 [112], though some unusual members were also detected, including phages encoding proteins with significant amino acid similarity to those infecting Lactococcus garvieae (on the chromosomes of NCDO702 and JM3), Corynebacterium xerosis (NCDO702), Streptococcus pyogenes (UC06), Streptococcus dysgalactiae (NCDO700), and Bacillus subtilis (UC023). Furthermore, a prophage similar to lactococcal skunaviruses was detected, as previously described by our group [29].

It has been shown that resident prophages can affect lactococcal host growth physiology and susceptibility to bactericidal agents including phages [32], a likely result of antiphage systems being encoded within the prophage genome itself, as has been observed in lactococci and other genera [87, 113].

Antiphage system survey

Having determined or predicted the mobile genetic elements as described above, all 43 genome sequences were then subjected to both PADLOC [7] and DefenseFinder [86] analysis to determine antiphage gene content. This cataloguing was supplemented with manual addition of systems and genes not currently detected by those tools (i.e. Sie systems [87], the Tab system [88], recently described Abi-like systems [9] and orphan R-M system components). An overview of the general distribution of mobile and non-mobile antiphage systems (N.B. only those determined to be complete are shown) is given in Fig. 3A. Fig. 3B provides a more detailed overview of antiphage system types, cross-correlated with genomic locations corresponding to detected mobile elements, with a detailed breakdown of all systems and their genomic locations given in Supplementary Dataset S2.

Figure 3.

Overview of the intragenomic distribution of dairy lactococcal phage resistance systems. System categories were assigned based on references [6–8, 137–153]. (A) Proportions of predicted antiphage systems on mobile and non-mobile components of 43 lactococcal genomes. (B) Percentage of antiphage systems according to system type and chromosomal or plasmid location. Absolute numbers of system category genes text in the relevant bars. PDC, phage defence candidate. (C) Distribution of all chromosomally located antiphage genes (dashes, coloured in a gradient from strain 1 (outside) to strain 43 (inside)) identified in the current study. The paucity of clearly discrete clusters of antiphage genes indicates the absence of obvious defence islands.

Fig. 3C shows the distributions of all antiphage system genes detected in this analysis across all 43 chromosomes, with all positions having been adjusted to account for varying chromosome sizes. It is noteworthy that the distribution of these genes seems to be remarkably even, indicating the apparent absence of discrete phage defence islands observed in other genera [70].

Overall, 59.6% of the deduced antiphage systems were also predicted to be located on mobile regions of the genomes—either plasmids, ICEs, or prophages (Fig. 3A), highlighting the potential for transfer or possibly the recent acquisition of these genes. Strikingly, 146 of the 269 plasmids sequenced in this study contain at least one antiphage system or component thereof. Furthermore, and based on an analysis of the number of bases predicted to encode antiphage systems in the present cohort of strains, an average of 0.5% of chromosomal DNA comprises antiphage systems, while the equivalent figure on plasmids is eight times higher at 4%. Extrapolating from this, it is likely that the dairy lactococcal plasmidome plays a significant role in the antiphage capabilities of these strains.

A discussion of five of the most prevalent system categories (Superinfection exclusion, toxin-antitoxin, abi-like, candidate systems, and R-M systems) follows below.

Superinfection exclusion systems

Sie systems are (generally) prophage-encoded antiphage systems which in lactococci have been reported to block or otherwise interfere with the DNA injection step of incoming phage infection [87, 114]. Eight protein sequences were used as probes to detect resident Sie systems in our strains (Supplementary Table S2), six of which have been experimentally verified as possessing antiphage activity [87] and two which have not [112]. Identical or near-identical proteins matching all eight examples were identified as being encoded on the genomes of the current strain cohort, indicating their importance in dairy lactococci. Following this, the protein encoded adjacently to the putative Sie was subjected to HHPred [78] analysis to assess its function as metallopeptidase, which may be required for certain Sie types to elicit their antiphage function [87]. In the absence of this gene, the Sie protein itself was categorised as a complete system if it was predicted by the DeepTMHMM [115] tool to contain membrane-spanning domains. Of the 48 Sie systems identified here, 47 are associated with prophage elements (either complete or incomplete), which usually harbour these systems at defined locations within the prophage genome [87]. Of note, the mobility of incomplete prophages has not been tested here, as such the proportion of incomplete prophages which are actively mobile (as satellites) is not known.

Toxin-antitoxin systems

A total of 16 toxin-antitoxin (T-A) systems were identified in the current study, 10 of which were classified as the RosmerTA system (previously shown to confer partial phage resistance—at 37°C—to E. coli [6]. Interestingly, 8 of these systems were detected within regions identified as ICEs or IMEs, and two were located on apparently non-mobile chromosomal regions. The presence of industrially relevant characteristics located on lactococcal ICEs has recently been explored [116], and included a number of other phage defence systems, however it is unknown if the RosmerTA system’s primary function is phage defence or an aspect of self-preservation on the part of the mobile element. Examples of the MazEF TA [117] and DarTG TA [118] were also found. According to BLAST searches, these are examples of only a handful of these genes being encoded by dairy lactococci, highlighting the diversity of apparent antiphage systems encoded by the genus.

Abi-like systems

Abi-like systems are a common feature of the lactococcal antiphage repertoire and have been well studied [119]. These systems block phage multiplication and cause premature bacterial cell death upon phage infection. In this manner, Abi-like systems decrease the number of progeny viral particles and limit phage spread to other cells allowing the overall bacterial population to survive. New systems are still being uncovered, with examples of four of the six newly identified Abi-like systems [9] being present in the current strain cohort. Recent insights in their expression and mode of action indicate that, behind diverse phenotypic and molecular effects, these systems share common traits with the well-studied E. coli systems Lit and Prr [14]. Abi-like systems are widespread in bacteria, and analysis indicates that they play additional roles other than conferring phage resistance [14]. Abi-like systems appear to be widely distributed amongst mobile and non-mobile lactococcal elements (Fig. 3B). Despite this, not all known lactococcal Abi-like systems were detected in this cohort, highlighting the apparent vast diversity and uneven distribution of these systems.

Hypothetical, candidate, and other systems

One of the more notable features of the breakdown of antiphage system categories (Fig. 3B) is the abundance of hypothetical and candidate systems. These candidates are distributed almost evenly between MGEs and non-mobile regions of the chromosome. In some cases, such candidates have been hypothesised to be defensive in nature due to their encoding genes being genomically surrounded by known antiphage systems [7]. In addition to these candidate systems, a small number of systems with diverse mechanisms were identified in these strains, including 19 examples of the Viperin system [120], 6 examples of plasmid-encoded Bunzi systems [6], and 5 or fewer (non-mobile) examples of CBASS, BREX, Dodola, and RloC were found, amongst others (Supplementary Dataset S2). The presence of these systems reflects the diversity of antiphage systems in this genus. Indeed, their presence offers an exciting glimpse of the ‘dark matter’ of the lactococcal antiphage landscape and may form the basis of future studies.

R-M systems

The prediction of lactococcal R-M systems and their cross-correlation with MGEs in these species offered some fascinating insights into how the components of these systems are distributed and combined. In order to elicit an effect, so-called orphan Type I S subunits must interact with other Type I R-M components which may or may not be encoded within predicted MGEs, thus rendering the enumeration of complete Type I systems challenging. Considering that the specific compatibilities of orphan Type I S, R, and M enzymes are not currently known, the M and R units combining plasmid-encoded orphan S subunit with were not assigned. Therefore, for the purposes of our study, each orphan, plasmid-encoded S subunit was considered to be a complete mobile R-M system in its own right. Further detail on the enumeration of complete systems is given in the Methods.

In spite of this disparate spatial distribution, it is evident (Fig. 3B) from this study that R-M systems (and components thereof) are preferentially encoded on plasmids. Of a total of 179 complete R-M systems identified herein, 64% of those were found to be encoded on plasmids—which increased to 71% when only active systems were considered. The prevalence of orphan S subunits (i.e. those whose encoding genes are not spatially linked to other components of the system) was somewhat surprising. While these orphans have been described in single strains of lactococci previously, our study indicates that they are extremely common in this genus and their deduced activity indicates that they are used to great effect in phage defence. In fact, a notable feature of the plasmids identified in this study was the number (n = 9) of plasmids that solely encode a replicase (Rep) and one or more orphan S subunits. Perhaps the most striking of these is pUC147G, which harbours three Rep-encoding genes and three Type I S subunit-encoding genes (Fig. 4), all with active and distinct DNA-binding specificities (see below). These findings indicate the presence of a large, mobilisable reservoir of Type I R-M functionalities, prompting a large scale, detailed study of the methylomes of the assessed lactococcal strains.

Figure 4.

Schematic of plasmid pUC147G, encoding three replication-associated genes (RepB) and three Type I R-M specificity subunits (each containing TRD1 and TRD2). The similarities of each TRD are indicated by colour coding, and methylated nucleotide sequences are indicated. All three S subunits are active in L. cremoris UC147.

Table 3 provides a summary of all actively methylated motifs detected in this study, broken down by Type and subtype of R-M system, specificity, location of responsible enzyme and assignment method. An expanded breakdown of the enzymes responsible for the 135 methylated motifs identified in this study is provided in Supplementary Table S3. The respective prevalence of R-M systems in Lactococcus has previously been described to occur in the order Type I, II, III, and IV (highest to lowest) [33], which is consistent with our findings. Below is a more detailed analysis of each of the R-M Types identified in our strain collection.

Table 3.

Summary of methylated DNA motifs in 43 strains of L. lactis and L. cremoris. Details of R-M system subtypes may be found on rebase.neb.com. Chromosomally located motifs not overlapping with a predicted mobile element are designated ‘non-mobile’. Motifs were assigned to their methylating enzymes by amino acid identity to previously described enzymes or target recognition domains, or by molecular cloning and subsequent SMRT sequencing. ^aUnique motifs indicate motifs not previously found in the REBASE repository at time of deposition. An expanded version of this table is given in Supplementary Table S3. *See Supplementary Tables S4 for cloning intermediates, plasmids, and primers used

	Total	Subtypes	Specificity	Location of responsible gene	Assignment	Motifs unique to this studya	Overlap of enzyme with predicted MGEs
Total methylated motifs	135	93 Type I 31 Type II 11 Type III	122 m6A 11 m5C 1 m4C 1 m6A-m5C	1 Chromosome (ICE/IME) 2 Chromosome (Prophage) 39 Chromosome (Non-mobile) 93 Plasmid-borne	110 by ID to validated genes 25 by molecular cloning*	46	95 overlap with MGEs 40 no overlap with MGEs
Type I	93	47 I 42 I gamma 4 I SP, gamma	93 m6A	1 Chromosome (Prophage) 16 Chromosome (Non-mobile) 76 Plasmid-borne	86 by ID to validated genes 7 by molecular cloning	41	77 overlap 16 no overlap
Type II	31	15 II 4 II alpha 2 II beta 7 II G, S 2 II G, S, alpha 1 II gamma	18 m6A 11 m5C 1 m4C 1 m6A-m5C	1 Chromosome (ICE/IME) 1 Chromosome (Prophage) 17 Chromosome (Non-mobile) 12 Plasmid-borne	16 by ID to validated genes 15 by molecular cloning	5	14 overlap 17 no overlap
Type III	11	11 III beta	11 m6A	6 Chromosome (Non-mobile) 5 Plasmid-borne	8 by ID to validated genes 3 by molecular cloning	0	5 overlap 6 no overlap

Type I R-M systems

Type I R-M systems are, by a considerable margin, the most common R-M systems encoded in the currently investigated strain collection, and (as based on a previous study) in Lactococcus as a whole [16]. These systems are composed of either one (‘Type IS’) or three subunits, which may be co-located (i.e. all genes in the system are adjacent on the genome) or dispersed (i.e. in which an orphan S subunit combines with other components encoded by genes at a distinct genetic locus). They can further be described as ‘active’ (i.e. producing a discernible and distinct methylation pattern) or inactive—however, this distinction may depend on cellular or environmental conditions giving rise to transcriptional up- or downregulation of these systems. All systems herein described as ‘active’ are considered active at the time of our analysis. Ninety-three such Type I systems were found in this study (Table 3).

In total, 149 Type I R-M systems were detected on the genomes of the 43 strains, eclipsing the number of Type II (n = 54), Type III (n = 14), and Type IV (n = 3) systems. Four active examples (methylating three distinct motifs) of the ‘I SP’ subtype of Type I R-M systems were also identified, in which all domains necessary for R-M (nuclease, helicase/ATPase, MTase, TRD) are encoded in a single polypeptide [121, 122]. In total, the identified active Type I systems were shown to methylate 93 motifs, 41 of which were unique to this study.

We further investigated elements of these Type I R-M systems by analysing the deduced number, genomic location, and identified activity of orphan S subunits (Fig. 5). Remarkably, not only did 47% of the predicted S subunits represent orphans, but 81% of these orphans were found to be plasmid encoded. This finding illustrates the highly mobile and flexible antiphage strategy employed by lactococci, who share a reservoir of plasmids containing (in effect) single genes which confer significant protective effects. As illustrated in Fig. 5, a significant (43%) portion of this reservoir could be described as ‘silent’, i.e. the genes in question appear intact, but the associated motif was not detected as being methylated upon SMRT sequencing—this is discussed further in a later section.

Figure 5.

Distribution of active and inactive (at time of sequencing) Type I R-M specificity (S) subunit-encoding genes amongst the chromosomes and plasmids of 43 dairy lactococcal strains. The numbers of orphan S subunit-encoding genes, as well as those adjacent to R and M component-encoding genes, are also given.

Type II R-M systems

Historically, it has been surmised that Type II R-M enzymes are the most abundant in the LAB [123]; however, this was challenged in the case of dairy lactococci [16], a finding which is substantiated here. The 149 Type I and 54 Type II R-M systems were identified in the lactococcal genomes analysed in the present study, renders the ratio of Type I to Type II systems 2.75:1.

Of the 31 Type II motifs identified, five were unique to this study at time of deposition: AGCYAC (by R-M.LcrJM1V), GRTAAAT by R-M.LcrJM3I, VTCGAB (by M.Lla303II), GGYAAG (by R-M.Lcr147VIII), and GMAGG (by LcrJM3II + M1/M2.LcrJM3II). In stark contrast to Type I systems, the majority (72%) of Type II R-M genes were chromosomally encoded. Interestingly, Type II R-M components were the only representative of the R-M types to be encoded by prophages, and the only active prophage-linked R-M system was located within a prophage that was predicted to be incomplete. This may indicate that these MTases remain dormant until such time as they are required by their host (in this case, the prophage).

Type III R-M systems

The 43 lactococcal strains analysed herein encode 14 Type III R-M systems, the majority of which are composed of one R enzyme and one M enzyme, however, other configurations (such as a second M) were also observed. All methylate an adenine base within motifs of either five or six bases, either symmetric or asymmetric. Interestingly, no orphan components were observed, i.e. each R, fused R-M and M is co-located with its partner gene on the host genome. No Type III methylated motifs were unique to this strain collection, and components were approximately evenly distributed between chromosomes and plasmids (6 and 5 systems, respectively).

Type IV R-M systems

Functional Type IV R-M systems have not been described in dairy lactococci, and the paucity of these systems in the current collection suggests that these strains are no exception. Nonetheless, a complete Type IV, methyl-directed system is encoded by L. cremoris UC06, with LLUC06_2391 encoding an McrC family protein and LLUC06_2392 encoding an McrB methyl-dependent restriction endonuclease, the functions of which have been described in detail elsewhere [124]. In addition, an orphan McrB enzyme is encoded by D.1.7 (LLD17_01400). As these systems appear to be rare in dairy lactococci, they may be inactive or may not confer significant benefit to these strains. In contrast to e.g. Type II systems, Type IV systems are relatively non-specific in terms of nucleotide motif targeting, and therefore difficult to conclusively prove as being active. As such, the activity of these systems were not tested in this study.

Unmethylated genomes and dormant R-M systems

Interestingly, SMRT sequencing of the genomes of 43 strains revealed that four (D.1.7, HP, W34, and WM1) apparently do not methylate their DNA to a detectable level, as evidenced by a lack of motifs detected using SMRT sequencing (Supplementary Table S3). Strains D.1.7 and HP are predicted to encode one Type I R-M system each—in the case of D.1.7, the system is encoded by ORFs LLD17_06665, _06670, _06690, and _06695, but is interrupted by two transposase-encoding genes which may thus negate its ability to produce a functional system. ORFs LLHP_06345, _06350, _06365, and _06370 are interrupted by a single transposase-encoding gene which may have a similar effect. Strains W34 and WM1 are predicted to encode single orphan S subunits only (LLW34_03640 and LLWM1_04435). In all cases, therefore, the absence of detected methylated motifs can be explained. Interestingly, the unmethylated genomes of the four strains discussed above did not appear to harbour an abundance of currently known non-R-M phage resistance systems, as might have been expected, though this does not preclude heretofore unknown systems being encoded. Indeed, strain W34 apparently harbours comparatively few systems (n = 7), compared to other non-R-M and R-M harbouring strains. While this could be the result of ‘domestication’ (i.e. adaptation of the strain to the relatively phage-free laboratory environment), it also raises the possibility of undescribed phage resistance systems being encoded by this strain and others.

Another interesting feature of these R-M systems is the frequency of those that, though appearing genetically intact, do not methylate their host genomes (and by inference, likely do not exhibit nuclease activity). For example, and as mentioned above, of a total of 156 detected Type I R-M specificity subunits, only 89 (57%) were active at the time of sequencing based on the methylation state of the DNA motifs they are predicted to target. R-M systems are known to be tightly regulated, due to the hazard of R activity without corresponding protection being conferred by the M (and S) component [125], and the knock-on effects of DNA methylation on the cell [126]. It has been shown in E. coli that low level expression of a Type II system is highly effective in reducing sensitivity to phages [127]. It may be the case that apparently ‘inactive’ R-M systems retain their effectiveness despite not methylating their hosts’ DNA to a level detectable using methods applied in this study. Regardless, the fact that apparently dormant R-M systems (or components thereof) are intact and prevalent within dairy lactococci suggests that they represent a readily mobilisable antiphage reservoir.

Antiphage activity of selected R-M systems

The activity of selected R-M systems and orphan components were then experimentally verified. One example of Type II and III systems, and two examples of Type I systems were provided to a sensitive host (Fig. 6  i, ii, iii, v), whereupon it was observed that each system provides a broad-acting, yet moderate level of protection against four Skunavirus phages and one Ceduovirus phage (a representative P335 phage infecting this strain is not currently available). This limited anti-phage activity of single systems may partly explain the need for ‘stacking’ of R-M systems (or their components) which is evident on lactococcal genomes and has been previously shown to increase phage resistance [128]. Indeed, of the 43 strains analysed, all but seven encode more than one active RM system (Supplementary Table S3).

Figure 6.

Reduction in the observed titer of five distinct phages on the R-M free L. cremoris VES7862 harbouring the pPTPi (or pPTPi + pNZ8048) plasmids containing distinct R-M systems or components thereof. EOP reduction is presented for (i) a complete Type I R-M system (Lcr023III) (ii) a Type II system (Lla7005III) (iii) a Type III system (Lcr007I) (iv) a Type I system (Lcr147I) in which the S subunit-encoding gene has been truncated before TRD2 (TRD1-S.Lcr147I) (v) the complete Lcr147I system (vi) Lcr147I with a distinct, additional, truncated S subunit (TRD1-S.Lcr033I) provided on pNZ8048 and (vii) Lcr147I with the full additional S.Lcr033I provided, encoded on pNZ8048. The DNA motifs methylated by each system or stacked systems are provided in bold above the graph, with blue text indicating the methylated base (or complement thereof). The numbers of these sites present on each phage genome are provided in text boxes below the motif. * indicates a palindromic methylated motif. ¹ indicates a predicted assignment. All experiments were conducted after nisin induction and the efficiency of plaquing (EOP) values, of which the logs to base 10 are provided for clarity, are the average of three independent experiments. Error bars indicate the standard deviations of the transformed values.

Interestingly, the two Type I systems tested (Lcr023III and Lcr147I) were variably active against the five phages tested, the provided resistance level of which correlated with the number of corresponding R-M target sites on the phage genomes (Fig. 6, boxed insets), as has been shown previously [57] and discussed further below. In general terms, the R-M systems tested (in particular the Type I and II systems) were less inhibitory of phage c2 than of the 936 phages. Although the reason for this is unknown, the species difference between the two phage groups may be a factor, with the possibility of heretofore unknown anti-defence systems potentially having a role in this differing response.

Truncated S subunits facilitate methylation and increased phage resistance

Previously, it has been suggested that the genes of Type I S subunits encoded by many bacterial genera have the potential to recombine leading to altered methylation profiles. Furthermore, truncated S subunits encompassing a single TRD have been shown to dimerise and consequently mediate methylation of palindromic DNA motifs [129–131]. However, to the best of our knowledge the methylation patterns of such systems have not been experimentally studied in light of their associated antiphage efficacies. We, therefore, endeavoured to prove the functionality of such systems in Lactococcus in terms of methylation (motifs) and phage resistance.

To that end, a phage susceptible host (VES7862) was supplied with R-M system components as follows: a Type I system (Lcr147I) in which the S subunit-encoding gene has been truncated before TRD2 (Fig. 6  iv), the complete Lcr147I system (Fig. 6  v), Lcr147I with a distinct, additional, truncated S subunit (which represents a single TRD: TRD1-S.Lcr033I) provided on pNZ8048 (Fig. 6  vi), and Lcr147I with the complete S.Lcr033I, again provided on pNZ8048 (Fig. 6  vii). The Lcr147I-S.Lcr033I pair was chosen due to their being classified as the same subtype (Type I gamma), though it is unknown if this affects the compatibility of the enzymes. The requirement or preference for Type I orphan S subunits to be ‘paired’, or in reality, complexed, with R and M components of the same subtype has yet to be experimentally tested in detail and may form a basis for future studies. All clones returned the predicted methylation profiles when subjected to SMRT sequencing, with full S subunits facilitating the methylation of bipartite motifs and truncated S subunits (i.e. encompassing a single, complete TRD) apparently dimerising to facilitate methylation of palindromic motifs (Fig. 6, bold inset). In addition, the reduction in EOP conferred by each R-M configuration was correlated with the number of target sites present on the genomes of the five phages tested (with the notable exception of phage c2). This is exemplified by the increased phage resistance conferred by the truncated S.Lcr033I (specifying the methylation of the palindromic motif TTAANNNNNTTAA, relatively abundant on the genomes of these phages) when provided in addition to the Lcr147I system. The observation that a truncated (single TRD-encompassing) S subunit provides a greater level of phage protection to its encoding cell may explain their observed presence on lactococcal genomes and highlights the remarkable plasticity of these antiphage systems.

Finally, complete and dimerised versions of TRD1 were visualised bound to DNA in silico as a representation of the capacity of truncated S subunits to dimerise, imitating the target recognition function of full-length subunits (Fig. 7). The structure predictions of the methyltransferase complexes RM2S(full) and RM2S2(truncated) are of high quality (Fig. 7) and reveal the two flipped adenines primed for methylation (Fig. 7 insets (ii)) located within the groove of the active site (Fig. 7 insets (i)). Furthermore, the two truncated S subunits superimpose well with the full-length protein. The functional (Fig. 6  iv, vi) and in silico structural confirmation that truncated S subunits function as well as full length units highlights the extensive potential for plasticity and flexibility of these antiphage systems as they are encoded in lactococcal species.

Figure 7.

Structure predictions of two methyltransferase complexes with dsDNA. (A) Ribbon view of the complex of R, 2xM and 2xS (truncated) with the target (palindromic) DNA sequence. (i) Surface view of a close up of one of the active sites with the flipped base. (ii) The DNA exhibits two flipped bases on either strand. The recognition sequences are underlined. Base colours are red (A), green (G), yellow (C), and blue (T). (B) Ribbon view of the complex of R, 2xM and S(full) with the target (non-palindromic) DNA sequence. (i) Surface view of a close up of one of the active sites with the flipped adenines. (ii) The DNA exhibits two flipped bases on either strand. The recognition sequences are underlined. Base colours are red (A), green (G), yellow (C), and blue (T).

Conclusions

A complete cataloguing of the antiphage architecture of a plasmid-rich bacterial strain requires complete and accurate genome resolution which, using current technologies, should involve a multifaceted sequencing and assembly approach. This resolution may then be followed by analysis with (preferably) more than one antiphage system retrieval software, a comparison of the results, and (if necessary) a subsequent gene-trait matching effort. In the current study, the complete genomes of 43 lactococcal strains were deduced using this approach and a comprehensive cross-correlation of the mobilome was performed to assess the extent of the ‘mobile defensome’, the first time this has been completed for the genus, or indeed any LAB member, at this scale and analytical breadth.

Antiphage systems are diverse and abundant in lactococci, and well over 50% of these are ‘mobile’, i.e. encoded on plasmids or other MGEs such as ICEs/IMEs or prophages. Though a comprehensive repository of R-M system components is provided by REBASE [45], allowing an accurate delineation of R-M systems encoded by these genomes, the fact that only previously described Abi-like, T-A and other phage defence systems can be detected (and other systems such as Sie only by manual intervention) allows the possibility of as-yet-undiscovered antiphage systems being encoded by this strain collection. Indeed, although we expect that the antiphage systems described herein comprise a broad representation of those found in L. cremoris and L. lactis, it is unlikely that they capture the full spectrum of antiphage system diversity in these species. Future studies may examine the existence of lactococcal ‘defence islands’, which may require gene candidate profiling by bacteriophage challenge. Currently it is not possible to distinguish clear defence islands in this genus, particularly as discrete as the recently described CRISPR-associated defence islands in S. thermophilus [70], with Fig. 3C showing the locations of antiphage genes being widely dispersed across all analysed chromosomes.

In characterising R-M elements of the lactococcal mobile defensome, 135 methylated motifs were identified, and each motif was matched to its responsible enzyme either by homology searches or heterologous expression combined with SMRT sequencing, in the first large scale matching effort for the genus Lactococcus. A remarkable reservoir of both active and silent Type I R-M system specificity units was identified on the deduced plasmidome, elucidating the high level of flexibility and adaptability of these species, undoubtedly in response to constant phage challenge in the dairy production environment. This plasticity is highlighted further by the demonstration of phage resistance conferred by R-M systems being cumulative, and in the case of Type I systems conveyed by systems harbouring a single TRD-encompassing S subunit. While this approach identifies the presence, activity and spatial distribution of these systems, future studies (e.g. transcriptomic or plasmid copy number analysis) may shed light on the level of expression of certain genes or systems, further characterising the role they may play under environmental conditions or phage pressure.

MGEs, and in particular plasmids, although generally thought of as important players in bacterial and viral communities, are now being recognised as crucial elements in the transfer of important traits such as phage defence systems [9], virulence factors and antibiotic resistance cassettes [105, 132]. Their selfish nature [133–135], and the continuous discovery of impressive self-preservation mechanisms [136] will undoubtedly cement this position as research into their biology continues.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

PdW and NvP are employees of dsm-firmenich. RJR, AF and BA are employees of New England Biolabs. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

Office of Cyber Infrastructure and Computational Biology; National Institute of Allergy and Infectious Diseases; National Institute of Health (Grant/Award Number: 2023-AD010714075); University of California, San Francisco. This publication has emanated from research conducted with the financial support of the Science Foundation Ireland under grant numbers (12/RC/2273-P2, 17/SP/4678, 20/FFP-P/8664), which is co-funded by dsm-firmenich, Taste, Texture & Health. For the purpose of open access, we have applied a CC BY public copyright license to any author accepted manuscript version arising from this submission. This work was performed in part using HPC resources from GENCI-IDRIS (Grant/Award Number: 2023-AD010714075). Funding to pay the Open Access publication charges for this article was provided by APC Microbiome Ireland.

Materials and correspondence

For material requests and correspondence, please contact Douwe van Sinderen: d.vansinderen@ucc.ie.

Data availability

Coordinates of predicted protein structures are accessible on Zenodo (https://doi.org/10.5281/zenodo.13987384). Complete genome sequences of all strains are available from the GenBank and RefSeq repositories managed by the National Centre for Biotechnological Information (NCBI; ncbi.nlm.nih.gov).