Short-range interactions dominate pneumococcal Hi-C and Pore-C data

Three independent 3C preparations were generated for each of S. pneumoniae RMV7_domi, RMV7_rare, RMV8_domi and RMV8_rare (S1 and S2 Tables). These were digested with NlaIII, which can cut at sites found in the pneumococcal chromosome at intervals of ~300 bp (S3 Table), and used to generate a 12-plex library for Nanopore sequencing. However, this yielded few reads, as the activity of the pores declined rapidly during the sequencing run (S1 Fig). Residual DNA from these samples was instead processed for Illumina sequencing, and successfully generated Hi-C data (S2 Table). A further three independent 3C preparations were generated for each of the four variants using MluCI, which can cut at sites found in the pneumococcal chromosome at intervals of ~125 bp (S3 Table). Previous applications of the Pore-C approach to human cells found an additional protease treatment with pronase increased the efficacy of de-cross-linking, thereby slowing the rate of nanopore blocking by proteins attached to DNA [38]. Correspondingly, including this step in the sample preparation protocol enabled more efficient data generation from individual flow cells (S1 Fig and S2 Table).

The three biological replicates for both methods for each of the variants were mapped to the appropriate dominant variant reference genome (S2 and S3 Figs and S1 Table). The necessity of mapping short read fragments meant the Pore-C data, which contains a higher rate of sequencing errors, mapped less efficiently than the Hi-C data. Nevertheless, a high proportion of the Nanopore data could be uniquely mapped to locations within the pneumococcal genome (S3 Fig). To establish the highest resolution at which analyses could be reliably conducted, the proportions of inferred relative mapping orientations of sequence fragments were plotted against the separation between their mapped locations if they came from the same read pair, for Illumina Hi-C data, or from the same read, for Pore-C data. This found that sequence fragments separated by less than 5 kb were typically mapped to opposite strands of the genome for Illumina Hi-C data (S4 Fig), or to the same strand of the genome, for Pore-C data (S5 Fig). These are the orientations expected for undigested DNA sequenced with these technologies, consistent with only a subset of restriction sites being cleaved during sample preparation. This is typical of 3C preparations, requiring a convergence distance to be defined as the separation at which sufficient digestion and re-ligation occurred to ensure all sequence fragment mapping orientations become equally frequent [39]. This was estimated to be between 5 and 10 kb for both technologies (S4 and S5 Figs).

Plotting the inferred contact frequency between loci against their separation highlighted particularly strong interactions at an interval of ~5 kb (Fig 1B), which may indicate a particular set of atypical contacts. However, plotting the locations of these high-level short-range interactions found they were distributed across the chromosome, and therefore may represent false positive contact inferences resulting from incompletely-digested DNA (S6 Fig). Therefore data were subsequently analysed at a resolution of 10 kb, to mitigate any effects of DNA fragments that had not undergone digestion and religation.

This locus size was used to analyse the contact frequencies between all loci across the chromosome for each replicate (S7 and S8 Figs), and combined across each set of replicates (Fig 1C). A strong diagonal was evident across all of the symmetrical contact frequency matrices. This suggested most interactions occurred over distances <25 kb, consistent with a simple genome organisation, dominated by short-range structures. There were relatively few signals of higher-level chromosomal organisation. Faint secondary diagonals, intersecting the orthogonal main diagonal around the terminus of replication, were evident in the Pore-C data. Both methods identified an elevated density of off-diagonal interactions around the origin of replication (ori; Fig 1D and S8 Fig). This was positioned ~1.6 Mb from the beginning of the RMV7 reference, resulting in vertical and horizontal stripes across the matrix in the Hi-C data, whereas the Pore-C data showed an enrichment of contacts in the upper right quarter of the matrix (Fig 1C). The RMV8 ori was at the start of the reference genome, resulting instead in a framing effect in the Hi-C data, with elevated intensities in each corner of the matrix.

Focussing on the origin of replication suggested the presence of short-range boundaries to interactions that were consistent across technologies at ~40 kb or ~20 kb downstream of ori in RMV7 and RMV8, respectively (Fig 1D). These coincided with the location of the only conserved parS site in this downstream replichore [40], which is more distant from ori in RMV7 as a consequence of the integration of PRCI_dnaN in the intervening sequence [12]. The parS site is recognised by the ParB chromosome segregation machinery, which is integral to organising the origin domain of the chromosomes of the Bacillota species B. subtilis [41] and S. pneumoniae [40]. Hence this motif is likely to underlie the observed patterns of chromosomal contacts in this region. Although two parS sites are proximal to ori in the upstream replichore, these do not appear to affect contact patterns in the same way, consistent with previous observations of parS sites having distinct effects depending on their location [40,41]. Hence both Hi-C and Pore-C data produced a consistent view of the contacts that dominate the pneumococcal genome’s conformation.

Differential interactions between variants span PRCIs

A genome-wide search for differences in chromosome conformation between the variants was undertaken using contact matrices that summarised contact counts within non-overlapping 10 bp loci. The DiffHiC (S9–S12 Figs) and multiHiCcompare (S13–S16 Figs) methods were used, both of which fit generalised linear models to independent biological replicate datasets to identify differences in contact frequencies [42,43]. However, they differ in their normalisation of the data [44]. Clustering of the biological replicates by the diffHiC pipeline suggested the variants could be resolved from one another by the genome-wide patterns of contacts in each comparison, except in the Hi-C analysis of the RMV7 variants (S17 Fig). However, these separations were not always associated with high bootstrap values, suggesting the differences were attributable to a small subset of loci, rather than broad changes in genome-wide contact frequencies. Correspondingly, specific sites exhibiting significantly different contact frequencies between the genotypes, following Benjamini-Hochberg corrections for multiple testing, could be identified using Manhattan plots (Fig 2A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Genome-wide analysis of differential contact frequencies.

(A) Manhattan plots comparing the contact densities between epigenetic variants. For each genotype, two different statistical methods (diffHiC and multiHiCCompare) were used to compare the three biological replicates generated using each technology (Hi-C and Pore-C) for both epigenetic variants. The negative logarithm of the q values, corresponding to false discovery rates following a Benjamini-Hochberg correction for multiple testing, are shown for each locus across the genome. The threshold for genome-wide significance is shown by the red dashed horizontal line. The vertical black dashed lines show the boundaries of the variable PRCIs (PRCI_dnaN in RMV7, and PRCI_malA in RMV8) within the chromosomes. (B) Sites of differential contact frequencies identified by all four statistical comparisons within PRCI_dnaN in RMV7, and PRCI_malA in RMV8. Each point represents an interaction that was found to be significantly different between the corresponding dominant and rare variants following a correction for multiple testing. Its position on the horizontal axis represents the site of the interacting locus with the lower numerical position. The dashed line connects the point to the interacting locus with the higher numerical position. The vertical position of the point represents the base two logarithmic fold difference in the rare variant relative to the dominant variant, such that values greater than zero represent a higher contact frequency in the rare variant. The shapes of the points represent the method used to generate the data.

https://doi.org/10.1371/journal.ppat.1013392.g002

For RMV8, the only loci that were consistently identified as significantly differing between the two variants by all four comparisons were within, or adjacent to, PRCI_malA [18]. The interacting loci, and the fold differences in contact frequency, were plotted against the annotation of this MGE (Fig 2B). The hits were either between the flanking region and the element, or between the two flanks. However, while the spatial distribution and magnitude of the effects were similar across all comparisons, the Hi-C data suggested higher contact frequencies in RMV8_rare, whereas the Pore-C data suggested higher contact frequencies in RMV8_domi. To rule out sample misidentification as a cause of this discrepancy, it was verified that the patterns of contact frequencies at the tvr locus matched the arrangements characteristic of the corresponding variants, and that a simple analysis of contact frequency ratios within PRCI_malA supported the genome-wide analysis results (S18 Fig). Detailed analysis of the Nanopore sequencing read lengths mapping to this region did not suggest this was an artefact of the Pore-C methodology preferentially sequencing shorter DNA fragments originating from the MGE (S19 Fig). Therefore the Hi-C and Pore-C analyses concurred that there was variation in the contact frequency within PRCI_malA, but disagreed on the variant in which the contact density was highest.

The comparison of the RMV7 variants provided an opportunity to test whether this observation could be independently replicated, as the analyses of Hi-C data were consistent in identifying PRCI_dnaN [12] as the region differing most significantly between RMV7_domi and RMV7_rare (Fig 2A). This MGE was also the location of the most significant difference in contact frequencies between the variants in the diffHiC comparison of Pore-C datasets, in addition to being the site of multiple significant differences in the multiHiCcompare analysis of the same data (Fig 1D). The detailed plot of these points relative to the PRCI_dnaN annotation demonstrated these hits were either within the element, or between the element and its flanking regions (Fig 2B). While the differences were again of similar magnitude and distribution across the two methods, the Hi-C and Pore-C data were once more discordant in identifying the variant in which the contact frequency was highest. In this comparison, the Hi-C data suggested the contact densities were higher in RMV7_domi, whereas the Pore-C data implied they were higher in RMV7_rare. Therefore multiple significant differences in contact frequencies between both pairs of variants were within or flanking PRCIs, although the methods did not reach a consensus on the nature of the altered pattern of interactions.

Changes in PRCI contact frequency do not reflect changes in XS

These discrepancies between the datasets for each variant could represent either a biological difference between the samples, or an artefact of the differing methodologies used to generate the sequence reads. Consequently, a generalised linear mixed-effects model was constructed that could resolve the correlations with different genetic and biological variables. This model was fitted to the four combinations of sequencing methodology and genotype at a resolution of 10 kb using a zero-inflated Poisson distribution with a logarithmic link function (see Methods; S20–S24 Figs). The fitted models were able to explain a substantial fraction of the variation in contact frequencies across loci and datasets (S25 Fig). The effect of distance between loci was estimated to be consistent between the genotypes, but contact frequencies declined more rapidly with distance in the Hi-C data, suggesting the Pore-C data were more efficient at detecting long-range interactions (Fig 3A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Statistical modelling of contact frequencies.

A zero-inflated Poisson generalised linear mixed effects model was fitted to the four combinations of genotype and sequencing technology using contact matrices calculated at a resolution of 10 kb. The two genotypes are distinguished by colours, and the two technologies are distinguished by line and point styles. (A) Coefficients estimating the effect of separation between loci on contact frequencies, normalised to the expected contact frequency between neighbouring loci, shown by the horizontal line at a value of one. (B) Coefficients estimating the effect of MluCI and NlaIII site densities on the detection of contacts; these coefficients reflect the impact of the number of sites per 1 kb. The horizontal line at one represents the coefficient value if the density of sites had no effect on the number of inferred contacts. (C) Coefficients estimating the effect of the fraction of bases (scaled upwards 10-fold to aid model fitting) that are guanine or cytosine (GC) on contact frequencies. (D) Coefficients estimating the contact frequencies within mobile genetic elements. The top panel shows the coefficients quantifying the relative contact frequencies in the MGEs, normalised to the frequencies observed across the rest of the chromosome, corresponding to the horizontal line at a value of one. The bottom panel shows the coefficients estimating the contact frequencies at the same loci in the rare variant relative to the dominant variant. (E) Locus-specific variation in the contact frequencies across the loci encompassing the two PRCIs highlighted in Fig 2: PRCI_dnaN within RMV7, and PRCI_malA within RMV8. Data are plotted at 10 kb intervals. The black vertical dashed lines indicate the boundaries of the PRCIs. (F) Differences between the variants at the loci shown in (E).

https://doi.org/10.1371/journal.ppat.1013392.g003

To test whether any of the differential contact frequency results were artefacts of the restriction enzymes used to generate the samples, the model also estimated the sensitivity of the inferred contact frequencies to the density of sites targeted by these endonucleases (Fig 3B). With the Pore-C data, there was no evidence of any strong relationship between restriction sites and inferred contact density. With Hi-C data, a higher density of NlaIII sites was associated with an increased sensitivity for detecting contacts, which is consistent with the experimental protocol. However, there was an inverse relationship between the inferred contact density and the density of MluCI sites, despite this enzyme not being involved in sample preparation. Given the AT-rich nature of the MluCI sequence motif (AATT), such an observation may be attributable to Illumina sequencing’s bias towards DNA with a neutral GC content, which is not shared with Nanopore data [45]. However, contact frequencies did not exhibit an overall inverse relationship with GC content. Rather, contact frequencies increased with rising GC content in RMV7, but showed the opposite relationship in RMV8 (Fig 3C). Both trends were more extreme for the Hi-C data than the Pore-C data. This suggested the Illumina data may be more affected by localised regions of extremely AT-rich sequence, correlating with the presence of MluCI sites, rather than the mean GC content averaged over a 10 kb locus. Nevertheless, no consistent evidence could be found of a pneumococcal XS system that preferentially condensed AT-rich sequences [46].

A more specific test for XS was quantifying the contact density across the MGEs within each of the genomes, relative to the rest of the chromosome. For RMV7, neither the Hi-C or Pore-C data found any evidence of elevated contact densities across the identified MGEs, with the confidence intervals of the level of contact density relative to the rest of the genome always spanning one (Fig 3D). When comparing the variants, differences in the density of contacts were only significant for PRCI_dnaN. In further agreement with the differential contact analyses (Fig 2B), the contact frequencies were higher for PRCI_dnaN in RMV7_rare for the Pore-C data, but lower in the Hi-C data. For RMV8, the Illumina Hi-C data again found no evidence of the ϕRMV8 or PRCI_malA MGEs being associated with elevated contact densities, although there was some evidence of higher contact frequencies within the latter in the Pore-C data. However, the Pore-C data also indicated that contact frequencies within PRCI_malA were reduced in the rare variant, suggesting there was no consistent evidence for increased contact densities across both variants. By contrast, the Illumina Hi-C data found that PRCI_malA was associated with higher contact densities in RMV8_rare. Therefore the model outputs were consistent with the differential contact analysis in both genotypes, while also suggesting there was no evidence of elevated contact densities within the pneumococcal MGEs that would have been indicative of XS.

To test whether there was evidence of XS at any location within MGEs, locus-specific effects on contact frequencies were estimated for the sites within and flanking PRCI_dnaN and PRCI_malA (Fig 3E). There was no evidence of heterogeneity within the elements, although the size of PRCI_malA meant it contained only a single locus, making the elevated contact density inferred for this element in the Pore-C data highly susceptible to confounding by locus-level effects (Fig 3D). There was also little variation between the variants at these sites (Fig 3F), which is consistent with the differences being captured by the MGE-level variables (Fig 3D). However, variation was evident in the flanking regions (Fig 3F), the direction of which mirrored the alterations in the elements themselves (Fig 3D). Therefore the differential contact densities at the PRCIs was not caused by changing levels of XS-driven condensation, but instead reflected altered interactions between the MGEs and the proximal regions of the host chromosome.

Contact frequency changes are associated with altered PRCI activity

Such an observation suggested the differences between the variants may reflect changes in excision, replication and activation of the PRCIs. Given the conflicting data between the Hi-C and Pore-C methods, previously-collected RNA-seq data were used to identify the variants in which the MGEs were more active. PRCI_dnaN had previously been identified as being transcribed at a significantly higher level in RMV7_domi than RMV7_rare during early exponential phase using both RNA-seq (Fig 4A) and quantitative reverse transcriptase PCR (qRT-PCR) data [12]. Analogously, the PRCI_malA genes were generally more highly expressed in RMV8_rare than RMV8_domi (Fig 4B), although these differences were not consistent enough to reach genome-wide significance [18]. However, no difference was observed when the tvr locus of each of the variants was removed (Fig 4C), indicating the differences might be associated with epigenetic modifications [18].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Differential activity of PRCIs in epigenetic variants.

(A) Quantification of PRCI_dnaN expression in RMV7_domi and RMV7_rare using previously-published RNA-seq data [12]. The annotated protein coding sequences (CDSs) within the PRCIs are shown by the arrows at the bottom of the panel, with their orientation indicating the direction in which they are transcribed. The lines in the main panel show the level of expression estimated for each CDS for each of the three independent biological replicates as normalised transcripts per million reads. The solid line indicates the median value for each variant, and the shaded area indicates the range between the maximum and minimum values. (B) Quantification of PRCI_malA expression in RMV8_domi and RMV8_rare using previously-published RNA-seq data [18], as displayed in panel (A). (C) Quantification of PRCI_malA expression in RMV8_domi tvr::cat and RMV8_rare tvr::cat using previously-published RNA-seq data [18], as displayed in panel (A). These two mutants do not have functional SpnIV RMSs. (D) Quantification of PRCI_malA expression in RMV8_domi and RMV8_rare using qRT-PCR assays of transcription of an integrase (int) and replication (dnaC) CDS. Three technical replicates are shown for each of three biological replicates, as indicated by the shapes of the points. The violin plot summarises their distribution, with the median indicated by the horizontal line. Measurements were made at both early and mid-exponential timepoints. (E) Diagram showing the primers used to quantify the integration, excision and replication of the PRCIs. (F) Differences in integration, excision and replication of the PRCIs between variants. Data are shown as in panel (D). For both PRCI_dnaN, in RMV7, and PRCI_malA, in RMV8, the level of excision was quantified as the ratio of attB sites to attL sites A higher ratio indicates a greater level of excision. Similarly, the levels of circularised PRCIs were quantified by measuring the concentration of attP.

https://doi.org/10.1371/journal.ppat.1013392.g004

To validate whether transcription of PRCI_malA substantially differed between the variants, qRT-PCR was used to assay the expression of int, encoding the element’s integrase, and dnaC, encoding a DNA replication protein, during early exponential (OD₆₀₀ = 0.2) and mid-exponential (OD₆₀₀ = 0.5) phase growth (Fig 4D). Although no substantial difference was evident in early exponential phase, the transcription of the replication gene rose to be ~4-fold higher in RMV8_rare than RMV8_domi during mid-exponential phase, consistent with the higher expression of these genes in the same variant in the RNA-seq data (Fig 4B). This suggested that PRCI_dnaN and PRCI_malA were less frequently transcribed in RMV7_rare and RMV8_domi, respectively. Hence transcription, as well as contact densities, differed between epigenetic variants at these PRCIs.

The readily detectable levels of gene expression, which changed in response to growth, were also consistent with an absence of XS. Furthermore, they were indicative of active excision and replication of these PRCIs, which would explain the differences in contacts within these elements, and between them and their flanking regions [47]. Therefore qPCR assays were used to measure the ratios of excised to integrated elements, and the copy number of circularised elements within the cell (Fig 4E, F). In RMV7, the ratio of excised to integrated PRCI_dnaN was significantly higher in RMV7_rare than RMV7_domi, although there was no substantial difference in the number of circularised elements within the cell. In RMV8, both the ratio of excised to integrated elements, and the frequency of circularised PRCI_malA genomes, were significantly higher in RMV8_domi than RMV8_rare. Therefore the altered contact frequency patterns between variants likely represented changes in the topological configuration of the elements.

For such a topological change to fully explain the significant differences of opposing magnitude in the Hi-C and Pore-C comparisons, it would need to be possible for differences in PRCI circularisation to be metastable [48]. Under such circumstances, a variant may exhibit levels of PRCI circularisation and replication that are stable over enough bacterial generations such that separate biological cultures grown in parallel have consistent properties. Yet the conflicting results between analyses demands that it is also possible that re-isolation of the variant produces a genotype with a different level of PRCI activity that is similarly persistent.

To test this, three independent colonies were cultured for each variant of RMV7 and RMV8. Each of these cultures was frozen, used to inoculate overnight growth, then diluted into two parallel tubes of fresh media until they reached exponential phase growth. Measurements of the level of excision, and extrachromosomal circularisation, were generally consistent across the parallel duplicate cultures inoculated from the same isolate (Fig 5A). However, there was substantial variation between cultures derived from different colonies for RMV8. These results also contrasted with the previous qPCR assays, which were performed on the same isolate used for the Pore-C analysis (Fig 4F). For RMV7, the results were more consistent across the colonies representing each variant (Fig 5A). However, RMV7_domi was generally associated with the higher median copy number across the replicates in this experiment, whereas this value had been higher for RMV7_rare in the previous assay (Fig 4F). Therefore PRCI copy number levels varied stochastically between even isogenic isolates, but are metastable during culturing in liquid media.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Metastability and inducibility of PRCI activity.

(A) Violin plot showing the variation in PRCI topology between cultures inoculated with independently-isolated colonies of the four variants. Two biological replicate cultures were grown from each of three independently-isolated colonies for each variant. Each violin plot represents a single colony, and describes the distribution of three technical replicate measurements for each of the paired biological replicates, with the shape of each point indicating the biological replicate to which they relate. The horizontal line across each violin represents the median value. (B) Violin plot showing the variation in PRCI_dnaN excision during a six-day passage of RMV7 genotypes. Four independent mutants, in which the RMV7_domi or RMV7_rare tvr locus had been introduced into a tvr::Janus RMV7 parental genotype, were passaged in parallel [12]. DNA was sampled from a single biological replicate of each mutant at the start of the passage, and after two, four and six days. Each violin summarises three technical replicate measurements of a single biological replicate. (C) Violin plot showing the effect of inducing stimuli on the topology of PRCI_malA. The excision and replication of PRCI_malA following exposure to competence stimulating peptide and mitomycin C were assayed two hours after exposure in early exponential growth. These experiments used both RMV8_rare, and a mutant derivative lacking the ϕRMV8 prophage. Data are shown as in panel (A). (D) Violin plot quantifying the effect of mitomycin C on PRCI gene expression. The expression of the int and dnaC genes of PRCI_malA were assayed after two hours of exposure to mitomycin C, in both RMV8_rare and a mutant derivative lacking the ϕRMV8 prophage. Data are shown as in Fig 4.

https://doi.org/10.1371/journal.ppat.1013392.g005

PRCI excision and expression are separately regulated

The variation in contact frequency and copy number between isolates of the same variant contrasted with the consistent differences in PRCI expression between RNA-seq and qRT-PCR experiments (Fig 4B, D) [12]. Correspondingly, previous passages of RMV7 genotypes, in which tvr loci with differing arrangements were independently introduced into a common parental genotype, found phenotypic and expression differences were stable over multiple days [12]. To test whether levels of excision were more variable, the topological arrangement of PRCI_dnaN was assayed in the same samples (Fig 5B). This found substantial differences in the integration-excision dynamics of this element across isolates of both variants. As these samples were directly transferred between liquid cultures, this demonstrated the changes in PRCI excision did not depend on culturing on solid media. Hence differences in PRCI expression may be stable, despite stochastic variation in the topology of the element.

This implied the excision and transcription of PRCIs were differentially regulated, consistent with PRCI gene expression in at least some samples being highest in the variants in which these elements were most frequently integrated into the chromosome (Fig 4D, F). This was tested using RMV8, which carries both the ɸRMV8 prophage and PRCI_malA [18]. RMV8_rare cells were exposed to competence stimulating peptide (CSP) and mitomycin C (MMC), both of which activate the ɸRMV8 prophage [20]. These assays were conducted with RMV8_rare and a mutant in which ɸRMV8 had been replaced with a chloramphenicol acetyltransferase (cat) resistance marker, to ascertain whether the PRCI’s activity was affected by the prophage (Fig 5C). Both the level of excision, and the concentration of circularised PRCI elements, rose in response to MMC, but not CSP. This behaviour was independent of the presence of a prophage within the same cell. This suggests the excision of PRCI_malA is controlled by the MGE itself in response to similar stimuli that activate prophage.

A qRT-PCR assay was then used to test whether MMC also activated PRCI_malA int and dnaC gene expression in both RMV8_rare and RMV8_rare ɸRMV8::cat (Fig 5D). An inconsistent rise in the transcription of both genes was recorded in MMC supplemented cultures, correlating with the elevated copy number of the element (Fig 5C). However, independently of treatment with MMC, transcription levels were consistently lower in the absence of the prophage (Fig 5D). This effect was greatest for the dnaC replication gene, the median expression of which was 4.8-fold lower in the absence of ɸRMV8 in unsupplemented media, and 3.4-fold lower in the absence of ɸRMV8 in MMC-supplemented cultures. This is consistent with the PRCI’s dependence on the intact prophage for its ability to transmit between host cells [49]. Therefore expression of the PRCI is regulated by additional factors, associated with the presence of prophage, that do not affect excision.

PRCIs do not cause a substantial fitness cost to their host cells

To test whether this regulation of PRCI activity limited their cost to their host cell, the growth of mutants in which PRCIs had been replaced with a Janus cassette [50] was compared to the corresponding unmodified genotypes in unsupplemented media (Fig 6A). Neither RMV7_rare PRCI_dnaN::Janus, nor RMV8_rare PRCI_malA::Janus, exhibited detectable changes in growth in unsupplemented media relative to the variant from which they were derived. By contrast, removal of the prophage ϕRMV8 substantially increased the growth of RMV8_rare (Fig 6B), consistent with the behaviour of a similar RMV8 prophage null mutant [20]. The deletion of PRCI_malA from this non-lysogenic mutant caused no additional change. Therefore PRCIs cause a substantially smaller impairment of host pneumococcus growth than prophage in the absence of an activating stimulus.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. The impact of induced and uninduced MGEs on the growth of RMV7 and RMV8 genotypes.

Each panel shows the 20 h growth curve of a set of pneumococcal genotypes, measured through assaying the optical density (OD₆₀₀) at 30 min intervals. In each plot, the solid line shows the median, and the shaded region shows the full range of values observed across three biological replicates. (A) A comparison of the effect of removing PRCI_dnaN from RMV7_rare and removing PRCI_malA from RMV8_rare. (B) A comparison of the effect of removing ϕRMV8 from RMV8_rare and RMV8_rare PRCI_malA::Janus. (C) A comparison of the effect of exogenous stimuli on the growth of RMV7_rare and RMV7_rare PRCI_dnaN::Janus. (D) The effect of increasing concentrations of mitomycin C on RMV8_rare and mutant derivatives lacking ϕRMV8, or PRCI_malA, or both.

https://doi.org/10.1371/journal.ppat.1013392.g006

Growth assays were then used to test whether this changed when PRCIs were induced to excise from the chromosome by MMC during early exponential phase. Both RMV7_rare and the corresponding PRCI_dnaN::Janus mutant grew similarly in the presence of MMC, consistent with the MGE not reducing host cell fitness upon excision (Fig 6C). Although MMC did cause a dose-dependent reduction in the density of RMV8_rare cultures, this sensitivity was substantially reduced in a non-lysogenic ϕRMV8::cat mutant (Fig 6D). By contrast, removal of PRCI_malA did not affect MMC sensitivity in either the wild type or non-lysogenic RMV8 cells. Therefore stimulating the excision and circularisation of PRCIs neither adversely affect pneumococcal host cell growth, nor interfered with the lysis of cells following prophage induction.

Culturing of S. pneumoniae

All S. pneumoniae strains (S1 Table) were cultured at 35°C in a 5% CO₂ atmosphere. Culturing on solid media used Todd-Hewitt broth (Sigma-Aldrich) supplemented with 0.5% yeast extract (Sigma-Aldrich), 1.5% agar (Sigma-Aldrich) and 200 U ml⁻¹ catalase (Sigma-Aldrich). To select mutant genotypes, media were additionally supplemented with antibiotics: streptomycin (Sigma-Aldrich) or kanamycin (Sigma-Aldrich) at 400 μg ml^-1, or chloramphenicol (Sigma-Aldrich) at 4 μg ml^-1.

Unless stated otherwise, liquid culturing utilised a ‘mixed’ media consisting of Todd-Hewitt broth supplemented with 0.5% yeast extract and Brain Heart Infusion broth (Sigma-Aldrich) mixed at a 2:3 ratio [12]. For comparing PRCI dynamics between variants, samples were collected during mid-exponential growth, at an optical density at 600 nm (OD₆₀₀) of ~0.6. For measuring the PRCI responses to stimuli, either unsupplemented media, MMC (850 ng), or competence stimulating peptide (CSP; 8.75 µg) was added to 7 ml of bacteria grown to an OD₆₀₀ of ~0.2. The added CSP was appropriate for inducing competence in the cultured genotype: CSP1 for RMV7, and CSP2 for RMV8. The cultures were then incubated at 35 ºC with 5% CO₂ for 2 h. Subsequently, 5 ml of culture was used for RNA extraction, and 2 ml was centrifuged at 3220 g for 10 min to produce a cell pellet for DNA extraction.

To measure growth curves, 2 x 10⁴ cells from titrated frozen glycerol stocks were grown in mixed liquid media in 96-well microtiter plates at 35 ºC with 5% CO₂ for 20 h. For all experiments measuring bacterial sensitivity to inducing stimuli, 200 µl of bacterial cultures with an OD₆₀₀ between 0.15 and 0.25 were mixed with MMC (10 ng), CSP (75 ng), or H₂O₂ (2 ng), then grown in microtiter plates at 35 ºC with 5% CO₂ for 20 h. The OD₆₀₀ was measured at 30 min intervals using a FLUOstar Omega microplate reader (BMG LABTECH). Three or four replicate wells were assayed for each tested genotype and condition in each experiment.

DNA purification and PCR amplification

Following overnight incubation, bacteria were pelleted by centrifugation at 3220 g for 10 min. Supernatants were discarded, and bacterial pellets were then resuspended in 480 μl of lysis buffer (Promega) and 120 μl of 30 mg ml^-1 lysozyme (Sigma-Aldrich). Samples were incubated at 35 ºC for 45 min and centrifuged for 2 min at 8000 g. Bacterial genomic DNA was then extracted using the Wizard Genomic DNA Purification Kit (Promega), following the manufacturer’s instructions. For experiments requiring subsequent PCR product purification, 1000 ng of genomic DNA, 25 μl of 2x DreamTaq Master Mix (Thermo Fisher), 2 μl each of the 10 µM forward and reverse primers listed in S4 Table, and nuclease-free water (Thermo Fisher) were used for PCR amplification in a total reaction volume of 50 µl. Otherwise, 500 ng of genomic DNA, 7.5 μl of 2x DreamTaq Master Mix (Thermo Fisher), 1 μl each of a 10 μM forward and reverse primers (S2 Table), and nuclease-free water were combined in a total reaction volume of 15 μl.

Construction of the S. pneumoniae RMV8 PRCI_malA mutant

To generate the PRCI_malA::Janus construct, the ~1 kb region flanking upstream (UP) and downstream (DOWN) of PRCI_malA were amplified by PCR using primers that inserted restriction enzyme sites on the internal sides of each region. The UP region was generated with primers malA_KO_UP_For and malA_KO_UP_Apal, whereas the DOWN region was generated with primers malA_KO_DOWN_For_BamHI and malA_KO_DOWN_Rev. The Janus cassette was amplified with primers that added corresponding restriction sites (S4 Table). Purified PCR amplicons were then obtained via gel electrophoresis with a 1% agarose gel dyed with SYBR Safe (Thermo Fisher) in Tris/Borate/EDTA buffer, followed by gel extraction with a GenElute Gel Extraction Kit (Sigma-Aldrich).

After extraction, purified PCR amplicons were digested at 37 ºC for 2 h with the appropriate restriction enzyme: ApaI (Promega) for UP, and BamHI (Promega) for DOWN. Digested UP and DOWN products were then mixed with the digested Janus cassette at a 2:2:1 ratio and incubated with T4 DNA ligase (New England Biolabs) at room temperature (RT) overnight. The ligation mix was then used as the template for a touchdown PCR, and the amplified PRCI_malA::Janus construct was purified by separation through electrophoresis on a 1% agarose gel, and extraction using a QIAquick Gel & PCR Cleanup kit (Qiagen). To transform recipient bacteria (RMV8_rare or RMV8_rare ϕRMV8::cat) with the PRCI_malA::Janus construct, a 1 ml sample of a bacterial culture at an OD₆₀₀ of ~0.2 was incubated with 5 µl of 500 mM CaCl₂ (Sigma-Aldrich), 1250 ng of CSP2, and 1 μg of the PRCI_malA::Janus construct at 35°C for 2 h. Samples were then spread onto solid media supplemented with antibiotics (400 μg ml^-1 of either kanamycin or streptomycin) to select for resistant transformants that grew after incubation at 35 ºC with 5% CO₂ for at least 16 h.

RNA purification and reverse transcription

A 5 ml sample of a bacterial culture was mixed with an equal volume of RNAprotect (Qiagen) and incubated at room temperature for 5 min. RNA samples were extracted from this mixture using the SV Total Isolation System (Promega) according to the manufacturer’s instructions, with the exception that the initial extraction of RNA required treating samples with 480 μL 30 mg mL⁻¹ lysozyme (Promega) for 30 min at 35 °C. The purified RNA was treated with amplification-grade DNase I (Invitrogen), then used in reverse transcription reactions using the First-Strand III cDNA synthesis kit (Invitrogen). Each reaction used 0.2 μg RNA, 100 units of the SuperScript III reverse transcriptase, 1 μL of 100 μM random hexamer primers (Thermo Fisher Scientific) and 1 μL of 10 mM dNTP mix (Bioline). After a 5 min annealing period at 25 °C, reverse transcription was conducted at 50 °C for 30 min, then 55°C for 30 min. The reverse transcriptase was then heat inactivated at 70 °C for 15 min.

Quantitative PCR

The oligonucleotides used for qPCR were designed to generate amplicons that were 150–200 bp in length, and are listed in S4 Table. Template cDNA was diluted in a 1:25 ratio with nuclease-free water (Qiagen). Each 15 μL reaction used 3.75 μL of DNA template (either genomic DNA or cDNA), 0.75 μL of 10 μM forward and reverse primer solutions (Invitrogen), 7.5 μL of PowerUp SYBR Green Master Mix (Thermo Fisher Scientific) and 2.25 μL of DNase-free and RNase-free water. Reactions were run using MicroAmp Optical 96-well reaction plates (Thermo Fisher Scientific) and the QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). The mixtures were initially heated to 50 °C for 2 min. Subsequently, 40 amplification cycles were run, each consisting of denaturation at 95 °C for 15 s, followed by annealing and elongation at 60 °C for 1 min. The purity of the amplicon at the end of the reaction was assessed through a melt curve analysis. Three technical replicate measurements were made for each of three biological replicates, unless otherwise specified.

The expression of PRCI_malA genes was quantified using the ∆∆Ct method. The rpoA gene was used as the reference gene, against which target gene expression levels were normalised [12]. The ∆Ct values were calculated as the difference between the mean Ct of the target gene and rpoA across technical replicates within the same biological replicate. The ∆∆Ct values were then calculated relative to the reference ∆Ct, which was that of RMV8_rare at an OD₆₀₀ of 0.2. The fold gene expression difference between genotypes was then quantified as 2^-∆∆Ct.

For the quantification of PRCI dynamics, the amplicons produced across the integration site, and the rpoA reference gene, were generated from genomic DNA (S2 Table). The concentration of each purified PCR product was measured using a Qubit Broad Range Kit and Qubit 4 Fluorometer (Thermo Fisher). The Ct values for copy numbers of 3x10³, 3x10⁴, 3x10⁵ and 3x10⁶ of these amplicons were used to generate a standard curve to convert the Ct values from experiments into absolute copy numbers of DNA molecules.

Construction of Hi-C and Pore-C sequencing libraries

Three independent cultures of each of the four variants (S. pneumoniae RMV7_domi, RMV7_rare, RMV8_domi and RMV8_rare) were grown to mid-exponential phase, corresponding to an OD₆₀₀ of ~0.60. The samples processed for Illumina Hi-C and Pore-C analyses were cultured independently. To crosslink DNA and proteins, cells were mixed with 1% v/v formaldehyde (Sigma-Aldrich) for 30 min at RT, followed by a 30 min incubation at 4 ºC. The formaldehyde was then quenched through the addition of 1 ml 2.5 M glycine (Sigma-Aldrich), which was incubated at RT for 5 min, then at 4 ºC for 15 min. Fixed cells were then collected by centrifugation at 3220 g for 10 min at 4 ºC, flash frozen on dry ice, and stored at -80 ºC, to permit consistent processing of all samples in parallel.

Frozen pellets were thawed on ice, washed with 5 ml water, and centrifuged at 3220 g for 10 min. Pellets were then resuspended in 50 μl 30 mg ml^-1 lysozyme, followed by incubation at 37 ºC for 30 min. Both 10% v/v sodium dodecyl sulphate (SDS; Fisher Bioreagents) and 10% v/v Triton-X (Sigma-Aldrich) were added, followed by 10 min RT shaking incubation, then 10 min incubation at 55 ºC. A 400 µl sample of the lysed cells was then combined with 130 µl of digestion mix, comprising NEB CutSmart buffer (NEB), 10% v/v Triton-X, and 1 U μl^-1 of either NlaIII (NEB; Illumina Hi-C library preparation), or MluCI (NEB; Pore-C library preparation). The chromatin was digested at 37 ºC for 3 h. The NlaIII enzyme was then heat inactivated through a 65 ºC incubation for 20 min. The less stable MluCI enzyme did not require inactivation.

Digested chromatin was then ligated through the addition of 460 μL ligation mix, comprising ligation buffer (NEB), 0.1 µg μl^-1 recombinant albumin (Promega), and 20 U μl^-1 T4 DNA ligase (NEB). This mixture was incubated at RT for 1 h, followed by a 4 ºC incubation for 48 h. After ligation, the DNA-protein crosslinks were reversed through incubation of the mixture with 500 μl 20% Tween-20 (Sigma-Aldrich), 100 µl 10% v/v SDS, and 100 µl 20 μg μl^-1 proteinase K (Sigma-Aldrich) at 65 ºC overnight. For the Pore-C library preparation, another round of protein digestion was performed by adding 118 μl 20 mg ml^-1 pronase (Sigma-Aldrich) and 235 μl 10% v/v SDS, followed by a 63 ºC incubation for 1 h.

Following de-cross-linking, DNA was purified with 2.5 ml of a chilled 25:24:1 ratio mixture of phenol:chloroform:isoamyl alcohol (Sigma-Aldrich), followed by separation by centrifugation at 3220 g for 30 min at 4 ºC. DNA was then precipitated from the aqueous phase of each sample using 20 μl 5 M NaCl (Sigma Aldrich), 200 μl 3 M sodium acetate (EMD Millipore Corporation) and 6 ml isopropanol (Sigma-Aldrich). These mixtures were incubated at -80 ºC overnight, then centrifuged at 3220 g for 30 min at 4 ºC. The resulting pellets were washed with 1 ml of chilled 70% ethanol (VWR Chemicals BDH), followed by a second centrifugation (3220 g, 15 min, 4 ºC). All remaining ethanol was removed through pipetting and drying at 65 ºC in a heat block for 15 min. The DNA was the resuspended in 100 µl nuclease-free water.

To enrich for longer molecules likely to contain multiple ligated DNA fragments, a 100 μl sample of each preparation was mixed with 50 μl of KAPA Pure Beads (Roche) through vortexing, followed by incubation together at RT for 10 min. The beads were then magnetically captured, and the supernatant discarded. The beads were washed twice using 200 μl 80% ethanol and dried at RT for 4 min. DNA was then eluted from the beads through a 10 min incubation in 100 µl of pre-heated nuclease-free water at 37 ºC. The concentration and size distribution of the DNA molecules was checked using an Agilent 2200 Tape Station (Agilent Technologies) and NanoDrop 1000 (Thermo Fisher).

DNA sequencing

Illumina Hi-C data were generated at the Wellcome Sanger Institute. Libraries were prepared using standard approaches [68], enriching for a fragment size of approximately 450 bp. Libraries corresponding to the three biological replicates for each of the four variants were then combined into a 12-plex library, which was sequenced on a single lane of the Illumina Novaseq 6000 platform. This generated 152 nt paired end reads.

Pore-C data were generated at Imperial College London using the native 12-plex barcoding library preparation kit (Oxford Nanopore Technologies, ONT; product SQK-NBD114.24) to process ~400 ng of DNA from each of the three biological replicates for each of the four variants. Each DNA sample was individually treated to ensure fragments were blunt ended with the NEBNext FFPE DNA Repair Mix (NEB) and the NEBNext Ultra II End Repair/dA-tailing Module (NEB). Samples then underwent native barcode ligation with NEB Blunt/TA Ligase Master Mix, before being pooled together. The NEBNext Quick Ligation Module (NEB) was then used to attach sequencing adapters. The completed libraries were then purified from the mixture using the short fragment buffer to obtain DNA fragments of all sizes. The concentration of the final DNA library was measured on a Qubit 4 fluorometer (Thermo Fisher). Two flow cells (ONT product FLO-MIN114) were each used to sequence 15 fmol of the final library using the MinION platform, recording data with a minimum quality score of 9 using high-accuracy basecalling with Guppy version 6.4.6 (ONT).

Bioinformatic processing

The Illumina Hi-C data were processed using version 2.1.0 of the hic workflow [69] implemented using Nextflow version 24.02.0 [70]. This workflow filtered the Illumina reads using FastQC [71], then mapped reads to the appropriate reference genome (S1 Table) using bowtie2 [72], and generated raw contact maps using iced, as implemented within the HiC-Pro pipeline [73]. Genome-wide contact maps between loci were then generated at resolutions of 100, 500, 1000, 5000, 10000 and 50000 bp using cooler [74]. A quality control report was then generated with MultiQC [75], with additional visualisation files produced with juicer [76].

The Pore-C data were processed using version 1.1.0 of the wf-pore-c workflow [77] implemented using Nextflow version 23.10.1 [70]. This workflow mapped reads to the appropriate reference genome (S1 Table) using minimap2 [78] with the non-default settings “-x map-ont -k 7 -w 5”. These alignments were then processed with Pore-c-py [77] and pairtools [39] to generate raw contact maps, which were processed with cooler [74] and juicer [76], as for the Hi-C data.

Differential analysis of contact matrices

Analyses of differential contacts used matrices calculated at a resolution of 10 kb, with all self interactions along the matrix diagonals removed, and the ivr and tvr loci excluded, as they contained known rearrangements.

The diffHiC package [42] was applied using the recommended workflow [79], with modifications to parameter settings as appropriate for these datasets. Loci with fewer than five contacts across all replicates in each comparison were first removed, as there was low power to identify differential contacts at such positions. The counts across the filtered dataset were then normalised across replicates within each comparison between variants using a cyclic LOESS process, run for ten iterations. The biological coefficient of variation was then calculated across different contact densities, which informed the dispersion of the negative binomial generalised linear model that was fitted to the data. This used a tagwise, rather than trended, dispersion calculation, as there was not a strong relationship between variance and contact densities in some datasets. Differential contact densities between loci in the two variants were then identified from this model using quasi F-tests. The resulting p values were subject to a Benjamini-Hochberg correction for multiple testing to generate q values.

Similarly, analysis with the multiHiCcompare package [43] used a modified version of the recommended workflow [80]. This started by first both removing any loci for which the minimum frequency across replicates was zero. The remaining data were normalised using a cyclic LOESS process run for ten iterations. Significant differences were inferred through using a negative binomial exact test, as implemented in edgeR [81], with a Benjamini-Hochberg correction for multiple testing.

Statistical modelling of contact densities

The 10 kb resolution contact matrices were analysed using a generalised linear mixed effects model with the structure:

The terms all represent fixed effects, except those with the ‘|’ notation, which denotes a grouping variable for a random effect. Asterisks indicate an interaction between fixed effects. A zero-inflated Poisson distribution, with a logarithmic link function, was used to fit the model with the glmmTMB package [82]. Parameter estimates reproducibly converged using the default settings. However, attempts analyse the data using an equivalent model with a negative binomial link function, which would account for any overdispersion in the data, failed to converge on robust parameter estimates. Analysis of parameter estimates used the sjPlot package [83], model fits were assessed with the performance package [84], and data processing used the tidyverse packages [85].

Each interaction was categorised by the distance between the loci: < 12.5 kb (i.e., neighbouring 10 kb loci); 12.5-25 kb; 25–50 kb, and >50 kb. Only loci within 100 kb of each other were considered, to ensure parameter estimation was feasible, and avoid the model fitting primarily to the noisy, sparse interactions between distant loci.

Loci were also classified based on the annotation of the corresponding reference genomes. In both genomes, the four rRNA operons were annotated, as the difficulty of mapping to these relatively GC-rich repeats reduces the detection of contacts involving these loci. In RMV7, four MGEs were annotated: PRCI_dnaN, PRCI_uvrA, ICE_rpsI, and Tn916 [12]. In RMV8, three MGEs were annotated: the prophage remnant CIPhR, prophage ɸRMV8 and PRCI_malA [18]. These classifications were fitted separately for each variant, to enable intervariant differences to be inferred at these loci.

The additional fixed effects analysed the properties of each DNA locus, which were calculated using biopython [86]. The effect of the distribution of NlaIII and MluCI sites was accounted for through calculating the mean number of restriction sites across the two interacting loci, and dividing this value by ten. The GC content was calculated as the mean proportion of bases in the two interacting loci, which was scaled upwards by a factor of ten. Similarly, the distance to the origin was calculated as the proportion of the genome between the modelled site and the origin of replication. This was included because the DNA-protein crosslinks were fixed in mid-exponential phase, and therefore the detected contact frequencies were higher near the origin of replication, as some cells would have initiated, but not completed, chromosomal replication. These values were scaled upwards by a factor of ten when included in the model. These adjustments were necessary for numerical stability of the model fitting process.

The identity of the biological replicate was included as a random effect, to correct for differences in the amount of sequence data generated for each dataset. A second random effect estimated the level of contact frequency for each locus in the dominant variant, and the difference in contact frequencies at each of these loci in the rare variant.