Reconstitution of the BPP-1 DGR in E. coli

We set out to reconstitute the archetypal BPP-1 DGR in E. coli, leveraging the extensive genetic tools available in this bacterium to better understand the mechanism of DGR. We encoded the essential BPP-1 DGR components on two plasmids (Fig 2A): pDGR0 carried the avd-TR-brt operon under an arabinose inducible promoter, while pTarget contained a target gene immediately followed by the essential regulatory elements: GC14, IMH, and stem-loop.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Reconstitution of BPP-1 DGR-mediated editing in E. coli.

(A) Schematic of the DGR-mediated kanamycin repair reporter assay [63]. Left, Different pDGR variants, which encode all essential DGR components except for the target. The TR consists of the last thirty-six nucleotides of the kanR gene, followed by tandem stop codons (TAATAA). Right, pTarget encodes the target gene, which is a defective kanR gene terminated by TGATAA. DGR-mediated repair of the reporter was monitored by quantifying the frequency of kanamycin-resistant clones (repair frequency). (B) Bar graph showing the repair frequency for pDGR0, pDGR1, and pDGR2 in E. coli. Each of these plasmids was used in tandem with pTarget to transform E. coli strain MG1655. Freshly transformed colonies were used to inoculate LB + Tet10 + Spec50. The cultures were grown for 1.5 hours at 37°C, after which arabinose was added to induce expression. The cultures were then grown for an additional 4.5 hours. Finally, the cells were then plated on rich media agar supplemented with Kan25 to select for resistant clones. See Methods for details. (C) Bar graph comparing the repair frequency of pDGR1 and variants with the indicated mutations and deletions. These include: bRT(SMAA), YMDD>SMAA active site mutations in brt; Δavd, three nonsense mutations in avd; ΔTR, TR deletion; ΔSL, stem-loop deletion. Data are presented as mean ± standard deviation (n = 3 biological replicates). (D) Weblogo showing the recovery of the missing 18-nt in pTarget upon induction of DGR expression. Plasmids were harvested from a pool of kanamycin resistant clones containing both pTarget and pDGR1. The plasmids were then used as a template, and the target region was amplified by PCR for deep-sequencing. Only sequencing reads showing an 18-nucleotide insertion in the reporter were used to generate the Weblogo. (E) Scheme for specifically amplifying the cDNA-TR-RNA for deep-sequencing while avoiding DNA template amplification. (F) Agarose gel showing the RT-PCR product is dependent on an intact bRT. (G) Weblogo showing the results of cDNA sequencing. The reverse complement of the cDNA sequence is shown to facilitate easy comparison to the reporter in panel D. This sequence is shorter because the primer binding region is excluded.

https://doi.org/10.1371/journal.pgen.1012038.g002

To assess DGR activity, we used a reporter system previously used to characterize the BPP-1 DGR in its native host [10]. In this assay, the target gene was a defective kanamycin resistance cassette lacking the sequence for its last six amino acids. The TR was programmed with the sequence encoding the final twelve amino acids of the full-length cassette (Fig 2A). This design should enable DGR to transfer the missing sequence to the target through a cDNA intermediate, thereby “repairing” the defective kanamycin resistance gene. We then measured DGR activity by quantifying the frequency of kanamycin-resistant clones, which we termed the repair frequency.

Our initial attempts to reconstitute DGR activity in E. coli MG1655 using pTarget and pDGR0 yielded kanamycin-resistant clones at a meager ~10^-8 frequency (Fig 2A–2B). Absence of DGR function likely stemmed from the lack of a clear ribosomal binding site (RBS) upstream of the brt gene in the native BPP DGR operon. Introducing a consensus E. coli RBS before brt in the new construct (pDGR1) increased kanamycin-resistant colonies 100-fold (Fig 2B). Reasoning that TR-RNA level might limit DGR activity, we created pDGR2 (Fig 2A). In this construct, we removed the TR-RNA from the operon and placed it under the control of P_J23119, a strong constitutive synthetic promoter often used for expressing single-guide RNAs in CRISPR applications [14]. This modification improved repair frequency by 35-fold over pDGR1 (Fig 2B), a result likely mediated by the elevated production of both TR-RNA and bRT in pDGR2 (S1 Fig).

This reconstituted DGR system recapitulates key genetic features observed in the native host. The gain of kanamycin resistance did not occur through homologous recombination (S2 Fig), but requires intact essential DGR components: inactivating bRT, Avd, or TR reduced repair frequency to ~10^-9 (Fig 2C). For reasons that remain unclear, repair frequency was reproducibly lowest for DGR lacking avd. Additionally, deleting the inverted repeat downstream of IMH severely—but not completely—impaired repair. In Bordetella, residual DGR activity was also observed in the absence of this inverted repeat. These results confirmed that this stem-loop motif is not strictly required for mutagenic retrohoming but is clearly important for optimal efficiency (Fig 2C).

To directly confirm that resistance was acquired through DGR-dependent repair of reporter, we deep-sequenced the reporter isolated from resistant cells. Roughly half of the VRs have recovered the missing 18 nucleotides (S3A Fig). This heterogeneity reflects the co-existence of repaired and unrepaired pTarget within the same cells, as Sanger sequencing of pTarget from individual clones often showed sequence heterogeneity at the site of insertion (S3B Fig). In contrast, only 3–4% of pTarget isolated from background kanamycin-resistant clones (from our catalytically inactive bRT mutant control, arising at a frequency of <10^-8) acquired the missing sequence (S3A Fig). Thus, TR-VR recombination is rare, and the majority of background resistance likely emerged through suppressor mutations elsewhere in the genome. Collectively, these results confirm DGR-dependent restoration of the reporter.

The sequencing analysis also identified the characteristic A-to-N mutations within the restored reporter (Fig 2D). Mutations were limited to the four adenines in the stop codons, even though cDNA sequencing showed that bRT produces significant mutations throughout the cDNA (Fig 2E–2G). Because programmed synonymous mutations at non-adenine positions were incorporated at 100% efficiency without affecting the repair frequency (S4A-B Fig), we conclude that our selection excluded missense mutations in coding adenines due to reporter inactivation. Consequently, the observed repair frequency underestimated the true frequency of DGR-mediated target editing.

Together, these results demonstrate that DGR’s characteristic mutagenic retrohoming activity can be faithfully reconstituted in E. coli with its core components alone.

Sequence context biases nucleotide incorporation for A-to-N mutations

Our cDNA sequencing analysis confirms that misincorporation frequency and identity vary significantly depending on the adenine’s position within the RNA template (Fig 2G) [7]. While this suggests that the local sequence context directly influences error-prone reverse transcription, this relationship has remained systematically uncharacterized. To resolve this, we constructed a pDGR2 variant featuring a synthetic template encoding all 16 possible (NAN) trinucleotides within a single TR-RNA. To decouple contextual influence from a potential broader positional effect, we generated two more variants where the template was partitioned and rearranged, allowing us to evaluate how an adenine’s placement at the 5’, middle, or 3’ region of the RNA affects its mutagenic profile (Fig 3A). Following expression in E. coli, we deep-sequenced the resulting cDNAs to establish a profile of bRT’s context-dependent behavior.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Effect of sequence context on the fidelity of BPP-1 bRT.

(A) Left, Scheme for cellular profiling the mutational signature of bRT across various template sequence contexts. Right, pDGR2 was programmed with TR1, TR2, and TR3, each containing all 16 possible NAN triplets (underlined in orange) in different orders. Weblogos display the error profiles of cDNA synthesized by bRT from TR1, TR2 and TR3 RNA templates. Each WebLogo was generated from 10 million subsampled reads pooled from triplicate sequencing runs. (B) Heatmap shows the percentage of A, C, G and T incorporation opposite template adenines within all 16 NAN contexts, using TR1, TR2 and TR3 as RNA templates. The Y-axis shows the RNA sequence contexts. (C) Heatmap shows the percentage of base incorporations opposite template adenines in the NAU, NAG, NAC and NAA contexts, highlighting the effect of the nucleotide immediately 3’ (-1 position) of the adenine. (D) Heatmap shows the percentage of base incorporations at template adenines in the UAN, GAN, CAN and AAN contexts, highlighting the influence of the nucleotide immediately 5’ (+1 position) of the adenine. (E) Heatmap shows the percentage of base incorporations opposite template adenines in the indicated contexts, highlighting the combined influence of nucleotides at both the -1 and -2 positions relative to adenine. (F) For each expected AAU, AAC, AAA, and AAG triplet within the TR1, TR2, and TR3 templates, the corresponding observed triplets were identified in Illumina sequencing reads. The proportion of amino acids encoded by these observed triplets was calculated for each anticipated triplet and visualized as a heatmap. Heat maps were generated from 10 million subsampled reads pooled from all three TRs.

https://doi.org/10.1371/journal.pgen.1012038.g003

This analysis revealed notable context-dependent effects on the mutagenic tendency of bRT that were reproducible across the three RNA templates (Fig 3A–3B), indicating that the sequences immediately adjacent to adenines are the primary determinants of the observed variations. However, our results diverged significantly from a prior in vitro finding. We measured an average misincorporation rate of 30.65% opposite all template adenines, which is significantly lower than the previously reported 51.2% rate (Fig 3B) [7]. Moreover, the error distributions also differ. In our study, cytosine (C) misincorporation predominated (59.3%), followed by guanosine (G, 26.4%) and adenine (A, 14.3%). It should be noted that the guanosine misincorporation rate is artificially inflated by a specific context with an unusual guanosine bias (see below); in most contexts, guanosine is disfavored. This contrasts with previously reported in vitro error profiles, which showed a significant preference for adenine misincorporation (A, 43% > C, 34.5% > G, 22.5%) [7].

While the non-uniform cellular dNTP pools—where dCTP is most abundant [15]—may partially account for bRT’s preference for cytosine misincorporations in our dataset, this factor is likely secondary to the intrinsic sequence context of the template. The natural BPP-1 template used in the in vitro study is heavily biased, with 20 of its mutable 23 adenines residing within an AAC context [1]. As demonstrated below, this unique error profile imposed by this specific motif can largely explain the discrepancies between in vivo and in vitro studies.

Our analysis clearly showed that bases immediately 3’ of template adenines (position -1) influenced the mutagenesis profile more profoundly than those immediately 5’ (position +1) (Fig 3C–3D). The most prominent example of this context dependence was seen with adenines adjacent a cytosine at the -1 position (Fig 3C). In this 5’-NAC-3’ context (where N is any base at position +1), guanosine (G) misincorporation was markedly elevated at 26.8%, contrasting sharply with the significantly lower rates observed in NAA (2.4%), NAG (0.08%) and NAT (2.5%) contexts (Fig 3C). This strong, context-dependent bias explains a previous observation: the template adenines found to induce unusually elevated G misincorporation in an earlier study [7] were, in fact, those residing in the same NAC context reported here. This preference for G was further magnified when a template adenine is flanked by two 3’ cytosines (ACC), where G misincorporation reached 44.3% (Fig 3E).

Interestingly, the influence of the 3’ cytosine in the AC context extends to the 5’ adenine one base upstream. In this AAC context, adenine misincorporation (23.8%), displacing cytosine (14.2%) as the most common error (Fig 3E). This effect was absent in closely related AAU, AAA, AAG motifs. Coupled with the heightened G misincorporation in the NAC context, the AAC motif exhibits a mutagenic profile distinct from any other adenine contexts. This finding is particularly noteworthy because the AAC motif is overly represented in native DGR templates, almost exclusively encoding for asparagine in-frame within the target gene [16].

Despite an overall 68% mutational rate, our analysis revealed that DGR-mediated mutagenesis of the AAC codon permits sampling of a highly restricted amino acid space. Only eight out of the 14 different accessible amino acids reached a frequency of at least 1% (Fig 3F), with the mutations dominated by tyrosine (14%), serine (12%) and aspartic acid (10%). Although the AAU codon theoretically explores the same amino acid space as AAC, the fact that cytosine (rather than adenine) is the most common misincorporation opposite the first template adenine biases mutations toward glycine and aspartic acid instead of the tyrosine favored by AAC. This divergence illustrates how context-dependent effects constrain the accessible evolutionary landscape.

Moreover, we observed that the ~ 1% conversion of AAA and AAG to stop codons (Fig 3F) falls significantly below the 6.36% and 7.08% respectively expected if the 30.65% misincorporation rate is equally distributed among G, C and A (see Methods). Thus, the bRT appears to have evolved an intrinsic misincorporation bias to not only favor certain amino acids but also minimize unfavorable premature truncation.

Genetic screen identifies ExoI as an inhibitor of DGR activity

Having reconstituted DGR and characterized its mutagenic properties in E. coli, we next tried to improve upon its editing efficiency (~10⁻⁵), which remains too low for practical applications. We hypothesized that certain cellular factors antagonize DGR activity. To identify these potential inhibitors, we combined the kanamycin repair assay and transposon-sequencing (Tn-seq) to screen for genes whose inactivation enhance DGR activity [17].

To streamline the screen, we integrated the reporter cassette directly into the E. coli genome. Among the four tested chromosomal loci (Fig 4A), we chose the 317° locus (strain HCL26) for our screen, as it exhibited the highest repair frequency. We then mutagenized HCL26 using a mariner-based transposon harboring the chloramphenicol resistance gene for selecting cells with transposon insertions. We pooled ~150,000 chloramphenicol-resistant clones (Fig 4B) and transformed this mutant library with pDGR2 to initiate reporter repair. The library was then plated on kanamycin to select for cells where the reporter had been repaired by DGR. We deep-sequenced the transposon mutant library before (Input) and after kanamycin selection (Output) to identify enriched transposon mutants.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. ExoI constrains DGR efficiency through a nuclease-independent mechanism.

(A) Left, Chromosomal locations of reporter insertion. Right, Bar graph comparing the repair efficiency of reporters at the indicated chromosomal locations in MG1655. E. coli strains used in this experiment were: 63°(HCL34), 156°(HCL19), 291°(HCL24), and 317°(HCL26). (B) Schematic of Tn-seq workflow to identify factors affecting DGR-mediated reporter repair. See Methods for details. (C) Normalized Illumina read counts from each transposon insertion of the Input and Output libraries. Bump out shows changes in transposon distribution within the reporter cassette in the Input and Output libraries. (D) Volcano plot of transposon sequencing results. Vertical dotted lines represent thresholds for a four-fold increase or decrease in read counts. Data above the horizontal dotted line have Mann-Whitney p-value of 0.01 or less. (E) Normalized Illumina read counts from each transposon insertion surrounding the sbcB gene in the Input and Output libraries. (F) Bar graph showing the repair frequency of a chromosomal reporter in E. coli with the MG1655 (HCL26) vs ∆sbcB (HCL84) with and without a functional bRT. Error bars represent standard deviation of at least biological triplicates. bRT(SMAA) denotes catalytically inactive bRT expressed from a derivative of the pDGR2 plasmid. (G) Experiments were performed as in (F) except that reporter repair was read-out by sequencing reporters in the cell population without kanamycin selection. Shown are the frequency (in reads per million) of sequencing reads containing the expected 18 nucleotide insertion. n = 2 biological replicates. (H) Bar graph comparing the repair frequency of a chromosomally integrated reporter in MG1655 (WT) or ∆sbcB (HCL84) strain. These strains also contained either an empty vector pNEB309 (–), pNEB310 (expressing sbcB) or pNEB311 (expressing the nuclease-deficient sbcB^Mut). (I) Cell-lysate-based ssDNA nuclease assay. A 5’ Alexa488 fluorescently labelled ssDNA (see S3 Table for sequence) was incubated for 10 minutes at room temperature with lysates prepared from strains used in (H) or commercially available purified ExoI (NEB MS293). These samples and the unreacted ssDNA substrate (–) were resolved on a 15% polyacrylamide gel before fluorescence imaging. (J) Bar graph showing the relative abundance of TR-RNA-cDNA hybrid in MG1655 vs ∆sbcB (HCL84) strain as measured by qRT-PCR. Error bars: standard error of the mean (SEM). n = 3 biological replicates. p-value was calculated by unpaired, two-tailed t-test.

https://doi.org/10.1371/journal.pgen.1012038.g004

The result indicates that the screen identified genes altering the cell’s sensitivity to kanamycin as anticipated (Fig 4C–4D, S1 Table). Transposon insertions within the reporter cassette (Fig 4C) and genes coding for ATP synthase subunits (Fig 4D, S5 Fig) were severely depleted after selection. Inactivation of ATP synthase enhances the proton motive force (PMF), which consequently enhances kanamycin uptake [18–20]. This increases the cell’s antibiotic burden, causing impaired growth. This indicates that other hits in this screen, including many with metabolic function, may affect kanamycin sensitivity.

Despite this confounding factor, sbcB emerged as a promising inhibitor of DGR activity. In our screen, transposon insertions in the sbcB gene were enriched 10.5-fold in the Output relative to the Input (Fig 4D–4E). We validated this result by deleting the sbcB gene from the HCL26 reporter strain, which triggered a 19-fold increase in repair efficiency (Fig 4F). Importantly, this enhancement is strictly dependent on an intact bRT (Fig 4F). While sbcB mutations are known to activate a recombination pathway involving RecF under certain genetic backgrounds [21], the effect observed here is entirely independent of RecF (S6 Fig).

To rigorously exclude the possibility that sbcB deletion merely confers enhanced kanamycin tolerance, we used an orthogonal, antibiotic-independent assay to quantify DGR-mediated target editing. We deep-sequenced the reporter cassette isolated from the whole culture following induction, bypassing kanamycin selection entirely. Reporters isolated from the ΔsbcB background were found to be 10-fold more likely to have acquired the 18-nucleotide insert provided by the TR template compared to those from the wild-type strain (Fig 4G). Notably, both deep-sequencing and kanamycin repair assays yielded editing frequencies of the same order of magnitude. This high degree of quantitative consistency confirms that both methods accurately report the same underlying DGR activity, establishing that the observed enhancement reflects a genuine increase in mutagenic retrohoming.

The sbcB gene codes for Exonuclease I (ExoI), a processive 3’ to 5’ single-stranded DNA nuclease. Given that ExoI elimination enhances recombineering efficiency, presumably by stabilizing the ssDNA oligos [22], we reasoned that ExoI might similarly inhibit DGR by degrading the cDNA intermediate. To test this, we generated a nuclease-defective variant ExoI^Mut, by substituting two conserved catalytic residues (D15 and E17) with alanines. Surprisingly, expression of ExoI^Mut in trans suppressed the elevated repair frequency in the ΔsbcB background as effectively as the wild-type nuclease (Fig 4H).

This unexpected result implies that the nuclease activity of ExoI is not required for inhibiting DGR. To confirm that ExoI^Mut was indeed nuclease-defective, we performed a cell-lysate-based nuclease assay [23]. As anticipated, lysates derived from ΔsbcB cells expressing wild-type ExoI rapidly degraded a labeled ssDNA substrate, whereas those expressing ExoI^Mut did not show this activity, confirming its nuclease deficiency (Fig 4I).

Direct quantification of the cDNA-TR-RNA—the presumptive substrate for ExoI—further contradicted a degradative role. Rather than being stabilized in the absence of the nuclease, the hybrid level was reduced by 6.6(± 3.0)-fold in the ΔsbcB background (Fig 4J). This paradoxical finding suggests that ExoI does not act by degrading the hybrid. Instead, the accumulation of hybrid in its presence suggests that it hinders downstream processing. Consequently, the rise in DGR efficiency upon sbcB deletion likely stems from the relief of this processing bottleneck, allowing the cDNA to be more efficiently channeled towards integration.

Additional evidence suggests how ExoI may hinder processing. We observed in the nuclease assay that overexpressing wild-type ExoI in trans triggered a massive increase in nuclease activity (Fig 4I, compare Lanes 2 and 4), yet this did not further reduce the repair frequency below that of the wild-type strain (Fig 4H). This saturation of the inhibitory effect, combined with the lack of nuclease dependence, suggests ExoI-mediated inhibition is not a result of cDNA degradation. Instead, it likely involves a stoichiometric interaction with a yet-to-be-identified partner, which caps the rate of DGR-mediated diversification.

Chromosomal location and orientation of target genes affect DGR editing efficiency

In Fig 4A, we observed a significant variation in the repair frequency among four genomic loci, differing by as much as 190-fold. To investigate whether target orientation also influences DGR editing, we constructed four more reporter strains, in which the orientation of the target at those chromosomal loci was reversed. Orientation reversal significantly affected repair frequency across all four loci, with the two reporters at the 291˚ locus showing the highest difference of 44-fold based on their orientation (Fig 5A). These results indicate that the efficiency of DGR editing depends on both the genomic location and orientation of the target.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Chromosomal location and orientation of the target gene influence DGR activity.

(A) Bar graph comparing the repair efficiency of reporters inserted at the indicated chromosomal locations in both orientations in E. coli. DGR components were expressed from pDGR2. n = 3; error bars indicate standard deviation. (B) Schematic of a high-throughput workflow for profiling the efficiency of DGR-dependent editing of thousands of chromosomally integrated reporters at once. (C) Schematic showing the reporter orientation within the transposon. Transposon sequencing workflow was used to map the right transposon-chromosome junction. In this transposon design, the reporter is co-directional with the sequencing read. Normalized Illumina read counts (reads per million, RPM) of mapped transposon insertion sites across the MG1655 genome for both the Input and Output libraries. Note that the genomic coordinate is adjusted to center the data around oriC (3,925,797 bp). (D) Same as described in C, except that the reporter orientation is reversed in the transposon used in this experiment. Consequently, the reporter orientation is contra-directional with the sequencing read. Data in C and D, are from a single experiment.

https://doi.org/10.1371/journal.pgen.1012038.g005

To profile these effects at the genome scale, we used the Tn5 transposase to introduce reporters at roughly 5000 chromosomal loci and used the Tn-seq workflow to measure their repair frequency in a single experiment (Fig 5B). Our analysis identified about 5000 unique target insertions throughout the chromosome in both orientations in the starting library (Fig 5C, top). However, the post-selection output library, which contained ‘repaired’ reporters, showed a dramatically different distribution (Fig 5C, bottom). This altered representation recapitulated the location and orientation bias seen in our low-throughput experiment (Fig 5A).

Reporters located near the origin of replication (oriC) became significantly overrepresented, indicating a higher repair frequency in this region. The higher copy number of targets near oriC, resulting from multi-fork replication in E. coli [24,25], likely explains the enhanced target editing efficiency in this region. Beyond this positional effect, we also observed a striking pattern dependent on reporter orientation (Fig 5C). Across the genome, one orientation is always preferred from the other, but the favored orientation abruptly switches at oriC. Specifically, reporters oriented leftward were preferentially enriched on the left side of oriC, while the opposite orientation (right-facing) was enriched on the right side of oriC. This pattern demonstrates that the DGR mechanism consistently prefers reporters that point away from oriC.

To confirm that the observed genome-wide trends were not an artifact of our transposon design, we reversed the reporter orientation in a new transposon. This control resulted in a corresponding reversal in the enriched transposon orientation (Fig 5D). Thus, DGR editing efficiency is inherently influenced by reporter orientation, and the preferred orientation is dictated by the target’s genomic position relative to oriC. This finding suggests a programmable mechanism for using a single DGR to diversify multiple targets at differential rates within the same cell.

Transcription cannot explain the orientation bias

We next sought to understand why target genes pointing away from oriC exhibited higher DGR editing efficiency. Bacterial chromosomes typically orient highly transcribed genes co-directionally with DNA replication to prevent genome instability arising from head-on collisions between the RNA polymerase and the replisome [26]. We therefore hypothesized that the directional enrichment of stronger promoters could indirectly enhance the transcriptional output of reporters oriented away from oriC, thereby increasing their DGR editing efficiency. The kanamycin reporter assay is ill-suited to investigate this hypothesis, as altering the expression of the reporter would confound measurements of resistance. Instead, we used sequencing to measure the rate of A-to-N mutations directly as a function of the target’s promoter strengths.

For this assay, we designed a new reporter (VR4) with 18 adenines and drove its expression with either a weak (P_J23112) or strong (P_J23118) constitutive synthetic promoters (Fig 6A). To maximize mutation detection efficiency, we strategically incorporated three key design features learned from our study: we included many highly mutagenic AAC (Fig 3E); we used the highly active ΔsbcB strain (Fig 4); and we inserted the reporter near oriC (291°) (Fig 5C–5D). We also inactivated DNA mismatch repair by deleting mutS. While the 18-nt insertion is not a substrate for MutS in our kanamycin repair assay (where it showed no effect, S7 Fig) [27], deleting mutS should help preserve any DNA substitutions introduced by DGR in this assay. Finally, to control against spontaneous adenine mutation, we introduced a G-to-T signature mutation into the template (TR4; Fig 6A shaded grey), which was otherwise identical in sequence to VR4. The co-transfer of this mutation alongside the adenine mutations allowed us to definitively identify DGR-specific mutations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Target orientation relative to replication directionality determines DGR efficiency.

(A) Sequences of TR4 and VR4 used for deep sequencing analysis of DGR-mediated adenine mutations. Co-transfer of T from TR4 to replace G in VR4 (colored grey) was used as a signature to distinguish DGR-induced adenine mutations from spontaneous adenine mutations. (B) qPCR analysis of reporter expression levels in E. coli MG1655. Total RNA was extracted from cells harvested at OD₆₀₀ ~ 0.6. The VR4 reporter, controlled by a weak (P_J23112) or strong (P_J23118) constitutive promoter, was integrated at the 291˚ locus in a derivative of MG1655 strain lacking ΔsbcB ΔmutS. Reporters were inserted in either co-directional or anti-directional orientations with replication, generating the strains HCL124 (P_J23112, co-directional), HCL121 (P_J23112, anti-directional), HCL126 (P_J23118, co-directional) and HCL123 (P_J23118, anti-directional). For each orientation, reporter expression levels were normalized to tufA and then to the expression level of the P_J23112. Error bars denote the standard error of the mean for three biological replicates. (C) Deep-sequencing quantification of DGR-mediated editing rates of the VRs described in A. Cells of the indicated E. coli strains were grown in LB + 0.2% arabinose until OD₆₀₀ ~ 2. Genomic DNA was isolated from harvested cells, and the VR regions were PCR amplified for sequencing. The mutagenesis rate of VR4 represents the fraction of sequencing reads with both at least one A-to-N mutation and the designed G-to-T mutation in the TR4 region and then dividing it to the total number of reads in that sample. Error bars denote the standard error of the mean for three biological replicates. Regardless of the transcriptional output, opposing reporters showed significantly different mutational rates: p-values are 0.004 (P_J23112) and <0.0001 (P_J23112). (D) Schematic of a plasmid with a p15A origin of replication. The dominant replication fork replicates almost of the entire plasmid (marked by orange arrow). Reporters were inserted in both co-directional and anti-directional orientations relative to the dominant replication fork at the indicated sites, creating pNEB298 (right, anti-directional), pNEB299 (right, co-directional), pNEB300 (left, anti-directional), and pNEB301 (left, co-directional). (E) Repair frequency of reporters described in D. Measurements were performed in the E. coli MG1655 strain using the kanamycin repair assay, with each p15A-based plasmid co-transformed with pDGR2. Bar heights are the mean of n = 6 biological replicates; error bars indicate the standard deviation.

https://doi.org/10.1371/journal.pgen.1012038.g006

After confirming that these promoters exhibited the expected difference in transcriptional output (Fig 6B), we transformed these strains with a pDGR2 variant to express the corresponding TR4-RNA alongside bRT and Avd. We then sequenced the reporter to quantify the rate of A-to-N mutations. Consistent with our previous result (Fig 3E), adenines in the AAC context exhibited significantly higher mutations rates (S8 Fig). Furthermore, targets oriented co-directionally with replication accrued mutations at a higher rate than those in the reverse orientation (weak promoter: 7.6-fold, strong promoter: 24.6-fold) (Fig 6C). However, for reporters with the same orientation, an orders-of-magnitude difference in transcriptional output only mildly affected the rate of A-to-N mutations (Fig 6C). Thus, while transcription contributes to mutagenic retrohoming, it cannot explain the observed orientation effects.

Replication directionality dictates the preferred target orientation

As with most bacteria, chromosome replication in E. coli, initiates at the oriC, forming two replication forks that migrate in opposite directions [25,28]. This bifurcating replication direction aligns precisely with the chromosomal reporter orientations that favor DGR editing (Fig 5C–5D). To test whether DNA replication direction causes this target orientation bias, we measured the repair frequency of reporters encoded on a plasmid with a p15A origin. In contrast to bidirectional chromosomal replication, p15A-based plasmids are replicated largely in one direction by a dominant replication fork [29] (Fig 6D). Unlike what was observed for chromosomal targets, a single reporter orientation exhibited consistently higher repair frequency, regardless of the reporter’s location with respect to the origin (Fig 6E). Crucially, the preferred target orientation on this plasmid aligned with the direction of the dominant replication fork, thus demonstrating that replication directionality dictates the preferred editing orientation.

We next sought to understand why targets co-directional with replication exhibit much higher DGR editing efficiency than those in the reverse orientation. We hypothesized that this effect stems from the inherent asymmetry in which the two complementary DNA strands are replicated. While the leading strand is replicated continuously with little exposed ssDNA region, the lagging strand is replicated in short, disjointed segments, separated by large ssDNA gaps [30]. This is significant because, in the preferred (co-directional) target orientation, the sequence complementary to the cDNA is located on the lagging-strand template (S9A Fig). Therefore, the higher efficiency may be attributed to enhanced cDNA homing into the extended single-strand gap.

To directly test this model, we attempted to narrow the size of the lagging-strand ssDNA gap by overexpressing the primase DnaG, expecting a corresponding reduction in lagging-strand editing efficiency. We observed no change in the reporter repair frequency (S9B Fig). We also measured DGR-mediated editing in a reporter strain that bears the dnaG(K580A) allele. The K580A mutation is known to diminish recruitment of DnaG to the site of lagging strand synthesis and therefore increase the size of the ssDNA gap [31]. However, the reporter repair frequency was unaffected by this genetic alteration (S9C Fig).

We consider these negative results inconclusive. The cDNA (~70 nucleotides) in our reporter assay is much smaller than the typical ~1–2 kilobase ssDNA gap [30]. It is still uncertain whether our genetic manipulations altered this gap sufficiently and reproducibly to yield a detectable change in DGR efficiency. Nonetheless, our data clearly implicate DNA replication as a major driving force in DGR efficiency, even as more sensitive assays remain needed to clarify the precise underlying mechanism.

Plasmid construction

Plasmids were either constructed by GenScript Biotech or generated in-house using NEBuilder HiFi DNA Assembly (NEB, E2621) or NEB KLD Enzyme Mix (NEB, M0554). The sequences of all plasmids used in this study are listed in S2 Table.

Strains and growth conditions

All bacterial strains used in this study are listed in S3 Table. With few exceptions, all the reported experiments were performed in E. coli strain MG1655 and its derivatives. Cells were grown in grown in lysogeny broth (LB; 1% tryptone, 1% NaCl, 0.5% yeast extract; Teknova, L8000) or Lennox LB (1% tryptone, 0.5% NaCl, 0.5% yeast extract; Teknova, L5000) as indicated. Plasmids were maintained in NEB5α, except for those with an R6K origin, which were maintained in One Shot PIR2 (Invitrogen, C111110). Whenever appropriate, the following antibiotics were used: 100 μg/mL ampicillin (Amp100), 15 μg/mL chloramphenicol (Cm15), 25 μg/mL kanamycin (Kan25), 50 μg/ml Spectinomycin (Spec50), and 10 μg/mL tetracycline (Tet10).

E. coli strain construction

Reporter insertion: E. coli strains with genomically integrated reporters were constructed by amplifying desired sequences from synthesized gene fragments ordered from IDT or Twist. The primers used for this amplification had 40–60 nucleotides of complementarity to the intended genomic insertion site. The purified PCR products were then electroporated into electrocompetent MG1655 cells that contained the Lambda Red recombineering plasmid pKD46. After a one-hour recovery in SOC (NEB, B9020), the cells were plated on the appropriate selective media to isolate single recombinants.

Gene deletion: Mutations from relevant strains in the KEIO E. coli knockout collection (Horizon Discovery) [47] were first verified by PCR, as described by Datsenko et al. [48]. The verified mutations were then transferred into an E. coli MG1655 background using P1 phage transduction [49].

To remove selectable markers flanked by FRT sites, the strain was transformed with pCP20 [50]. Final insertions were verified by PCR. All primers used in this study are listed in S4 Table.

Kanamycin repair assay

3-5 colonies were inoculated into 3 mL LB with the appropriate antibiotics in 24-well deep well round bottom plates and sealed with a breathable film. The cultures were shaken at 300 rpm at 37°C for 90 minutes. Arabinose was added to 0.2% final concentration and grown for another 4.5 hours. OD₆₀₀ of the cultures was recorded (~ 2 in most genetic backgrounds). To select for kanamycin resistance, appropriate dilutions of cultures were plated on LB supplemented with Kan25. Colony forming units (CFU) were recorded after a 16-hour incubation at 37°C. To calculate repair frequency, calibrated CFU/mL was divided by the total cells plated based on OD₆₀₀ of the culture.

To sequence the chromosomal reporter prior to kanamycin selection, 1 mL of culture was collected by centrifugation and genomic DNA was extracted using Monarch Spin gDNA extraction kit (NEB, T3010). 75 ng of genomic DNA per sample was PCR amplified with Q5 DNA Polymerase (NEB, M0491) using equimolar mixture of forward primers (oNEB-203, 204, 205, 206) and reverse primers (oNEB-170, 171, 172). PCR product was used as the template for a second PCR to add Illumina barcodes. The amplicons were cleaned up with 1.2X NEBNext DNA Sample Purification Beads and sequenced on Illumina iSeq 2x 150 paired end run.

To sequence the reporter following kanamycin repair assay, resistant colonies were pooled and pelleted. Plasmids were isolated using Monarch Plasmid Miniprep Kit (NEB, T1010) and used as the template for amplifying the reporter region on pTarget. PCR were carried out with Q5 DNA Polymerase (NEB, M0491) using an equimolar mixture of forward primers (oNEB-145, 170, 171, 172) and an equimolar mixture of reverse primers (oNEB-173, 174, 175, 176). The first PCR was used as the template for a second PCR to add Illumina barcodes. The amplicons were cleaned up with 1.2X NEBNext DNA Sample Purification Beads and sequenced on Illumina NextSeq 500 2 x 150 paired end run. For RNA extraction, cells were pelleted and stored at -80°C.

RNA Extraction, TR RNA-cDNA hybrid sequencing, and quantification with RT-qPCR.

Total RNA was extracted from log-phase E. coli with Monarch Total RNA Extraction Kit (NEB, T2010). The on-column DNase I treatment step was skipped to preserve the cDNA-TR-RNA hybrid.

cDNA sequencing: To sequence the cDNA within cDNA-TR hybrids, 1 μg of the extracted RNA was reverse transcribed using Induro RT (NEB, M0681) with primers that bind to the 3’ end of the cDNA. These primers were oNEB-344 (for pDGR1, Fig 2E–2G) and oNEB-356 (for pDGR2 variants, Fig 3). The resulting RT reactions then served as templates for a PCR amplification across the cDNA-TR-RNA junction using Q5 DNA Polymerase (NEB, M0492). For the pDGR1 construct, a mixture of forward primers (oNEB-319, 320, 321, 322) and a mixture of reverse primers (oNEB-535, 536, 537, 538) were used. For the pDGR2 variants, the same forward primer mixture was used, but with a different reverse primer mixture (oNEB-460, 461, 462, 463). A subsequent PCR step was performed to add barcodes for multiplexed sequencing.

Quantitative PCR (qPCR) analyses: For cDNA-TR RNA hybrid qPCRs; 2 µg of RNA (DNase I treatment skipped) was reverse transcribed with SuperScript IV (Invitrogen, 18090010) using a specific primer (oNEB-344) or random hexamers (NEB, S1230S) according to manufacturer’s protocol. RT reactions were diluted 1:10 and 5 µL of the dilutions were used as template in a 20 µL qPCR reaction with 250 nM of forward and reverse primers (oNEB-301 and oNEB-345 for the cDNA-TR RNA hybrid and oNEB-313 and oNEB-314 for tufA internal control) and 2X Luna Universal qPCR Master Mix (NEB, M3003). For TR-RNA qPCRs; 10 µg of the RNAs were treated with DNase I (NEB, M0303) according to manufacturer’s recommendations and purified using Monarch Spin RNA Cleanup Kit (NEB, T2030). RT and qPCR reactions were then performed as described above with specific primers for TR-RNA only (oNEB-343 for RT and oNEB-300 and oNEB-346 for qPCR). tufA expression level was used as the internal control. All reactions were performed on a Bio-Rad CFX96 system, and the average of three technical replicates was taken for each independent biological replicate.

Data analysis for A-mutational rate

Illumina sequencing reads were aligned to the reference substrate using the minimap2 aligner with the -x sr parameter, optimized for short read mapping [51]. For each of the sixteen sequence contexts (NAN) within the reference substrate, the frequency of each nucleotide observed at the central position was recorded. The percentage of misincorporation at the central adenine was calculated and is presented as a heatmap. Following and preceding nucleotide heatmaps were generated similarly. For the weblogos, Illumina reads from the VRs of three biological reads were combined, the target region was trimmed with cutadapt, a custom python script was used to filter only the 36-nt region that corresponds to the VR. A random 10,000,000 reads were selected from the triplicate read pools for each TR and fed into WebLogo (version 3.7.12) [52,53] with -U ‘probability’ argument to show the probability on the y-axis instead of the default bits. To generate the amino acid sampling heatmaps, we used the same randomly selected pool of 10,000,000 combined triplicate reads. For each expected AAN triplet in the TR1, TR2, and TR3 templates, the corresponding observed triplets were identified. The data for all three templates was combined, and the proportion of amino acids encoded by these observed triplets was calculated for each anticipated triplet and visualized as a heatmap.

Theoretical stop codon conversion rate calculation

The mutation probability for an adenine (A) residue is 0.3065. Under the assumption of equal misincorporation probabilities—causing A to G, C, or T mutations with equal frequency—each specific mutation occurs at a rate of approximately 0.1022 (0.3065 ÷ 3). Conversely, the probability of the adenine remaining unchanged is 0.6935.

Following this logic, the probability of an AAG codon mutating to the stop codon TAG is approximately 7.08% (0.1022 × 0.6935 × 1.0000). Given that the DGR mechanism can diversify AAA into all three termination codons (TAA, TAG, and TGA), we calculated the specific probability for each transition:

• AAA → TAA: ≈ 0.1022 × 0.6935 × 0.6935 ≈ 0.0491 or 4.91%
• AAA → TAG: ≈ 0.1022 × 0.6935 × 0.1022 ≈ 0.0072 or 0.72%
• AAA → TGA: ≈ 0.1022 × 0.1022 × 0.6935 ≈ 0.0072 or 0.72%.

Consequently, the aggregate probability of an AAA codon mutating to any stop codon is approximately 6.36%.

Transposon sequencing

Genetic screen for identifying genes influencing DGR activity (Fig 4): The plasmid pNEB99 was constructed to generate the transposon mutant library. This plasmid encodes a hyperactive Himar1 mariner transposase under the control of an arabinose-inducible promoter, along with a transposon containing a chloramphenicol resistance gene. The pNEB99 plasmid was introduced into the reporter strain HCL26 via puddle mating. Overnight cultures of the MFDpir donor strain with pNEB99 (in LB with 0.2 mM DAP and Amp100) and the HCL26 recipient strain (in LB) were mixed at a 4:1 ratio and pelleted. The pellet was resuspended, spotted on a Lennox LB agar plate topspread with 0.2 mM DAP, and incubated for 24 hours at room temperature.

To select for cells with transposon insertions, cells were scraped from the mating spot, resuspended in Lennox LB, and plated on Lennox LB agar plates supplemented with 0.2% glucose and Cm15. Approximately 120,000–180,000 clones were pooled and used to prepare electrocompetent cells. These cells were then transformed in duplicate with pNEB132 (pDGR2) and immediately recovered at 30°C in 25 mL of Lennox LB with Amp100. Spot dilutions estimated that about 20 million cells received the plasmid, which is 100 times the size of the starting library. When the OD₆₀₀ reached 0.5, this culture was used to inoculate a 500 mL Lennox medium culture supplemented with Amp100 and 0.2% arabinose to induce DGR expression. After a 16-hour incubation at 30°C, 1.6 × 10⁹ cells were harvested by centrifugation and designated as the Input. Simultaneously, cells were plated on Lennox LB agar supplemented with Kan25 and incubated at 25°C for 48 hours. Approximately 4.8 million CFUs, or more than 26 times the size of the starting library, were pooled and designated as the Output.

Profiling the effect of target gene location and orientation (Fig 5): Two transposon versions (defective kanamycin reporter in either orientation) were created. They were generated by PCR amplification from plasmids pNEB34 and pNEB35 using Q5 DNA Polymerase (NEB, M0491) and primers oNEB-192 and oNEB-193 to add the inverted repeats. The amplicons were purified and incubated with EZ-Tn5 Transposase to assemble the transposomes according to the manufacturer’s protocol. These transposomes were then electroporated into HCL1 cells. Following a 90-minute recovery in SOC (NEB, B9020), the cultures were plated on LB supplemented with Cm15 and incubated overnight at 37°C. Approximately 5,000 transformants per condition were pooled.

The resulting reporter libraries were made electrocompetent and transformed with pDGR2. The transformants were selected by plating on LB with Tet10 at 37°C. About 1.5 million CFUs were scraped and resuspended in LB with Tet10 to prepare a starter culture. A 200 mL starter culture was inoculated at an initial OD₆₀₀ of 0.05 and grown for 50 minutes at 37°C with shaking at 230 rpm. DGR activity was then induced by adding 0.2% arabinose, and the culture was grown for an additional 6 hours. A 20 mL sample from each culture was frozen as the Input population before kanamycin selection. To select for cells with repaired reporters, 2 × 10¹⁰ cells were plated on LB with Kan25 and incubated overnight at 37°C. Approximately 25 million kanamycin-resistant colonies were pooled and frozen as the Output population.

Tn-seq library preparation and deep sequencing

Genomic DNA was extracted from cells with Monarch Genomic DNA Purification Kit (NEB, T3010). 5 μg of genomic DNA was sheared with Covaris ML230 instrument with these parameters: peak intensity power 260 W; duration 5 seconds; duty factor 6; cycles per burst 50; iterations 5; dithering Y = 3.0; speed = 20.0. The fragments were size selected with 0.6X NEBNext DNA Sample Purification Beads (NEB, E6178) to enrich for 1-1.5 kb DNA fragments. Purified DNA was end repaired with NEBNext End Repair Module (NEB, E6050) and cleaned up again with 0.6X NEBNext DNA Sample Purification Beads. Eluted DNA was treated with 10 units of Terminal Transferase (NEB, M0315) in a 50 mL reaction with 1 μL 9.5 mM dCTP, 0.5 mM ddCTP, 4 μL 10X TdT Buffer at 37°C for 30 minutes to add a poly-C tail to the DNA. Following heat inactivation at 75°C for 20 min, poly-C-tailed DNA was purified with 0.6X NEBNext DNA Sample Purification Beads. Purified DNA was used as the template to amplify the transposon-genome junction. 16 cycles of PCR were carried out with Q5 DNA Polymerase (NEB, M0491) using an equimolar mixture of oNEB-198, 199, 200, and 201 as forward primers and oNEB-202 as the reverse primer (S4 Table). PCR products were cleaned up with 0.7X NEBNext DNA Sample Purification Beads and eluted in 40 µL water. A second nested PCR was performed with 20 µL of the cleaned-up PCR products in 100 µL reaction with barcoded i5 and i7 primers for 14 cycles. Barcoded samples were combined and purified with 0.7X NEBNext DNA Sample Purification Beads. Libraries were then sequenced on Illumina NextSeq 500 1 x 150 single end run.

Data analysis for transposon-based screens

Sequencing reads were analyzed using a previously described pipeline [54]. Briefly, Cutadapt (version 4.5) [55] was used to remove adaptors and trim low-quality sequences. Trimmed reads were mapped to HCL26 (NC_000913 with reporter inserted at 317°) using Bowtie 1.0.0 [56]. Fold change in normalized sequencing reads for a given gene in the Output versus Input library were calculated and statistical significance was determined by applying the Mann-Whitney U test. Genes that had at least 2-fold change and p-value less than 0.05 were deemed hits. Gene Ontology [57,58] was used to annotate gene function.

To analyze the effect of reporter location and orientation, NovaSeq Illumina reads were trimmed for poly G tails using fastp program with the –trim_poly_g parameter [59,60]. Reads with the expected transposon junction sequence (AAGAGACAG) were selected and trimmed of this sequence. Only reads at least 25 nt long after trimming were retained for analysis. The trimmed reads were then mapped to the E. coli strain MG1655 reference genome (GenBank NC_000913) using the minimap2 aligner with the -x sr parameter, optimized for short read mapping [51]. Only primary alignments with a mapping quality of 60 were considered. The first mapped position and reference strand were saved as CSV files and plotted.

ExoI exonuclease assay

Cells harboring plasmids encoding ExoI variants were cultured in LB supplemented with 1 mM IPTG and (antibiotics) for 16 hours at 37°C. A total of 15 OD₆₀₀ units of cells were collected and lysed using 0.75 mL of NEB Express E. coli Lysis Buffer (NEB, P8116) and rotated for 20 minutes at room temperature. Total cell lysates were clarified by centrifugation at 10000 x g for 10 minutes at 4°C. For the assay, 40 µL reactions containing 20 µL clarified cell lysates and 0.1 µM 5’ Alexa488-labeled ssDNA probe with the sequence Alexa488N/AAGGGCAGGCTGGGAAATAACGCATCGCCTTCTATCGCCTTCTTGACGAGTTCTTCTAATAACGGGGCGCGCGGCGTCTG (ordered from IDT) was prepared in 1X ExoI reaction buffer (NEB, MS0293) and incubated at room temperature for 10 minutes. A positive control was included by adding 40 units of purified, commercially available ExoI (NEB, MS0293) instead of the clarified cell lysate. Reactions were stopped with Novex 2X TBE-Urea Sample Buffer (ThermoFisher, LC6876) and the products were resolved on a 15% TBE-Urea gel (ThermoFisher, EC68855BOX). Gels were imaged using a Cytivia Amersham Typhoon in fluorescence mode with the Alexa488 filter.

Quantification of the DGR mutational frequencies to understand the effect of transcription

Cultures of reporter strains (weak promoter: HCL121 and 124, strong promoter: HCL123 and 126) with chromosomally integrated VR4 carrying the pNEB153 plasmid were grown in LB supplemented with Tet10 and Ara0.2 for 16 hours at 37°C. Cells were harvested by centrifugation, and genomic DNA was extracted from cell pellets using Monarch Spin gDNA Extraction kit (NEB, T3010) and used as the template to amplify VR4 for deep sequencing. Twelve PCR cycles were carried out with Q5 DNA Polymerase (NEB, M0491), an equimolar mixture of forward primers (oNEB-203, 204, 205, 206) and an equimolar mixture of reverse primers (oNEB-528, 529, 530, 531) (S4 Table). A second nested PCR (6 cycles) was performed with 1 µL of the PCR products in 50 µL reaction with barcoded i5 and i7 primers (NEB, E6441A. Barcodes columns C6-D8). Barcoded samples were combined and purified using 1.2X NEBNext DNA Sample Purification Beads. Libraries were then sequenced on Illumina NovaSeq 2 x 150 paired end run. To confirm the promoter strengths, RNA was extracted from 1mL of log phase cells from each strain with Monarch Total RNA Extraction Kit (NEB, T2010). 1 μg of the extracted RNA was reverse transcribed using Induro RT (NEB, M0681) and random hexamers. RT reactions were then diluted 1:10 and 5 µL of the dilution was used in 20 µL qPCR reaction with 250 nM oNEB-573 and 250 nM oNEB-574 primers and 2X Luna Universal qPCR Master Mix (NEB, M3003). tufA expression level was used as an internal control.

Computational analysis for mutational frequencies to understand the effect of transcription

Paired end reads were merged into single reads with BBMap bbmerge [61] to lower the sequencing error rates. Primer binding sites and any sequence before and after were trimmed from the merged reads with Cutadapt (version 4.5) [55] to obtain the VR4 sequences with --discard-untrimmed argument to only include sequences containing primer binding sites. Reads that passed this filter were subjected to two analyses:

1. 1) Quantify the fraction of VR4 edited by DGR: A custom python script was written to identify VR4 sequences containing both a G-to-T mutation 8 nucleotide 5’ of the GC14 site (designed to distinguish between DGR-mediated and nonspecific adenine mutations, see Fig 6C for details) and at least one A-to-N mutation. The number of reads satisfying these criteria was then normalized to the total read number and plotted for the different conditions (Fig 6E).
2. 2) Determine the mutagenesis rate of VR4 at the nucleotide level: VR4 sequences considered to be edited by DGR (see above) were mapped to the unmodified VR4 sequence using minimap2 (version 2.26) [51] with the --sr argument for short reads. Primary alignments were then selected and sorted using samtools (version 1.17) [62]. To detect mutations with location information, samtools mpileup was used with these arguments on sorted primary alignments: -d 0 --min-MQ 20 --min-BQ 30 --output-BP-5 --no-output-ins --no-output-ins --no-output-del --no-output-del --no-output-ends. The mutation counts per location were then normalized to total read counts and plotted for each condition through the VR sequence (S8 Fig).