The bacterial strains and plasmids used in this study are listed in Table 1 and Supplementary. Table S2, respectively. Plasmids pCRn100 days to pCRn125 days, pMLi002 and pMLd009 are based on the pK18 mobsacB  vector (Schäfer et al., 1994a). pCRn100 days to pCRn125 days were used for successive genomic deletion of ISCg elements in MB001. pMLi002 allows for genomic integration of the coding sequence of the Bxb1 phage integrase at the former locus of ISCg3a deleted by pCRn110 days pMLd009 is used to remove groEL1 in the genome. The plasmid pML10 was build using pUC19 as backbone. It is used for cloning and replication in E. coli and as a transposition construct in C. glutamicum. It carries a kanamycin resistance cassette flanked by the ISCg1 inverted repeats as transposable element as well as the transposase encoded by tnp1. Expression of this gene in turn depends on a Bxb1 integrase-based switch (Bonnet et al., 2012). E. coli strains carrying plasmids were routinely grown on solid Antibiotic Medium No. 3 (PA) (Oxoid, Wesel, Germany) at 37°C. C. glutamicum strains were grown on solid brain-heart broth (BH) (VWR International, Darmstadt, Germany) at 30°C. Antibiotics used for selection of plasmids and strains were nalidixic acid (50 μg/ml for corynebacteria) and kanamycin (50 μg/ml for E. coli, 25 μg/ml for corynebacteria).

Standard procedures were employed for molecular cloning and transformation of E. coli DH5α, as well as for electrophoresis (Sambrook et al., 1989). Transformation of C. glutamicum was performed by electroporation using the methods of Tauch et al. (2002).

Sequence similarity-based searches with nucleotide sequences were performed using the basic local alignment search tool (BLAST, Altschul et al., 1990).

Plasmids pCRn100 days to pCR125 days as well as pMLi002, pMLd009 and pML10 were constructed using Gibson assembly (Gibson et al., 2009). The primers used are listed in Supplementary. Table S1. In case of “genome healing” constructs, the final insert was amplified directly from either “Brevibacterium lactofermentum” ATCC 13 869, “Brevibacterium flavum” ATCC 14 067, or “Corynebacterium crenatum” AS 1.451 gDNA using the respective primers _d1 and _d4. In all other cases, two products (primer pairs _d1 and _d2 respectively _d3 and _d4) were amplified from C. glutamicum MB001 gDNA. The insert(s) were then assembled with linearized pK18 mobsacB amplified via the primers pK18_ga1 and pK18_ga2. Primers pk18_dgroel1 and pk18_dgroel2 were used to linearize pK18 mobsacB for groEL1 deletion. The homologous flanks were amplified with the primers d1_groel1-d4_groel1 from ATCC 13 032. Linearization of pCRn110 days for insertion of the Bxb1 integrase coding sequence was achieved with the primer pair pK18_bxbI_1 and pK18_bxbI_2 and the sequence of the Bxb1 integrase was amplified with the primer pair bxbI_ins1 and bxbI_ins2 from Mycobacterium smegmatis. The backbone for pML10 was linearized with the primer pair switch_ga1 and switch_ga2 from pUC19. The switch, consisting of the Bxb1 integrase attB and attP flanking the original constitutive Anderson promoter (Anderson, 2013) was amplified via the overlapping primers attP_Pcons and Pcons_attB. Kan ^r originates form pK18 mobsacB and the imperfect inverted repeats for ISCg1 are from ATCC 13 032. All PCRs were performed using Phusion High–Fidelity DNA polymerase, the assembly was done using Gibson Assembly Master Mix (both Thermo Fisher Scientific, Germany). The assembly mixture was used to transform E. coli DH5αMCR, the transformants were selected on PA plates containing 50 μg/ml kanamycin and 40 mg/L X-Gal (5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside) and verified by Sanger sequencing.

Site-specific gene disruption was performed using the non-replicable integration vector pK18mobsacB which allows for marker-free deletion of the target gene (Schäfer et al., 1994c). The resulting plasmids pCR100 days to pCR124 days and pMLi002and pMLd009 were transformed into C. glutamicum MB001 respectively subsequent deletion mutants by electroporation (Tauch et al., 2002). Integration of the introduced plasmids into the chromosome by single-crossover was tested by selection on BH plates containing 25 μg/ml kanamycin. For the deletion of the target gene, the kanamycin-resistant (Km^r) cells were grown overnight in liquid BH and spread on BH plates containing 10% sucrose. Cells growing on this plate were tested for kanamycin sensitivity (Km^s) by parallel picking on BH plates containing either kanamycin or sucrose. Sucrose-resistant and kanamycin-sensitive cells were then tested for the deletion by PCR and checked for IS activity using Southern blotting and hybridization (Evans et al., 1994) if deemed appropriate.

Genomic DNA was isolated using the NucleoSpin Microbial DNA Mini kit (Macherey-Nagel, Germany). Single mutant colonies were sequenced using the Rapid Barcoding Kit SQK-RBK004 (Oxford Nanopore Technologies, UK) on R9.4.1 flowcells on a GridION.

For sequencing of complete mutant libraries, all colonies were pooled before genomic DNA was isolated. Subsequently, the gDNA was fragmented to approximately 10,000  bp fragments using a Covaris g-TUBE. The ends of the fragments blunted and an A-overhang was added with the NEBNext Ultra II End Repair/dA-Tailing Module. The A-overhang is then used for ligation of dsDNA adapters derived from annealing the oligos D7 and D7_top carrying part of the TruSeq D7 sequence. In a first round of PCR using Phusion High–Fidelity DNA polymerase, fragments are amplified between a biotinylated primer tnp_o, binding in the kanamycin resistance marker about 500 nt upstream of the inverted repeat, and the TruSeq D7 primer. After amplification, fragments were purified with Dynabeads M-270 Streptavidin binding the biotinylated fragment ends. During washing of the trapped DNA on a magnet rack, 0.1 mM NaOH was applied to retain ssDNA of the amplified region. This is followed by a half-nested second round of PCR with LongAmp Taq DNA Polymerase. Here, primer D7 is paired with a phosphorylated nested primer tnp_i, located 150 bp from the 3′-end of the kanamycin resistance cassette.

After purification, the DNA was used in the adapter ligation step of the Ligation Sequencing Kit SQK-LSK109 (Oxford Nanopore Technologies) and sequenced on an R9.4.1 flowcell on a GridION. The sequencing reads where mapped to pML10 and C. glutamicum CR101 respectively using minimap2 (Li, 2018) with options-cx map-ont–eqx–secondary = no. Reads mapping to both references were processed using custom Python scripts to identify transposition sites based on the local alignments. These were visualized in a polar genome plot using the Python library Matplotlib (Hunter, 2007). Sites with at least two mapped reads were subjected to an analysis of the target site of ISCg1 by creating a sequence logo with WebLogo3 (Crooks, 2004) based on the strand-specific 8 bp of genomic sequence downstream the inverted repeat at each transposition site.

Cultivations to check for differences in growth behaviour of MB001, CR101, ML102 and ML103 were performed using a BioLector I (m2p-labs). Four biological replicates of each strain were cultivated in four technical replicates. Samples were cultivated in 1 ml BHI at 30°C, 1,100 rpm, and at 85% humidity. Measurements were taken every 15 min and the growth rates were calculated for each sample with the BioLection software using a sliding window of eight data points (i.e. 2 h). Afterwards, for each strain first the average growth rate for each biological replicate was calculated, followed by calculation of the mean and standard deviation from those averages.

The transformation efficiency was determined as described by Baumgart et al. (2013) with one modification: In case of the integrative plasmid, 1,000 ng instead of 500 ng were used. For each strain, three independent cultivations, i.e. biological replicates, were performed to create competent cells as described by Tauch et al. (2002). From each batch prepared in that manner, four aliquots were used for electroporation with the respective plasmid, serving as technical replicates.

The complete genome sequence of C. glutamicum CR101 was determined using a hybrid assembly approach as described by Ballas et al. (2020). All sequences from other C. glutamicum strains were obtained from GenBank (Agarwala et al., 2017): BX927147 (C. glutamicum ATCC 13 032), CP005959 (C. glutamicum MB001), CP016335 (“B. lactofermentum” ATCC 13 869), CP022614 (“B. flavum” ATCC 14 067), and LOQT01 (“C. crenatum” AS1.452). All data generated during this project is available via BioProject PRJNA750341.

To construct strain CR101, containing no potentially active IS elements (Kalinowski et al., 2003; Pfeifer-Sancar et al., 2013), strain MB001 lacking all prophages (Baumgart et al., 2013) was used as a basis. Interestingly, the genome of this strain contains five copies of ISCg1, named ISCg1e, which is located at position 541,217 to 542,527 (CP005959) and disrupts gene cgp_0611. To determine the boundaries of the necessary deletions as well as to amplify the inserts used for double-crossovers, we applied a “genome healing” approach: First, using BLAST we searched for the sequence of the ISCg elements with 2,500 bp flanks in the genome sequences of 3 C. glutamicum strains available at our lab (“Brevibacterium lactofermentum” ATCC 13 869, “Brevibacterium flavum” ATCC 14 067, and “Corynebacterium crenatum” AS1.452) to identify the orthologous regions in the three genomes. In those cases where 1) there was a syntenous region present, but 2) without the ISCg element found in MB001, we then compared the number of differences between the target sequence and that of MB001 in the 750 bp directly up- and downstream of the missing region. If these showed >99% identity, primers were designed to amplify a deletion construct with 500–600 bp flanking sequences. In case of more than one potential candidate, i.e., a region satisfying the rules present in more than one genome, the candidate with the highest similarity to the region in MB001 was selected. Using this approach, 12 regions were identified in ATCC 13 869, three in ATCC 14 067, and two in AS1.452 (see Supplementary Table  S2). For ISCg1e, a fifth copy of ISCg1 that was absent in the ATCC 13 032 wildtype (Kalinowski et al., 2003, Table two), the region was compared to the wildtype genome and genomic DNA of the wildtype was used to amplify the construct for “genome healing”. For the remaining eight regions, deletion constructs were build conventionally, based on RNAseq data (Pfeifer-Sancar et al., 2013) to remove the CDS as well as the corresponding promoter.

These constructs were then used to sequentially remove the IS elements, starting with ISCg1 and ISCg2 as the supposedly most active. This resulted in strain CR099 (see Supplementary Figure  S2) which was already successfully used as the basis in other genome reduction projects (Baumgart et al., 2018). To ensure that no further jumps of IS elements still remaining in the respective strain had occurred, the presence of the various IS elements was checked by Southern blotting and hybridization at key points, usually at least after removal of all instances of an element present in several copies or after deletion of three to four copies of elements present only once. Thereby, we were able to detect a total of seven jumps of ISCg1 (see Supplementary Figure S1 for an example), two jumps of ISCg2, one of ISCg5, one of ISCg13 and one of ISCg16 (data not shown). The strain with those active as well as several other elements removed was dubbed CR100 (see Supplementary  Figure S2). Removal of the remaining seven elements progressed without further complications, resulting in the final, prophage- and IS-free strain CR101.

To ensure the complete removal as well as to detect all other mutations introduced during the 26 rounds of genetic manipulation, the complete genome sequence of C. glutamicum CR101 was established using a hybrid assembly approach based on Illumina short read and Oxford Nanopore Technologies (ONT) long read data. The final genome sequence has a size of 3,034,563 bp, compared to 3,079,25 bp for MB001. Comparison of the annotated genomes revealed that 11 genes were “healed” through the removal of the various IS elements. While a large number of synonymous and missense mutations was introduced in the flanking regions, they all corresponded to the sequences found in the donor genomes and are thus unlikely to change the function of genes encoded in these regions. On the other hand, only four missense mutations and one frameshift were detected in the remainder of the genome, indicating an overall low rate of spontaneous mutations in C. glutamicum. The final strains were also checked for changes in growth rate and transformability. Strains CR101, ML102, and ML103 showed no significant difference in growth rate compared to the progenitor C. glutamicum MB001 (Supplementary  Table  S3). In case of transformability, a slight increase in the number of transformants per μg DNA could be observed (Supplementary  Figure S3), but analysis via two-way ANOVA with replication indicates that this is not significant in case of the replicative plasmid (p = 0.44 for an alpha of 0.05) as the variance between biological replicates far exceeds that among the four strains. In case of the integrative plasmid, the observed change is statistically significant (p = 0.023), but again the variability between batches of competent cells is extremely high.

As random insertion mutagenesis is of use in some experimental contexts, e.g., adaptive laboratory evolution Hennig et al. (ALE 2020), we decided to re-purpose ISCg1, using a system similar to that developed by Suzuki et al. (2006).

In most scenarios, only a single transposition event per mutant is desired. Therefore, pUC19 was used as a transposition construct, as it is a non-replicating vector in C. glutamicum and allows for easy replication and isolation in E. coli (Yanisch-Perron et al., 1985). We then used the regular ISCg1 sequence from ATCC 13 032 as transposase and a kanamycin resistance cassette flanked by the 24 bp imperfect terminal inverted repeats as transposon. This way, only a successful transposition into the genome provides resistance to kanamycin. However, problems occurred when we tried to isolate the plasmid from E. coli because ISCg1 transposition turned out to be highly active in E. coli already. This resulted in mostly “empty” plasmids, i.e., plasmids from which the transposon cassette had jumped into the E. coli chromosome. To acquire sufficient quantities of active plasmid for large-scale transposon mutagenesis, the plasmid pML10 was designed using a Bxb1 phage integrase genetic switch inserted upstream of ISCg1 (Figure 1). The switch consists of a constitutive promoter flanked by the bacterial and phage attachment sites targeted by the Bxb1 integrase (Bonnet et al., 2012). By default, the promoter is in opposing orientation to ISCg1 such that no expression is occurring and subsequently no transposition takes place. To invert the switch and enable expression of ISCg1, the coding sequence for the Bxb1 integrase has been inserted into the genome of CR101 at the former locus of ISCg3a creating ML102. As soon as pML10 is transformed into ML102 the Bxb1 integrase targets the attB and attP attachment sites flanking the promoter, forming the hybrid attachment sites attL and attR by recombination. In the process, the orientation of the promoter is inverted and drives the expression of ISCg1.

During sequencing of the first mutants, we noticed the occurrence of integration of the whole transposition construct pML10 into groEL1 with high frequency (data not shown). Examination of the sequences of several mutants revealed that all these insertions occurred at the same site and with the same directionality. The subsequent comparison of the bacterial attachment site in M. smegmatis to the integration site of our mutants showed a significant degree of similarity, indicating a partially conserved attB site in groEL1. Recognition of this site by the Bxb1 integrase then results in integration of the plasmid via the attP site of the genetic switch. Interestingly, many nucleotides defining attB functionality as well as the central GT dinucleotide thought to be essential (Singh et al., 2013) are not well conserved in the site present in C. glutamicum groEL1 (Figure 2).

To circumvent the genomic integration via the Bxb1 integrase as a competing event to transposition via ISCg1, we deleted groEL1. While this gene is essential in many organisms, there are two copies present in C. glutamicum ATCC 13 032 and groEL1 was actually disrupted by ISCg1c prior to its removal in CR099 and thus CR101. In subsequent mutant libraries of the strain ML103 no further display of integration via Bxb1 integrase was observed.

To test our genetic-switch-based system for random insertion mutagenesis, 200 transformations via electroporation of 800 ng pML10 each into ML103 were performed. On most agar plates less than 0.5 ⋅ 10² colonies were observed, but several cases with hundreds of colonies were also detected. From the resulting transposon mutants generated in ML103, we selected 12 single colonies for whole genome sequencing. One mutant colony was picked because of its noticeably intense yellow phenotype (Supplementary  Figure  S4), the others were selected randomly. In each of these mutants, only one transposition took place, each located at a different genomic position. The complete genome sequence of the phenotypically distinct mutant revealed a transposition into crtR which has been characterized as encoding a MarR-type repressor. CrtR represses the crt operon which is involved in decaprenoxanthin synthesis, a yellow carotenoid. Deletion of crtR increases transcription of the crt operon by 10 to 30-fold (Henke et al., 2017), probably explaining the striking phenotype of our mutant.

To further determine the suitability of ISCg1 in terms of target site bias, all mutants were pooled and ligation-based ONT sequencing libraries were constructed by targeted enrichment of the region downstream of the selection marker. Sequencing resulted in 3,207 independent transposition sites indicating no obvious target site preference on a genomic scale (Figure 3). Allocation of transposition sites with respect to strand directionality is evenly distributed with 1,582 sites on the leading strand and 1,625 sites on the lagging strand. Weblogo analysis of the eight bp target site of ISCg1 (Figure 4) suggests a bias towards loci with high AT content in the central four base pairs with strong discrimination against G and C at position four and 5. Frequency analysis of the central tetranucleotide (Supplementary Table S4) shows that no specific single sequence appears to be predominant which is in line with previous analyses (Inui et al., 2005).