Acinetobacter strains ED23-35, ED45-23, ED9-5a, EK30A, and VS15 used in this study were previously isolated from 15 thousand to 3 million years old permafrost sediments collected from different regions of Kolyma Lowland. We here in referred to these strains as “ancient” strains, which never come into contact with clinical isolates. Initially they were identified as A. lwoffii based on metabolic testing and from 16S rRNA gene sequence analysis (Petrova et al., 2002; Kurakov et al., 2016; Mindlin et al., 2016). In this work, based on data from a recent article by Nemec et al. (2018), revising the taxonomy of the Acinetobacter lwoffii group we showed that strain ED9-5a actually belongs to the Acinetobacter pseudolwoffii species. The remaining permafrost strains do belong to A. lwoffii. In addition, we analyzed two sets of Acinetobacter strains obtained from publicly available databases (“modern” strains). These were (i) 38 strains of A. baumannii described by Shintani et al. (2015), and (ii) 30 strains belonging to other Acinetobacter species obtained from public databases before January 2018. Bacteria were grown in lysogeny broth (LB) medium or solidified LB medium (LA) at 30°C.

Species identification was confirmed by comparison of partial rpoB gene sequences with the reference Acinetobacter genomes¹. The rpoB gene nucleotide identity level of 98–100% was regarded as evidence that the compared genomes belong to the same species.

The genomes were sequenced with a Roche GS FLX genome sequencer (Roche, Switzerland) using the Titanium protocol to prepare shotgun genomic libraries. The GS FLX reads were assembled into contigs using the GS De Novo Assembler (Roche); protein coding genes were identified and annotated using the RAST web server. To identify potential plasmid contigs we used read coverage depth and annotation information, such as existence of genes for mobilization and/or replication of plasmids. Using 454 assembly graph file (output of the GS De Novo Assembler) we extended and merge our potential plasmid contigs with each other. Some additional merging of contigs was done using PCR. All manually assembled regions were checked using PCR. A plasmid was considered as finished if it was circular according to the 454 assembly graph and PCR check. The 454 reads were mapped back to the finished sequences using 454 GS Reference Mapper, and read mapping was visualized in IGV browser and manually inspected for potential missassemblies.

A collection of plasmids isolated from different Acinetobacter species (Supplementary Table S1) was screened for the presence of pdif sites by BLAST software. The XerC/XerD (3916–3943 bp) and XerD/XerC (6899–6926) pdif sites of plasmid pM131-6 [NC_025120] were used as a reference. Only sites that were at least 75–80% identical to the reference were taken into account. The same algorithm was used in the search for the main and additional dif sites in Acinetobacter chromosomes.

For the assembly and analysis of plasmid genomes from ancient strains, the program UGENE² was used. Similarity searches were performed using BLAST (Altschul et al., 1997) and REBASE (Roberts et al., 2009). Conserved domains and motifs were identified using the NCBI Conserved Domain Database (CDD) (Marchler-Bauer et al., 2011) and the Pfam database (Finn et al., 2009). Dif sites were clustered using blastclust from blast package with 95% identity threshold. Fasta files with dif sites were sorted according to blastclust results using custom perl script (dif sites from the same cluster follow each other, and clusters are sorted by size). The sorted fasta files were visualized in AliView as color blocks. The fasta files with dif sites were used to create sequence logos using WebLogo³ (Crooks et al., 2004).

Identification of dif modules in permafrost plasmids and determination of their mobility was carried out manually in several stages. Initially, plasmids containing at least two different pdif-sites were selected. Then, the regions between the pairs of pdif-sites were analyzed, and, based on the presence of a potentially adaptive gene (s) in this region candidates for mobile modules were revealed. At the final stage, using the Blast software, the distribution of the identified modules among the ancient (isolated from permafrost) and modern strains of Acinetobacter’species was investigated. The presence of the same module with an identity level of 98–99% in different regions of different plasmids (contigs) isolated from Acinetobacter strains belonging to different species was regarded as evidence of its mobility.

The level of resistance was determined by the agar dilution method (Mindlin et al., 2016). Overnight cultures of bacterial strains were diluted 10-fold with fresh LB and grown at 30°C with shaking for 3–4 h; the cultures were then induced for 1 h by the addition of tert-butyl hydroperoxide (tBHP) at 0.1 mM. Five μl of the bacterial suspension (about 5 × 10⁶ cells per ml) were plated onto LA supplemented with Na₂TeO₃ (Tel) or tBHP. The concentrations tested were as follows: Tel – 0.01; 0.02; 0.04 and 0.1 mg/ml and tBHP – 0.05; 0.1; 0.2; 0.3 and 0.4 mM. The plates were incubated at 30°C for 24 h and visually inspected. In each case, three independent experiments were performed.

The accession numbers for nucleotide sequences of the plasmids deposited in this work in the GenBank database are as follows: CP032112.1 (pALWED1.2); CP032113-CP032116.1 (pALWED1.4-1.7); CP032117.1-CP032124.1 (pALWED2.2-2.9); CP032287.1-CP032289.1 (pALWED3.2-3.4); CP032290.1 (pALWED3.6); CP032102.1 (pALWEK1.1); CP032105.1-CP032107.1 (pALWEK1.2-1.4); CP032108.1- CP032111.1 (pALWEK1.6-1.9); CP032103.1 (pALWEK1.10); CP032104.1 (pALWEK1.11).

For analysis of the occurrence of pdif sites in Acinetobacter plasmids, we studied plasmids found in strains of various Acinetobacter species, divided into three groups (see Materials and Methods): (1) plasmids of permafrost strains of A. lwoffii and A. pseudolwoffii isolated and studied in our group (35 plasmids, see below) together with plasmids from A. lwoffii ZS207 (10 plasmids, see Supplementary Table S1); (2) plasmids from A. baumannii strains, described in Shintani et al. (2015) (65 plasmids, Supplementary Table S1); (3) plasmids isolated from strains belonging to others Acinetobacter species available from public databases before January 2018 (73 plasmids, Supplementary Table S1).

The presence of pdif sites and the number of their copies in each plasmid were determined by the BLAST tool. The results obtained for each set of plasmids are presented in Figure 1, Table 1, and Supplementary Table S1. Plasmid pdif sites were found in more than 50% of the Acinetobacter plasmids in all three groups (Table 1). In most cases, the number of sites was 1–4, while some plasmids contained 5–8 copies of pdif or more (Table 1). Most pdif sites were found in small to middle size plasmids (6–30 kb), while very small plasmids (<6 kb) as a rule did not contain them.

Interestingly, multiple copies of pdif sites were found both in very large and in middle-size plasmids (22–400 kbp) (Figure 1 and Table 2).

To determine the localization of pdif sites relative to the backbone and accessory plasmid regions, we analyzed their distribution in four plasmids, two middle size (pALWED2.3 and pOXA58-AP_882) and two large (pALWED2.1 and pXBB1-9), each containing more than 10 sites (Table 2). For all four plasmids, pdif sites were distributed unevenly across the plasmid sequence; notably, they were absent from the backbone regions and formed clusters in certain plasmid regions (Figure 2).

To our knowledge, there is currently no data on the distribution of typical pdif sites in the plasmids of bacteria from other systematic groups. To get insight into their abundance in closely related genera to Acinetobacter, we analyzed a total of 31 plasmids from 16 strains of Psychrobacter (2117–44793 bp) and 8 plasmids from 6 strains of Moraxella (1313–44215 bp) described by Shintani et al. (2015) and Moghadam et al. (2016). No pdif sites were detected in any of these plasmids. Therefore, the presence of a large number of pdif sites may be a unique feature of Acinetobacter plasmids.

The primary search for dif modules was performed on the set of plasmids from five permafrost strains from our collection (ED23-35, ED45-23, ED9-5a, EK30A and VS15; Supplementary Table S1) (Petrova et al., 2002; Mindlin et al., 2016). In five fully sequenced genomes, a total of 35 plasmids of various sizes were found. In addition to previously described 8 plasmids (Mindlin et al., 2016, 2018), we have now assembled 27 new plasmids from the same strains (see details in Materials and Methods).

In 10 of the plasmids, sequences homologous to pdif sites were absent or only a single pdif site was detected (Supplementary Table S1). The remaining 25 plasmids with two or more pdif sites were analyzed for the presence of dif modules containing adaptive genes surrounded by two pdif sites (see Materials and Methods). In total, putative dif modules were detected in 12 plasmids; 6 of them contained one module; 4 contained two modules, and 2 contained three different modules (Supplementary Table S2).

In addition to the three modules identified previously (chrAB dif, kup dif, and sulP dif) (Mindlin et al., 2018), we revealed four new putative modules with adaptive genes: terC dif, add dif, ohr dif, and sulP-uspA dif (Figure 3).

In most cases, plasmids containing the same modules originated from different permafrost strains (Supplementary Table S2) suggesting their horizontal transfer. To test this hypothesis, we analyzed the distribution of all modules among plasmids (contigs) of modern Acinetobacter strains belonging to different Acinetobacter species (Supplementary Table S2). Below, we briefly describe characteristics and distribution of each of these modules (Table 3).

We revealed the terC dif module (1221 bp) in four permafrost plasmids, pALWED1.4 from A. lwoffii ED23-35, pALWED2.1 from A. lwoffii ED45-23, pALWED3.6 from A. pseudolwoffii ED9-5a and pALWEK1.1 from A. lwoffii EK30A (Supplementary Table S2). It contained a single orf, identified as terC (978 bp), encoding resistance to tellurium. We also found the terC dif module in two modern plasmids and several contigs from modern strains belonging to different species of Acinetobacter (Supplementary Table S2 and Figure 3).

Up to date, genetic determinants of resistance to tellurium were found in the genomes of different bacteria both in chromosomes and plasmids (Chasteen et al., 2009). We also detected the terC gene in the sequenced chromosomes of Acinetobacter strains, belonging to different species including Acinetobacter baylyi ADP1 [NC_005966.1] widely used in genetic studies.

However, the plasmidic and chromosomal terC genes were related only distantly (42–60% identity at the protein sequence level), and the chromosomal terC gene was not flanked by dif sites.

To determine the functional activity of the terC dif module, we compared the level of tellurium resistance of the permafrost strains of A. lwoffii and A. pseudolwoffii containing the terC dif module with the strain A. lwoffii VS15 that lacked this module in its plasmids. The results of three independent experiments showed that strains ED23-35, ED9-5a and EK30A have a significantly higher resistance level (MIC from 0.04 to 0.1 mg/ml) compared to strain VS15 (MIC < 0.01 mg/ml); whereas strain ED45-23, characterized by very slow growth, does not differ from VS15. Thus, additional studies are needed to confirm the contribution of the terC dif module into the tellurium resistance of corresponding host strains.

Most prokaryotic chromosomes as well as many plasmids contain toxin-antitoxin (TA) systems consisting of a pair of genes that encode 2 components, a stable toxin and its cognate unstable antitoxin. TA systems are also known as addiction modules (Yamaguchi et al., 2011; Schuster and Bertram, 2013). In particular, many type II toxins are mRNA-specific endonucleases that arrest cell growth through RNA cleavage, thus preventing the process of translation (Bertram and Schuster, 2014). We revealed a number of type II TA modules in permafrost plasmids. Amongst them we found two related modules surrounded by pdif sites (both 768 bp long). Both modules contained two genes, one (297 bp) encoding a killer protein and another (291 bp) – an antidote protein. The first module was located in large plasmid pALWED2.1; the second was detected in medium-sized (22771 bp) plasmid pALWED2.3 and large plasmid pALWEK1.1. The difference in nucleotide sequences of the modules was 4%.

We also found variants of both add dif modules in plasmids and contigs of modern strains belonging to different Acinetobacter species (A. lwoffii, A. baumannii, Acinetobacter towneri, Acinetobacter johnsonii) (Supplementary Table S2). In most cases they were almost identical to the basic prototype modules (96–99% nucleotide identity level).

This genetic element contains two genes: a gene encoding organic hydroperoxide resistance, ohr (429 bp) and a second gene ohrR (432 bp) encoding transcription regulator from the MarR family of transcription factors (Sukchawalit et al., 2001; Atichartpongkul et al., 2010).

The ohr dif module was detected in plasmids from three permafrost strains A. lwoffii: pALWED1.2 from strain ED23-35; pALWVS1.1 from strain VS15 and pALWEK1.1 from strain EK30A (identity level 99%). However, the size and the boundaries of this module differed between the plasmids. We therefore analyzed its structure and distribution in more detail.

The ohr/ohrR genes with plasmid localization were first found by Dorsey et al. (2006) in the plasmid pMAC [AY541809.1] from a reference clinical strain A. baumannii ATCC 19606. The authors showed that A. baumannii ATCC 19606 is resistant to the organic peroxide-generating compounds cumene hydroperoxide (CHP) and tert-butyl hydroperoxide (t-BHP). We analyzed the genome of pMAC and detected typical pdif sites on both flanks of the ohr genes suggesting that pMAC contains an ohr dif module. The permafrost and clinical versions of the ohr dif module are overall closely related (96% nucleotide identity) but reveal some specific differences: (i) the ohr module from the modern strain is 5 bp longer than the ancient one (1068 bp vs.1063 bp); (ii) both modules contain identical ohrR genes but their ohrA genes differ by 3%; (iii) region adjacent to the XerD/XerC site contains multiple substitutions. We therefore designated the ancient variant ohr1 dif and the modern variant ohr2 dif and studied the distribution of each of them separately.

Although we did not find the ohr1 dif module in the plasmids of modern strains, it was revealed among unassembled genomic sequences of various Acinetobacter species, in most cases in small-size contigs (Supplementary Table S2). The ohr2 dif module was detected both in plasmids and in contigs from various Acinetobacter species (Supplementary Table S2).

Interestingly, two plasmids (plasmid unnamed 2 [CP027180] and plasmid unnamed 4 [CP027186] from the modern strains of A. baumannii AR_0070 and A. baumannii AR_0052, respectively) contained numerous copies of the ohr 2 module, 14 in the first plasmid and 8 in the second.

The location of both ohr dif modules in various sequence contexts in different plasmids and contigs (Figure 3) strongly suggests their ability to move. Both modules likely originated from a common ancestor and then distributed independently among Acinetobacter species.

To test the functional activity of the ohr1 dif module, we compared resistance of permafrost strains, containing (group 1: ED23-35, VS15 and EK30A) or lacking (group 2: ED45-23 and ED9-5a) this module, to tert-butyl hydroperoxide (t-BHP) (see Materials and Methods). The results of three independent experiments demonstrated a higher level of resistance of group 1 strains (MIC = 0.2–0.4 mM) in comparison with group 2 strains (MIC = 0.1 mM).

The gene sulP encodes sulfate permease (484 aa) that perform transport of inorganic anions into cells (Price et al., 2004). It is a member of the large and ubiquitous family of SulP proteins present in archaea, bacteria, fungi, plants and animals (Alper and Sharma, 2013).

We revealed two different mobile elements containing the sulP gene. The first module sulP dif (2159 bp), first found in our previous work (Mindlin et al., 2018), contained a small orf (198 bp) encoding hypothetical protein in addition to sulP; the second module sulP-uspA dif (2739 bp), detected in this work in the plasmid pALWEK1.10 from A. lwoffii EK30A, contained the uspA gene encoding a universal stress protein in addition to sulP. The sulP genes from the two modules are related (the identity level of 75%). The distribution of both dif modules in modern plasmids (contigs) is presented in Supplementary Table S2 and Figure 3 illustrates the location of the sulP-uspA dif module in different plasmids.

A putative 2549 bp dif module containing the kup gene (1878 bp) encoding a potassium uptake transporter (Zakharyan and Trchounian, 2001) and an orf of unknown function (210 bp) was first revealed in our previous work (Mindlin et al., 2018). In this work we detected almost identical kup dif module in plasmids of three other permafrost strains (Figure 3) and studied its distribution in modern strains. The kup dif modules with identity level of 99% were found in plasmids and contigs from Acinetobacter strains belonging to different species (Supplementary Table S2). Importantly, seven out of 10 modern plasmids/contigs containing this module were isolated from environmental strains. The origin of the host strains for the remaining three plasmids [CP026424.1; APOG01000001.1; JWHB01000051.1] could not be established. Therefore, this module, which likely controls the uptake of potassium, may increase the fitness of Acinetobacter strains living in the environment but not in the clinic.

To reveal whether the presence of multiple dif sites is restricted to Acinetobacter plasmids, we next analyzed the number and location of dif sites in the Acinetobacter chromosomes. Usually a single dif locus involved in the resolution of chromosomal concatenates is located near the chromosome terminus (Carnoy and Roten, 2009). Surprisingly, we found that the Acinetobacter chromosomes are organized in a different way. Besides the main dif site (dif1), we revealed additional dif sites in most completely sequenced circular chromosomes of different species of the Acinetobacter genus. In total, we analyzed 20 sequenced chromosomes of 13 Acinetobacter species, and only one of them did not contain additional dif sites.

As a rule, the number of discovered dif sites varied from one to five and they were located far from dif1 (e.g., A. lwoffii WJ10621 [CM001194.1]; A. lwoffii ZS207 [NZ_CP019143.1]; A. johnsonii XBB1 [CP010350.1]; A. nosocomialis SSA3 [CP020588.1]). Interestingly, we found two Acinetobacter strains that contained multiple additional dif sites in their chromosomes, Acinetobacter indicus SGAir0564 with 14 additional chromosomal dif sites [CP024620.1] and Acinetobacter sp. SWBY1 with 15 additional dif sites [CP026616.1].

It should be mentioned that strains containing multiple chromosomal dif sites were found occasionally when studying the distribution of different dif modules. In particular, the terC dif module, almost identical to that from plasmid pALWED1.4 (99% identity), was revealed in the chromosome of A. indicus SGAir0564 (CP024620.1). It was flanked by typical pdif sites (XerC/D and XerD/C) and was located at 2363397–2364617 bp, far from the main chromosomal dif1 site at position 1526563. Similarly, we found a variant of the add dif module in the chromosome of the Acinetobacter sp strain SWBY1 [CP026616.1].

To get insight into the origin and specific structure of the additional chromosomal dif sites, we compared their sequences with the main dif1 sites from Acinetobacter chromosomes and with pdif sites from A. baumannii and A. lwoffii (see Materials and Methods). In total 109 pdif sites from 23 plasmids of A. lwoffii, 115 pdif sites from 33 plasmids of A. baumannii, 20 main dif1 and 40 additional chromosomal Acinetobacter dif sites were analyzed.

All main features of classical dif sites were revealed in all studied groups (Figure 4). All of them have a length of 28 bp and are composed of two 11 bp regions comprising the binding sites for XerD (more conserved) and XerC (less conserved), separated by a more variable 6 bp central region. The inner parts of the XerC and XerD binding motifs contain inverted repeats forming a palindrome and are highly conserved while their outer regions are more variable (Figure 4).

Comparison of dif sites from different groups revealed that (Figure 4): (i) pdif sites from plasmids of distant species A. baumannii and A. lwoffii (A. pseudolwoffii) are very close to each other in their consensus structure and form two clusters with different consensus sequences of five central nucleotides; (ii) additional chromosomal dif sites differ significantly in their structure from the main chromosomal dif1 site, especially in their central part, while (iii) they are more similar to the plasmid pdif sites from both clusters. It should be mentioned that the structure of the main chromosomal dif1 sites of Acinetobacter corresponds well to the consensus of dif sites from Proteobacteria chromosomes (Carnoy and Roten, 2009). Thus, it can be hypothesized that the plasmid pdif sites and additional chromosome dif sites have originated from a common genetic element, whose identity remains to be established.