The evolution of infectious transmission promotes the persistence of mcr-1 plasmids

ABSTRACT Conjugative plasmids play a vital role in bacterial evolution and promote the spread of antibiotic resistance. They usually cause fitness costs that diminish the growth rates of the host bacteria. Compensatory mutations are known as an effective evolutionary solution to reduce the fitness cost and improve plasmid persistence. However, whether the plasmid transmission by conjugation is sufficient to improve plasmid persistence is debated since it is an inherently costly process. Here, we experimentally evolved an unstable and costly mcr-1 plasmid pHNSHP24 under laboratory conditions and assessed the effects of plasmid cost and transmission on the plasmid maintenance by the plasmid population dynamics model and a plasmid invasion experiment designed to measure the plasmid’s ability to invade a plasmid-free bacterial population. The persistence of pHNSHP24 improved after 36 days evolution due to the plasmid-borne mutation A51G in the 5′UTR of gene traJ. This mutation largely increased the infectious transmission of the evolved plasmid, presumably by impairing the inhibitory effect of FinP on the expression of traJ. We showed that increased conjugation rate of the evolved plasmid could compensate for the plasmid loss. Furthermore, we determined that the evolved high transmissibility had little effect on the mcr-1-deficient ancestral plasmid, implying that high conjugation transfer is vital for maintaining the mcr-1-bearing plasmid. Altogether, our findings emphasized that, besides compensatory evolution that reduces fitness costs, the evolution of infectious transmission can improve the persistence of antibiotic-resistant plasmids, indicating that inhibition of the conjugation process could be useful to combat the spread of antibiotic-resistant plasmids. IMPORTANCE Conjugative plasmids play a key role in the spread of antibiotic resistance, and they are well-adapted to the host bacteria. However, the evolutionary adaptation of plasmid-bacteria associations is not well understood. In this study, we experimentally evolved an unstable colistin resistance (mcr-1) plasmid under laboratory conditions and found that increased conjugation rate was crucial for the persistence of this plasmid. Interestingly, the evolved conjugation was caused by a single-base mutation, which could rescue the unstable plasmid from extinction in bacterial populations. Our findings imply that inhibition of the conjugation process could be necessary for combating the persistence of antibiotic-resistance plasmids.

between bacteria via plasmids is strong evidence of bacterial adaptation, becoming a considerable threat to human health worldwide in recent decades (4)(5)(6). As the horizontal transfer (HT) of resistance genes between microorganisms via plasmids is highly efficient, conjugative plasmids are considered the main drivers of antimicrobial resistance spread among clinically relevant pathogens (3). Thus, the emergence of plasmid-mediated resistance genes, especially those conferring resistance to last-line antimicrobials such as colistin, has caused global concern. The mcr-1 gene confers resistance to colistin by modifying the lipid A of lipopolysaccharide (7)(8)(9)(10)(11). So far, mcr-1 has spread globally and has recently been classified as the highest-risk ARG in an omics-based framework for assessing the health risk of ARGs based on gene mobility, human-associated-enrichment, and host pathogenicity (12). Some resistance plasmids show high persistence even without antibiotic selection, especially F-like plasmids from Enterobacteriaceae (13)(14)(15)(16). The persistence of plasmids in bacterial populations is usually influenced by plasmid stability function (including efficient replication, partition system, and post-segregational killing system), plasmid cost, and conjugational transfer.
Poor plasmid persistence in bacterial populations can be improved by compensatory evolution occurring in the bacterial host, plasmid, or both by reducing the fitness burden (17)(18)(19)(20)(21)(22). Several studies have shown that mutations in host-encoded genes can often reduce plasmid cost and improve persistence (18,19,21,22). For example, chromosomal compensatory mutations in Pseudomonas fluorescens that target the two-component system GacA/GacS or the hypothetical protein PFLU4242 reduce the fitness cost of pQBR plasmids and increase their permissiveness (18,22). Several studies have shown that plasmid cost and persistence can often be improved by mutations in host-enco ded helicases (19,21). These chromosomal mutations compensated for the cost of plasmid carriage by reducing the metabolism burden caused by plasmid replication or transcription of the plasmid-encoded genes. Besides, a poor plasmid-host association can improve due to the evolution of plasmids (17,20). Plasmid persistence can improve after losing costly conjugative transfer genes during long-term experimental evolution (20). In Shewanella oneidensis MR-1, loss of the helicase binding domain (DnaB) in the replication initiation protein encoded by an IncP-1β plasmid reduced the plasmid cost and improved long-term plasmid survival (17).
In addition to the selective benefit of plasmid-encoded genes and compensatory adaptation, plasmid conjugative transfer also influences its persistence. A previous study showed that a high transfer frequency could offset plasmid loss resulting from segrega tional loss and fitness cost (23,24). We recently also found that the master conjugative regulator PixR promotes the persistence of mcr-1-positive IncX4 plasmids by increasing their transfer frequency (25). However, high plasmid transfer rates might impose a fitness burden on the host and slow its growth. In addition, high-throughput analysis of plasmid sequences showed that over half of known plasmids in nature lack a conjugation system (26)(27)(28)(29). Hence, whether conjugation is the principal mechanism for plasmid persistence is debatable.
In this study, we evolved an unstable mcr-1-positive plasmid pHNSHP24 under colistin selection in Escherichia coli C600 for 36 days. Our results demonstrated that plasmidencoded mutations that increase the plasmid transfer rate could help stabilize mcr-1positive plasmids and rescue the costly mcr-1 plasmid from extinction in the bacterial populations.

RESULTS
E. coli C600 harboring pHNSHP24 was obtained by conjugation from E. coli SHP24 isolated from a swine rectal swab sample of the pig farm where we first discovered mcr-1 in 2013 (30). pHNSHP24 is a hybrid between an IncFII plasmid and a phage-like pO111 plasmid carrying a copy of ISApl1-flanked mcr-1. pHNSHP24 carries the whole conjuga tive transfer region of the IncFII plasmid. It contains a putative stbAB operon located in the IncFII plasmid part, known as a partitioning system, and two putative toxin-antitoxin systems, hok/sok and doc/phd, located in the phage-like pO111 part, which may mediate the stable maintenance of plasmid (30). The persistence of pHNSHP24 in E. coli C600 was investigated by serial passaging for 14 days without colistin. Since pHNSHP24 could not persist stably in E. coli C600 (40% plasmid-free cells after 14 days) (Fig. 1), E. coli C600(pHNSHP24) was used to investigate plasmid-host adaptation in the presence of colistin. Here, we chose the transconjugant C600(pHNSHP24) to perform a laboratory evolution experiment instead of the native strain because E. coli SHP24 contains four other plasmids besides the pHNSHP24 (30).
FIG 1 Experimental overview. Three lineages (E1-E3) of the Escherichia coli C600 containing pHNSHP24 plasmid were cultured by serial batch transfer in LB broth and in the presence of colistin. The plasmid persistence was done at day 0 as well as following 7, 14, and 36 days of serial transfer. The E1 lineage evolved for 7, 14, and 36 days was named C600(pHNSHP24-7D), C600(pHNSHP24-14D), and C600(pHNSHP24-36D), respectively. To determine if the improvement of plasmid persistence was due to evolution of the plasmid, the host, or both, the plasmid persistence was measured for all possible host-plasmid permutations: the ancestral host (H A ) containing the ancestral plasmid (P A ); the ancestral host (H A ) containing the evolved plasmid (P E ); the evolved host (H E ) containing the ancestral plasmid (P A ); the evolved host (H E ) containing the evolved plasmid (P E ). The genome sequences of the evolved host-plasmid associations were analyzed to determine the key mutations on plasmids or chromosomes.

Experimental evolution of pHNSHP24 during antibiotic selection in E. coli
To investigate whether plasmid-host adaptation could enhance the persistence of pHNSHP24, we performed a plasmid evolution experiment in liquid medium containing colistin (Fig. 1). E. coli C600 carrying pHNSHP24 was cultured consecutively for 36 days. We monitored plasmid persistence following 7, 14, and 36 days of serial transfer in Luria-Bertani (LB) broth containing colistin. As shown in Fig. 1, the 36-day evolved lineage E1 (E1-36D) showed a pronounced gain in plasmid persistence compared to the non-evolved lineage. No improvement was observed after 7 and 14 days or with the two other evolved lineages. Then, we purified the lineage E1 evolved for 36 days by streaking on LB plates containing colistin and randomly picked three clones (E1-1, E1-2, and E1-3) to determine the plasmid persistence. As shown in Fig. 2A, all three strains exhibited improved plasmid persistence. To determine whether the enhanced plasmid persistence was due to the evolution of the plasmid or host, we transferred evolved pHNSHP24-36D (P E ) from the E1-1 clone to the ancestral host E. coli C600 A (H A ), and the resulting strain was named E. coli C600 A (pHNSHP24-36D)(H A P E ). The plasmid-free evolved strain E1-1 was obtained from the plasmid persistence experiment. Then, the ancestral plasmid pHNSHP24 (P A ) was transferred into the plasmid-free strain E1-1 designated as E. coli C600 E (H E ), and the resulting strain was named E. coli C600 E (pHNSHP24)(H E P A ). Then, we determined the plasmids (P E and P A ) persistence in H A or H E . As shown in Fig. 2B, the evolved plasmid pHNSHP24-36D in both E. coli C600 A and E. coli C600 E shows dramatic improvement in persistence compared to the ancestral plasmid pHNSHP24. In contrast, we did not detect improved persistence of pHNSHP24 or pHNSHP24-36D in E. coli C600 E compared with E. coli C600 A , indicating the evolution occurred in the plasmid, not the chromosome. We also compared the plasmid persistence profiles of H A P A , H E P E , H A P E , and H E P A , using the HT plasmid population dynamics model, as described in previous studies (21,31). We used the Bayesian information criterion (BIC) to reflect the difference in plasmid persistence dynamics between each strain. As shown in Table 1, the plasmid persistence profiles of P E -bearing strains (H A P E , H E P E ) showed a significant difference with P A -bearing strains (H A P A , H E P A ) (the ΔBIC score was −38.6 to −30.17; the more negative the ΔBIC, the larger the difference between strains). However, the plasmid persistence profiles of host H A P A or H A P E were indistinguishable from that of the host H E P A or H E P E , respectively (ΔBIC=19.17 or 19.47), further implying that the improved persistence was attributable to plasmid evolution.

The evolved 5′UTR of traJ is responsible for plasmid persistence improve ments
To examine the underlying mechanism responsible for improving plasmid persistence, we compared the plasmid sequences of pHNSHP24 and pHNSHP24-36D. As shown in Fig. 2C, we observed point mutations in cDmt (cytosine-specific DNA methyltrans ferase gene) and the upstream sequence (A51G) of traJ in pHSNHP24-36D. Besides, we also sequenced the plasmids pHNSHP24-7D and pHNSHP24-14D purified from the clones that evolved for 7 and 14 days in the E1 lineage and found that the two plasmids only contained the C1999T mutation in cDmt (Fig. 2C). To determine whether the mutations in cDmt and traJ were required for increased plasmid persis tence, we further measured the plasmid persistence of pHNSHP24, pHNSHP24-36D, and pHNSHP24-14D in E. coli BW25113. As shown in Fig. 2D, pHNSHP24-36D showed better persistence in BW25113 than pHNSHP24 and pHNSHP24-14D, while pHNSHP24 and pHNSHP24-14D showed similar persistence. These findings indicated that the mutation in cDmt does not affect plasmid persistence as it was the only SNP between pHNSHP24 and pHNSHP24-14D. Likewise, the difference between the persistence ability of pHNSHP24-36D and pHNSHP24-14D suggests that the mutation A51G in the 5′UTR of traJ improves plasmid persistence but whether it is dependent on the presence of the cDmt mutation is unknown.

A point mutation at 5′UTR of traJ increases the plasmid conjugation rate
Given that the mutation A51G in the 5′UTR of traJ was responsible for the improve ment of plasmid persistence, we investigated the effect of this mutation on conjuga tion. Conjugation assay was carried out using E. coli BW25113 containing pHNSHP24 or the evolved plasmids pHNSHP24-14D and pHNSHP24-36D as donors, and E. coli The evolved plasmids from E1-1, E1-2, and E1-3 clones were sequenced, and found that the sequences of each plasmid were identical. This evolved plasmid was named as pHNSHP24-36D. The pHNSHP24-36D plasmid contains C1999T mutation in cDmt gene and A51G in 5′UTR of traJ. The evolved plasmids from the lineage E1 evolved for 7 and 14 days were named as pHNSHP24-7D and pHNSHP24-14D, respectively. The sequences of pHNSHP24-7D and pHNSHP24-14D were identical, and they only contain C1999T mutation in cDmt gene (D) Comparison of the persistence of pHNSHP24, pHNSHP24-36D, and pHNSHP24-14D in E. coli BW25113. Each point represents the mean fraction of plasmid-containing cells (n = 3) and bars represent SD of the three biological replicates.
Research Article mBio BW25113::kan was used as recipient. As shown in Fig. 3A and Table S1, the transfer rate of pHNSHP24-36D increased ~1,600-fold compared with pHNSHP24 (P = 0.031). However, there were no significant differences in the conjugation transfer rate between pHNSHP24 and pHNSHP24-14D (P = 0.4171). In addition, we confirmed that the cDmt gene has little effect on the conjugation of pHNSHP24 and pHNSHP24-36D plasmids ( Fig. 3A and Table  S1). This result suggests that the mutation A51G in the 5′UTR of traJ led to increased transfer rate. To confirm this finding, we constructed pHNSHP24∆traJ and complemen ted traJ (wild type) or mutated traJ (A51G) under the control of its native promoter to test their ability to conjugate. As expected, the complementation of mutated traJ significantly outperformed the transfer rate of traJ (wild type)-complemented pHNSHP24 ∆traJ ( Fig. 3B and Table S1). These results indicate that the point mutation A51G in the 5′UTR of traJ could indeed increase the plasmid conjugation efficiency.
TraJ is an activator of the tra operon (16,(32)(33)(34)(35). To examine the effect of the mutation A51G on the expression level of traJ, we fused the upstream region (−217 to +242) of traJ or mutated traJ with a promoterless lacZ, yielding P traJ -lacZ and P traJ(A51G) -lacZ, respectively. As shown in Fig. 3C, this mutation led to a considerable increase in the β-galactosidase activity, suggesting that this mutation could upregulate the expression of TraJ. Thus, we determined the expression of tra genes in both BW25113(pHNSHP24) and BW25113(pHNSHP24-36D) by qPCR. Expectedly, the transcription level of tra genes, including traY, traA, traC, traG, traD, and traI, in BW25113(pHNSHP24-36D) strongly increased compared with that of BW25113(pHNSHP24) (Fig. 3D). These results confirmed that the mutation A51G in the 5′UTR of traJ increased the plasmid transfer ability by enhancing the expression of traJ and tra genes.
In IncF plasmids, the expression of traJ is negatively regulated by the antisense RNA FinP, which interacts with the leader region of the traJ transcript and prevents the translation of traJ ( Fig. 4A) (36). FinP contains two stem-loop structures, SLI and SLII, which were complementary to the stem-loops of the traJ UTR, SLIc, and SLIIc, respectively. Since the point mutation in the 5′UTR of traJ (position +51) is located at SLIIc (Fig. 4B), we speculated that this mutation could impact the inhibitory effect of FinP. We predicted the structure of SLIIc of pHNSHP24-36D and found that its loop is smaller than that of pHNSHP24 (Fig. 4B). We hypothesized that the reduced loop of SLIIc impairs the complementary pairing between the 5′UTR of traJ mRNA and FinP.
To test the effect of the loop structure of SLIIc on the expression of traJ, we changed the loop size of SLIIc by introducing mutations in the 5′UTR of traJ (+51 or +56) (Fig.  4B) and fused the upstream region (−217 to +242) with promoterless lacZ, yielding P traJ(A51C) -lacZ and P traJ(C56U) -lacZ, respectively. As shown in Fig. 4B, the mutation C56U is predicted to create the same loop size as A51G of pHNSHP24-36D. The β-galactosidase activity of both P traJ(C56U) -lacZ and P traJ(A51G) -lacZ increased significantly compared with P traJ -lacZ(wild-type). Likewise, the mutation A51C is predicted to create the same loop size as in pHNSHP24 and P traJ(A51C) -lacZ shows similar activity to P traJ -lacZ (Fig. 4C). These results suggest that the loop size of SLIIc indeed affects the expression of traJ, and the reduction of the loop size caused by the point mutation at 5′UTR of traJ (+51) probably impaired the inhibitory effect of FinP on traJ. Notably, the P traJ(C56U) -lacZ fusion shows lower activity than P traJ(A51G) -lacZ implying that the stability of SLIIc also affects joint=ΔBIC was used to assess the magnitude of the difference between two plasmid persistence profiles (31). BIC.sep represents the BIC of the model which assumes the two plasmid persistence data are governed by different dynamics. BIC.joint represents the BIC of the "null" model which assumes that the stability dynamics of two data sets are the same. More negative ΔBIC values indicate larger differences between the two plasmid persistence data. b H A P A , the ancestral host harboring the ancestral plasmid; H E P E , the evolved host harboring the evolved plasmid; H A P E , the ancestral host harboring the evolved plasmid; H E P A , the evolved host harboring the ancestral plasmid.
the inhibitory effect of FinP, since the SLIIc harboring A51G, which contains GC base pair, is probably more stable than the SLIIc harboring C56U, which contains AU base pair.

The increased plasmid transfer rate accounts for the improved plasmid persistence
Previous studies have shown that increased plasmid transfer frequency often incurs fitness costs on host bacteria (37,38). We further compared the costs of pHNSHP24 and pHNSHP24-36D by measuring the growth rates of the plasmid-harboring strains relative to the plasmid-free strain E. coli BW25113. As shown in Fig. S1, the growth rates of BW25113 harboring pHNSHP24 and pHNSHP24-36D decreased compared to E. coli BW25113, while the growth rate of BW25113(pHNSHP24-36D) was lower than BW25113(pHNSHP24). Then, we also evaluated the fitness of BW25113(pHNSHP24-36D) relative to BW25113(pHNSHP24) in a competition assay. To facilitate the screening of All plasmid conjugation assays were performed with three biological replicates. One-way analysis of variance was used to determine the statistical differences among groups. The β-Galactosidase assay and qPCR experiment were carried out with three biological replicates, and Student's t test was used for comparison of differences between the two groups.
Research Article mBio BW25113 harboring pHNSHP24-36D in competition cultures, we introduced a kanamycin resistance gene at the cDmt locus since the deletion of cDmt in pHNSHP24-36D did not affect fitness (Fig. S2). We used BW25113(pHNSHP24-36DΔcDmt::kan) instead of BW25113(pHNSHP24-36D) for competition assays, since both strains show comparable fitness (Fig. S2). As expected, BW25113(pHNSHP24-36DΔcDmt::kan) showed a significant reduction in fitness relative to BW25113(pHNSHP24) (Fig. 5A), indicating that the evolved plasmid pHNSHP24-36D indeed imposes a higher cost on host bacteria than pHNSHP24. These findings suggest that the evolved 5′UTR of traJ increased the plasmid conjugation efficiency but also incurred a higher fitness cost on host bacteria. Plasmid persistence usually correlates with segregation loss rate (λ), fitness cost (σ) as well as conjugation rate (γ) (21,31,39). To investigate how these parameters affect the persistence of plasmids before and after laboratory evolution, we used the plasmid population dynamic model to estimate model-based parameters and evaluate the role of these parameters in plasmid persistence. Initially, to determine the best-fitting model, the plasmid persistence profiles were fitted to both segregation and selection (SS) and the HT models as previously described (21). As shown in Table S2, the HT model provided a better fit, suggesting that conjugation was considered important for the persistence of plasmids in this study. Then, these parameters were estimated by the HT model SLIIc on the expression of traJ. Activity of traJ and its derivates (A51C, C56U, and A51G) was monitored from lacZ fusion. The β-Galactosidase assay was performed with three biological replicates. One-way analysis of variance was used to determine the statistical differences among different groups.
Research Article mBio and listed in Table S3. The maximum likelihood estimates for the segregation rate (λ), ranging from 1.72 × 10 −6 to 1.99 × 10 −5 , are negligibly low. The model-based conjugation frequency (γ) of BP E (evolved plasmid pHNSHP24-36D) is more than eight times of BP A (ancestral plasmid pHNSHP24); however, its cost is twofold of BP A . To better evaluate the Research Article mBio role of the model-based conjugation frequency (γ) and cost (σ) on plasmid persistence, we introduced a concept of the persistence threshold γ/θ described by Ponciano et al. (39). θ represents the fraction of the plasmid-carrying cells at which the conjugation frequency is half its maximum. When inequality γ/θ≥1−(1−λ)/2^σ holds, this formula indicates that a high transfer frequency can balance or even offset the frequent loss of plasmids resulting from segregation and fitness cost for the persistence of plasmids in the populations. The γ/θ of BP E satisfied the inequality mentioned above while BP A did not, suggesting that the enhancement of transfer frequency of pHNSHP24-36D compensates for plasmid loss.
We also performed a plasmid invasion experiment to determine the invasion abilities of ancestral and evolved plasmids. The invasion abilities of pHNSHP24-36D and pHNSHP24 were compared individually or in competitive co-cultures. For each plasmid individually, the evolved plasmid pHNSHP24-36D invaded most cells after 24 h (Fig.  5B), indicating that the HT of pHNSHP24-36D was sufficient to offset the segregation loss and fitness cost. In contrast, the ancestral plasmid pHNSHP24 failed to invade the plasmid-free population (Fig. 5C). Given that the deletion of cDmt in pHNSHP24-36D did not affect the plasmid transmission ability and fitness cost ( Fig. 5D; Fig. S2), we used BW25113(pHNSHP24-36DΔcDmt::kan) instead of BW25113(pHNSHP24-36D) for plasmid invasion in competitive co-cultures. As shown in Fig. 5E, pHNSHP24-36DΔcDmt::kan invaded the cell population faster than pHNSHP24, as expected from the results of the persistence assay.
In addition, we investigated the effect of mcr-1 on plasmid persistence since mcr-1 usually exerts a fitness cost to host bacteria. We constructed in-frame dele tion of mcr-1 on pHNSHP24 and pHNSHP24-36D and monitored the persistence of pHNSHP24 ∆mcr-1::kan and pHNSHP24-36D∆mcr-1::kan. Interestingly, the deletion of mcr-1 improved the persistence of pHNSHP24 considerably, even better than the A51G mutation of pHNSHP24-36D, while the deletion of mcr-1 slightly improved the persis tence of pHNSHP24-36D (Fig. 6). Notably, pHNSHP24 ∆mcr-1::kan and pHNSHP24-36D ∆mcr-1::kan showed very similar maintenance abilities. These findings suggest that the poor persistence of pHNSHP24 is partly due to the cost incurred by mcr-1. The evolved 5′UTR of traJ improved the persistence of pHNSHP24 in the presence of mcr-1; however, without mcr-1, it has little effect on plasmid persistence. Altogether, these results confirmed that high transmissibility could offset plasmid loss due to fitness cost

DISCUSSION
The plasmid-mediated dissemination of resistance genes has become a great concern to human health (3,(40)(41)(42). Most antibiotic resistance plasmids not only stably persist in host bacteria for long periods but also confer little fitness cost to host bacteria, implying that these plasmids are well-adapted to the host bacteria (6). However, the evolutionary adaptation of plasmid-bacteria associations is not well understood. Here, we investigated an evolutionary adaptation in laboratory conditions of the persistence of the mcr-1-bearing IncFII-pO111 plasmid pHNSHP24 in E. coli. We observed that the evolved plasmid had improved persistence in E. coli.
In this study, conjugation rate but not fitness cost explained the improved persis tence of the evolved plasmid. Although pHNSHP24 contains two putative TA system, hok/sok and doc/phd, it still displayed poor persistence in the bacterial populations, suggesting that two putative TA modules may be not involved in post-segregational killing mechanism for maintaining this plasmid. The persistence of the evolved plasmid was improved significantly by the point mutation A51G in the 5′UTR of traJ, which boosted the plasmid conjugation rate by ~1,600-fold compared to the ancestral plasmid. However, we observed that the conjugation rate increased at the expense of host fitness. This finding is consistent with the concept that plasmid usually displays a tradeoff between conjugation rate and fitness (37,38). Since fitness costs produced by plasmids often play a key role in the persistence of plasmid-carrying strains, we evaluated the impact of conjugation rates, fitness cost, and loss rate in the maintenance of the evolved plasmid by the plasmid population dynamic HT model. We found that, under laboratory conditions, an increase in conjugation rate improved the persistence of pHNSHP24. The plasmid invasion experiment also showed that the increased transfer rate of pHNSHP24-36D could compensate for the plasmid fitness cost. The evolved plasmid was more stable in host bacteria, presumably because more conjugation meant plasmid-free bacteria were more likely to be reinfected.
Previous observations of plasmid evolution showed that compensatory evolution could increase plasmid persistence significantly, often by reducing the fitness cost of a costly plasmid-bacteria association (18,19,21). Furthermore, compensatory mutations can also target the transfer region and inhibit or abolish conjugation (20,43). Unex pectedly, we observed that the mcr-1 plasmid pHSNHP24 improved its persistence by the evolution of conjugation transfer, even though the evolved plasmid incurred a higher fitness cost than the ancestral plasmid. Deletion of mcr-1 could enhance the plasmid persistence to some extent, suggesting that the plasmid cost on the host partly resulted from mcr-1. These observations are consistent with our previous study, which suggests that efficient conjugative transfer is crucial for mcr-1-bearing IncX4 plasmid to alleviate fitness cost and promote its persistence in the bacterial populations (25). Although the evolved infectious transmission favors the maintenance of the unstable mcr-1-bearing plasmid pHNSHP24, this evolution has little effect on the persistence of mcr-1-deficient plasmid pHNSHP24∆mcr-1. These results reflect that, in addition to low copy number (44), high conjugation rate may be another important contributor to the maintenance of mcr-1 plasmids. A limitation of our study is that the plasmid evolution and persistence experiment were conducted in the laboratory environment which is very different from the natural environment. The plasmid transfer and persistence in the natural microbial communities are very complex. The species diversity and interspecies interactions could impact the plasmid dynamics in these communities (45). For example, bacteriophages and protist predation could limit conjugative plasmid persistence by affecting the microbial community structure (45). Less permissive species may hinder the spread of conjugative plasmids in communities, and diversity in fitness effects of plasmids within different species may facilitate the persistence of costly plasmids in bacterial communities (45). In addition, since the host bacteria used for the evolution of pHNSHP24 plasmid was not the natural host of this plasmid, the difference in genetic background of host bacteria may also influence the evolution of this plasmid.
The FinOP system that regulates the expression of the tra operon in IncF plasmids has been investigated in detail (46). The antisense RNA FinP represses TraJ translation by pairing with its traJ mRNA target, facilitated by FinO (47,48). Pairing between FinP and traJ mRNA is thought to initiate between complementary loops in the two RNAs (49). Here, we show that point mutations located in traJ 5′UTR that reduce the predicted loop size of SLIIc from 6 to 4 nt negatively affect the inhibitory effect of FinP on traJ. This finding is reminiscent of the binding reaction between CopA and CopT RNAs of plasmid R1. Likewise, CopA/CopT-binding process initiates with the formation of a transient "kissing complex" between two RNAs (50). The size alterations of CopA loop II have been shown to strongly affect the binding rates of CopA and CopT (51). Notably, the most efficient binding rates for the two RNAs were obtained with loop II of 6-7 unpaired nucleotides, while small (4 nt) loop II led to significantly decreased binding rates and, hence, the inhibitory effect of CopA on CopT (51). We propose that mutations reducing the loop size of SLIIc impair the binding rates between FinP and 5′UTR of traJ mRNA and therefore decrease the inhibitory effect of FinP on its target. It should be noted that the loop size of the C56U mutant was 4 nt, which is consistent with that of the A51G mutant. However, the C56U mutant shows lower expression level of traJ than that of the A51G mutant. The C56U mutant contains an AU base pair, while the A51G mutant contains a GC base pair that is the only difference between them, indicating that the stability of SLIIc also affects the inhibitory effect of FinP on traJ. In fact, the 5′UTR of traJ from various F-like plasmids contain highly conserved stems of SLIIc structure but various loop regions (52) , suggesting that various loop regions of SLIIc may represent one evolutionary strategy for F-like plasmids transfer. In addition, the amino acid sequence of TraJ from F-like plasmids varies greatly, and the promoter sequences of tra operon which is activated by TraJ are also differential (16). It is worth noting that some F-like plasmids could co-exist in the same host bacterium without interference in regulation of plasmid transfer (16), indicating that F-like plasmids have evolved specialized transfer regulatory systems.
In conclusion, we showed that long-term positive selection on mcr-1-bearing plasmid pHNSHP24 can lead to the evolution of conjugation and an increased conjugation transfer rate could enhance its maintenance in pure culture conditions. The loop size and stability of SLIIc in 5′UTR of traJ can influence the transfer rate of pHNSHP24. The plasmid conjugative transfer is important for maintaining the mcr-1-bearing plasmid. Our findings emphasize that inhibition of the conjugation process could be another approach apart from the plasmid-curing strategy to combat the spread and persistence of antibiotic-resistance plasmids.

Bacterial strains and growth conditions
The bacterial strains used in this study are listed in Table S4. All bacterial strains were cultured in LB broth at 37°C with shaking (180 rpm) or on LB agar plates at 37°C. Strains containing pHSG575 plasmid and its derivatives were cultured in LB broth supplemen ted with chloramphenicol (30 μg/mL), and strains containing pHGR01 plasmid and its derivatives were cultured in LB broth supplemented with kanamycin (30 μg/mL). The strain E. coli BW25113 and transconjugants carrying pHNSHP24 or pHNSHP24-36D were grown in LB broth overnight at 37°C, then diluted 1:100 into 100 mL fresh medium and shaking (180 rpm) at 37°C for 12 h. Bacterial growth was measured OD 600 every hour by Multiskan spectrum microplate spectrophotometer (Thermo Labsystems, Franklin, MA, USA).

Experimental evolution and plasmid persistence experiment
The E. coli C600 harboring pHNSHP24 plasmid (GenBank accession number: CP065023) was cultured serially for 36 days in the presence of colistin (2 μg/mL). The E. coli C600(pHNSHP24) strain was grown in LB broth containing colistin overnight at 37℃, then 3μL culture was transferred into 3 mL LB broth containing colistin and shaken at 37℃ each day. The plasmid persistence was measured following 7, 14, and 36 days of serial transfers. Briefly, the evolved cultures or strains were cultured serially in the absence of colistin for 14 days. The cultures which were passaged for 4, 7, 10, and 14 days were spread on non-selective LB agar plates with proper dilution, then 50 colonies were randomly selected to confirm the presence of plasmid by PCR using primers FrepB-F/ FrepB-R, mcr-1-F/mcr-1-R, and pO111-F/pO111-R (Table S5).

Comparison of plasmid persistence
The plasmid persistence profile describes the plasmid persistence over time (from days 0 to 10). The plasmid persistence profile was analyzed and compared using plasmid population dynamics models as described previously (21,31). The two models named SS model or HT model, describe the plasmid dynamics based on maximum likelihood estimates (MLEs) of the segregation rate (λ), conjugation rate (γ), and cost (σ). The SS model involves only two parameters, λ and σ, but the HT model needs all parameters. Initially, plasmid persistence profiles were fit to both models to find the best-fitting model. The best-fitting model is reflected by the lowest negative log-likelihood score. Then, different profiles were compared with each other using the best-fitting model. The BIC reflects the difference between two persistence profiles (the smaller the ΔBIC values, the greater difference between two plasmid persistence dynamics) (31). Moreover, we use the HT model to estimate plasmid cost, segregational loss frequencies, and conjugation frequency based on plasmid persistence profiles. A formula described by Ponciano et al. was used to reflect the effect of conjugation, cost, and segregation on plasmid persistence (39).
where γ is an asymptotic maximum conjugation frequency within a period of time, and θ represents the fraction of the plasmid-carrying cells when the frequency of conjugations is half the maximum. γ/θ represents the transfer frequency threshold needed to ensure the persistence of plasmids. λ is segregational loss frequency and σ is plasmid cost. This formula states that plasmid loss due to segregation and cost must be balanced by a high transfer frequency for the plasmids to persist in the populations. It should be noted that γ described here is a model-based conjugation transfer frequency, and it is not the same as the experimental estimated transfer rate γ estimated by the conjugation experiment based on the Simonsen method. All these analyses were conducted using the R package "StabilityToolkit" (https://github.com/jmponciano/Stabi lityToolkit/blob/master/RunningStabToolsPack.zip) created by Ponciano et al. (39).

Plasmid sequencing and analysis
The whole genomes of the clones that evolved for 36 days (E1-1, E1-2, and E1-3) and clones that evolved for 7 and 14 days in E1 lineage were extracted and subjected to perform whole-genome sequencing (Illumina, San Diego, CA, USA) and Nanopore sequencing (Oxford Nanopore Technologies, UK). Nanopore reads and Illumina reads were combined to produce a de novo hybrid assembly using Unicycler version 0.4.3. The plasmid sequence comparison was performed using BLAST.
The kanamycin-resistant gene was inserted in the truncated lacZ gene of E. coli BW25113 to construct the kanamycin-resistant E. coli BW25113::kan using λ Red recombination system. The kanamycin-resistant gene fragment with homology extension was cloned from the pKD4 plasmid using primers insert-kan-F/insert-kan-R. The mutant was verified by PCR using primers check-kan-F/check-kan-R.

Bacterial conjugation assays
The conjugation experiments were set up based on the Simonsen method (54) to estimate the plasmid conjugation rate (γ). All plasmid conjugation assays were performed with three biological replicates. The colistin-resistant strains BW25113(pHNSHP24) and its derivatives were used as donors (D), and the kanamycinresistant E. coli BW25113::kan was used as recipient (R). Thus, donors (D), recipients (R), and transconjugants (T) could be estimated by selective LB plates containing colistin (2 μg/mL) and/or kanamycin (30 μg/mL). The donor and recipient were individually cultured in 5 mL LB broth overnight at 37℃. The overnight cultures were adjusted to the same density (OD 600 =1), and 3 µL of each culture was transferred into tubes containing 3 mL LB broth. At the start point of mating, 100 µL mixture was sampled to quantify the total initial cell densities (N 0 ). The mating cultures were incubated at 37℃ with shaking for 12 h. During the exponential growth phase, the optical density was measured at t b and t a to calculate the growth rate of the mating cultures, using the following formula (54): where ψ is the growth rate of the mating cultures, OD b and OD a represent the optical density at t b and t a , respectively. This model assumes that donors (D), recipients (R), and transconjugants (T) have the same growth rate. However, if they do not have the same growth dynamics, it could lead to biased estimates (54,55).

Competition experiments in vitro and plasmid invasion assays
Competition experiments were used to measure the relative fitness of strains harboring pHNSHP24 or its derivatives. The overnight cultures of the two strains were diluted 1:1,000 in LB broth and mixed at a 1:1 ratio. The mixture was incubated at 37°C with shaking (180 rpm) for 24 h and then the mixed population was again 1:1,000 diluted into fresh LB broth. This procedure was repeated until the competition experiment lasted for 4 days. The samples of each competitive mixture with proper dilution were spread on LB agar every 24 h. The properly diluted cultures were plated to antibiotic-free, kanamycin-containing, or colistin-containing LB agar plates. The formula RF= log 10 S1 dtlog 10 S1 d0 )/(log 10 S2 dt -log 10 S2 d0 ) was used to calculate relative fitness (RF), where RF is the relative fitness of S1 strain (compared to S2 strain), S1 dt and S1 d0 represent the densities of cells (CFU/mL) of S1 and the end and beginning of the competition, and S2 dt and S2 d0 represent the densities of cells of S2 and the end and beginning of the competition. All competition experiments were carried out with three biological replicates.
Plasmid invasion assays were performed according to the method described in the previous study (56). All plasmid invasion assays were performed with three biologi cal replicates. Overnight cultures of BW25113 were diluted 1:10 into 2 mL LB broth and mixed with 1:1,000 dilutions of BW25113(pHNSHP24), BW25113(pHNSHP24-36D), or BW25113(pHNSHP24-36DΔcDmt::kan) overnight cultures, respectively. The 2 mL mixed cultures grew in a 50-mL tube at 37℃ with shaking (45 rpm), and diluted 1:100 into fresh media every 24 h for 96 h. Viable counts were performed at 24, 48, 72, and 96 h. For each time point, dilutions of the cultures were plated to non-selective, colistin-containing and kanamycin-containing media. Cells containing pHNSHP24 or pHNSHP24-36D grew on non-selective and colistin-containing plates but were killed by kanamycin-containing media. Cells containing pHNSHP24-36DΔcDmt::kan grew on all plates. Plasmid-free cells were killed by both antibiotics. Therefore, the number of cells containing pHNSHP24, pHNSHP24-36D, and pHNSHP24-36DΔcDmt::kan were counted from colistin-containing plates, colistin-containing plates, and kana mycin-containing plates, respectively. The number of plasmid-free cells BW25113 was calculated by the following formula: BW25113 (plasmid-free) cells=antibioticfree plate counts minus colistin-containing/kanamycin-containing plate counts. The experimental procedure for a mixture containing BW25113, BW25113(pHNSHP24), and BW25113(pHNSHP24-36DΔcDmt::kan) is consistent with the steps described above. The cell numbers of three strains were calculated by the following formula: BW25113 (plasmid-free) cells=antibiotic-free plate counts minus colistin-containing plate counts, BW25113(pHNSHP24) cells=colistin-containing plate counts minus kanamycin-contain ing plate counts, BW25113(pHNSHP24)-36DΔcDmt::kan cells=kanamycin-containing plate counts.

RNA extraction, cDNA synthesis, and qPCR
Equal numbers of bacteria during the log phase of growth were collected and total RNA was extracted using E.Z.N.A. Bacterial RNA Kit (OMEGA BIO-TEK, USA), according to the product manual. Then, the integrity of total RNA was confirmed by agarose gel electrophoresis and quantified by NanoDrop 2000c. The RNA samples with 23S/16S rRNA ratio≈2.0 and the A 260 /A 280 between 1.8 and 2.2 were used for cDNA synthesis. About 1 µg total RNA was used for reverse transcription using Goldenstar RT6 cDNA Synthesis Mix (TsingKe Biotech, China), according to the production protocol. qPCR was performed using the TB Green Premix Ex Taq II reagent (TaKaRa, Japan) and Bio-Rad CFX96 Real-Time PCR Detection System (Bio-Rad, USA). The reaction mixture contained 10 µL of TB Green Premix Ex Taq II, 1 µL of cDNA (diluted 1:10), 0.5 µL of each primer (10 µM), and 8 µL of double distilled water. The amplification was performed as the following conditions: 95℃ for 5 min, 40 cycles of 95℃, 30 s; 60℃, 30s; 72℃, 20s, followed by the melting reaction from 65℃ to 95℃. The relative gene expression levels were calibrated by the 2 −ΔΔCt method and gapA gene was used as reference gene. The tenfold gradient-diluted cDNA mixture from all samples was amplified for construction of standard curve. The standard curve was used for calculating the amplification efficiency (E) and correlation coefficient values (R 2 ) of the primers (57). The amplification efficiency (E) for each pair of primers ranges from 90% to 110%, and the correlation coefficient values (R 2 ) range from 0.98 to 1. The relative expression level of genes was obtained by the 2 −ΔΔCt method (58). All experiments were carried out in three biological replicates and two technical replicates.

β-Galactosidase assay
Cells carrying lacZ fusion vectors were cultured in LB broth containing kanamycin overnight at 37℃. Overnight cultures diluted 1:10 into fresh media with shaking (180 rpm) at 37℃ for 4 h. Cells were collected and resuspended in lysis buffer (50 mM Tris-HCl pH 7.3, 1 mM DTT, 5% glycerol, and 1 mM EDTA) and lysed by sonication. The lysates were centrifuged at 12,000× g for 10 min at 4℃ to obtain supernatants, adding 4 μL of the supernatant to Bradford Protein Assay Kit (TaKaRa, Japan) to determine protein concentration. About 50 μL of the supernatant was taken to mix with 950 μL of the lysis buffer. Whereafter, recording began immediately after the addition of 0.2 mL of o-nitrophenyl-β-D-galactopyranoside (4 mg/L), with all reactions performed at room temperature. Reactions were terminated when the color of mixtures turned yellow by the addition of 500 μL of 1 M Na 2 CO 3 . The optical density at 420 nm (OD 420 ) was measured by Multiskan spectrum microplate spectrophotometer (Thermo Labsystems, Franklin, MA, USA). The units of activity were calculated as previously described (59). The authors declare no competing interests.

DATA AVAILABILITY STATEMENT
The sequence data in this study have been submitted to the Sequence Read Archive (SRA) in National Center for Biotechnology Information under accession number PRJNA936493. The authors declare that all data supporting the findings of this study are available within the article and the Supplementary data.