Aquaculture sewage samples were collected in Shandong Province, China, and spread onto Luria–Bertani (LB)-agar plates (0.5% yeast extract, 1% tryptone, 1% sodium chloride, 2% agar) containing 8 mg/L PMB (Sigma Co. Shanghai, China) and then incubated at 28°C for 24 h. All colonies with different phenotype on the plates were selected and cultivated three consecutive times on LB-agar with PMB to obtain single colonies. After further purification, one single colony, named PL22-22A, was selected and grown in pure culture, then 16 S rDNA was amplified by PCR and then sequenced to verify the purity. Antimicrobial susceptibility tests were performed to determine the minimum inhibitory concentrations (MICs) of nine antibiotics, including meropenem, cefixime, tetracycline, ciprofloxacin, florfenicol, amikacin, sulfamethoxazole, PMB, and CST, toward PL22-22A. Data were analyzed based on breakpoints defined by the Clinical and Laboratory Standards Institute (CLSI, 2020).

The genome sequence of Chryseobacterium sp. PL22-22A was sequenced by using the Nanopore and BGISEQ-500 platforms at BGI Co., Ltd. (Wuhan, China). The PL22-22A genome was assembled by glimmer3¹ with hidden Markov models (Tatusova et al., 2016). Genome annotation was performed using the Prokaryotic Genome Annotation Pipeline of NCBI.² Genome alignment was performed by the PATRIC server (Brettin et al., 2015). A phylogenetic tree, including PL22-22A, was produced by using neighbor-joining algorithms in Molecular Evolutionary Genetics Analysis 7 (MEGA 7) software and 16S rDNA gene sequences (Jeong et al., 2016).

Antibiotic resistance genes were analyzed by RAST and BLASTp based on the core dataset in the Antibiotic Resistance Genes Database (Liu and Pop, 2009). Multisequence comparison was carried out using Clustal Omega (Madeira et al., 2019) and ESPript (Robert and Gouet, 2014) software.

A fragment of the promoter sequence of the ampicillin-resistance gene (Pamp^r) from pMD-18 T with a KpnI site at the 5′-end was amplified using primers Pamp-F3 and Pamp-R3. The entire target gene cprB was amplified from genomic DNA of PL22-22A by PCR using primers CprB-F3 and CprB-R3. cprA and cprC with an XbaI site at the 3′-end were amplified by PCR using primers CprA-F3 and CprC-R3 from genomic DNA of PL22-22A. These fragments were fused to obtain fragment CdABC, which was digested with XbaI and KpnI and inserted into vector pMD18-T digested with XbaI and KpnI. The resulting recombinant plasmid pMD-CdABC was transformed into Escherichia coli DH5α, and this strain was named E. coli EcABC3. E. coli containing cprB (E. coli EcB1), cprA (E. coli EcA1), or cprC (E. coli EcC1), respectively, was also prepared. Combinations of two genes cprB + cprA (E. coli EcAB), cprB + cprC (E. coli EcBC), and cprA + cprC (E. coli EcAC) were constructed by using the same process as described above. E. coli DH5α and DH5α harboring pMD18-T vector (strain DT) were used as controls. Primer sequences are listed in Supplementary Table S1.

The MICs of PMB and CST for the various E. coli strains were determined in the presence of PMB and CST at 28°C for 24 h. MICs of PMB and CST in the presence of efflux pump inhibitors (EPIs) were also determined. Verapamil, carbonyl cyanide m-chlorophenyl hydrazone (CCCP), phenylalanine-arginine β-naphthylamide dihydrochloride (PAβN), reserpine (RES), and baicalin (BAC) were used as EPIs at final concentrations of 8.0, 0.1, 8.0, 8.0, and 8.0 mg/L, respectively. As control, MICs of BAC were also determined to check the viability of the bacteria as above. Trypan blue staining was operated when bacteria grow into OD₆₀₀ = 0.8 and microscope direct counting method used to evaluate live/dead ratio. This experiment was performed three times independently. An aliquot of the initial inoculum for the MIC plate was similarly processed. The cells were resuspended in fresh media, plated onto LB, and the colonies enumerated after incubating for 24 h at 37°C. The MBC is defined as the first drug dilution which resulted in a 99.9% decrease from the initial bacterial titer of the starting inoculum or no growth. MBC/MIC ratio with tolerance defined by a ratio ≥ 32.

EC₅₀ values of PMB and CST were determined according to growth inhibition. Briefly, 10 μL of bacterial suspension was added to 5 mL LB medium containing diluted concentrations of PMB and CST. OD₆₀₀ values of the tested suspensions were measured when the control suspensions increased to OD₆₀₀ = 1.0. The log of percentage inhibition based on OD₆₀₀ values were regressed on the log of compound concentrations, and EC₅₀ values were calculated. This experiment was performed three times independently. To assess the effect of EPIs on bacterial growth, 1 mL samples were taken from each bottle at selected time intervals (0, 1, 2, 3, 4, 6, 8, 10, 12, 14, 16, 20 and 24 h after the start of the experiment) and serial 10-fold dilutions were plated in triplicate onto LB agar plates. The numbers of cfu were counted after incubation for 24 h at 37°C.

Transmembrane (TM) regions of proteins were identified using the TMHMM 2.0 server.³ Signal peptides were predicted using SignalP-6.0.⁴ Protein homology modeling and molecular docking analyses were performed as previously described (Zong et al., 2020). The homology model constructed for the selected MFS transporter was analyzed using Rosetta software (Leman et al., 2020) and Discovery Studio (Biovia, 2017) by using EmrAB. Hydrophobic surface features were analyzed using Discovery Studio (Biovia, 2017). The structure of polymyxins, for application as ligands, was obtained from Chemspider.⁵ The binding of CprB with PMB and CST was modeled using the CDOCKER protocol of Discovery Studio 2.0 (Biovia, 2017). Chemical bonds between CprB and polymyxins were demonstrated by 3D and 2D methods. Allosteric sites of CprB were analyzed by using AlloSite 2.0 (Huang et al., 2013).

Strain PL22-22A was isolated from aquaculture sewage under selection with 8 mg/L PMB. MIC analysis of different kinds of antibiotic, including a series of β-lactam, fluoroquinolone, tetracycline, chloramphenicol, aminoglycoside, sulfonamide, and glycopeptide antibiotics, revealed that PL22-22A is a multidrug-resistant strain (Supplementary Table S2). Interestingly, strain PL22-22A showed notable resistance to PMB (MIC 96 mg/L) and CST (MIC 128 mg/L).

Based on phylogeny of 16S rDNA genes, PL22-22A was identified as Chryseobacterium sp. (Figure 1A), which is an opportunistic pathogen. Chryseobacterium sp. PL22-22A harbors a 4,858,345 bp chromosome with 36.12 mol% G + C content (Figure 1B). The complete genome sequence was annotated; 857 functional proteins were assigned Enzyme Commission numbers, and 746 were assigned Gene Ontology terms (Supplementary Table S3). Subsystem analysis of PATRIC annotations indicated that strain PL22-22A contained 88 stress response-, defense-, and virulence-related genes, and 88 membrane transport-related genes (Figure 1C).

To investigate the mechanisms of multidrug resistance, antimicrobial resistance genes were identified in the genome. Genes that are responsible for resistance to β-lactams and tetracycline, including those encoding β-lactamase, metallo-hydrolase, and TetX, were identified (Supplementary Table S4). In addition, a considerable number of efflux pump-related genes were identified; MFS and ATP-binding cassette transporters were prominent classes, with 48 and 57 related genes (Supplementary Table S4), respectively. However, genes that were obviously responsible for polymyxin resistance, such as the lipopolysaccharide modification gene pmrK, which was reported to be the main mechanism of polymyxin resistance in Gram-negative bacteria (Moffatt et al., 2019), were not found in the genome.

In the Chryseobacterium sp. PL22-22A chromosome, the only three tripartite MFS efflux system genes, including PFY10_00045 (encoding a TolC family protein), PFY10_00050 (encoding a membrane fusion protein), and PFY10_00055 (encoding an inner membrane protein), were identified in a single operon. The components of this system PFY10_00055, PFY10_00050, and PFY10_00045, showed low identities (21.16–26.97%) with currently known MFS tripartite transporters, such as the three-component multidrug MFS-type efflux pumps EmrAB-TolC and EmrKY-TolC in E. coli (Supplementary Figures S1–S3). These cpr (Chryseobacterium Polymyxins Resistance) proteins in Chryseobacterium sp. PL22-22A were named CprABC.

Potential functions of Chryseobacterium sp. PL22-22A CprABC in multidrug resistance were analyzed by expression of the genes in E. coli. MICs of nine antibiotics toward E. coli EcABC3 (containing cprA, cprB, and cprA) and EcAB (containing cprA and cprB) were tested. Compared with the control strains DH5α and DT (MICs <2 mg/L), EcABC3 (containing all three CprAB-TolC-encoding genes) gained resistance to PMB and CST (MICs 64 mg/L and 48 mg/L, respectively) (Figure 2A). This strain showed no resistance to the other seven tested antibiotics (meropenem, cefixime, tetracycline, ciprofloxacin, florfenicol, amikacin and sulfamethoxazole) (Supplementary Table S2). These data indicate that the CprABC-encoding genes confer polymyxin-specific resistance.

Expression of the genes for incomplete transporter systems (i.e., CprAB of CprABC) gave different results. In the EcAB-expressing E. coli strain, PMB resistance (MIC 48 mg/L) and CST resistance (MIC 32 mg/L) increased compared with those of control strains DH5α and DT. Thus, in the presence of cprB, E. coli can gain resistance to PMB and CST.

The EC₅₀ of PMB and CST were significantly lower for E. coli DH5α (0.411 ± 0.013 and 0.534 ± 0.02 mg/L respectively) and DH5α with pMD-18 T (0.455 ± 0.047 and 0.509 ± 0.02 mg/L respectively), compared with Chryseobacterium sp. PL22-22A (37.35 ± 1.102 mg/L and 61.61 ± 1.053 mg/L respectively), indicating that a lower PMB and CST concentrations are needed in DH5α to reach 50% of the maximum effect. While, after expressing cprA, cprB or cprABC in E. coli, the EC₅₀ of PMB were significantly raised to 19.74 ± 1.103 and 22.85 ± 1.153, the EC50 of CST raised to 11.99 ± 1.063 and 15.29 ± 1.127 (Figure 2B). EC₅₀/MIC showed the consistent results (Supplementary Table S5). This confirmed that MFS tripartite transporter system CprABC contributes to resistance to polymyxins.

CprA is a membrane fusion protein of 370 amino acid residues and shorter than EmrA and EmrK (390 and 387 amino acid residues, respectively) (Supplementary Figure S1). CprA showed 26.97 and 24.56% sequence identity with EmrA and EmrK, form from E. coli, respectively. Similarly to EmrA, CprA has an N-terminal cytoplasmic domain preceding a TM helix (Figure 3A). The N-terminal TM helix (Hf) is directly attached to the β-barrel domain in CprA. This helix may lead to localization in the cell membrane and assembly as an inner membrane (IM) component of the tripartite MFS system.

CprC of CprABC in Chryseobacterium sp. PL22-22A is a 370-amino acid protein. Compared with TolC in the tripartite multidrug efflux proteins AcrAB-TolC (PDB: 2VDD) and MacAB-TolC (PDB: 5NIK), CprC is shorter at the C-terminus (Supplementary Figure S1). The shortened C-terminal fragment of CprC was constituted by a minor α-helix and a β-sheet, compared with the equatorial domains of TolCs in AcrAB-TolC and MacAB-TolC (Figure 3B). The CprC monomer projects across the intermembrane periplasmic space and is anchored in the OM by a contiguous barrel domain (Figure 3B).

The IM component of the tripartite MFS system in Chryseobacterium sp. PL22-22A, CprB, consists of 530 amino acid residues; it is thus larger than the tripartite MFS system IM proteins EmrB and EmrY (both 512 amino acid residues). Additional sequence, which is absent from EmrB and EmrY, is located at the C-terminus of CprB (from G489-L530, this sequence is named CprBc hereafter) (Supplementary Figure S3). Homology modeling showed that CprB contains 14 putative TM helices, of which helices H1–H12 form two domains (the C-domain and N-domain) that were observed previously for MFS transporters such as PepTSo (Newstead et al., 2011) and EmrB (Yousefian et al., 2021) (Figure 3C, left). As in previous MFS transporter structures, the N- and C-terminal six-helix bundles, formed by helices H1–H6 and H7–H12 respectively, come together to form a “V”-shaped transporter, related by a pseudo two-fold symmetry axis running perpendicular to the membrane plane. Similarly to PepTSo, CprB has two additional TM helices, HA and HB, which are inserted into the cytoplasmic loop connecting the N- and C-terminal bundles (Figure 3C, right).

In Gram-negative bacteria, tripartite efflux pumps span both the inner and outer membranes, providing a continuous seal for drugs to bypass the periplasm. To analyze the details of polymyxin efflux mediated by CprABC, this tripartite transporter system was modeled and then assembled by using SWISS-MODEL and Discovery Studio software. Based on EmrAB from E. coli (Tanabe et al., 2009), membrane fusion component CprB is modeled as a homohexamer, CprC as a typical homotrimer, and IM component CprB as an MFS-type monomer (Figure 4A). CprABC is a 303.7-Å long transporter complex. A slice through the model shows the continuous conduit that runs from the CprB funnel through the CprB porin domain (Figure 4B). In this conduit, there were three conspicuous domains, one in each component of CprABC, crossing the channel-like gates; we named these gates G, F, and T, respectively (Figure 4B).

In this model, CprC has a typical closed state of the entrance, which is constituted by 12 tunnel-forming α-helices. This entrance must be opened for substrates to access the exit duct (Andersen et al., 2002). Similar to TolC in E. coli (Andersen et al., 2002), three inner coiled coils (comprising helices H7 and H8), and also gate T in CprABC, are proposed to be allosteric for tunnel opening (Figure 4C). Because of the homologous polymerization, homohexamer CprA presents a loop region in the middle-upper part (Figure 4D), which is different compared with the model of the CprA monomer. Gate F in CprA is composed of four inner loops (loops 1–4) located in that loop region. Helices α2, α4, and α11 in CprB constitute a narrow entrance, gate G (Figure 4E), and may be essential for the transition between the inward and outward conformations of the MFS-type transporter (Jiang et al., 2013).

In tripartite efflux systems such as EmrAB-TolC and EmrKY-TolC, IM components were a key factor in the efflux of free fatty acids (Li et al., 2015). In the CprABC tripartite efflux system, MFS-type IM proteins CprB has two adjacent binding sites that combine with PMB and CST, respectively, in its intermediary channel (Figure 5A), although their binding amino acids are different. Y34, T35, N56, and F284 seem to be essential amino acids for combination with PMB and CST (Figure 5A). Redundant sequence from G487 to L530, i.e., CprBc, forms a specific α-helix fragment of CprB. Its function was unknown. Moreover, we found that CprBc contains an arginine (R)-rich domain and a glycine (G)-rich domain, which is important for interaction between AcrZ and AcrB that in RND-type AcrAB-TolC complex (Newstead et al., 2011) (Figure 5B). The G-rich domain of CprBc is at its N-terminus, and the R-rich domain is in the middle; this contrasts with AcrZ, in which the G-rich domain is C-terminal, and there is no R-rich domain (Figure 5C). Considering that AcrZ enhanced the ability of the AcrAB-TolC pump to export certain classes of substrates (Hobbsa et al., 2012), we investigated the function of CprBc in substrate selectivity.

CprB was truncated by deletion of CprBc (M1-F487 were retained) and this incomplete CprB was expressed in E. coli. EcAB_dC3 (containing the CprABC-encoding genes with truncated cprB) was sensitive to PMB and CST (MIC <2 mg/L). This sensitivity was the same as that of the control strains E. coli DH5α and DT. In comparison, EcABC3, which contains the complete CprABC encoding genes, was resistant to PMB and CST (MICs 64 and 48 mg/L, respectively) (Figure 5D). A similar phenomenon was observed when truncated cprB was expressed alone in strain EcB_d1, compared with EcB1 (containing complete cprB) (Figure 5D). Loss of resistance to PMB and CST showed that CprBc activates CprB to excrete PMB and CST.

EPIs are molecules that inhibit efflux pumps by one or more mechanisms, leading to inactive drug transport (Sharma et al., 2019). MIC analysis showed that the typical EPIs verapamil, CCCP, PaβN, and reserpine had no or only slight decreasing effects on PMB and CST resistance in EcG1 strain. The MICs toward PMB and CST showed no significant difference in the presence and absence of CCCP (24.0 mg/L). Although verapamil, PaβN, and reserpine decreased the MIC of PMB from 24.0 mg/L to 16.0 mg/L, strain EcB1 retained resistance to polymyxins (Figure 6A). BAC (CAS: 21967-41-9) is a potential flavonoid EPI screened from a Traditional Chinese Medicine Active Compound Library that was proven to be an MFS transporter EPI in Brevundimonas sp. M20 (unpublished data). Here, the effect of BAC on CprB was investigated. At the concentration of 8 mg/L, BAC showed no bacteriostatic activity (Supplementary Figure S4) for strain EcG1, while, it decreased the MIC of PMB and CST to <2.0 mg/L for both drugs (Figure 6A). MBC result also showed a significant reduction in the present of BAC (<2 mg/L) and BAC restored the sensitivity for PMB and CST. While other traditional EPIs showed no equivalent effects (Figure 6B) and the strain maintained tolerance. The minimum inhibitory/bactericidal concentration (MIC/MBC) ratio for polymyxins confirmed that.

Log-linear growth rates in the EPIs-free control showed no significant difference for EcB1 strain in the traditional EPIs (CCCP, verapamil, PaβN, and reserpine) as determined. While, in the presence of BAC, cfu of EcG1 for PMB (Figure 6C) and CET (Figure 6D) were significantly reduced. Considering low doses of BAC (8 mg/L) has no bacteriostasis but could deduce MIC for more than 8-fold (Figure 6A) at this concentration. These results indicated that polymyxins and BAC reveal a synergistic effect in terms of bacterial resistance.

MFS transporters pump their substrates based on conformational changes between the outward and inward conformations (Reddy et al., 2012; Yan, 2015). According to analysis using AlloSite software, CprB was predicted to have four allosteric sites (AS1–AS4) in the protein (Figure 7A). AS1 was located at the top of the central TM channel of CprB (adjacent to the periplasmic space) (Figure 7A, blue). AS2 was predicted to be adjacent to cytoplasmic side and was beneath AS1 (Figure 7A, orange). AS1 and AS2 were surrounded by the channel-forming helices H1, H2, H4, H5, H7, H8, H11, and H12. AS3 and AS4 were located outside of the substrate channel (Figure 7A, green and red), between Ha and Hb (Figure 7A), whose function is still unclear in 14-TM-helix-containing MFS transporters (Newstead et al., 2011). This result indicated a possibility of allosteric functions of Ha and Hb in the novel MFS-type pump CprB.

To investigate the mechanism of inhibition of transport by flavonoid compound BAC, it was docked with CprB. Three BAC-binding sites (BBSs) were predicted in CprB, near AS1, AS3, and AS4, respectively (Figure 7B). It was predicted that ten, seven, and four amino residues bind with BAC in BBS1, BBS2, and BBS3, respectively. Q151 and N56 in BBS1 site, with are also in AS1 site, may play a crucial role. Most of the binding was by hydrophobic amino residues via π-alkyl and unfavorable bump interactions (Figure 7B). Thus, BAC may act as an allosteric inhibitor-type EPI. Furthermore, BBS1 is located in the central TM channel (Figure 7B); PMB and CST must pass this when they are pumped-out. Thus, in addition to allosteric inhibition, BAC appears to directly obstruct the passage of substrate through the transporter. This competitive effect caused a dysfunction for CprABC pump out polymyxins and resulted in a polymyxins sensitive.