Antimicrobial susceptibility profiles and tentative epidemiological cutoff values of Legionella pneumophila from environmental water and soil sources in China

Legionnaires’ disease (LD), caused by Legionella, including the most prevalent Legionella pneumophila, has been treated primarily with antibiotics. Environmental water and soil are the reservoirs for L. pneumophila. Studying antimicrobial susceptibility using a large number of isolates from various environmental sources and regions could provide an unbiased result. In the present study, antimicrobial susceptibility of 1464 environmental L. pneumophila isolates that were derived from various environmental water and soil sources of 12 cities in China to rifampin (RIF), erythromycin (ERY), clarithromycin (CLA), azithromycin (AZI), ciprofloxacin (CIP), moxifloxacin (MOX), levofloxacin (LEV), and doxycycline (DOX) was investigated, and minimum inhibitory concentration (MIC) data were obtained. We show that regarding macrolides, ERY was least active (MIC90 = 0.5 mg/L), while CLA was most active (MIC90 = 0.063 mg/L). A total of three fluoroquinolones have similar MICs on L. pneumophila. Among these antimicrobials, RIF was the most active agent, while DOX was the most inactive one. We observed different susceptibility profiles between serogroup 1 (sg1) and sg2-15 or between water and soil isolates from different regions. The ECOFFs were ERY and AZI (0.5 mg/L), RIF (0.002 mg/L), CIP, CLA and MOX (0.125 mg/L), LEV (0.063 mg/), and DOX (32 mg/L). Overall, two fluoroquinolone-resistant environmental isolates (0.14%) were first documented based on the wild-type MIC distribution. Not all azithromycin-resistant isolates (44/46, 95.65%) harbored the lpeAB efflux pump. The MICs of the ERY and CLA on the lpeAB + isolates were not elevated. These results suggested that the lpeAB efflux pump might be only responsible for AZI resistance, and undiscovered AZI-specific resistant mechanisms exist in L. pneumophila. Based on the big MIC data obtained in the present study, the same defense strategies, particularly against both CLA and RIF, may exist in L. pneumophila. The results determined in our study will guide further research on antimicrobial resistance mechanisms of L. pneumophila and could be used as a reference for setting clinical breakpoints and discovering antimicrobial-resistant isolates in the clinic, contributing to the antibiotic choice in the treatment of LD.


Introduction
Legionnaires' disease (LD), which was first identified in 1977, manifests as atypical pneumonia and can evoke severe disease and death (Fields et al., 2002). Legionella pneumophila (L. pneumophila) is the most common causative agent of LD. It is a Gram-negative, facultative intracellular opportunistic pathogen of humans and is responsible for both sporadic and epidemic community-acquired and hospital-acquired cases of LD. Legionella pneumophila is ubiquitously present in aquatic environments, including moist soil and water systems, where aquatic protozoa act as its natural hosts (Lau and Ashbolt, 2009). The co-evolution and everlasting pathogen-host interactions make L. pneumophila have the ability to infect metazoans and enable them to replicate within accidental hosts, such as lung macrophages and epithelial cells of humans, and then cause LD (Gomez-Valero and Buchrieser, Park et al., 2020). Aquatic environments, such as rivers, lakes, ponds, and garden soils, are the shelter and reservoir for L. pneumophila and from where it can colonize man-made environments, such as cooling towers and hot water systems, and then spread in aerosols, infecting susceptible persons (Zhan et al., 2010). Direct contact with L. pneumophila-contaminated water or dust is another means of infection. L. pneumophila is the first identified agent for LD, which accounts for approximately 91.5% of the LD cases, of which serogroup 1 (LP sg1) appears to be responsible for over 84% of diagnosed cases and 80-90% of community-acquired LD (Yu et al., 2002). Almost all human infections are due to environmental Legionella contamination, and only one case of probable human-tohuman transmission has been reported, suggesting that LD is an environment-borne disease and L. pneumophila, which is hidden in the environment, is responsible for the infection (Correia et al., 2016).
In the present study, we compiled a large (∼1500 isolates) and comprehensive collection of L. pneumophila isolates to investigate the in vitro activity of RIF, ERY, CLA, AZI, CIP, MOX, LEV, and DOX against L. pneumophila derived from diverse environments (ponds, rivers, lakes, small streams, seawaters, puddles, fountains, potting soil, and garden soils), which could represent the most common aquatic environmental reservoirs of L. pneumophila in China. Antimicrobial susceptibility profiles of these isolates belonging to different serogroups (sgs) or from different environmental sources and regions were investigated. Furthermore, we aimed to determine an unbiased tentative ECOFF for these antibiotics against L. pneumophila since EUCAST provides widely accepted ECOFFs for these antimicrobial agents that are used for the treatment of LD.

Ethics statement
Ethical approval was not required as only Legionella bacterial isolates from environmental sources were used.

Strains, antibiotics, and culture conditions
A total of 1464 L. pneumophila isolates (including 329 sg1 and 1135 sg2-15 or 1079 water and 385 soil isolates) derived from 138 samples, which were obtained from 12 cities in seven provinces of China between April 2019 and January 2021, were investigated in this study. These isolates were from various aquatic sources, including water sources of rivers, sea, lakes, ponds, small streams, and grass puddles, and soil sources of the garden and potting soils (Supplementary Figure 1). The sampling and processing of water and soil samples, the isolation of Legionella-like bacteria, and the identification of L. pneumophila and L. pneumophila sg1 isolates were described in our previous study (Zhan et al., 2022

Minimum inhibitory concentration determination
Minimum inhibitory concentrations (MICs) were determined by using the broth microdilution (BMD) test using buffered yeast extract broth supplemented with 1g/L α-ketoglutarate, 0.25 g/L L-cysteine, and 0.4 g/L ferric pyrophosphates (BYE) on 96-well microtiter plates as it can provide unbiased MICs due to the absence of charcoal and it was the internationally agreed gold standard for Legionella species (Portal et al., 2021b). In brief, a single colony of Legionella was isolated from a BCYEα plate and inoculated in a BYE broth, adjusting the optical density of the bacterial suspension to OD600 = 0.1 (∼1.0 × 10 8 /mL) in a 1.5-mL tube. A final Legionella bacterial concentration of ∼10 5 /mL in each well was obtained by dilution. A volume of 20 µL of each antibiotic solution was then added to microtiter wells in 2-fold decreasing concentrations. The final volume of each well was 100 µL, containing 10 4 Legionella bacteria. A blank well consisting of 100 µL of the BYE broth was used as a negative control. The plates were sealed and incubated for 48 h without agitation at 37 • C before determining the MIC by observing the presence or absence of turbidity. The MIC was defined as the lowest concentration of antibiotics yielding complete inhibition of visible growth. Each isolate was tested at least two times to verify the results.
MIC 50 /MIC 90 , epidemiological cutoff value, and detection of antibiotic resistance isolates The MICs that inhibited the growth of 50% and 90% of strains were defined as MIC 50 and MIC 90 , respectively. The ECOFF is used to separate bacterial populations based on MIC distributions and is defined as the highest MIC value of isolates that are not known to have resistance (Chidiac et al., 2012). The ECOFFinder program (version 2010-v21), which was raised by the EUCAST, was applied to fit a log-normal distribution curve for the number of isolates that had MICs for L. pneumophila. Theoretically, no less than 95% of WT isolates should be encompassed in the ECOFF; thus, the 97.5% endpoints of value were defined as the ECOFF of a specific antibiotic against L. pneumophila, expressed as WT ≤ X mg/L. We defined isolates with an MIC higher than ECOFF are resistant isolates, whereas others are susceptible ones.

Determination of possible resistance mechanism
More and more antibiotic-resistant isolates are found in clinical L. pneumophila isolates (Bruin et al., 2014;Natas et al., 2019;Cocuzza et al., 2021). These isolates are mainly AZI-resistant, and resistant isolates against other antibiotics are rare (Bruin et al., 2014;Jia et al., 2019). L. pneumophila is a typical environment-borne pathogen. Investigation of antibiotic susceptibility profiles of L. pneumophila from various environmental sources is crucial in generating antibiotic treatment recommendations for clinical infection. Because macrolides, including AZI, are the first-line antimicrobial agents for the treatment of LD, we focus on the molecular mechanism of AZI-resistant isolates as they are shown to have a high frequency in L. pneumophila (Jia et al., 2019;Natas et al., 2019;Cocuzza et al., 2021). Many macrolide resistancemediating mechanisms were found in bacteria, including enzymatic modifications and 23S rRNA gene mutation, as well as ribosomal accessory protein-coding genes including rplD (protein L4) and rplV (protein L22) (Prats-van der Ham et al., 2018;Abushaheen, 2020). However, an efflux pump encoded by lpeA and lpeB (designated as lpeAB) is mainly the cause of AZI resistance in L. pneumophila Jia et al., 2019;Natas et al., 2019). Therefore, we designed a multiplex PCR method to detect the presence of the lpeAB efflux pump in the environmental isolates with AZI resistance. A total of 123 reference L. pneumophila strains that harbored lpeAB efflux were utilized for an efficient primer design. The detailed information of these strains including names, genomic sequences, GenBank accession numbers, and lpeA/lpeB gene locations in the sequences is shown in Supplementary Table 1. Overall, six alleles were found both on lpeA and lpeB genes. Based on the conservative sequence of lpeAB genes, two pairs of primers were designed as follows: lpeA_F: CTGTWGTAAGTATTTACGACCC (W indicates degenerate bases and is equal to A and T), lpeA_R: GGTGTCTTCGTCGAGCA; lpeB_F: CA TCCTGTAATCACCATCATC, lpeB_R: CAACGGAAGCA ACACCTTG. These primers could not only detect the presence of lpeAB but also define if the two genes were correctly arranged in spatial.

Statistical analysis
Continuous variables (log MICs) were compared using Student's t-test or one-way ANOVA. For categorical variables, we calculated the frequencies and percentages of isolates in each category by chi-square or Fisher's exact tests. Non-parametric Spearman's correlation analysis was used to analyze the relationship between different antibiotics against L. pneumophila isolates. Statistical analyses were performed using SPSS 25.0 (IBM software). A principal component analysis (PCA) for MICs of the eight antibiotics was performed using Origin 2021b (OriginLab). P < 0.05 was considered statistically significant. Simpson's diversity index was used to measure the homogeneity of antibiotic susceptibility of the isolates using the formula: D = 1-(Ni/N) 2 .

Results and discussion
Minimum inhibitory concentration ranges and MIC 50 /MIC 90 for Legionella pneumophila isolates from environmental sources The in vitro susceptibility profiles of the 1464 L. pneumophila isolates against the eight antibiotics, as judged by MIC distributions, are shown in Table 1. From the overall MIC distribution of the 1464 L. pneumophila isolates, almost all tested ones were inhibited by low concentrations of RIF, fluoroquinolones, and macrolides. ERY yielded MICs ranging between 0.031 and 0.5 mg/L. Similar results were found in CIP (0.008-0.5 mg/L), which had the same upper limit but a lower limit as compared with ERY. By contrast, AZI produced a higher upper limit of MIC than CIP (0.008-1 mg/L). This result was following the average MICs of each antibiotic against L. pneumophila (Figure 1). One dilution lower MIC range of MOX and LEV was observed (0.016-0.25 mg/L) as compared with ERY. RIF was most active, with the highest MIC being 0.002 mg/L, whereas DOX was most inactive, with the lowest MIC being 2 mg/L. Based on this result, DOX seems to be not recommended to treat LD, although it is the second choice for patients with minor pneumonia (Torre et al., 2018). We did not find significant MIC range differences between sg1 and sg2-15 isolates. Regarding the control strain, L. pneumophila indicating more mechanisms might be existed in confronting ERY for L. pneumophila, and vice versa. Previous studies using the BMD method to perform susceptibility testing showed variability in the MIC 50 , MIC 90 , and MIC ranges (García et al., 2000;Choi et al., 2010;Xiong et al., 2016;Koshkolda and Lück, 2018;Wilson et al., 2018;Assaidi et al., 2020;Cocuzza et al., 2021) (Table 2). Our data are comparable to some of these results ( Table 2). The variability may be because of the limited number of isolates used in those studies and the influence of regional differences for the L. pneumophila isolates, raising the importance of using numerous isolates from different sources and using the standard protocol to define ECOFFs for L. pneumophila. Given that LD is a typical environment-borne disease and all the strains that caused infections are from environments, our results obtained based on numerous environmental isolates could indicate the antimicrobial susceptibility to the clinical isolates.
Distinct antimicrobial susceptibility profiles of Legionella pneumophila isolates with different serogroups, environmental origins, or regions Average MICs obtained by antibiotics against sg2-15 isolates were significantly higher in ERY than those against the sg1 isolates (Figure 2A). We could also observe that MICs obtained by ERY against the isolates derived from China's water sources were higher than those from soil sources ( Figure 2B). PCA indicated that sg2-15 isolates from water samples displayed a specific nature when compared to sg2-15 isolates from soil or sg1 from both soil and water, of which CLA, ERY, and RIF might have a homologous contribution to the nature ( Figure 2C). Further study suggested that sg2-15 isolates from water showed significantly decreased susceptibility to RIF, ERY, and CLA ( Figure 2D). These results together indicated that antimicrobial susceptibility profiles of L. pneumophila belonging to different sgs or from different environmental sources were distinct. Many studies demonstrated antimicrobial susceptibility in a limited number of L. pneumophila isolates from small geographic areas (Xiong et al., 2016;Graells et al., 2018;Koshkolda and Lück, 2018;Torre et al., 2018;Wilson et al., 2018;Cocuzza et al., 2021). As China is a country with a huge land area, antimicrobial susceptibility in L. pneumophila from different geographic may be distinct. Therefore, MICs of the eight antibiotics against isolates from northern, central, eastern, and southern China were compared ( Figure 2E). We found that average MICs of RIF and AZI against northern isolates were significantly higher than those against isolates from other regions of China ( Figure 2F), indicating decreased susceptibility of these antibiotics to those isolates from northern China. Antibiotics enter the environment via waste produced by humans, agricultural food and animal   production, pharmaceutical production, etc. This process may lead to a susceptibility decrease by natural selection. Thus, the RIF and AZI susceptibility decrease might be caused by relatively high environmental concentrations of the two antibiotics in north China, although some articles reported that the Pearl River basin, in which many south cities are located, has the highest antibiotics emission densities (Qiao et al., 2018).
Principal component analysis and correlation networks for minimum inhibitory concentrations revealed possible shared strategies within Legionella pneumophila against antibiotics A biplot via PCA indicated the configuration of MIC distributions of the eight antibiotics on the 1464 L. pneumophila isolates in China, as shown in Figure 3A. Loading signs of the antibiotic MICs indicates whether they are positively or negatively correlated. We found that the susceptibility of antibiotics belonging to the same class (e.g., macrolides and fluoroquinolones) was positively correlated, for they have signed with the same directions. This is in line with our expectations because antibiotics belonging to the same class always have their particular mechanism to kill microorganisms, and a confrontation mechanism has also been developed. To our surprise, RIF, which had a distinct direct antimicrobial mechanism, may have a positive correlation with macrolides ( Figure 3A). Correlation matrix analysis revealed that MICs of antibiotics belonging to the same class were positively correlated (e.g., among ERY, CLA, and AZI, or among MOX, LEV, and CIP), indicating the most similar mechanism against L. pneumophila by the same class of antibiotics or confrontation mechanism against them by L. pneumophila ( Figure 3B). MICs of RIF to L. pneumophila were positively correlated with CLA, ERY, and CIP, with CLA having a moderate correlation (r = 0.32, P < 0.0001, Figure 3B). This also holds for isolates belonging to different sgs or from different environmental sources, especially for sg1 and water isolates (Supplementary Figure 2). RIF plus CLA was a standard of care for many infections, including Mycobacterium ulcerans and Mycobacterium avium infections (Akiyama et al., 2019;Almeida et al., 2019). However, the interaction of RIF and CLA to the cytochrome P450 may influence the therapy (Akiyama et al., 2019). Thus, clinical use of the RIF-CLA combination to treat LD may not benefit not only because of unsatisfactory clinical outcomes that have been observed in other intracellular bacteria, such as Mycobacteria (Akiyama et al., 2019), and in L. pneumophila pneumonia patients who had a longer length of stay and a trend toward higher bilirubin levels when using CLA plus RIF (Grau et al., 2006), but also because of the probability of reduced susceptibility against L. pneumophila. The RpoB gene mutation in the RIF resistance-determining region (RRDR) is the major cause of decreased RIF susceptibility in many bacteria (Campbell et al., 2001). In addition, a few putative efflux pumps (viz., Rv1258c, Rv1410c, Rv1819c, and PstB) have been observed to play a role in regulating RIF susceptibility in another intracellular bacterium Mycobacterium tuberculosis (Pang et al., 2013;Narang et al., 2019). Efflux pumps that used two or more antibiotics as substrates were observed in bacteria, including the RND efflux pumps (fluoroquinolones, macrolides,  and aminoglycosides), the MFS efflux pumps (isoniazid, fluoroquinolones, rifampicin, tetracyclines, etc.), and the ABC efflux pumps (macrolides, streptogramins, and fluoroquinolones) (Thakur et al., 2021). We speculated that special defense strategies against both RIF and CLA may exist in L. pneumophila, probably due to undiscovered efflux pumps used as substrates or similar infiltration mechanism of the two antibiotics in L. pneumophila, which deserve further research. Although DOX showed less efficiency in inhibiting L. pneumophila in vitro (Figure 1 and Table 1), the uncorrelated susceptibility to all other antimicrobials implied it was a potential supplement in confronting resistant strains of other antibiotics ( Figure 3B).

Epidemiological cutoff value and the presence of antibiotic resistance isolates
The primary MIC distribution of the eight antimicrobials against L. pneumophila is shown in Figures 4A-H. Raw and fitted curves were obtained by ECOFFinder. Due to the limited number of MICs for LEV and DOX (only four for each), the normal distribution could not be calculated for the two antibiotics (Figure 4I), they still showed good shape distributions as the other six antibiotics. As a result, the ECOFF was defined as 0.002 mg/L for RIF; 0.5 mg/L for ERY; 0.125 mg/L for CIP, CLA, and MOX; 0.5 g/L for AZI; 0.063 mg/L for LEV; and 32 mg/L DOX (Figures 4A-H). These results were comparable to those obtained by using BMD methods but may be more accurate due to a large number of isolates (Table 3). Jia et al. (2019) defined ECOFFs based on 149 strains using the E-test method and obtained a 2-fold higher value for AZI (1 mg/L) and ERY (1 mg/L); 8-fold higher value for MOX (1 mg/L) and LEV (0.5 mg/L); and 16-fold higher values for the and RIF (0.031 mg/L) when compared with our data. Bruin et al. established the ECOFFs of 10 antimicrobials for sg1 using 183 clinical isolates by the E-test (Bruin et al., 2012). Sharaby et al. (2019) established the ECOFFs of 10 antimicrobials for L. pneumophila using 93 environmental and 12 clinical isolates by using the E-test (Sharaby et al., 2019). They showed 2-to16-fold higher MICs in most of the antibiotics, which were similar to Jia et al.'s report. Conversely, their E-test (E-test methods as indicated in Table 3) for DOX in L. pneumophila sg1 gained about 1/80∼1/2-fold MICs and 1/64∼1/4 ECOFFs when compared with our data and Portal et al.'s data (Tables 1, 3) (Bruin et al., 2012;Portal et al., 2021b). Given that charcoal in the BCYEα plate biases the MICs for many antibiotics, including RIF, macrolides, and tetracycline (Bornstein et al., 1985), two or more folds of high MICs or ECOFFs in most of the antimicrobials could be obtained when compared with the BMD method (Portal et al., 2021b); however, this seemed not to hold for DOX. Thus, a standardized method for Legionella is urgent to ensure that the MIC data remain consistent (Portal et al., 2021a). Portal et al. (2021b) obtained an ECOFF for AZI based on the BMD methods (Portal et al., 2021b). Natas et al. (2019), Portal et al. (2021b) defined MICs of AZI based on a bimodal MIC distribution of 122 isolates. They all obtained an ECOFF WT ≤ 0.25 mg/L for AZI. Natas et al. (2019) also reported that isolates with MICs ≥ 0.5 mg/L all harbored resistant genes (lpeAB) and could be defined as resistant isolates. These results were roughly consistent with our results obtained by the ECOFFinder, although we did not find an obvious bimodal distribution of MIC on the 1464 isolates. Vandewalle-Capo et al. (2017) reported the ECOFF WT ≤ 2 mg/L for AZI by using the BYD method. However, owing to the seemly trimodal of MIC distribution, a WT ≤ 0.25 mg/L might be the "real" ECOFF (Vandewalle-Capo et al., 2017), which was also roughly consistent with our results. For consistency, we also tentatively defined those isolates with MIC ≥ 0.5 mg/L for azithromycin as resistant ones. Therefore, based on the WT ECOFFs of the present study and AZIresistant thresholds reported by other studies, some resistant isolates to all the fluoroquinolones (ciprofloxacin, moxifloxacin, and levofloxacin) (two isolates, 0.14%) and azithromycin (46 isolates, 3.14%) were found (Figure 4I). Few studies have reported fluoroquinolone-resistant L. pneumophila (Bruin et al., 2014). Bruin et al. (2014), Shadoud et al. (2015) reported the isolation of ciprofloxacin-resistant L. pneumophila sg1 isolates (MIC = 2 mg/L, E-test method) from a pneumonia patient, probably due to the gyrA mutation during antimicrobial therapy. Fluoroquinolone-resistant strains from environmental Antimicrobial susceptibility profiles of L. pneumophila belonging to different sgs or from different environmental sources or regions in China.
(A) Antibiotic susceptibility profiles of isolates belonging to different sgs. (B) Antibiotic susceptibility profiles of isolates from different sources. W, water sources; S, soil sources. (C) PCA biplot of MICs of the isolates. Isolates are shown as dots and colored by groups based on the sources and sgs. Indices are given as lines with arrows and colored. The configuration of indices in the biplot represents the relationship between variables and principal components. The gray shadow indicates 95% confidence for sg2-15 isolates from water sources. A dashed ellipse indicates significantly correlated variables. (D) Antimicrobial susceptibility profiles of isolates belonging to different sgs and from different sources. W1, sg1 isolates from water sources; W2-15, sg2-15 isolates from water sources; S1, isolates from soil sources; S2-15, sg2-15 isolates from water sources. Data are shown as floating bars with the max and min and mean values. Dotted lines indicate that log MICs are different in W2-15 isolates when compared with the other three types of isolates. (E) Regions of China where the tested isolates were obtained are shown. Cities where the isolates were obtained in each region are shown. (F) Antibiotic susceptibility profiles of isolates from different regions of China.
Colored regions indicate the provinces where the isolates are from. We defined those MICs were 1.25 times higher/lower (log MIC gap > 0.969) than the contrast with P < 0.05 as significant. ****P < 0.0001. sources were not documented yet, and our findings revealed the hidden isolates in the environments and emphasized the need for environmental surveillance and susceptibility testing of L. pneumophila isolates. Given that the two isolates were resistant to all fluoroquinolones (CIP, MOX, and LEV) based on the WT distribution of MIC, resistance-associated mutations in the QRDR of DNA gyrase and topoisomerase IV coding genes may exist and requires further research. The two studies that utilized limited numbers (122 and 149 strains) of L. pneumophila composed of environmental and clinical strains from Norway or China showed that frequencies of AZI-resistant ones were both about 17% (21/122 and 25/149), and environmental strains displayed similar frequencies of resistant ones with the clinical strains (Jia et al., 2019;Natas et al., 2019). In a study with 58 environmental isolates from northern Italy, 6.9% of AZI-resistant L. pneumophila were found with an MIC ≥ 0.5 mg/L (BMD method) (Cocuzza et al., 2021). In addition, Vandewalle-Capo et al. demonstrated that 23.85%    (26/109) of L. pneumophila clinical sg1 strains had MICs ≥ 0.5 mg/L, and 11.01% (12/109) of which were AZI-resistant ones based on the ECOFF WT ≤ 2 mg/L they obtained (BMD method) (Vandewalle-Capo et al., 2017). However, our big data indicated that the frequency of AZI-resistant L. pneumophila in the environmental reservoir might be less (3.14%) than ever thought.

Detection of azithromycin resistance genes
To uncover possible resistance mechanisms associated with L. pneumophila against AZI, a multiplex PCR was conducted not only to detect the efflux pump coding genes lpeA and lpeB but also to detect the genetic arrangement of the two genes because the arrangement may have a potential impact on the gene expression (Ueda et al., 2005;Mernke et al., 2011). Our design differed from previous studies that only detected the presence of the two genes Vandewalle-Capo et al., 2017;Natas et al., 2019). The design and multiplex PCR conditions, as well as the examples of PCR results, are shown in Figure 5. Based on the genomic study of 123 L. pneumophila strains that harbor lepAB efflux pump, the spatial arrangement of the two genes was resolved. LpeB is located downstream of lpeA, with an interval of four DNA bases (Figure 5A and Supplementary Table 1). All 46 AZI-resistant isolates and representative AZI-sensitive isolates that covered all types of MICs (0.008-0.25 mg/L) were tested. Among the 46 AZIresistant isolates, 44 (95.65%) harbored lpeAB efflux; 13 AZIresistant isolates that had an MIC of 1 mg/L all (100%) harbored lpeAB efflux. For the 33 AZI-resistant isolates that had an MIC of 0.5 mg/L, 31 (93.94%) harbored lpeAB efflux. All the lepAB-positive (lpeAB +) isolates had the right spatial arrangement of two genes. Conversely, none of the AZI-sensitive isolates (MIC < 0.5 mg/L) harbored the lpeAB efflux according to the PCR for 273 representative isolates ( Figure 6A and Supplementary Table 2). A biplot via PCA indicated the configuration of MIC profiles of L. pneumophila isolates, as shown in Figure 6B. The second component (PC2) could roughly separate lpeAB + isolates from lpeAB-isolates, with AZI having 35.7% contribution for PC2, but only 2.5% for PC1. LpeAB + isolates were not observed to have an elevated MIC of the other seven antimicrobials, including the other two macrolides (ERY and CLA), but an elevated MIC of CLA in lpeAB-isolates was observed (Figure 6C), indicating that the lpeAB efflux pumpmediated resistance in the L. pneumophila environmental isolates were restricted to AZI, but not the other macrolides such as ERY and CLA. This partially contradicted Massip et al. report that lepAB mutant could also lead to decreased MICs of other macrolides including ERY, CLA, and spiramycin against L. pneumophila strain Paris . We did not observe differences in the distribution of lpeAB efflux between the sg1 and sg2-15 isolates, but significantly different distribution patterns of lpeAB between water and soil isolates or among isolates from different Frontiers in Microbiology 13 frontiersin.org Design of lepAB primers with genomic sequences from L. pneumophila harboring the lpeAB efflux, and multiplex PCR to detect lpeAB genes and their spatial arrangement. (A) Arrangement of the lpeAB is shown using L. pneumophila str. Paris as a reference. Primers and DNA sequences of lpeAB, and GenBank accession numbers are shown, and the blue italics indicate the gene location in the selected sequences of the strains. (B) PCR amplification conditions. (C) Representative results for detection of lpeAB efflux in L. pneumophila isolates. DNA electrophoresis shows specific lpeA targets (green arrow, 152 bp), lpeB targets (blue arrow, 499 bp), and lpeA-lpeB combined targets (red arrow, 1009 bp), which indicate the right arrangement of the two genes. Lanes 1-9 show positive for lpeAB efflux; lanes 10-11 show negative for lpeAB efflux; n.c. indicates negative control which used sterile water as a template.
regions were found ( Figure 6D). Together, these results indicated that the presence of the lepAB efflux pump was the most important AZI but not the other seven antimicrobials resistance mechanism for L. pneumophila.
The distribution patterns of AZI-resistant genes (lpeAB) were distinct between isolates from different environmental sources or regions in China. Natas et al. and Jia et al. both showed that all AZI-resistant L. pneumophila isolates harbored an lpeAB efflux pump; however, we did find 2 AZIresistant isolates without lpeAB. These isolates also showed similar MICs of other two macrolides, as well as other types of antibiotics, when compared with those lpeAB-AZIsensitive isolates (Supplementary Table 2), indicating some undefined AZI-specific resistance mechanisms such as special ribosomal mutations, and other AZI-specific efflux pumps in L. pneumophila might exist and deserve further study .

Conclusion
A large number of L. pneumophila isolates may provide an unbiased antimicrobial susceptibility profile and ECOFFs for the clinically used antibiotics against LD. L. pneumophila, which is hidden in the environment, is responsible for LD. Limited data on antimicrobial susceptibility of L. pneumophila have been obtained in China over the past few decades. Our data will guide future studies of the antibiotic sensitivity and resistance mechanism of pathogenic L. pneumophila, which could serve as a reference for setting clinical breakpoints and discovering antibiotic-resistant L. pneumophila isolates in the clinic, as well as an effective treatment of L. pneumophila infection. Our data showed antimicrobial susceptibility of L. pneumophila belonging to different sgs or from different environmental sources and regions was distinct in some antimicrobial agents, such as RIF, ERY, CLA, and AZI. Most of the environmental isolates showed in vitro sensitivity to the tested antibiotics, in agreement with previously published data. In total, two isolates with fluoroquinolones (CIP, MOX, LEV) resistance and 46 isolates with AZI resistance were determined in the present study. Genomic research, multiplex PCR assay, and statistical analysis confirmed the correction of lepAB efflux with AZI-resistance in L. pneumophila. However, we found this mechanism was only restricted to AZI but no other macrolides (e.g., ERY and CLA). Furthermore, some undefined AZI-specific resistance mechanisms that are not the same as other macrolides (e.g., CLA and ERY) in L. pneumophila might exist. With the statistical analysis using our big data, we also proposed the existence of the shared defense strategies, particularly against both RIF and CLA in L. pneumophila, which deserves further research.