Evaluation and comparison of antibiotic susceptibility profiles of Streptomyces spp. from clinical specimens revealed common and region-dependent resistance patterns

Notwithstanding the fact that streptomycetes are overlooked in clinical laboratories, studies describing their occurrence in disease and potential pathogenicity are emerging. Information on their species diversity in clinical specimens, aetiology and appropriate therapeutic treatment is scarce. We identified and evaluated the antibiotic susceptibility profile of 84 Streptomyces clinical isolates from the Czech Republic. In the absence of appropriate disk diffusion (DD) breakpoints for antibiotic susceptibility testing (AST) of Streptomyces spp., we determined DD breakpoints by correlation with the broth microdilution method and by the distribution of zone diameters among isolates. Correlation accuracy was high for 9 antibiotics, leading to the establishment of the most valid DD breakpoints for Streptomyces antibiotic susceptibility evaluation so far. Clinical strains belonged to 17 different phylotypes dominated by a cluster of strains sharing the same percentage of 16S rRNA gene sequence identity with more than one species (S. albidoflavus group, S. hydrogenans, S. resistomycificus, S. griseochromogenes; 70% of isolates). AST results showed that Streptomyces exhibited intrinsic resistance to penicillin, general susceptibility to amikacin, gentamycin, vancomycin and linezolid, and high percentage of susceptibility to tetracyclines and clarithromycin. For the remaining antibiotics, AST showed inter- and intra-species variations when compared to available literature (erythromycin, trimethoprim-sulfamethoxazole), indicating a region-dependent rather than species-specific patterns.


Materials and methods
Bacterial isolates acquisition and identification. The study included 84 non-duplicated human clinical isolates collected between 2009 and 2018 at the Trutnov Regional Hospital (29 strains, strain coding TR), the Ostrava Public Healthcare Institute (53 strains, strain coding OS), and the Příbram Regional Hospital (2 strains, strain coding PR). All strains were isolated during routine diagnostics in mycobacteriology laboratories. Strains have been deposited in the Collection of Actinomycetes of the Biology Centre Collection of Organisms (BCCO, www. actin omyce tes. bcco. cz). DNA extraction and identification of all isolates was performed according to 41 . The 16S rRNA gene sequences were compared against the type strains database using the Basic Local Alignment Search Tool 42 . The phylogenetic tree was constructed using Geneious (v 8.1.6, http:// www. genei ous. com, neighbour joining, Tamura-Nei genetic distance model, 1000 replicates). The strains were assigned to different clades according to 43 . The list of all strains and the reference strains most closely related according to 16S rRNA gene similarity and their classification into clusters are reported in Table 1. Strain TR1341 (here assigned to S. murinus based on nucleotide similarity of the 16S rRNA gene) was discussed in the previous study 6 . The nucleotide sequences of the 16S rRNA genes of the isolates were deposited in GenBank under the accession numbers MZ393577-MZ393782 (Supplementary Table S1).
Antibiotic susceptibility testing. Antimicrobials. The antimicrobials used in the study are listed in Table 2. The abbreviations and disk contents are included.
Clinical breakpoints setting. To develop a criterion for interpreting DD results, we followed three different approaches. (1) To correlate DD with BM results for antibiotics available in a commercial BM kit. The minimum inhibitory concentration (MIC) breakpoints chosen to the derive zone diameter (ZD) breakpoints are reported in Table 2. We selected 29 clinical isolates and the number of strains belonging to a given cluster was selected proportionately. To increase the robustness of the correlation analysis, 18 additional soil Streptomyces isolates (also deposited in BCCO) and the type strains S. rameus DSM 41685 and S. violaceoruber DSM 40783 were included in the study. The soil and type strains were subjected to the same procedures as the clinical isolates, except for the growth temperature, which was set to 28 °C. The soil isolates belonged to the same clusters as the clinical ones when possible. A total of 49 Streptomyces strains were used in the correlation analysis (Table 1). (2) For antibiotics not available in a commercial BM kit, we proposed at least tentative breakpoints based on the distribution of the ZD data and the class of the antibiotic. (3) For antibiotics not available in a commercial BM kit and without a clear cut-off in the distribution of ZDs, we used arbitrary ZD breakpoints.
Minimum inhibitory concentration (MIC) determinations. The BM method for aerobic Actinomycetes described by CLSI in the M24 manual 45

Results
Strain taxonomy. The selected 84 clinical Streptomyces strains belong to 16 different phylotypes (Table 1).
Most strains (83%) have the same percentage of sequence identity (PID) with more than one Streptomyces species due to insufficient variation in their 16S rRNA gene sequence (all clinical strains PID > 99%, except OS587 with PID 98.88%). The phylogenetic tree of the clinical Streptomyces strains was complemented with environmental and type strains (Fig. 1). The clinical strains were then assigned to 13 phylogenetic clusters, with cluster C (S. albidoflavus group/S. hydrogenans/S. resistomycificus/S. griseochromogenes) comprising 70% of all clinical isolates.

ZD breakpoints setting.
To derive ZD interpretive criterion, 3 different approaches were followed: i) Correlation of BM MICs values with zone diameters when antibiotics available in commercial BM kits; ii) Tentative breakpoints setting based on the distribution of ZD data and class of the antibiotic for antibiotics without correlation analysis; iii) Arbitrary ZD breakpoints when the distribution of ZDs did not reveal a clear cut-off for S-R category for antibiotics without correlation analysis.
Correlation analysis-based breakpoints. We found a strong negative correlation (Pearson´s correlation r ˂  Table 2. Discrepancy percentages of proposed breakpoints were in acceptable ranges. Scattergrams and discrepancy percentages are shown in Fig. 2 and Supplementary Figures S1-S7. Evaluation of the correlation between MIC and ZD values for SXT was problematic due to difficulties in determining the MIC and ZD endpoints, which is likely reflected in a weak correlation (Pearson´s r = −0.60, n = 36, data not shown). In addition, strong disagreement between cat-  Figures S8 and S9).
Arbitrary ZD breakpoints. The ZD distribution of DOX, OFX and SMN did not allow us to visually define the R-S cut-offs, and with the lack of MIC-ZD correlation analysis, we arbitrarily defined the ZD breakpoint values as R ≤ 20 mm, I = 21-29 mm, S ≥ 30 mm (Supplementary Figure S10A-C).  Table 2. The CO WT values defining the cut-offs for the wild type and non-wild type population of the strains in cluster C are also shown in Table 2. The resistance patterns of each Streptomyces phylotype and the assignment of the corresponding graphs are summarized in Table 3. A high percentage of resistant clinical strains was found in case of penicillin group antibiotics: PEN (100%), AMP (82%) and AMX (81%). The enrichment of AMX with clavulanic acid rapidly decreased the resistance of clinical isolates to 7%. A high resistance frequency was also found in case of CZN, CRO, ERY and SXT: 98%, 88%, 87% and 79%, respectively. In contrast, all clinical isolates were susceptible to AKN, GEN and VAN. A high frequency of susceptible clinical strains was found in case of tetracycline group antibiotics (MNO, DOX, TET), CLR, CMP and CIP: 94-91%, 89%, 85% and 49% (plus 45% in "I" category), respectively. A high frequency of susceptible or intermediate susceptible clinical strains was also found for RIF (62% and 27% as "S" and "I" category) and OFX (7% and 76% as S and I category). However, due to the lack of reliable breakpoints (ZD breakpoints were set arbitrarily) these results need further validation. Within the tested cluster C strains, 3.4%, 3.4%, 6.7%, 8.5%, 1.7% and 1.7% were non-wild type in case of AMX, CZN, CIP, TET, DOX and VAN, respectively. For AKN and RIF, non-wild type populations were not determined, as the cluster C zone diameter datasets have an abnormal distribution. For the remaining tested antibiotics, all strains were wild types.
For AMS and LIZ, we performed only the BM method and only 29 clinical isolates were included. All isolates were susceptible to LIZ. In case of AMS, the MIC values of 83% of tested clinical isolates decreased by 1 or more dilutions compared to the MIC value of AMP alone, i.e., the susceptibility of the strains increased (24% of clinical isolates changed susceptibility category from R to S).
Most isolates were resistant to 7 drugs (53 isolates), with a minimum resistance to 1 drug (TR1144 resistant only to PEN) and with maximum multi-drug resistance to 12 drugs occurring in 2 isolates in cluster B (OS282 and OS1126B) and strain TR1341 of cluster G (Supplementary Figure S11).
The comparison of AST results between clinical isolates of the same cluster (cluster C) collected in Spanish provinces (25 isolates) 14 and those in our study (59 isolates) is shown on Fig. 3. The proportion of resistant isolates significantly differs by country for ERY (X 2 = 42.314; p ˂ 0.01) and SXT (X 2 = 23.789; p ˂ 0.01). The frequency

Discussion
Although Streptomyces is increasingly emerging in clinical settings, there is little information on the aetiology, species distribution and antibiotic susceptibility profiles, with respect to the high number of Streptomyces species described to date. These are mostly case reports 5,19,[28][29][30][32][33][34][49][50][51] , with only scarce works addressing the large cohorts of clinical isolates [12][13][14] . Together with these, our work brings a new perspective on the presence and diversity of Streptomyces spp. in human microbiome, and points, that streptomycetes are important in clinical samples and should receive greater attention.
Here we present identification and AST of 84 clinical Streptomyces isolates collected in Central Europe (Czech Republic). For the first time, we performed a correlation between the BM and DD method in AST of Streptomyces. Most of the strains in our study were isolated from patients suffering from chronic respiratory disease (59.5%), although it is not clear whether they were the cause of the disease, represented a secondary bacterial infection, or they were common colonizers of the human body. Although 17 different phylotypes were identified among the clinical Streptomyces in our study, only 3 of them have been reported as causative organisms of human diseases before: S. albus and S. thermoviolaceus in pulmonary infections 27,50,52 , S. albus as a causative agent of mycetoma 33 and S. thermocarboxydus in keratitis 53 . A recent study 14 identified 6 other phylotypes identical to those presented in our study (S. albidoflavus, S. rutgersensis, S. rochei, S. drozdowiczii, S. xylanilyticus and S. carpaticus), but  A  1  1  1  0  0  1  1  0  1  1  1  1  1  0  1  0  1  1  0  0  1   www.nature.com/scientificreports/ their clinical relevance is unknown. The phylotypes reported in our study as clusters H, I, L, M, and S. ginkgonis, with S. xiamenensis of cluster A are associated with clinical specimens for the first time. It is interesting to note that cluster C, which is the most abundant in our study (70.2% of clinical isolates), has been reported only in another study so far, and moreover with much less proportion (~ 13% of clinical isolates) 14 . This could be due to regional differences in agriculture, industry and lifestyle, that could play a role in the colonization of humans by different Streptomyces species, as well as due to complicated taxonomic issues. Taxonomy of Streptomyces species is a complex task involving the identification of genotypic and phenotypic characters 54 . The sequences in the GenBank database are updated literally every minute, and therefore the reports on species identified decades ago are questionable (e. g. 28 Streptomyces isolates of S. griseus 12 ) and thus difficult to compare.

S. plicatus/S. rochei/S. vinaceusdrappus/S. enissocaesilis
To evaluate the AST of Streptomyces spp. we chose the modified Kirby-Bauer method correlated with broth microdilution method (CLSI guidelines for Nocardia and other aerobic actinomycetes 45 ) for 10 antibiotics. Since establishing clinical breakpoints is a challenging task, all available MIC breakpoints for Streptomyces spp. were selected as alternatives based on their taxonomic classification. If official MIC breakpoints for Streptomyces spp. are established by an international committee, the DD breakpoints proposed in this study can be easily adjusted using the scattergrams provided.
The correlations were excellent for all tested antibiotics (Pearson´s correlation coefficient ranged from −0.89 to −0.96), except for PEN and SXT. Streptomyces species are known for their benzylpenicillin resistance 55 , thus all the strongly resistant values outside the MIC range (> 8 mg/L) had to be excluded from the correlation calculation (28 strains). This might have led to less accurate results. In case of SXT, the inconsistency in reading the MIC endpoint occurred. Due to the uneven growth of streptomycetes in liquid cultures (growth in clumps), it is difficult to detect wells with partial growth (20% lower detection rate comparing to control well as recommended by CLSI) or determine a clear cut-off, while slight growth is neglected. Moreover, the unresolved disagreement between methods occurred in cluster D (S. gougerotii). Although SXT is recommended worldwide for the treatment of actinomycetoma 56 as an empirical antibiotic therapy, we believe that the treatment of Streptomyces infections with SXT is probably inappropriate because of the low accuracy of the results as well as the high percentage of resistant strains in our study.
The variability of the species in our study is limited and is mainly represented by cluster C strains (70.2%), which have the same or very similar resistance patterns. Considering the species-specific susceptibility profile of Streptomyces 57 , the notably high percentages of resistance to some antibiotics (AMP, SXT and ERY) may be biased by the large proportion of one phylotype in our data set. Therefore, we compared AST profiles from our study with those for clinical species previously published (Supplementary Table S2). Our data confirmed a general susceptibility of Streptomyces to AKN, GEN, VAN and LIZ. The only discrepancy found in the literature is a strain of S. griseus resistant to AKN with MIC of 16 mg/L 12 described as susceptible in the study, but resistant according to current guidelines 45 . However, the result is questionable, since the AST procedures as well as the methods for identification have changed since the study was published (1990). These antibiotics are associated with the treatment of complicated multi-drug resistant infections and are often reserved as drugs of last resort, some of which have significant side effects [58][59][60][61][62] . On the contrary, there is intrinsic resistance to PEN, usually recommended as first-line therapy for the treatment of respiratory diseases and pneumonia 63 . For the remaining antibiotics, there is a considerable variability in the susceptibility profiles of Streptomyces spp., for some of them with high percentage of resistance (cephalosporins, CIP, ERY) or susceptibility (CLR, tetracyclines, imipenem). Thus, if antibiotic other than the safe one must be used, AST of the causative organism is recommended as well as species identification.
The most interesting property of Streptomyces is their ability to produce antibiotics. Since antibiotic resistance genes are thought to originate from antibiotic-producing bacteria, the likelihood of multi-drug resistance (MDR) occurrence is high. For example, there are more than 100 drug resistance gene homologues in the chromosome of Streptomyces coelicolor A3(2) 64 and the presence of the Van cluster (vanSRJKHAX) associated with inducible resistance to vancomycin has been reported, too 65 . Nevertheless, there are indications that most of the MDR systems are suppressed under laboratory conditions and Streptomyces species are therefore generally considered to be drug sensitive 66 . In our study, we found strains with a wide range of resistance patterns, from  www.nature.com/scientificreports/ generally susceptible to MDR strains. It is noteworthy that one of the most resistant clinical strains in our study, Streptomyces sp. TR1341, originated from a patient with multiorgan tuberculosis, relapsing respiratory infections and chronic obstructive pulmonary disease with a long-term and repeated antibiotic therapy 6 . This phenotype was supported by genomic analysis, which revealed the presence of 41 known resistance models in its genome 15 . Even though non-wild type strains were found in clinical isolates of cluster C, there is no obvious shift in susceptibility category within strains of same phylotype. The ZD value of the only one non-wild type isolate of cluster C that changed susceptibility category compared to wild type isolates balances at the ZD breakpoint value. Therefore, our data suggest a rather species-specific susceptibility profile of Streptomyces. To confirm our findings, we compared the antibiograms of the cluster C strains with those presented in the only study on clinical isolates of S. albidoflavus (Spain) in the literature to date 14 . All our isolates from cluster C were resistant to ERY and SXT, as reflected also in a low CO WT value in case of ERY (CO WT for SXT was not calculated since all strains lacked an inhibition zone). Contrary, only 24% of S. albidoflavus isolates from Spain were resistant to ERY. This discrepancy suggests local adaptation of the cluster C species to unique selection pressures in different regions, including differences in agriculture, industry, lifestyle, and also antibiotic and other drug policy (higher consumption of SXT and macrolides in Czech Republic compared to Spain) 67,68 . Spontaneous mutations conferring ERY-resistance under selection pressure have already been demonstrated in the model actinomycetes S. coelicolor and S. lividans 69 and are associated with point mutations in rrnA-23S rRNA and rrnC-23S rRNA 70 . Since SXT resistance determinants are located on mobile elements such as small plasmids and gene cassettes 71 , horizontal gene transfer under appropriate conditions is a possibility, although it has not yet been documented for Streptomyces.
In conclusion, we proved a generally high suitability and accuracy of the disk diffusion method for the AST of Streptomyces spp. by correlation with the gold-standard microdilution broth method for 9 antibiotics (SXT remains the questionable due to unresolved ambiguity in cluster D). This led to the determination of DD susceptibility breakpoints derived from MIC breakpoints for Streptomyces-related organisms. To the best of our knowledge, these are the most valid DD breakpoints for Streptomyces reported to date. Tentative breakpoints have been proposed for 10 additional antibiotics, however, these breakpoints were designed primarily for the purpose of this study. Further analyses, such as correlation analyses, are recommended. All tested clinical isolates were susceptible to AKN, GEN, VAN and LIZ which is in agreement with literature, therefore these antibiotics can be chosen as empiric treatment for Streptomyces-associated infections. A low percentage of resistant isolates (˂ 10%) was found in our study for tetracyclines, CLR, CIP and AMC, however, data for CIP and AMC susceptibility differs in literature (Supplementary Table 2). In contrast, Streptomyces are intrinsically resistant to penicillin, and have a high percentage of resistance to cephalosporins. Other antibiograms (ERY and SXT) appear to be regionally driven, rather than species-specific and thus AST must be performed prior therapy. The treatment with ERY and SXT associates with a high risk of failure due to acquired resistance and should be reconsidered. Our study also emphasizes that Streptomyces are emerging in clinical practice, although still largely neglected, and points out the importance of optimizing techniques for selective isolation from clinical specimens. As awareness of streptomycetes infections in humans has increased, it is desirable to continue investigating their virulence factors and clinical relevance.

Data availability
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.