Highly sensitive detection of antimicrobial resistance genes in hospital wastewater using the multiplex hybrid capture target enrichment

ABSTRACT Wastewater can be useful in monitoring the spread of antimicrobial resistance (AMR) within a hospital. The abundance of antibiotic resistance genes (ARGs) in hospital effluent was assessed using metagenomic sequencing (mDNA-seq) and hybrid capture (xHYB). mDNA-seq analysis and subsequent xHYB targeted enrichment were conducted on two effluent samples per month from November 2018 to May 2021. Reads per kilobase per million (RPKM) values were calculated for all 1,272 ARGs in the constructed database. The monthly numbers of patients with presumed extended-spectrum β-lactamase (ESBL)-producing and metallo-β-lactamase (MBL)-producing bacteria, methicillin-resistant Staphylococcus aureus (MRSA), and vancomycin-resistant enterococci (VRE) were compared with the monthly RPKM values of blaCTX-M, blaIMP, mecA, vanA, and vanB by xHYB. The average RPKM value for all ARGs detected by xHYB was significantly higher than that of mDNA-seq (665, 225, and 328, respectively, and P < 0.05). The average number of patients with ESBL producers and RPKM values of blaCTX-M-1 genes in 2020 were significantly higher than that in 2019 (17 and 13 patients per month and 921 vs 232 per month, respectively, both P < 0.05). The average numbers of patients with MBL-producers, MRSA, and VRE were 1, 28, and 0 per month, respectively, while the average RPKM values of blaIMP, mecA, vanA, and vanB were 6,163, 6, 0, and 126 per month, respectively. Monitoring ARGs in hospital effluent using xHYB was found to be more useful than conventional mDNA-seq in detecting ARGs including blaCTX-M, blaIMP, and vanB, which are important for infection control. IMPORTANCE Environmental ARGs play a crucial role in the emergence and spread of AMR that constitutes a significant global health threat. One major source of ARGs is effluent from healthcare facilities, where patients are frequently administered antimicrobials. Culture-independent methods, including metagenomics, can detect environmental ARGs carried by non-culturable bacteria and extracellular ARGs. mDNA-seq is one of the most comprehensive methods for environmental ARG surveillance; however, its sensitivity is insufficient for wastewater surveillance. This study demonstrates that xHYB appropriately monitors ARGs in hospital effluent for sensitive identification of nosocomial AMR dissemination. Correlations were observed between the numbers of inpatients with antibiotic-resistant bacteria and the ARG RPKM values in hospital effluent over time. ARG surveillance in hospital effluent using the highly sensitive and specific xHYB method could improve our understanding of the emergence and spread of AMR within a hospital.


Comparison between ARGs in hospital effluent and clinical ARB isolates
ESBL-producing E. coli, K. pneumoniae, K. oxytoca, and P. mirabilis were detected, averaging 10, 3, 1, and 1 patient per month, and the monthly average was significantly higher in 2020 than in 2019 (17 vs 13 patients/month and P < 0.05) (Fig. 3). In parallel with this increase, the average monthly RPKM values of bla CTX-M-1 group was signifi cantly higher in 2020 than in 2019 (921 vs 232, and P < 0.05), while the average RPKM values of bla CTX-M-9 group in 2019 and 2020 were similar (1,016 vs 636 per month and P ≥ 0.05).
Significantly fewer patients carried metallo-β-lactamase (MBL) producers, including E. coli, K. pneumoniae, Enterobacter cloacae, Pseudomonas aeruginosa, and Acinetobacter spp., than carried ESBL producers (1 vs 14 patients per month and P < 0.05) (Fig. 3). The mean RPKM value of bla IMP in effluent was significantly higher than that of bla CTX-M (6,163 vs 1,327 per month and P < 0.05). Significantly more patients carried methicillinresistant Staphylococcus aureus (MRSA) than carried ESBL producers (28 vs 14 patients per month and P < 0.05). The average RPKM value of mecA was significantly lower than that of bla CTX-M (6, 1,327 per month, and P < 0.05). VRE was not clinically isolated during the study period, and although vanA was not detected in the hospital effluent, vanB was detected, with an RPKM value of 126 per month.

DISCUSSION
This study demonstrated that xHYB could detect clinically important ARGs, including bla CTX-M and bla IMP , most of which were not well detected by conventional mDNA-seq and can effectively depict the ARG profile in hospital effluent. ARGs frequently detected among clinical ARB in Japan, including bla CTX-M-1 , bla CTX-M-9 , and bla IMP-1 genes, were not detected by mDNA-seq. However, xHYB revealed that they were more abundant in the hospital effluent than other ARGs of the same type. An increase in the frequency    of ESBL-producing isolates was observed, paralleled by an increased abundance of bla CTX-M-1 genes in the effluent. Although most ESBLs in clinical isolates in Japan are bla CTX-M (8,9), in this study, the RPKM value of bla TEM was higher than that of bla CTX-M . Although all previously identified bla CTX-M genes confer an ESBL phenotype, the operational taxonomic unit (OTU) of bla TEM in the AMROTU database in the present study contains narrow-spectrum bla TEM genes, including bla TEM-1 and bla TEM-2 , which are widely distributed among gram-negative rods in water environments (10). Among the bla CTX-M genes detected in the hospital effluent, bla CTX-M-9 was most abundant, with a total RPKM of 765, followed by bla CTX-M-1 , with 556, accounting for 99% of the bla CTX-M genes. Among the major ESBL producers isolated, E. coli was most abundant (71%). These results were consistent with previous clinical epidemiological studies of ESBL producers in Japan (6,9). The recent rapid increase in the prevalence of bla CTX-M , the largest group of ESBLs, has become a major public health concern worldwide (6). In Japan, the prevalence of  In the present study, the increased abundance of bla CTX-M-1 group in hospital effluent paralleled the frequency of ESBL-producing bacterial isolates. The most frequent and clinically important carbapenemase genes are bla NDM-1 , bla VIM , bla KPC , and bla IMP (12). None of these genes was detected by mDNA-seq, whereas bla VIM and bla IMP were detected by xHYB. bla IMP-1 group and bla VIM , with a total RPKM of 6,130 and 224, accounted for 96% and 4% of the carbapenemase genes in the hospital effluent, respectively, which is consistent with previous clinical epidemiological studies showing that the carbapenemases detected in Japan were mostly bla IMP-1 group genes, whereas bla VIM , bla NDM-1 , and bla KPC were rarely detected (7,13). In recent years, there has been a reported increase in the prevalence of clinical isolates of bla VIM -producing P. aeruginosa in Japan (14); however, the origin of the bla VIM in this study remains uncertain. The RPKM value of bla IMP in this study was higher than that of bla CTX-M , while fewer patients had MBL producers than ESBL producers. This finding may be attributed to the fact that many MBL producers, such as P. aeruginosa, Acinetobacter spp., and Aeromonas spp., are opportunistic pathogens with wide environmental distribution, including in wastewater, and are not part of the human flora, unlike Enterobacterales, to which ESBL producers mainly belong (15,16). Contamination of wastewater systems with ARB indicates a potential transmission risk to inpatients (17). Monitoring sewage for MBL genes in regions where MBL-type carbapenemases are endemic may provide insight into the extent of hospital water contamination by MBL-producing bacteria.
Among the colistin-resistance genes, mcr-1, which is the predominant mcr gene worldwide (18), was not detected. xHYB primarily detected mcr-5, as well as mcr-3 and mcr-9, although at low RPKM values. Although rare in Japan, clinical isolates carrying mcr-5 and mcr-9 were detected in a nationwide survey (19). Since strains carrying mcr-5 and mcr-9 may be susceptible to low colistin concentrations (19), monitoring their dissemination within hospitals using culture-independent methods could help prevent their spread. Research Article mSphere xHYB also detected resistance genes for quinolone (qnr), aminoglycosides (aac(6′)-Ib and aph), macrolides (ermB and ermF), and sulfonamides (sul1) at significantly higher RPKM values than those obtained using mDNA-seq. The most significant quinolone resistance mechanism in Enterobacterales involves the accumulation of mutations in the quinolone resistance-determining regions of DNA gyrase and topoisomerase IV, which cannot be detected by either mDNA-seq or xHYB (20,21); however, the wide spread emergence of plasmid-mediated quinolone resistance (PMQR) genes, including aac(6′)-Ib-cr, a variant of aac(6′)-Ib, and qnr, has become a worldwide concern (22). The abundance of PMQR and macrolide resistance genes in sewage reflects the respective clinical use of quinolones and macrolides (21). The presence of sul1 in wastewater strongly correlates with anthropogenic inputs and is associated with horizontal gene transfer (HGT) (23). Therefore, monitoring effluent ARGs may help assess the extent of hospital dissemination and estimate antimicrobial usage and the rate and extent of HGT.
Although we infrequently detected mecA in hospital effluent, the number of patients with MRSA exceeded the number of patients carrying ESBL or MBL producers. S. aureus is a commensal bacterium colonizing skin and nasal passages that is less prevalent in gastrointestinal and genitourinary tracts than Enterobacterales (24). Thus, surveying resistance genes in sewage may be insufficient for monitoring inpatient spread of pathogens not typically found in gastrointestinal or genitourinary tracts.
The study found no VRE clinical isolates, and vanA was not detected in the hospi tal effluent; however, vanB was detected by the xHYB method. VRE strains carrying vanA are resistant to vancomycin and teicoplanin, while strains carrying vanB are resistant to vancomycin and susceptible to teicoplanin (25), rendering them undetect able through conventional drug-susceptibility testing methods. As such, they may go clinically unrecognized.
This study has several limitations. First, the xHYB-detectable ARGs are limited to those in the QIAseq xHYB AMR Panel, and the xHYB method is unable to distinguish the number of reads for variants within the same group, such as bla CTX-M-14 and bla CTX-M-27 , bla IMP-1 and bla IMP-6 , and aac(6′)-Ib and aac(6′)-Ib-cr. However, the performance of xHYB seems sufficient for comprehensive detection of ARGs in hospital effluent, since the 2,786 target ARGs cover 93.1% of those registered in the NCBI/ResFinder database, and Research Article mSphere those remaining are not worldwide concerns. Second, this study did not evaluate the residual antimicrobials in hospital effluent that may influence the abundance of AMR in wastewater (1,21). Third, as the number of inpatients with ARB was analysed retrospec tively, it may be an underestimate since not all patients underwent available culture tests and only prevalent ARB were targeted. Fourth, the culture-independent methods including mDNA-seq and xHYB are unable to determine the specific bacterial strain responsible for the detected ARGs. Nonetheless, correlations were observed between the numbers of inpatients with ARB and the ARG RPKM values in hospital effluent over time. In conclusion, this study demonstrates that xHYB appropriately monitors ARGs in hospital effluent for sensitive identification of nosocomial AMR dissemination. ARG surveillance in hospital effluent using the highly sensitive and specific xHYB method could also improve our understanding of the emergence and spread of AMR within a hospital.

mDNA-seq analysis of water samples
mDNA-seq libraries were prepared using the QIAseq FX DNA library kit (Qiagen, Hilden, Germany) with an index for each sample, followed by short-read sequencing using the NexSeq 500 platform (2 × 150-mer paired-end) (Illumina, San Diego, CA, USA). Adapters and low-quality sequences were trimmed using Sickle version 1.33 (https://github.com/ najoshi/sickle) with the following parameters: average quality threshold "-q 20" and minimum length threshold "-l 40" Subsequently, mDNA-seq analysis was performed using clean reads for homology searches without de novo assembly.

Resistome analysis
In subsequent analyses, mDNA-seq was performed using clean reads for homology searches without de novo assembly. Before resistome analysis, an ARG database was constructed from the National Center for Biotechnology Table S1 shows the xHYB and mDNA-seq RPKM values of the ARGs in AMROTU. We also compared ARGs with high environmental fitness commonly associated with mobile genetic elements, as suggested by Berendonk et al.: bla CTX-M , bla TEM , bla NDM-1 , bla VIM , bla KPC , qnrS, aac-(6')-Ib, aph, ermB, ermF, tetM, sul1, sul2, vanA, and mecA (27). bla IMP , the most prevalent MBL gene in clinical isolates in Japan (28); mcr, which confers resistance to colistin, a last-resort antimicrobial (29); and vanB, found in most clinically important VRE like vanA (30), were also considered.

Hospital microbiology database
From the hospital microbiology database, we collected monthly data on clinical specimens of ceftriaxone-resistant ESBL-producing bacteria (Escherichia coli, Klebsiella pneumoniae, Klebsiella oxytoca, and Proteus mirabilis) (8), MBL-producing gram-nega tive bacteria, MRSA, and VRE. Species identity and antimicrobial susceptibility were determined using the Vitek-2 system (Siemens Healthcare Diagnostics Japan, Tokyo, Japan). MBL production was detected in gram-negative bacteria resistant to both ceftazidime and sulbactam-cefoperazone by the sodium mercaptoacetic acid disk test (31). All specimen types were included (blood, urine, fecal, other fluid and tissue, and indwelling plastic); duplicate specimens from the same patient collected within a month were excluded. The monthly number of patients with ARB isolates was compared to the monthly average RPKM of bla CTX-M , bla IMP , mecA, vanA, and vanB.

Statistical analysis
Fisher's exact test was used to compare categorical variables. Welch's t-test and Tukey's honestly significant difference test were used to compare continuous variables between two groups and more than two groups, respectively. All analyses were performed using JMP Pro 16 (SAS Institute Japan, Tokyo, Japan). A P value <0.05 was considered statisti cally significant. of data. All authors discussed the results and contributed to the writing of the final manuscript.
The authors declare no conflict of interest.

ETHICS APPROVAL
This study was approved by the Institutional Review Board of Tohoku University Graduate School of Medicine (IRB Number: 2020-1-061).

ADDITIONAL FILES
The following material is available online.