An ISO-certified genomics workflow for identification and surveillance of antimicrobial resistance

Abstract

Realising the promise of genomics to revolutionise identification and surveillance of antimicrobial resistance (AMR) has been a long-standing challenge in clinical and public health microbiology. Here, we report the creation and validation of abritAMR, an ISO-certified bioinformatics platform for genomics-based bacterial AMR gene detection. The abritAMR platform utilises NCBI's AMRFinderPlus, as well as additional features that classify AMR determinants into antibiotic classes and provide customised reports. We validate abritAMR by comparing with PCR or reference genomes, representing 1500 different bacteria and 415 resistance alleles. In these analyses, abritAMR displays 99.9% accuracy, 97.9% sensitivity and 100% specificity. We also compared genomic predictions of phenotype for 864 Salmonella spp. against agar dilution results, showing 98.9% accuracy. The implementation of abritAMR in our institution has resulted in streamlined bioinformatics and reporting pathways, and has been readily updated and re-verified. The abritAMR tool and validation datasets are publicly available to assist laboratories everywhere harness the power of AMR genomics in professional practice.

Fig. 1. Overview of abritAMR pipeline.

Assembled bacterial genomic sequence data (short- or long-read or hybrid assemblies, fasta format) are inputted in to the abritAMR pipeline to identify acquired AMR genes and mutations. The pipeline implements the AMRFinderPlus tool, identifying AMR genes (BLASTx search), and optionally identifies mutations associated with AMR for specified species where mutational data are available. Identified AMR genes + /− mutations are then binned into functional AMR classes according to the classification database (Detailed Report output, Fig. 2 and Supplementary Data 1). An additional step then applies reporting logic tailored to local requirements to produce per-isolate reports with acquired AMR genes, excluding common intrinsic AMR genes from being reported in specific species (Final AMR Gene Report), and phenotypic inference for specified species, where validated (Inferred Antibiogram Report, Fig. 2 and Supplementary Data 1). AMR, antimicrobial resistance.

Fig. 2. Examples of abritAMR pipeline outputs.

This figure demonstrates abritAMR features and outputs across four different species (genome sequence A, Escherichia coli; genome sequence B, Klebsiella pneumoniae; genome sequence C, Acinetobacter baumannii; genome D, Salmonella enterica). The horizontal lanes represent the different output stages: top lane, raw AMRFinderPlus output; middle lane, abritAMR Detailed Report output (aligned to and binned by Enhanced Subclass from abritAMR’s classification database); bottom lane, abritAMR’s Final AMR Gene Report (genomes A-C) after application of tailored reporting logic to meet local requirements, and Inferred Antibiogram Report (genome D), currently validated for Salmonella spp. Key features include: (i) simplification of mechanism or drug class bins, (ii) identification and separation of high-priority AMR groups (such as notifiable AMR mechanisms) from lower-priority groups, (iii) identification of clinically-relevant AMR mechanisms within a drug class (e.g. separation of ESBLs and AmpCs from genes encoding first-generation cephalosporin resistance; separation of metallo-beta-lactamase (MBL) carbapenemases and other carbapenemases due to differences in patient treatment), and (iv) application of tailored reporting logic to separate reportable and non-reportable genes according to local requirements, thus de-cluttering the report for clinicians and public health teams. Note, only ‘Exact’ matches (100% sequence identity and coverage) and ‘Close’ matches (90-<100% identity and 90-<100% sequence coverage) from the AMRFinderPlus tool are reported by abritAMR. Abbreviations: MBL, metallo-beta-lactamase; ESBL, extended-spectrum beta-lactamase.

Fig. 3. Validation of abritAMR outputs compared to PCR and synthetic read data.

A Validation compared to PCR data – Assembled short-read sequence data from bacterial isolates are run through the abritAMR pipeline and compared to multiplex real-time PCR results for the same AMR genes. B Synthetic data – where no validation dataset or PCR assay was available for comparison, synthetic read data were generated from publicly-available closed reference genomes by fragmentation with the art-illumina tool (using the error profile from local sequencing platforms) to mimic library preparation from bacterial DNA. Synthetic reads then underwent the same analytical processes as for usual usage (genome assembly and input into abritAMR pipeline). These results were then compared to AMRFinderPlus results on the complete bacterial reference genomes, minimizing the risk of discordant results due to disparities between AMR databases. Abbreviations: AMR, antimicrobial resistance; PCR, polymerase chain reaction. Source data are provided as a Source Data file.

Fig. 4. Performance of abritAMR pipeline to detect AMR mechanisms compared to PCR.

Each panel details the identification of AMR mechanisms by abritAMR compared to the ‘gold standard’ multiplex PCR assays used in our laboratory. True positive, detected by both PCR and abritAMR; true negative, not detected by either PCR or abritAMR; false positive, detected by abritAMR but not multiplex PCR, and within the known range of the PCR assay; false negative, detected by multiplex PCR but not by abritAMR. Panel A, mec genes compared to multiplex PCR (mecA/mecC, no mecC detected by either method). Panel B, van genes compared to multiplex PCR (vanA/B/C). Panel C1, detection of genes within carbapenemase and ESBL gene families compared to multiplex PCR panel (AusDiagnostics CRE panel); asterisks represent groups where discrepancies were identified, and expanded out in Panel C2 to show the specific gene discrepancies between the two methods. Abbreviations: AMR antimicrobial resistance, PCR polymerase chain reaction, CP carbapenemase, ESBL extended-spectrum beta-lactamase. Source data are provided as a Source Data file.

Fig. 5. Performance of abritAMR pipeline compared to synthetic read data.

This figure shows the presence or absence of 415 AMR genes across 321 genomes identified by abritAMR (performed on assembled synthetic read data) compared to AMRFinderPlus on complete genome sequences (‘gold standard’). True positive, detected by AMRFinderPlus and abritAMR; true negative, not detected by either AMRFinderPlus or abritAMR; false positive, detected by abritAMR but not by AMRFinderPlus; false negative, detected by AMRFinderPlus but not by abritAMR. Panel A, abritAMR results by enhanced subclasses (further grouped to simplify visualisation); asterisks represent classes with any discrepant result (false positive or false negative), and are examined in more detail in Panel B. Panel B shows a detailed view of genes with discrepant results within each subclass. A full list of AMR subclasses included in the validation set can be found in Supplementary Table 3. Abbreviations: EBL extended-spectrum beta-lactamase, Rmt ribosomal methyltransferase, Agly aminoglycoside, Macro., Linco. & Strepto, macrolides, lincosamides and streptogramins (combined class); Carb., carbapenemase; Pen., penicillin resistance (S. aureus); Other BL, other beta-lactamase; Other discrep. Agly, other discrepant aminoglycoside subclass; Ami./Quin., amikacin/quinolone subclass; Quin., quinolones; Mac., Lin. & Strep., macrolides, lincosamides and streptogramins; Sulfon., sulfonamides; Rif., rifampicin; Tet., tetracyclines. Source data are provided as a Source Data file.

Fig. 6. Performance of inferred phenotype from abritAMR compared to antimicrobial susceptibility testing (AST).

Classification of genotype (AMR mechanism detection) compared to the ‘gold standard’ phenotypic AST for each isolate and antimicrobial. True positive, genotypic and phenotypic resistance; true negative, no AMR mechanisms detected in phenotypically susceptible isolate; false positive, AMR mechanism identified in phenotypically susceptible isolate; false negative, no AMR mechanisms detected in phenotypically resistant isolate. For ciprofloxacin, ‘true positive’ defined as concordant intermediate or resistant results (phenotype and genotype). Abbreviations: AST, antimicrobial susceptibility testing; AMR, antimicrobial resistance; Trim-sulfa, trimethoprim-sulfamethoxazole. Source data are provided as a Source Data file.