Metagenomic nanopore sequencing for analysis of orthopaedic device-related infection - Teresa Street

Teresa Street (University of Oxford) began her talk by highlighting the extent of infections - a "devastating side-effect" - following joint replacement surgery. Displaying data from the National Joint Registry, she showed that the proportion of patients who develop an infection, or suspected infection, after a joint replacement implant procedure can reach as high as 25% (for knee, elbow, or ankle). Orthopaedic device-related infections, Teresa explained, are currently difficult to treat, often requiring revision surgery and long courses of antibiotics. The gold-standard for the process of identifying the cause of an infection currently requires culturing the pathogen from periprosthetic tissue, but this is relatively insensitive, enabling identification in only ~65% of cases. Tests are often culture negative: for example, it may not be possible to culture a bacterium with particular nutritional requirements.

Teresa displayed a typical workflow for samples for surgery which are sent to their microbiology lab, focusing on bacterial infections. Crucially, the methods used currently take a long time as they rely on culturing of the samples: tissue, biopsy, or bone samples are cultured for 18-24 hours, joint fluid for up to 5 days, and bacteria on devices or metalwork for up to 10 days. These samples are then analysed with a wide range of tests.

Teresa then asked: "What about culture-free diagnostics?" This is the fundamental principle of metagenomics: taking DNA straight from clinical samples, without the need to culture samples. She discussed some of the currently used molecular methods, which generally rely on PCR. Specific PCR can be used to target specific bacterial species; however, this only enables detection of species for which primers are present. Broad-range 16S PCR, followed by sequencing, can detect all bacterial species, but problems can arise from contamination. Mass spectrometry, meanwhile, helps identify coagulase-negative Staphylococcus species, but isn't strictly culture-free. Teresa then showed how a handful of publications have used short-read sequencing for clinical metagenomics to identify the cause of orthopaedic device-related infections. Teresa and her team instead decided to use long-read nanopore sequencing test this in a proof-of-principle study; their work was published last year.

For the pilot study, DNA was first extracted from 7 culture-positive and 2 culture-negative sonication fluid samples; the DNA was amplified and the library prepared for sequencing with the Ligation Sequencing Kit. The samples were then sequenced on the MinION, with one sample per MinION Flow Cell. The team developed an analysis pipeline, CRuMPIT (Clinical Real-time Metagenomics Pathogen Identification Test), to identify species that represented >10% of bacterial bases and >1% of mapped bases as "present"; this uses Centrifuge for taxonomic ID and minimap2 for alignment to reference genomes. In the resulting data, for every culture-positive sample, the species identified by culture matched the majority species identified by sequencing. Teresa also noted how "in a couple of cases, we actually improved on the culture results": in one case, two Bacillus species were identified where culturing only identified one, and in another sample, an additional bacterium representing a "plausible cause of infection" was revealed in the long-read data. The method enabled rapid identification of the species, with the majority species evident in the data as early as ten minutes into sequencing. Teresa noted that the samples used were originally extracted with short-read sequencing in mind, and that extracting DNA to preserve longer fragments would enable identification of genomic context in sequencing. She also highlighted how in this initial study, the data was dominated by human DNA from the samples; furthermore, their use of one sample per flow cell proved uneconomical.

Teresa then discussed how the team optimised their workflow to tackle these limitations. In the initial study, 80-97% of sequenced bases were classified as human. To reduce the presence of human DNA, the team utilised a saponin-based treatment from Charalampous et al. (Nature Biotech, 2019) to selectively lyse human cells, degrade the released DNA and then take intact bacteria into DNA extraction. The team used ethanol precipitation DNA extraction to preserve high molecular weight DNA, enabling longer reads and providing more genomic context. As the depletion of human DNA reduced the overall DNA yield, they then used the PCR Barcoding Kit to amplify the bacterial DNA prior to sequencing, also based on the method from the same paper. The addition of PCR barcodes enabled multiplex sequencing, improving the cost per sample.

The optimised workflow was used to sequence 85 individual patient sonication fluid samples; of these, 47 bacterial isolates had been identified via culture. Samples were sequenced in multiplex on the GridION platform, with 4-7 samples sequenced per flow cell. In the resulting dataset, 40 of the 47 culture-positive samples were identified to genus level and 36 to species level. A further 5 were detected below the 10% imposed read threshold, one was misclassified and two were not detected. They discovered that the sample that was misclassified resulted from the lack of a full reference in the database used. For the remaining three samples, where Enterobacter cloacae was previously classified, upon sequencing, E. hormaechei was identified. Furthermore, three additional species were identified which were not observed in culture, all of which represented plausible causes of infection. In the culture-negative samples, meanwhile, three putative pathogenic species were identified, two of which corresponded to culturing results from tissue collected during the same surgery. Teresa pointed out that three identified species likely represent "kitome" contaminants, whilst three more may have resulted from skin flora or reagent contamination. The 36 species identified obtained >70% genome coverage; Teresa pointed out that depleting human DNA via an optimised method using 5% saponin improved the breadth and depth of coverage.

The team are now in the process of investigating antimicrobial resistance genes in the samples. Teresa described how they were able to detect genes accounting for the resistance phenotypes observed by susceptibility testing for multiple antibiotics, with a few discordant results seen. Using their optimised method, they are able to generate "near whole-genome sequences".

Teresa concluded her discussion of nanopore metagenomic sequencing for orthopaedic device-related infection with: "it looks promising!"; she highlighted how near-whole genomes were sequenced, with pathogen identification taking place in real-time and genes conferring antimicrobial resistance successfully identified.