Host-response-based diagnostics can improve the accuracy of diagnosing bacterial and viral infections, thereby reducing inappropriate antibiotic prescriptions. However, the existing cohorts with limited sample size and coarse infections types are unable to support the exploration of an accurate and generalizable diagnostic model. Here, we curate the largest infection host-response transcriptome data, including 11,247 samples across 89 blood transcriptome datasets from 13 countries and 21 platforms. We build a diagnostic model for pathogen prediction starting from a pan-infection model as foundation (AUC = 0.97) based on the pan-infection dataset. Then, we utilize knowledge distillation to efficiently transfer the insights from this "teacher" model to four lightweight pathogen "student" models, i.e., staphylococcal infection (AUC = 0.99), streptococcal infection (AUC = 0.94), HIV infection (AUC = 0.93), and RSV infection (AUC = 0.94), as well as a sepsis "student" model (AUC = 0.99). The proposed knowledge distillation framework not only facilitates the diagnosis of pathogens using pan-infection data, but also enables an across-disease study from pan-infection to sepsis. Moreover, the framework enables high-degree lightweight design of diagnostic models, which is expected to be adaptively deployed in clinical settings.

Objective: To integrate an enhanced molecular diagnostic technique to develop and validate a machine-learning model for diagnosing sepsis. Methods: We prospectively enrolled patients suspected of sepsis from August 2021 to August 2023. Various feature selection algorithms and machine learning models were used to develop the model. The best classifier was selected using 5-fold cross validation set and then was applied to assess the performance of the model in the testing set. Additionally, we employed the Shapley Additive exPlanations (SHAP) method to illustrate the effects of the features. Results: We established an optimized mNGS assay and proposed using the copies of microbe-specific cell-free DNA per milliliter of plasma (CPM) as the detection signal to evaluate the real burden, with strong precision and high accuracy. In total, 237 patients were eligible for participation, which were randomly assigned to either the training set (70 %, n = 165) or the testing set (30 %, n = 72). The random forest classifier achieved accuracy, AUC and F1 scores of 0.830, 0.918 and 0.856, respectively, outperforming other machine learning models in the training set. Our model demonstrated clinical interpretability and achieved good prediction performance in differentiating between bacterial sepsis and non-sepsis, with an AUC value of 0.85 and an average precision of 0.91 in the testing set. Based on the SHAP value, the top nine features of the model were PCT, CPM, CRP, ALB, SBPmin, RRmax, CREA, PLT and HRmax. Conclusion: We demonstrated the potential of machine-learning approaches for predicting bacterial sepsis based on optimized mcfDNA sequencing assay accurately.

To the Editor: Sepsis is a clinical syndrome characterized by life-threatening organ dysfunction caused by a dysregulated host response to infection. Early intervention with antibiotics, intravenous fluids, and other supportive measures can significantly improve the chances of recovery. For every hour of delay in diagnosing and treating patients with septic shock, there is a 7.6% increase in the mortality rate.[1] Despite advances in diagnostic technology, clinicians are still unable to detect the origin of sepsis in approximately one-quarter (28%) of patients with septic shock by the end of their intensive care unit (ICU) stay.[2] Therefore, a more rapid test to detect a broad spectrum of pathogens is essential for the diagnosis and treatment of sepsis. Although metagenomic next-generation sequencing (mNGS) is a novel solution for pathogen detection, it is labor-intensive and time-consuming. A typical mNGS experiment takes ~24 h, which is significantly longer than that of serological and polymerase chain reaction (PCR)-based tests. Rapid mNGS is required by clinicians to obtain accurate results within a rapid timeframe. A user-friendly and rapid procedure would aid clinicians in decision-making, which may eventually benefit patients. Therefore, we designed an mNGS workflow based on the Illumina platform with a theoretical turnaround time (TAT) of 7 h. This study was approved by the Research Ethics Board of the Peking University People's Hospital (No. 2021PHB410-001). Informed consent was obtained from all patients that were enrolled in the study. To expedite a standard mNGS procedure with a ~24 h turnaround [Figure 1A], we modified a previously validated experimental protocol[3] and designed an ultra-rapid mNGS workflow according to the following: (1) automation in nucleic acid extraction and library preparation through the use of a cartridge-based point-of-care device.[3] The device comprised four chambers, each of which was equipped with liquid handling, temperature control, and magnetic separator modules to facilitate DNA extraction, enzymatic fragmentation, end repair, dA-tailing, adaptor ligation, and library purification; (2) PCR-free library preparation in which only one nucleic acid purification step was needed[3]; (3) Miniseq rapid reagent kit was used (~25 million reads, 3 pmol/L of pooled library input) and 50 base pairs were sequenced instead of 100; (4) one plasma and one negative control (NC) were sequenced in each run, simplifying the pooling processes; and (5) the bioinformatics pipeline was optimized to reduce runtime. The theoretical TAT for the ultra-rapid mNGS was 7 h, representing one of the fastest mNGS tests performed on the Illumina platform [Figure 1B]. Microbial reads identified from a library were reported if: (1) the sequencing data passed quality control filters (library concentration >50 pmol/L, Q20 >85%, Q30 >80%); (2) the NC in the same sequencing run does not contain the species or reads per million (RPM) (sample)/RPM (NC) is ≥5.Figure 1: Experimental steps for ultra-rapid (A) and standard mNGS (B). Contingency tables for the ultra-rapid mNGS results compared to the initial culture, CMTs, and clinical adjudications (C). Sankey diagram showing the clinical actions in response to mNGS results and patient outcomes (D). CMTs: Conventional microbiological tests; mNGS: Metagenomic next-generation sequencing; PPA: Positive percentage agreement. PPV: Positive percentage; QC: Quality control; qPCR: Quantitative polymerase chain reactionTo explore whether the ultra-rapid mNGS using blood samples has real-world benefits, particularly in the ICU where patients with sepsis have exhibited an increase in mortality with a delay in effective antimicrobial initiation, 36 patients were enrolled from the ICU department at Peking University People's Hospital, Beijing, China according to the following criteria: (1) 18 years and older, suspected of sepsis (body temperature >38°C or <36°C with elevated serum C-reactive protein [CRP] or procalcitonin [PCT] levels); (2) sequential organ failure assessment (SOFA) score of +2 or higher; (3) expected survival time of ≥8 h; and (4) providing informed consent [Supplementary Figure 1, https://links.lww.com/CM9/C38]. Bilateral double bottles (aerobic and anaerobic) were collected for blood cultures. In some cases, specimens other than peripheral blood were sent for culture. Positive culture results were recorded within 3 days before or after mNGS for analytical performance evaluation. Additional microbiological tests were ordered by clinicians when deemed necessary, including the interferon-gamma release assay (IGRA), acid-fast stain, Gram stain, cytomegalovirus (CMV)/Epstein–Barr virus (EBV) quantitative real-time PCR (qPCR), β-D-glucan test (G test), galactomannan test (GM test), cryptococcal capsular antigen (CrAg) test, and influenza A/B antigen test. Moreover, an in-house standard mNGS test that utilized Nextseq 550Dx (Illumina, California, San Diego, USA) with a 24-h TAT was performed in five cases (designated as routine mNGS in the manuscript). The details and results of these tests are presented in Supplementary Table 1, https://links.lww.com/CM9/C38. The clinical characteristics of the enrolled patients are summarized in Supplementary Tables 2 and 3, https://links.lww.com/CM9/C38. All patients were administered empirical antibiotics prior to microbiological testing. Three different reference standards were used to evaluate the diagnostic accuracy of the ultra-rapid mNGS: (1) blood cultures resulted in a positive percentage agreement (PPA) and negative percentage agreement (NPA) of 72.73% and 12.00%, respectively; (2) a composite standard that included all conventional microbiological tests (CMTs) that generated a PPA and NPA of 44.83% and 0%, respectively; and (3) clinical adjudications based on the examination of medical records, imaging scans, microbiological findings, and responses to antibiotics (whether symptoms improved or exacerbated), which resulted in a PPA and NPA of 82.86% and 100.00%, respectively [Figure 1C]. Combined with laboratory and clinical data, the pathogen results were classified as clinically relevant (definite, probable, and possible) or clinically irrelevant (unlikely) according to the composite microbiological and clinical criteria outlined in the Karius test.[4] The microorganisms detected using different methods in each patient are shown in Supplementary Figure 2A, https://links.lww.com/CM9/C38. Compared with a routine mNGS, the ultra-rapid mNGS exhibited the same results in 3/5 cases and detected more bacteria in blood or sputum by culture in 2/5 cases. The time from sample collection to the results of all microbiological tests was recorded [Supplementary Figures 1 and 2B and Supplementary Table 1, https://links.lww.com/CM9/C38]. The average TAT for the ultra-rapid mNGS was 10.53 h (minimum 7.4 h), which was ostensibly faster than other microbiological methods, especially culture (average TAT 97.72 h). In a real clinical setting, qPCR is performed no quicker than a G-test (average TAT 26.66 h vs. 19.87 h). The delay in TAT is caused by experimental scheduling; upon sample arrival in a clinical laboratory, technicians need to wait for more samples to arrive to start batch processing, which is also the case for routine mNGS in which 10–20 samples are handled simultaneously (average TAT, 55.4 h). We also investigated whether faster mNGS reporting could lead to better antibiotic management. To this end, we analyzed all cases and categorized the clinicians' actions into (1) escalation of antibiotics, (2) de-escalation of antibiotics, (3) increase in the types of antibiotics, (4) reduction in the types of antibiotics (in cases of combination antibiotic therapy), (5) validation/confirmation of the empirical therapy and no change in antibiotics, and (6) irrelevant results and no change in antibiotics. As shown in Supplementary Figure 3A, https://links.lww.com/CM9/C38, the impact of mNGS was the validation of empirical therapy (n = 14), followed by additional antibiotics (n = 10), and fewer antibiotics (n = 9), antibiotic escalation (n = 2), and no change (n = 1). Next, we evaluated whether these clinical managements affected patient outcomes. Among the 36 mNGS reports, 30 (83%) were deemed clinically relevant based on a retrospective review of medical records. On day 30 following the mNGS test, 17 of the 30 patients survived and 13 (43%) died, compared to four and two (33%) of six patients whose mNGS results were considered irrelevant, respectively [Figure 1D and Supplementary Figure 3B, https://links.lww.com/CM9/C38]. However, 9/10 of the patients whose empirical antibiotics were validated by mNGS survived, which was higher than that of other types of clinical actions [Figure 1D]. Lastly, we analyzed the monetary expenditure associated with antibiotic use during the 24-h time window before and after ultra-rapid mNGS testing. A change in antibiotic costs occurred in 20 of the 36 patients. A total reduction of 10,909.52 Chinese Yuan (~1558.5 US dollars) was observed in 15 cases [Supplementary Figure 4, https://links.lww.com/CM9/C38]. In five cases, an increase in antibiotic charge was seen (1413.12 Chinese Yuan, ~201.9 US dollars), largely due to the use of additional antibiotics targeting pathogens identified by mNGS that were not covered by empirical treatment [Supplementary Table 1, https://links.lww.com/CM9/C38]. To date, the fastest TAT of mNGS was 6 h, which was performed on a Nanopore sequencer.[4] The cost of Nanopore-based mNGS was approximately $300/sample, as compared to $100/sample for the Illumina platform.[5] Moreover, the sequencing output of the Nanopore was lower than that of the Illumina with a higher error rate in base calling. Therefore, it would be cost-effective if an Illumina-based mNGS could be expedited to a level comparable to that of the Nanopore. Our work demonstrated that 7 h-mNGS was plausible using an Illumina sequencer. The clinical implementation of this workflow yielded an average sample-to-result time of 10.6 h, with a minimum of 7.4 h [Supplementary Table 3, https://links.lww.com/CM9/C38]. We employed three different standards to evaluate the diagnostic performance of ultra-rapid mNGS [Figure 1C] and observed high PPA and NPA when using clinical adjudication as a reference, in which clinicians reached a definitive microbiological diagnosis based on a more systematic, thorough, albeit subjective review of clinical cases, which reflected a more accurate account of clinical situations. Our study has several limitations. First, no control group was included. Second, the sample size was small and biased towards elderly Han Chinese males, which prevented a more comprehensive evaluation of the technique in broader ethnic and age groups. Third, the study was conducted in the ICU of a single tertiary hospital; therefore, we could not fully assess the cost-benefit from a wider perspective, especially in low-resource settings. The ultra-rapid mNGS workflow can be easily implemented in a clinical setting with the help of a point-of-care automation device; thus, only one person is required to complete the procedure. Currently, the protocol can only handle one plasma sample using the MiniSeq sequencer owing to the limited data output. Other platforms with higher throughputs and faster sequencing time can accommodate more samples in a single run. Acknowledgement The Miniseq rapid reagent kits in this study were provided by Illumina (China) Scientific Co., Ltd. This study was supported by the National Natural Science Foundation of China (No. 82241048) and Beijing Major Epidemic Prevention and Control Key Specialty Project-Medical Laboratory Excellence Project (2022). Conflicts of interest Z. Du, J. Wang, and C. Liu are employees of Hangzhou Matridx Biotechnology Co., Ltd. The rest of the authors declare no conflict of interest.

We read with interest the study by Peng and colleges showing the performance of metagenomic Next-Generation Sequencing (mNGS) in detecting pulmonary pathogens was not superior to conventional microbiological tests (CMT) in a cohort of 101 immunocompromised adults. 1 Peng J.M. Du B. Qin H.Y. Wang Q. Shi Y. Metagenomic next-generation sequencing for the diagnosis of suspected pneumoni in immunocompromised patients. J Infect. 2021; 82 (AprPubMed PMID: 33609588): 22-27 Abstract Full Text Full Text PDF PubMed Scopus (17) Google Scholar Indeed, although mNGS enables untargeted “pan-pathogen” detection that covers a broad array of microorganisms with known genomic sequences, clinical application of this test has encountered challenges. For instance, the diagnostic sensitivity is affected by the quantity of host DNA, which varies considerably from sample to sample. 2 Schlaberg R. Chiu C.Y. Miller S. Procop G.W. Weinstock G. Professional practice CValidation of metagenomic next-generation sequencing tests for universal pathogen detection. Arch Pathol Lab Med. 2017; 141 (JunPubMed PMID: 28169558): 776-786 Crossref PubMed Scopus (216) Google Scholar Host cell depletion techniques have been used to improve the sensitivity of mNGS but may lead to unspecific removal of pathogens. 3 Charalampous T. Kay G.L. Richardson H. Aydin A. Baldan R. Jeanes C. et al. Nanopore metagenomics enables rapid clinical diagnosis of bacterial lower respiratory infection. Nat Biotechnol. 2019; 37 (JulPubMed PMID: 31235920): 783-792 Crossref PubMed Scopus (186) Google Scholar In our independent study, we developed a spike-in internal control to assess the abundance of host and microbial DNA in bronchoalveolar lavage fluid (BALF) and evaluated the analytical and diagnostic performance of mNGS with and without host depletion in a cohort of 205 patients suspected of lower respiratory tract infections (Supplementary Table 1).

BackgroundDifferential diagnosis of patients with suspected infections is particularly difficult, but necessary for prompt diagnosis and rational use of antibiotics. A substantial proportion of these patients have non-infectious diseases that include malignant tumors. This study aimed to explore the clinical value of metagenomic next-generation sequencing (mNGS) for tumor detection in patients with suspected infections. MethodsA multicenter, prospective case study involving patients diagnosed with suspected infections was conducted in four hospitals in Shanghai, China between July 2019 and January 2020. Based upon mNGS technologies and chromosomal copy number variation (CNV) analysis on abundant human genome, a new procedure named Onco-mNGS was established to simultaneously detect pathogens and malignant tumors in all of the collected samples from patients. ResultsOf 140 patients screened by Onco-mNGS testing, 115 patients were diagnosed with infections; 17 had obvious abnormal CNV signals indicating malignant tumors that were confirmed clinically. The positive percent agreement and negative percent agreement of mNGS testing compared to clinical diagnosis was 53.0% (61/115) and 60% (15/25), vs. 20.9% (24/115) and 96.0% (24/25), respectively, for conventional microbiological testing (both P <0.01). Klebsiella pneumoniae (14.8%, 9/61) was the most common pathogen detected by mNGS, followed by Escherichia coli (11.5%, 7/61) and viruses (11.5%, 7/61). The chromosomal abnormalities of the 17 cases included genome-wide variations and local variations of a certain chromosome. Five of 17 patients had a final confirmed with malignant tumors, including three lung adenocarcinomas and two hematological tumors; one patient was highly suspected to have lymphoma; and 11 patients had a prior history of malignant tumor. ConclusionThis preliminary study demonstrates the feasibility and clinical value of using Onco-mNGS to simultaneously search for potential pathogens and malignant tumors in patients with suspected infections.

Background Metagenomic next-generation sequencing (mNGS) offers the promise of unbiased detection of emerging pathogens. However, in indexed sequencing, the sequential paradigm of data acquisition, demultiplexing, and analysis restrain read assignment in advance and real-time analysis, resulting in lengthy turnaround time for clinical metagenomic detection. Methods We described the utility of internal-index adaptors with different lengths of barcode in multiplex sequencing. The base composition for each position within these adaptors was well-balanced to ensure nucleotide diversity and optimal sequencing performance and to achieve the early assignment of reads by first sequencing the barcodes. Combined with an automated library preparation device, we delivered a rapid and real-time bioinformatics pathogen identification solution for the Illumina NextSeq platform. The diagnostic performance was evaluated by testing 153 lower respiratory tract specimens using mNGS in comparison to culture, 16S/internal transcribed spacer amplicon sequencing, and additional PCR-based tests. Results By calculating the average F1 scores of all read lengths under different threshold values, we established the optimal threshold for pathogens identification, and found that 36 bp was the optimal shortest read length for rapid mNGS analysis. Rapid detection had a negative percentage agreement and positive percentage agreement of 100% and 85.1% for bacteria and 97.4% and 80.3% for fungi, when compared to a composite standard. The rapid mNGS solution enabled accurate pathogen identification in about 9.1 to 10.1 h sample-to-answer turnaround time. Conclusions Optimized internal index adaptors combined with a real-time analysis pipeline provide a potential tool for a first-line test in critically ill patients.

Pythium insidiosum is a rare fungus-like pathogen that is known to cause pythiosis in mammals with high morbidity and mortality. Identification of the pathogen is essential for timely treatment and rational use of antibiotics. However, Pythium insidiosum is difficult to detect via conventional microbiological tests. The current gold standard is polymerase chain reaction, which is lacking in most hospitals since human pythiosis is rare in China. In this study, we used metagenomic Next-Generation Sequencing and identified Pythium insidiosum in a 56-year-old Chinese male who was hospitalized due to severe edema in the right lower limb with scattered darkening indurations. The patient had a history of cirrhosis and occupational exposure to swamp water. Serological level of immune biomarkers indicated immunodeficiency, and Proteinase 3-Anti-Neutrophil Cytoplasmic Antibody was positive. Surgical incision of the lesions revealed radiating and reticular cutaneous ulcers. Microbial infections were suspected but conventional tests failed to discover the etiology. Empirical use of penicillin, vancomycin, and ceftriaxone had no effect. As a result, the peripheral blood and tissue biopsies were sent for metagenomic Next-Generation Sequencing, which reported Pythium insidiosum. This finding was corroborated by pathological staining, whole-genome sequencing, and internal transcribed spacer sequencing. Notably, antifungal treatment was ineffective, but the patient responded well to oral trimethoprim–sulfamethoxazole, which may be due to the folp gene found in Pythium insidiosum genome. Our study prompts future studies to determine the optimal treatment of skin pythiosis.

Background: Lung biopsy tissue samples can be used for infection detection and cancer diagnosis. Metagenomic next-generation sequencing (mNGS) has the potential to further improve diagnosis. Methods: From July 2018 to May 2020, lung biopsy samples of 133 patients with suspected pulmonary infection or abnormal imaging findings were collected and subjected to clinical microbiological testing, Illumina and Nanopore sequencing to identify pathogens. The neural networks were pretrained by extracting features of human reads from 2,095 metagenomic next-generation sequencing results, and the human reads of lung biopsy samples were entered into the validated pipeline to predict the risk of cancer. Findings: Based on the pathogen-cancer detection pipeline, the Illumina platform showed 77·6% sensitivity and 97·6% specificity compared to the composite reference standard for infection diagnosis. However, the Nanopore platform showed 34·7% sensitivity and 98·7% specificity. mNGS identified more fungi, which was confirmed by subsequent pathological examination. M. tuberculosis complex was weakly detected. For cancer detection, compared with histology, the Illumina platform showed 83·7% sensitivity and 97·6% specificity, diagnosing an additional 36 cancer patients, of whom half had abnormal imaging findings (pulmonary shadow, space-occupying lesions, or nodules). Interpretation: For the first time, we have established a pipeline to simultaneously detect pathogens and cancer based on Illumina sequencing of lung biopsy tissue. This pipeline efficiently diagnosed cancer in patients with abnormal imaging findings. Funding: This work was supported by the National Key Research and Development Program of China and National Natural Science Foundation of China.

An acquired cholesteatoma generally occurs as a consequence of otitis media and eustachian tube dysfunction. Patients with acquired cholesteatoma generally present with chronic otorrhea and progressive conductive hearing loss. There are many microbes reportedly associated with acquired cholesteatoma. However, conventional culture-based techniques show a typically low detection rate for various pathogenetic bacteria and fungi. Metagenomic next-generation sequencing (mNGS), an emerging powerful platform offering higher sensitivity and higher throughput for evaluating many samples at once, remains to be studied in acquired cholesteatoma. In this study, 16 consecutive patients from January 2020 to January 2021 at the Second Affiliated Hospital of Zhejiang University School of Medicine (SAHZU) were reviewed. We detected a total of 31 microbial species in patients, mNGS provided a higher detection rate compared to culture (100% vs. 31.25%, p = 0.000034). As the severity of the patient's pathological condition worsens, the more complex types of microbes were identified. The most commonly detected microbial genus was Aspergillus (9/16, 56.25%), especially in patients suffering from severe bone erosion. In summary, mNGS improves the sensibility to identify pathogens of cholesteatoma patients, and Aspergillus infections increase bone destruction in acquired cholesteatoma.

AIMS:Metagenomic next-generation sequencing (mNGS) has been utilized for diagnosing infectious diseases. It is a culture-free and hypothesis-free nucleic acid test for diagnosing all pathogens with known genomic sequences, including bacteria, fungi, viruses and parasites. While this technique greatly expands the clinical capacity of pathogen detection, it is a second-line choice due to lengthy procedures and microbial contaminations introduced from wet-lab processes. As a result, we aimed to reduce the hands-on time and exogenous contaminations in mNGS. METHODS AND RESULTS:We developed a device (NGSmaster) that automates the wet-lab workflow, including nucleic acid extraction, PCR-free library preparation and purification. It shortens the sample-to-results time to 16 and 18·5 h for DNA and RNA sequencing respectively. We used it to test cultured bacteria for validation of the workflow and bioinformatic pipeline. We also compared PCR-free with PCR-based library prep and discovered no differences in microbial reads. Moreover we analysed results by automation and manual testing and found that automation can significantly reduce microbial contaminations. Finally, we tested artificial and clinical samples and showed mNGS results were concordant with traditional culture. CONCLUSION:NGSmaster can fulfil the microbiological diagnostic needs in a variety of sample types. SIGNIFICANCE AND IMPACT OF THE STUDY:This study opens up an opportunity of performing in-house mNGS to reduce turnaround time and workload, instead of transferring potentially contagious specimen to a third-party laboratory.

The clustered regularly interspaced short palindromic repeat (CRISPR)-associated endonuclease Cas13a can specifically bind and cleave RNA. After nucleic acid pre-amplification, bacterial Cas13a has been used to detect genetic mutations. In our study, using a transcription-mediated amplification together with Cas13a, we can isothermally amplify and detect mitochondrial point mutations under non-denaturing conditions from human genomic DNA. Unlike previous reports, we prepared CRISPR DNA with T7 promoter sequences and generated CRISPR RNA via transcription-mediated amplification instead of synthesizing and adding CRISPR RNA in a separate step. As a proof-of-concept, we showed that both m.1494C > T and m.1555A > G mutations were detected within 90 min. In addition, we explored various designs of CRISPR DNA to improve assay specificity, including the location and number of nucleotide mismatches, length of protospacer sequence, and different buffering conditions. We also confirmed the possibility of a “one-step single-tube” reaction for mutation detection. This assay can robustly distinguish circular DNA templates that differ by a single nucleotide. It has the potential to be adapted for automated applications, such as the screening of mitochondrial diseases.