Nucleosome-targeted host DNA depletion enables automated plasma metagenomic sequencing for sensitive detection of bloodstream pathogens

Abstract

Background

Bloodstream infections (BSIs) are leading causes of sepsis-related mortality. Although metagenomic next-generation sequencing (mNGS) enables culture-independent pathogen detection, its clinical utility in plasma is limited by the overwhelming abundance of host cell-free DNA (cfDNA) and labor-intensive manual workflows.

Methods

A plasma host DNA depletion mNGS (HD-mNGS) assay was developed which integrated nucleosome-targeted host DNA depletion with automated DNA extraction and library preparation. Analytical performance was evaluated through limit of detection, linearity, precision, and contamination control. Clinical performance was assessed in a cohort of 107 patients with suspected BSI and benchmarked against blood culture (BC), conventional microbiological testing (CMT), and standard mNGS without host depletion, using a composite clinical reference standard.

Results

Nucleosome depletion markedly reduced host DNA background by an average of 66-fold, consequently enriching microbial reads by approximately 46.73-fold. The automated HD-mNGS assay exhibited robust analytical sensitivity, with limits of detection (LoD) ranging from 9.1 to 38 genome equivalents (GE) /mL for bacteria and fungi, and from 283 to 321 GE/mL for viruses and excellent linearity across tested concentrations (R² = 0.915–0.989). Furthermore, the automated workflow maintained strong quantitative correlation with manual protocols while significantly reducing common skin and environmental contaminants by 71.7% and 83.7%, respectively. In a cohort of 107 patients, HD-mNGS demonstrates improved diagnostic performance for BSI, achieving a significantly higher pathogen detection rate (64.49%) and clinical positive percent agreement (PPA: 95.24%) than standard mNGS, BC, and CMT (P < 0.001). Crucially, HD-mNGS demonstrates enhanced performance in detecting rare, fastidious, and intracellular pathogens (such as Mycobacterium tuberculosis and Rickettsia) that yield extremely low concentrations of circulating DNA, overcoming the limitations of traditional methods while maintaining high overall diagnostic total percent agreement (TPA: 88.79%).

Conclusions

Nucleosome-targeted host DNA depletion integrated with a fully automated mNGS platform significantly enhances microbial detection in plasma and provides a scalable approach for standardized BSI diagnostics.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12967-026-08597-x.

Introduction

Bloodstream infections (BSIs) cause significant global morbidity and mortality, particularly in critically ill and immunocompromised patients. Rapid, accurate pathogen identification is essential for guiding targeted therapy, improving outcomes, and minimizing broad-spectrum antibiotic use [1]. Conventional microbiological testing (CMT), including blood culture, remains the clinical gold standard but is limited by slow turnaround times (2–5 days), low sensitivity (particularly after prior antibiotic use), and inability to detect fastidious, non-culturable, or novel pathogens [2]. Recently, metagenomic next-generation sequencing (mNGS) has emerged as a culture-independent diagnostic method that enables comprehensive, unbiased detection of bacteria, fungi, viruses, and parasites directly from clinical samples [3]. However, despite its proven utility across neurological, respiratory, and bloodstream infections [4–6], several technical challenges continue to limit the widespread clinical adoption of mNGS for BSI diagnosis.

A major technical challenge in plasma mNGS-based pathogen detection is the overwhelming dominance of host-derived DNA. In plasma samples, host cell-free DNA (cfDNA) typically constitutes 95–99% of sequencing reads, reducing pathogen-derived sequences to a negligible fraction of the library [7]. This stark imbalance drastically limits the sequencing depth available for pathogen identification, especially in low-biomass samples like blood. A clear understanding of the biological mechanisms driving this imbalance is therefore essential for designing effective depletion strategies.

Host cfDNA enters the bloodstream primarily through cellular turnover, which involves apoptosis, necrosis, and neutrophil extracellular trap formation (NETosis). Among these, apoptosis serves as the dominant source under both normal and pathological conditions [8]. During apoptosis, endonuclease-mediated chromatin cleavage at internucleosomal sites generates DNA fragments that remain bound to histone octamers as nucleosomes. Consequently, most circulating host cfDNA exists as nucleosome-protected fragments with a characteristic size of ~ 167 bp, corresponding to ~ 147 bp of DNA wrapped around a histone octamer plus linker DNA [9]. Genome-wide sequencing studies consistently confirm this 167 bp peak, along with di- and tri-nucleosomal multimers, establishing that host cfDNA circulates predominantly in nucleosome-associated form [10, 11]. In contrast, pathogen-derived DNA released during infection through microbial replication, immune-mediated lysis, or antimicrobial-induced cell disruption lacks nucleosome-based chromatin organization. As a result, microbial DNA in plasma exists as free or randomly fragmented molecules rather than nucleosome-bound complexes. This fundamental structural distinction offers a potential basis for selectively depleting host DNA from plasma samples [12, 13].

Several strategies have been developed to reduce host DNA background in mNGS, including differential lysis, CpG methylation-based capture, saponin treatment, and CRISPR-Cas-based depletion [14–16]. However, each has inherent limitations for plasma-based BSI diagnosis. Saponin-based selective lysis relies on intact cellular structures and is therefore inapplicable to cell-free nucleic acid matrices such as plasma-derived cfDNA [14]. Methylation-based approaches exploit the relative hypomethylation of microbial DNA, yet their efficacy varies with microbial GC content and methylation patterns, with reduced performance for certain fungi and DNA viruses [16]. CRISPR-Cas9-based methods afford highly specific host sequence depletion but necessitate custom guide RNA design and entail technically demanding workflows that are difficult to standardize [15]. By contrast, nucleosome-targeted immunocapture operates independently of DNA sequence or methylation status, directly targeting the physical packaging of host cfDNA. As outlined above, this strategy leverages the intrinsic structural distinction between nucleosome-bound host cfDNA and naked microbial DNA. By employing antibody-mediated immunocapture to selectively remove histone-nucleosome complexes, nucleosome-targeted depletion offers a biology-guided, sequence-agnostic approach that is uniquely suited for plasma-based BSI diagnosis.

Another key challenge hindering the clinical implementation of mNGS is workflow complexity. Conventional protocols require multiple manual steps, such as DNA extraction, purification, and library preparation, leading to increased hands-on time, introducing inter-laboratory variability, and elevating contamination risk. Automated platforms integrating these steps with sequencing and bioinformatic analysis are under development to improve reproducibility and minimize human error [17]; however, rigorous analytical and clinical validation of such systems, particularly for plasma-based BSI diagnosis, remains limited. Plasma is inherently well-suited to streamlined mNGS workflows. Unlike other body fluids types, it requires only simple centrifugation for preparation, eliminating the need for homogenization or cell lysis. Plasma cfDNA is already fragmented and acellular, with fragment sizes of 150–200 bp that closely match the optimal insert size for most NGS library preparation protocols [8, 18], rendering the enzymatic or mechanical shearing otherwise required for intact-cell genomic DNA unnecessary [19]. These properties collectively simplify the workflow, reduce hands-on time, and minimize potential bias and contamination, making plasma a well-suited specimen type for automated clinical mNGS, particularly for bloodstream pathogen detection.

In the present study, we describe the development of a plasma host DNA depletion mNGS (HD-mNGS) assay that combines a nucleosome-mediated host DNA depletion strategy with a fully automated DNA extraction and library preparation platform. We first detail the nucleosome depletion workflow designed to selectively remove host cfDNA while preserving microbial DNA, followed by comprehensive analytical validation of the automated platform (including limit of detection, linearity, precision, interference, and cross-contamination control). We then assess the clinical performance of HD-mNGS in patients with suspected bloodstream infections, benchmarking its diagnostic yield against conventional microbiological methods. Our findings demonstrate that integrating nucleosome-targeted host DNA depletion with automated mNGS workflows substantially enhances the sensitivity and clinical utility of plasma-based pathogen detection.

Methods

Specimen collection

This single center, retrospective laboratory validation study was conducted at Peking University Third Hospital. Residual whole blood specimens were collected from patients with suspected BSI who had undergone routine clinical microbiological testing as part of standard care. Following standard-of-care testing, residual samples were anonymized and stored at 4 °C for no more than 24 h prior to HD-mNGS processing. Additionally, all samples underwent standard mNGS testing. The study included a total of 107 residual specimens: 63 from patients ultimately diagnosed with BSI and 44 from non-BSI controls, as determined by comprehensive clinical diagnostic criteria (see Clinical Analysis). The study protocol was approved by the Institutional Review Board of Peking University Third Hospital.

Clinical analysis

Two independent infectious disease physicians, blinded to the mNGS results, retrospectively reviewed all available clinical data for each enrolled patient, including: (1) demographic and clinical characteristics (e.g., presence of fever, hypotension, or organ dysfunction); (2) laboratory findings (e.g., white blood cell count, C-reactive protein, procalcitonin); (3) microbiological results from all conventional tests (including blood cultures, other sterile-site cultures, serology, and antigen testing); (4) radiological imaging reports; and (5) clinical response to antimicrobial therapy. A diagnosis of definite BSI was assigned if at least one of the following criteria was met: (i) a recognized pathogen was isolated from one or more blood cultures in a patient with compatible clinical signs of infection (e.g., fever, chills, hemodynamic instability), and no other source explained the syndrome; (ii) a common skin contaminant (e.g., Staphylococcus epidermidis, Cutibacterium acnes) was isolated from two or more separate blood cultures drawn on different occasions, accompanied by clinical signs of infection and/or a positive response to targeted antimicrobial therapy; or (iii) in culture-negative cases, there was strong clinical and laboratory evidence of systemic infection (e.g., persistent fever, elevated inflammatory markers, sepsis-related organ dysfunction) with corroborating microbiological evidence from a non-blood sterile site (e.g., cerebrospinal fluid, pleural fluid) or a positive result from an alternative validated molecular assay (e.g., mNGS or PCR), and a clear clinical improvement following pathogen-directed therapy [6, 20, 21]. Patients who did not meet any of these criteria were classified as non-BSI controls. In cases of disagreement between the two reviewers, a third senior clinician adjudicated the final diagnosis.

Nucleosome depletion workflow

Whole blood (1.5 mL) was collected in EDTA tubes and processed within 24 h. Cell-free plasma was isolated by sequential centrifugation: 1,600 × g for 10 min at 4 °C to pellet intact blood cells, followed by 16,000 × g for 10 min at 4 °C to remove residual cellular debris, apoptotic bodies, and cell fragments. The resulting cell-free plasma was immediately used for nucleosome-based host DNA depletion. For plasma pre-treatment, cell-free plasma was mixed 1:1 (v/v) with binding enhancer, incubated at room temperature with thorough mixing, then transferred to ice for 10 min. Glycine was added to 2% (final concentration) to quench nucleosome–binding enhancer cross-linking, yielding pre-treated plasma.

For magnetic bead preparation, beads (50 µL, 1.5 mg) were resuspended by vortexing (≥ 30 s) or end-over-end inversion (5 min), collected on a magnetic separator, and the supernatant discarded. Anti-histone antibody diluted in PBS-Tween 20 was added, and the suspension was incubated with rotation at room temperature for 30 min. Beads were re-collected magnetically, washed twice with 200 µL PBS-Tween 20 to remove unbound antibody, and the final pellet was retained for immunocapture.

For nucleosome depletion, the antibody-conjugated beads were resuspended in pre-treated plasma and incubated at room temperature for 40 min. After capturing the bead-nucleosome-antibody complexes on a magnetic stand, the clarified supernatant, depleted of nucleosome-bound host DNA, was transferred to a new tube. Proteinase K was added, and the mixture was incubated at 56 °C for 10 min, followed by centrifugation at 12,000 × g for 10 min to pellet debris. The clarified supernatant, enriched for pathogen-derived DNA, was subsequently processed using an automated platform for DNA extraction and library construction.

DNA extraction

Total DNA were extracted from 400 µL of clarified supernatant using the PathoXtract^® Automated Nucleic Acid Extraction Kit (Willingmed, Beijing, China) on the PathoLab automated instrument (Cat#WYAED001, Willingmed, Beijing, China), following the manufacturer’s instructions. Prior to extraction, an exogenous internal control (IC)-comprising non-human, non-pathogenic synthetic DNA sequences-was spiked into each sample and negative control to monitor workflow efficiency across the entire process, from extraction through sequencing and bioinformatic analysis.

Library preparation and sequencing

Extracted DNA was loaded onto the PathoLab automated instrument (Cat#WYAED001, Willingmed, Beijing, China) for library construction using the PathoLib Automated Library Prep Kit (Cat#WYLA04405D-12, Willingmed, Beijing, China). The integrated closed-system workflow encompassed end-repair, adapter ligation with unique dual-index (UDI) barcodes, PCR amplification, and purification, minimizing contamination risk. Libraries were quantified using the Qubit dsDNA HS Assay Kit on a Qubit 4.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA), normalized to a uniform concentration, and pooled in equimolar amounts. Sequencing was performed on an illumina NextSeq 550Dx platform (Illumina, San Diego, CA, USA) using Mid- or High-Output v2.5 reagent kits to generate single-end 75-bp reads, with a minimum of 10 million reads per sample. Standard mNGS was performed using an identical workflow without the nucleosome depletion step.

Bioinformatic analysis

Raw sequencing data in FASTQ format were processed through a standardized bioinformatic pipeline. Quality control and filtering were performed using Trimmomatic v0.40, with removal of low-quality reads, adapter-contaminated sequences, high-coverage duplicate reads, and reads below the minimum length threshold, retaining only high-quality reads for downstream analysis. The filtered reads were subsequently aligned to the human reference genome GRCh38 (hg38) using Bowtie2 v2.4.3 to remove host-derived sequences, yielding clean non-host reads for pathogen identification. Taxonomic classification of the remaining non-host reads was performed using Kraken2 v2.1.0 against the pre-built core_nt reference database (release 20251015; built October 15, 2025), the most comprehensive Kraken2 index. The core_nt database is a curated, large-scale collection derived from the NCBI core nucleotide (core_nt) collection, inclusive of sequences from GenBank, RefSeq, TPA, and PDB, and encompasses reference sequences across bacteria, fungi, viruses, archaea, and parasites. Final microbial identification results were generated based on species-level taxonomic annotation of the classified reads. Microbial identification was based on uniquely mapped reads, normalized to reads per ten million mapped reads (RPTM) to account for sequencing depth variation. The positive thresholds for pathogenic microorganisms were set as follows: Bacteria and fungi: RPTM ≥ 8; Viruses: RPTM ≥ 3; Special pathogens: For clinically highly relevant pathogens such as Mycobacterium tuberculosis complex and Cryptococcus species, RPTM ≥ 1. RPTM of IC values were assessed for each run to confirm assay validity. These thresholds were established based on clinically validated cut-offs from established plasma mNGS platforms [7, 19, 22] and were further verified against our analytical validation data, including LoD measurements, linearity assessments, and cross-contamination control results, which collectively confirmed that the selected cut-offs reliably distinguished true pathogen signals from background noise. The lower threshold for special pathogens reflects the clinical priority of maximizing detection sensitivity for organisms known to circulate at extremely low cfDNA concentrations [23].

Host DNA depletion efficiency

To confirm the effectiveness of the nucleosome depletion procedure, depletion efficiency was assessed in a subset of 30 plasma samples obtained from BSI patients. Each sample was processed in parallel with and without host depletion. Host DNA abundance was quantified by real-time PCR targeting the human β-actin gene (forward primer: 5’-CGGCCTTGGAGTGTGTATTAAGTA-3’; reverse primer: 5’-TGCAAAGAACACGGCTAAGTGT-3’) [24], performed in triplicate for each aliquot, with depletion efficiency expressed as the cycle threshold difference (ΔCt = Ct[non-depletion]-Ct[depletion]). Pathogen abundance enrichment fold-change, and pathogen genome coverage improvement fold-change were also measured.

External controls preparation

To evaluate the performance of the automated plasma mNGS pipeline, we prepared spiked-in reference materials encompassing ten representative pathogens across major DNA-based pathogen categories: Gram-positive bacteria, Gram-negative bacteria, Fungi (yeasts and Aspergillus spp.), and DNA viruses. Genomic DNA was extracted from ten representative pathogens: Enterococcus faecium (ATCC 19434), Enterococcus faecalis (ATCC 23241), Staphylococcus aureus (ATCC 29213), Staphylococcus epidermidis (ATCC 12228), Escherichia coli (ATCC 25922), Klebsiella pneumoniae (ATCC 700603), Candida albicans (ATCC 18804), Aspergillus fumigatus (GDMCC 3.627), Human Cytomegalovirus (CMV, AD169 strain), and Human adenovirus type C (HAdV-C, ATCC VR-7). Extracted DNA were enzymatically fragmented to 100–200 bp to mimic circulating microbial cell-free DNA, then spiked into pooled pathogen-free human plasma from healthy donors to generate positive controls (PCs) [7, 25]. The concentration of each pathogen in the undiluted PC stock was standardized to 10,000 genomic equivalents per milliliter (GE/mL). Negative controls (NCs) consisted of sterile deionized water.

Evaluation of mNGS analytical performance characteristics

Analytical performance of the automated mNGS assay was evaluated as follows:

LoD and linearity: PC samples prepared from ten representative pathogens were serially diluted in 4-fold increments (1:1 to 1:4,096), spanning a theoretical concentration range of ~ 2.4–10,000 GE/mL, with four technical replicates per dilution. LoD was defined as the lowest concentration at which all four of replicates yielded a positive result. For linearity, a standard curve was constructed by plotting the log10 of the input pathogen concentration (GE/mL) against the log10 of the detected abundance (RPTM), with goodness-of-fit assessed by the coefficient of determination (R²).

Precision: Intra-assay (within-run) precision was assessed by testing 20 replicates of the same PC sample within a single sequencing run; inter-assay (between-run) precision was evaluated by testing the same PC sample across 20 independent runs conducted on different days. For both assessments, precision was expressed as the coefficient of variation (CV%) of RPTM values for each microorganism.

Interference: The assay’s capacity to distinguish highly homologous species was assessed using mixed PC samples prepared at defined input ratios: E. faecalis vs. E. faecium and S. epidermidis vs. S. aureus. Performance was considered acceptable when detected species identities and relative abundance ratios were concordant with input proportions.

Cross-contamination and residual contamination: To assess instrument carryover control, high-concentration PC samples and negative controls (NCs; nuclease-free water) were loaded in alternating adjacent wells across two separate assay plates, with PC and NC positions swapped between plates. Any pathogen signal detected in NC wells was used as an indicator of cross-contamination or carryover from preceding runs or neighboring wells.

Accuracy: Clinical accuracy was evaluated using 107 clinical samples. Diagnostic performance was assessed against two reference standards: (1) blood culture (BC) alone and (2) a composite microbiological testing (CMT) panel comprising BC and additional standard culture and serology methods. Performance was further benchmarked against a composite clinical diagnostic standard established by a multidisciplinary team through comprehensive integration of clinical, laboratory, microbiological, and therapeutic response data. Positive percent agreement (PPA), negative percent agreement (NPA), and total percent agreement (TPA) were calculated for HD-mNGS and comparator methods against this reference standard.

Statistical analysis

Categorical variables are presented as counts and percentages; continuous variables as medians with interquartile ranges (IQRs). Sensitivity, specificity, PPA, NPA, and TPA were calculated with 95% confidence intervals (CIs), and differences in positive detection rates between methods were assessed by McNemar’s test [26]. A two-sided P < 0.05 was considered statistically significant. All analyses were performed using R software (version 4.3.0).

Results

Nucleosome-based host DNA depletion improves microbial signal

A nucleosome-based host DNA depletion strategy was employed to minimize host-derived background in plasma metagenomic sequencing. This approach employs antibody-mediated immunocapture of histone proteins to selectively remove nucleosome-associated host DNA fragments, thereby enriching non-nucleosomal microbial DNA. The overall workflow was illustrated in Fig. 1.

Fig. 1.

Schematic illustration of the plasma-based host depletion workflow for mNGS pathogen detection. The workflow comprises three sequential modules. Module 1 (Plasma Separation): EDTA-anticoagulated whole blood is centrifuged to isolate plasma, which contains both host nucleosome-bound DNA and pathogen-derived cell-free DNA (cfDNA). Module 2 (Host Depletion): Anti-histone antibody-conjugated magnetic beads are incubated with plasma to selectively capture host nucleosome–DNA complexes. Following magnetic separation, the resulting antibody–bead–nucleosome complexes are removed, and the supernatant enriched for pathogen cfDNA is collected for downstream processing. Module 3 (Downstream Automated & Bioinformatics Workflow): The pathogen-enriched cfDNA supernatant undergoes automated DNA extraction and library preparation, followed by high-throughput sequencing. Raw sequencing reads are processed through a standardized bioinformatics pipeline for pathogen identification and clinical reporting

To evaluate host DNA depletion efficiency, 30 plasma samples from the clinical BSI patients were processed with or without nucleosome capture prior to sequencing. Depletion efficacy was quantified by measuring the Ct values of the human β-actin gene, where higher Ct values reflect greater host DNA removal. qPCR analysis demonstrated a mean β-actin Ct difference of 6.06 (median: 5.38, IQR: 5.05–5.63, P < 0.0001) between depleted and non-depleted samples, corresponding to an approximately 66-fold reduction in human DNA abundance (Fig. 2A).

Fig. 2.

Performance comparison of host depletion strategies in HD-mNGS analysis. Thirty plasma samples from BSI patients (biological replicates) were processed in parallel with and without nucleosome-targeted host DNA depletion. (A) Distribution of human DNA abundance (β-actin Ct value) before and after host-reads removal. Higher Ct values indicate greater host DNA removal. Violin plots show the median (solid line) and quartiles (dotted lines). Statistical significance was assessed using the Wilcoxon signed-rank test. (B) Fold-change in microbial read enrichment for overall microbiota and individual pathogen categories (bacteria, fungi, and virus) following host DNA depletion across all samples. (C) Genome coverage fold change for bacteria, fungi, and virus after host-read removal. For (B) and (C), data are presented as median with interquartile range. ***, P < 0.001

Consistent with the reduction in host background, nucleosome depletion substantially enriched microbial signal, yielding an approximately 46.73-fold increase in relative microbial read abundance compared with non-depleted samples. Enrichment varied across microbial kingdoms, with bacteria showing the greatest fold change [median: 54.62 (IQR: 41.99–81.50)], followed by fungi [27.00 (IQR: 17.84–42.87)] and viruses [25.01 (IQR: 16.54–42.63)] (Fig. 2B). Genome coverage of detected microbial species likewise improved markedly, with median coverage increases of 42.89-, 15.25-, and 31.56-fold for bacteria, fungi, and viruses, respectively (Fig. 2C).

Analytical performance of the host DNA depletion mNGS assay (HD-mNGS)

Analytical performance was validated using a ten-pathogen reference panel designed to assess potential sources of bias, comprising gram-positive bacteria (E. faecium, E. faecalis, S. aureus, S. epidermidis), gram-negative bacteria (E. coli, K. pneumoniae), fungi (C. albicans, A. fumigatus), and the viruses CMV and HAdV-C. Two pairs of closely related organisms (S. epidermidis/S. aureus and E. faecalis/E. faecium) were included to verify accurate species discrimination during coinfections.

LoD and linearity

Ten representative organisms were used to evaluate LoD and linearity. Each organism underwent 4-fold serial dilution (1:1 to 1:4096) from an initial concentration of 10,000 GE/mL, with each dilution tested in quadruplicate. The LoD in PC samples ranged from 9.1 to 38 GE/mL for bacteria and fungi, and from 283 to 321 GE/mL for viruses (Table 1; Supplementary Table 1). Microorganism concentration and detected RPTM values showed a strong linear correlation (R² = 0.915–0.989; Fig. 3).

Table 1.

Performance characteristics for the automated equipment assay

Metrics	Method/Pathogen type	Representative organism	Results
LoD	Detection of PC dilution by probit analysis
	Bacteria, gram- positive	Enterococcus faecium ATCC 19,434	34 GE/mL
		Enterococcus faecalis ATCC 23,241	29 GE/mL
		Staphylococcus aureus ATCC 29,213	33 GE/mL
		Staphylococcus epidermidis ATCC 12,228	36 GE/mL
	Bacteria, gram-negative	Escherichia coli ATCC 25,922	38 GE/mL
		Klebsiella pneumoniae ATCC 700,603	10.4 GE/mL
	Fungi	Candida albicans ATCC 18,804	9.1 GE/mL
		Aspergillus fumigatus GDMCC 3.627	9.5 GE/mL
	Virus	Human Cytomegalovirus AD169	283 GE/mL
		Human adenovirus type C ATCC VR-7	321 GE/mL
Precision	Intra-assay consistency: 20 replicates of the same sample processed in parallel on the same run		100% concordance CV: 18.01%-24.48%
	Inter-assay consistency: the same sample across 20 independent runs		100% concordance CV: 13.10%-20.33%
Interference	Co-infection interference (high genomic similarity microorganisms) performance was assessed in the same genus at different levels		Not interfered by high genomic-similarity microorganisms

Fig. 3.

Evaluation of linearity for the HD-mNGS assay. Correlation between nominal and observed concentrations for the ten spiked species was evaluated. Mean quadruplicate concentrations at each dilution were plotted against expected concentrations on a log scale, and R² was calculated

Precision

Intra-assay precision was assessed by testing one PC sample within the same HD-mNGS run using reagents from different batches, while inter-assay reproducibility was evaluated using the same sample across 20 HD-mNGS runs. Both demonstrated 100% reproducibility, with coefficients of variation ranging from 18.01% to 24.48% for intra-assay and 13.10% to 20.33% for inter-assay analyses (Table 1; Supplementary Table 2).

Interference

Interference from high genomic-similarity microorganisms (S. epidermidis vs. S. aureus, and E. faecalis vs. E. faecium) within the same genus was evaluated. The types and abundance ratios (1:1, 1:3, 3:1) of the detected species are consistent with that input, indicating that the HD-mNGS assay can reliably distinguish the source species (Table 1; Fig. 4).

Fig. 4.

Evaluation of interference testing for closely related species. To assess potential cross-interference under co-infection conditions, genetically related microbial cfDNA reference materials were spiked at defined ratios into sterile blood matrices. Plasma cfDNA concentrations were quantified for Staphylococcus epidermidis and Staphylococcus aureus (left panel), or Enterococcus faecalis and Enterococcus faecium (right panel) at simulated co-infection ratios of 1:1, 1:3, and 3:1 (ratios indicate the relative cfDNA concentrations of the first-listed to the second-listed species, respectively). Expected plasma cfDNA concentrations were derived from samples spiked with each individual strain. Data represent the mean of three independent replicates; error bars denote the standard deviation (SD). No significant differences were observed between expected and detected concentrations at any ratio (P > 0.05 for all comparisons)

Cross-contamination and residual contamination

To assess cross-contamination and residual contamination control, high-concentration PC and NC samples were placed alternately. During the second round of testing, the positions of PC samples and NC samples were swapped (Supplementary Fig. 1). In panel 1, only 2 of 11 NCs yielded low-abundance pathogen detections (S. epidermidis, RPTM = 5; C. albicans, RPTM = 3). In panel 2, only 1 of 10 NCs had 2 CMV reads detected. These results demonstrate the equipment’s effective control of cross-contamination and residual contamination.

Baseline characteristics of the study patients

Blood samples from 107 patients were enrolled to evaluate the clinical accuracy of the automated HD-mNGS assay (Table 2). Patient age ranged from 18 to 98 years (median: 62 years), and 67.3% (72/107) were male. The most prevalent underlying conditions were diabetes mellitus (22.4%), hypertension (21.5%), and malignancy (20.6%). Fever was the predominant presenting symptom (70.1%), followed by localized pain (15.0%) and dyspnea (8.4%). Impaired immune function was documented in 72.9% (78/107) of patients. Based on comprehensive integration of clinical manifestations, laboratory findings, microbiological results, and treatment responses, 63 patients received a final diagnosis of bloodstream infection.

Table 2.

Patient characteristics

Characteristics	Value
Patient demographics (n = 107)
Age(years)
Median (IQR)	62 (43–74)
Range	18–98
Gender, n (%)
Male	72 (67.29)
Female	35 (32.71)
Underlying disease, n (%)
Diabetes	24 (22.43)
Hypertension	23 (21.50)
Cancer	22 (20.56)
Symptom, n(%)
Fever	75 (70.09)
Pain	16 (14.95)
Dyspnea	9 (8.41)
Fatigue	7 (6.54)
Consciousness disturbance	4 (3.74)
Department, n (%)
Emergency	38 (35.51)
Infectious Diseases	19 (17.76)
Hematology	15 (14.02)
Intensive Care Unit	8 (7.48)
Respiratory Medicine	8 (7.48)
Surgery	5 (4.67)
Geriatrics	2 (1.87)
Cardiology	2 (1.87)
Ophthalmology	2 (1.87)
Rheumatic Immunology	1 (0.93)
Orthopaedics	1 (0.93)
Holding Unit	1 (0.93)
General Practice	1 (0.93)
Neurosurgery	1 (0.93)
Nephrology	1 (0.93)
Cardiovascular Medicine	1 (0.93)
Cardiac Surgery	1 (0.93)

Diagnostic performance of plasma mNGS for BSI

Among the 107 patients, HD-mNGS achieved the highest pathogen detection rate (64.49%), followed by standard mNGS (50.47%), CMT (30.84%), and blood culture (BC; 11.21%); HD-mNGS positivity rates were significantly higher than all comparator methods (P ≤ 0.05) (Fig. 5A). When combined with other methods, 56.07% of patients tested positive, with 8.41% positive exclusively by HD-mNGS (Fig. 5B). Using each comparator method as the reference standard, HD-mNGS demonstrated sensitivities of 100% (95% CI: 69.87–100%) versus BC, 87.88% (95% CI: 70.86–96.04%) versus CMT, and 100% (95% CI: 93.36–100.00%) versus standard mNGS, with corresponding specificities of 40.00% (95% CI: 30.24–50.58%), 45.95% (95% CI: 34.44–57.87%), and 71.70% (95% CI: 58.43–82.03%), respectively. McNemar’s test revealed that HD-mNGS sensitivity was significantly higher than that of BC, CMT, and standard mNGS (all P < 0.001; Fig. 5C). Among double-positive samples across all method pairs, pathogen concordance exceeded 80%. Additionally, except when compared with CMT, where the proportion of partial matches was slightly lower than that of complete matches, the proportion of complete matches was higher in both BC and standard mNGS, exceeding 66% of the double-positive samples (Fig. 5D).

Fig. 5.

Evaluation of diagnostic performance of the plasma HD-mNGS method. (A) Positive rate of HD-mNGS, BC, CMT and standard mNGS. (B) Pie chart of pathogen detection distribution. (C) Contingency tables for HD-mNGS results verses BC, CMT and standard mNGS. (D) Concordance of pathogen detection between HD-mNGS and three methods (BC, CMT, and standard mNGS) in double positive samples

Of the 107 enrolled patients, 63 were diagnosed with BSI per the composite diagnostic criteria. Using BSI status as the reference standard, significant differences in diagnostic performance were observed among HD-mNGS, BC, CMT, and standard mNGS. HD-mNGS achieved the highest PPA of 95.24%, higher than that of BC (19.05%, P < 0.001), CMT (44.44%, P < 0.001) and standard mNGS (76.19%, P < 0.001). Regarding NPA, BC demonstrated perfect specificity (100%), which was significantly higher than that of HD-mNGS (79.55%, P = 0.004), while no significant differences were found between CMT and HD-mNGS, and standard mNGS and HD-mNGS. HD-mNGS (88.79%) showed higher TPA than BC (52.34%, P < 0.001), CMT (62.62%, P < 0.001) and standard mNGS (80.37%, P = 0.035) (Fig. 6). Collectively, these findings demonstrate the improved diagnostic performance of HD-mNGS for BSI, with higher pathogen detection sensitivity while maintaining clinically acceptable specificity.

Fig. 6.

Analysis of the coincidence rate of clinical diagnosis. Contingency tables for BC, CMT, standard mNGS and HD-mNGS results vs. Clinical diagnose results for BSI. PPA, Positive percent agreement; NPA, negative percent agreement; TPA, total percent agreement

False-negative results were observed in three cases of BSI using HD-mNGS. Among them, two were due to RNA virus infections (Dengue virus and SARS-CoV-2), which could not be detected by the current DNA-based HD-mNGS platform. The third case involved a patient with pyogenic liver abscess, in whom Fusobacterium nucleatum was identified by mNGS of drainage fluid but was undetected in the blood sample by HD-mNGS.

Pathogens identification

Across the 107 clinical cases, a total of 44 distinct pathogens were identified (Fig. 7). HD-mNGS-alone or in combination with other methods-demonstrated the improved diagnostic sensitivity, contributing to the detection of the following pathogens, including fastidious or non-culturable agents such as Coxiella burnetii (n = 3), Rickettsia sibirica (n = 2), and M. tuberculosis (n = 2). By contrast, blood culture detected only a single pathogen, failing to identify all viruses, most fungi, and obligate intracellular bacteria. These findings highlight the complementary value of HD-mNGS in pathogen identification, particularly for uncultivable, slow-growing, or unexpected etiologies that are routinely missed by conventional diagnostics.

Fig. 7.

Pathogen detection profiles by different diagnostic methods. Bacteria, fungi, viruses, and special pathogens identified by BC, CMT, standard mNGS, or HD-mNGS were ranked by the number of positive detections

Performance comparison between automated and manual workflows

To evaluate the performance of the automated workflow, all the 107 plasma samples were processed in parallel using the automated platform and a conventional manual protocol. Across all 107 samples, the automated and manual workflows demonstrated a qualitative detection agreement of 65.79% (Fig. 8A). Among samples in which a given pathogen was detected by both workflows, the RPTM values generated from the automated workflow showed strong correlation with those obtained from the manual workflow for bacteria (R² = 0.95), fungi (R² = 0.91), and virus (R² = 0.97) (Fig. 8B). Automated processing significantly reduced the detection of common contaminants. Notably, S. epidermidis, the most prevalent skin commensal, showed a detection rate reduction from 45.2% to 12.8% (71.7% decrease). Environmental contaminants demonstrated even greater reductions, with Pseudomonas brassicacearum decreasing from 15.3% to 2.5% (83.7% decrease) (Fig. 8C).

Fig. 8.

Automated vs. manual HD-mNGS workflow comparison. (A) Venn diagram of detected pathogen species. (B) Quantitative correlation analysis. Scatter plot comparing pathogen abundance (expressed as log10-transformed RPTM values) between automated (y-axis) and manual (x-axis) workflows. (C) Detection rates of six common contaminant bacteria in manual (green bars) and automated (blue bars) HD-mNGS workflows. ***, P < 0.001

Discussion

In this study, we present a fully automated plasma HD-mNGS assay integrating a biology-guided, nucleosome-targeted host DNA depletion strategy for the diagnosis of bloodstream infections. The assay exhibited robust analytical performance across key metrics, including dynamic range, reproducibility, species discrimination, and cross-contamination control. In clinical validation, HD-mNGS achieved higher positive detection rates and stronger concordance with the composite reference standard relative to blood culture, CMT, and standard mNGS. Together, these results establish HD-mNGS as a valuable complementary diagnostic approach for BSI, with particular utility in cases refractory to conventional microbiological methods.

As introduced, multiple host DNA depletion strategies have been evaluated for plasma mNGS, each with distinct limitations. In the present study, nucleosome-targeted immunocapture achieved nearly 47-fold enrichment of microbial reads alongside a statistically significant reduction in host-derived sequences, consistent with the biology-guided rationale underlying this approach. The sequence-agnostic nature of this depletion strategy, which operates independently of GC content, methylation status, or guide RNA design, provides significant advantages for clinical BSI diagnosis, where a broad and unpredictable spectrum of pathogens may be encountered. We observed that the fold-enrichment of microbial reads following host DNA depletion varied by pathogen category, with bacteria (mean 54.62-fold; IQR: 41.99–81.50) showing higher enrichment than fungi (mean 27.00-fold; IQR: 17.84–42.87) and viruses (mean 25.01-fold; IQR: 16.54–42.63). This gradient is primarily explained by differences in genomic packaging and release mechanics. Because bacteria lack histones, their cfDNA circulates as entirely naked fragments unaffected by anti-histone immunocapture [27]. For fungi, two mechanisms likely contribute to reduced enrichment. First, fungi are eukaryotes that possess canonical histone proteins and conserved nucleosomal chromatin organization [28, 29]; a fraction of circulating fungal cfDNA likely retains nucleosomal association, leading to partial co-depletion. Additionally, the rigid fungal cell wall can restrict direct cfDNA shedding into plasma, rendering a portion of fungal biomass cellular rather than free-floating [30]. For DNA viruses, several species (e.g., Epstein-Barr virus) undergo chromatinization using host histones during their lifecycle, generating nucleosome-bound cfDNA that is partially cleared during host depletion [31]. Furthermore, the shorter fragment length and lower baseline concentration of viral cfDNA in plasma limit the theoretical upper bound of relative enrichment. Nevertheless, the significant overlap in enrichment IQRs suggests these dynamics are highly dependent on patient-specific factors, such as infection stage and pathogen load. Most importantly, the minimum average enrichment of ≥ 25-fold achieved across all categories significantly lowers the limit of detection for difficult-to-sequence fungal and viral pathogens.

A further barrier to clinical mNGS adoption is workflow complexity, as traditional protocols require multiple sequential manual steps, including sample preparation, DNA extraction, library construction, and sequencing, each introducing variability and contamination risk [3, 22]. Our fully automated platform performs end-to-end DNA extraction and library preparation within a closed system, substantially reducing hands-on time and eliminating inter-operator variability. Notably, automation markedly reduced contaminant detection rates, with S. epidermidis falling from 45.2% to 12.8% and the environmental contaminant P. brassicacearum from 15.3% to 2.5%, which is a meaningful gain given that skin commensals and environmental organisms represent a persistent source of false positives that complicates clinical mNGS interpretation [22, 32]. While the overall qualitative agreement of 65.79% between the automated and manual workflows reflects acceptable performance for highly challenging, low-biomass plasma matrices, it requires careful contextualization. This binary concordance metric is heavily influenced by samples containing trace pathogen levels near the assay’s LoD, rather than discordance among high-confidence detections. This metric captures several independent sources of discrepancy: (i) stochastic sampling effects at low pathogen concentrations, where near-LoD signals may be detected in one replicate but not another due to Poisson sampling variation [19, 33]; (ii) systematic differences in extraction efficiency between the closed automated system and the open manual protocol, which may differentially affect recovery of trace-level pathogens, with a subset of low-abundance pathogens detected by the automated workflow but missed by the manual protocol likely reflecting genuine sensitivity gains rather than false positives; (iii) differences in low-abundance pathogen recovery arising from variable extraction efficiency between the two workflows; and (iv) the reduction of contaminant signals by the closed-system automated workflow improves specificity but lowers qualitative concordance with the manual protocol by excluding contaminant organisms, representing a clinically beneficial rather than detrimental outcome. Importantly, the strong quantitative correlation (R² > 0.91) among shared positive detections confirms that when a pathogen is detected by both workflows, its abundance is measured with high consistency, supporting the analytical reliability of the automated system. This pattern of moderate qualitative agreement driven largely by detection at or near the LoD, coupled with high quantitative concordance for robustly detected pathogens, is consistent with expectations for low-biomass specimens and aligns with other mNGS platform comparisons, where diagnostic concordance rates improve significantly when stratified by pathogen abundance [22, 33].The analytical performance of the HD-mNGS assay compares favorably with previously published plasma mNGS platforms [34]. LoD ranged from 9.1 to 38 GE/mL for bacteria and fungi, and 283 to 321 GE/mL for viruses. The assay exhibited high precision, with 100% reproducibility and coefficients of variation consistently below 25% under both intra- and inter-assay conditions. Strong linearity (R²= 0.915–0.989) spanning over three orders of magnitude supports semi-quantitative pathogen monitoring. Cross-contamination was effectively minimized, with only trace signals detected in negative control wells across two alternating-layout runs, mitigating a known limitation of highly sensitive amplification-based diagnostics.

The incremental benefit of host DNA depletion over standard mNGS was evident in the higher positive detection rate (64.49% vs. 51.40%). This improvement is consistent with prior studies demonstrating that host DNA depletion enhances microbial downstream detection sensitivity in plasma mNGS [12, 13]. Importantly, the strong concordance in pathogen identification between HD-mNGS and standard mNGS double-positive samples (> 80% consistent pathogens detected) confirms that depletion did not introduce novel false-positive signals but rather improved detection of true low-abundance pathogens. The higher proportion of complete matches in BC and standard mNGS double-positive comparisons (> 66%) further supports the analytical reliability of the assay. Furthermore, HD-mNGS demonstrated improved diagnostic yield over all comparator methods, with positive detection rate, PPA and TPA against the composite reference standard substantially exceeding those of BC, CMT, and mNGS. The improved sensitivity of HD-mNGS was further evidenced by its detection of M. tuberculosis, C. burnetii, and Rickettsia in plasma samples where standard mNGS yielded negative results. These pathogens share a common feature of yielding extremely low concentrations of circulating DNA, which is attributable to the lysis resistance of M. tuberculosis and the obligate intracellular lifestyle of C. burnetii and Rickettsia [35], rendering their trace-level signals effectively masked by host cfDNA background under standard mNGS conditions. By reducing host cfDNA through nucleosome-targeted immunocapture, HD-mNGS improves the sequencing signal-to-noise ratio and reallocates sequencing depth toward pathogen-derived reads, enabling detection of pathogens below the practical sensitivity threshold of standard mNGS without increasing the absolute quantity of pathogen DNA released into circulation.

In this study, three false-negative cases were identified, involving dengue virus, SARS-CoV-2, and F. nucleatum, respectively. For the two RNA virus cases, the false-negative results were attributable to the intrinsic limitation of the DNA-only sequencing strategy employed by the HD-mNGS platform. For the F. nucleatum case, the pathogen was detected in abscess drainage fluid but not in blood. This discrepancy is likely explained by the localized nature of the infection, in which the pathogen is confined to the abscess cavity with minimal bacteremia, resulting in insufficient cfDNA shedding into the circulation [36, 37]. Blood-based mNGS has substantially lower sensitivity than site-directed specimens, as pathogen cfDNA loads in blood are significantly lower than in samples obtained directly from the infection site [38]. Additionally, cfDNA-based mNGS assays have pathogen-specific detection thresholds of 10–1000 copies/mL, below which false-negative results occur. To mitigate false-negative results, we recommend three strategies: (1) supplementing HD-mNGS with serological testing or RT-PCR for patients with suspected RNA virus infections; (2) prioritizing site-directed sampling over blood alone for localized infections such as abscesses; and (3) adopting a multimodal diagnostic approach that integrates mNGS findings with culture, serology, and clinical-radiological data, particularly when mNGS results are discordant with clinical presentation.

Several limitations warrant consideration. This single-center retrospective study enrolled a relatively modest cohort, which may limit generalizability across diverse patient populations, clinical settings, and regional pathogen spectra. Although the composite reference standard integrates clinical, microbiological, and therapeutic response data to reduce misclassification, diagnostic uncertainty may persist in culture-negative cases. Additionally, the analytical validation panel encompassed DNA-based pathogens exclusively; the clinical detection performance of RNA viruses remains to be evaluated. Multicenter prospective trials with larger and clinically diverse cohorts will be required to validate these findings and assess the impact of HD-mNGS on patient outcomes.

In conclusion, integrating biology-guided, nucleosome-targeted host DNA depletion into a fully automated plasma mNGS workflow improved pathogen detection sensitivity without compromising specificity or reproducibility. HD-mNGS demonstrated robust analytical performance and improved clinical diagnostic yield relative to blood culture, CMT, and standard mNGS. Together, these findings establish automated HD-mNGS as a scalable and standardizable approach to BSI diagnosis, with the potential to meaningfully improve outcomes in patients with life-threatening infections.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

BSIs

Bloodstream infections

CMT

Conventional microbiological testing

cfDNA

Cell-free DNA

mNGS

Metagenomic next-generation sequencing

HD-mNGS

Host DNA depletion mNGS

RPTM

Reads per ten million mapped reads

BC

Blood culture

PPA

Positive percent agreement

NPA

Negative percent agreement

TPA

Total percent agreement

IQRs

Interquartile ranges

CIs

Confidence intervals

LoD

Limits of detection

Author contributions

N.S. and JJ.Z. contributed to the study design. TY.W. X.H. J.L., and SS.M. conducted experiments and collects data. SH.Z. YF.Z., and TY.W. analyzed and interpreted the data. SH.Z. and YF.Z. drafted the manuscript. All authors have reviewed and approved the final version of the manuscript.

Funding

This work was supported by Beijing Natural Science Foundation (F252052), Peking University Third Hospital Fund for Interdisciplinary Research (BYSYJC2023005), State Key Laboratory of Vascular Homeostasis and Remodeling Open Research Fund (2025-VHR‐O‐SY‐21), and National Key Research and Development Program of China (No.2025YFC2609702 and 2025YFC2609700), Beijing Municipal Science and Technology Plan Project (No. Z251100004625067).

Data availability

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

Declarations

Ethics approval and consent to participate

This study follows the Helsinki Declaration. The research plan received approval from the Medical Science Research Ethics Committee at Peking University Third Hospital (M20250641). According to the national laws and the hospital’s policies, obtaining written informed consent from participants for this study was not necessary.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.