Optimization of Metagenomic Next-Generation Sequencing Workflow with a Novel Host Depletion Method for Enhanced Pathogen Detection

Abstract

Introduction

Sepsis is a critical condition requiring timely and accurate pathogen identification. Traditional blood cultures are slow and often yield low sensitivity. Metagenomic next-generation sequencing (mNGS) offers broad and rapid pathogen detection but is hindered by excessive human DNA background in blood samples. This study evaluated a novel Zwitterionic Interface Ultra-Self-assemble Coating (ZISC)-based filtration device designed to deplete host cells and enhance microbial DNA recovery for improved mNGS diagnostics.

Methods

We assessed the novel filter’s performance in depleting white blood cells (WBCs) while preserving microbial integrity using spiked blood samples. Comparisons were made with other host depletion techniques, including differential lysis and CpG-methylated DNA removal. Analytical sensitivity was tested using spiked microbial communities at varying genome equivalents (GEs). Clinical validation involved eight blood culture-positive sepsis patient samples, processed with and without filtration, for both genomic DNA (gDNA) and cell-free DNA (cfDNA)-based mNGS. All libraries were sequenced on a NovaSeq600 with at least 10 million reads per sample.

Results

The novel filter achieved > 99% WBC removal across various blood volumes and allowed unimpeded passage of bacteria and viruses. Compared to other depletion methods, the novel filtration was more efficient, less labor-intensive, and preserved microbial reads. mNGS with filtered gDNA detected all expected pathogens in 100% (8/8) of clinical samples, with an average microbial read count of 9351 reads per million (RPM), over tenfold higher than unfiltered samples (925 RPM). In contrast, cfDNA-based mNGS showed inconsistent sensitivity and was not significantly enhanced by filtration (1251–1488 RPM). Finally, the novel filtration did not alter the microbial composition, making it suitable for accurate pathogen profiling.

Conclusion

The workflow with the novel host depletion method significantly enhanced the analytical sensitivity of gDNA-based mNGS by reducing the host DNA background and enriching microbial content. This approach improved diagnostic yield in sepsis and may be a valuable tool for further clinical infectious disease diagnostics.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40291-025-00797-3.

Key Points

The Zwitterionic Interface Ultra-Self-assemble Coating (ZISC)-based filtration device for host depletion enables > 99% removal of white blood cells, significantly reducing human DNA background and improving microbial signal in blood-derived metagenomic next-generation sequencing (mNGS) samples.

The novel filtration preserves microbial composition to ensure clinical diagnostic reliability.

Genomic DNA-based mNGS with host depletion outperforms cell-free DNA-based methods, achieving 100% pathogen detection in culture-positive sepsis samples with > tenfold enrichment of microbial reads.

Introduction

Sepsis is a life-threatening condition triggered by the host’s immune response to a bloodstream infection [1]. If not promptly diagnosed and appropriately treated, sepsis can progress to septic shock and result in mortality [2, 3]. Currently, the diagnosis of sepsis relies heavily on traditional blood culture (BC) methods, which, despite their utility, are hindered by prolonged turnaround times. This limitation has been extensively documented [4–6].

In recent years, metagenomic next-generation sequencing (mNGS) has emerged as a promising diagnostic alternative, offering enhanced speed and sensitivity for pathogen detection [7–9]. mNGS employs an unbiased approach to extract and sequence all nucleic acids within a patient sample, enabling simultaneous analysis of host and microbial genomes. This method has expanded diagnostic strategies for pathogen identification, particularly in cases where traditional methods fall short. A notable example of its clinical utility involved the diagnosis of neuroleptospirosis in a 14-year-old boy with meningoencephalitis, where mNGS enabled timely pathogen identification, guided targeted antimicrobial therapy, and facilitated recovery [10].

One significant challenge in applying mNGS to blood samples is the overwhelming abundance of human DNA, which consumes valuable sequencing capacity. Strategies to mitigate this include pre-extraction methods such as differential lysis of human cells [11] and post-extraction techniques like methylated human DNA removal [12]. However, these methods often face limitations in efficiency or operational complexity [13–15]. Recently, a novel device called Devin (Micronbrane, Taiwan) employs Zwitterionic Interface Ultra-Self-assemble Coating (ZISC) technology, which selectively binds and retains host leukocytes and other nucleated cells without clogging, regardless of filter pore size. It has shown promise in addressing this issue. Integrating this novel device into the mNGS workflow has demonstrated significant enrichment of microbial content, improving assay sensitivity and reducing sequencing resource wastage.

In sepsis diagnosis from blood samples, cell-free DNA (cfDNA) extracted from plasma is a common starting material for mNGS library preparation [4, 16, 17]. However, cfDNA is not amenable to pre-extraction host-cell depletion methods. Alternatively, microbial genomic DNA (gDNA) derived from cell pellets can also be used. Unlike cfDNA, microbial cell pellets offer an opportunity for enrichment if host cells are effectively depleted before high-speed centrifugation to isolate microbial cells.

The aim of this study was twofold: first, to validate the efficacy of the novel ZISC-based fractionation filter in depleting host cells and enriching microbial sequences within a gDNA-based mNGS workflow, and second, to compare the performance of gDNA-based and cfDNA-based workflows, evaluating whether the use of cell pellets, particularly in combination with ZISC-based host-cell depletion, enhances diagnostic sensitivity.

Methods

Analytical Performance of the Novel Zwitterionic Interface Ultra-Self-assemble Coating (ZISC)-Based Host Depletion

To evaluate the white blood cell depletion efficiency of the novel ZISC-based host depletion filter, varying volumes of human blood (ranging from 3 to 13 mL) were processed using the ZISC-based filtration system. White blood cell counts in the pre-filtration inputs (I) and post-filtration outputs (O) were measured using a complete blood cell count analyzer.

To assess the filter’s bacterial retention properties, blood samples spiked with 10⁴ CFU/mL of Escherichia coli, Staphylococcus aureus, or Klebsiella pneumoniae were subjected to ZISC-based filtration. Bacterial counts in the filtrates were determined using standard plate-enumeration techniques to quantify the passage efficiency of these microorganisms through the filter. Similarly, blood samples were spiked with feline coronavirus to evaluate viral passage through the ZISC-based filter. Viral concentrations in the input and output samples were quantified using quantitative polymerase chain reaction (qPCR).

Comparison of the Novel ZISC-Based Host Depletion with Other Host Depletion Procedures

The performance of the ZISC-based host depletion filter was evaluated in parallel with two alternative host depletion methods: the QIAamp DNA Microbiome Kit (Qiagen), which removes white blood cells through differential lysis, and the NEBNext^® Microbiome DNA Enrichment Kit (New England Biolabs), which targets CpG-methylated host DNA for removal. A negative control, without any host depletion applied, was included for comparison.

A blood sample spiked with ZymoBIOMICS reference material D6320—containing two bacterial species, Imtechella halotolerans and Allobacillus halotolerans, at 10⁴ GE/mL—served as the test material for all four experimental conditions. After host depletion, DNA was extracted and used for NGS library preparation. The resulting libraries were sequenced on the MinION (Oxford Nanopore, UK) and the MiSeq (Illumina, California) NGS platforms.

Analytical Sensitivity of Metagenomic Next-Generation Sequencing (mNGS) Workflow Incorporating the Novel ZISC-Based Host Depletion

To assess the analytical sensitivity of the mNGS workflow, molecular biology-grade water (MBG) or human whole blood samples were spiked with either 10² or 10⁴ GE of the D6331 microbial community reference material (ZymoBIOMICS) and processed using the novel ZISC-based host depletion filter. The D6331 standard comprises 21 bacterial and fungal species at defined compositions (Online Supplemental Material (OSM), Table S1), of which 17 microorganisms were relevant to this study, as their expected GEs at 10² GE was ≥ 1.

Control MBG or whole blood samples were processed in parallel and were not subjected to the novel ZISC-based filtration. All samples underwent library preparation using the Ultra-Low Library Prep Kit (Micronbrane) and were sequenced on the MiSeq platform (Illumina). The resulting sequencing data were analyzed using a customized, in-house bioinformatics pipeline to evaluate microbial recovery and the efficacy of host depletion.

Patient Samples

Blood samples were collected from patients admitted to the emergency department of Taipei Veterans General Hospital between April 2021 and September 2022 who were clinically suspected of sepsis (n = 8). The study was approved by the Institutional Review Board of Taipei Veterans General Hospital (Institutional Review Board number: 2021-03-013AC; approval date: 22 March 2021). Written informed consent was obtained from all participants before sample collection. Blood samples for this study were collected concurrently with routine clinical blood culture samples to ensure comparability.

Each blood sample was divided into two equal portions of approximately 4 mL. One portion underwent host-cell depletion filtration, while the other was processed directly for mNGS without host depletion. For this analysis, only blood samples from patients with subsequent positive blood culture results were included in the final evaluation. All blood samples were processed fresh. No frozen samples were used for this study.

Host Depletion of Patient Samples

To evaluate the effectiveness of host depletion using the novel ZISC-based filter, blood samples were processed in parallel: one portion underwent host depletion via the novel ZISC-based fractionation filter (Micronbrane Medical, Taiwan), while the other portion was processed without filtration. The filter was securely connected to a syringe, where approximately 4 mL of whole blood was transferred. The syringe plunger was then gently depressed, pushing the blood sample through the filter into a 15 mL Falcon tube. Both the filtered and the unfiltered blood samples were subsequently used for downstream assays, allowing for comparison of host-cell depletion efficiency in relation to microbial DNA recovery.

Sample Processing and DNA Extraction of Patient Samples

As shown in Fig. 1, the ZymoBIOMICS Spike-in Control I (High Microbial Load) (Zymo Research, Irvine, CA), which includes two extremophile bacterial species, Imtechella halotolerans and Allobacillus halotolerans, was routinely added to all samples, including the No Template Control, at a concentration of 10⁴ genome copies/mL. This served as an internal reference control for microbial detection.

Fig. 1.

Experimental workflow for the patient samples of the study

To isolate the plasma, filtered and unfiltered blood samples were subjected to low-speed centrifugation (400g for 15 min at room temperature). The plasma was then further processed by high-speed centrifugation (16,000g) to obtain a sample pellet for DNA extraction using the ZISC-based Microbial DNA Enrichment Kit (Micronbrane Medical, Taiwan). In addition, the supernatant collected after high-speed centrifugation was used for cfDNA extraction, utilizing the iCatcher® Circulating cfDNA1000 Kit (CatchGene Co., Ltd., Taiwan) to isolate cfDNA for additional analysis.

Library Construction and NGS of patient samples

Library preparation for DNA extracted from the pellet was performed using the Illumina DNA Prep Library Kit (Illumina, USA), following the manufacturer’s instructions. After DNA augmentation, polymerase chain reaction amplification was conducted under the following conditions: top lid set to 100 °C; initial denaturation at 68 °C for 3 min, followed by 98 °C for 3 min; then 15 cycles of denaturation at 98 °C for 45 s, annealing at 62 °C for 30 s, and extension at 68 °C for 2 min; final extension at 68 °C for 1 min, and storage at 10 °C indefinitely. After amplification, the reaction was purified using Sample Purification Beads, and the product was eluted with 21 µL of Resuspension Buffer (Illumina, USA). A 20 µL aliquot of the supernatant was transferred to a new Lobind Eppendorf tube and stored at − 20 °C for future use.

The prepared libraries were sent to a service provider for sequencing on the Illumina NovaSeq platform, aiming for a minimum of 20 million reads per sample at a read length of 150 base pairs. For cfDNA extracted from plasma, NGS library preparation was carried out using the xGen™ ssDNA & Low-Input DNA Library Preparation Kit (Integrated DNA Technologies, USA), according to the manufacturer’s protocol.

Bioinformatics Pipeline and Results Interpretation

Sequencing reads were processed using fastp v0.23.2 [18], which removed adapter sequences and trimmed low-quality bases (quality score < Q30). Host-derived reads were identified and removed by mapping to the human genome (GRCh38) using the bwa-mem algorithm in bwa v0.7.17 [19]. The remaining reads were subsequently aligned to a microbial reference database using bwa v0.7.17. This database comprised approximately 1400 representative genomes of microorganisms, including bacteria, viruses, fungi, protozoa, and other multicellular eukaryotic pathogens, curated from the NCBI Nucleotide and Genome databases. The software identifies microorganisms based on the fold-change of microbial composition (microbial reads_%) over a no-template control (NTC).

Optimization of Read-Length of Analysis

Given the critical importance of turn-around time in pathogen detection workflows, we assessed the microorganism classification performance across different sequencing read lengths for eight known positive samples. As detailed in OSM Table S2, sequencing analysis was conducted for gDNA mNGS using read lengths of 150 bp, 120 bp, and 100 bp, both with and without host depletion filtration. To ensure sample comparability, the read counts were normalized to reads per million (RPM) of quality-controlled reads.

The results demonstrated high concordance across all read lengths for the eight samples. Notably, shorter read lengths (100 bp) did not increase the risk of “drop-outs” compared to longer reads. Therefore, using 100 bp reads, offering improved efficiency without compromising accuracy, was selected as the optimal approach for this study.

Results

Analytical Performance of Host Depletion and Microbial Passing Efficiency of the Novel ZISC-Based Filter

Using up to 13 mL of whole blood resulted in > 99% reduction of host cells, as shown in Fig. 2a. The novel ZISC-based filtration achieved consistent reductions of up to 6-log for blood volumes of 3 mL, 5 mL, and 8 mL. However, at volumes ≥ 10 mL, white blood cell binding saturation became evident. Given that the genome size of a human white blood cell is approximately 1000 times larger than that of a bacterium, even minimal residual human cells can still influence DNA sequencing outcomes. Enumeration of bacterial and viral particles spiked into whole blood, both pre- and post-ZISC-based filtration, confirmed that these particles passed through the filter unobstructed (Fig. 2b).

Fig. 2.

A Comparison of host depletion efficiency using the novel Zwitterionic Interface Ultra-Self-assemble Coating (ZISC)-based filter across different whole blood volumes. White blood cell counts before and after Devin filtration are denoted as I (input) and O (output). N.D. (not detected) indicates that white blood cells were below the detection limit after filtration. B Evaluation of microbial passing efficiency through the novel ZISC-based filter using whole blood samples spiked with (i) Escherichia coli, Klebsiella pneumoniae, Staphylococcus aureus, and (ii) feline coronavirus. C Comparison of performance and processing time of the novel ZISC-based filtration with other host depletion methods. D Microbial targets in molecular biology grade water or blood samples spiked with D6331 at 10² or 10⁴ Genome Equivalents/sample, with or without the novel ZISC-based filtration. W/o filter without filter

Comparison of the Novel ZISC-Based Host Depletion with Other Host-Depletion Procedures

The results demonstrated the novel ZISC-based filtration was the most effective method for enriching microbial reads by reducing human reads, regardless of the NGS platform used for DNA sequencing (Fig. 2c). Notably, the ZISC-based filtration was also the fastest and most operationally straightforward method. Quantitative polymerase chain reaction analysis further confirmed a significant reduction in host DNA, with Ct values decreasing from 28.5 pre-extraction to 33.4 post-filtration. In contrast, Ct values for I. halotolerans and H. halotolerans remained unchanged. In contrast, differential lysis and methylated-DNA removal methods, while effectively reducing host DNA, also resulted in the loss of microbial targets. Additionally, these alternative methods were significantly more time-consuming and labor-intensive than the novel ZISC-based filtration procedure.

Analytical Sensitivity of mNGS Workflow Incorporating the Novel ZISC-Based Host Depletion

Figure 2d presents the detection results for 17 target organisms spiked into D6331 at levels of 10² or 10⁴ (GE)/mL in molecular biology-grade water or whole blood. At 10⁴ GE in MBG, one target organism (Saccharomyces cerevisiae) was not detected, likely due to its low abundance, suggesting the assay’s limit of detection (LOD) may be approximately 150 GE. Notably, three target organisms were missed in the 10⁴ GE MBG sample without filtration. At the spiked-in level of 10² GE in MBG, all 17 target organisms were successfully identified, suggesting high analytical sensitivity for bacteria. The molecular biology-grade water samples showed some human DNA contamination. The filter may have enriched microbial detection due to some host DNA presence. Additionally, using human blood at spike-in levels of 10² and 10⁴ GE revealed that host DNA significantly affected the assay’ sensitivity. The estimated 150 GE LOD was achieved in the filtered samples.

Efficacy of the Novel ZISC-Based Filtration on the Number and Ratio of Microbial Reads of Patient Samples

Eight known-positive whole blood samples, identified in Table 1, were included in this study. These samples were obtained from patients admitted to the emergency department who were subsequently diagnosed with sepsis based on positive blood culture results.

Table 1.

Summary of next-generation sequencing (NGS) results in correlation with clinical microbial culture and reference microbial culture findings

Sample	Filtration	mNGS-gDNA					mNGS-cfDNA					Clinical diagnosis	mNGS-gDNA	Reads count	mNGS-cfDNA	Reads count
		QC reads	Human reads	Human (%)	Microbial reads	Unclassified reads	QC reads	Human reads	Human (%)	Microbial reads	Unclassified reads
Patient 1	Yes	21,047,043	19,081,056	91.75	1,055,832	659,848	12,944,153	10,921,003	84.37	1679	2,021,444	Escherichia coli	E. coli	1440	Escherichia coli	119
	No	20,334,492	20,260,019	99.80	31,069	10,246	15,445,795	14,030,099	90.83	1389	1,414,258		E. coli	47	Escherichia coli	133
Patient 2	Yes	30,831,571	30,024,457	98.89	209,581	126,839	15,727,170	15,340,577	97.54	5392	381,025	Proteus mirabilis	Proteus mirabilis	6234	Not detected	0
	No	40,191,141	39,900,756	99.56	122,485	54,525	15,220,843	15,015,917	98.65	11,806	192,600		Proteus mirabilis	1189	Not detected	0
Patient 3	Yes	32,747,888	32,698,973	99.94	15,576	4340	15,738,122	15,544,231	98.77	2819	191,066	Klebsiella pneumoniae	Klebsiella pneumoniae	7771	Klebsiella pneumoniae	1885
	No	22,253,181	22,250,570	99.99	1320	1238	12,134,274	10,196,735	84.03	33,510	1,904,021		Klebsiella pneumoniae	645	Klebsiella pneumoniae	669
Patient 4	Yes	24,234,465	24,200,363	99.94	9119	5256	4,225,531	515,902	12.21	5982	3,703,627	MRSA	S. aureus	55	Not detected	0
	No	27,464,006	27,459,234	99.99	1166	1850	13,997,776	13,954,709	99.69	117	42,944		Not detected	0	S. aureus	20
Patient 5	Yes	21,900,792	21,726,149	99.72	40,076	20,884	5,195,068	943,043	18.15	2378	4,249,610	Escherichia coli + Candida albicans	E. coli/C. albicans	525 / 139	E. coli/C. albicans	31 / 7
	No	26,747,021	26,708,028	99.94	9447	5543	4,216,824	104,201	2.47	2154	4,110,441		E. coli/C. albicans	66 / 24	E. coli/C. albicans	20 / 1
Patient 6	Yes	19,808,419	19,782,900	99.98	2100	1604	11,007,159	1,156,173	10.50	49,847	9,801,105	Pseudomonas aeruginosa	Pseudomonas aeruginosa	378	Pseudomonas aeruginosa	18
	No	22,249,899	22,247,175	99.99	314	861	11,815,395	3,183,334	26.94	38,548	8,593,466		P. aeruginosa	54	Pseudomonas aeruginosa	38
Patient 7	Yes	20,852,636	19,326,723	97.61	243,857	228,563	13,393,903	12,612,294	94.16	2079	779,421	Achromobacter species	Achromobacter insolitus	9407	Achromobacter insolitus	1424
	No	27,361,083	27,086,096	99.58	62,132	52,338	5,229,878	788,980	15.09	4122	4,436,733		Achromobacter insolitus	8168	Achromobacter insolitus	26
Patient 8	Yes	22,938,400	22,715,917	99.26	77,141	92,178	9,669,278	604,357	6.25	45,415	9,019,491	Escherichia coli	Escherichia coli	1253	Escherichia coli	59
	No	20,696,706	20,691,316	99.98	1670	1535	13,118,788	7,638,602	58.23	23,769	5,456,321		E. coli	16	Escherichia coli	63

gDNA genomic DNA, cfDNA cell-free DNA, QC quality control, mNGS metagenomic next-generation sequencing

A comparison of microbial reads per million quality-controlled reads between filtered and unfiltered samples revealed a significant increase in microbial RPM in eight of the nine sample pairs after filtration. For blood culture-positive samples, host depletion via filtration enhanced microbial RPM from an average of 925 RPM to 9351 RPM, representing a tenfold increase (Table 2). Concurrently, human reads decreased from 997,116 RPM to 973,172 RPM after filtration. Despite the reduction, human reads remained the predominant class of sequences in whole blood samples. However, even a modest decrease in human RPM yielded a substantial increase in microbial RPM.

Table 2.

Normalized read counts (reads per million) for cell-free DNA (cfDNA), genomic DNA (gDNA), and human DNA, along with gDNA enrichment efficiencies

Sample	Filtration	Reads per million			gDNA efficiency
		cfDNA	gDNA	Human
Patient 1	Yes	130	50,165	906,591	32.83
	No	90	1528	996,338
Patient 2	Yes	343	6798	973,822	2.23
	No	776	3048	992,775
Patient 3	Yes	179	476	998,506	8.02
	No	2762	59	999,883
Patient 4	Yes	1416	376	998,593	8.86
	No	8	42	999,826
Patient 5	Yes	458	1830	992,026	5.18
	No	511	353	998,542
Patient 6	Yes	4529	106	998,712	7.51
	No	3263	14	999,878
Patient 7	Yes	155	11,694	926,824	5.15
	No	788	2271	989,950
Patient 8	Yes	4697	3363	990,301	41.68
	No	1812	81	999,740
Average	Yes	1488	9351	973,172
Average	No	1251	925	997,116
Fold-change (p-value)		1.19 (p = 0.351)	10.11 (p = 0.041)	0.98 (p = 0.084)

A plot of microbial RPM before and after filtration (Fig. 3a) demonstrated a wide range of enrichment efficiencies across samples, with fold enrichment ranging from 2.23 to 41.68 (Table 2). Microbial read enrichment was more pronounced in samples with lower initial microbial reads entering the filtration process.

Fig. 3.

A Comparison of sample reads (in reads per million, RPM) with and without filtration, is arranged in order of decreasing reads for the “Without Filter” samples. The Y-axis (sample output in RPM) is shown on a log scale. B Plot of microbial genome % from “Without Filter” samples vs. “With Filter” samples

The novel ZISC-based filtration must not alter the microbial composition (microbial%) of individual species in a sample, as the proportion of pathogen-specific reads (% microbial reads) is clinically significant, especially during infection when pathogen abundance exceeds that of normal baselines [20]. To assess this, the microbial% reads for each species in filtered samples were compared to those in unfiltered samples. The results showed a strong correlation (R² = 0.90) between filtered and unfiltered microbial% reads, closely aligning with a 1:1 ratio (y = 0.9x) (Fig. 3b).

Comparison Between gDNA-Based and cfDNA-Based mNGS for Pathogen Detection

The results revealed that gDNA-based mNGS with host-depletion filtration successfully detected sequences representing the known pathogens in all samples (100%, 8/8). In contrast, gDNA-based mNGS without host-depletion filtration identified expected pathogen sequences in 87.5% (7/8) of the samples, with one sample failing to detect the expected pathogen. The cfDNA-based mNGS detected most of the expected pathogens, though with lower read counts (Table 1). Specifically, 75% (6/8) and 87.5% (7/8) of samples exhibited expected pathogen reads using cfDNA-based mNGS with and without filtration, respectively. Detected microorganisms included Gram-negative and Gram-positive bacteria and one yeast species.

Among the three tested conditions—gDNA with filtration, gDNA without filtration, and cfDNA—the gDNA-based mNGS with host-depletion filtration demonstrated the highest microbial detection sensitivity. This was attributed to the enrichment of microbial reads via filtration. Without filtration, the microbial RPM was similar between gDNA (925) and cfDNA (1251). However, with filtration, gDNA-based mNGS achieved a significantly higher microbial RPM (average 9351), while cfDNA-based mNGS only reached an average of 1488, slightly lower than the unfiltered cfDNA process. For gDNA, the enrichment (fold-change) was statistically significant (p = 0.041), while the corresponding values for cfDNA and human DNA were not significant (p = 0.351 and p = 0.084).

These findings indicate that gDNA, unlike cfDNA, is amenable to host depletion and enrichment of microbial sequences. The cfDNA-derived signal, closer to baseline noise, poses a greater risk of false-positive detections. Overall, gDNA-based mNGS with host-depletion filtration performed best, followed by gDNA without filtration and cfDNA (with or without filtration).

Discussion

This study investigated the analytical performance of the novel ZISC-based filter to remove nucleated human cells in whole blood samples. Our findings demonstrated that the novel filtration method efficiently removes human cells while leaving bacterial and viral particles largely unaffected. Theoretically, this host depletion strategy could be applied to whole blood and extended to other biological fluids, such as cerebrospinal fluid, for similar removal of human cells. However, it was important to recognize that even after eliminating several orders of magnitude of white blood cells, human DNA can still be detected in clinical samples. This was primarily because the human genome was approximately 1000 times larger than a typical bacterial genome allows residual human DNA to dominate sequencing reads. Additionally, the presence of cell-free human DNA that has not been removed by cell-based filtration methods may further contribute to the persistence of host-derived sequences.

A major challenge in mNGS was the overwhelming background of host DNA in human bio-specimens, particularly whole blood. Although various host depletion strategies have been explored, many are impractical for clinical implementation due to limited efficacy and inconsistent performance. In this study, the novel filtration-based method for depleting host cells and DNA resulted in a marked increase in microbial read enrichment, especially in samples with low microbial loads. Notably, this host depletion approach enhanced the analytical sensitivity of the assay without altering the relative abundance of microbial species in the sample. As anticipated, mNGS with host depletion successfully detected the expected microbial sequences in all eight test samples. In contrast, samples processed without filtration or using only cfDNA exhibited “drop-outs” at lower microbial concentrations.

It was essential to thoroughly characterize the analytical performance of clinical laboratory tests to understand their capabilities and limitations and to ensure they were appropriate for their intended clinical use [21]. In clinical molecular pathology, the ideal LOD for bacterial genome detection should be as low as reasonably achievable while still maintaining analytical sensitivity and specificity [22]. Generally, LODs in the range of 100–1000 genome equivalents per milliliter (GE/mL) were considered acceptable for clinical applications [23]. In this study, the LOD was determined to be 150 GE/mL per sample, and this threshold appeared to be validated across eight positive blood culture samples, reflecting its practical clinical utility. Because the high background of host DNA can significantly reduce the sensitivity of mNGS, sample filtration was used to reduce the abundance of human DNA, which otherwise interferes with microbial signal detection. This step was crucial in preserving both the sensitivity and specificity of the assay.

cfDNA was compared to gDNA-based mNGS in this study, as many mNGS-based pathogen detection workflows rely on cfDNA. As expected, no microbial enrichment was observed in cfDNA samples when comparing filtered versus unfiltered conditions, indicating that host depletion by filtration is not applicable to cfDNA-based approaches. We also found that cfDNA samples without host depletion yielded 1488 RPM, consistent with the Liu et al. study [16], a range of 50–2000 RPM observed in cfDNA samples without host depletion. In contrast, gDNA samples processed with host depletion achieved 9351 RPM, indicating that gDNA-based mNGS was amenable to host depletion and microbial sequence enrichment. These findings suggested that gDNA combined with host depletion filtration not only enhances sensitivity but also optimizes sequencing efficiency by reducing the proportion of host-derived reads.

Limitations

This study had several limitations. First, the D6331 reference material used for analytical evaluation lacked sufficient representation of clinically relevant pathogens, including fungal, protozoan, and viral organisms. Second, the detection of intracellular pathogens was not assessed, raising the possibility that such organisms may be unintentionally removed during the filtration process. For more comprehensive validation, a well-defined panel of clinically important pathogens representing major microbial groups, such as Gram-positive and Gram-negative bacteria, yeasts, filamentous fungi, DNA viruses, and protozoa, should be developed and used in analytical testing. It was also essential to determine the extraction efficiency for each organism type to ensure reliable detection is not compromised by poor nucleic acid recovery. For clinical validation, the diagnostic relevance of pathogens identified by mNGS, particularly in culture-negative samples, should be carefully evaluated. A third limitation was that some detected organisms, commonly considered environmental contaminants, may be excluded by reference databases, potentially affecting detection accuracy. Finally, distinguishing true infection from colonization remains a significant interpretive challenge, especially when mNGS detects organisms in culture-negative specimens.

Conclusion

In conclusion, this study evaluated the impact of host depletion on the enrichment of microbial reads in both gDNA and cfDNA using known culture-positive blood samples. The results demonstrated that host-depletion filtration significantly increased microbial read recovery in gDNA samples while maintaining the relative microbial composition (% microbial content) within each sample. Importantly, the workflow provided improved analytical sensitivity compared to conventional mNGS approaches, highlighting its potential utility as a good tool for clinical pathogen detection.

Supplementary Information

Below is the link to the electronic supplementary material.

Declarations

Funding

The study was supported by the research support scheme of Taipei Veterans General Hospital Grant No. T21006. Micronbrane (Taiwan), the manufacturer of the Devin filter, did not provide any funding or personnel support for this study.

Conflict of Interest

The authors (CYC, LPH, CYW, YHT, HCK, CCM) have no conflicts of interest related to this study.

Data Availability

The data supporting the findings of this study are available within the article and its supplementary materials. Additional datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Code Availability

Not applicable.

Ethics Approval

This study was conducted according to the ethical standards of the Institutional Review Board of Taipei Veterans General Hospital (Institutional Review Board number: 2021-03-013AC; approval date: 22 March 2021). All procedures involving human participants complied with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Consent to Participate

Written informed consent was obtained from all participants before sample collection.

Consent to Publish

The authors affirm that all participants provided informed consent to publish relevant data and images in the study.

Authors’ Contributions

Yen Chia Chen: Conceptualization, methodology, and writing – original draft. Po Hsiang Liao: Software, data curation, and formal analysis. Yen Wen Chen: Visualization, investigation, and data curation. David Hung Tsang Yen: Investigation and supervision. Chorng Kuang How: Supervision and validation. Chia Ming Chang: Conceptualization, writing – reviewing and editing, and project administration.