Comparative diagnostic performance of metagenomic and two targeted sequencing methods in lower respiratory infection

Categories: Disease & Virus

July 28, 2025

Study population

In this study, a total of 205 BALF samples were collected from patients suspected with lower respiratory tract infections (Supplementary Table 1). Each sample underwent testing using mNGS, amplification-based tNGS, capture-based tNGS, and CMTs. Based on the comprehensive clinical diagnosis, 174 patients were identified as lower respiratory tract infection, and the other 31 were diagnosed as non-infection. Among 205 samples, capture-based tNGS, mNGS, and amplification-based tNGS reported microorganisms in 193, 188, and 179 samples, respectively (Fig. 1).

Cost and turnaround time comparisons of three NGS

The cost and turnaround time (TAT) of three NGS used in this study were compared (Fig. 2). The cost of mNGS (840 USD) were far higher than those of the two tNGS (130–250 USD). The cost of amplification-based tNGS (130 USD) was the lowest in this study. As can be seen from the workflow, this huge difference was mainly due to the data size of sequencing. Which was 20 million single end (SE) 75 bp reads in mNGS, 0.1 million SE 100 bp reads in amplification-based tNGS, and 5 million SE 100 bp reads in capture-based tNGS.

The TAT of mNGS (20 h) was also longest compared to that of amplification-based tNGS (12.5 h) and capture-based tNGS (16.5 h). The amplification-based tNGS was the fastest approach in this study. For nucleic acid extraction, mNGS took more time due to the independent extraction of DNA and RNA, while that were extracted together in the two tNGS. For library preparation, mNGS was faster than the two tNGS because it did not need to enrich specific nucleic acid. Compared to multiplex PCR amplification, probe hybridization was slower, which was why the capture-based tNGS took the most time in this section. For sequencing and data analyzing, the two tNGS both took 5.5 h, while mNGS took more than twice that time (12 h). Because mNGS required a platform with higher throughput due to its large data size requirement, which also means it needs more time in sequencing and data analyzing.

Comparison of microorganisms reported by three NGS

To clarify the species spectrum size of the three NGS in real clinical cases, the reported species were counted (Fig. 3A). There were obvious differences among the species spectra of the three NGS. The species spectrum size of mNGS was the largest, covering 80 species. Whereas the value was 71 and 65 for the capture-based tNGS and amplification-based tNGS respectively. Specifically, mNGS solely identified 21 species, including 9 gram-negative bacteria, 6 gram-positive bacteria, 2 fungi, 2 DNA virus, and 2 RNA virus. Despite this, the two tNGS still solely identified some species, involving 5 gram-negative bacteria, 3 gram-positive bacteria, 3 DNA virus, and 2 RNA virus for the capture-based tNGS, and 4 RNA virus, 3 fungi, and 2 gram-positive bacteria for the amplification-based tNGS.

In the 205 samples, the capture-based tNGS detected the most microorganisms, identifying 678 in total, followed by mNGS with 535 and the amplification-based tNGS with 481 (Fig. 3B; Supplementary Fig. 1). The three NGS consistently detected 280 microorganisms, mainly consisting of DNA virus, fungi, and gram-negative bacteria. Besides, 149 microorganisms were consistently identified by mNGS and the capture-based tNGS, with a high proportion of gram-positive bacteria. In comparison, the two tNGS consistently identified 89 microorganisms, with the exception of RNA viruses (N = 47), the majority were DNA viruses and bacteria. The capture-based tNGS (78.38%, 58/74) and the amplification-based tNGS (76.32%, 58/76) showed a high degree of concordance in the identification of RNA viruses. 43 microorganisms were consistently identified by mNGS and the amplification-based tNGS, which had a high fungi proportion. For the microorganisms reported solely, there was the most in the capture-based tNGS (N = 160), followed by the amplification-based tNGS (N = 69) and mNGS (N = 63). More DNA virus and less fungi were in the capture-based tNGS, more gram-positive bacteria and RNA virus were in the amplification-based tNGS, and more gram-negative bacteria was in mNGS (Fig. 3B).

In order to gain a deeper understanding of the characteristics of the three NGS in reporting, the distribution of microorganisms reported was analyzed. The most common microorganisms included human betaherpes virus 5 (HHV-5), human gammaherpes virus 4 (HHV-4), severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), human alphaherpes virus 1 (HHV-1), Candida albicans, Acinetobacter baumannii, Streptococcus pneumoniae, and Pneumocystis jirovecii (Supplementary Fig. 2). Among these, HHV-5, C. albicans, HHV-4, P. jirovecii, A. baumannii, HHV-1 were frequently reported consistently by the three NGS (Fig. 3C). Moreover, mNGS mainly missed HHV-5 (N = 29), S. pneumoniae (N = 27), HHV-1 (N = 25), and HHV-4 (N = 23), which were reported by the other two NGS, especially the capture-based tNGS (Fig. 3D). But for SARS-CoV-2, that was because 165 (80.49%) samples were not extracted RNA for mNGS considering the sample volume in this study. However, even within the remaining 40 samples where RNA extraction was performed, mNGS exhibited a lower sensitivity for SARS-CoV-2 detection, missing 47.37% (9/19) of positive cases compared to the combined results from the two tNGS (Supplementary Fig. 3). The amplification-based tNGS mainly missed HHV-5 (N = 35), Enterococcus faecium (N = 32), HHV-4 (N = 30), Corynebacterium striatum (N = 29), S. pneumoniae (N = 23), and HHV-1 (N = 22) (Fig. 3E). Notably, among these microorganisms, E. faecium and C. striatum were not included in the panel of the amplification-based tNGS used in this study. While the capture-based tNGS missed fewer microorganisms, mainly C. albicans (N = 14), Human betaherpes virus 7 (N = 13), and Fusobacterium nucleatum (N = 12) (Fig. 3F).

To figure out the reason of different reported microorganisms between mNGS and tNGS, the RPM of consistent microorganisms reported by mNGS and the capture-based tNGS were compared (Fig. 3G). The amplification-based tNGS was not included in the comparison because it was based on the principle of PCR amplification, and it is not possible to obtain the RPM it originally enriched. In the microorganism species commonly missed by mNGS, the capture-based tNGS reported significant more reads, indicating the sensitivity of the capture-based tNGS was significantly higher than mNGS. This is probably why the capture-based tNGS reported more microorganisms than mNGS.

Diagnostic value for lower respiratory tract infection

Benchmarked against the comprehensive clinical diagnosis, the diagnostic value of three NGS were evaluated in the Table 1. The capture-based tNGS demonstrated the highest overall accuracy and sensitivity among three NGS, achieving 93.17% and 99.43%, respectively. In comparison, mNGS showed an accuracy of 90.73% and sensitivity of 91.95%, while amplification-based tNGS showed an accuracy of 90.24% and sensitivity of 92.53%. However, the specificity of capture-based tNGS was the lowest, which was just 58.06%. While that of mNGS and amplification-based tNGS were 83.87% and 77.42%. Overall, the capture-based tNGS performed significantly higher diagnostic value (Padj < 0.017), including accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). There was no difference between mNGS and the amplification-based tNGS.

Table 1 Overall performance of three NGS.

To gain a deeper understanding of the differences among the three NGS, an assessment was conducted at the pathogen type level (Table 2). Specifically, mNGS showed high accuracy and sensitivity across most of pathogen types, with values ranging from 92.20 to 96.10% and 90.11–98.85%, respectively. However, for RNA virus, these metrics were significantly lower, with accuracy at 76.10% and sensitivity dropping to just 24.62%. Even focusing solely on samples subjected to mNGS RNA extraction, the assay’s sensitivity remained at 68.18% (15/22) (Supplementary Table 2). Despite this, the specificity of mNGS remained consistently strong across all pathogen types, ranging from 91.15 to 100.00% (Table 2). Besides, the specificity of amplification-based tNGS was also robust, which was ranging from 91.53 to 98.25%. The amplification-based tNGS demonstrated high accuracy in DNA virus (90.73%), RNA virus (94.15%), and fungi (93.17%), but reduced accuracy in gram-positive bacteria (69.76%). Sensitivity was particularly strong for RNA virus (93.85%) and fungi (91.95%), yet notably poor for gram-positive bacteria (40.23%) and gram-negative bacteria (71.74%). For capture-based tNGS, aside from a specificity of 74.78% for RNA virus, the accuracy, sensitivity, and specificity across all pathogen types were consistently high, ranging from 83.90 to 98.91%. In comparison, for both gram-positive and gram-negative bacteria, the amplification-based tNGS exhibited a lower diagnostic value (Padj < 0.017). In DNA virus, the capture-based tNGS showed inferior identification performance than the amplification-based tNGS (Padj < 0.017). For the identification of RNA virus, mNGS performed worse than the two tNGS (Padj < 0.017). While the diagnostic value of the three NGS showed no significant difference in fungi (Padj > 0.017).

Table 2 Performance of three NGS in specific pathogen types.

The diagnostic value of the three NGS at the pathogen level was evaluated. The identification accuracy of mNGS for SARS-CoV-2, Aspergillus fumigatus, and Haemophilus influenzae was lower than that of both tNGS, with the largest discrepancy observed for SARS-CoV-2 (Fig. 4A). Notably, even in samples not subjected to RNA extraction for mNGS analysis, the diagnostic accuracy of mNGS for SARS-CoV-2 detection remained suboptimal (85.00% vs. 92.50% vs. 92.50%) (Supplementary Fig. 4). Conversely, the amplification-based tNGS demonstrated reduced accuracy for certain pathogens, including E. faecium, Haemophilus parainfluenzae, C. striatum, and Streptococcus mitis (Fig. 4B). Notably, H. parainfluenzae and S. mitis were also absent from its detection panel. In contrast, the capture-based tNGS showed significantly lower accuracy for HHV-5, S. pneumoniae, and HHV-4, which was primarily due to reduced specificity (Fig. 4C). However, for pathogens like Mycobacterium tuberculosis, P. jirovecii, Klebsiella pneumoniae, and A. baumannii, there were no significant differences in identification accuracy across the three NGS (Fig. 4D).

Genotypes, AMR genes, and VFs

In addition, the capability to identify pathogen genotypes was assessed in this study. For SARS-CoV-2, mNGS was limited to species-level identification, prompting a comparison just between the two tNGS. Among all the SARS-CoV-2 genotypes consistently identified by both tNGS methods, it was always the capture-based tNGS that provided a more refined genotype distinction (Fig. 5A).

Remarkably, examination of the outputs from the two tNGS unveiled the presence of numerous antimicrobial resistance (AMR) genes and virulence factors (VFs). The two tNGS consistently identified several genes, including blaKPC, mecA, iucA, peg344, among others (Fig. 5B and C). However, the capture-based tNGS approach demonstrated superior detection capabilities, identifying a broader spectrum of AMR genes and VFs.

Source link