A total of 130 patients (184 specimens) collected from Ruijin Hospital, Shanghai Jiao Tong University School of Medicine between February 2019 and December 2020 were investigated. Specimens were subjected to regular clinical microbiological assays and mNGS testing in parallel. The study was exempted from ethical review as it was a retrospective study and patient data were anonymized.

The inclusion criteria were in accordance with previous studies, that is a PaO² <60 mmHg in room air. Baseline data including demographic characteristics, comorbidities, immunosuppressive state, treatment process, routine blood, and inflammatory markers were recorded. Different specimens, such as sputum, bronchoalveolar lavage fluid (BALF), blood, pleural effusion, ascitic fluid, and urine, were collected.

1) Bacteria, fungi, and mycobacteria were identified using the culture method as follows. Bacteria: specimens were inoculated into media containing blood agar, brain heart infusion broth, chocolate agar, and thioglycollate. The colonies were used to identify bacterial species using the VITEK 2 compact automated system. Fungi: specimens were inoculated into media containing Sabouraud agar and potato glucose agar media. The fungi were identified according to characteristics of the colonies and microscopic characteristics of hyphae and spores. Mycobacteria: sputum was digested with 4% NaOH digestive juice, and fluid specimens were collected after centrifugation for mycobacterium testing. In a 37°C incubator, 0.5 ml of resuspended sputum pellet and 1 ml fluid were inoculated in MGIT. MGIT was put into BACTEC 960 system (BD Microbiology Systems), which could detect automatically. If a positive result appeared, direct acid-fast staining of the smear was done to confirm the positive samples of the mycobacteria. 2) Viruses, Mycoplasma, and Legionella pneumophila were confirmed using the PCR assay as previously described. Briefly, nucleic acids of Mycoplasma and L. pneumophila were extracted from each specimen using commercial assays (Qiagen, Germany). They were eluted in distilled water and stored at −20°C for future use. PCR reaction was performed using Applied Biosystems 7500 Real-Time PCR system (Applied Biosystems, USA). Distilled water was used as negative control.

For mNGS sequencing, about 300 μl plasma was separated from 3 ml of blood, and DNA was extracted using the TIANamp Micro DNA Kit. With regard to other specimens (BALF, sputum, pleural effusion, ascitic fluid, and urine), 600 μl specimen, enzyme, and 1 g 0.5 mm glass bead were mixed and agitated vigorously at 2,800–3,200 rpm for 30 min, then 300 μl specimens were used for DNA extraction using the TIANamp Micro DNA Kit. DNA was fragmented and constructed by an end-repair, adapter-ligation, and PCR amplification. Quality control of the DNA libraries was performed by Agilent 2100, and then quality qualified libraries were sequenced by MGISEQ-2000 platform (Jeon et al., 2014). Next, raw sequencing data were exposed to quality control, including the removal of low-quality reads (Q score cutoff, 20), elimination of human host sequences through mapping to the human reference genome (containing hg19) using Burrows-Wheeler Alignment (Li and Durbin, 2009). The remaining data were aligned to the curated non-redundant Microbial Genome Databases (bacteria, fungi, viruses, parasites). The mapped reads were annotated and the number of species-specific sequences of the specimens were counted.

All of the taxa in the classification reference databases were downloaded from NCBI RefSeq genome database (ftp://ftp.ncbi.nlm.nih.gov/genomes/).

The multiplex PCR-based targeted gene sequencing technology was performed to identify targeted pathogens, and all of the relevant pathogens explored in sequencing are listed in Supplementary Table S5 . The workflow was conducted as previously described. Briefly, nucleic acids were extracted according to the instructions of the manufacturer (ZymoBIOMICS™ DNA/RNA Miniprep Kit, Zymo, R2002). Then, a PCR reaction system (including enzyme mix, DNA, RNA, and nuclease-free water) was performed as follows: denaturation at 95°C (3 min); 25 cycles of denaturation at 95°C (20 s), annealing for 20 s, and extension at 60°C for 4.5 min; final extension at 72°C (5 min). The PCR reaction products were purified after amplification. After sequencing adapter assembly, qualified DNA library was sequenced. The reads were filtered, <60 bp reads or non-specific primer binding was removed, and finally, the clean reads were used for sequencing analysis.

Categorical variables were expressed as percentages, and continuous variables were expressed as mean ± SD if normally distributed. Comparative analysis was conducted by Pearson’s test, Fisher’s exact test, or the McNemar test where appropriate. Data analysis was performed with SPSS 22.0 and Prism. P < 0.05 was considered significant.

A total of 130 patients were enrolled in the study. Their demographic characteristics, comorbidities, laboratory examination results, and type of specimen used are summarized in Supplementary Table S1 . The data showed that 51 out of 130 (39.2%) patients were female, and 79 were male, with a median age of 62.0 years. Among the patients included in the study, 21 (16.15%) patients were immunocompromised, 17 (13.07%) patients had autoimmune diseases, and 29 (22.3%) patients had been diagnosed with cancer. A total of 184 specimens were tested, consisting of 47 sputum, 55 BALF, 74 blood, 6 pleural effusion, 1 ascitic fluid, and 1 urine specimen.

Figure 1 shows the etiological results; a total of 110 microbes were detected by mNGS, including 68 bacteria, Mycobacterium tuberculosis (MTB), 4 Mycoplasma, Legionella pneumophila, 17 fungi, and 18 viruses.

We found that the most frequent bacterial pathogens were K. pneumoniae, A. baumannii, Pseudomonas aeruginosa (P. aeruginosa), Enterococcus faecium (E. faecium), Staphylococcus aureus (S. aureus), and Prevotella melaninogenica (P. melaninogenica), all of which were more likely to be detected in BALF and sputum than in other specimen types. Microbes like P. melaninogenica, Lautropia mirabilis (L. mirabilis), Streptococcus pseudopneumoniae (S. pseudopneumoniae), Acinetobacter nosocomialis (A. nosocomialis), Rothia mucilaginosa (R. mucilaginosa), Tropheryma whipplei (T. whipplei), and Burkholderia multivorans (B. multivorans) were present in BALF and sputum only ( Figure 1A ). The results also showed that 10, 19, and 3 bacteria were exclusively detected in BALF, sputum, and blood, respectively ( Supplementary Figure S1 ). The most commonly detected fungi were P. jirovecii, C. albicans, C. glabrata, C. tropicalis, and Aspergillus fumigatus (A. fumigatus). P. jirovecii was more commonly detected in BALF, sputum, and blood, while C. albicans and C. tropicalis identifications were limited to BALF and sputum, and A. fumigatus was detected in blood and sputum specimens only ( Figure 1A ). We also found that the most frequently detected viruses were the Epstein–Barr virus (EBV), CMV, and HSV1. Mycoplasma orale (M. orale), Mycoplasma pneumoniae (M. pneumoniae), Ureaplasma parvum (U. parvum), Legionella pneumophilia (L. pneumophilia), and Ureaplasma urealyticus (U. urealyticus) were also detected by mNGS ( Figure 1A ).

Subsequently, we compared the total read numbers of detected microbes in different specimen types. The results show that the sequences of K. pneumoniae in BALF (n = 9) were higher than those in blood (n = 3) and sputum (n = 8). The average sequences of A. baumannii in BALF (n = 10) exceeded the read numbers in blood (n = 3) and sputum (n = 5). We also noted that the identified sequences of P. aeruginosa in BALF outnumbered those in blood and sputum. However, they did not reach statistical difference ( Figure 1B ). Among the fungal sequences, the sequences of P. jirovecii in BALF (n = 8) were higher than those in blood (n = 13) and sputum (n = 7). The total reads of C. albicans in blood were significantly higher than those in BALF ( Figure 1C ). Besides, we found that, compared with blood and sputum specimens, the viral sequences of CMV were more abundant in BALF specimens, even though it was not statistical significant ( Figure 1D ). The read numbers of HSV1 in BALF were significantly higher than those in blood and sputum ( Figure 1D ).

Unlike mNGS, among the microbes isolated by conventional methods, Stenotrophomonas maltophilia (S. maltophilia) was the most common pathogen, followed by A. baumannii, K. pneumoniae, P. aeruginosa, and S. aureus ( Figure 2A ). The most frequent fungi detected by the conventional method were C. albicans, C. tropicalis, C. glabrata, A. fumigatus, and C. parapsilosis ( Figure 2C ). Mycoplasma pneumoniae were the only microbes isolated by the conventional method ( Figure 2D ). The viral detection revealed EBV, CMV, and HSV1 as the top 3 viruses, as determined by only mNGS ( Figure 2E ).

The comparisons of microbial detection by the mNGS and conventional methods (184 specimens) can be visualized in Figure 2 . The percentage of mNGS-positive specimens was significantly higher than that of conventional-positive specimens with regard to MTB (OR, ∞; 95% CI, 1–∞; P < 0.05), bacterial (OR, 3.7; 95% CI, 2.4–5.8; P < 0.0001), fungal (OR, 4.37; 95% CI, 2.7–7.2; P < 0.0001), mycoplasma (OR, 10.5; 95% CI, 31.8–115; P = 0.005), and viral (OR, ∞; 95% CI, 180.7–∞; P < 0.0001) detections.

In detail, E. faecium, P. melaninogenica, Enterococcus faecalis (E. faecalis), Corynebacterium striatum (C. striatum), L. mirabilis, S. pseudopneumoniae, A. nosocomialis, R. mucilaginosa, Rothia dentocariosa (R. dentocariosa), and Streptococcus mitis (S. mitis) exhibited a significantly higher detection rate by the mNGS method compared with that noted by the conventional method (P < 0.05) ( Figure 2A ). However, A. baumannii was observed to have a significantly higher yield rate by conventional than by mNGS (P < 0.05) ( Figure 2A ). The frequent K. pneumoniae and P. aeruginosa showed similar detection rate by both methods ( Figure 2A ). We also found that mNGS showed a higher detection rate for both G+ and G− bacteria than the conventional method ( Figure 2B ).

With regard to fungal detection, Aspergillus flavus (A. flavus) showed a higher detection rate by the mNGS method compared with the conventional method (P < 0.05) ( Figure 2C ). Of note, 20 patients (28 specimens) were exclusively detected with P. jirovecii by mNGS, and most of the patients were immunocompromised ( Supplementary Table S2 ). For patients #6, #56, #60, #69, and #83, blood mNGS identified 23, 240, 164, 72, and 1,097 reads of P. jirovecii. The clinical treatment of patients #6, # 69, and #83 was changed, and the treatment of patients #56 and #60 remained unchanged with the continuation of sulfamethazine (SMZ). Sputum mNGS of patients #79 and #90 detected 100 and 31 reads of P. jirovecii, the treatment of voriconazole for patient #79 was discontinued, while therapy of patient #90 was not changed. Positive results of BALF mNGS were detected in four patients (#106, 4,627 reads; #114, 790 reads; #115, 1,020 reads; #116, 1,490 reads), of which three patients were added with SMZ therapy. Five patients (#9, #24, #25, #47, #62) had P. jirovecii detected in both blood and sputum specimens by mNGS; their treatment was altered, with replacement or discontinuation of broad-spectrum antibiotics and the addition of SMZ. In addition, both BALF and blood mNGS were positive in three patients (#43, #50, #78), and oseltamivir therapy of patient #78 was replaced with SMZ and fluconazole ( Supplementary Table S2 ).

The percentage of mNGS-positive specimens was higher than that of the conventional method for C. albicans and C. glabrata, although the differences were not significant (P > 0.05). The percentages of positive specimens were similar in the identification of A. fumigatus, C. tropicalis, and C. parapsilosis by both methods ( Figure 2C ).

Interestingly, some patients were also detected with M. orale (n = 5 specimens), CMV (n = 45 specimens), HSV4 (n = 38 specimens), HSV1 (n = 25 specimens), Human parvovirus B19 (n = 6 specimens), HSV7 (n = 4 specimens), and Torque teno virus (n = 4 specimens); additionally, their yield rate was significantly higher by the mNGS method than by the conventional method ( Figures 2D, E ).

On the basis of conventional detection results (130 patients), 70 (53.8%), 33 (25.3%), 14 (10.7%), and 5 (3.8%) patients were diagnosed with mixed infection, bacterial infection, fungal infection, and viral infection. With regard to mNGS results, the microbial detection rate (100%) was significantly higher in viral infection, followed by mixed infection (73.9%), fungal infection (63.2%), and bacterial infection (59%) ( Figure 3A ). Among the patients with mixed infection, we noted that 31 (44.3%), 25 (35.7%), 5 (7.1%), 4 (5.7%), 4 (5.7%), and 1 (1.4%) patients had mixed bacterial–fungal, bacterial–fungal–viral, bacterial–viral, bacteria–mycoplasma, bacterial–viral, bacterial–MTB, and fungal–viral infections ( Figure 3A ). The potential detection rate of mNGS in viral infection was the highest, followed by mixed infection, fungal infection, and bacterial infection. However, the positive rate of the conventional method in viral infection was significantly lower than infection from other etiologies ( Supplementary Table S3 ).

Among all 130 patients, the pathogens of 58.46% patients were uncovered by an empirical antibiotic regimen, and the antibiotic prescriptions were modified in these patients ( Figure 3B ). However, there was no difference in death rate between patients with modified therapy and those without change (P > 0.05) ( Supplementary Figure S2 ). Among the patients with different infection types, most patients had modified therapies, especially mixed infection and viral infection. However, the death rate was not significantly different between those with modified therapy and those without change (P > 0.05) ( Figure 3B and Supplementary Figure S2 , data not shown).

With respect to bacterial identification, mNGS showed positive results in 12 conventional-negative specimens (sputum—62, sputum—75, BALF—82, sputum—100, sputum—104, sputum—108, BALF—121, BALF—125, blood—132, sputum—151, sputum—156, sputum—157). For fungal identification, mNGS yielded positive detection in 10 conventional-negative specimens (blood—11, sputum—75, BALF—82, sputum—88, blood—99, sputum—100, sputum—104, blood—124, sputum—153, BALF—161). In addition, we found in patient #55 (sputum–88) that mNGS and conventional methods produced conflicting data, with mNGS results revealing positive detection of K. pneumoniae, while A. baumannii and S. aureus were detected by the conventional method ( Supplementary Table S4 ).

mNGS detection rate of bacteria and virus was higher in patients with orotracheal intubation and type 2 diabetes mellitus (T2DM) ( Figures 4A, C ). Immunocompromised patients were found to have a significantly higher detection rate of fungi and virus, indicating the predisposition of fungal and viral infections in patients with immune deficiencies ( Figure 4B ). We also demonstrated a high mNGS-positive rate of bacteria in patients whose detection time was 15 days after being admitted to the hospital ( Supplementary Figure S3B ). However, the clinical outcome (death) and symptom (fever) were not associated with the mNGS microbial detection rate ( Supplementary Figures S3A, C ). We further assessed the effect of laboratory examinations on mNGS detection rate. Blood WBC >100 * 10⁶/L and procalcitonin (PCT) >5 μg/L were related with significantly higher mNGS detection rate of bacteria and lower detection rate of fungi ( Figures 4D, F ). C-reactive protein (CRP) >90 mg/L was associated with lower rate of fungal detection, while there was no relationship between ESR levels and mNGS detection rate ( Figure 4E and Supplementary Figure S3D ). These results indicate more common bacterial infections than fungal infections in patients with high WBC, CRP, and PCT. Additionally, further measurement on the influence of IL-1β, IL-2, and IL-6 levels on mNGS detection rate showed that in CNS infections, mNGS detection rate was significantly higher in patients with blood IL-2 >2,000 pg/ml, and IL-6 >40 pg/ml was related with higher mNGS detection rate of fungi or viruses, respectively ( Supplementary Figures S3E–G ).

A total of 13 patients were enrolled. mNGS and pathogen targets NGS (ptNGS) were performed simultaneously in 13 paired specimens, including #001 blood, #002 sputum, #003 sputum, #004 BALF, #005 sputum, #006 BALF, #007 BALF, #008 sputum, #009 sputum, #010 blood, #011 BALF, #012 sputum, and #013 sputum ( Supplementary Table S6 ). We performed the multiplex PCR-based targeted gene sequencing technology to identify targeted pathogens and compared the results of ptNGS and mNGS. All of the relevant pathogens explored in sequencing are listed in Supplementary Table S5 . ptNGS results were consistent with mNGS in the detection of abundant bacteria, fungi, and mycoplasma in specimens ( Supplementary Table S6 ).

For bacterial detection, both ptNGS and mNGS yielded negative results in 3 out of the 13 patients (#001, #010, #013). In patients with positive results, 6 out of the 10 patients were detected with similar abundant species by both methods (#002, #003, #004, #006, #008, #011) ( Supplementary Table S6 ).

For fungal detection, ptNGS and mNGS yielded negative results in 7 out of the 13 patients (#001, #004, #005, #008, #009, #011, #012). In patient #007, both methods showed very low sequence. Pneumocystis jirovecii and C. albicans were detected by ptNGS and mNGS analysis in patient #002. However, C. parapsilosis was only found to be positive by mNGS in patient #003 ( Supplementary Table S6 ).

For viral detection, ptNGS and mNGS yielded negative results in 3 out of the 13 patients (#004, #007, #009). Both methods only detected very low viral sequences in patients #002 and #006. Both analysis showed low abundance of CMV in patient #001 and high abundance of CMV in patient #010. However, the two methods showed discrepant results in two patients (#003, #008). In patient #003 (sputum), mNGS presented high sequences of EBV, while ptNGS showed low sequences. In patient #008 (sputum), mNGS exhibited low sequences of Torque teno virus and CMV, while ptNGS demonstrated opposite results ( Supplementary Table S6 ).

For mycoplasma detection, mNGS and ptNGS generated negative results in 10 out of 13 patients. Patients #004 (BALF) and #006 (BALF) showed high abundance of Mycoplasma hominis by both methods. In patient #007 (BALF), mNGS presented low sequences of EBV, while ptNGS showed high sequences ( Supplementary Table S6 ).