Clinical metagenomics assessments improve diagnosis and outcomes in community-acquired pneumonia

We recruited 159 patients admitted into the Respiratory ICU of the People’s Liberation Army General Hospital in Beijing, China from December 2018 to November 2019 with a diagnosis of CAP according to Chinese guidelines [2, 27, 28]. Patients who met the following criteria (1 + 2) and at least one of the criteria (3)–(7) were enrolled this prospective study and randomly assigned into either the control or mNGS groups with informed consents signed by patients or surrogates: (1) Admitted at our ICU and considered for pneumonia acquired outside of the hospital setting; (2) A new or progressive pulmonary infiltration with/without pleural effusion on a chest radiograph; (3) New or increased cough with or without sputum production; (4) Purulent sputum or a change in sputum characteristics; (5) Fever; (6) Signs of lung consolidation or moist rales; (7) Peripheral white blood cell (WBC) count ≥10 × 10⁹/L or ≤ 4 × 10⁹/L. Demographic characteristics of the cohort are provided in Table 1.

Sample collection was reviewed and approved by the Chinese People’s Liberation Army General Hospital Ethics Committee Review Board. Informed consents were signed by patients or surrogates. Patients were classified using APACHE II criteria on the first day of ICU admission [29]. The cohort was random divided into two groups. In the control group (100 cases), only standard non-NGS methods (culture or smear, acid-fast staining, T-spot and X-pert MTB/RIF for M. tuberculosis) were employed for pathogen detection. In the mNGS group (59 cases), samples underwent traditional clinical microbiological assays and mNGS testing in parallel.

Results of mNGS were reviewed along with other clinical evidence by the physicians, and changes in the treatment plans were made when warranted. We evaluated the impact of mNGS-based testing on clinical management and outcomes by categorizing each patient’s clinical outcome into three groups as of his/her last day in the Intensive Care Unit (ICU): Improved, Resolved, or Mortality. In the improved group, patients had resolution of abnormal vital signs, including heart rate, respiratory rate, blood pressure, oxygen saturation, and temperature; ability to eat; and normal cognition [28]. In the Resolved group, patients showed all the above signs of clinical stability and no lesions on chest CT. Patients who died during hospitalization comprised the Mortality category. In the mNGS group who had beneficial clinical outcomes, levels of alanine aminotransferase, aspartate aminotransferase, albumin, blood urea nitrogen, and creatinine were examined to evaluate the impact on patients’ liver and kidney function.

Within 2 days of enrollment, samples of blood, sputum, or bronchoalveolar lavage (BALF) were collected and transported to the laboratory following standard procedures (https://​emergency.​cdc.​gov). Cerebrospinal fluid (n = 3), pleural effusion (n = 2), tissue (n = 1), and urine were also collected in a small number of cases. About 3 mL of samples were collected from the patients and stored at room temperature (for blood) or at − 80 °C (for other specimen types) prior to testing. Blood were stored in EDTA tubes, from which plasma was seperated by centrifuging at 1600 g for 10 min at 4 °C. Trypsin-liquefied sputum and BALF were centrifuged at 8000 g for 5 min. Pellets were resuspended and vortexed at 3000 rpm for 30 min in lysis buffer with the aid of glass beads to break the cell walls. DNA extraction was performed with 300 ul of specimens as described above (plasma, liquefied sputum or BALF). One hundred ng of DNA, as measured by a Qubit Fluorometer (Invitrogen, Carlsbad, CA), were then subjected to library preparation with a transposase-based method. Pooled libraries were sequenced on an Nextseq 550 sequencing system (Illumina, San Diego, CA) using a 75 bp, single-end sequencing kit.

Raw sequencing data were first subjected to a quality control process including removal ow-quality, low-complexity, short reads (< 35 bp) and adapter trimming before further analyses. At least 10 M reads were generated for each sample. Prior to microbial classification, reads derived from the human genome were depleted by aligning to the human reference genome (hg38) using the Burrows-Wheeler Alignment (BWA) tool [30]. Microbial classification were performed by mapping the remaining sequencing reads to a reference microbial database comprising genomes of archaea, bacteria, fungi, protozoa, viruses and parasites, which was curated from the NCBI genome databases.

As shown in Fig. 1a, this prospective study enrolled patients who met our inclusion criteria and randomly assigned them into either the control or mNGS groups with informed consents. The average Acute Physiologic Assessment and Chronic Health Evaluation II (APACHE II) scores were 12.6 ± 7.9 and 11.1 ± 7.0 in the control and mNGS groups, respectively (Fig. 1b). The lack of significant differences (P = 0.24) in APACHE scores between the two groups suggests balanced enrollment with regard to disease severity.

In the control group, a total of 109 samples were collected from 100 patients, including BALF (53), sputum (46), tissue (2), blood (8). For the mNGS group, a total of 104 samples were collected from 59 patients, including bronchoalveolar lavage fluid (BALF) (47), blood (45), sputum (5), cerebrospinal fluid (3), pleural effusion (2), tissue (1) and urine (1) (Fig. 1c). Generally, BALF and blood samples were tested in pairs in the mNGS group.

In this study, an NGS test was only considered positive when potential pathogen(s) were identified in the specimens. Tests that detected no microbes or only bacteria that were clinically considered commensal (for instance, Propionibacteria spp., Veillonella spp., Rothia spp., and Neisseria flavescens) were defined as negative (Supplementary Table S1). A total of 284 pathogens were detected in the overall cohort. These included 105 pathogens (78 bacteria and 27 fungi) in the control group and 179 pathogens in the mNGS group (113 bacteria, 32 fungi and 34 viruses). In line with these findings, 79.7%(47/59) of those in the mNGS group were positive compared to 37.0% (37/100) in the controls (P < 0.001, Fig. 2a).

mNGS results agreed with clinical diagnosis in 40 (60.8%) infection cases and 6 (10.2%) non-infection cases (Fig. 2b). NGS-based assays identified significantly more pathogens (179 vs 39) in significantly more cases (29 out of 59, 49.2% vs 47 out of 59, 79.7%, P < 0.01) (Fig. 2c) compared to conventional techniques. A total of 34 bacterial and 5 fungal pathogens were reported by conventional testing, whereas 113 bacteria, 34 DNA viruses and 32 fungi were found by metagenomics (Fig. 2d). Discordant positive results included unculturable bacteria, viruses, and eukaryotic pathogens. The most common organisms detected by mNGS were A. baumannii (23.7%), P. jirovecii (23.7%), and cytomegalovirus (18.6%) (Fig. 2e). Our mNGS assay confirmed the diagnosis in all four cases of tuberculosis (Table 2). Multiple pathogens were identified in a substantial portion of the cases (30/59, 50.8%); in 3 cases, co-infections with bacteria, fungi, and viruses were detected (Fig. 2f, g).

In 37 of 59 mNGS patients, a pair of plasma and BALF samples were collected for testing. Among those, the same pathogens, including bacteria, viruses and fungi, were detected in both samples in 11 cases (29.7%). There were 3 additional cases (0.8%) in which both BALF and blood samples were negative for pathogen identification (Fig. 2h). In most of the remaining samples, BALF samples yielded better pathogen detection results than blood samples (Table 3). As indicated by the number of sequencing reads detected, the abundance of bacteria and fungi in cell-free DNA was generally lower than in the corresponding respiratory specimen (Table 3).

Compared to the control group, the mNGS group has a considerably lower mortality rate (13.6% vs 26.0%, or 8/59 vs 26/100; Fig. 3a) and a significantly higher rate of complete symptom resolution (55.9% vs 7.0%, or 33/59 vs 7/100, P < 0.001; Fig. 3b). The duration of mechanical ventilation was also significantly reduced in the mNGS group (average of 7.4 days versus 17.3 days in the control group) (P < 0.05, Fig. 3c). No significant differences were found in the length of stay or medical costs in the ICU (data not shown).

In the mNGS group, the results of metagenomics testing led to a) change in clinical management in 11 patients (18.6%) and b) confirmation of ongoing treatment in 24 patients (40.7%). In addition, one patient was transferred to a specialized hospital after confirmation of tuberculosis by mNGS. In the 11 cases with treatment changes in the control group, 8 showed complete resolution of symptoms (72.7% vs 7.0%, P < 0.001) and a decreased mortality rate (2 out of 11, 18.2% vs 26.0%, Fig. 3a, b, Supplementary Table S2). None of the 11 cases required mechanical ventilation.

To assess whether mNGS affected ongoing clinical management, we evaluated clinical outcomes in the 11 patients with changed treatment and the 24 patients with confirmed treatment. Consistently, compared with the control group, outcomes improved, manifested by a significantly higher rate of disease resolution (22/35, 62.9% vs 7/100, 7.0%, P < 0.001) and a reduced mortality rate during hospitalization (3/35 8.6%, vs 26/100, 26.0%, P = 0.055) (Fig. 3a, b). The duration of mechanical ventilation was also significantly shorter in 2 of the 36 cases, with an average of 8.5 days (Fig. 3c). Consistently, multivariate logistic regression analyses with factors of sex, age and concurrent conditions included also showed significant correlations between mNGS testing and improved sympton resolution (P<.001, control vs. mNGS groups; P<0.001, control cases vs. cases with treatment changed/confirmed based on mNGS results; Supplementary Table S3).

The levels of alanine aminotransferase, aspartate aminotransferase, albumin, blood urea nitrogen, and creatinine were examined in the 51 patients from the mNGS group who had beneficial clinical outcomes. As shown in Fig. 3d, fewer patients in the mNGS group had abnormal levels of these clinical indicators at discharge (compared to admission).