This prospective multicenter study used clinical information to diagnose infectious encephalitis and/or meningitis in six teaching hospitals in Beijing, China. The inclusion criteria were as follows: high-level clinical suspicion of CNS infectious diseases; and lumbar puncture in uninfected patients. The exclusion criteria were as follows: refusal to undergo lumbar puncture; any contraindication for such puncture; and a diagnosis of autoimmune encephalitis. Eligible patients were divided into five groups according to their final diagnoses: viral encephalitis and/or meningitis, TBM, bacterial meningitis, fungal meningitis, and CNS non-infection. We included patients with definite or probable diagnoses of CNS viral infections (Jeffery et al., 1997), TBM (Marais et al., 2010) and bacterial meningitis (Mudaliar et al., 2006). Fungal meningitis was confirmed via conventional microbiological testing. Cryptococcal meningitis was confirmed via India ink staining or fungal culture, as well as histopathological evidence of cerebral aspergillosis. All patients had undergone at least 3 months of follow-up. Figure 1 shows the flow chart of study enrolment.

CSF specimens were collected in accordance with standard aseptic procedures, snap frozen, stored at −20°C, and subjected to mNGS within 24 h. The DNA of blood samples collected from healthy volunteers was fragmented and mixed with water in a certain proportions (negative controls). Glass beads were added to CSF samples, followed by vigorous agitation; DNA extraction; and DNA library construction. Quality-controlled libraries were sequenced on a BGISEQ-500/50 platform (BGI-Tianjin, Tianjin, China); an average of 20 million reads was obtained for each sample. After removal of human sequences, the remaining data were aligned to bacterial, virus, fungal and protozoan databases, as described in detail elsewhere (Xing et al., 2018, 2019).

A final sequencing list of suspected pathogenic microorganisms was obtained after removal of common background microorganisms and those that had appeared in >50% of samples during the past 3 months, compared to the negative controls. Next, the sequencing data list was analyzed in terms of species-specific read number (SSRN), genome coverage (%), and depth. Instead of SSRN, genus-specific read number was used for the Mycobacterium tuberculosis complex, because M. tuberculosis complex members exhibit >99.99% genomic sequence similarity (Supply and Brosch, 2017). Probable causative microorganisms were identified with reference to the literature and when the pathogenicity was consistent with the clinical manifestations; they were verified using traditional methods (serological tests, smears, cultures, and/or polymerase chain reaction assay). Herpes simplex virus (HSV), varicella zoster virus, Epstein–Barr virus, and cytomegalovirus were considered to be mNGS false-positives in patients with non-viral infections. Common pathogens of bacterial meningitis (Streptococcus pneumoniae, Staphylococcus aureus, Haemophilus influenzae, Klebsiella pneumoniae, and Neisseria meningitidis) were considered to be mNGS false-positives in patients with non-bacterial meningitis. Aspergillus fumigatus, Aspergillus flavus, Aspergillus niger, and Aspergillus terreus (Supply and Brosch, 2017) were considered to be mNGS false-positives in patients who did not have CNS aspergillosis.

Continuous data were considered to be nonparametric. Quantitative variables are expressed as medians (ranges) and qualitative variables are expressed as percentages. We drew receiver operating characteristic curves to compare diagnostic tests; test accuracies were represented by the area under the curve (AUC), such that a larger area implied a better test. Data processing was performed using SPSS Statistics version 21.0 (IBM Corp., Armonk, NY, USA).

In total, 213 patients treated in six hospitals were finally enrolled, including 75 with presumed viral encephalitis and/or meningitis, 44 with presumed TBM, 43 with presumed bacterial meningitis, 18 with fungal meningitis, and 33 with presumed non-infectious CNS diseases (Figure 1). All patients with infectious encephalitis and/or meningitis were screened by a single investigator (X.-W. X.) to determine diagnostic classification. Most patients (178/213, 83.57%) were treated in the First Medical Center of the PLA General Hospital. Non-infectious CNS conditions included meningeal carcinoma (6/33, 18%), encephalopathy associated with connective tissue disease (6/33, 18.18%), metabolic encephalopathy (5/33, 15.15%), and other non-infectious diseases (16/33, 48.48%). The clinical data are summarized in Table 1.

Seventy-five patients with viral encephalitis and/or meningitis (54 definite and 21 probable) were enrolled. An mNGS result was considered positive if the CSF sample exhibited SSRNs ≥1, 2, 3, 5, or 10; the diagnostic probability consistencies for viral encephalitis and/or meningitis were 32, 30.7, 29.3, 26.7, and 22.7%, respectively. The negative consistency rates were 84.8, 89.1, 89.9, 89.9, and 92.8%, respectively; the total consistency rates were 66.2, 68.5, 68.5, 67.6, and 68.1%, respectively. The AUCs for the five positive mNGS criteria were 0.584, 0.599, 0.596, 0.583, and 0.577, respectively. When an SSRN ≥2 was considered positive, the corresponding AUC (0.599) was relatively larger than when other SSRNs were considered positive. For the 54 patients with definite viral encephalitis and/or meningitis, if an SSRN ≥2 was considered positive, the AUC was largest (0.659, 95% confidence interval [CI] = 0.566–0.751; Figure 2A); the positive, negative, and total consistency rates were 42.6, 89.1, and 76.0%, respectively (Table 2, Table S1).

Forty-four patients were diagnosed with TBM (six definite and 38 probable). An mNGS result was considered positive if the CSF samples exhibited genus-specific read numbers ≥1, 2, 3, 5, and 10; the associated positive consistency rates were 27.3, 20.5, 18.2, 13.6, and 6.8%, respectively. The negative consistency rates were 96.4, 97.6, 98.2, 99.4, and 100%, respectively; the total consistency rates were 82.2, 81.7, 81.7, 81.2, and 80.8%, respectively. When a genus-specific read number ≥1 was considered positive, the AUC (0.619, 95% CI = 0.516–0.721) was largest. Among the six patients with definite TBM (including five who were positive on Xpert MTB/RIF and one who was positive on acid-fast staining of nerve tissue), the sensitivity and specificity of mNGS were 66.67% (4/6) and 96.45% (163/169), respectively. Among the 44 patients with presumed TBM (six definite and 39 probable), the CSF Xpert MTB/RIF positivity rate was 16.13% (5/31) and the mNGS positivity rate was 27.27% (12/44) (Table S2); the combined mNGS and Xpert MTB/RIF positivity rate was 29.55% (13/44) (Figure 2B1). Unfortunately, acid-fast bacillus staining and M. tuberculosis culture results were negative for all CSF samples.

Forty-three patients were diagnosed with bacterial meningitis (15 definite and 28 probable). An mNGS result was considered positive if the SSRNs were ≥2, 3, 5, 10, and 15; the positive consistency rates were 62.8, 62.8, 55.8, 55.8, and 51.2%, respectively. The negative consistency rates were 82.9, 87.6, 95.9, 95.9, and 95.9%, respectively; total consistency rates were 78.9, 82.6, 87.8, 87.8, and 86.9%, respectively. The AUCs for the five positive mNGS criteria were 0.729, 0.752, 0.758, 0.758, and 0.735, respectively. When SSRNs ≥5 or 10 were considered positive, the AUC (0.758, 95% CI = 0.663–0.854) was largest. The sensitivity, specificity, positive predictive value, and negative predictive value of mNGS in the diagnosis of 15 patients with definite bacterial meningitis were 73.3, 95.9, 61.1, and 97.6%, respectively (Table S3) and the AUC was largest (0.846, 95% CI = 0.711–0.981). In brief, among the 43 patients with presumed bacterial meningitis (15 definite and 28 probable), mNGS identified a bacterial pathogen in 24 (55.8%, 24/43); conversely, the CSF combined Gram stain/culture-positive rate was only 32.56% (14/43). Combination of the two methods increased the positive rate to 62.79% (27/43) (Figure 2B2). In the 24 patients with positive mNGS results (SSRNs ≥5 or 10), the top three pathogens were S. pneumoniae (41.7%, 10/24), K. pneumoniae (12.5%, 3/24), and Listeria monocytogenes (8.3%, 2/24). Of note, among the patients with non-bacterial meningitis, 4.1% (7/170) were false-positive on mNGS for S. aureus (57.1%, 4/7), S. pneumoniae (28.6%, 2/7), and H. influenzae (14.3%, 1/7).

Among the 13 patients with definite cryptococcal meningitis, the first-time-positive rates of CSF India ink staining and fungal culture were 0 and 38.4%, respectively. After multiple stainings/cultures, the positive rate of the above two methods were 84.62%. Cryptococcal antigen was detected in all (12/12) CSF samples. When an SSRN ≥2 was considered mNGS-positive, the sensitivity, specificity, positive predictive value, and negative predictive value were 76.92% (10/13), 99.52% (1/207), 90.91% (10/11), and 98.56% (206/209), respectively (Table S4). In addition, among the 13 patients, five were co-infected with Cryptococcus neoformans sensu lato and Cryptococcus gattii sensu lato, as has been described in detail elsewhere (Xing et al., 2019). Among the five patients with confirmed cerebral aspergillosis, when an SSRN ≥2 was considered mNGS-positive, the sensitivity, specificity, positive predictive value, and negative predictive value were 80% (4/5), 79.3% (165/208), 8.5% (4/47), and 99.40% (165/166), respectively (Table S5).

The mNGS-positive rate for diagnosis of the four types of definite or probable infectious encephalitis and meningitis, including viral encephalitis and/or meningitis, TBM, bacterial meningitis, and fungal meningitis, was 40.6% (73/180) (Table 3). The positivity rate for the four types of definite CNS infectious diseases was 57.0% (52/93).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.