The study used CSF samples collected from patients with suspected CNS infection enrolled in a hospital-based surveillance program conducted in Vietnam during December 2012–October 2016 (5). The study was conducted as part of the Vietnam Initiative on Zoonotic Infections (VIZIONS) project (5), and patient recruitment was carried out at 7 provincial hospitals across Vietnam. After collection, as per the study protocol, all CSF samples were tested for a range of pathogens by using the diagnostic work-up of the clinical study (Appendix Table 1). The remaining volume of the CSF samples were stored at −80°C for further testing.

We focused our metagenomic analysis on patients of unknown origin from 4 provincial hospitals in central (Hue and KhanhHoa), highland (DakLak), and southern (DongThap) Vietnam (Figure 1), representing 3 distinct geographic areas in Vietnam. To increase the chance of detecting a virus in the CSF samples, we only selected patients with CSF leukocyte counts >5 cells/mm³ and an illness duration <5 days.

mNGS assay was carried out as previously described (8). Before viral nucleic acid (NA) isolation, 100 μL of each CSF sample was treated with Turbo DNase (Ambion, Life Technology, ThermoFisher, https://www.thermofisher.com) and RNase I enzyme (Ambion). Then viral NA was isolated using a QIAamp viral RNA kit (QIAGEN GmbH, https://www.qiagen.com), and recovered in 50 μL of elution buffer provided with the extraction kit. Double-stranded DNA was synthesized from the isolated viral NA by using a set of 96 nonribosomal primers (FR26RV–Endoh primers) and then was randomly amplified by using the FR20RV primer (5′-GCCGGAGCTCTGCAGATATC-3′). Finally, the amplified product was subjected to a library preparation step by using Nextera XT sample preparation kit (Illumina, https://www.illumina.com), following the manufacturer’s instructions, and sequenced by using a MiSeq reagent kit, version 3 (600 cycles) (Illumina) in a MiSeq platform (Illumina).

Potential viral reads were identified by using an in-house viral metagenomic pipeline running on a 36-node Linux cluster as described previously (9). In brief, after duplicate reads and reads belonging to human or bacterial genomes were filtered out, the remaining reads were assembled de novo. The resulting contigs and singlet reads were then aligned against a customized viral proteome database by using an approach based on BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi). Next, the candidate viral reads were aligned against a nonredundant nonvirus protein database to remove any false-positive reads (i.e., reads with expected values higher than those against viral protein databases). Any virus-like sequence with an expected value <0.00001 was considered a significant hit. Finally, a reference-based mapping approach (Genious 8.1.5; Biomatters, https://www.geneious.com) was used to assess the levels of identity and genome coverage of the corresponding viruses.

PCR assays were conducted to confirm mNGS hits for each specific virus identified from the viral metagenomic pipeline. Depending on availability of CSF, the PCR confirmations were performed either on leftover NA or newly extracted NA. A viral mNGS result was considered positive only if it was subsequently confirmed by PCR analysis of the original NA samples. The nucleotide sequences of primers and probes used for PCR confirmatory testing are shown in Appendix Table 2 (8).

For enterovirus serotype determination based on the obtained sequences generated by viral mNGS, we used a publicly available genotyping tool (10). To determine the relationship between enterovirus strains we sequenced and global strains, we first performed pairwise alignment by using the ClustalW tool in Geneious 8.1.5, and then reconstructed a maximum-likelihood phylogenetic tree by using IQ Tree 1.4.3 (11). A similar phylogenetic approach was used for other viruses. The generated sequences of this study were submitted to GenBank (accession no. PRJNA561465).

The study was approved by the corresponding institutional review broad of local hospitals in Vietnam, where the patients were enrolled, and the Oxford Tropical Ethics Committee. Informed consent was obtained from each study participant or a legal guardian.

From the clinical study described previously, a total of 841 patients were enrolled from Hue, Khanh Hoa, Dak Lak, or Dong Thap provincial hospitals. Of these, 609/841 (72%) patients had no etiology identified. The etiologic profiles of the patients in whom a pathogen was detected will be reported separately. Of the patients in whom a pathogen was not identified, 204 met our selection criteria, and their CSF samples were included for viral mNGS analysis (Figure 1).

The baseline characteristics and outcome of the 204 study patients are described in Table. Male patients were predominant. A substantial proportion of the patients were seriously ill; fatal outcome was recorded in 22 (11%), whereas incomplete recovery was recorded in 17% (n = 35) and deterioration (reflected by being transferred to other hospitals) in 16.5% (n = 34).

A total of 204 CSF samples were subjected to 3 NGS runs, and 108 million reads were obtained (median number of reads per sample 445,412 [range 430–908,890]). Of these, viral reads accounted for 0.64% (n = 692,731; median number of reads per sample 2,001 [range 4–268,933]). Excluding common contaminants and commensal viruses such as torque teno virus, which are not reported in this article, sequences related to a total of 8 distinct viral species were identified in 107/204 (52.4%) patients. These viruses are either known to be infectious to humans (e.g., enteroviruses, rotavirus, molluscum contagiosum virus [MCV], human papillomavirus, HIV, and hepatitis B virus [HBV]) or are without evidence of human infections besides previous detection in sterile human samples (e.g., cyclovirus-VN and gemycircularvirus) (Figure 2).

After virus-specific PCR confirmatory testing, the proportion of patients in whom a virus was found by mNGS was reduced from 53% (108/204) to 14.7% (30/204). Accordingly, the number of virus species was reduced from 8 to 5 (Figure 2); enteroviruses were the most common virus detected, accounting for 11.3% (23/204) of the included patients, followed by HBV (n = 3), HIV (n = 2), gemycircularvirus, and MCV (1 each) (Figure 2). Because of the focus of our study and the unavailability of the PCR assays, confirmatory testing for human papillomavirus was not performed.

All 23 enterovirus-infected patients were admitted to hospitals from the central or highland areas (Table), and none were from Dong Thap Province. Male patients were slightly predominant, accounting for 56%. Notably, the enterovirus-infected patients were younger than those who were mNGS-negative (Table). At discharge, incomplete recovery or transfer to other hospitals because of disease deterioration were recorded in 21.7% (Table).

Enterovirus cases were not detected during January 2015–December 2016. During 2013 and 2014, two main peaks were observed during March–July and September–December (Figure 3, panel A); cases from Dak Lak and Khanh Hoa contributed to the first peak (Figure 3, panels B and C), and cases from Khanh Hoa and Hue contributed to the second (Figure 3, panels C and D). The general baseline characteristics of patients with HBV, gemycircularvirus, and MCV are shown in Appendix Table 3.

mNGS generated sufficient sequence information for an enterovirus serotyping assessment in 11/23 cases. Subsequently, results of serotyping analysis based on the NGS sequences showed that echovirus 30 (E30) was the most common serotype detected (n = 9, 39% of enteroviruses), followed by enterovirus A71 and enterovirus B80 (1 each, 4.3%). Phylogenetically, the 9 E30 strains sequenced in our study belonged to 2 distinct genogroups, V and VIIb, and showed close relationship with E30 strains circulating in Russia and elsewhere in Asia, including China (Figure 4).

In additional to enterovirus sequences, a gemycircularvirus genome was obtained from a 12-year-old boy. Phylogenetic analysis revealed that this gemycircularvirus strain was closely related to a gemycircularvirus species previously found in CSF sample from a patient with a CNS infection of unknown origin in Sri Lanka (12); the level of amino acid identities between the 2 strains were 98.79% for replication-coding sequences and 99.3% for capsid protein–coding sequences.