Clinical characteristics of subjects with febrile illness

From October to December 2013, 94 children admitted to Tororo District Hospital were enrolled (Table 1). Their mean age was 16.4 (IQR: 8.0–21.0) months, and 66 (70.2%) were female. Chief symptoms reported in addition to fever were cough (88.3%), vomiting (56.4%), diarrhea (47.9%), and convulsions (27.7%). Top admitting diagnoses were respiratory tract infection (57.4%), gastroenteritis/diarrhea (29.8%), and septicemia (11.7%) (Table A in S1 Data). Of the 90 blood samples that were collected, thick blood smears identified P. falciparum in 12 samples that underwent mNGS analysis (Table B in S1 Data).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Overview of the patients enrolled in the study.

The other category includes: unknown (10), urinary infection (1), meningitis (2), hepatitis (1), and fever (1). Information on gender was missing for three and on age for two children.

https://doi.org/10.1371/journal.pone.0218318.t001

Metagenomic sequencing findings

mNGS was performed on RNA extracted from 90 serum, 90 NP swab, and 10 stool samples following library preparation. A mean of 11.5 million (IQR 6.4–15.2 million) paired-end reads were obtained per sample; sequencing statistics are in S2 Data. For one batch of serum samples, only a single read, rather than paired-end reads, was produced. Bioinformatic analysis was carried out using the IDseq pipeline, a cloud-based, open-source platform designed for detection of microbes from metagenomic data (https://github.com/chanzuckerberg/idseq-dag) that incorporates several features of previously developed pathogen detection pipelines [11–17].

In this section, we discuss the mNGS results identified in a given sample type. Detailed findings on microbes identified in every patient, along with the total reads per million (rpM) are reported in Table B and Table C in S1 Data, respectively.

mNGS of serum

At least one microbial species was detected in 60 (66.7%) of the serum samples; more than one microbe was detected in 11 (12.2%) samples (Fig 1A). No microbial species were identified in the serum of 30 (33.3%) individuals. The most commonly identified microbes were Plasmodium falciparum (46, 51.1%) and parvovirus B19 (4, 4.4%). P. falciparum was detected in 10/12 samples from patients reported as smear-positive. mNGS detected Plasmodium spp. in 37 additional samples that were smear negative (36 P. falciparum, 1 P. malariae). Viruses detected in serum included human immunodeficiency 1 virus (HIV-1), hepatitis A virus, rotavirus A, human herpesvirus (HHV) type 6, HHV type 4, HHV type 7, human rhinovirus (HRV)-C, HRV-A, enteroviruses (enterovirus A71, Coxsackievirus B2 and echovirus E30), human parechovirus 2, hepatitis B virus, a novel orthobunyavirus (described in greater detail below), human cardiovirus (Saffold virus), mamastrovirus 1 and Norwalk virus (Fig 1A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1.

(A) Microbial landscape found in serum samples in Ugandan children. Each column represents a febrile child. Results for GB virus C and torque teno virus, which are of uncertain clinical significance, are not included. (B) Microbial landscape found in nasopharyngeal (NP) swab samples in Ugandan children. Note that bacterial species were not considered in Fig 1B. Each column represents a febrile child, and the color bars represent the total reads per million (rpM) of a particular microbe present in the sample. Results for GB virus C and torque teno virus, which are of uncertain clinical significance, are not included.

https://doi.org/10.1371/journal.pone.0218318.g001

Multiple viruses were detected from serum in patients with Plasmodium infections (10 of 46 (21.7%) samples; S1A Table). Three of the four identified parvovirus B19 cases were associated with P. falciparum. Additionally, GB virus C and torque teno virus (TTV), which are of unknown clinical significance [18], [19], were identified in the serum of 25 (27.8%) and 37 (41.1%) children, respectively. There have been reports on associations between immunosuppression and TTV abundance [20, 21]. Interestingly, a prior study has reported a higher abundance of TTV in children with fevers [22].

mNGS of NP swabs

90 NP swabs were collected and processed; 52 (57.7%) of these were from patients with admission diagnoses of pneumonia, respiratory tract infection, or bronchiolitis (Table 1). Chest imaging was not available to further assess these diagnoses. 72 NP samples (80%) contained at least 1 viral species (Fig 1B), with no microbes meeting our required cut-offs in 18 samples (20%). HRV-A and HRV-C were the most prevalent, followed by respiratory syncytial virus (RSV), cytomegalovirus (HHV-5), influenza B, and coronavirus OC43. Other respiratory viruses identified included influenza A (H1N1), HRV-B, Human mastadenovirus B (type 7), three human parainfluenza virus types (type 1, 3 and 4), metapneumovirus, coronavirus NL63, avian coronavirus, coxsackievirus A2, coxsackievirus B2, polyomaviruses (KI), HHV-6 and HHV-7. Other viruses identified that are not typically considered respiratory pathogens included hepatitis A virus, hepatitis B virus, parvovirus B19, mamastrovirus 1, Bwamba orthobunyavirus, betapapillomavirus 1, and rotavirus. Additionally, TTV was found in 49 (54.4%) NP swab samples, including one sample with both gemykrogvirus and TTV. For 26 (28.8%) patients, mNGS identified respiratory viral co-infections, most commonly with HRV-C (n = 11) and HRV-A (n = 5) (S1B Table). The same microbial species was identified in the NP swab and serum samples in six patients (on independent sequencing runs), one each with HRV-A, HRV-C, hepatitis A virus, hepatitis B virus, rotavirus A, and parvovirus B19.

Bacteria identified in NP samples included four dominant genera, which together comprised 79% of all microbial reads—Moraxella (39.4%), Haemophilus (16.7%), Streptococcus (16.2%), and Corynebacterium (6.6%). Given that diversity loss in the microbial flora in lower respiratory tract samples correlates with pneumonia [23, 24], we compared the Simpson Diversity Index (SDI) in patients with and without clinical diagnoses of respiratory tract infection. We found no significant difference in the SDI of upper airway samples between patients with (mean SDI = 0.51, IQR 0.37–0.65) or without (mean SDI = 0.51, IQR = 0.42–0.65; p = 0.86) diagnoses of respiratory infection (S1 Fig). [25–27].

mNGS of stool

Among the 10 stool samples collected and sequenced, potential non-bacterial pathogens were detected in 9/10 samples. The three most common microbes identified were rotavirus A (50%), Cryptosporidium (40%), and human parechovirus (40%). Our sequencing did not provide enough information to type the identified human parechovirus. Seven children had additional microbes identified: two rotavirus A and human parechovirus, and one each rotavirus A and Cryptosporidium, rotavirus A and enterovirus, Cryptosporidium and human parechovirus, Cryptosporidium and Norwalk virus, and Giardia and human parechovirus. Five samples also had Blastocystis hominis, a protozoan of uncertain pathogenicity.

Genomic characterization of viruses

Representative genomes for all viruses identified in this study were assembled and deposited in GenBank (Accession numbers: MH685676-MH685701, MH685703- MH685719, MH684286-MH684293, MH684298-MH684334). In this section, we describe in detail a novel orthobunyavirus identified in the serum of one individual, two novel HRV-C species identified in two individuals and diversity of influenza B viruses assembled in the nasopharynx from five individuals. We studied genomic diversity of influenza B in detail, given its potential implication to inform vaccine design. We did not include influenza A in this analysis, as we were able to assemble the complete genome sequence only from one individual.

Orthobunyaviruses

Serum from one patient admitted with a clinical diagnosis of malaria and pneumonia contained a novel orthobunyavirus in addition to P. falciparum. Assembly of a near-complete genome and comparison with existing orthobunyavirus genomes indicated that this sequence includes 97.5%, 100% and 91% of the L, M, and S coding regions, respectively (Fig 2). Average read coverage across the segments was 86-fold. Phylogenetic comparison showed that the novel virus was significantly divergent from known orthobunyaviruses, sharing 44.9–55.1% amino acid identity with the closest known relatives, Calchaqui virus, Kaeng Khoi virus, and Anopheles A virus (Figs 2 and S2, S3 and S4). The virus was isolated from a patient from Nyangole village, Tororo District—hence, we propose the name “Nyangole virus”, consistent with nomenclature guidelines for the family Bunyaviridae.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Characterization of the novel orthobunyavirus identified in a febrile child.

(A) Schematic representation of the large (L) or RNA dependent RNA polymerase, medium (M) or polyprotein of Gn, NSm and Gc proteins and small (S) segment encoding the nucleocapsid (N) protein of Nyangole virus and percentage identity with the most closely related virus. Phylogenetic tree of all complete orthobunyavirus genome sequences along with Nyangole virus are represented in (B) for the RNA dependent RNA polymerase and (C) for the glycoprotein.

https://doi.org/10.1371/journal.pone.0218318.g002

In addition, a second orthobunyavirus, Bwamba virus, was identified in the NP swab sample from a patient admitted with rash, sepsis, and diarrhea. Insufficient sample and sequencing reads precluded genome assembly of this virus.

Human rhinoviruses

Within the species rhinovirus, we assembled de novo a total of 13 HRV-C (mean coverage: 39-fold) and 13 HRV-A (mean coverage: 268-fold) genomes (> 500 bp). Of these, 10 HRV-A and nine HRV-C genomes had complete coverage of the VP1 region, which is used to define enterovirus types [28]. Unique HRV types are defined by <73% similarity in the VP1 gene. As such, we found three HRV-A and eight HRV-C types in this cohort. One individual harbored two distinct HRV-A types (genome pairwise identity = 75.3%, VP1 pairwise identity = 67.1%). Additionally, we assembled two novel HRV-C species from two patients admitted with gastroenteritis (patient ID: EOFI-014) and with pneumonia, malaria and diarrhea (patient ID: EOFI-133), that shared 70.1% and 70.7% nucleotide sequence identity at VP1 compared to the closest known HRV-C (Accession JQ245968 and KF688606, respectively) (Fig 3). The Picornavirus Working Group has established that novel HRV-Cs should exhibit at least 13% nucleotide sequence divergence in the VP1 gene [29], qualifying these two as novel.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Phylogenetic tree of all complete HRV-C genomes from NCBI (with information on country of isolation) and HRV-C genomes assembled in this study.

https://doi.org/10.1371/journal.pone.0218318.g003

Influenza B virus

We assembled influenza B genome segments (>500bp, mean-coverage: 4-fold) from six of seven samples containing influenza B virus (one sample had insufficient sequencing reads). The viruses assembled were >99% similar to each other and >99% identical to the B/Massachusetts/02/2012-like virus included in the vaccine recommended by WHO for the 2013–2014 northern hemisphere and 2014 southern hemisphere influenza seasons (Accession numbers: KC891816.1, KC891879.1 and KC892119.1). For the Ugandan viruses, three of the four major epitopes (150 loop, 160 loop and 190 helix) and their surrounding regions were 100% identical; a Ser->Thr substitution was observed at amino acid 136 (in the 120 loop) compared to the B/Massachusetts/02/2012-like virus.

Enrollment of study subjects

We studied children admitted to Tororo District Hospital, Tororo, Uganda, with febrile illnesses. Potential subjects were identified by clinic staff, who notified study personnel, who subsequently evaluated the children for study eligibility. Inclusion criteria were: 1) age 2–60 months; 2) admission to Tororo District Hospital for acute illness; 3) documentation of axillary temperature >38.0°C on admission or within 24 hours of admission; and 4) provision of informed consent from the parent or guardian for study procedures. The only exclusion criterion was unwillingness or inability of parents/guardians to provide consent.

Sample collection

Serum and nasopharyngeal (NP) swab samples were collected from 90 children each; for four children, only one of the two sample types was successfully collected. Although 45 (47.9%) of the children had a presenting symptom of diarrhea, stool samples were available for only 10 due to logistical constraints. All samples that were collected were processed and included in the analysis.

Study specimens

NP swabs and serum were collected from each enrolled subject within 24 hours of hospital admission. Approximately 5 ml of serum was collected by phlebotomy, the sample was centrifuged at room temperature, and serum was then stored at -80°C. NP swab samples collected with FLOQSwabs swabs (COPAN) were placed into cryovials with Trizol (Invitrogen), and stored at -80°C within ~5 min of collection. For subjects with acute diarrhea (≥ three loose or watery stools in 24 hours), stool was collected into clean plastic containers and stored at -80°C within ~5 min of collection. Samples were stored at -80°C until shipment on dry ice to UCSF for sequencing.

Clinical data

Clinical information was obtained from interviews with parents or guardians, with specific data entered onto a standardized case record form that included admission diagnosis and physical examination as well as malaria blood smear results. For malaria diagnosis, thick Blood smears were Giemsa stained and evaluated by Tororo District Hospital laboratory personnel following routine standard-of-care practices. No efforts were made to improve on routine practice, so malaria smear readings represent routine standard-of-care rather than optimal quality controlled reads.

Metagenomic next-generation sequencing (mNGS)

After shipment to University of California, San Francisco, RNA was extracted from clinical samples as well as positive (HeLa cells) and negative (water) controls, and unbiased cDNA libraries were generated using previously described methods (see sections “Sequencing library preparation” and “Metagenomic Library Preparation”, respectively) [62, 5]. Barcoded samples were pooled, size selected (Blue Pippin), and run on an Illumina HiSeq2500 to obtain 135 base pair (bp) paired-end reads.

Bioinformatic analysis and pathogen identification

Microbial pathogens were identified from raw sequencing reads using the IDseq (v1.6) Portal (https://idseq.net), a cloud-based, open-source bioinformatics platform designed for detection of microbes from metagenomic data. IDseq is a scalable and cloud based implementation of a previously published pipeline (see Fig 1 in [11]). In brief, initial host read filtering is performed using Spliced Transcripts Alignment to a Reference (STAR) algorithm [17], followed by removal of duplicate/low quality and low complexity sequences [14], [16]. Next, reads are aligned once again to the host genome of interest using bowtie2 [12] to remove any remaining host reads. The non-human reads are then aligned to the NCBI nucleotide and protein database, using GSNAPL and RAPSearch, respectively [13], [15]. Additionally, reads that were identified as HHV-5 were assessed individually using BLAST, to verify specificity to this virus. All IDseq scripts and user instructions are available at https://github.com/chanzuckerberg/idseq-dag and the graphical user interface web application for sample upload is available at https://github.com/chanzuckerberg/idseq-web. To distinguish potential pathogens from ubiquitous environmental agents including laboratory reagent contaminants and skin commensal flora, a Z-score was calculated for both nucleic acid and protein alignments, for each genus relative to a background of non-templated (“water only”) controls in addition to a previously published set of uninfected clinical mNGS samples [11]. CSF samples acquired through lumbar puncture from uninfected controls are included in this background model as there is extensive representation of the skin microbiome given the need to puncture the skin. Only human papillomavirus was identified in the positive (HeLa cell) controls.

For this study we report species greater than 0 reads per million (rpM) and a Z-score > 0 (for both nucleic acid and protein alignments) detected in the serum, stool, and NP samples. We chose to report all species satisfying these criteria, rather than restricting to particular species, to offer an unbiased representation of microbes present in the sample. IDseq uses the CD-HIT-DUP tool to compress duplicate reads, hence final assignment of the rpM in a given sample represents coverage across the genome, rather than a single portion. Consistent with previous studies, low levels of “index bleed through” or “barcode hopping” (assignment of sequencing reads to the wrong barcode/index) was observed within the non-templated water control samples [63]. To account for barcode mis-assignment, when a microbe was found in more than one sample, it was reported only when present at levels at least four times the level of mis-assigned reads observed in the control samples. Given the extremely high levels of rotavirus found in stool samples, these samples were run in duplicate, and only microbes identified in both replicates and present at levels at least four times the number of reads mis-assigned in the control samples were reported. If the reads identified for a given microbe were not species-specific, we reported the corresponding genus. For NP and stool samples, because the nasopharynx and intestines are normally colonized with commensal bacteria [45–48], and because of a lack of healthy Ugandan NP and stool samples to serve as controls, only non-bacterial species were reported though we did analyze NP microbiome diversity (see below).

Genome assembly, annotation and phylogenetic analysis

To more comprehensively characterize the genomes of identified microbes, the paired-read iterative contig extension (PRICE) assembler [16] and the St. Petersburg genome assembler (SPAdes) [64] were used to de novo assemble short read sequences into larger contiguous sequences (contigs). Assembled contigs were queried against the National Center for Biotechnology Information (NCBI) nucleotide (nt) database using the basic local alignment search tool (BLAST) to identify the closest related microbes. GenBank annotation files from genome sequence records corresponding to the highest scoring alignments were used to identify potential features within the de novo assembled genomes. Geneious v10.3.2 was used to annotate newly assembled genomes. Reference genomes for multiple sequence alignments and phylogenetic analyses were downloaded from NCBI. Multiple sequence (nucleotide) alignments were generated using the default settings in MUSCLEv3.8.1551 [65], and ModelTest-NGv0.1.5 was used to identify the best-fitting evolutionary model. Using the best-fitting model of for evolution, we reconstructed a maximum-likelihood phylogeny using RAxML-ng v0.6.0 using default settings [66]. Annotation of protein domains in the novel orthobunyavirus was performed using the InterPro webserver [67] as well as direct alignment against previously known orthobunyaviruses. The TOPCONS webserver [68] was used for the identification of transmembrane regions and signal peptides, and the NetNglyc 1.0 Server (http://www.cbs.dtu.dk/services/NetNGlyc/) for the identification of glycosylation sites.

Evaluation of NP microbiome diversity

We applied SDI to evaluate alpha diversity of microbes identified in NP samples. For this analysis, patients were stratified into two categories based on clinical assignment: respiratory infections (admitting diagnosis of pneumonia, respiratory tract infection, or bronchiolitis; n = 52) and all other syndromes (n = 39); cases with unknown admitting diagnosis were excluded. SDI was calculated in R using the Veganv2.4.4 package on genus-level reads per million values for all microbes, including bacteria. A Wilcox Rank Sum test was used to evaluate differences in SDI between patients in the two categories.