Clinical samples

The samples were obtained from the Colombo Dengue Study (CDS), which is a single centre prospective cohort study recruiting clinically suspected adult dengue patients presenting within the first 96 hours of fever to the National Hospital of Sri Lanka [15]. The work described in this paper was limited to samples collected prior to the pandemic (2017–2020). The patients are followed up daily during the hospital admission to record the clinical outcomes, including plasma leakage and severe dengue [16]. The design of the cohort, recruitment strategy and findings from other CDS projects have been published previously [2,15,17,18]. All patients recruited to CDS undergo two diagnostic tests: an NS1 dengue antigen test (a bedside immunochromatography test) and an RT-PCR test. The latter test is done retrospectively via batch processing for logistical purposes. Thus, the treating physicians are unaware of the final diagnosis when the patient is admitted to the hospital, given the relatively low sensitivity of the NS1 test [15]. Hence, all clinically suspected patients are managed as dengue fever. Neither NS1 antigen nor RT-PCR testing is regularly available for patients treated in the public healthcare system of Sri Lanka due to their cost [19]. If an enrolled patient was suspected of having a bacterial infection (e.g., leptospirosis) and were treated as such during their hospital stay, or if an alternative diagnostic test including blood culture became positive, they were excluded from CDS. There were 345 patients (39%) from Phase 1 recruitment in whom both NS1 and RT-PCR tests were negative for dengue, and so they were considered likely to have had an alternative infection or a non-infective febrile illness (termed here non-dengue fever patients or NDF).

Nucleic acid extraction

Consecutive NDF cases (n = 60) with available stored plasma samples were selected for this study. Total nucleic acids (TNA) were manually extracted from plasma samples using the MagMAX Viral/Pathogen Nucleic Acid Isolation Kit (Catalog No. A42352, Thermo Fisher Scientific, USA), following the manufacturer’s instructions using a starting volume of 200 µL of plasma. Briefly, plasma was mixed with 10 μL of Proteinase K, 530 µL of binding solution, 20 µL of magnetic beads, and incubated at 65°C for 5 minutes, and shaken at 1,050 rpm for 5 minutes. The beads were separated with magnets (10 minutes, supernatant discarded), washed sequentially with 1 mL wash buffer and ethanol (1 mL and 500 µL of 80% ethanol), air-dried for 5 minutes before adding 60 µL of elution solution, followed by incubation at 65°C for 10 minutes. The eluted TNA were separated from the beads using a magnet. The TNA purity was assessed with NanoDrop (Thermo Fisher Scientific) and the elution was split into two aliquots for DNA and RNA library preparation. Plasma samples known to be infected with dengue (n = 2), Hepatitis C (n = 2), Hepatitis B (n = 1) and culture supernatant containing cytomegalovirus were used as controls.

Library preparation and sequencing

For DNA library preparation, the Illumina DNA Prep kit (Catalog No. 20060059, Illumina, Inc., USA), which employs bead-bound transposons to simultaneously fragment genomic DNA (gDNA) and attach sequencing primers was used. This process produces DNA fragments with a uniform size of approximately 300 base pairs, followed by PCR amplification to enrich the library. The fragment size distributions of libraries were assessed using an Agilent 2100 Bioanalyzer and DNA amount was quantified with the Quant-iT™ PicoGreen™ assay. All the above steps were done at room temperature.

For RNA library preparation, the SMARTer Universal Low Input RNA Kit (Catalog number 634940, Takara Bio, USA) was used to convert RNA into double-stranded cDNA (dscDNA). Following the manufacturer’s instructions, 10 μL of TNA was mixed with the 3’ SMART N6 CDS Primer II A to synthesize the first cDNA strand using SMARTScribe Reverse Transcriptase and the SMARTer II A Oligonucleotide at 42°C for 90 minutes. The resulting first-strand cDNA was purified using SPRI AMPure magnetic Beads. The purified first-strand cDNA, still bound to the SPRI beads, was directly used for PCR amplification with Advantage 2 Polymerase and PCR Primer IIA to produce double-stranded cDNA. This amplified cDNA library was further purified, during which the beads were washed with 80% ethanol and eluted in a purification buffer. The amplified and purified dscDNA was digested with the restriction enzyme RsaI to excise the SMART adapter sequences and purified again with SPRI beads as before.

The DNA and RNA libraries were sequenced on the Illumina NovaSeq 6000 with a 2 x 150 bp paired-end sequencing platform (S4 flow cell) to generate an average of 120–150 million reads (each 150–300 bp) per sample.

Data analysis

Sequencing data was analyzed using the open-source Illumina mNGS Pipeline v8.3 available on the CZID platform (https://czid.org). De-multiplexed reads were uploaded to CZID in FASTQ format. The pipeline checked data for quality control using fastp v 0.23.4 [20]. Human reads were removed using aligners Bowtie2 v 2.5.4 [21] and Hisat2 v 2.2.1 [22], and duplicate reads were eliminated by CZID-dedup v 0.1.2 [23]. The remaining reads were aligned to the NCBI nucleotide (NT) database using Minimap2 v 2.28 [24] and the non-redundant (NR) protein database using Diamond v 2.1.9 [25]. Short reads were also de novo assembled into contigs using SPAdes v 4.0.0 [26], which were then queried against the NT and NR databases using GSNAPL [27] and RAPSearch2 [28], respectively. Finally, the pipeline generated a taxonomic report based on pathogen-specific reads per million total reads (rPM), providing a detailed profile of the microbial taxa present at the genus level. All scripts and user instructions for the CZID workflows are available at https://github.com/chanzuckerberg/czid-workflows.

If the same pathogen was confirmed in at least 5 samples, we compared the symptoms, signs, and haematological and biochemical parameters recorded in the first 96 hours of fever against those of confirmed dengue patients in a 1:2 case control design matched for age and sex using Mann-Whitney U Test (for continuous variables) and chi-square test (for categorical variables), adjusted for multiple comparisons with Bonferroni correction. A sample size calculation was not done as this was only an exploratory study.

Confirmation of missed infections identified by mNGS

Assuming other arboviral infections (Zika and Chikungunya) to be the most common missed infections mimicking dengue, we employed the TaqMan Arbovirus Triplex Kit (Catalog Number A31747) which detects Zika, chikungunya and dengue infections to validate any positive mNGS findings for these viruses according to manufacturer’s instructions. The qPCR assays were conducted using the QuantStudio 7 Real-Time PCR System (Applied Biosystems) under the following thermal cycling conditions: reverse transcription at 50°C for 20 minutes to synthesize cDNA from RNA, initial activation at 95°C for 2 minutes, followed by 40 cycles of amplification consisting of denaturation at 95°C for 15 seconds and annealing/extension at 60°C for 1 minute. Fluorescence data were collected at each cycle, and threshold cycle (Ct) values were calculated using QuantStudio software. Each run included no-template controls (NTC) to confirm the absence of contamination.

Phylogenetic analysis

For non-dengue RNA virus infections identified, a phylogenetic analysis was performed to compare the sequence similarity to other contemporary circulating genomes of the same pathogen globally if a consensus sequence covering at least 40% of the reference genome could be recovered from mNGS reads. Complete viral genomes for comparison were downloaded from the Bacterial and Viral Bioinformatics Resource Center (https://www.bv-brc.org/), and these were aligned with consensus sequences generated in this study using multiple sequence comparison using log-expectation (MUSCLE) algorithm [29] implemented in Geneious Prime (2022.1.1, Biomatters, New Zealand). The alignment was manually edited to remove gaps in the mNGS generated consensus sequences (entire columns of genome positions were deleted). The remaining positions were concatenated and realigned, and then used to build a maximum likelihood consensus phylogenetic tree with 100 bootstrap replicates using RaXML (Randomized Axelerated Maximum Likelihood) [30]. The consensus tree was visualized by collapsing nodes with a bootstrap support of less than 70%.

Ethics approval

The research was approved by the Human Research Advisory Panel (Biomedical) at the University of New South Wales (HC220320). The CDS and its sample collection and storage processes were approved by the Ethics Review Committee of the University of Colombo, Sri Lanka (EC-17–080). All patients provided written informed consent for data and sample collection for research. Patient recruitment for samples described in this study started on 01/10/2017 and ended on 10/07/2018. The data were accessed for research purposes between 01/10/2017 and 01/11/2024. Deidentified clinical data were transferred to UNSW for analysis as approved by the Ethics Review Committee of the University of Colombo, Sri Lanka (EC-17–080) and stored in encrypted UNSW Onedrive, only accessible by the researchers. The sequence data generated were also securely transferred and stored in UNSW Onedrive.

Pathogen characterisation by mNGS

Of the 345 NDF patients recruited to CDS during the pre-COVID phase, samples from 191 patients were available to be tested. Of these the first 60 samples collected were (60/191, 31%) to be tested on the mNGS workflow described above alongside the 6 control samples. The total sequencing data generated by mNGS for this study was 790.15 GB. Each sample produced 13.17 GB of data, with an average read count of 115 million.

In the analysis of RNA virus sequence data, filters were applied in the CZID workflow to retain reads from known pathogens and samples that had at least 100 pathogen specific reads (an arbitrary, conservative cut-off). With these filters, all infections in the control samples plus 6 likely chikungunya virus infections (read count range: 114–60143, median: 2197) and another 5 dengue virus infections (read count range: 460–1125971, median: 2754), were detected. When mNGS reads were mapped across corresponding pathogen reference genomes in all samples except one chikungunya infected sample, a consensus sequence covering a minimum of 1000 nucleotides were recovered. The triplex arboviral qPCR independently confirmed 5 chikungunya infections (but not the sixth sample mentioned above), and the 5 dengue infections. The number of pathogenic virus specific reads reported in each additional sample, when the above conservative filters were removed, are detailed in S1 and S2 Tables. However, given the low read counts and limited coverage across the corresponding reference genomes, these are unlikely to be true infections.

For bacterial (DNA) data analysis, samples with likely or probable acute viral infections (n = 11) were excluded as the clinical picture was most consistent with a viral infection. Of the remaining samples (n = 49), only those with genomic reads from a known pathogen with an abundance > 100 pathogen-specific reads per million of total reads, aligning throughout the length of the reference genome, were retained. Of these samples where only a single pathogen met all the above criteria (notionally equivalent to a positive blood culture with a single bacterial growth) was assigned as a likely infection. Such stringent criteria were necessary given the plethora of commensal and opportunistic pathogen DNA recovered due to likely contamination during sample collection and processing. On this basis, five samples with positive hits of Pseudomonas putida (n-2), Aeromonas caviae (n-2), and Klebsiella pneumoniae (n-1) were identified. Given that all 5 patients had no history of an immunocompromising illness (where they would be susceptible to Pseudomonas or Aeromonas infections), and as all recovered without antibiotics, only the Klebsiella infection was considered as a possible differential diagnosis for the presenting AUF.

Characteristics of confirmed chikungunya infections

The sixty samples were collected between November 2017 and June 2018, and the five confirmed chikungunya infections were recruited in January (1), April (1), May (1) and June (2) of 2018. When the clinical records of the five confirmed Chikungunya infections were compared with those of confirmed dengue infections in a 1:2 case control design, matched for age and gender, no statistically significant differences were identified in: the frequency of symptoms or signs present on admission, the duration of the hospital stay, or in the haematological or biochemical parameters recorded in the first 96 hours of fever, except for the haematocrit value being significantly higher in dengue patients (Mann Whitney U test, p-value = 0.003, S3 Table).

The chikungunya genomes recovered from the five confirmed infections were phylogenetically compared with contemporary circulating chikungunya sequences as mentioned above. In the BV-BRC database, 933 complete chikungunya genomes sequenced from human hosts from 57 countries were available on 27^th August 2024. From this, one representative sequence per calendar year, per country, was selected if they were isolated on or after 2015 (n = 44), except for Sri Lankan sequences for which there was no time limit. These sequences were aligned with 5 sequences generated in this study, and the alignment was manually edited (see Methods) to generate a concatenated alignment of 4815 nucleotide positions (41% coverage across the genome) after removing gaps. The 5 sequences from this study clustered together and were embedded within a larger cluster of Asian sequences, which stood apart from two clusters of American (North, Central, South American and Caribbean) sequences. The phylogenetically closest sequences were from Malaysia (2017), Singapore (2015), India (2019) and Maldives (2019). All previous Sri Lankan sequences (isolated between 2006–2008) were more phylogenetically distant (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. The maximum likelihood phylogenetic tree of contemporary circulating chikungunya sequences isolated from around the world.

These sequences were selected from 933 complete chikungunya genomes available in the BV-BRC sequenced from human hosts from 57 countries. One representative sequence per calendar year, per country, was selected if they were isolated on or after 2015 (n = 44), except for Sri Lankan sequences for which there was no time limit. The sequences from the current study are in blue and previous sequences isolated from Sri Lanka are in green. Each tip of the tree is named as follows; Genbank accession number_Country_Year of isolation. The tree nodes collapsed to show only those with bootstrap support > 70%.

https://doi.org/10.1371/journal.pone.0326995.g001