Development of an improved RT-qPCR Assay for detection of Japanese encephalitis virus (JEV) RNA including a systematic review and comprehensive comparison with published methods

Background Japanese encephalitis virus (JEV) is a major cause of encephalitis in Asia, and the commonest cause of mosquito-borne encephalitis worldwide. Detection of JEV RNA remains challenging due to the characteristic brief and low viraemia, with 0–25% of patients positive, and the mainstay of diagnosis remains detection of anti-JEV IgM antibody. Methods We performed a systematic review of published RT-PCR protocols, and evaluated them in silico and in vitro alongside new primers and probes designed using a multiple genome alignment of all JEV strains >9,000nt from GenBank, downloaded from the NCBI website (November 2016). The new assays included pan-genotype and genotype specific assays targeting genotypes 1 and 3. Results Ten RT-qPCR assays were compared, a pre-existing in-house assay, three published assays and six newly designed assays, using serial RNA dilutions. We selected three assays, one published and two novel assays, with the lowest limit of detection (LOD) for further optimisation and validation. One of the novel assays, detecting NS2A, showed the best results, with LOD approximately 4 copies/ reaction, and no cross-reaction on testing closely related viruses in the JEV serocomplex, West Nile Virus and St. Louis Virus. The optimised assays were validated in consecutive patients with central nervous system infections admitted to hospitals in Laos, testing paired CSF and serum samples. Conclusions We succeeded in developing a JEV specific RT-qPCR assay with at least 1 log10 improved sensitivity as compared to existing assays. Further evaluation is required, field-testing the assay in a larger group of patients.


Results
Ten RT-qPCR assays were compared, a pre-existing in-house assay, three published assays and six newly designed assays, using serial RNA dilutions. We selected three assays, one published and two novel assays, with the lowest limit of detection (LOD) for further optimisation and validation. One of the novel assays, detecting NS2A, showed the best results, with LOD approximately 4 copies/ reaction, and no cross-reaction on testing closely PLOS

Introduction
Japanese encephalitis virus (JEV) represents a major cause of encephalitis in Asia, with an estimated 70,000 cases per year, and 20,000 deaths [1][2][3][4][5]. It is a highly neurotropic arthropodborne virus (arbovirus), a member of the genus Flavivirus and family Flaviviridae, and closely related to West Nile virus (WNV) and St. Louis encephalitis virus (SLEV). Five JEV genotypes have been identified, G1-5, with at least 12% nucleotide difference between them, and concomitant geographical variations. G1 and G3 have been isolated and characterised most extensively in humans [6][7][8][9][10][11][12][13][14]. G3 was the main genotype identified in Asia until the 1990s. However, there has been a recognised genotype displacement to G1 which is now predominant [10,15]. G1 is the only genotype that has been isolated in Laos, but it is likely that G3 co-circulates as well [6,16].
JEV is transmitted between birds and swine in enzootic cycles with mosquito vectors (mainly Culex spp.) [17]. Humans are 'dead-end' hosts, infected when they encroach upon enzootic cycles, mainly involving transmission from swine, which live close to humans and have high and prolonged viraemias. The viraemia elicited in humans is typically brief and low, and in endemic areas only a small proportion of those infected (1 in 300) experience symptoms [18][19][20].
There are recognised limitations in current diagnostics for JEV [21][22][23][24]. The mainstay of diagnosis remains the detection of anti-JEV IgM antibody using IgM antibody capture ELISA (JEV MAC-ELISA) from blood and/or cerebrospinal fluid (CSF) samples [25][26][27]. However cross-reactivity of Flavivirus immune-assays is well-documented, as seen with Dengue virus infection [28]. Moreover, recent dynamics in Flavivirus epidemiology, seen with the epidemic of Zika virus, further questions our dependence on detection of anti-JEV IgM [28][29][30][31]. Field studies suggest the sensitivity of JEV MAC-ELISA is only 50-93%, and more recently the positive predictive value has also been questioned [22,23,32,33]. Additionally, the use of serology for diagnostic confirmation limits knowledge of molecular epidemiology and vaccine efficacy.
Reasons for reliance on JEV MAC-ELISA are the short period of JEV viraemia in humans and potential low sensitivity of existing JEV RT-PCR techniques. Data suggest that JEV RNA is detected in blood and/or CSF in a very small proportion (0-25%) of clinical cases [32,34]. Studies reporting higher proportions, 25-30%, involve selected patient populations, such as patients with less than 3 days onset of fever and more severe symptoms [35,36].
We aimed to perform a systematic review of the JEV RT-PCR assays available and to compare their performance. In addition, we developed new assays, in order to obtain the highest efficiency. This required extensive sequence analysis to design an in-silico optimal system followed by optimization of the experimental protocol. A set of selected assays were then assessed on patient samples.
Selection of the best performing RT-qPCR systems. The ten different systems were compared by testing them on 7 tenfold dilutions of JEV G1-1326 RNA in duplicate, and G3 RP9-190 RNA tested once, using 'Superscript-III kit' following standard conditions, see Table 1. On the basis of these results, the three RT-qPCR systems with the lowest limit of detection (LOD), i.e. highest dilution found positive for all replicates, were selected for optimisation.

Validation
For the three selected RT-qPCR assays, efficiency, repeatability and LOD, were assessed using optimised RT-qPCR assay protocols (based on results of 2.4.1 and 2.4.3) using JEV G1-769 or JEV G3-RP9-190 RNA dilutions (10 −3 to 10 −10 ) in triplicate, and repeated on two different days. Cross-reaction of the assays were tested using RNA from viruses in the JEV serocomplex at 10 −5 dilutions: two WNV strains and one SLEV strain.

Evaluation using patient samples with Central Nervous System (CNS) infections
Two groups of consecutive patients with CNS infection, admitted 2009-2015, were retrospectively tested: group 1 with acute encephalitis syndrome determined using the clinical WHO criteria [44] and admitted within 7 days of fever onset to Mahosot Hospital; and group 2 with suspected CNS infection from three other hospitals in Vientiane, the Friendship, Children and Setthathirat Hospitals. CSF was collected by lumbar puncture (LP) from consenting patients without contraindications to LP, according to the judgment of the responsible physician. Blood was collected on admission for all patients. Samples were sent to the Mahosot Hospital Microbiology Laboratory. CSF and serum, after blood centrifugation, were aliquoted and stored at -80˚C. 200 μl serum and CSF samples were extracted and eluted in 60μl on a EZ1 Advanced machine using EZ1 Virus Mini Kit v2.0 (Qiagen, UK) following manufacturer's instructions. The validated JEV RT-qPCR assays were performed as per the optimised protocols for each sample. Positive and negative (no template) controls were performed for each RT-qPCR run. An internal control (MS2 phage) was added to each patient sample. After extraction, MS2 RT-qPCR was performed to control the extraction process and to exclude inhibition as previously described [45].

Sequencing
Single positive RT-qPCR results were further investigated by sending the RT-qPCR product for next-generation sequencing using Ion S5 system (Thermo Fisher, Waltham, USA), to Unité des Virus Emergents, Faculty of Medicine, Marseille, France.

Pan-flavivirus RT-PCR
The standard diagnostic algorithm at Mahosot Hospital did not include a JEV RT-PCR assay, but it did include prospective testing using a pan-flavivirus hemi-nested SybrGreen RT-qPCR with sequencing for identification. This was performed as per a published protocol [46], with results presented alongside JEV RT-qPCR results.

Panbio ELISA
Detection of anti-JEV IgM using JEV MAC-ELISA is the World Health Organisation recommended test for diagnosing JEV infection. Japanese Encephalitis-Dengue IgM Combo ELISA kit (Catalogue number E-JED01C; Panbio ELISA; Inverness Medical Innovations, Brisbane, Australia, formerly Panbio Ltd) is a commercial JEV MAC-ELISA kit that is combined with an anti-Dengue IgM test to exclude cross-reactivity. The test was performed according to the manufacturers' instructions, reported as anti-JEV or anti-Dengue IgM Positive, Equivocal or Negative [22]. For the purposes of this analysis, anti-Dengue IgM Positive or Equivocal, or anti-JEV IgM Equivocal were interpreted as anti-JEV IgM Negative.

Ethical considerations
The study was part of an ongoing study on the causes of CNS infections in Laos. Ethical clearance was granted by the Ethical Review Committee of the Faculty of Medical Sciences, National University of Laos, and the Oxford University Tropical Ethics Research Committee, Oxford, UK. Diagnostic testing was performed on anonymised and frozen serum and CSF samples previously collected. Venepuncture and LPs were performed if written consent was given by patients or their parents/guardian.

Systematic review and bioinformatics analysis
Forty-six studies evaluating JEV RT-PCR assays were identified (see S1 File, S1 Fig and Table A in S2 File), including 7 assays utilising hydrolysis probes [27,[47][48][49][50][51][52]. The first published record of JEV RT-PCR was performed as a conventional RT-PCR at the National Institute of Health, Maryland, USA, in 1991 [53]. Since then, a variety of techniques have been harnessed: conventional and nested RT-PCR; and RT-qPCR using hydrolysis probes or SYBR green. Genotype-specific assays have also been reported, specifically for G1 and G3 [52,54]. Thirteen (28%) studies reported detection of JEV RNA in patients (see Table A in S2 File). Among these studies, the median number of published JEV cases confirmed by the assay was 12 (range 1-75). These mainly involved CSF samples, but the assays also detected JEV RNA in serum. A case report has recently demonstrated detection of JEV RNA in urine, although this was not substantiated by a larger case series of 52 JEV patients [55,56].
The primers and probes from the five published hydrolysis probe RT-qPCR assays specifically designed for detection of JEV RNA in human cases (Table C in S2 File) were aligned with the 303 complete genome sequence (see Table B in S2 File for Accession numbers, Table D [27,48,50] systems showed the best fit with the sequence alignment in-silico, with published validation data including genotypes tested, and were chosen for subsequent analysis. Six new primer/probe sets were also designed, targeting conserved regions of 1) all Genotypes, 2) G1 and G3 (Table 4). It is notable that the G1 and G3 assays were designed using two separate multiple genome alignments containing only strains belonging to the corresponding genotype. However, they were not exclusively targeting a specific genotype, and for this reason the assays were tested against both G1 and G3.

Technical development of RT-qPCR assays
Comparison of RT-qPCR conditions using the pre-existing in-house system. Experiments were performed to establish the best mastermix kits, reaction and sample volumes using the pre-existing in-house system. All the possible combinations were not performed, but the kits were evaluated with 5 tenfold dilutions of G1-1326 RNA in duplicate or triplicate to adequately compare the limits of detection.  NS2Av2 was designed as NS2Av1 with a modification of the probe sequence as its reverse complement.
NS5v2 was designed as NS5v1 with a modification of the probe sequence, with a degeneracy Y inserted instead of C at position 16. NS3v2 was designed as NS3v1 with a modification of the forward primer sequence, with a degeneracy Y inserted instead of C at position 9.  (25μl vs 50μl). The manufacturers' instructions for both Superscript-III and Fastvirus kits advise performing RT-qPCR as 50μl reaction volumes. In an effort to conserve reagents, only Superscript-III kit was used to compare two reaction volumes, 25μl and 50μl. Increasing the reaction volume from 25μl to 50μl resulted in a lower LOD, 10 −6 vs 10 −7 (Table E in S2 File).
Sample volumes. The standard sample volume was 5μl. As suggested by the kit manufacturer, the sample volume may be increased by not adding water in the reaction mix. For Superscript-III kit, the sample volume may be increased to 8μl in a 25μl reaction volume, equivalent to 16μl in a 50μl reaction volume. For Fastvirus kit the sample volume may be increased to 15μl in a 25μl reaction volume, vs. 30μl in a 50μl reaction volume. The results demonstrate a lower LOD (10 −7 vs. 10 −6 ) with a larger sample volume ( Table E in S2 File).
The experiments thus demonstrated that the optimal conditions involved use of the Fastvirus kit with a 30μl sample volume and 50μl reaction volume. These optimised conditions were subsequently used for the validation experiments described in Section 3.3.
Selection of the best performing RT-qPCR systems. A total of ten RT-qPCR systems, the pre-existing in-house, three published systems, and six (two versions each of three) newly designed systems, were tested on JEV G1-1326 RNA in duplicate, and G3 RP9-190 RNA once, in tenfold serial dilutions using Superscript-III under standard conditions. RT-qPCR results are presented in Table F in S2 File. Three RT-qPCR systems with the lowest LOD were selected for further optimisation: two pan-genotype assays, the Pyke and NS2A version 1 (subsequently called NS2A); and a G3 assay, NS3 version 2 (subsequently called NS3).
Assay optimisation. The Tm, primer and probe concentrations were optimised using Superscript-III under standard reaction and sample volumes.
The performance of the three selected assays at different annealing temperatures from 52 to 64˚C was assessed by testing JEV G1-1326 RNA in triplicate in tenfold serial dilutions. A three-step protocol was used for the cycling: 1) denaturation 95˚C for 0:15, 2) annealing x˚C for 0:30, and then 3) extension 68˚C for 0:30. Adjusting the annealing temperature of the thermocycling protocol improved the LOD by one log 10 for two assays, Cq results presented in Table G in S2 File. The optimal annealing temperature for the Pyke assay was 60˚C, NS2A assay was 62˚C and NS3 assay was 56˚C (Table G in S2 File).
The assays were tested in duplicate at various combinations of forward and reverse primers concentrations, from 200-600nM, using G1-1326 RNA at 10 −6 , results presented in Table H in S2 File. There was minimal difference between Cq results, and a concentration of 600nM for both forward and reverse primers were selected for further experiments.
The assays were tested at different probe concentrations, from 100-400nM, using G1-1326 RNA in duplicate at 10 −6 , results presented in Table I in S2 File. 300nM was selected for further experiments.
Note that Cq values in Tables H and Table I in S2 File are not directly comparable to each other because different PCR runs employed different threshold settings.
Validation. In view of the results detailed above, the final conditions employed Fastvirus kit with a reaction volume of 50μL, sample volume of 30μl, and primer and probe concentrations of 600nM and 300nM, respectively. The optimal annealing temperature was different for each assay: 60˚C, 62˚C and 56˚C for the Pyke, NS2A and NS3 assays, respectively. This was performed as a two-step protocol, as per the Fastvirus manufacturer's instructions.
Efficiency was calculated from the standard curve for Day 1 and 2: 101.3% and 105.2% for the Pyke; 95.1% and 101.2% for NS2A; and 107.5% and 110.2% for NS3 (Fig 1). Assay repeatability was assessed by calculating the standard deviation (SD) for the Cq variance, presented in Table 5. SD for all assays was less than 0.40, excluding the dilution at the LOD, suggesting adequate repeatability of the assays.
No amplification was detected in testing the assays with WNV or SLEV.  RNA copies/μl; GenBank KF907505, EVA 001V-02344) with triplicates on 2 different days. N = replicate number. LOD (highest dilution found positive by all replicates) is shaded in grey. RT-qPCR performed with Fastvirus kit (TaqMan1 Fast Virus 1-Step) with a reaction volume of 50μL, sample volume of 30μl, and primer and probe concentrations of 600nM and 300nM respectively. Thermocycling conditions were 50˚C for 5 minutes, 95˚C for 20 seconds and 45 x (95˚C for 15 seconds + x˚C for 60 seconds). The optimal annealing temperature 'x˚C' was different for each assay: 62˚C, 60˚C and 56˚C for the Pyke, NS2A and NS3 assays respectively. https://doi.org/10.1371/journal.pone.0194412.t005

Evaluation using patient samples with Central Nervous System (CNS) infections
123 patients were tested with the three optimised RT-qPCR assays (conditions as described in  Table 6, alongside results for JEV MAC-ELISA, hemi-nested panflavivirus RT-qPCR and RT-qPCR sequencing results. None of the samples were considered inhibitory according to the results of MS2 testing. Two patients (2 CSF samples) were positive using the NS2A assay, of which 1 (1 CSF) was also positive by the NS3 assay. These 2 patients were known positives by cell inoculation and/or hemi-nested panflavivirus assay. 12 (8 CSF and 4 serum samples) were positive by the Pyke assay alone. As 11/12 (92%) of these patients were not previously detected by the panflavivirus assay or cell inoculation, the RT-qPCR products were sent for sequencing. This revealed that the amplicons generated by the Pyke assay with these clinical samples consisted of primer multimers and primer/probe concatemers resulting from non-specific amplification, possibly related to the unusually high GC content at 3' end of the reverse primer, GCCC.

Discussion
This systematic review confirmed that the detection of JEV RNA in suspected human cases by RT-PCR is uncommon. Forty-six studies were identified, involving a variety of RT-PCR techniques, however only a quarter had any published record of detection of JEV RNA in human samples. It was common for an article title to poorly identify the study as evaluating a JEV RT-PCR method, and while every effort was made to hand-search publications, it is possible that the search was not 100% comprehensive. Additionally, studies utilising the assays may not have cited them. Articles rarely adhered to the MIQE guidelines for evaluation of RT-PCR methods, such as reporting accession numbers of JEV strains used or details of the optimisation process [39]. Many were only optimised and validated on G3, whereas G1 is now the predominant strain. It is also important to differentiate studies developing RT-qPCR for detection in mosquitos as opposed to humans, as they are not consistently interchangeable. Mosquito surveillance may require a cost-effective and rapid method, but in clinical diagnostics analytical accuracy is of utmost important. To this end, a number of studies detecting JEV RNA in clinical samples incorporated nested RT-PCR methods. Although nested RT-PCR may improve both analytical sensitivity and specificity, it is cumbersome and prone to cross-contamination. In the study by Meiyu et al [57], it was reported that out of 52 clinical JEV cases, an extraordinary 45 (87%) were PCR positive for JEV RNA. However, there was no additional evidence to support the diagnosis, such as repeat testing, sequencing of the isolates or alternative confirmation. Existing RT-qPCR assays detect JEV RNA in 0-25% of clinical cases.
It is for these reasons that we focussed on identifying and validating the most effective RT-qPCR assay. Hydrolysis probe assays are the most extensively used methods for nucleic acid detection, both for clinical and research diagnostic purposes. From the results of the systematic review, 3 published systems incorporating hydrolysis probes were selected on the basis of best matching in silico. Assessment with JEV RNA dilutions of G1 and G3 showed the lowest LOD was obtained with the Pyke system: 2 logfold below both the Yang and Shirato assays with G1, and 1 logfold below both with G3 [27,50]. It is recognised that the conditions employed were not identical to those published, and it would have been interesting to compare the validated assays with existing assays performed with published conditions. Validation experiments confirmed good RT-qPCR efficiency and LOD of 1.3 JEV RNA copies/μl or 39 copies/reaction. However, in a study of patient samples with paired serum and CSF, all the positive RT-qPCR  Negative Negative Negative Non-specific amplification products that were sequenced, even results with Cq of approximately 30 showed non-specific amplification. Eleven positive samples were detected by the Pyke assay alone, of which all were sequenced and showed non-specific amplification.
The results highlight the importance of validating a newly designed RT-qPCR method against a panel of negative patients samples in addition to the no-template controls recommended by the MIQE guidelines [39]. The oligonucleotides tested in these experiments had been checked in-silico for specificity, and also experimentally tested for cross-reactivity against strains of the closely related viruses, WNV and SLE. All the experiments had been performed with at least one no template control per RT-qPCR plate. Nonetheless, the results of validation against patient CSF and serum samples suggested poor specificity of the Pyke assay. It is possible that further optimisation, increasing the annealing temperature and reducing the primer concentrations may have improved the specificity, although this may adversely affect sensitivity of the assay.
The new assays aimed to improve the sensitivity of detection, through targeting various conserved regions from a multiple genome alignment. The NS2A region was chosen for the pan-genotype assay, and NS5 and NS3 for the G1 and G3 systems, respectively. Two versions of each system were designed. An initial assessment showed poor performance of the NS5 assay, and this was not used in further experiments. Initial optimisation experiments demonstrated best results with Fastvirus mix, with the maximal reaction and sample volumes (although it is recognised that use of 30 μL of extract for a single assay may not be consistently possible in practice). The assays were individually optimised for annealing temperature, primer and probe concentrations. Validation showed one logfold lower LOD for the NS2A compared to the published Pyke assay for G1, NS2A assay was 1.3x10 -1 JEV RNA copies/μl or 4 copies/reaction. NS2A and NS3 showed equally good results for G3, with LOD 1.2 x10 -1 JEV RNA copies/μl or~4 copies/reaction. Individual assays targeting highly conserved regions of G1 and G3 did not improve the sensitivity of diagnosis. Nonetheless, it may be useful to target multiple regions of the viral genome to overcome false negatives due to mutations in any single region.
A study of patient samples, including paired serum and CSF, demonstrated marginally improved detection of JEV RNA as compared to the existing hemi-nested panflavivirus RT-qPCR assay. One case was detected with the hemi-nested RT-qPCR assay, and both this and an additional case were detected by the new assays. The low number of positives is consistent with previous data. It must also be taken into account that this involved retrospective testing using samples stored for many years. The potential greater sensitivity of the NS2A as compared to the NS3 is likely due to both JEV strains detected established previously by sequencing as G1.
In conclusion, the novel pan-genotype NS2A assay and G3 NS3 assay described here showed higher analytical sensitivity than previously described RT-qPCR assays for JEV RNA. However, it will be essential to field-test them in serum, CSF and other body fluids, to evaluate their sensitivity and potential for improved diagnostic capacity in a large prospective study. Since our systematic review, there have been no further JEV specific diagnostic assays published, but there has been evidence of the potential for improved detection of JEV RNA in other body fluids, whole blood [56], throat swabs (Bharucha et al, submitted) and urine samples [56,58], and this needs further evaluation.  Table A: Overview of JEV-specific RT-PCR assays Ã and Sample Types Used to Detect Human JEV cases. Table B: Accession numbers of all included sequences.