West Nile Virus, an Underdiagnosed Cause of Acute Fever of Unknown Origin and Neurological Disease among Hospitalized Patients in South Africa

West Nile virus (WNV), a mosquito-borne flavivirus, is endemic to South Africa. However, its contribution to acute febrile and neurological disease in hospitalized patients in South Africa is unknown. This study examined two patient cohorts for WNV using molecular testing and IgM serology with confirmation of serological results by viral neutralization tests (VNT) to address this knowledge gap. Univariate analysis was performed using collected demographic and clinical information to identify risk factors. In the first cohort, 219 cerebrospinal fluid (CSF) specimens from patients with acute neurological disease in Gauteng hospitals collected in January to June 2017 were tested for WNV. The study identified WNV in 8/219 (3.65%, 95.00% CI (1.59–7.07)) patients with unsolved neurological infections. The second cohort, from 2019 to 2021, included 441 patients enrolled between January and June with acute febrile or neurological disease from urban and rural sites in Gauteng and Mpumalanga provinces. West Nile virus was diagnosed in 40/441 (9.07%, 95.00% CI (6.73–12.12)) of patients, of which 29/40 (72.50%, 95.00% CI (56.11–85.40)) had neurological signs, including headaches, encephalitis, meningitis, and acute flaccid paralysis (AFP). Notably, most of the cases were identified in children although adolescents and senior adults had a significantly higher risk of testing WNV positive. This suggests a previously underestimated disease burden and that WNV might be underrecognized as a cause of febrile and neurological diseases in hospitalized patients in South Africa, especially in children. This emphasizes the importance of further research and awareness regarding arboviruses of public health concern.


Introduction
West Nile virus (WNV) is a vector-borne arbovirus (Family: Flaviviridae, Genus: Orthoflavivirus) first identified in a febrile woman from Uganda in 1937 [1].Typically, only 20.00% of infections manifest clinically, usually as a mild febrile illness [2].Neurological complications can occur in 10.00% of clinically ill patients [2].Of these, the case fatality rate (CFR) is around 10.00% but is significantly higher in the elderly, immunocompromised or those with WNV encephalitis or poliomyelitis [3].Diagnosing WNV infections is challenging due to the non-specific flu-like symptoms, and the neurological presentations are indistinguishable from other pathogens [3].Due to the fleeting viremia observed in WNV infection, the detection of WNV-specific IgM antibodies in serum or CSF is the gold standard of diagnosis of acute infections [2].As antibodies do not normally cross the blood-brain barrier, the detection of WNV IgM in the CSF is a marker of neuroinvasive disease, as IgM antibodies can only permeate the blood-brain barrier due to inflammation [4].However, the persistence of WNV IgM antibodies complicates the diagnosis of acute infections.Studies have shown that WNV IgM antibodies can persist in serum from anywhere between two months up to 8 years after infection [5].Thus, they may not be a true marker of acute infection [5].However, WNV IgM antibodies seem to persist for less time in CSF (up to 7 months), making CSF preferential for diagnosing acute neurological infections over serum [4].To complicate diagnosis further, a positive serological test might be due to cross-reactive antibodies from an infection with a related flavivirus [6].Therefore, accurate diagnosis of acute WNV infection should consider both clinical symptoms and molecular and/or serological results that have been confirmed with WNV VNTs.
Outbreaks of WNV in the human population before the early 1990s were sporadic, thought to be self-limiting, and neurological infections were uncommon [7].Information on the epidemiology and ecology of the disease was based primarily on work from the Mediterranean and Egypt in the 1950s [7].The first well-established link between WNV and neurological disease in humans was identified in Israel in 1957 following an outbreak in a nursing home, whereby encephalitis developed in one-third of the infected patients [8].One of the largest WNV outbreaks ever recorded occurred in 1974 in the Karoo and Northern Cape provinces of South Africa (SA), where over 10,000 febrile cases were reported, however, no neurological cases were noted [9].The subsequent expansion of WNV in West and Central Europe, and its emergence and continuous spread across the Americas since 2000 has made WNV the most widespread cause of arboviral encephalitis worldwide [10].
Currently, nine WNV lineages are proposed but lineages 1 and 2 have been associated with outbreaks [11].Lineage clade 1a circulates in North Africa, Asia, Europe, and the Americas, and clade 1b is confined to Australia [10].Lineage 2 was thought to be geographically confined to sub-Saharan Africa and Madagascar but emerged in Europe in 2004, following the identification of an isolate in Hungary obtained from an encephalitic goshawk [10].Lineage 2 has since swiftly expanded in Central Europe and the Mediterranean region, with outbreaks characterized by unprecedented numbers of neurological disease in humans and animals [10].Most recently, lineage 2 was identified in mosquitoes and humans for the first time in the Netherlands [12].In Europe, during the 2022 WNV transmission season, over 965 human cases of WNV, and 73 deaths, have been reported [13].These recent outbreaks have shifted the focus from lineage 1 to lineage 2 as being of high public health concern [14].
Despite being endemic in SA, only 5-15 human WNV cases are reported annually, although this is probably an underestimation due to a lack of clinical awareness [9].A single study focusing on patients experiencing acute febrile or neurological disease in the Tshwane region of SA from 2008 to 2009 identified serum-neutralizing antibodies (nAbs) to WNV in 40/206 (19.42%) patients, while 4/15 (26.67%)CSF specimens had nAbs [15].However, IgM antibodies could only be confirmed in 2/206 (0.97%) serum specimens due to the limited specimen volume available and 1/190 (0.53%) were PCR positive.In SA, no study has investigated risk factors for WNV infection, nor has the burden of the disease been investigated in hospitalized patients experiencing acute fever of unknown cause with or without neurological symptoms.To address this gap, we sought to investigate the association of WNV with neurological disease using molecular and IgM serodiagnosis coupled with WNV VNT for confirmation in two patient cohorts.The first cohort was derived from CSF specimens collected from patients presenting to Gauteng hospitals with neurological disease of unknown etiology in 2017.The second cohort focused on patients presenting with acute febrile or neurological disease at three hospitals in the Gauteng and Mpumalanga provinces from 2019 to 2021.These provinces have temperate and subtropical climates, respectively.Demographic and socioeconomic risk factors as well as clinical presentation for WNV infection were investigated.This allowed us to investigate the contribution of WNV in unsolved cases of fever, with or without neurological symptoms, in hospitalized patients in the Gauteng and Mpumalanga provinces of SA.

Ethical Statement
The University of Pretoria's Faculty of Health Sciences ethics committee (100/2017 and 101/2017) approved the study.Informed consent was obtained as outlined by the African Network for Improved Diagnostics, Epidemiology, and Management of Common Infections Agents (ANDEMIA) study [16].

Study Design and Sentinel Sites
Two patient cohorts were used in this study.The first used archived CSF specimens sampled from patients suffering acute neurological disease at 10 public sector hospitals in Gauteng Province (see Figure S1 for a map of the hospitals) that were submitted to the National Health Laboratory Service (NHLS) for viral diagnoses (hereinafter described as the neurological cohort).Cases were de-identified, assigned study numbers, and entered into a database containing no identifying information.Basic demographic data and diagnostic findings were recorded through the NHLS TrakCare database.Only specimens collected between January and June 2017 were tested for WNV, as this covers the South African mid-summer to autumn months, when the availability of mosquito breeding sites and vectorial capacity increases due to the warmer and wetter weather conditions [17].This seasonality has also been confirmed through sentinel WNV surveillance in horses over 8 years [18].
The second cohort performed prospective sampling as part of the larger ANDEMIA study and consisted of two sentinel sites in SA, namely Kalafong Provincial Tertiary Hospital, which is an urban teaching hospital in the Gauteng province, and the rural Matikwane and Mapulaneng hospitals, in the Mpumalanga province (see Figure S2 for a map of the hospitals) [16].Prospective febrile and neurological disease surveillance was conducted from January 2019 to December 2021 at both sentinel sites in patients meeting the AFDUC case definition with or without acute neurological symptoms (hereinafter described as the AFDUC cohort).The case definition aimed at identifying AFDUC patients with arboviral infections [19].Patients were subsequently enrolled under informed consent by surveillance officers, and a case investigation form was completed to capture basic demographic, socioeconomic, and clinical data.A one-month follow-up was performed where possible to determine the recovery rate.Cases were de-identified, assigned study numbers, and entered into a database containing no identifying information.

Sample Processing and RNA Extraction
For the neurological cohort, only CSF specimens were obtained and processed.For each patient enrolled into the AFDUC cohort, EDTA (ethylene diamine-tetra acetic acid) blood was taken.In some cases, serum blood was collected as well.CSF was only taken if deemed necessary by medical staff as part of the standard of care.CSF and EDTA blood were used for nucleic acid extractions using the Qiagen RNA Viral mini kit (Qiagen, Valencia, USA).For serological testing, serum was preferentially used or EDTA was centrifuged at 2000× g for 15 min to separate the plasma.If a single patient had both blood and CSF specimens collected, CSF was preferentially tested for the presence of IgM antibodies due to it being a marker of acute neuroinvasive disease.Therefore, a single specimen was tested for each patient.All methods using commercial kits were carried out according to manufactures recommendations.

Multi-Pathogen Detection in the AFDUC Cohort
Routine molecular surveillance for ANDEMIA specimens was performed using the Chipron LCD macroarray assay (Chipron GmbH, Berlin, Germany), which is a multi-plex PCR designed to simultaneously detect 30 pathogens, including WNV and other flaviviruses [20].

Flavivirus Detection in the Neurological Cohort
Flavivirus screening using real-time RT-PCR, subsequent Sanger sequencing and phylogenetic analysis was performed as described previously [15] on the 2017 CSF specimens in the neurological cohort.For phylogenetic comparison to other South African strains, WNV sequences obtained from mosquitoes [17] and animals suffering neurological disease in SA [21] were included.

Euroimmun WNV IgM ELISA and Neutralization Tests
Serum/plasma and CSF specimens from both cohorts were screened for the presence of anti-WNV IgM antibodies using the Euroimmun WNV IgM ELISA kit (Euroimmun, Lübeck, Germany).Sera or plasma was diluted at 1:101 with the sample buffer as recommended, while CSF was diluted at 1:2.Due to the cross-reactive nature of flaviviruses, all IgM positive or inconclusive results were confirmed using a WNV VNT with 10 5 TCID50 U/mL (50.00% tissue culture infectious dose) of WNV culture (HS101/08, accession number JN393308.passage 6) as previously described [22].Briefly, VNTs were performed in flatbottomed 96 well plates.Two-fold dilutions (1: 8, 1: 16, 1:3 2) of test sera/plasma/CSF were prepared using Earle's minimum essential medium (EMEM) containing 2.00% fetal calf serum and 100 µL was added in duplicate to the plate.To this, 100 µL of 100 TCID50/mL WNV prepared in 2.00% EMEM was added.To prepare the back titration, 10-fold serial dilutions of the 100TCID50 stock were created, until 0 TCID50 was reached, and 100 µL of these dilutions were added in quadruplicate to the plate.The plates were placed in humidity chambers and incubated for 1 h at 37 • C with 5.00% CO 2 .After which, 80 µL of 8 × 10 5 Vero cells/mL were added to each well.Negative control wells consisted of Vero cells and EMEM only.The plates were incubated for 4-7 days at 37 • C with 5.00% CO 2 .Only once the back titration reads 100.00% cytopathic effect (CPE) at 100 TCID50, 75.00%CPE at 10 TCID50 and 25.00% CPE at 1 TCID50 and 0 CPE at 0 TCID50 was the plate was read.Specimens showing VNT titers ≥ 1: 16 were considered positive for WNV.To rule out further cross-reactions, any WNV IgM positive cases that could not be confirmed by WNV VNTs, were tested for neutralizing antibodies against the following co-circulating flaviviruses: Wesselsbron virus, Banzi virus, Bagaza virus, and Usutu virus.See Table S1 for details of the virus isolates used.

Data Analysis
The patients' ages were categorized into four groups based on the Neural Network using the Face and Gesture Recognition Research Network (FG-NET) ageing database [23].The children were grouped as less than 12 years of age, adolescents aged 13 to 18 years, adults 19 to 59 years of age, and senior adults 60 years and above.In the AFDUC cohort, disease presentation was classified based on the symptoms observed by surveillance officers.As all cases were hospitalized, no cases were assigned as mild disease.Moderate disease was defined as having only febrile illness, and severe disease if at least one neurological syndrome, such as encephalitis, AFP, meningitis or seizures, were observed.Risk factors associated with WNV infection were calculated based on patient demographic and clinical features using a univariate odds ratio analysis, with associated 95.00% confidence intervals (CI), performed in EpiInfo (version 7.2.0.1) using Fisher's exact test.

Neurological Cohort
A total of 219 archived CSF specimens submitted to the NHLS Tshwane Virology Laboratory between the months of January and June 2017 were tested for the presence of WNV.The patient's demographic data is described in Table 1

AFDUC Cohort
Routine molecular testing was performed on 979 AFDUC patients at both sentinel study sites for three years, yet no WNV positive cases were identified (refer to Table S3 for the demographic data).Acute WNV infections were investigated through IgM antibody

Discussion
Despite originating in Africa, limited data exist on the burden of WNV disease in humans on the African continent.Given the global burden of WNV, there is an increasing need for improved information on the phylogeny, epidemiology, and circulation on the African continent [24].Currently, no studies determining demographic and socioeconomic risk factors for WNV infection in SA exist, nor does a study examining a large group of hospitalized patients experiencing acute fever of unknown cause with or without neurological signs.In this study, investigation of CSF specimens from hospitalized patients experiencing acute neurological disease in Gauteng in 2017 identified WNV in 8/219 (3.65%, 95.00% CI (1.59-7.07))patients, which suggests a higher incidence than a smaller study focusing on a similar population in South Africa [15].Phylogenetic analysis confirmed endemic lineage 2 strains circulating between mosquitoes, animals, and humans in SA.The interannual WNV detection rate in patients experiencing acute febrile disease with or without neurological symptoms at two sentinel sites in the Gauteng and Mpumalanga provinces ranged from 8.33-10.91%from 2019 to 2021.This detection rate is in line with that noted in horses in SA [18].The variability in detection rates could be due to annual changes in rainfall or changes in behavior during the COVID-19 pandemic.This high detection rate supports the notion that WNV is underestimated in SA [9].The presence of clinical disease that matches the case definition for arboviral infections, along with IgM antibodies confirmed with neutralization tests, suggests these are likely acute WNV infections, despite the known persistence of these antibodies.
Since most WNV disease occurs in adults [25], the true burden of WNV disease in SA may be underestimated as the median age in both cohorts was 10 and 4 years of age.Few other studies included young children as part of the study group.Moderate and severe disease was most noted in children compared to any other age category.The high detection of IgM antibodies in children and the detection of WNV lineage 2 in two children with neurological signs suggest that WNV should be considered in neurological cases in children.Despite over half the infections being identified in children, adolescent and senior adult patients had a higher risk of having WNV infection, although these statistics are likely affected by the high number of children enrolled into the ANDEMIA study.This bias could also explain why in contrast to other studies, disease was not associated with a compromised immune system due to HIV, as children are less likely to be HIV positive.The higher detection rate at the rural Mpumalanga site may be due to a higher abundance of mosquito breeding sites, amplification hosts, and increased contact between humans and mosquitoes through a more outdoor lifestyle and increased farming activity compared to the patients at the urban sentinel site.There is a geographical correlation between areas where WNV has been identified in animals and mosquitoes, suggesting areas of endemic transmission between mosquitoes, animals, and humans within the country [17,18].
West Nile virus accounted for 11.03% (29/263, 95.00% CI (7.51-15.45)) of AFDUC cases that presented with at least one neurological symptom in hospitals in Gauteng and Mpumalanga province from 2019 to 2021, and 72.50% (29/40, 95.00% CI (56.11-85.40)) of all WNV cases had a neurological component, suggesting WNV may contribute significantly to unsolved cases of neurological disease in hospitals in SA.The prevalence of AFP in WNV cases is variable but usually ranges between 5-15% [26], which is lower than the results presented in this study (6/29, 20.69%, 95.00% CI (7.99-39.72)).However, AFP may be more common in younger patients [27].This may explain the high percentage of AFP in the study, as the majority of the patients enrolled were children.A high recovery rate was observed, which is in concordance with previous studies [2].A single death was recorded in a 38-year-old HIV positive male who, at the time of death, was experiencing sepsis and AFP.
Although no PCR positive cases and phylogenetic data could not be obtained for the WNV positive patients in the AFDUC cohort (2019 to 2021), lineage 2 is well known to be the most common lineage circulating in SA and was identified in 2 PCR positive cases identified in the neurological CSF cohort from 2017.Syndromic surveillance of horses and other animals [21] from 2017-2020 also detected WNV lineage 2 in all cases in the same regions except a single lineage 1 case identified in a lion in the Kruger National Park in 2017.Molecular testing may be a suitable diagnostic tool during early infection, but after seroconversion has taken place, serological testing is more likely to correctly identify cases.However, VNTs are required for confirmation for accurate epidemiological studies.The number of confirmed WNV cases reduced from 14.29% (63/441, 95.00% CI (11.33-17.86)) to 9.07% (40/441, 95.00% CI (6.73-12.12))using VNTs.We were unable to identify if another flavivirus was responsible for these cases despite testing for co-circulating flaviviruses using VNTs.The detection of WNV IgM antibodies in serum or plasma shows a CSF specimen is not crucial for diagnosis.
This study has a few limitations, the first being the high enrolment numbers of children in the AFDUC cohort.The low enrolment numbers in the other age groups make the accurate determination of the relative proportion difficult.Secondly, IgM ELISA was used for screening of serum and CSF samples, however not all neurological cases had CSF specimens available and were diagnosed on serum alone.As CSF was preferentially tested over serum for WNV IgM, fewer neuroinvasive cases may have been missed.Both serum and CSF should be tested if available for a single patient.Since we did not have paired serum samples that would show an increase in antibody levels, we only considered IgM-positive cases that could be confirmed via virus neutralization tests as positive.As we only tested patient sera or CSF that failed to neutralization WNV for cross-reactivity to other flaviviruses after the WNV VNTs were performed, this leaves the possibility that a some WNV VNT results were false positives, as cross-reactivity needs to be performed simultaneously, and the decision reached is based on which titer is the highest.
There was limited access to patients placed in the intensive care unit, and patients who had unfortunately demised were unable to give consent.Therefore, the enrolments of patients experiencing severe AFDUC symptoms were likely affected.
To conclude, the findings identified WNV as the etiological agent in 8-11% of hospitalized patients experiencing acute febrile and/or neurological disease in SA.This study lays the groundwork for defining the true burden of WNV amongst undiagnosed cases of febrile and neurological disease in hospitalized patients in South Africa and suggests a significant disease burden not previously known for Africa.Our study confirms that WNV should form part of the differential diagnoses of AFDUC cases with neurological signs in hospitalized patients in SA, particularly in children.Increased clinical awareness and prospective surveillance across countries in which animal cases were previously detected as well as in other parts of Africa is needed to define the true burden.

Figure 1 .
Figure 1.Maximum Likelihood phylogram of the partial NS5 gene (nt = 226) used for confirmation of WNV human infection indicating bootstrap (bs) support values on the branches (36 taxa).Viral sequences obtained in this study from hospitalized patients in SA are indicated by filled squares.Filled black triangles indicate sequences identified in mosquito vectors[15].Filled circles represent the viral sequences identified in animals in South Africa[18].Values are shown if bs support was ≥70.The evolutionary history was inferred using the Kimura 2-parameter model plus gamma distribution.The tree is drawn to scale and branch lengths represent the number of nucleotide substitutions per site.Reference sequences were downloaded from GenBank.GenBank accession numbers for the WNV NS5 sequences are ZRUNH273/17 = OL790166, ZRUNH497/17 = OL790167.

Figure 1 .Figure 2 .
Figure 1.Maximum Likelihood phylogram of the partial NS5 gene (nt = 226) used for confirmation of WNV human infection indicating bootstrap (bs) support values on the branches (36 taxa).Viral sequences obtained in this study from hospitalized patients in SA are indicated by filled squares.Filled black triangles indicate sequences identified in mosquito vectors [15].Filled circles represent the viral sequences identified in animals in South Africa [18].Values are shown if bs support was ≥70.The evolutionary history was inferred using the Kimura 2-parameter model plus gamma distribution.The tree is drawn to scale and branch lengths represent the number of nucleotide substitutions per site.Reference sequences were downloaded from GenBank.GenBank accession numbers for the WNV NS5 sequences are ZRUNH273/17 = OL790166, ZRUNH497/17 = OL790167.Viruses 2023, 15, x FOR PEER REVIEW 7 of 14

Figure 2 .
Figure 2. The seasonality of WNV in hospitalized patients in South Africa, January-June 2017, 2019 to 2021.

Table 2 .
Risk factors associated with West Nile virus infection and clinical symptoms/syndromes in hospitalized patients with neurological disease in South Africa, January to June 2017.West-

Table 2 .
Risk factors associated with West Nile virus infection and clinical symptoms/syndromes in hospitalized patients with neurological disease in South Africa, January to June 2017.

Table 3 .
Demographic data of the subset of AFDUC patients enrolled for West Nile virus serological testing, January to June 2019 to 2021.

Table 4 .
Details of the West Nile virus testing, including number of AFDUC patients and specimens tested, January to June 2019 to 2021.

Table 5 .
Details of the West Nile virus positive AFDUC patients in South Africa.January to June 2019 to 2021.

Table 6 .
Demographic and socioeconomic factors associated with West Nile virus infection in AFDUC patients in South Africa, January to June 2019 to 2021.

Table 7 .
Clinical symptoms/syndromes, recovery rate, and case fatality rate associated with WNV infection in AFDUC patients in South Africa, 2019 to 2021.