Abstract
This study assessed the risks posed by noroviruses (NoVs) in surface water used for drinking, domestic, and recreational purposes in South Africa (SA), using a quantitative microbial risk assessment (QMRA) methodology that took a probabilistic approach coupling an exposure assessment with four dose-response models to account for uncertainty. Water samples from three rivers were found to be contaminated with NoV GI (80–1,900 gc/L) and GII (420–9,760 gc/L) leading to risk estimates that were lower for GI than GII. The volume of water consumed and the probabilities of infection were lower for domestic (2.91 × 10−8 to 5.19 × 10−1) than drinking water exposures (1.04 × 10−5 to 7.24 × 10−1). The annual probabilities of illness varied depending on the type of recreational water exposure with boating (3.91 × 10−6 to 5.43 × 10−1) and swimming (6.20 × 10−6 to 6.42 × 10−1) being slightly greater than playing next to/in the river (5.30 × 10−7 to 5.48 × 10−1). The QMRA was sensitive to the choice of dose-response model. The risk of NoV infection or illness from contaminated surface water is extremely high in SA, especially for lower socioeconomic individuals, but is similar to reported risks from limited international studies.
INTRODUCTION
Exposure to contaminated water is an important route for the transmission of diarrhoeal pathogens (Enger et al. 2012) and the waterborne transmission of viruses is of public health concern (Gibson 2014). Noroviruses (NoVs) have been identified as the major cause of waterborne disease outbreaks in many regions of the world (Gibson 2014) and are recognised as the predominant cause of sporadic and epidemic gastroenteritis accounting for greater than 90% of viral gastroenteritis and approximately 50% of all-cause outbreaks (Atmar 2010). NoVs are members of the Caliciviridae family and are small non-enveloped viruses with a single stranded, positive-sense RNA genome. NoVs contain seven genogroups (GI–VII) with GI, GII and GIV causing infections in humans. Following a mean incubation time of 24 hours, NoV illness, characterised by diarrhoea, vomiting, fever, headache, and muscle aches, is self-limiting with symptoms usually lasting between 24 and 48 hours (Vinje 2015).
In South Africa (SA), outbreaks of NoV-associated gastrointestinal illness were first described in 1993 (Taylor et al. 1993) and NoVs have been shown to be the second most important viral agent causing gastroenteritis in children after rotavirus (Mans et al. 2010). Norovirus was detected in 21–95% of river samples from selected rivers in the Gauteng Province, SA (Mans et al. 2013) and a high diversity of NoV GI and GII genotypes has been identified in wastewater from several provinces of SA (Mans et al. 2013; Murray et al. 2013a). To estimate the consequences from exposure to NoV, quantitative microbial risk assessment (QMRA) with the four key steps of hazard identification, exposure assessment, dose-response assessment, and risk characterisation can be applied (Hunter et al. 2003; Mena 2007; Haas et al. 2014). In SA to date, only one small study using simple deterministic risk assessment of NoVs in water has been carried out (Genthe et al. 2013). The aim of the present study was to quantify NoV in selected river water samples in Gauteng, SA and to assess the risks posed by these NoVs from exposure via drinking, domestic use and recreation.
METHODS
Study area, samples, and qualitative NoV detection
From January 2011 to December 2014, grab water samples (10 L) were collected at sampling points located on the Klip (KR), Suikerbosrant (SR), and Rietspruit (RV) rivers in the Gauteng province, SA. These rivers are all tributaries of the Vaal River, which forms one of the largest river systems in SA. Collected samples were transported cooled on ice blocks to the laboratory. On arrival the temperature and pH of the samples were measured and the samples were stored at 4 °C until processing – usually within 24 h of receipt in the laboratory. Viruses were recovered from the samples by glass wool adsorption-elution followed by secondary concentration using polyethylene glycol 6,000/sodium chloride precipitation as described previously (Mans et al. 2013; Kiulia et al. 2014). The recovered virus concentrates were resuspended in a final volume of 10 mL phosphate buffered saline (pH 7.4) (Sigma Aldrich Co., St Louis, MO). Nucleic acid was extracted from 1 mL virus concentrate using the MagNa Pure LC Total Nucleic Acid Isolation Kit (large volume) (Roche Diagnostics GmbH, Mannheim, Germany) on an automated MagNa Pure LC platform (Roche) according to the manufacturer's recommendations. The nucleic acid was eluted in a final volume of 100 μL. Norovirus GI and GII (using 5 μL RNA) were detected qualitatively by real-time reverse transcription polymerase-chain reaction (RT-PCR) using commercial NoV GI (norovirusGI@ceeramTools™ Kit, Ceeram S.A.S, La Chappelle-Sur-Erdre, France) and NoV GII (norovirusGII@ceeramTools™ Kit, Ceeram S.A.S) kits which had internal amplification controls. After NoV detection the remaining recovered virus concentrates were stored at −20 °C for between 1 and 4 years before NoV quantification in 2015.
Nov quantification
Norovirus-positive GI and GII water samples, with cycle threshold (Ct – the number of cycles required for the fluorescent signal to cross the background threshold level) values <40 and of sufficient volume, were selected for viral quantification. Samples were thawed and nucleic acid was extracted from 1 mL virus concentrate using using the semi-automated Nucli-SENS® EasyMAG® instrument (BioMérieux, Marcy l'Etoile, France) with the nucleic acid being eluted in a final volume of 100 μL. Prior to nucleic acid extraction, mengovirus (1.97 × 105 genome copies (gc)) was added to each sample as an extraction control. Standard curves for NoV GI and GII were prepared using commercial plasmid DNA quantification standards (Norovirus GI Q Standard & Norovirus GII Q Standard: Ceeram S.A.S). Mengovirus (5 μL RNA) was detected using a qualitative commercial assay mengo@ceeramTools™ Kit, Ceeram S.A.S). NoV GI and GII (5 μL RNA per assay) were quantified separately in each sample using commercial real-time RT-PCR assays (norovirusGI@ceeramTools™ Kit and norovirusGII@ceeramTools™ Kit, Ceeram S.A.S) with the inclusion of a DNA standard in each assay. The extraction efficiency (%) was calculated by comparing the mengovirus concentration after extraction with the input mengovirus concentration (% = final/input × 100). If the extraction efficiency was greater than or equal to 100%, then full extraction of NoV was assumed.
QMRA model
Statistical fitting was programmed in R software (R Foundation for Statistical Computing, Vienna, Austria) as was the QMRA model, which took a probabilistic approach (10,000 iterations) to account for variability and uncertainty. The exposure pathways considered were for both the lower and upper socioeconomic status populations (LSES and USES). For the LSES, the exposure pathways of: (a) consumption of contaminated river water for drinking; (b) incidental exposure by ingestion during domestic activities such as laundry or body washing were used; and (c) incidental exposure by ingestion during recreational activities (playing by river). For the USES, the exposure pathway considered was incidental exposure by ingestion during recreational activities such as swimming or boating.
Exposure assessment
Exposure assessment parameters
Variable . | Definition . | Units . | Distribution . | Value . | Description . | Reference . |
---|---|---|---|---|---|---|
CNoV GI | NoV GI Concentration | gc/L | Lognormal | μ = 5.650, σ = 0.7158 | Mean = 360 gc/L | Collected & analysed data (see Table 3 for raw data) |
CNoV GII | NoV GII Concentration | μ = 7.203, σ = 0.7412 | Mean = 1,780 gc/L | |||
Rlow | % recovery | % | Uniform | Min = 0.01, max = 3.80 | Tap & raw water | Ruhanya (2013) |
Rhigh | % recovery | % | Uniform | Min = 7, max = 60 | Range across 3 water sources | Lambertini et al. (2008) |
Inact | Inactivation by sunlight | Log reduction | Uniform | Min = 1, max = 3 | NTU = 40 to 80 | Best estimate; based on Flannery et al. (2013) and Lee & Ko (2013) |
Inf | % infectious (viable) | % | n/a | 100 | Conservative assumption | Best estimate |
Vcons, DW | Consumption DW | L/day | Lognormal | μ = −0.036, σ = 0.772 | Mean = 1.3 L per person per day | US Environmental Protection Agency (USEPA) (2006) and Schijven et al. (2011) |
Vcons, dom | Consumption of water during domestic activities | mL/day | Uniform | Min = 1, max = 10 | Ranges of values reported from 1 to 10 mL per person per day for Africa | Steyn et al. (2004), Katukiza et al. (2013a) and Yapo et al. (2014) |
Vcons, RW, s | Consumption RW, swimming | mL/event | Lognormal | μ = 2.92, σ = 1.43 | Mean = 55 mL/event | Dufour et al. (2006) and US Environmental Protection Agency (USEPA) (2014) |
nRW, s | Number of recreation (swimming) events per year | Events/yr | Negative binomial | r = 1, λ = 0.04 | Mean = 24 events/yr | Schets et al. (2011) |
Vcons, RW, p | Consumption RW, children playing | mL/hr | Triangle | Min = 0.1, mode = 2.5, max = 11.2 | Mean = 2.1 mL/hr | Dorevitch et al. (2011) and Sales-Ortells & Medema (2014) |
tRW, p | Time spent playing in or near water | Min/day | Lognormal | μ = 4.1, σ =0.80 | Mean = 82 min/day | Schets et al. (2011) and Sales-Ortells & Medema (2014) |
nRW, p | Number of days per year where child plays by water | Days/yr | Triangle | Min = 1, mode = 12, max = 95 | Mean = 15 days/yr | Donovan et al. (2008) and Sales-Ortells & Medema (2014) |
Vcons, RW, b | Consumption RW, boating | mL/hr | Triangle | Min = 0.1, mode = 3.9, max = 11.8 | Mean = 1.9 ml/hr; based on canoeing | Dorevitch et al. (2011) and Sales-Ortells & Medema (2014) |
tRW, b | Time spent boating | Hr/day | Triangle | Min = 1, mode = 2, max = 4 | Mean = 2.1 hr/day | Sales-Ortells & Medema (2014) |
nRW, b | Numbe of recreation (boating) days per year | Days/yr | Uniform | Min = 1, max = 108 | Boating can occur from 1 to 108 days per year | Sales-Ortells & Medema (2014) |
Variable . | Definition . | Units . | Distribution . | Value . | Description . | Reference . |
---|---|---|---|---|---|---|
CNoV GI | NoV GI Concentration | gc/L | Lognormal | μ = 5.650, σ = 0.7158 | Mean = 360 gc/L | Collected & analysed data (see Table 3 for raw data) |
CNoV GII | NoV GII Concentration | μ = 7.203, σ = 0.7412 | Mean = 1,780 gc/L | |||
Rlow | % recovery | % | Uniform | Min = 0.01, max = 3.80 | Tap & raw water | Ruhanya (2013) |
Rhigh | % recovery | % | Uniform | Min = 7, max = 60 | Range across 3 water sources | Lambertini et al. (2008) |
Inact | Inactivation by sunlight | Log reduction | Uniform | Min = 1, max = 3 | NTU = 40 to 80 | Best estimate; based on Flannery et al. (2013) and Lee & Ko (2013) |
Inf | % infectious (viable) | % | n/a | 100 | Conservative assumption | Best estimate |
Vcons, DW | Consumption DW | L/day | Lognormal | μ = −0.036, σ = 0.772 | Mean = 1.3 L per person per day | US Environmental Protection Agency (USEPA) (2006) and Schijven et al. (2011) |
Vcons, dom | Consumption of water during domestic activities | mL/day | Uniform | Min = 1, max = 10 | Ranges of values reported from 1 to 10 mL per person per day for Africa | Steyn et al. (2004), Katukiza et al. (2013a) and Yapo et al. (2014) |
Vcons, RW, s | Consumption RW, swimming | mL/event | Lognormal | μ = 2.92, σ = 1.43 | Mean = 55 mL/event | Dufour et al. (2006) and US Environmental Protection Agency (USEPA) (2014) |
nRW, s | Number of recreation (swimming) events per year | Events/yr | Negative binomial | r = 1, λ = 0.04 | Mean = 24 events/yr | Schets et al. (2011) |
Vcons, RW, p | Consumption RW, children playing | mL/hr | Triangle | Min = 0.1, mode = 2.5, max = 11.2 | Mean = 2.1 mL/hr | Dorevitch et al. (2011) and Sales-Ortells & Medema (2014) |
tRW, p | Time spent playing in or near water | Min/day | Lognormal | μ = 4.1, σ =0.80 | Mean = 82 min/day | Schets et al. (2011) and Sales-Ortells & Medema (2014) |
nRW, p | Number of days per year where child plays by water | Days/yr | Triangle | Min = 1, mode = 12, max = 95 | Mean = 15 days/yr | Donovan et al. (2008) and Sales-Ortells & Medema (2014) |
Vcons, RW, b | Consumption RW, boating | mL/hr | Triangle | Min = 0.1, mode = 3.9, max = 11.8 | Mean = 1.9 ml/hr; based on canoeing | Dorevitch et al. (2011) and Sales-Ortells & Medema (2014) |
tRW, b | Time spent boating | Hr/day | Triangle | Min = 1, mode = 2, max = 4 | Mean = 2.1 hr/day | Sales-Ortells & Medema (2014) |
nRW, b | Numbe of recreation (boating) days per year | Days/yr | Uniform | Min = 1, max = 108 | Boating can occur from 1 to 108 days per year | Sales-Ortells & Medema (2014) |
Legend: gc, genome copies; NTU, nephelometric turbidity unit; DW, drinking water; dom, domestic water use; RW, recreational water; s, swimming; p, playing by riverside; b, boating; min, minimum; max, maximum.
Based on US data the volume of water consumed for drinking water (Vcons,DW) was estimated to be approximately 1.3 L per person per day (US Environmental Protection Agency (USEPA) 2006; Schijven et al. 2011). For domestic activities, the volume of water consumed (Vcons, dom) was estimated to be in the range of 1–10 mL per person per day for Africa (Steyn et al. 2004; Katukiza et al. 2013a; Yapo et al. 2014). The mean volume of water consumed during recreational activities (Vcons, RW, x) depended on the activity (s = swimming, p = playing, b = boating) ranging from approximately 2–55 mL per person per event while the range for the exposure events was from 1 to 108 exposure events (days) per year (nRW, x) (Donovan et al. 2008; Dorevitch et al. 2011; Schets et al. 2011; Sales-Ortells & Medema 2014; US Environmental Protection Agency (USEPA) 2014).
Dose-response assessment
A variety of NoV dose-response models estimating the probability of infection have been proposed for the secretor positive fraction of the population, i.e. those that are capable of binding NoV and becoming infected (Lindesmith et al. 2003; Hutson et al. 2005). In addition, these models make a variety of assumptions about the aggregation of the dose (Teunis et al. 2008; Messner et al. 2014; Schmidt 2015). In this QMRA, to account for uncertainty in the predicted outcomes, multiple NoV dose-response models were used because they represent the extreme predicted values (Van Abel et al. 2017). The four models used (Table 2) include two disaggregated models, the 1F1 hypergeometric (1F1) and the 2F1 hypergeometric with immunity (2F1i) with an ‘a’ approaching 0, and 2 aggregated dose-response models, the 2F1 hypergeometric (2F1) and the fractional Poisson (FP). Aggregated and disaggregated models were selected to take both a mechanistic view, the published models describe the mechanisms of norovirus dose-response and can be decomposed so a → 0 can be assumed, and an empirical view, where similarity between the environmental conditions and the dose-response conditions is assumed so a = 0.9997 or μa = 1,106. These two views, as well as the dose-response parameters, have been described in depth previously (Van Abel et al. 2017) and are briefly defined as follows: (α, β) are shape parameters of the beta distribution that describe variation in host susceptibility, ‘a’ is the aggregation parameter (0 ≤ a < 1) where a → 0 assumes disaggregation, μa is the mean number of genomes per aggregate, P is the fraction of individuals who are secretor positive that are fully susceptible, and φ is the fraction of secretor positive individuals who are immune to norovirus (Teunis et al. 2008; Messner et al. 2014; Schmidt 2015). It should be noted that these dose-response models relate to the mean ingested dose to the outcome of probability of infection (Haas 2002; Haas et al. 2014). Also, harmonisation of the PCR procedures between the environmental and human challenge data (McBride et al. 2013) was not completed because the environmental data included two genogroups and the human challenge data was compiled from four studies with two genogroups; thus, estimating one adjustment factor to harmonise all the data was not judicious.
Dose-response models and parameters
DR model . | Equation . | Parameters . | Reference . |
---|---|---|---|
Disaggregated models | |||
1F1 | ![]() | α = 0.04, β = 0.055 | Teunis et al. (2008) |
2F1i | ![]() | α = 2.91, β = 2,734 φ = 0.2754 a → 0 | Schmidt (2015) |
![]() | |||
Aggregated models | |||
2F1 | ![]() | α = 0.04, β = 0.055, a = 0.9997 | Teunis et al. (2008) |
FP | ![]() | P = 0.72, μa = 1,106 | Messner et al. (2014) |
DR model . | Equation . | Parameters . | Reference . |
---|---|---|---|
Disaggregated models | |||
1F1 | ![]() | α = 0.04, β = 0.055 | Teunis et al. (2008) |
2F1i | ![]() | α = 2.91, β = 2,734 φ = 0.2754 a → 0 | Schmidt (2015) |
![]() | |||
Aggregated models | |||
2F1 | ![]() | α = 0.04, β = 0.055, a = 0.9997 | Teunis et al. (2008) |
FP | ![]() | P = 0.72, μa = 1,106 | Messner et al. (2014) |
1F1, 1F1 hypergeometric; 2F1i, 2F1 hypergeometric with immunity; 2F1, 2F1 hypergeometric; FP, fractional Poisson.
(α, β) are shape parameters of the beta distribution that describe variation in host susceptibility, φ is the fraction of secretor positive individuals who are immune to NoV, ‘a’ is the aggregation parameter (0 ≤ a < 1), P is the fraction of individuals who are secretor positive that are fully susceptible to NoV infection, and μa is the mean number of genomes per aggregate.
The probability of illness was estimated based on the conditional likelihood that the person will become ill if infected (Pill|inf) and was calculated by Pill = Pill|inf X Pinf. The probability of becoming ill once infected (Pill|inf) was defined by a triangular distribution (minimum = 0.3, maximum = 1, mode = 0.6) based on the number of illnesses observed from the infected individuals in a human challenge study (Teunis et al. 2008).
Risk characterisation
The risk characterisation determined if any exposure scenarios exceeded the allowable health benchmarks. For drinking and domestic water exposures, the daily allowable risk of less than approximately one in 1,000,000 infections was used (Signor & Ashbolt 2009). For recreational (bathing) water, the annual illness benchmark of less than three illnesses per 100 events was used, which is the revised benchmark that excludes fever as a symptom of highly credible gastrointestinal illness (EU Directive 2006; Soller et al. 2010). The risk characterisation also described the impact of risk based on socioeconomic status.
RESULTS
Virus concentration
Norovirus concentrations could be estimated in 50% (17/34) NoV GI and 74% (25/34) NoV GII positive water samples, which had a detection Ct < 40. The concentration in the surface waters ranged from approximately 80 to 1,900 gc/L for GI and from 420 to 9,760 gc/L for GII (Table 3). The turbidity as reported by Rand Water ranged from 40 to 80 NTU while the recorded temperature of the water ranged from 8 to 18 °C and the pH ranged from ∼7 to 8.
Norovirus GI and GII concentrations in selected surface water in Gauteng, South Africa
Sample . | NoV GI . | NoV GII . | |||||||
---|---|---|---|---|---|---|---|---|---|
Date NoV detected . | River . | Ct value . | Raw concentration (gc/L) . | Extraction efficiency (%) . | Final concentration (gc/L) . | Ct value . | Raw concentration (gc/L) . | Extraction efficiency (%) . | Final concentration (gc/L) . |
2011/02/28 | RV | 33.99 | 2.94 × 102 | 110 | 2.94 × 102 | ||||
2011/05/23 | KR | 35.95 | 6.92 × 102 | 54 | 1.28 × 103 | ||||
2011/05/30 | RV | 34.41 | 2.30 × 102 | 43 | 5.31 × 102 | 35.73 | 7.50 × 102 | 43 | 1.73 × 103 |
2011/06/27 | SR | 37.94 | 4.20 × 102a | 325 | 4.20 × 102 | ||||
2011/07/04 | KR | 34.63 | 1.20 × 103 | 320 | 1.20 × 103 | ||||
2011/07/25 | RV | 34.66 | 1.98 × 102a | 126b | 1.98 × 102 | 33.45 | 2.24 × 103 | 126 | 2.24 × 103 |
2011/08/22 | RV | 33.51 | 3.96 × 102 | 266 | 3.96 × 102 | 35.21 | 9.20 × 102 | 266 | 9.20 × 102 |
2011/08/29 | KR | 36.24 | 8.32 × 101a | 563 | 8.32 × 101 | 35.52 | 8.10 × 102 | 563 | 8.10 × 102 |
2011/09/19 | RV | 33.98 | 2.96 × 102 | 119 | 2.96 × 102 | 34.15 | 1.52 × 103 | 119 | 1.52 × 103 |
2011/10/17 | RV | 34.68 | 1.96 × 102a | 181 | 1.96 × 102 | 34.62 | 1.20 × 103 | 181 | 1.20 × 103 |
2011/10/24 | KR | >40 | |||||||
2011/11/14 | RV | 32.34 | 8.42 × 102 | 100c | 8.42 × 102 | 35.07 | 9.78 × 102 | 100c | 9.78 × 102 |
2011/12/05 | RV | 32.87 | 5.94 × 102 | 31 | 1.91 × 103 | ||||
2012/01/30 | KR | 35.2 | 1.46 × 102a | 283 | 1.46 × 102 | 34.96 | 1.02 × 103 | 283 | 1.02 × 103 |
2012/02/20 | RV | 34.13 | 2.72 × 102 | 79 | 3.45 × 102 | ||||
2012/02/27 | KR | >40 | |||||||
2012/03/19 | RV | 35.17 | 1.48 × 102a | 75 | 1.96 × 102 | ||||
2012/05/21 | RV | 35.59 | 7.90 × 102 | 191 | 7.90 × 102 | ||||
2012/06/25 | KR | 35.41 | 8.48 × 102 | 155 | 8.48 × 102 | ||||
2012/07/02 | KR | 36.98 | 5.08 × 102a | 81 | 6.24 × 102 | ||||
2012/07/30 | KR | 37.05 | 5.00 × 102a | 349 | 5.00 × 102 | ||||
2012/08/20 | RV | 33.48 | 2.22 × 103 | 83 | 2.67 × 103 | ||||
2013/02/18 | RV | 35.41 | 1.33 × 102 | 60 | 2.23 × 102 | 36.12 | 6.54 × 102 | 60 | 1.10 × 103 |
2013/02/25 | KR | 37.94 | 4.20 × 102a | 334 | 4.20 × 102 | ||||
2013/03/18 | RV | >40 | |||||||
2013/04/15 | RV | >40 | |||||||
2013/05/13 | RV | 34.38 | 1.35 × 103 | 270 | 1.35 × 103 | ||||
2013/06/10 | RV | 34.25 | 1.44 × 103 | 121 | 1.44 × 103 | ||||
2013/09/16 | RV | 35.89 | 1.36 × 103 | 14 | 9.76 × 103 | ||||
2013/10/21 | KR | 34.89 | 1.78 × 102 | 235 | 1.78 × 102 | ||||
2013/11/11 | RV | 36.02 | 9.60 × 101 | 63 | 1.52 × 102 | 34.83 | 2.10 × 103 | 63 | 3.33 × 103 |
2014/02/17 | RV | 33.34 | 4.52 × 102 | 147 | 4.52 × 102 | 35.15 | 1.81 × 103 | 147 | 1.81 × 103 |
2014/03/17 | RV | >40 | 36.18 | 1.23 × 103 | 21 | 5.85 × 103 | |||
2014/04/14 | RV | 34.73 | 1.95 × 102 | 176 | 1.95 × 102 | 34.29 | 2.72 × 103 | 176 | 2.72 × 103 |
Sample . | NoV GI . | NoV GII . | |||||||
---|---|---|---|---|---|---|---|---|---|
Date NoV detected . | River . | Ct value . | Raw concentration (gc/L) . | Extraction efficiency (%) . | Final concentration (gc/L) . | Ct value . | Raw concentration (gc/L) . | Extraction efficiency (%) . | Final concentration (gc/L) . |
2011/02/28 | RV | 33.99 | 2.94 × 102 | 110 | 2.94 × 102 | ||||
2011/05/23 | KR | 35.95 | 6.92 × 102 | 54 | 1.28 × 103 | ||||
2011/05/30 | RV | 34.41 | 2.30 × 102 | 43 | 5.31 × 102 | 35.73 | 7.50 × 102 | 43 | 1.73 × 103 |
2011/06/27 | SR | 37.94 | 4.20 × 102a | 325 | 4.20 × 102 | ||||
2011/07/04 | KR | 34.63 | 1.20 × 103 | 320 | 1.20 × 103 | ||||
2011/07/25 | RV | 34.66 | 1.98 × 102a | 126b | 1.98 × 102 | 33.45 | 2.24 × 103 | 126 | 2.24 × 103 |
2011/08/22 | RV | 33.51 | 3.96 × 102 | 266 | 3.96 × 102 | 35.21 | 9.20 × 102 | 266 | 9.20 × 102 |
2011/08/29 | KR | 36.24 | 8.32 × 101a | 563 | 8.32 × 101 | 35.52 | 8.10 × 102 | 563 | 8.10 × 102 |
2011/09/19 | RV | 33.98 | 2.96 × 102 | 119 | 2.96 × 102 | 34.15 | 1.52 × 103 | 119 | 1.52 × 103 |
2011/10/17 | RV | 34.68 | 1.96 × 102a | 181 | 1.96 × 102 | 34.62 | 1.20 × 103 | 181 | 1.20 × 103 |
2011/10/24 | KR | >40 | |||||||
2011/11/14 | RV | 32.34 | 8.42 × 102 | 100c | 8.42 × 102 | 35.07 | 9.78 × 102 | 100c | 9.78 × 102 |
2011/12/05 | RV | 32.87 | 5.94 × 102 | 31 | 1.91 × 103 | ||||
2012/01/30 | KR | 35.2 | 1.46 × 102a | 283 | 1.46 × 102 | 34.96 | 1.02 × 103 | 283 | 1.02 × 103 |
2012/02/20 | RV | 34.13 | 2.72 × 102 | 79 | 3.45 × 102 | ||||
2012/02/27 | KR | >40 | |||||||
2012/03/19 | RV | 35.17 | 1.48 × 102a | 75 | 1.96 × 102 | ||||
2012/05/21 | RV | 35.59 | 7.90 × 102 | 191 | 7.90 × 102 | ||||
2012/06/25 | KR | 35.41 | 8.48 × 102 | 155 | 8.48 × 102 | ||||
2012/07/02 | KR | 36.98 | 5.08 × 102a | 81 | 6.24 × 102 | ||||
2012/07/30 | KR | 37.05 | 5.00 × 102a | 349 | 5.00 × 102 | ||||
2012/08/20 | RV | 33.48 | 2.22 × 103 | 83 | 2.67 × 103 | ||||
2013/02/18 | RV | 35.41 | 1.33 × 102 | 60 | 2.23 × 102 | 36.12 | 6.54 × 102 | 60 | 1.10 × 103 |
2013/02/25 | KR | 37.94 | 4.20 × 102a | 334 | 4.20 × 102 | ||||
2013/03/18 | RV | >40 | |||||||
2013/04/15 | RV | >40 | |||||||
2013/05/13 | RV | 34.38 | 1.35 × 103 | 270 | 1.35 × 103 | ||||
2013/06/10 | RV | 34.25 | 1.44 × 103 | 121 | 1.44 × 103 | ||||
2013/09/16 | RV | 35.89 | 1.36 × 103 | 14 | 9.76 × 103 | ||||
2013/10/21 | KR | 34.89 | 1.78 × 102 | 235 | 1.78 × 102 | ||||
2013/11/11 | RV | 36.02 | 9.60 × 101 | 63 | 1.52 × 102 | 34.83 | 2.10 × 103 | 63 | 3.33 × 103 |
2014/02/17 | RV | 33.34 | 4.52 × 102 | 147 | 4.52 × 102 | 35.15 | 1.81 × 103 | 147 | 1.81 × 103 |
2014/03/17 | RV | >40 | 36.18 | 1.23 × 103 | 21 | 5.85 × 103 | |||
2014/04/14 | RV | 34.73 | 1.95 × 102 | 176 | 1.95 × 102 | 34.29 | 2.72 × 103 | 176 | 2.72 × 103 |
KR, Klip River; RV, Rietspruit River; SR, Suikerbosrant River.
aThese values were outside the range of the standard curve and were extrapolated.
bExtraction efficiencies exceeding 100% are ascribed to clumping of mengovirus stocks.
cThe extraction efficiency was not estimated for this sample and was assumed to be 100%.
Risk estimates from the QMRA
A difference in the daily probabilities of infection was observed depending on the type of exposure, NoV genogroup, and dose-response model used (Figure 1). Daily risk estimates from all four dose-response models were lower for recreational water (RW) exposure (min–max: 5.29 × 10−7 to 6.42 × 10−1) in contrast to drinking water (DW) exposure (min–max: 2.33 × 10−5 to 7.24 × 10−1). Also, daily risk estimates using all four dose-response models were lower for GI (min–max: 2.91 × 10−8 to 7.02 × 10−1) versus GII (min–max: 1.19 × 10−7 to 7.24 × 10−1). Daily risk estimates were highest for the 1F1 hypergeometric dose-response model when the mean dose was less than approximately 2,000 gc.
Comparison of probability of infection (Pinf) values for exposure to drinking water (DW) and recreational (swimming) water (RW): (a) Pinf for DW exposure to norovirus GI; (b) Pinf for RW exposure to norovirus GI; (c) Pinf for DW exposure to norovirus GII; (d) Pinf for RW exposure to norovirus GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson.
Comparison of probability of infection (Pinf) values for exposure to drinking water (DW) and recreational (swimming) water (RW): (a) Pinf for DW exposure to norovirus GI; (b) Pinf for RW exposure to norovirus GI; (c) Pinf for DW exposure to norovirus GII; (d) Pinf for RW exposure to norovirus GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson.
A difference in the estimated daily probabilities of infection was observed for the drinking water exposure for GII versus GI (Figure 2) as well as for the different dose-response models. In general, the estimated probability of infection was higher for GII (min–max: 2.33 × 10−5 to 7.24 × 10−1) as opposed to GI (min–max: 1.04 × 10−5 to 7.02 × 10−1) and was highest for the 1F1 hypergeometric dose-response model. All exposures, no matter which genogroup or dose-response model, exceeded the allowable daily risk of infection of <1 in 106 infections.
Daily probability of infection values for drinking water (DW) exposure to norovirus GI and GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson; Box = Interquartile range (IQR) (upper = 75%, middle = median, lower = 25%), Whiskers = 1.5*IQR, Points = outliers (i.e. distant from rest of data).
Daily probability of infection values for drinking water (DW) exposure to norovirus GI and GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson; Box = Interquartile range (IQR) (upper = 75%, middle = median, lower = 25%), Whiskers = 1.5*IQR, Points = outliers (i.e. distant from rest of data).
The estimated probabilities of infection were lower for domestic water exposure (Figure 3) than for drinking water. Similar to the drinking water exposure, the largest estimates were for GII (min–max: 1.19 × 10−7 to 5.19 × 10−1) and for the 1F1 hypergeometric dose-response model. The exceedances for GI were slightly lower for this exposure scenario at 94% for the fractional Poisson, 96% for the 2F1 with immunity, 98% for the 2F1 hypergeometric, and 100% for the 1F1 hypergeometric. The exceedances for GII were ∼100% for all dose-response models.
Daily probability of infection values for domestic water exposure to norovirus GI and GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson; Box = Interquartile range (IQR) (upper = 75%, middle = median, lower = 25%), Whiskers = 1.5*IQR, Points = outliers (i.e. distant from rest of data).
Daily probability of infection values for domestic water exposure to norovirus GI and GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson; Box = Interquartile range (IQR) (upper = 75%, middle = median, lower = 25%), Whiskers = 1.5*IQR, Points = outliers (i.e. distant from rest of data).
To compare to an annual risk of illness benchmark for recreational water, the probability of illness was estimated. The annual probabilities of illness for recreational exposures varied depending on the type of exposure with the highest risks being from boating (min–max: 3.91 × 10−6 to 5.43 × 10−1) and swimming (min–max: 6.20 × 10−6 to 6.42 × 10−1) while lower risks were observed from playing by the riverside (min–max: 5.30 × 10−7 to 5.48 × 10−1). Again, a difference was observed between the different dose-response models (Figure 4). Also, the risks were higher for GII as opposed to GI. The exceedances summarised in Table 4 varied greatly depending on the genogroup and dose-response model selected with GII having more exceedances as compared to GI and 1F1 hypergeometric consistently predicting the most exceedances.
Exceedances (%) for recreational water exposures for norovirus GI and norovirus GII from all dose-response models
Norovirus genogroup . | Swimming . | Playing . | Boating . | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
1F1 . | 2F1i . | 2F1 . | FP . | 1F1 . | 2F1i . | 2F1 . | FP . | 1F1 . | 2F1i . | 2F1 . | FP . | |
GI | 96 | 49 | 58 | 46 | 97 | 5 | 9 | 4 | 100 | 38 | 50 | 34 |
GII | 96 | 76 | 82 | 74 | 100 | 24 | 33 | 21 | 100 | 74 | 82 | 71 |
Norovirus genogroup . | Swimming . | Playing . | Boating . | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
1F1 . | 2F1i . | 2F1 . | FP . | 1F1 . | 2F1i . | 2F1 . | FP . | 1F1 . | 2F1i . | 2F1 . | FP . | |
GI | 96 | 49 | 58 | 46 | 97 | 5 | 9 | 4 | 100 | 38 | 50 | 34 |
GII | 96 | 76 | 82 | 74 | 100 | 24 | 33 | 21 | 100 | 74 | 82 | 71 |
1F1, 1F1 hypergeometric; 2F1i, 2F1 hypergeometric with immunity; 2F1, 2F1 hypergeometric; FP, fractional Poisson.
Annual probability of illness values for all recreational water (RW) exposures to norovirus: (a) GI and (b) GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson, BM = benchmark, which is the recreational water illness benchmark. Box = Interquartile range (IQR) (upper = 75%, middle = median, lower = 25%), Whiskers = 1.5*IQR, Points = outliers (i.e. distant from rest of data).
Annual probability of illness values for all recreational water (RW) exposures to norovirus: (a) GI and (b) GII. 1F1 = 1F1 hypergeometric, 2F1i = 2F1 hypergeometric with immunity, 2F1 = 2F1 hypergeometric, FP = fractional Poisson, BM = benchmark, which is the recreational water illness benchmark. Box = Interquartile range (IQR) (upper = 75%, middle = median, lower = 25%), Whiskers = 1.5*IQR, Points = outliers (i.e. distant from rest of data).
In general, the percentage of exceedances for the corresponding benchmark was greatest for drinking water exposures followed by domestic water exposures followed by the recreational water exposures.
DISCUSSION
The use of surface water for recreation, drinking, and domestic purposes is associated with potential public health risks. This study provided a unique opportunity to investigate the health risks posed by exposure to selected water sources in SA because the concentration of NoV was quantified for the QMRA. Moreover, the concentrations of NoV GI and GII were estimated separately. While the samples had been frozen for a number of years, GI and GII genogroups from 50 and 74% positive samples, respectively, were quantified. A review of the data indicated that the time frozen did not significantly impact the NoV concentration results presented in this paper (data not shown). This was advantageous as many QMRAs do not have quantification data for viruses but extrapolate the concentration from indicator and surrogate data (Howard et al. 2006; Seidu et al. 2008; Labite et al. 2010; Machdar et al. 2013; Petterson et al. 2016). This is especially true in developing countries where few data exist on the concentration of pathogens in water sources (Enger et al. 2012).
The estimated concentration of NoV GI and GII in these SA surface water samples (8 × 101 to 9.76 × 103 gc/L: Table 3) was similar to reported viral concentrations in water sources in sub-Saharan Africa. For open storm drains in Uganda, the estimated concentration of human adenovirus (HAdV) F & G ranged from 1.35 × 102 (±1.9 × 102) to 2.65 × 104 (±1.9 × 104) gc/L and for rotavirus (RV) from 3.44 × 102 (±4.86 × 102) to 2.98 × 104 (±3.66 × 104) gc/L, while unprotected spring water had an estimated concentration of HAdV of 7.64 × 100 (±1 × 101) gc/L (Katukiza et al. 2013a, 2013b). In SA, water samples from the Buffalo River had estimated corrected viral concentrations in gc/L ranging from 3.11 × 101 to 2.48 × 103 for HAdV, 2.25 × 103 to 9.68 × 104 for hepatitis A virus (HAV), 7.93 × 102 to 2.79 × 103 for RV, and 2.79 × 101 to 1.25 × 102 for enterovirus (EV) (Chigor et al. 2014), while in Accra, Ghana, the concentration of NoV in stream water, drains, and waste stabilisation pond influent and effluent ranged from <3.13 × 103 to 1.58 ± 0.28 × 105 gc/L (Silverman et al. 2013). The concentration of NoV was, however, lower than the concentration of sapoviruses found in the water samples from the same SA source where average monthly samples ranging from 4.24 × 106 to 6.04 × 108 gc/L (Murray et al. 2013b). The concentration of NoV GI and GII in these surface water samples was considered to be clinically relevant because the median infectious dose for NoV has been estimated at approximately 26–2,800 gc (Teunis et al. 2008; Atmar et al. 2011). The observed higher prevalence of NoV GII compared to NoV GI in the water sources was similar to previous studies in SA where GII was found in 43% and GI in 14% surface water samples from the same sites, respectively (Mans et al. 2013). Similarly, the prevalence of NoV GII was 63% while NoV GI was 29% in wastewater samples from SA (Murray et al. 2013a). The difference in NoV GI and NoV GII prevalence in surface water sources was not unusual and has been reported in recreational water from nine countries in Europe (Wyn-Jones et al. 2011), in surface water in Singapore (Aw & Gin 2011), in urban surface water in Kenya (Kiulia et al. 2014) and river water samples in Japan (Kishida et al. 2012).
The daily risk of infection from the consumption of untreated surface water as DW ranges from 2.3 × 10−5 to 6.4 × 10−1 for NoV GI and for GII, from 7.3 × 10−5 to 7.2 × 10−1. At these high probability of infection values, the daily health benchmark of <1 infection in 106 individuals is exceeded 100% of the time (Figure 2). The daily risk of infection from domestic water use is less (range GI: 1.3 × 10−7 to 4.8 × 10−1, GII: 3.6 × 10−7 to 5.1 × 10−1). The daily health benchmark for domestic water was however still exceeded 97% of the time for GI and 100% of the time for GII when all dose-response models were used to quantify the risk. The daily risk of infection, which was higher for NoV GII when compared to NoV GI, was not unexpected because the concentration and prevalence of GII was higher than for GI in the surface water samples. In addition, the daily risk of infection was higher when using the water for drinking water rather than for domestic use because the volume of water consumed was orders of magnitude higher for drinking (1.3 L) than domestic water (1–10 mL) use. However, even with the small volume of water consumed during domestic exposures, the daily risk of infection still consistently exceeds the benchmark. It should be noted that protective measures are not often taken during domestic exposures, which could contribute to the higher risks from this pathway (Yapo et al. 2014). The risk of infection after drinking and domestic water use exposures are similar to other QMRA results from sub-Saharan Africa. In SA, the consumption of untreated drinking water stored in containers led to an unacceptable risk of infection from Escherichia coli and Salmonella spp. (Steyn et al. 2004), while an individual consuming just 100 mL of untreated contaminated surface water had up to a 26% chance of falling ill from seven enteric pathogens (Le Roux et al. 2012). Similar risk results, albeit slightly higher for the maximum risk, were estimated by a study from Ghana that found for RV the risk from drinking water ranged from 1.5 × 10−5 to 3.3 × 10−1 (Machdar et al. 2013). In Cote d'Ivoire collecting and washing plastic bags in wastewater canals for domestic water use, a common practice in this country, was associated with the highest risk of infection (Yapo et al. 2014).
The risk from recreational exposure was dependent on the type of exposure (Figure 4), with playing by the riverside demonstrating the least percentage of exceedances of the annual recreational water illness benchmark of 0.03 (GI: 4–97%, GII: 21–100%) and swimming yielding the highest risk and exceedances (GI: 46–96%, GII: 74–96%) (Figure 4). Variable results were reported from other QMRAs depending on exposure and water source. In SA, recreational exposure to HAdV-contaminated water was within the acceptable USEPA limits (van Heerden et al. 2005), but exposure to HAV was not (Venter et al. 2007). In a later QMRA study from SA, that did not specify the recreational exposures, benchmark exceedances for exposure to dam water contaminated with viruses were reported, while the non-dam water exposures were within acceptable limits (Chigor et al. 2014). A QMRA in Ghana that found the major transmission route for waterborne diseases was the ingestion of contaminated water by children playing by roadside drains (Labite et al. 2010), while in Uganda incidental ingestion exposure to contaminated surface water from open drainage channels contributed the highest disease burden in an urban slum (Katukiza et al. 2013a). The differences in QMRA results found to be within acceptable benchmark values could be due to different assumptions made including the volume of water ingested, the number of events per year, and the choice of dose-response model.
An important result from this study was that the recreational exposure risk was dependent on the dose-response model used. This was previously observed in a deterministic QMRA of drinking and water exposures that compared all available NoV dose-response models published in the literature (Van Abel et al. 2017). For recreational water exposures, where the intake doses are lower than for consumption of untreated drinking water, the risk of illness estimates from the four dose-response models was disparate (Figure 4). The 1F1 hypergeometric model consistently predicted the largest probability of infection values leading to the highest percentages of exceedances at >95% (Table 4). The disparity was most extreme for playing by the river where the risk from GI ranges from ∼3 to 100% depending on the dose-response model used. This a consideration because the 1F1 hypergeometric function, which leads to the largest risk estimates, is the dose-response model most commonly used in QMRA (Van Abel et al. 2017). Use of only this dose-response model in QMRA would be extremely protective of human health, but may also be overestimating the risk burden and could lead to different risk management strategies than if another model was selected for use in the QMRA. The recommendation to use multiple dose-response models in NoV QMRA, even though the risk estimates predicted by the four dose-response models are disparate (∼6 orders of magnitude), ensures the complete picture of risk is described including the conservative estimates that may be more protective of public health. Without more information on which norovirus dose-response model is ‘best’ then no one model can be put forward, thus the recommendation is to select multiple appropriate NoV dose-response models considering aggregation, genogroup, and secretor status for each NoV QMRA (Van Abel et al. 2017).
The calculated risk of infection for the drinking and domestic use of NoV-contaminated water places the greatest risk burden on individuals of LSES, as these individuals may have regular access to safe water in their community, but intermittent interruptions in the water supply may result in communities using surface water for drinking or domestic purposes. Even a few days of interrupted water supply has been shown to destroy the health benefits of treated water (Hunter et al. 2009). In contrast, recreational exposure to contaminated surface water can impact both LSES and USES individuals. Both LSES and USES individuals could swim in the surface water, which places both of these groups at increased risk of NoV infection or illness. The risk from playing next to the river, mainly associated with LSES, had the lowest risks. Boating was also associated with an increased risk of illness, which would probably only impact the USES individuals. Overall, the risk burden falls more heavily on LSES than USES which confirms QMRA results for HAV infection in individuals from the same geographical area (Venter et al. 2007). In addition to SES, vulnerable populations must be addressed. Susceptible individuals who are at a higher risk of (NoV) infection include the very young, the elderly, the malnourished, pregnant women and immunocompromised individuals, including human immunodeficiency virus/acquired immunodeficiency patients (HIV/AIDS) who may present with chronic NoV infection due to supressed immune systems (Gerba et al. 1996; Wingfield et al. 2010; Haas et al. 2014). The high prevalence of HIV/AIDS in SA predisposes a significant portion of the population to chronic NoV infection or illness, which could lead to a serious public health threat in this vulnerable population (Mans et al. 2013).
One of the most important findings of this QMRA study was that the choice of dose-response model was critical to the outcome of the QMRA, a finding that has been described previously (Teunis & Havelaar 2000). Therefore, to account for the uncertainty in NoV dose-response, multiple models (Van Abel et al. 2017) were used in the present study to formulate a complete picture of the risk. Of concern was that the four dose-response models were determined from predominantly NoV GI.1 data. Therefore, the assumption must be made that GII is as infectious as GI, which has not yet been demonstrated. Overall, more information on NoV dose-response for both GI and GII is needed. The lack of adequate information on the recovery of NoV is another limitation. This limitation has also been identified by others and a request for more pathogen recovery data was made (Vivier et al. 2002; de Man et al. 2014). For NoV, the recovery efficiencies estimated for glass-wool adsorption elution were very disparate. Ultimately, the recovery data from two studies (Lambertini et al. 2008; Ruhanya 2013) were used to address this uncertainty.
Further uncertainty in the risk estimates results from the lack of NoV infectivity information and this work included the assumption that all viruses are infectious. For the quantification of NoV in water, PCR was used, which is a method that cannot distinguish between infectious and non-infectious particles, but does demonstrate, at the very least, viral contamination (Vergara et al. 2016). However, until recently no cell culture method for NoV has repeatedly demonstrated success (Jones et al. 2015; Ettayebi et al. 2016) so an assumption about the percentage of infectious NoVs was made (Girones et al. 2010; World Health Organization 2011). This QMRA also assumed that the LSES individuals consumed the raw surface water without boiling or treating it in any manner prior to consumption. Others have reported that in-home treatment interventions for drinking water may occur, but the follow-up periods on the review of these techniques is short and the results incomplete (Hunter et al. 2009). Ultimately, it is unclear what water treatment measures would be taken and whether they would be carried out completely to ensure inactivation of the virus. Thus, the assumption was made that the water was consumed without treatment. The daily risk of infection from drinking or domestic water exposures is exceptionally high, leading to exceedances of the daily infection health benchmark. For recreational water exposures, the results for the annual risk of illness, as well as the percentage of exceedances, demonstrated huge variability. It should be noted that both health benchmarks were developed for use in the United States and the European Union, which may not be representative of a developing country like SA. Considering the limitations and assumptions, an overestimation of the drinking water risk may be demonstrated. However, although the results of this analysis indicate a high risk of NoV infection and illness from exposure to river water in SA, this is congruent with other QMRA results. Unacceptably high risks to water users in the Olifants River catchment in SA from exposure to seven pathogens was demonstrated and the greatest risk to water users was found to be exposure to NoV (Le Roux et al. 2012; Genthe et al. 2013).
CONCLUSIONS
The results of this assessment demonstrate that selected surface waters in SA are contaminated with NoV which can lead to unacceptable risks of infection from exposure via direct drinking of untreated surface water as well as domestic and recreational use. Furthermore, this work presented quantitative data on the concentration of NoV GI and GII in surface waters, which is useful for QMRA. The results of the QMRA found that: (1) a higher risk was observed for GII as compared to GI; (2) the daily probability of infection risk was higher for drinking water exposures as compared to domestic use of water; (3) the annual probability of illness from recreation exposures was highest for boating and swimming and lowest for playing by the river; and (4) the risk burden was greater for individuals of LSES who may use untreated surface water for multiple household and recreational uses. Finally, the choice of dose-response model was critical to the outcome of the QMRA, so multiple dose-response models must be used in NoV QMRA to account for uncertainty and to completely describe the risk.
ACKNOWLEDGEMENTS
The authors thank Rand Water for the collection of water samples and permission to publish these data. The assistance of Mr Johannes Christiaan (Ian) Botha with the norovirus quantification is acknowledged. Dr Nicole Van Abel acknowledges a postdoctoral fellowship from the University of Pretoria and Rand Water Chair in Public Health. This study was funded, in part, by the Rand Water Chair in Public Health.
Ethics approval: Approval for this study was obtained from the Faculty of Health Sciences Research Ethics Committee, University of Pretoria (Protocol 91/2015).
Competing interests: None.