Validation of the bag‐mediated filtration system for environmental surveillance of poliovirus in Nairobi, Kenya

Abstract Aims This study compared the bag‐mediated filtration system (BMFS) and standard WHO two‐phase separation methods for poliovirus (PV) environmental surveillance, examined factors impacting PV detection and monitored Sabin‐like (SL) PV type 2 presence with withdrawal of oral polio vaccine type 2 (OPV2) in April 2016. Methods and Results Environmental samples were collected in Nairobi, Kenya (Sept 2015–Feb 2017), concentrated via BMFS and two‐phase separation methods, then assayed using the WHO PV isolation algorithm and intratypic differentiation diagnostic screening kit. SL1, SL2 and SL3 were detected at higher rates in BMFS than two‐phase samples (P < 0·05). In BMFS samples, SL PV detection did not significantly differ with volume filtered, filtration time or filter shipment time (P > 0·05), while SL3 was detected less frequently with higher shipment temperatures (P = 0·027). SL2 was detected more frequently before OPV2 withdrawal in BMFS and two‐phase samples (P < 1 × 10−5). Conclusions Poliovirus was detected at higher rates with the BMFS, a method that includes a secondary concentration step, than using the standard WHO two‐phase method. SL2 disappearance from the environment was commensurate with OPV2 withdrawal. Significance and Impact of the Study The BMFS offers comparable or improved PV detection under the conditions in this study, relative to the two‐phase method.

Environmental surveillance for PV in Nairobi, Kenya began in October 2013 (Borus et al. 2015;WHO 2017). Kenya's final clinical WPV case occurred in July 2013, and the last detected WPV environmental sample was in October 2013 (Centers for Disease Control and Prevention (CDC) 2013; Kamadjeu et al. 2014;Borus et al. 2015;WHO 2015aWHO , 2017. In 2015, environmental surveillance expanded to Mombasa, Garissa and Kisumu (WHO 2017). Environmental surveillance in Kenya utilizes the standard WHO procedure (two-phase method): a 500 ml grab sample is concentrated by two-phase separation, for a 50-fold concentration factor and 10 ml final volume (WHO 2015a). This method has been used for over 30 years; nevertheless, the GPEI recommended evaluation of alternative environmental surveillance methods (P€ oyry et al. 1988;WHO 2015aWHO , 2015b. Consequently, the bag-mediated filtration system (BMFS) was developed to enable primary concentration of 3-6 l in the field, followed by secondary concentration in the laboratory. This method increased the concentration factor to 300-to 600-fold with a final volume of 10 ml (Fagnant et al. 2014WHO 2015a;Zhou et al. 2018). A previous study was conducted in Nairobi to identify and address complications from conducting a multi-national study, and compare PV detection between environmental samples concentrated by the two-phase method and BMFS, using a limited data set .
The objectives of the study described here were to (i) validate the BMFS for PV environmental surveillance with the two-phase method, (ii) examine sample processing factors that may impact PV detection and (iii) monitor environmental Sabin-like PV type 2 (SL2) presence before and after the withdrawal of OPV2.

Study design
From 29 September 2015 to 14 February 2017, samples were collected in Nairobi (n = 133) twice per month from four sites: Starehe, Eastleigh A, Eastleigh B and Kibera (described in Supporting Information). Single water samples were collected within 5 min and a 1-m radius of each other for parallel testing by the BMFS and two-phase concentration methods. Each collected BMFS water sample was concentrated using two ViroCap TM filters, resulting in two replicate BMFS samples for each BMFS sampling event.
Primary concentration for two-phase and replicate BMFS samples occurred at Kenya Medical Research Institute (KEMRI) in Nairobi throughout the study. Additional processing and analyses occurred at multiple locations (KEMRI, University of Pretoria (UP), and/or CDC) during this study. From 29 September 2015 to 15 February 2016 (Fig. 1a), replicate BMFS filters were treated with preservatives at KEMRI, shipped to UP in Pretoria, South Africa, for processing, and then to CDC in Atlanta, United States, where a randomized portion of BMFS samples was analysed. After two-phase separation was performed at KEMRI, all two-phase sample concentrates were shipped to CDC for analysis. On 16 February 2016, KEMRI personnel were trained to fully process BMFS samples by the University of Washington personnel to perform virus isolation on environmental samples by WHO-AFR personnel. From 16 February 2016 to 14 February 2017 (Fig. 1b), one BMFS filter was treated with preservatives at KEMRI, shipped to UP for processing and then to CDC where a randomized portion was analysed. The second BMFS filter received no preservative treatment and remained at KEMRI for processing and analysis. All BMFS samples remaining at KEMRI were analysed, and all two-phase samples were processed and analysed at KEMRI.

BMFS samples
Eight-litre samples were collected in a collection bag, then sealed and placed into a water-tight, insulated bucket, with cold packs, for transport to KEMRI within 4 h (i.e. bucket protocol) , and filtration within 24 h. Collection bags were hung on a tripod stand outside on KEMRI's campus, allowed to settle 15 min and approximately 0Á5 l was drained as waste to remove settled solids. A Y-adapter was connected to the bag's outlet, two replicate ViroCap filters preseeded with a known titre of bacteriophage MS2 as previously described (Zhou et al. 2020) were attached to either end and samples were filtered simultaneously by gravity. All two-phase and BMFS-1, and a randomized subset of BMFS-2 samples analyzed by tissue culture and ITD at CDC n = 33 two-phase n = 33 BMFS-1 n = 3 BMFS-2 Twice per month, co-located, sequentially collected two-phase and BMFS samples collected at 4 field sites in Nairobi by KEMRI n = 33 two-phase For filters shipped to UP, a 2% sodium benzoate (Becton Dickinson, Sparks, MD) and 0Á2% calcium propionate (Becton Dickinson) preservative mixture was passed through the filter at KEMRI (Fagnant et al. 2017a). All filters were processed by a single 30-min elution using 100 ml pH 9Á5 eluent containing 1Á5% beef extract (Becton Dickinson) and 0Á05 mol l À1 glycine (Fisher Scientific, Hampton, VA (KEMRI); Merck KGaA, Darmstadt, Germany (UP)) (Fagnant et al. 2017bZhou et al. 2018). Secondary concentration was performed on the eluate by polyethylene glycol (PEG) precipitation (Meleg et al. 2008;Kiulia et al. 2010), with addition of 14 g PEG 8000 (Sigma Aldrich (KEMRI); Amresco LLC, Solon, OH (UP)) 1Á17 g sodium chloride (NaCl) (Sigma Aldrich), overnight incubation (room temperature (KEMRI) or 4°C (UP)), and centrifugation (2500 g (KEMRI) or 6500 g (UP), 30 min). The pellet was resuspended in 10 ml PBS.
For BMFS samples processed at UP the following controls were included. Infectious MS2 preseeded onto the ViroCap filters as a BMFS process control, was enumerated in the filter eluate via the double agar layer method using an E. coli F À amp host as previously described (Adams 1959;US EPA 2000;Zhou et al. 2020). MS2 recovery efficiency ranged from 0 to 5900%, with a median of 9Á9%. Additionally, an aliquot of the resuspended secondary concentration pellet remained at UP. These samples were chloroform extracted, seeded with 5 9 10 4 copies of mengovirus as an extraction control and nucleic acid extracted via the semi-automated NucliSENSâ easyMAGâ instrument (bioM erieux, SA, Marcy-I' Etoile, France) (Zhou et al. 2020). The median extraction efficiency for mengovirus was 32Á04% (interquartile range = 20Á18-52Á61%). Real-time reverse transcription polymerase chain reaction (rRT-PCR) analysis using CeeramToolsâ (bioM erieux) showed that 98Á9% samples were positive for mengovirus.
The RNA from samples that tested negative for mengovirus was diluted 10-fold and all tests were repeated.

Two-phase samples
One-litre samples were collected, placed in a cooler with ice packs, transported to KEMRI within 4 h, and concentrated by two-phase separation within 48 h (WHO 2015a). A 500 ml aliquot was centrifuged to pellet debris and saved. The supernatant was combined with 287 ml 29% PEG 6000, 39Á5 ml 22% dextran T40 (Pharmacosmos, Holbaek, Denmark) and 35 ml 5 mol l À1 NaCl, and placed into a separation funnel at 4°C overnight. The lower-and inter-phases were collected, and the pellet was added to the concentrate. Secondary concentration was not performed, according to the standard WHO protocol (WHO 2015a).

Assay
Concentrates were chloroform extracted and assayed at KEMRI or CDC via the WHO Poliovirus Isolation Algorithm, utilizing L20B (mouse L cell expressing the PV receptor, CD155) and human rhabdomyosarcoma (RD) cell lines (WHO 2015a). Samples positive for cytopathic effects (CPE) were screened by rRT-PCR using a suite of assays included in the Poliovirus Intratypic Differentiation rRT-PCR Kit (CDC, Atlanta, USA) on an Applied Biosystemsâ 7500 thermocycler (Applied Biosystems, Foster City, CA), as previously described (Gerloff et al. 2018). Briefly, reaction cycling conditions included: reverse transcription (RT) at 50°C for 30 min, RT inactivation and initial denaturation at 95°C for 1 min, followed by 40 cycles of 95°C for 15 s, 50°C for 45 s and 72°C for 5 s with a 25% ramp rate between the annealing and elongation step. Results were reported following the PV diagnostic algorithm (Kilpatrick et al. 2009;WHO 2015a;Gerloff et al. 2018). Briefly, samples positive for CPE in RD cells, but negative in L20B cells are reported as nonpolio enterovirus (NPEV). Additionally, samples positive for CPE, but negative for the following assays (pan enterovirus (PanEV), Sabin-like PV type 1 (SL1), Sabin-like PV type 2 (SL2), Sabin-like PV type 3 (SL3), pan poliovirus (PanPV), WPV1, PV type 2, WPV3-I and WPV3-II) are reported as non-enterovirus. If the PanEV assay is positive and others negative, samples are reported as NPEV. Samples presumptively positive for Sabin-like PV are further assayed for VDPV type 1 or VDPV type 3, and sequenced if determined to be non-Sabin-like (NSL) or reported as Sabin-like. Any NSL, PV2 positive or indeterminate samples are sequenced for final confirmation (Gerloff et al. 2018).

Analyses
Statistical analyses were conducted using Microsoftâ Excel 2016 (concentration factor, effective volume assayed, and Pearson's chi-squared) or RStudioâ ver. 1.0.143 using the lme4, dplyr and Rcpp packages (McNemar mid-P, generalized linear mixed model (GLMM) and logistic regression) (additional details on these methods are provided in the Statistical Methods section of the Supporting Information).
The concentration factor (ratio between the original and final sample volumes) and effective volume assayed (product of the concentration factor and assay volume) were calculated. The WHO algorithm assay volume is 3 ml.
The McNemar mid-P test was used to determine significant differences between BMFS and two-phase samples (Objective 1)  . Generalized linear mixed model and logistic regression models were performed to determine the effect of multiple variables on detection of SL1, SL2, SL3, NPEV and any PV (Tables 2 and 3 and Table S2). The effect of concentration method (BMFS or two-phase) on PV detection was tested (Objective 1). Factors tested for their impact on PV detection in BMFS samples included (i) filtration volume and time (GLMM), (ii) processing time (CDC samples, GLMM; KEMRI samples, logistic regression), (iii) refrigerated shipping conditions (shipping time, GLMM; temperature, logistic regression) and (iv) assay location (GLMM) (Objective 2). Logistic regression was used to determine effect of assay location on PV detection for two-phase samples (Objective 2). The effects of volume and time filtered on PV detection were not tested for two-phase samples, as the processed volume did not vary (500 ml) and no filtration occurred. Processing time was not tested for two-phase since all two-phase samples were received the day of collection, and processed within 2 days (WHO 2015a). The effect of refrigerated shipping of BMFS filters on PV detection was tested for BMFS samples assayed at CDC (BMFS samples assayed at KEMRI were not shipped), but not for two-phase and  BMFS concentrates, as these were shipped frozen, with minimal temperature fluctuation. The GLMM accounted for random and fixed effect variables, and binary outcomes (Tables 2 and 3 and   Table S2). Pairs were treated as clusters and assigned random effect variables, to enable analysis of replicate BMFS sample results without bias. Pairs were defined as twophase and BMFS (individual or replicate) samples, Variables other than pairs and the target were considered precision variables and assigned fixed effects. Analyses that did not include random effect variables (i.e. pairs) were analysed by logistic regression. The logistic regression accounted for fixed effect variables and binary outcomes (Table 3 and Table S2). All variables were considered precision variables and assigned fixed effects, other than the target variable. For both the GLMM and logistic regression, all assayed samples were included for analysis of SL1, SL3, NPEV and any PV. For analyses on factors impacting SL2 detection, only samples prior to 18 July 2016 were considered, as SL2 was presumed absent from the environment 3 months after the bOPV switch (Huang et al. 2005). Results from these analyses included the OR of positive virus detection with an increase in the predictor of interest by one unit, while holding all other factors constant, the 95% CI for the OR, and the P-value. The Pearson's chi-squared test was used to determine the likelihood that differences in virus detection before and after the bOPV switch were due to chance (Objective 3) . The BMFS samples included are the same as used during the McNemar mid-P analysis ( Table S2). The OR of virus detection during tOPV use, compared to detection during bOPV use, was calculated.

Factors impacting PV detection
The volume passed through each filter ranged between 1Á4 and 4Á0 l and averaged 2Á7 AE 0Á16 l (95% CI). The average concentration factor was 270-fold, and average effective volume assayed was 815 AE 18 ml (95% CI). SL1, SL2 and SL3 detection were not statistically impacted by BMFS filtration volume (P = 0Á272, 0Á287 and 0Á211, respectively) or filtration time (P = 0Á862, 0Á945, and 0Á244, respectively) ( Table 3). For BMFS samples analysed at KEMRI, an increased time from collection to obtaining primary concentrate resulted in significantly decreased odds of SL3 detection (P = 0Á015). Assay location (KEMRI or CDC) did not statistically impact PV detection in BMFS or two-phase samples (Fig. 1, Table 3).

Discussion
The BMFS method detected SL1, SL2 and SL3 more frequently in environmental samples than the two-phase method (P = 0Á036, 0Á009, and 2 9 10 À5 (McNemar mid-P) and P = 0Á007, 0Á007, and 1 9 10 À5 (GLMM), respectively). Processing variables (e.g. time, volume, transit conditions, location, etc.) did not impact SL1 and SL2 detection in BMFS samples, although processing time and maximum shipment temperature impacted SL3 detection. SL2 was detected less frequently after the bOPV switch, indicating a successful switch in Nairobi. This study demonstrated that BMFS can be an acceptable method for PV environmental surveillance and resulted in successful PV environmental surveillance in Nairobi.
Increased PV detection by BMFS may be due to the higher concentration factor , with an average of 270-fold for BMFS versus 50-fold for twophase samples. However, the increase in PV detection is not directly proportional to the concentration factor and effective volume assayed, as the assay is nonquantitative.
When examining BMFS samples only, filtration volume (and consequently, concentration factor) did not impact SL1, SL2 or SL3 detection (P = 0Á272, 0Á287 and 0Á211, respectively, Table 3). As 96Á5% of BMFS samples filtered 2Á0-3Á5 l, additional data at lower and higher volumes may help examine the full effects of filtration volume on PV detection. Future research may determine an optimal volume at which PV would be detected in a majority of samples (with PV presence in the system).
The effective volume assayed in BMFS samples is due to concentration of the original volume (2Á7 l) to the WHO algorithm target volume (10 ml), using both primary and secondary concentration. Secondary concentration is not used in the two-phase method because it increases sample manipulation, and modifying this existing standard method would complicate harmonization around the global network, and the method already results in the current target volume for the WHO algorithm. Incorporating secondary concentration into the two-phase method could be explored to increase the effective volume assayed if cell culture independent PV detection was used or if a greater concentration factor was desired.
Shipping and processing variables did not impact SL1 and SL2 detection in BMFS samples (P > 0Á1), indicating viruses on BMFS filters can withstand shipment delays, long filtration times and cold chain loss. Odds of SL3 detection were reduced with an increased maximum shipment temperature (P = 0Á027), suggesting SL3 sensitivity to temperature fluctuation. Preservative agent treatment on the filters helped maintain sample integrity when cold chain was lost during shipment (Fagnant et al. 2017a). Filtration time ranged from 89 to 240 min, indicating PV can be detected when the filtration system is placed in direct sunlight for extended times (Nairobi maximum temperatures average 24-28°C (Egondi et al. 2015)). SL2 was the most frequently detected PV during tOPV use. While SL1, SL2 and SL3 are shed at similar rates, SL2 circulates more widely among unvaccinated individuals, thus increasing SL2 environmental prevalence (Troy et al. 2014;Ferreyra-Reyes et al. 2017), and possibly contributing to frequent SL2 detection. After the switch to bOPV, SL2 detection decreased, suggesting its absence from the environment following OPV2 withdrawal in Nairobi. SL1 and SL3 were detected more frequently after the switch to bOPV and these results were statistically significant for SL1 (Table 4).
The study had several limitations ). Ten per cent of BMFS samples analysed at KEMRI were collected following the tOPV campaign from 9 to 13 April 2016 and experienced filter hold times of 19-26 days. As the tOPV would increase Sabin-like PV shedding and subsequently increase environmental PV concentrations, these samples may have disproportionately impacted SL3 detection analyses. These showed improved odds of SL3 detection, with increased time from collection to primary concentration when BMFS samples were analysed at KEMRI (OR = 1Á15; Table 3). While BMFS and two-phase samples were collected sequentially within a 1-m radius, they were not processed from the same homogenous source, thus natural virus distribution is reflected in the results. Additionally, as the Poliovirus Isolation Algorithm was utilized for analysis and is designed for PV detection, PV presence may have impacted NPEV reporting, which is used as a site and sample control. As the BMFS detected PV more frequently than two-phase during this study, it is difficult to compare the rate of NPEV reporting between BMFS and two-phase samples due to potential masking of NPEVs in a PV background. Finally, use of MS2 as an internal process control yields inconsistent recoveries, potentially due to disaggregation, integrity of the MS2, challenges with the double agar layer assay or other issues. Future work should examine the use of alternative process controls, be it seeded or indigenous organisms such as adenovirus, pepper mild mottle virus, other bacteriophages, or direct detection of NPEV.
The BMFS resulted in frequent PV detection. Generally, filtration, processing and filter shipping variables did not impact PV detection in BMFS samples, indicating that BMFS retains sample integrity even under nonoptimal conditions. SL2 was detected less frequently after the bOPV switch, indicating that the gradual decrease of SL2 is commensurate with OPV2 withdrawal. The BMFS offers comparable or improved PV detection under the conditions of this study, relative to the two-phase method. Future BMFS work should explore its ability to detect additional targets, including other viruses, bacteria, parasites and antimicrobial resistance genes.