Technicians from the Kenya Medical Research Institute (KEMRI) Center for Virus Research in Nairobi, Kenya and the University of Pretoria in Pretoria, South Africa were trained in-person on the use of the BMFS field kit and laboratory protocols (Fagnant et al. 2014, 2017, 2018). Trainees received detailed illustrated protocols, a one-page “quick-sheet” reference protocol, and a training video describing the BMFS field sampling protocol. Training on a supplemental “bucket” modification was completed by telephone with a detailed illustrated protocol. This modification was developed to safely transport large water samples on cold chain for sites with safety or security concerns (Zhou et al. 2018).

Environmental samples were collected from April 14, 2015 to September 7, 2015 at four sites in Nairobi, Kenya, twice per month, and analyzed for PV (Fig. 1). Samples were obtained from the Motoine River bordering the Kibera informal settlement (Kibera), a latrine waste stream that connects to the Mathare River (Starehe), and two sewer conveyance lines in the Eastleigh neighborhood (Eastleigh A and B). In 2013 and 2014, these sites were chosen for the WHO environmental surveillance program based on discussions among KEMRI, the Kenya Ministry of Health, and WHO. Sampling was conducted by a field staff team made up of two KEMRI field technicians and a supervisor. At some sites, a water utility staff member and/or community health worker were present. BMFS sampling frequency was set to coincide with routine WHO two-phase sampling activities, and occurred over 12 sampling days (Online Resource, Table S4). Environmental samples processed by the two-phase separation method were collected first, and a second sample was collected for the BMFS. Samples were collected within 5 min and a 1-m radius of each other. Samples were collected in the late morning/early afternoon, with sampling beginning between 10:30 and 14:15 at the Kibera site (n = 15), 09:00–12:15 at the Starehe site (n = 15), 09:50–11:15 at the Eastleigh A site (n = 13), and 09:45–11:45 at the Eastleigh B site (n = 13). During the first 14 sampling events of the study period (April to May 2015), two BMFS samples were collected sequentially (within 5 min) at the same site to determine variability in BMFS sampling: four each at Kibera and Starehe, and three each at Eastleigh A and Eastleigh B (Table 1).Fig. 1

Workflow of collected environmental samples. Bag-mediated filtration system (BMFS) samples were collected concurrently with 1 L grab samples for two-phase concentration according to World Health Organization protocols. KEMRI Kenya Medical Research Institute, UP University of Pretoria, CDC US Centers for Disease Control and Prevention, ITD intratypic differentiation

This study was approved by KEMRI, the National Commission for Science, Technology, and Innovation (NACOSTI), and the Research Ethics Committee, Faculty of Health Sciences, University of Pretoria Ethics Reference no: 119/2017. Appropriate import and export permits were obtained from the Department of Health South Africa and the United States Centers for Disease Control and Prevention (CDC) to facilitate intercountry transfer of samples and derivatives.

Two-phase samples were collected according to WHO protocols (WHO 2015). One-liter grab samples were collected and transported on cold packs to KEMRI (Fig. 1, B1). Aliquots of 500 mL per sample were concentrated at KEMRI using the two-phase separation PEG/dextran method as previously described (Fig. 1, B2) (WHO 2015). Briefly, samples were centrifuged and the pellet was saved to add at a later step. PEG 6000 (Sigma-Aldrich, St. Louis, MO, USA), dextran (Pharmacosmos, Holbaek, Denmark), and NaCl (Sigma-Aldrich) were added to the supernatant, mixed for one hour, added to a separatory funnel, and held overnight at 4 °C. The lower phase and interphase in the separation funnel were collected, combined with the pellet, and extracted using chloroform. The upper aqueous phase from the chloroform extraction was collected (approximately 10 mL; 7–12 mL), and sample concentrates were sent to the CDC in Atlanta, GA, USA. Two-phase samples were analyzed at the CDC according to the WHO virus isolation algorithm (WHO 2015) followed by intratypic differentiation (ITD) for SL PV type 1 (SL1), SL PV type 2 (SL2), SL PV type 3 (SL3), WPV type 1 (WPV1), WPV type 3 (WPV3), panPV, and panEV (Fig. 1, B3) (WHO algorithm, see above “WHO Algorithm” section). This algorithm, combined with VP1 sequencing, can distinguish VDPV from SL PV and WPV (Gerloff et al. 2018).

A generalized linear mixed model was performed to determine if there was a correlation between MS2 and PV detection. Sample site was treated as a non-random variable effect. The McNemar mid-p test was used to determine the significance of the difference between samples (i.e., BMFS and two-phase samples collected within 5 min and a 1-meter radius of each other, and individual BMFS samples measured by different detection methods) (Fagerland et al. 2013; McNemar 1947). The odds ratio (OR) and the 95% confidence intervals (CI) on the OR were also calculated. Results were considered significant with a mid-p value < 0.05. Analyses were conducted using MS Excel 2016 and R version 3.4.3.

Results from the two sequentially collected BMFS samples (BMFS-1 and BMFS-2; April to May 2015) were analyzed by three methods. First, they were treated as a single composite sample result (n = 14) by the (1) concordant positive method (Table 1). Additionally (2) BMFS-1 (n = 14) and single BMFS samples (n = 28) and (3) BMFS-2 (n = 14) and single BMFS samples (n = 28) were combined into two separate sample sets. The concordant positive method considered discordant PV results to be negative and concordant PV results as a single sample, as to not bias the comparison between the BMFS and two-phase methods. Single BMFS samples were collected from June to September 2015 (n = 28), for a total of 42 BMFS samples analyzed. Single two-phase samples were collected throughout the study period (n = 42).

This study established a study design and logistics required to conduct a future environmental surveillance study using the BMFS by identifying the scheme logistics, shipping logistics, and processing timeframes needed. An average of 2.9 ± 0.1 L (95% CI; n = 54) was filtered using the BMFS, and filtration took an average of 70.1 ± 3.9 min (95% CI; n = 55). Use of the bucket modification at two sites due to safety and security concerns enabled continued collection of BMFS samples alongside two-phase samples. This modification was a preferred strategy by the field staff as it enabled concurrent filtration of all samples outside near KEMRI; therefore, it was implemented at all sites during the last 9 sampling days (Online Resource, Table S4). During these 9 sampling days, overall sample collection (from the beginning of the first sample collection to arrival back at KEMRI) took an average of 4.7 h (n = 9) (Online Resource, Table S4).

Protocols were instituted to inform on and enhance sample integrity, including the addition of MS2 and preservatives, timely shipping, and cold-chain techniques. MS2 was seeded in all filters prior to sample filtration (n = 56), and MS2 recovery efficiency ranged from no detection to 500%, with a median recovery value of 8.8%. No correlation existed between MS2 recovery and positive or negative PV detection in BMFS (p = 0.81, 0.27, and 0.54 for SL1, SL2, and SL3, respectively) and two-phase (p = 0.23, 0.11, and 0.62 for SL1, SL2, and SL3, respectively) samples when measured using the WHO algorithm. After filtration, filters were stored (4 °C) and preservatives were added prior to shipping of the filters to the University of Pretoria for processing (n = 56; Online Resource, Table S4). After elution and secondary concentration at the University of Pretoria, portions of the BMFS sample concentrates were stored prior to shipment to the CDC (− 80 °C, 4.5–9.5 months). Samples were shipped to the CDC in two batches of 30 samples each with dry ice replenished throughout (shipping time 14 and 17 days, respectively).

Samples received at the CDC were stored (− 20 °C or − 80 °C, 1 week to 3.5 months) before processing using the WHO algorithm. For BMFS samples at the University of Pretoria measured by direct RT-PCR, 1-mL aliquots of chloroform-extracted secondary concentrates were frozen (− 20 °C) prior to nucleic acid extraction. The nucleic acids were stored until real-time RT-PCR was completed (− 80 °C, < 1 month). For samples at the University of Pretoria measured by ICC-RT-PCR, cell culture was initiated within 9 days after secondary concentration for 86% of samples, and the chloroform-extracted secondary concentrates were stored at 4 °C until inoculation into cell lines. For the other 14% of samples, the secondary concentrates were stored (− 20 °C, 20 days) prior to inoculation of the cell lines due to the large quantity of samples received.

No statistical difference in PV detection was noted between the two sequentially collected BMFS samples (BMFS-1 and BMFS-2) for all three detection methods used (WHO algorithm, direct RT-PCR, and ICC-RT-PCR: p = 0.22, 1, and 0.5 for SL1, p = 0.63, 0.5, and 0.63 for SL2, and p = 0.25, 0.13, and 0.25 for SL3; n = 14). Combining sequentially collected BMFS samples (n = 14; April to May 2015) and single BMFS samples (n = 28; June to September 2015) results in a total of 42 samples. When detected by the WHO algorithm and analyzed using the concordant positive method (Table 1), SL1, SL2, and SL3 were detected in 19.0%, 76.2%, and 52.4% of BMFS samples, respectively (Table 2; n = 42). When considering the BMFS-1 and single BMFS sample dataset, SL1, SL2, and SL3 were detected in 28.6%, 78.6%, and 57.1% of BMFS samples, respectively (n = 42). When considering the BMFS-2 and single BMFS sample dataset, SL1, SL2, and SL3 were detected in 21.4%, 81.0%, and 52.4% of BMFS samples, respectively (n = 42). The following analyses (comparing BMFS samples analyzed by different detection methods, and comparing between the BMFS and two-phase results) utilized the concordant positive method.Table 2

PV detection in 1-L grab samples with 500-mL processed by two-phase concentration and measured by the WHO algorithm versus BMFS samples by three different detection methods

The rate of PV detection using the three different detection methods varied (Table 2 and Online Resource, Table S6). PV detection in BMFS samples was not significantly different when measured by the WHO algorithm compared to ICC-RT-PCR (p = 0.092, 0.125, and 0.092 for SL1, SL2, and SL3, respectively; Online Resource, Table S6). PV detection in BMFS samples was statistically less frequent by direct RT-PCR compared to WHO algorithm (p = 3.9 × 10⁻³, 1.49 × 10⁻⁸, and 6.3 × 10⁻³ for SL1, SL2, and SL3, respectively) and ICC-RT-PCR (p = 1.2 × 10⁻⁴, 3.73 × 10⁻⁹, and 3.05 × 10⁻⁵ for SL1, SL2, and SL3, respectively; Online Resource, Table S6). Discordant results favored the WHO algorithm when compared to direct RT-PCR, in all cases except with one SL3. Discordant results favored ICC-RT-PCR when compared to direct RT-PCR in all cases. The most frequently detected PV also varied with the detection method. SL2 was the most frequently detected PV, followed by SL3, and SL1, when measured using the WHO algorithm and ICC-RT-PCR (Table 2). In contrast, SL3 was the most frequently detected PV when measured using direct RT-PCR. PV detection also varied after amplification in the three cell lines in ICC-RT-PCR (Table 2 and Online Resource, Table S7). PV prevalence in BMFS samples was not statistically different between PLC/PRF/5 and L20B cells (p = 0.344, 0.388, and 0.774 for SL1, SL2, and SL3, respectively; Online Resource, Table S7). In contrast, PV was statistically detected less frequently after amplification in BGM cells than after amplification in L20B cells (p = 2.0 × 10⁻³, 7.63 × 10⁻⁵, and 2.4 × 10⁻³ for SL1, SL2, and SL3, respectively) or PLC/PRF/5 cells (p = 1.6 × 10⁻², 5.2 × 10⁻⁴, and 2.6 × 10⁻³ for SL1, SL2, and SL3, respectively; Online Resource, Table S7).

The BMFS and two-phase samples were both concentrated to approximately 10 mL resulting in a 290- and 50-fold concentration, respectively (effective volume assayed of 870 mL per BMFS sample vs. 150 mL per two-phase sample after inoculation). Poliovirus was detected in a majority of samples using the standard WHO algorithm (Table 3) with at least one PV in 88.1% (37 of 42) and 76.2% (32 of 42) of BMFS and two-phase samples, respectively (Fig. 2). WPV1, WPV3, and VDPV were not detected in samples by any of the detection methods. There was statistically more frequent detection of SL3 in BMFS than in two-phase samples (52.4% [22 of 42] and 33.3% [14 of 42], respectively; p = 0.035), with an increased odds ratio of SL3 detection in BMFS samples compared to two-phase samples (OR 3.7 [1.0–13 95% CI]) (Table 3). For SL1, there was no statistical difference in the frequency of detection between BMFS and two-phase samples (19.0% [8 of 42] and 21.4% [9 of 42], respectively; p = 0.80) (Table 3). There was no statistical difference in the frequency of SL2 detection in BMFS and two-phase samples (76.2% [32 of 42] and 64.3% [27 of 42], respectively; p = 0.18) (Table 3). SL2 was the most frequently detected PV in both BMFS and two-phase samples, followed by SL3, and SL1 (Table 3).Table 3

Comparison of PV detection in matching BMFS and two-phase samples as measured by WHO algorithm

Dataset grouping did not result in a change in the statistical significance of SL1, SL2, or SL3 detection when comparing between BMFS and two-phase samples. SL1 and SL2 detection was not statistically different when comparing two-phase samples with the BMFS-1 and single BMFS samples dataset, the BMFS-2 and single BMFS samples dataset, or the BMFS concordant positive method dataset (p = 0.481, 1.0, and 0.80 for SL1, respectively, and p = 0.118, 0.057, and 0.18 for SL2, respectively). SL3 detection was statistically more frequent in the three BMFS datasets when compared individually to two-phase samples (p = 0.013, 0.035, and 0.035, respectively) (Tables 3, S7).

The goal of poliovirus environmental surveillance is to supplement AFP surveillance and assist in determining locations where PV is circulating or where it has been eliminated. It can also help with monitoring the elimination of SL PV after type-specific OPV cessation or complete cessation of OPV. PV (SL1, SL2, and/or SL3) was detected in a majority of BMFS and two-phase samples. The BMFS method demonstrated equivalent or greater PV detection when compared to WHO’s two-phase method based on the McNemar mid-p test (Table 3).

The feasibility of the BMFS method was demonstrated during implementation of this study for PV environmental surveillance. The method was successfully applied by staff from the national polio laboratory in Nairobi, Kenya (KEMRI) and the University of Pretoria laboratory personnel. Study sites were previously established by WHO, Kenya Ministry of Health, and KEMRI personnel for environmental surveillance. Due to security concerns at sites during the study, a bucket modification was developed to minimize field exposure. Transporting samples to a centralized processing location reduced the overall processing time when multiple sites were sampled on a single day by permitting parallel filtration. The bucket modification is recommended for security-compromised locations but, if security is not a concern, processing filters on-site may reduce the cost of transporting sewage or human-impacted residual waters if they are shipped.

MS2 was initially chosen as an internal control because it is non-pathogenic, found in the environment, and is similar to enteric viruses (Grabow 2001). In a previous study, with MS2 pre-seeded onto ViroCap filters, 10 L PV1-spiked Seattle influent wastewater filtered through these ViroCap filters, and preservatives added to the filters prior to elution, the average MS2 recovery (84.6%) was greater than the average PV1 recovery spiked into the 10 L influent wastewater (53.5%) (Fagnant et al. 2017). The erratic recovery of MS2 seen in this pilot study may have been due to various factors including deaggregation, MS2 presence in the environment, complications with the double agar layer assay, and/or seeding at different levels than anticipated. It is also possible the MS2 stock was left at room temperature for longer than prescribed by the protocol, thereby reducing the total PFUs seeded onto the filter. The relatively high frequency of PV detection in two-phase and BMFS field samples in comparison with the seeded MS2 may be due to a reduced robustness of laboratory-strain organisms when compared with environmental strains (Online Resource Table, S5) (Silverman and Nelson 2016). The inconsistent recovery of MS2 indicated its lack of suitability as an internal control and MS2 will not be included in future BMFS samples.

This study also showed that sample integrity could be maintained even when shipping filters to an out-of-country laboratory for processing and analysis, demonstrating the potential for use of this system in low-resource settings without local access to a processing laboratory. Samples were shipped on ice packs at 4 °C and a preservative mixture was added to the filters prior to shipment. For future BMFS studies, if filters are shipped for processing, the addition of preservatives is recommended, to stabilize virus and reduce bacterial and fungal growth in the case of cold-chain failure during shipment. The addition of preservatives has been found to increase virus survival in ViroCap filters held for up to 7 days at temperatures up to 25 °C (Fagnant et al. 2017). Future research to determine the effect of these preservatives on unconcentrated water samples should be examined for potential use with other sampling methods.

While there was an overall trend of increased SL detection by the BMFS method, there was statistically more frequent SL3 detection in BMFS than in two-phase samples, and no statistical difference in SL1 and SL2 detection (Fig. 2). The greater volumes processed with the BMFS (2.9 ± 0.1 L) compared to the two-phase method (0.5 L) may have contributed to the more frequent SL3 detection, the most frequent PV detected. The BMFS method includes a secondary concentration step utilizing PEG precipitation after filtration through ViroCap filters, which increases the concentration factor by tenfold. The two-phase method recommended by the WHO GPLN includes primary concentration only, which facilitates establishment of standardized environmental surveillance processing methods in a large number of laboratories. Additionally, incorporation of secondary concentration into the two-phase method would require additional sample manipulation (with increased chance of cross-contamination), longer processing time, and reduced sample volume to save in reserve. At the time of this study, the secondary concentration method would have required expensive equipment and therefore fewer laboratories would have the capability to implement it. Recent studies have improved the BMFS secondary concentration method to lower the centrifuge speed and eliminate overnight shaking at 4 °C (Falman et al. 2019), and field validation has been completed.

The differences observed between the BMFS and two-phase PV detection results could also be due to the variable recovery rates between the three PV types. Previous efforts have found PV recoveries of 38–48% for PV1, 56–89% for PV2, and 70–88% for PV3 during seeded studies with ViroCap filters and a variety of water sources (Fagnant et al. 2014). PV recovery could be affected by charge interactions as the viral capsid is negatively charged when at a pH below the isoelectric point. As SL3 has a lower isoelectric point (6.34) than SL1 (7.42) and SL2 (7.18), SL3 likely has a higher affinity to the positively charged ViroCap filters during filtration (Thomassen et al. 2013). SL1 detection in BMFS (8 of 42 samples) and two-phase samples (9 of 42 samples) and SL2 detection in BMFS (32 of 42 samples) and two-phase samples (27 of 42 samples) were not statistically different (p = 0.80 and 0.18, respectively). The stringent negative control process of sample amplification in cell culture, followed by PCR provides confidence in positive samples being true positives. While SL3 detection was more frequent using the BMFS than two-phase method, three samples were discordant in favor of two-phase.

The frequency of SL2 detection in both BMFS and two-phase samples can provide an important baseline to compare SL2 prevalence and persistence after the switch from trivalent OPV to bivalent OPV. The frequency of SL2 detection was comparable to other settings (Esteves-Jaramillo et al. 2014; Nakamura et al. 2015; Wahjuhono et al. 2014; Wang et al. 2014). This may be due to the high rate of PV2 shedding (88%), compared to PV1 (42%) and PV3 (58%) shedding 1 week after OPV use (Laassri et al. 2005) as demonstrated by frequent PV2 detection within 3 weeks after OPV use (n = 10) compared to prior to OPV use (n = 32) (SL1: 40% vs. 38%, SL2: 90% vs. 84%, and SL3: 60% vs. 59% in BMFS and two-phase samples, respectively).

Frequency of PV detection by BMFS varied among the three methods (Table 2). Similar PV detection rates between the WHO algorithm (WHO 2015) and ICC-RT-PCR suggest that both methods are able to detect PV even though ICC-RT-PCR is not specifically targeted towards PV. The greater frequency of PV detection by the WHO algorithm and ICC-RT-PCR compared to direct RT-PCR is likely due to the increased volume assayed and virus amplification by cell culture. After concentration, 3 mL was inoculated into cell culture in ICC-RT-PCR. In comparison, 1 mL of the concentrate was used for nucleic acid extraction, and 5% of the extracted volume was used as the input in direct RT-PCR.

Non-polio enterovirus (NPEV) was detected in 87% of samples negative for PV, and 22% of samples positive for PV (Fig. 2). The WHO virus isolation algorithm is designed for PV detection, and if no PV is isolated, NPEV can be determined via two routes. NPEV presence can be determined if cytopathic effects are present solely in the RD cell culture flask, as this indicates NPEV presence and no isolated PV. Therefore, when PV is detected in a flask, it masks NPEV in that flask. NPEV can also be reported from ITD results if the panEV assay is positive and all PV assays are negative. It is possible the more frequent detection of any PV in BMFS samples (37 of 42) compared to two-phase samples (32 of 42) may have contributed to the less frequent detection of NPEV in BMFS samples (13 of 42) compared to two-phase samples (24 of 42).

Study limitations existed including those related to sample collection and processing. While BMFS and two-phase samples were collected within 5 min and a 1-meter radius of each other, there is natural variation in viral distribution that could affect the results and lead to discrepancies between matched samples. Furthermore, the full sample was not assayed during detection. If a low virus concentration was present in the sample, then it is possible the virus was not present in the assayed portion of the sample. This may also partially explain discordance in virus detection. Further, sample collection occurred in the late morning or early afternoon. If samples were collected earlier in the day as recommended by the WHO GPLN (e.g., by 9 a.m.), the detection rates may potentially have been greater. Due to the study design and laboratory capacity, BMFS samples were processed at the University of Pretoria and BMFS and two-phase samples were analyzed by the WHO algorithm at CDC. The shipping and storage times and conditions may have affected the detection of PV. Finally, as only SL PV was detected in this study, additional evaluation in WPV endemic regions is necessary to show BMFS capacity to detect circulating WPV (Zhou et al. 2018).

Below is the link to the electronic supplementary material.

Electronic supplementary material 1 (DOCX 64 kb)