We selected two areas, one in Hidalgo State and one in Mexico City, for environmental sampling based upon the following criteria: (1) convergent wastewater system; (2) presence of risk factors for VDPV emergence such as low polio vaccination coverage and high population density; and (3) proximity to a polio laboratory. Coverage by 1 year of age with three doses of IPV in 2013 was 67% for Hidalgo State and 94% for Mexico City. In the absence of water treatment plants in the areas chosen, samples were collected at access points to underground wastewater canals. The catchment population was estimated as 270,000 persons for the Hidalgo site and 4 million for the Mexico City site.

Prospective collection of samples was planned around two OPV campaigns conducted in 2016 and 2017. The 2016 campaign distributing trivalent OPV was implemented between February 16th and March 20th and the 2017 campaign, which distributed bivalent OPV, was conducted between February 25th and March 20th. A campaign planned for October 2016 was canceled because of insufficient bivalent OPV supply. Assuming that Sabin viruses were unlikely to be detected more than 2 months after a campaign (Esteves-Jaramillo et al. 2014; Mas Lago et al. 2003), samples were collected during the week before each campaign, then bi-weekly during the duration of the campaigns and up to 1 month after the first campaign. Samples were collected weekly between April 26 and September 27, 2017; and monthly between September 27 and February 21, 2018.

At each study site, trained staff collected two serial samples of wastewater. Using the BMFS, about 5 L of wastewater was collected with a 6 L polyurethane-coated nylon bag. This bag, which was the third-generation BMFS model (Fagnant et al. 2018), had a mesh over the opening to exclude refuse and debris, an open tubing adapter port at the bottom for discarding settled solids into a second vessel, and a 15.5° slope toward the opening that facilitated the bleeding of sediment. After filling with wastewater water, the bag was placed on a tripod and connected to the ViroCap filter (Scientific Methods, Granger, Indiana, USA) to use gravity to force water through the filter. Before starting the filtration, the sample was allowed to sit for 10 min, and the bottom sediment was transferred to a Whirl–Pak bag through the adapter port to avoid solids clogging the filter. The sediment was added back to the bag before finishing the filtration process, if possible. Once all the water in the bag had passed through the filter (or the filter had become obstructed), the filter was disconnected and placed in a cold box with frozen ice packs. ViroCap filter housings were re-used up to ten times for the first 63 samples and were single-use for the remaining samples (Fagnant et al. 2018).

While filtration was taking place with the BMFS, staff collected another sample of wastewater at the same site, using a bucket. The sample was poured into a 1-L container and placed in the same cold box.

The sequential samples were transported directly to the poliovirus laboratory at the Instituto de Diagnóstico y Referencia Epidemiológico (InDRE) in Mexico City. In Hidalgo, the samples were taken to the Laboratorio Estatal de Salud Publica (LESP) in Hidalgo State for administrative processing, and then shipped, within 12 h of collection, to the InDRE in Mexico City.

A portion of each 1-L wastewater sample was concentrated in the laboratory using the two-phase separation (TPS) method recommended by the WHO (Global Polio Eradication Initiative 2015). Briefly, the sample was centrifuged for 20 min at 1500×g and 4 °C. Following this, 500 mL of the supernatant were mixed with Polyethylene Glycol 6000 (Sigma-Aldrich, Damstad, Germany) and Dextran T40 (Pharmacocosmos A/S, Holbaek, Denmark). The mixture was shaken vigorously and incubated overnight at 4 °C in a separatory funnel. The lower phase and interphase were collected and treated with chloroform, then centrifuged as above, and the supernatant was supplemented with penicillin and streptomycin before storage at − 20 °C for future testing (Blomqvist et al. 2004; Shulman et al. 2006). Reported recovery efficiency for a sample spiked with 10⁵ CCID₅₀ of type 1 Sabin with this method were between 4 and 39% (CDC, unpublished results). However, wide variations in poliovirus recovery within the same sample have been reported in parallel testing, especially when samples have enterovirus mixtures (Global Polio Eradication Initiative 2015; Hovi et al. 2005).

The ViroCap filters were eluted as described previously (Fagnant et al. 2014, 2018). Sterile 1.5% beef extract (100 mL) with 0.05 M glycine buffer, pH 9.5, was pumped into the filter and allowed to stand for 30 min to allow viruses to detach from the filter. The eluate was collected with a manual bilge pump and pH was adjusted to 7.0–7.5. An aliquot of 12 mL was treated with chloroform and antibiotics as described above, and stored at − 20 °C for future testing. Poliovirus recovery from surface water, secondary effluent and a 50:50 mix using ViroCap filters averaged 44% for type 1, 70% for type 2 and 81% for type 3 (Fagnant et al. 2014).

The concentrates obtained with both methods were inoculated onto five 25 cm² flasks (0.5 mL per flask) containing monolayers of L20B cells (mouse fibroblast cells that express the human poliovirus receptor) and onto one 25 cm² flask containing a monolayer of human rhabdomyosarcoma (RD) cells, following the standard WHO methodology (World Health Organization 2004, 2010). The presence of viral cytopathic effect (CPE) was monitored daily according to the approved virus isolation algorithm. The presence of poliovirus or other enteroviruses and the poliovirus serotype was determined by real-time reverse transcriptase polymerase chain reaction (rRT-PCR) versions 4.0, 4.1 and 5.0 (termed intratypic differentiation; ITD) (Kilpatrick et al. 2009; Gerloff et al. 2018). Sabin isolates were screened by the real-time PCR assay for VDPVs as described previously (Kilpatrick et al. 2009; Gerloff et al. 2018; Kilpatrick et al. 2014), and standard sequencing in the VP1 region was performed at CDC laboratory on a subset of isolates as needed (Burns et al. 2016; Kilpatrick et al. 2014).

Staff involved in the collection and transport of samples recorded information about procedures on standardized forms (one per sample). Relevant data included: (a) dates of sample collection; (b) times when collection started and when sample was placed in the cold box; and (c) volumes collected, and, for BMFS, volumes remaining after removing the sediment and after filtration (using volume markings in the bag). Staff also described problems that arose during sample collection.

Laboratory staff recorded the following information: (a) date, time, and temperature of samples on arrival at the laboratory; (b) volume of sample concentrate obtained with each method; and (c) results per sample of virus isolate, rRT-PCR, and (when performed) sequencing assays.

The primary study outcome was the proportion of samples positive for poliovirus types 1, 2, or 3, or for NPEV, with either the two-phase separation or the BMFS methodology. Secondary outcomes included description of volumes obtained and time invested in sample collection and concentration, and estimates of costs per sample. Proportions are presented with Wilson 95% confidence intervals, and continuous variables are presented as mean ± standard deviation.

We used McNemar’s concordance test to compare the proportion of samples that tested positive for poliovirus or NPEV with each method. Statistical significance was defined as a mid-p value of McNemar’s test below 0.05. Continuous variables were compared using t test.

To determine sample size, we assumed that the difference in percent of positive samples between the two procedures would be 15%, with a discordance between samples of 30%. Under these assumptions, 125 pairs of samples would be required for a power of 85%. Sample size calculations were performed using PASS version 14. Analyses were performed with SAS version 9.3 and R version 3.4.3.

The 125 pairs of samples were collected between February 16, 2016 and April 18, 2017. Of these, 62 pairs were collected in Mexico City and 63 in Hidalgo. Because of delays in shipment of supplies, five samples that were supposed to have been processed in the field with the BMFS had to be collected (6 L) and transported to the laboratory in cold chain, where they were stored at − 20 °C, and thawed and filtered at a later time. Information on collection and transportation for these five samples was excluded from the analysis for field logistics.

Staff initially collected an average (± standard deviation) of 5.2 ± 0.5 L with the BMFS collection bag and filtered 4.8 ± 0.8 L in situ (Table 1). The volume filtered reached > 4 L for 90% of the samples (111/123), 3–4 L for 9% of the samples (11/123) and < 3 L for one (1%) sample. Staff collected a slightly larger initial sample in the Mexico City site than in the Hidalgo site (5.4 ± 0.7 L in Mexico City vs. 5.1 ± 0.2 L in Hidalgo, p < 0.001), and were able to filter a slightly larger volume (5.0 ± 0.9 vs. 4.7 ± 0.7 L, p = 0.039). Field staff reported that filtration was incomplete because of obstructed filter in six samples and because of leaks in the filter housing in four samples. Reasons were not provided for other occasions when all the volume collected was not filtered. The volume of samples collected by grab was 1 ± 0 L.

Field staff spent 4.3 ± 2.2 min collecting a 1 L sample for TPS, and 73.5 ± 30.5 min collecting and filtering a sample with the BMFS (Table 1). The filtration process took longer in Hidalgo than in Mexico City (87.9 ± 27.6 vs. 58.9 ± 26.1 min, p < 0.001).

Duration of transportation measured the time between sample placement in the cold box and arrival at the local laboratory (LESP in Hidalgo) and reference laboratory in Mexico City (InDRE). Transportation took less than 2 h for samples collected in Mexico City, which were taken directly to the InDRE. Transportation took less than 4 h for samples collected in Hidalgo, after combining transportations to the LESP, and then to the InDRE (Table 1). Temperature inside the cold boxes on arrival to the InDRE laboratory was 3.8 ± 1.9 °C.

Samples collected for TPS were initially stored at − 20 °C, then thawed and concentrated in batches of four using the WHO standard method (Global Polio Eradication Initiative 2015). The average volume of the concentrate obtained from an original sample of 500 mL was 14.4 ± 3.6 mL (concentration factor of 35) for an effective volume of 104 mL. The laboratory staff estimated that the concentration process required about 5 h of hands-on work (2 h for sample processing plus 3 h for reagent preparation and clean-up) and 16 h of overnight incubation.

The ViroCap filters were eluted using the manual bilge pump as described (Fagnant et al. 2014), upon arrival to the laboratory (except for five samples from the Hidalgo site that were collected by grab, frozen, and later filtered at the InDRE laboratory). The average volume of elute obtained was 99.9 ± 5.3 mL, for a concentration factor of 48 from the average 4.8 L of sample filtered and an effective volume assayed of 144 mL. The staff estimated that the process required approximately 3.5 h (1 h for the elution process plus 2.5 h for reagent preparation and clean-up).

Table 2 shows a calculation of the cost of non-reusable supplies and reagents required for collecting and concentrating the sample (up to addition of chloroform and antibiotics) with each method. We excluded supplies required for transportation and storage of samples in cold chain, because they were partially provided by the shipping company. In addition, we separated the cost of supplies for personal protective equipment and for cleaning and disposal of infectious materials, which were similar for the two methods, from the cost of supplies for procedures specific to each method.

The cost of collecting one sample was > 30-fold higher with the BMFS than for the TPS method (~ $101 vs. $3 per sample collected for TPS), whereas the cost of the concentration process was similar for both methodologies ($4 with the BMFS versus $5 with TPS). Adding the two steps, the cost was $105 for each sample collected and processed with the BMFS compared with $8 per sample collected and concentrated with the current TPS method. The total up-front cost of the pump and other reusable materials required for the elution of the sample was about $530. The cost of reusable materials for the two-phase concentration was $121. We did not include the cost of equipment, because most of it should be available in a virology laboratory (i.e. biosafety cabinet, centrifuges, shakers, etc.). The only specific equipment that most laboratories will not have are high-speed refrigerated centrifuges for large volumes (i.e. 250 mL), which may cost $15,000–20,000 and a separate refrigerator to store the samples during the overnight incubation (estimated cost $6000–8000).

A total of 125 pairs of sequential samples were collected between February 16, 2016 and April 18, 2017; 62 pairs in Mexico City and 63 pairs in Hidalgo (Fig. 1).

As expected, most of the samples were positive for Sabin poliovirus during the two campaigns and for about 1 month after the campaigns (Fig. 1). Sabin strains could be detected up to 13 weeks after the end of the tOPV campaign (March 20th, 2016), either alone or in mixtures with other Sabin serotypes or NPEV. After that, samples remained negative for polioviruses until the bOPV campaign conducted in March 2017 (1 year after the tOPV campaign).

The comparison of sequentially collected paired samples with McNemar’s test indicated significant discordances between the two concentration methods for detecting Sabin type 1, Sabin type 3 and NPEV, as well as for detecting any poliovirus or NPEV (Table 3). As shown on Table 4, the TPS method identified Sabin 1, Sabin 3 and NPEV in a higher proportion of samples than the BMFS. Sabin 1 poliovirus was detected in 37 samples [30%; 95% confidence interval 22.3–38.1%] with the two-phase method versus 24 samples [19%; 13.3–27.0%] with the BMFS (McNemar’s mid p value = 0.004). Sabin 3 poliovirus was detected in 59 [47%; 38.7–55.9%] versus 49 samples (39%; 31.1–48.0%, p = 0.043), and NPEV was detected in 67 samples [54%; 44.9–62.1%] versus 40 samples [32%; 24.5–40.6%, p < .001). There were no significant differences for detection of type 2 Sabin strains (20%; 13.9–27.9% with TPS vs. 24%; 17.4–32.2% with BMFS, p = 0.18). The difference in detection of NPEV between methods was maintained after exclusion of mixtures of poliovirus and NPEV, with 48/107 [45%] samples positive for NPEV with the two-phase method versus 31/107 [30%] with the BMFS (p = 0.004).