We collected samples during both the rainy (March) and dry (May-July) seasons in La Paz, Bolivia (March/2018, May-June/2018, March/2019, June-July/2019) where a network of rivers receives untreated sewage discharge, industry effluent, and stormwater runoff. Most of the waterway flows in a series of engineered channels (Alarcon Calderon, 1996; Poma et al., 2016) consisting of highly impacted surface waters, which we refer to here as open wastewater canals (OWCs). The main waterway, the Choqueyapu River, begins as a small stream at Pamapalarama, flowing south through central La Paz, home to 900,000 people (1900 persons/km²) (INE, 2018; Ohno, A; Marui, A; Castro, ES; Reyes, AA; Elio-Calvo, D; Kasitani, H; Ishii, Y; Yamaguchi, 1997). As it flows through the city, the Choqueyapu is joined by tangential tributaries including the Orkojahuira, Irpavi, and Achumani rivers (Figure 1).

During the first three sampling events, we identified 13 sites along the Choqueyapu River meeting the following criteria: (1) proximity to OWCs, (2) accessible at ground level, and (3) unintrusive to residents or passers-by. We identified two control sites >1 km from known concentrated wastewaters or other contaminated sources: (1) Chacaltaya, a weather station and environmental observatory located at 5380 m in elevation and far from human habitation and (2) Pampalarama, a pristine site near the Choqueyapu headwaters. In a final sampling event (June-July 2019), we chose an expanded range of sites to further interrogate the spatial relationship of targets in relation to the OWCs, with particular attention to distances of 200 m or less from OWCs. We based selected sampling locations on the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM) version 2 data for the study area (NASA/METI/AIST/Japan Space Systems and Team, 2009) and a hydrographic model (ESRI Inc., 2019) in ArcGIS Version 10.7.1 to generate the major streamlines according to the topography for the La Paz metro area. Based on the four main OWCs, we randomly selected sites ≤ 200 m from the mid-point of OWCs and at least 150 m from the nearest other site. Given the diurnal cycles of atmospheric stability and the resulting impact on bioaerosol dissemination (Jacob, 1999; Jones and Harrison, 2004), we assessed the potential temporal associations of ARG densities in the air through sampling both in the morning and afternoon at each site.

We used the Six-Stage Viable Andersen Cascade Impactor (ACI) with selective media in six partitioned chambers at a flow rate of approximately 28.5 L/min to assess viability of E. coli at a subset of sites (Andersen, 1958) (ACI, Thermo Scientific^™, USA), as an indicator of aerosolized fecal material. We used AquaTest Medium (Sisco Research Laboratories PVT. LTD., India) to select for E. coli (Bain et al., 2015; Brown et al., 2020; Genter et al., 2019; Magro et al., 2014). We incubated plates overnight at 37°C. The limit of detection for the culture analysis was determined by dividing 1 colony-forming unit (CFU, the minimum count per plate) by the volume of air sampled.

To collect larger volumes for molecular analysis, we sampled for approximately 4 hours per sampling event using the ACD-200 BobCat Dry Filter Air Sampler (InnovaPrep, Drexel, MO, USA) with 52 mm electret filters and a flow rate of 150 L/min, to yield a total sample volume of 36 m³ of air per sample. We used a single-use wet foam carbon compressed elution kit (InnovaPrep, Drexel, MO, USA) to flush the filter following the manufacturer’s instructions, yielding approximately 6 mL of liquid. In a subset of locations, we collected 150 mL grab samples of OWC water for further molecular analysis.

We treated filter eluant with a guanidine thiocyanate-based lysis buffer (UNEX; Microbiologics, St. Cloud, MN, USA) in a 1:1 volumetric ratio, stored these in SK38 bead tubes (Bertin Corp, MD, USA) and cryovials and transported them to our laboratory in Atlanta. We used 900 μL of sample eluant and UNEX mixture for extraction. After DNA extraction per the bacterial/viral UNEX protocol (Hill et al., 2015), we stored extracted nucleic acids in 50–75 μL of 10 mM Tris-1 mM EDTA (pH 8) in a −80°C freezer until further analysis. We vortexed OWC samples and then followed the same extraction and storage procedures as for eluted air samples.

In total, we collected 27 ACI samples near OWCs (13 wet season, 14 dry season) for E. coli enumeration by culture, 71 high-volume samples for molecular analysis near OWCs (13 wet season, 58 dry season), and 4 grab samples of OWC water. The distance to nearest OWCs was a mean of 93 m (range: 1 – 778 m). We further collected 4 ACI (3 wet season, 1 dry season) and 4 high-volume (2 wet season, 2 dry season) control samples in the unimpacted settings far from human habitation. Sampling locations are shown in relation to OWCs in Figure 1.

We conducted absolute quantification of ARGs via droplet digital PCR (ddPCR^™, Bio-Rad, Hercules, CA, USA). ARG targets spanned three major antibiotic groups commonly used in low-income settings and whose ARGs have been detected previously in environmental samples: tetracyclines (tetA) (Guarddon et al., 2011), fluoroquinolones (qnrB) (Cavé et al., 2016), and β-lactams (bla_TEM) (Lachmayr et al., 2009). Resistance to tetracycline in clinical isolates has been reported in multiple studies in Bolivia, including in two urban cities of Bolivia where 93% of E. coli isolates from children’s stool were not susceptible to the antibiotic and another study in La Paz, Bolivia where 28% of ETEC isolates from children’s stool were not susceptible (Bartoloni et al., 2006; Rodas et al., 2011). Furthermore, tetracycline resistant Enterobacteriaceae were detected in ambient air in the same area as ours at 28% prevalence (Salazar et al., 2020) and 50% prevalence (Medina et al., 2020). Fluoroquinolone resistance has been reported in one study in Bolivia where overall, qnr genes were detected in 63% of commensal enterobacteria isolated from healthy children’s stools and specifically, qnrB was detected in 60% of isolates (Pallecchi et al., 2009). Additionally, another study in Bolivia reported fluoroquinolones as a common antibiotic used to treat upper respiratory infections (Cordoba et al., 2017). Class A β-lactamases such as TEM varieties are known to be present in South American countries due to the widespread distribution of β-lactam antibiotics (Villegas et al., 2008). Though the other three assays are specific to the individual gene, the bla_TEM assay used in our analysis incorporates 135 variants within the TEM family of β-lactam resistance including resistance to penicillins, cephalosporins, carbapenems and other antibiotics that have a β-lactic ring in their structure. The TEM β-lactamases can be both ESBL (extended spectrum β-lactamases) and inhibitor resistant and are the most clinically significant. Therefore this assay accounts for a wide range of resistance mechanisms and target drugs that are commonly used in this sampling locations (Lachmayr et al., 2009).

An additional target, integron class 1 (intI1), was included to assess potential genetic mobility and has also been previously detected in environmental media. Through enabling gene cassette movement and their physical association to MGEs, integrons aid in the spread of antibiotic resistance in gram negative bacteria and when present in the environment or in bacteria, they indicate that resistance has either been acquired or may be acquired in the future (Barraud et al., 2010; Gillings et al., 2015; Mazel, 2006). One study by Leverstein-Van Hall et al. found that the detection and presence of integrons in Enterobacteriaceae is strongly associated with resistance to multiple antibiotics and in the case of some antibiotics, predictive of their resistance. (Leverstein-Van Hall et al., 2003).

We experimentally determined LODs for each assay using a probit analysis outlined by Stokdyk et al. (Bivins et al., 2020; Stokdyk et al., 2016). Reaction mixes, conditions, ARG target sequences, and experimentally determined limits of detection (LOD) for each target are described in Table 1.

We conducted a spatial analysis to estimate the relationship between lateral distance from OWCs and ARGs in aerosols, using data from the fourth sampling event (2019). First, we calculated the two-dimensional distance from each sampling location to the nearest OWC segment and then, to account for the city’s variable topography and elevation, we adjusted distances by applying the elevations we extracted from the ASTER GDEM (NASA/METI/AIST/Japan Space Systems and Team, 2009) to the horizontal distances of the sampling points in ArcGIS, calculating the adjusted distance through using Pythagorean theorem. We performed a linear regression analysis in RStudio version 1.1.383 to assess the significance of the relationship between lateral distance from OWCs and ARGs density based on 95% confidence (alpha=0.05), subsequently disaggregating the data based on time of sampling (morning or afternoon) and conducting a multiple linear regression analysis to assess the potential diurnal impacts.

We analyzed high-volume samples for ARG and MI targets (mean volume 48 m³, range 9–216 m³). We set all densities below the experimentally derived LOD to zero. Our LODs can be interpreted as the density at which we are 95% confident that the density detected is accurate. Among samples located <1 km from OWCs (n=71), we detected the MI, intI1, in 47% of samples (n=33) and we detected ARGs qnrB, tetA, and bla_TEM in 11%, 17%, and 68% of samples respectively (n=8,12, and 48; Table 3). We then plotted the density distribution and the corresponding standard error for each target (Figure 2).

To confirm the presence of aerosolized fecal material, we measured culturable E. coli in aerosol samples (Cronholm, 1980; Dueker, 2012; Farling et al., 2019; Rocha-Melogno et al., 2020; Salazar et al., 2020). We detected E. coli in 52% of samples (n = 14) with an average density of 11 CFU/m³ air across all positive samples (Figure 2). We detected culturable E. coli in aerosols with aerodynamic particle sizes of 0.6–7 μm, in 27 % of culturable E. coli under 2.1 μm, the size cutoff for fine aerosol particles. These data indicate that fecal indicator bacteria may linger in the air on a scale of hours, with a settling velocity in air of 0.5 m/hour for a typical particle with 2 μm diameter (Flagan and Seinfeld, 1988), indicating high transport potential in air near OWCs. While ARGs can be free-floating or contained within viable or non-viable cells, we make no inference about the relationship between E. coli and ARGs in this context, instead using E. coli only as a marker for aerosolized fecal waste that may be attributed to many sources in a contaminated urban setting such as La Paz. We do note that E. coli cultured from aerosols near the Choqueyapu River has indicated the presence of a range of resistant phenotypes in members of the coliform group (Medina et al., 2020; Salazar et al., 2020).

We detected all targets above the LOD among OWC-adjacent sites (Table 3). We detected bla_TEM at one control site. Other studies in high-and-middle-income countries have reported comparable ARG absolute densities in contaminated settings where fecal wastes may be enriched, primarily indoors (Gao et al., 2018; Ling et al., 2013). Studies reporting relative abundances of ARGs in outdoor ambient air, such as Li et al. (Li et al., 2018), show that a wide range of ARG and MI subtypes may persist in the environment and shed light on the potential threat of urban aerosol transmission of these contaminants. However, where relative abundance reveals target diversity and presence within the microbial communities of a particular sample, absolute quantification in environmental media allows for characterization in a broader context that can be directly applied to population and environmental exposures through fate and transport modeling and risk assessments. Our results further confirm the range of ARGs present in outdoor ambient air at detectable levels and additionally through absolute quantification allow for public health exposure applications. This suggests widespread distribution of ARGs and associated MIs in ambient air, with uncontained urban wastewater flows now implicated as a potential source in cities with poor sanitation. Once airborne, ARGs and MIs in microorganisms may be transported through the air, inhaled or deposited on surfaces and fomites and subsequently ingested (de Man et al., 2014). Free floating ARGs and MIs may be picked up by other microbes in the environment through horizontal gene transfer. Both mechanisms may contribute to the dissemination of AR in the environment and potentially pose a public health threat to the affected populations.

To confirm the presence of targets in OWCs, we collected samples during the third sampling event (n = 4). We detected intI1 and bla_TEM in all OWC samples and at the highest averages (1.4×10⁸ and 6.4×10⁷ gc/100mL respectively). We detected qnrB and tetA in 3/4 OWC samples at averages of 1.7×10⁷ and 1.4×10⁷ gc/100 mL respectively, indicating AR contaminated surface waters that are likely contributing to the dissemination and proliferation of AR in the surrounding environment. Comparably, ESBL-producing bacteria, such as bla_TEM, have been detected in the Choqueyapu River previously (Guzman-Otazo et al., 2019; Poma et al., 2016). Although there are limited data on antibiotic usage in Bolivia, previous studies have reported that in some areas in Bolivia and more widely in South America, the most commonly used antibiotics include penicillin, ampicillin, and amoxicillin (Bartoloni et al., 1998; Cordoba et al., 2017), all members of the β-lactam family. High usage of these drugs combined with poor wastewater treatment may lead to the release and spread of resistant organisms. A study in Cochabamba, Bolivia, a city south-east of La Paz, detected the presence of β-lactam resistance encoding ARG variants (also covered by the bla_TEM ARG target we used in our study) in rivers that flow throughout the city (Saba Villarroel et al., 2017). Our results support these studies that β-lactam resistance encoding ARGs, in addition to tetracycline and fluoroquinolone resistance encoding ARGs are highly prevalent in these urban settings.

We sought to assess the relationship between ARGs and one MI near OWCs. From the randomly generated points (n=50) for the final sampling event, we only sampled at 25 sites due to time restraints and in some cases, lack of accessibility. One site we sampled at was observed to be much further (~450 m) than within the intended buffer because the actual river location varied from the model near this point. We excluded the 2 samples taken at this site from further analysis and the sample taken at the control site. Hence, we assessed 46 samples from 23 sites in our spatial model, all located within 150 m from OWCs. We performed a linear regression calculation for target densities as distance from OWCs increased and examined the relationship between lateral distance and OWCs for all targets detected (Figure 3).

Despite observed mean higher densities of targets at locations close to OWCs, specifically for bla_TEM, overall regression results showed no significance between distance from OWCs and ARG density based on 95% confidence both when separated by time of day and when not separated. However, when not separated by time of day we observed a tendency of bla_TEM density in aerosols to decrease (p=0.082) as distance from OWCs increases within the 150 m. The apparent decrease may indicate that proximity to OWCs is important when assessing human risk through exposure, though we cannot rule out that this observation is attributable to chance.

Our interpretation of findings is constrained by the study limitations. Our data are observational and limited in time and space, where phenomena such as aerosolization of waste and aerosol transport may be highly variable according to local conditions. We identified a limited number of ARGs to assess a priori, and these may or may not be the most relevant targets for the study area or for exposure relevance more generally. Additionally, we acknowledge that though identification of low gene target densities via ddPCR is improved when compared to qPCR (Cavé et al., 2016), false positives are possible (Cao et al., 2015) even though we have applied conservative estimates of LODs derived experimentally. Though the detection of E. coli in air near OWCs indicates the presence of fecal waste in aerosols, we cannot unambiguously attribute E. coli to these OWCs. Additionally, though ARGs are present in air near OWCs where the same ARGs are also present, we cannot unambiguously identify the OWCs as the source.