Sampling sites corresponded to sites representing different land-use areas in the Maipo and Mapocho Rivers, both located in the central macrozone of Chile. The Maipo crosses the Metropolitan and Valparaiso Regions. The Maule crosses the Maule Region of the country. The Maipo is the primary source of drinking water for the Metropolitan region and supplies approximately 70% of the drinking water and 90% of the irrigation demands (DGA, Dirección General de Aguas, 2004a). Similarly, the Maule provides irrigation to approximately 118,263 hectares and drinking water to approximately 99,6% of the urban population (DGA, Dirección General de Aguas, 2004b). Fecal coliforms levels in the Maipo and Maule Rivers have exceeded levels permitted by the Chilean water quality standard (INN, Instituto Nacional de Normalización, 1978). For example, fecal coliforms in the Maipo River ranged between <2 and 8.50 × 10⁶ fecal coliforms/100 ml in past reports (DGA, Dirección General de Aguas, 2004a), while the Maule River between 540 and 1.1 × 10⁴ fecal coliforms/100 ml (DGA, Dirección General de Aguas, 2004b).

Land use throughout each river system was categorized using information from the Land Use Atlas of the Chilean Military Geographic Institute (Instituto Geográfico Militar de Chile, 2005). The land-use categories considered were as: (i) natural, (ii) agriculture, (iii) urban, and (iv) livestock and/or forestry. The Maipo and Maule Rivers share similar land use classifications, allowing a comparison between land use zones and both river courses. The natural zone represents an area with low influence of urban, agricultural, and/or livestock activity. At the same time, the other sites have large urban, agricultural, or livestock/forestry activities, respectively. Both the Maipo River’s natural and agricultural land use areas receive the discharge of two wastewater treatment plants (WWTP), causing these zones to receive urban discharges.

For this reason, river water samples were collected before and after WWTP effluent discharge to determine whether this action influences the microbial source tracking by introducing Cryptosporidium spp. and Giardia spp. species that are human host specific into agricultural or natural areas. The selected urban sites corresponded to the discharge of rivers that have an important influence of major cities; for instance, the Mapocho River crosses the Chilean capital and major metropolis (Santiago) and discharges in the studied site in the Maipo River, and the Claro River receives the urban influence of the capital and major city in the Maule region (Talca) and discharges in the area sampled in the Maule River (Figure 1).

Water samples were collected from each river every 3 months between June 2017 and April 2019. The timing of sampling was selected to ensure sampling during each season: summer (December–March), fall (March–June), winter (June–September), and spring (September–December). Fall and winter sampling events correspond to the rain event, while spring and summer sampling events correspond to the dry season. All sampling sites were georeferenced using Google Maps. Water was collected in a sterile plastic bottle at each sampling site, which was rinsed three times in the river before collecting the sample. The sample was collected by submerging the bottle at least 10 cm below the surface. Immediately after sample collection, bottles were placed in a cooler and transported on ice. Samples were all processed within 24 h. Physical parameters of pH, water temperature (°C), conductivity (μS or mS), salinity (ppm or ppt), and total dissolved solvents (ppm or ppt) were measured in-situ in each site using the Yalitech AM 006 Waterproof Multiparameter Meter Combo 6. Weather conditions (sunny, cloudy, or raining) were also recorded at the time of sampling.

Water samples (15 ml) were used for fecal coliform enumeration following the Chilean Regulation NCh2313/22:1995 protocol for the detection of fecal coliforms in wastewater using EC medium [Instituto Nacional de Normalización (INN), 1995] to determine if each sample was complaint with Chilean water quality standards (INN, Instituto Nacional de Normalización, 1978) and whether the river complied with the water quality standards for surface waters. Specifically, based on this regulation, we considered that a sample was compliant with the standard if fecal coliforms levels were below 1,000 fecal coliforms /100 ml of water; compliance is based on the assumption that fecal coliform levels above 1,000 fecal coliforms/100 ml is indicative of fecal contamination. The most probable number (MPN) of fecal coliforms in 100 ml of water was conducted following the instructions of the Chilean Regulation NCh2313/22:1995 protocol.

To identify possible sources of fecal contamination, 20 L water samples were concentrated to 200 ml using Hollow-Fiber Ultrafiltration and B Braun DIACAP HIPS 20 (catalog code: 10-10-149144-5) hemodialysis filters following the methodology described by Hill et al. (2005) and Bambic et al. (2011). Cryptosporidium spp. oocysts and Giardia duodenalis cysts in the ultra-filtrated water were quantified according to the United States Environmental Protection Agency (EPA) method 1623 (Miller et al., 2005). Cryptosporidium spp. oocysts and Giardia duodenalis cysts were concentrated using Dynabeads Cryptosporidium/Giardia Combo kit (Applied Biosystems, IDEXX Laboratories Inc., Maine, United States) immunomagnetic separation (IMS). IMS was conducted following Miller et al. (2005) with some modifications. Since metallic sediments interfered with the IMS step, a sediment wash was performed before adding the 100 μl IMS Dynal Cryptosporidium spp. and Giardia duodenalis beads to remove metallic sediments. Levels of each target were stained using Waterborne kit reagents (Aqua-Glo G/C Direct; Waterborne Inc., New Orleans, LA) per the manufacturer’s instructions and enumerated according to Environmental Protection Agency (EPA) method 1623 (Miller et al., 2005). Briefly, 3-well slides (well diameter 14 mm) were stained with fluorescein isothiocyanate (FITC), and DAPI (4,6-diamidino-2-phenylindole) and the entire well was examined for protozoa using an Olympus epifluorescence microscope. Organisms were visualized at ×200 magnification, and identifications were confirmed at × 400 (Miller et al., 2005). Cryptosporidium parvum and C. hominis oocysts were identifiable as 5–7 μm-diameter spheres with apple green outlines, often with a midline seam. Cryptosporidium spp. oocysts were identified as 5–7 μm-diameter spheres outlined in apple green and often with a midline seam, whereas Cryptosporidium andersoni/C. muris like organisms were identified as 5-by-7-μm elliptical forms (Miller et al., 2005). Giardia cysts were also apple green but oval and 9–14 μm long (Miller et al., 2005). The same experienced microscopist read all slides. As described below, samples were then subjected to DNA extraction and conventional PCR for confirmation and sequencing of Cryptosporidium spp. and Giardia duodenalis.

A modified version of a previously published method (Miller et al., 2005) was used for extracting DNA from all the slides. After protozoa quantification the slide well for each sample was scraped with a scalpel blade and washed with sterile PBS into a separate microcentrifuge tube. Proteinase K (40 μl) was added to the sample and kept at 56°C overnight for sample digestion. DNA was extracted using Qiagen DNA Minikit (Qiagen Inc., Valencia, CA), and subsequently purified using a QIAamp column. Extracted DNA was stored at −20°C until PCR analysis.

The Cryptosporidium PCR was based on molecular characterization of the 18S rRNA gene using nested primers that amplify ~825 bp product (Xiao et al., 1999, 2000), and using conventional primers that amplified a 298-bp DNA fragment (Morgan et al., 1997). Amplification conditions were described by Adell et al. (2014). The Giardia PCR was based on molecular characterization of the glutamate dehydrogenase (GDH) and the SSU rRNA genes. GDH confirmation amplifies a ~432-bp fragment using a semi-nested-PCR protocol and primers (Read et al., 2004), while a ~292 bp final fragment of the SSU rRNA gene was amplified using a nested PCR protocol and primers (Appelbee et al., 2003). All primers are listed in Supplementary Table S1. All DNA was purified and submitted to Macrogen (Macrogen, Inc. Korea) for sequencing.

The protozoan parasites Cryptosporidium spp. and Giardia spp. have been used as an MST tool because both provide knowledge of host specificity (Ryu et al., 2011; Prystajecky et al., 2014; Xiao et al., 2018). For instance, the genus Cryptosporidium is composed of genotypes that are host specific and genotypes that infect various hosts (Xiao et al., 2000). For example, the major hosts for C. parvum are cattle, other ruminants, and humans; for C. andersoni cattle and bactrian camels; for C. baileyi, poultry and birds; for C. meleagridis, humans, turkeys, and birds; for C. canis, dogs; and other 10 species that have specific hosts (Xiao et al., 2004; Cacciò et al., 2005).

A situation also occurs with Giardia genotypes and assemblages (Feng and Xiao, 2011). Giardia spp. are divided into six species based on morphological characteristics, among which Giardia duodenalis (synonyms G. lamblia and G. intestinalis) infects a variety of mammals, including humans and livestock (Cacciò et al., 2005) and will be referred to as G. duodenalis throughout the manuscript. G. duodenalis also have several genetic groupings (henceforth assemblages), each of them having specific hosts. Thus, assemblage A can infect humans, other primates, livestock, dogs, cats, rodents, and other wild animals; similarly assemblage B is specific to humans and other primates, dogs, some species of wild mammals, assemblage C and D (also reported as Giardia canis) are specific to dogs and other canids; assemblage E (also reported as G. bovis) to cattle and other hoofed livestock; assemblage F (also reported as G. cati) to cats; and assemblage G (or G. simondi) to rats (Cacciò et al., 2005; Monis et al., 2009).

Therefore, in our study the idnetifying the genotype of Cryptosporidium and assemblage for Giardia duodenalis will be used to provide information on the host source (animal or human) of origin of the fecal contamination, its infective potential to productive animals (cattle), and its zoonotic potential.

To allow for the identification of G. duodenalis and Cryptosporidium subtypes associated with specific hosts and, therefore, fecal sources, three custom databases (Giardia_SSU, Giardia_GDH, and Crypto_18S) were built by searching the NCBI-nr database and publicly available literature. DNA sequences were aligned using ClustalW-2.1 (GNU Lesser GPL, 2010), and phylogenies were built in Molecular Evolutionary Genetics Analysis (MEGA) version 10.2.5 software (Kumar et al., 2018) using maximum likelihood and Tamura 3. Phylogenetic tree topological reliability was determined using the bootstrap method, with 1,000 replicates and bootstrap lower than 70% were deleted from the tree. Accession numbers with the nucleotide sequence data for PCR products used for phylogenetic classification obtained during this study are available in Supplementary Table S2.

Conditional inference trees (CTree) are a type of modified classification and regression tree (CART) to address fundamental limitations of traditional CART algorithms (e.g., bias toward using categorical variables with many levels/continuous variables for splits over categorical variables with fewer levels). It is important to note that this is a multivariable and not a multivariate analysis and is a standard approach used in environmental microbiology literature, including by water-focused studies throughout the past decade (Bae et al., 2010; Strawn et al., 2013; Verhougstraete et al., 2015; Weller et al., 2016, 2020a; Green et al., 2021; Guo and Lee, 2021; Hannan and Anmala, 2021). Ctrees and CARTs are ideal tools for capturing complex (e.g., hierarchical) relationships within data and are thus, well suited for observational studies in non-controlled environments (Kuhn and Johnson, 2016). For example, identifying combinations of factors (specific scenarios) associated with an increased or decreased likelihood of detecting each microbial target, which is something regression-based approaches cannot do (Kuhn and Johnson, 2016). Tree-based approaches also create easy-to-understand visuals and do not rely on significance testing, thereby avoiding limitations associated with multiple comparison testing (Kuhn and Johnson, 2016). Moreover, tree-based analyses do not generate effect estimates or odds ratios to quantify the strength of associations, which is worth mentioning as other studies have reported substantial variability in water quality within a waterway over time; therefore, the duration of our study was insufficient to generate reliable effect estimates (Kuhn and Johnson, 2016; Weller et al., 2020b,c).

Conditional tree analysis was used to characterize the relationship between environmental factors and (i) fecal coliform levels log transformed (log10 fecal coliforms/100-ml), (ii) whether fecal coliform levels exceeded the Chilean water quality standard of 1,000 fecal coliforms/100-ml (Yes/No), and (iii) if Giardia spp. were detected by PCR (Yes/No); tree analyses were implemented using the mlr and party packages and visualized using the partykit as previously described (Strobl et al., 2007a,b, 2009; Boulesteix et al., 2015; Weller et al., 2020b,c). Briefly, hyperparameters were tuned, models trained, and performance assessed using three-fold cross-validation repeated 10 times; cross-validation was performed to optimize R-squared (for continuous outcomes) and area under the curve (AUC; for binary outcomes). For imbalanced, binary outcomes (i.e., Giardia detection versus non-detection), SMOTE resampling was performed as part of tuning (Kuhn and Johnson, 2016; Weller et al., 2021).

To minimize the potential for overfitting we: (i) tuned the maxdepth parameter, (ii) limited the upper bound of maxdepth to 6, and (iii) used a mincriterion of 0.95. Conditional forests analysis was used since it is robust to correlation and missingness in explanatory factors (by enabling surrogate splits), can handle both categorical and continuous outcomes, and can capture complex, hierarchical relationships (e.g., interactions) between explanatory factors. The environmental features considered were the river system (i.e., Maule versus Maipo); predominant land use at the site [i.e., plant-based agriculture (henceforth agricultural), livestock operations, natural, urban sites not at a wastewater discharge (henceforth urban) and urban sites with a wastewater discharge (henceforth treated urban)]; season (i.e., wet versus dry); if feces, garbage, or livestock were present at the sampling site; pH; water temperature; total dissolved solids; salinity; if it rained during the 24 h before sampling; and weather conditions at the time of sampling (i.e., rainy, sunny, partly cloudy, or cloudy). For Giardia, coliform levels and if coliform levels met the Chilean water standard (i.e., were above or below 1,000 fecal coliforms/100-ml) were also considered explanatory factors. Separately, nonparametric Spearman’s one-tailed tests were conducted to assess the correlation between continuous outcome variables (i.e., MPN of fecal coliforms/100 ml, Giardia cysts/100 ml, and Cryptosporidium oocysts/100 ml) to enable comparison of the findings reported here with previous studies (e.g., Wu et al., 2011; Korajkic et al., 2018). A generalized linear model (GLM) was implemented to determine whether compliance with the Chilean water quality standard (i.e., fecal coliforms above versus below 1,000 fecal coliforms/100-ml) was associated with protozoa detection (i.e., if Cryptosporidium, Giardia, or no protozoa were detected). In both analyses (Spearman’s correlation and GLM), differences with p < 0.05 were statistically significant. All statistical analyses were performed using RStudio v1.1.456 or R v3.6.3 (R Core Team, 2020; RStudio Team, 2020).

To determine if multi-drug resistant E. coli were present in water samples, a subset of samples (from the first four of the eight total samplings) were tested (n = 48 total samples were tested). Briefly, 1 ml ultra-filtered samples were enriched in 5 ml of Buffer Peptone Water and incubated at 37°C for 24 h. Following incubation, 100 μl of the samples were streaked onto plates with MacConkey agar supplemented with Ciprofloxacin (1 mg/L), Cefotaxime (1 mg/L), and Tetracycline (4 mg/L) and incubated at 37°C for 24 h. Colonies were selected according to morphology using a magnifying glass. This selection was made according to standard patterns of color (pink), as described previously (Jung and Hoilat, 2020). All isolates were identified using MALDI-TOF (BioMérieux, Marcy l’Etoile, France) and only the isolates identified as E. coli were tested against a panel of 15 antibiotics using the disk diffusion method to assess their antimicrobial susceptibility profile following CLSI guidelines (Clinical and Laboratory Standards Institute, 2018a). The antibiotic used were: Ampicillin (AMP, 10 μg); Cefazolin (CFZ, 30 μg); Ceftriaxone (CRO, 30 μg); Cefepime (FEP, 30 μg); Imipenem (IPM, 10 μg); Ciprofloxacin (CIP, 5 μg); Amikacin (AMK, 30 μg); Gentamicin (GEN, 10 μg); Fosfomycin/Trometamol (FOF, 200 μg); Amoxicillin-Clavulanate (AMC, 30 μg); Cefuroxime (CXM, 30 μg); Aztreonam (ATM, 30 μg); Chloramphenicol (CHL, 30 μg); Tetracycline (TET, 30 μg); and Trimethoprim/Sulfamethoxazole (SXT, 1.25/23.75 μg). All antibiotics were supplied by OXOID (Hampshire, England). Isolates resistant to three or more antimicrobial classes were cataloged as MDR (Multi-Drug-Resistant Bacteria) following previously standardized criteria (Magiorakos et al., 2012).

To further characterize environmentally resistant bacteria, 50 ml of water samples from Maipo and Maule Rivers, were passed through a 0.22-μm filter. This filter was resuspended in 0.85% saline solution, and 200 μl were seeded in TSB and R2A agar media in 120 mm square plates. These plates were incubated at 37°C, 25°C and 15°C, and the colonies were classified according to the standard pattern: shape, color, texture, and shape of the colony border and designated as morphotypes (Higuera-Llantén et al., 2018). A total of 2631 morphotypes were isolated at all sampling points, media, and temperatures. Minimum Inhibitory concentrations (MICs) for Ampicillin (AMP), Erythromycin (ERY), Ceftazidime (CAZ), Imipenem (IMI), Ciprofloxacin (CIP), Gentamicin (GEN) and Tetracycline (TET) were determined by agar dilution methods as recommended by CLSI (Clinical and Laboratory Standards Institute, 2018b). Briefly, serial two-fold dilutions ranging from 0.1 to 2,048 μg/ml of any antibiotic were added into Mueller-Hinton agar (Difco). After overnight growth, all isolates were resuspended in 0.85% saline solution and inoculated in plates using a 96-needle replicator. The isolates that grew at 37°C were incubated for 24 h, while those that grew at 25°C and 15°C were incubated for 48 h. A Mueller-Hinton agar plate without antibiotics was used for growth and possible contamination control. These experiments were performed in triplicate. MIC was defined as the lowest concentration of antibiotic that inhibits the growth completely at least in two of three plates seeded at any temperature. As a reference, we used the strain Escherichia coli ATCC 25922, as recommended by CLSI (Clinical and Laboratory Standards Institute, 2018b).

Fecal coliform levels in the Maule River ranged between 1 and 130 MPN/100-ml, not exceeding the Chilean water quality standard of 1,000 fecal coliforms/100-ml (Table 1). Conversely, fecal coliform levels in the Maipo ranged between 2 and 30,000 MPN/100-ml in 11 samples exceeding the water quality standard. Most of the Maipo River samples that exceeded the standard were mainly collected from urban sites. In both rivers, fecal coliform levels varied based on season, with higher levels during the dry season (mid-September to mid-March) compared to the wet season (mid-March to mid-September; Table 1). Fecal coliform levels also appeared to vary based on land use (Table 1). The samples with the highest coliform levels were collected from urban sites in both rivers, while the samples with the lowest coliform levels were collected from natural sites (Table 1). While the range in fecal coliforms levels from treated urban sites (i.e., sites in the river at a wastewater discharge; <2–8,000 CFU/100-ml) overlapped the range for urban sites (i.e., sites not at a discharge; 34–30,000 CFU/100-ml), the mean level in treated sites (6,887 CFU/100-ml) was considerably lower than the mean in urban sites (32,705 CFU/100-ml); this difference was even more pronounced when only Maipo River sites were considered.

According to conditional tree analysis, fecal coliform levels were significantly lower in the Maule compared to the Maipo River (p < 0.001; Figures 2A,B). Specifically, fecal coliforms were lowest in Maule River sites and highest in samples from Maipo River sites where the predominant land use was agricultural (i.e., plant-based production) or urban (i.e., not at a wastewater discharge) as opposed to livestock, natural, or treated urban (i.e., urban sites at a wastewater discharge; Figure 2A). All samples collected from the Maule River met the Chilean water quality standard, and this river was significantly associated with if a water sample was compliant with the standard (p < 0.01; Figure 2B). For Maipo River, samples were most likely to be non-compliant if collected from sites where the predominant land use was agricultural, urban, or treated urban during the dry season. Samples were most likely to be compliant if collected from sites where the land use was natural or livestock-based production (Figure 2B). For urban, treated urban and agricultural sites, samples were more likely to meet the standard if collected during the wet season (p = 0.002; Figure 2B).

Higher Cryptosporidum spp. counts were found in water samples collected from the Maipo River (range = 0–60 oocysts/10 L; mean = 3.18 oocysts/10 L) compared to the Maule River (range = 0–31 oocysts/10 L; mean = 2 oocysts/10 L; Supplementary Figure S1). Higher Giardia spp. counts were also found in the Maipo (range = 0–171 cysts/10 L, and a mean of 7 cysts/10 L) compared to the Maule River (range = 0–4 cysts/10 L, and a mean of <1 cysts/10 L; Supplementary Figure S1). In Maipo River samples, two assemblages of G. duodenalis were identified, including assemblages A or B (Figure 3; Supplementary Tables S2, S3). Only two G. duodenalis were detected in water samples collected from the Maule River. One of them corresponds to assembly A or B, and the other is assembly G. Also, four Cryptosporidium spp. were identified in the Maipo River, including C. parvum, C. andersoni, C. meleagridis, and C. canis (Figure 4; Supplementary Tables S2, S3). Sequencing analysis did not identify any Cryptosporidium spp. in the Maule River (Figure 4; Supplementary Tables S2, S3).

Most Cryptosporidium-positive samples were collected during the wet season. Whereas Giardia-positive samples were collected in both seasons, more positives were collected during the dry season. While fecal coliform levels were highest in samples collected from agricultural and urban sites, G. duodenalis assemblages A or B, and C. meleagridis were only detected in samples collected from agricultural sites (i.e., sites with plant-based production; Table 1; Figure 3; Supplementary Tables S2, S3). While all, except one sample, collected from sites upstream of wastewater discharges (i.e., natural or agricultural sites) were Giardia-negative, multiple samples (n = 8 samples) collected downstream of wastewater discharges (i.e., treated urban sites) were positive for G. duodenalis assemblages A or B (Supplementary Table S3). Indeed, conditional tree analysis indicated that the likelihood of detecting Giardia by PCR was greatest in samples collected downstream of wastewater discharges (i.e., treated urban sites; p < 0.001) and lowest in samples collected from all other sites during the wet season (Figure 5).

Spearman’s correlation analysis indicated that coliforms counts (MPN/100 ml) significantly, and positively correlated with Giardia counts (cysts/100 ml; correlation coefficient = 0.31; p = 0.002) but were not significantly correlated with Cryptosporidium counts (oocysts/100 ml; correlation coefficient = 0.03; p = 0.75). However, the lack of a correlation between coliform and Cryptosporidium spp. levels may be due to the large number of Cryptosporidium-negative samples. According to GLM analysis, neither G. duodenalis (OR = 0.84; p = 0.896; 95% Confidence Interval = 0.20, 6.33) or Cryptosporidium spp. (OR = 0.48; p = 0.531; 95% Confidence Interval = 0.06, 9.95) detection were associated with fecal coliforms levels being compliant with the Chilean water quality standard of 1,000 CFU of fecal coliforms/100-ml.

To evaluate antimicrobial resistance in enteric bacteria, E. coli were isolated, and their resistance profile was characterized. Twenty-eight out of 48 samples analyzed (58.3%) contained Gram-negative bacteria resistant to at least one of the antibiotics tested (Ciprofloxacin, Cefotaxime, and Tetracycline). We obtained 66 isolates; 26 (lactose positives) isolates were characterized at the species level by MALDI-TOF MS. Finally, we identified eight E. coli isolates and three of them showed a multi-drug resistant profile (Figure 6; Supplementary Table S4). Regarding the environmental bacteria, 2,631 morphotypes were isolates at different temperatures (37°C, 25°C, and 15°C) and media. The major number of colonies were obtained at 25°C and in the R2A medium, demonstrating that the optimal temperature for environmental bacterial growth is 25°C. To determine the AMR profile, the MICs for Ampicillin (AMP), Erythromycin (ERY), Ceftazidime (CAZ), Imipenem (IMI), Ciprofloxacin (CIP), Gentamicin (GEN), and Tetracycline (TET) were determined. Bacteria that showed resistance to at least three antibiotics of different families were classified as MDR bacteria. The percentage of MDR-bacteria was evaluated by season and land use and summer was the season with higher values in most of the land use analyzed (>2.5%), followed by the winter season where a high percentage of resistance was observed in both urban and livestock land uses (>2%; Figure 7).