Widespread occurrence of fecal indicator bacteria in oligotrophic tropical streams. Are common culture-based coliform tests appropriate?

Study site

This study was carried out at 21 streams in the Panama Canal Watershed (PWC), Panama, for one year during 2018 to 2019 (Figs. 1A–1D). Water samples were collected from 17 streams in the western part of the Panama Canal watershed (Fig. 1C) and in four experimental watersheds within the Agua Salud facility of the Smithsonian Tropical Research Institute (STRI) in the eastern Panama Canal watershed (Fig. 1D). Each stream was bordered by one of four land uses that are common in the Neotropics: (1) mature forests (>80 years in age (MF)); (2) secondary forests (<25 years in age (SF)); (3) silvopastures (SP) and; (4) traditional cattle pastures (CP). Silvopastures differ from traditional cattle pastures in that the stream channels are bordered by riparian forest, streams are fenced to allow cattle drinking water access at only one point, and more trees are left in the pastures. All streams were specifically selected because their adjacent land use was not confounded by other upstream uses. Landowners granted permission to sample water on their properties. This research had sampling permission from the Republic of Panama Ministry of Environment by permits (SE/PO-2-19) and (SE/AP-2-18).

Sampling and physicochemical measurements

Biweekly water samples and physicochemcial measurements were collected from each stream following protocols described in Chavarria et al. (2021), Murphy & Stallard (2012) and Paton (2020). Most samples were collected between 8:00 A.M. and 12:00 P.M. during baseflow conditions, and all samples were transported at 4 °C to STRI’s laboratory for processing within three hours of collection (Chavarria, 2021).

Following Chavarria (2021), turbidity, pH, temperature, conductivity, total dissolved solids and dissolved oxygen were measured and recorded on site during sample collection. Turbidity was measured with a field-portable turbidimeter (MicroTPW, HF Scientific Inc., Fort Myers, FL, USA). Temperature, pH, conductivity and total dissolved solids were measured with a HANNA portable multiparameter meter (HANNA Instruments, Smithfield, RI, USA) and dissolved oxygen was measured using a portable dissolved oxygen meter (HANNA Instruments, Smithfield, RI, USA), following manufacturer’s recommendations.

Total coliform and E. coli concentrations were determined from 100 mL water samples using the IDEXX Colilert^®/Quanti-tray 2000 culture assay, following the manufacturer’s recommendations (IDEXX Laboratories, Inc., Westbrook, ME, USA) (Chavarria, 2021). Samples were incubated at approximately 35 °C for 24 h and positive wells were counted. The IDEXX Quanti-Tray^®/2000 MPN Table was used to convert the number of positive wells to the most probable number (MPN). Microbial detection limits for total coliform and E. coli had a lower bound of <1 MPN/100 mL and an upper bound of >2419.6 MPN/100 mL (Chavarria, 2021). Field and laboratory positive (internal pure-culture samples) and negative controls (consisting of autoclaved water exposed to the field and laboratory environment) were measured periodically during sampling, and contamination was not detected.

16S rDNA metabarcoding was also used to assess the presence of coliforms and E. coli in the total bacterial community. Water samples (1L) were collected and processed according to protocols described in Chavarria et al. (2021). Briefly, the water was filtered in a sterile fashion using a vacuum pump filtration system (Laramie et al., 2015) and DNA was isolated from 0.22 µm mixed cellulose filters (EDM Millipore, Burlington, MA, USA) using the DNeasy PowerSoil Kit (QIAGEN, Hilden, Germany). The V4 region of the 16S rRNA gene was amplified in triplicate 12.5 uL reactions using a two-step PCR protocol using the phased primers 515F (Parada, Needham, & Fuhrman, 2016) and 806R (Apprill, Parsons & Weber, 2015) (Caporaso et al., 2011). Negative and positive controls containing nuclease-free water and previously PCR-positive samples were included in all PCR amplification steps. After a second PCR to add Illumina adaptors and sample specific barcodes, products were cleaned and normalized using normalization plates (Charm Biotech, San Diego, CA, USA) and pooled into two sequencing libraries (Chavarria, 2021). The libraries were further concentrated using magnetic beads, quantified by Qubit High Sensitivity dsDNA Assay (Thermo Fisher Scientific, Waltham, MA, USA), and quality checked by a Bioanalyzer dsDNA High Sensitivity assay. The resulting libraries were sequenced on an Illumina MiSeq with V2 chemistry on a 2,250 paired-end runs at the STRI molecular laboratory (Chavarria, 2021). A more in depth look at bacterial communities from this sequencing effort for the Agua Salud sites can be found in Chavarria et al. (2021).

Data analysis

Statistical tests for total coliforms and E. coli data were performed on transformed log values following (Chavarria, 2021). One-half of the lower detection limit was substituted for values below the detection limit, and the upper detection limit was substituted for values above the maximum limit. Untransformed data were used for physicochemical parameters. All tests of significance used the non-parametric Kruskal-Wallis test with Benjamini–Hochberg corrections, as implemented in R v3.6.0 and the Coin package (R Core Team, 2012) (Table S1). Values were considered significant at p <0.05 (Chavarria, 2021).

The QIIME2 platform v2022.4 was used for the majority of the analysis of paired-end sequences, as described in Chavarria et al. (2021). Briefly, preprocessing consisting of primer trimming, sequence quality control, and forward and reverse read merging using QIIME2’s Cutadapt and DADA2 plugins to produce amplicon sequence variants (ASVs) (Bolyen et al., 2019; Callahan et al., 2016; Martin, 2011). Taxonomy was assigned using a naïve Bayesian classifier trained on the SILVA 99% sequence similarity database (v132) (Pruesse et al., 2007; Wang et al., 2007) and ASVs classified as mitochondria or chloroplast were removed from further analysis. The sequencing dataset was further filtered to only include ASVs assigned to genera classified as coliforms and genera that have previously been identified to cause false positives in culture-based indicator tests for total coliform and E. coli (Chao, 2006; Pisciotta et al., 2002). Genera classified and grouped as coliforms include Escherichia, Citrobacter, Enterobacter, Hafnia, Serratia, Klebsiella, and Yersinia (Leclerc et al., 2001; Yoon et al., 2013).

Physicochemical parameters

Between 2018 and 2019, 21 streams were monitored and regularly sampled which resulted in a total of 315 samples collected, with a distribution of 13–19 samples per site. Water quality based on physicochemical parameters varied across land uses and sites (Table 1). Notably, temperature was found to be significantly different across land uses, with overall temperatures close to one degree Celsius higher in CP and SP sites than forested sites (Table 1, Table S1). pH was significantly higher in forested sites (MF and SF) than sites with cattle present (SP and CP). Turbidity was significantly higher in CP sites than forested sites (Table S1). Highest conductivity and total dissolved solids (Table S1) measurements were found in MF sites, while the lowest oxygen concentrations were found in SP and CP sites, although they were not significantly different from forested sites (Table 1, Table S1).

Coliform indicators and seasonal changes

Total coliforms were found to be ubiquitous across all sites and there were no significant differences in concentrations between the different land uses when aggregating all measurements by the different land-use types (Table S1, Fig. 2A). Only two samples had undetectable levels of coliform bacteria (two MF samples). While some differences between sites within a land use and seasons within sites were observed, the majority of these comparisons were not significant (Figs. S1 and S2).

Escherichia coli was also detected at all sites and no significant differences were found between land use types (Table S1, Fig. 2B). Eighty seven percent of samples had measurable concentrations of E. coli (274/315 samples) and undetectable concentrations within each land use was below 15% (MF = 14.0%, SF = 12.3%, SP = 10.8%, CP = 14.3%). However, concentrations were more variable, and no seasonal shifts were found; only a few differences in concentrations were observed between the dry and wet seasons at sites where cattle were present (Fig. S3).

Identification of coliform bacteria with high-throughput sequencing

Using 16S rDNA metabarcoding, we detected several genera that are classified as total coliforms and found that their relative abundances were more variable across land uses, sites, and seasons than seen in our culture-based tests results (Fig. 3A and Fig. S4). Total coliform represented less than 2% of the bacterial communities across all sites (for full Agua Salud sites data please see Chavarria et al., 2021) and coliform bacteria abundance was significantly lower in CP sites than other land uses (Table S1, Fig. 3A). As coliform bacteria are known to be ubiquitous in the environment and their presence alone does not signify fecal contamination in water environments (McMath et al., 1999; Somaratne & Hallas, 2015), it was not surprising to detect them across all land uses. Although the relative abundance of coliform bacteria was not significantly different across MF, SF, and SP (for Agua Salud sites), significant differences were found in the overall bacterial communities across these land-use sites (Chavarria et al., 2021). However, metabarcoding detected E. coli at only five of our 21 sampling sites (seven samples in total: MF = 3, SF =1, SP = 2, CP = 1) and where it was found, it represented only a small proportion of the total coliform community (Figs. 3A and 3B).

To better understand why Colilert^® tests results showed high concentrations with no differences across land uses while metabarcoding detected variable abundances of total coliform and few samples containing E. coli, we did further analysis of ASVs assigned to taxa that have been reported to cause false positive results in culture-based tests like Colilert^® that are based on the enzymatic properties β-galactosidase (total coliforms) and β-glucuronidase (E.coli) (Maheux et al., 2015). Aeromonas (Gammaproteobacteria) and Pseudomonas (Gammaproteobacteria) were among the most abundant genera capable of producing false positives results in total coliforms (Fig. 4A, Fig. S5) and are both known to be ubiquitous in aquatic systems (Usui et al., 2016; Valério, Chaves & Tenreiro, 2010). Flavobacterium (Bacteroidetes), Klebsiella (Gammaproteobacteria), Pantoea (Gammaproteobacteria), and Serratia (Gammaproteobacteria) were the most abundant genera capable of causing false positives for E. coli (Fig. 4B, Fig. S6). These genera are common in soil, water and associated plants (Bernardet & Bowman, 2006; Johnson et al., 2007; Lopez-Torres, Hazen & Toranzos, 1987; Thompson et al., 2007). All sites showed presence of one or more of these genera with higher relative abundances in mature forest sites (Fig. S6, Table S1), creating multiple opportunities for false positives in the Colilert^® test. Thus, non-target bacteria that are part of the natural ecology of these tropical oligotrophic water environments and their surroundings may confound traditional FIB culture-based water quality indicators in these environments.

To further assess the potential for fecal contamination, we searched for sequences that matched other bacterial members known to be part of the intestinal tract of warm-blooded animals and common fecal contaminants. We found several families in the order Bacteroidales with relative abundances that were two times higher in streams influenced by pastures than forested streams (Fig. 5). Bacteroidales (phylum Bacteroidetes) are commonly found in the feces of humans and other warm-blooded animals and their presence in water is an indication of fecal pollution and the possible presence of enteric pathogens (Ridley et al., 2014; Tambalo et al., 2012).