Effects of chronic exposure to arsenic on the fecal carriage of antibiotic-resistant Escherichia coli among people in rural Bangladesh

Antibiotic resistance is a leading cause of hospitalization and death worldwide. Heavy metals such as arsenic have been shown to drive co-selection of antibiotic resistance, suggesting arsenic-contaminated drinking water is a risk factor for antibiotic resistance carriage. This study aimed to determine the prevalence and abundance of antibiotic-resistant Escherichia coli (AR-Ec) among people and drinking water in high (Hajiganj, >100 μg/L) and low arsenic-contaminated (Matlab, <20 μg/L) areas in Bangladesh. Drinking water and stool from mothers and their children (<1 year) were collected from 50 households per area. AR-Ec was detected via selective culture plating and isolates were tested for antibiotic resistance, arsenic resistance, and diarrheagenic genes by PCR. Whole-genome sequencing (WGS) analysis was done for 30 E. coli isolates from 10 households. Prevalence of AR-Ec was significantly higher in water in Hajiganj (48%) compared to water in Matlab (22%, p <0.05) and among children in Hajiganj (94%) compared to children in Matlab (76%, p <0.05), but not among mothers. A significantly higher proportion of E. coli isolates from Hajiganj were multidrug-resistant (83%) compared to isolates from Matlab (71%, p <0.05). Co-resistance to arsenic and multiple antibiotics (MAR index >0.2) was observed in a higher proportion of water (78%) and child stool (100%) isolates in Hajiganj than in water (57%) and children (89%) in Matlab (p <0.05). The odds of arsenic-resistant bacteria being resistant to third-generation cephalosporin antibiotics were higher compared to arsenic-sensitive bacteria (odds ratios, OR 1.2–7.0, p <0.01). WGS-based phylogenetic analysis of E. coli isolates did not reveal any clustering based on arsenic exposure and no significant difference in resistome was found among the isolates between the two areas. The positive association detected between arsenic exposure and antibiotic resistance carriage among children in arsenic-affected areas in Bangladesh is an important public health concern that warrants redoubling efforts to reduce arsenic exposure.

Introduction Antimicrobial resistance (AMR) is one of the leading threats to public health in the 21st century. The number of hospitalizations and deaths due to AMR infections has been increasing in recent years. In 2019, globally there were an estimated 4.95 million deaths associated with bacterial AMR, including 1.27 million deaths attributable to bacterial AMR [1]. The emergence and spread of AMR infections are more common in resource-poor countries due to the inappropriate use of antibiotics in healthcare, agriculture, and community settings. Bangladesh is known as a hotspot of AMR, with numerous reports showing the widespread prevalence of antibioticresistant organisms (ARO) in humans, food-producing animals, wild animals, and the environment including drinking water, wastewater, surface water, and courtyard soil [2][3][4][5][6][7][8][9][10].
AMR develops when microorganisms are exposed to sub-lethal concentrations of antibiotics and/or antimicrobial agents [11]. In addition to anthropogenic drivers, there are lesserknown natural elements that contribute to the emergence and dissemination of AMR in certain geographic locations. Heavy metals such as arsenic, cadmium, lead, and chromium can function as antimicrobials, and exposure to these heavy metals can co-select resistance to antibiotics [12][13][14][15][16]. Heavy metals are also more stable and can continue to exert selective pressure on bacteria over a longer period than antibiotic residues [17,18], which decrease in the environment due to degradation, absorption, and sequestration over time [19].
Arsenic is a toxic metalloid ubiquitously present either in inorganic or organic forms in the environment (soil, air, and water). The toxicological effects of arsenic depend on its chemical form and oxidation state. Arsenite is the most toxic form and is commonly detected in water, soil, and food [20]. High concentrations of arsenic in groundwater and soil have been an issue that impacts human health in many parts of the world, particularly in the South Asia [21]. Geological, anthropological, and industrial activities are the major sources of arsenic pollution in these areas [22]. In Bangladesh, arsenic contamination in the groundwater (e.g., tube well water) was first detected in 1993 and is now widespread in 50 out of 64 districts of the country [23]. Currently, arsenic is present at a very high level (>50 μg/L) in at least 30% of the tube wells used to provide drinking water and approximately 50 million people use arsenic-contaminated water for their daily activities [23,24]. One of the major contributors to arsenic mobilization in groundwater is phosphate (PO 4 ) rich fertilizer leaching [25]. Therefore, in addition to contaminated drinking water, Bangladeshis also face the risk of arsenic exposure through the consumption of fish raised in aquaculture ponds and rice irrigated by arsenic-polluted water [26,27]. Chronic exposure to arsenic causes serious health issues such as arsenicosis, type 2 diabetes, adverse pregnancy outcomes, and cardiovascular diseases [28][29][30][31][32]. In children, arsenic may cause lower respiratory diseases and diarrhea [33]. In a previous study in Bangladesh, it has been shown that the composition of gut microbiota was altered in children exposed to drinking water containing a high level of arsenic (>50 μg/L) compared to those exposed to drinking water containing low levels of arsenic [34]. The study also reported a higher abundance of antibiotic resistance and virulence genes in the gut microbiome of children that were exposed to higher arsenic levels. Based on this work, arsenic-exposed populations may face further risks of increased gut colonization with ARO due to arsenic-antibiotic co-resistance in enteric bacteria, but empirical evidence is lacking.
Resistance to arsenic might be ubiquitous among various microbial species. Arsenic-resistant bacteria have been isolated from various sources, including industrial wastewater [35], aquatic environments [36][37][38], and soil [39]. Arsenic resistance has also been reported among the members of the Enterobacteriaceae family [40,41]. Exposure to arsenic might play a role in the emergence and spread of antibiotic resistance among organisms in metal-contaminated environments [42]. One study has shown arsenic-induced antibiotic resistance in bacteria in laboratory conditions [15]. Bacteria may use similar mechanisms to become resistant to arsenic and antibiotics [43,44]. Therefore, it is likely that humans and animals that are exposed to arsenic may be colonized with ARO even without being exposed to antibiotics, but there is a lack of evidence to support this claim.
Given the public health significance of AMR, it is important to understand the drivers of AMR carriage, including the relationship between exposure to arsenic through drinking water and the fecal carriage of ARO. In this study, we apply a cross-sectional natural experiment type design to assess how arsenic exposure in drinking water correlates to antibiotic resistance in E. coli in drinking water samples and stool samples collected from households in Bangladesh. Specifically, we assess AMR in water and the feces of both mothers and children in two rural areas in which there is high (Hajiganj) and low (Matlab) levels of arsenic present in drinking water.

Arsenic concentrations in household water samples
Household drinking water samples collected from the Hajiganj area had arsenic concentrations ranging from 223 μg/L to 729 μg/L with a median value of 481 μg/L (arsenic concentrations of individual households are listed in S1 Table). In contrast, arsenic concentrations in household water samples from the Matlab area ranged from 0 μg/L to 20 μg/L with a median value of 0 μg/L. Therefore, the households enrolled in Hajiganj were exposed to high arsenic concentrations, and households in Matlab were exposed to low arsenic concentrations.

Prevalence and abundance of antibiotic-resistant E. coli in high and low arsenic areas
A total of 84% (251 of 300) samples across both sites, including 60 water samples, 100 stool samples from mothers (MS), and 91 stool samples from children (CS) were found to be positive for E. coli. Third-generation cephalosporin-resistant (3GCr) E. coli was found in a total of 67% (201 of 300) samples (24 water, 95 MS, and 82 CS), while 65% (194 of 300) samples (26 water, 98 MS, and 70 CS) were positive for fluoroquinolone-resistant (FQr) E. coli. A significantly higher proportion of water samples in Hajiganj were positive for 3GCr E. coli, FQr E. coli, or both compared to samples in Matlab (Chi-squared test, p = 0.0001; Table 1). Similarly, a significantly higher proportion of children in Hajiganj (94%) were colonized with 3GCr E. coli, FQr E. coli, or both than children in Matlab (76%) (p = 0.0004). However, no significant difference was found in the number of mothers colonized with 3GCr E. coli between Hajiganj and Matlab (p = 0.5164).
The abundance of total E. coli, 3GCr E. coli, and FQr E. coli in each sample was also determined. The mean number of total E. coli in water samples from Hajiganj (1.48 log 10 CFU/100 ml; SD: 0.75) was significantly higher than that of Matlab (0.76 log 10 CFU/100 ml; SD: 0.81) (Mann-Whitney U test, p = 0.0003; Fig 1A). However, no significant difference was found for the 3GCr and FQr E. coli in water samples between Hajiganj and Matlab. Likewise, no significant difference in the mean count of total E. coli, 3GCr E. coli, and FQr E. coli was found in MS and CS samples between the Hajiganj and Matlab areas (Fig 1B and 1C).

Antibiotic susceptibility of E. coli isolates
A total of 251 E. coli isolates (Hajiganj = 132, Matlab = 119) (one isolate per sample positive for E. coli) were tested for susceptibility against a panel of 16 antibiotics encompassing 11 antibiotic classes (S1 Fig). A total of 94% (235 of 251) of E. coli isolates across both sites [54% (126 of 235) from Hajiganj and 46% (109 of 235) from Matlab] were resistant to at least one class of antibiotic. Based on their susceptibility to different antibiotics, isolates were classified as multidrug-resistant (MDR), extended-spectrum β-lactamase (ESBL)-producing, 3GCr, and FQr. A significantly higher proportion of E. coli isolates from drinking water in Hajiganj were found to be MDR, ESBL-producing, 3GCr, and FQr compared to the isolates from Matlab (Chisquared test, p <0.05; Table 2). Similarly, a significantly higher proportion of MDR E. coli, Table 1  , suggesting that isolates in Hajiganj are frequently exposed to environments contaminated with antimicrobial agents ( Table 2).   4 ESBL, extended-spectrum β-lactamase-producing. 5 3GCr, third-generation cephalosporin-resistant. 6 FQr, fluoroquinolone-resistant. 7 MAR, multiple antibiotic resistance (MAR) index of 0.2 or higher.

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms

Prevalence of ESBL and diarrheagenic genes in E. coli from high and low arsenic areas
Of the four ESBL genes tested among 251 E. coli isolates from both sites, bla CTX-M-1 was identified as the most prevalent (73%, n = 184) gene followed by bla TEM ( Table 3.

Arsenic resistance among E. coli isolates
About 80% (200 of 251) of all E. coli isolates including 82% (108 of 132) of isolates from Hajiganj and 77% (92 of 119) of isolates from Matlab were identified as arsenic-resistant based on the minimum inhibitory concentration (MIC) of �5 mM of sodium arsenite [As(III)]. A significantly higher proportion of isolates from water samples in Hajiganj (79%, 27 of 34) were found to be resistant to arsenic compared to the isolates from water samples in Matlab (54%, 14 of 26) (Chi-squared test, p = 0.0002; Table 4). However, there was no significant difference in the prevalence of arsenic-resistant E. coli in MS or CS samples between the Hajiganj and Matlab areas ( Table 4). The presence of characteristic arsenic resistance genes including arsA, arsB, arsC, arsR, and plasmid-specific arsR and arsD was tested by PCR assays. None of the E. coli isolates tested positive for the arsA gene. The arsB gene was present in 52% (131 of 251) of E. coli isolates from both sites, including in 67% (89 of 132) of isolates from Hajiganj and in 35% (42 of 119) aatA 2 (5.9) 3 (11.5) 4 (8) 3 (6) 6 (12.5) 9 (20.9) 12 (9.1) 15 (12.6) bfpA of isolates from Matlab. A significantly higher proportion of E. coli isolates from water, MS, and CS samples from Hajiganj were positive for arsB compared to the corresponding group of isolates from Matlab (Chi-squared test, p <0.01; Fig 2A). In the case of FQr and 3GCr E. coli, water, MS, and CS isolates from Hajiganj were predominantly positive for arsB compared to the isolates from Matlab (p <0.01; Fig 2B and

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms

Correlation between arsenic resistance and antibiotic resistance among 3GCr E. coli isolates
Co-resistance to arsenic and antibiotic classes including penicillin, cephalosporin, tetracycline, fluoroquinolone, macrolide, and trimethoprim-sulfamethoxazole was found in a significantly higher proportion of E. coli isolates from water samples in Hajiganj compared to Matlab (Chisquared test, p <0.05; Fig 3A). Similarly, a significantly higher proportion of E. coli isolates from MS samples in Hajiganj had co-resistance to arsenic and antibiotic classes such as penicillin, cephalosporin, tetracycline, and macrolide than E. coli isolates from MS in Matlab (p <0.05; Fig 3B). In the case of CS samples, co-resistance to arsenic and antibiotic classes (penicillin, cephalosporin, fluoroquinolone, and macrolide) was found in a significantly higher proportion of E. coli isolates from Hajiganj than Matlab (p <0.05; Fig 3C). Next, the association between arsenic and antibiotic resistance was observed among all E. coli isolates from both areas.
Analysis of the arsenic resistance gene operon among isolates revealed that complete arsRBC operon was present in 80% (24 of 30) of the isolates. Six isolates had incomplete arsRBC operon; among these, three isolates (one isolate from Hajiganj and two from Matlab) did not have the arsR and arsB genes and three other isolates from Matlab had a truncated arsR gene (Fig 5). Three isolates that were negative for arsR and arsB genes also had the lowest MIC against As(III). All three isolates belonged to the phylogroup B2 and ST131.
The resistome analysis showed that both Hajiganj and Matlab isolates shared antibiotic resistance genes (Fig 6A). However, isolates from Hajiganj were distinctively positive for a few antibiotic resistance genes that were not present in isolates from Matlab areas. These include bla TEM-128 and bla TEM-135 for β-lactam, qnrS4 for fluoroquinolone, sul3 for sulfonamide, aph (3')-Ia for aminoglycoside, cmlA1 for phenicol, and tet(b) and tet(D) for tetracycline. The ESBL E. coli isolates from both Hajiganj and Matlab were positive for a catalog of virulence genes that are broadly categorized as adhesins, toxins, iron acquisition, and regulators. Most of

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms the virulence genes were commonly distributed between isolates from low and high arsenic areas resulting in no clustering of isolates ( Fig 6B). However, several genes including papA, agg3D, agg3A, agg3C, agg3B, and agg5A were present in the Matlab isolates, while absent in Hajiganj isolates. Overall, isolates from the Matlab areas were positive for a higher number of virulence genes compared to the isolates from the Hajiganj areas, although the difference was not statistically significant.

Factors associated with co-resistance to arsenic and antibiotics among E. coli isolates
Fecal carriage of 3GCr and FQr E. coli among children in the Hajiganj and Matlab areas was not associated with the children's demographic characteristics including sex, maternal use of antibiotics, and mode of delivery (Table 5). Similarly, no significant association was observed in the fecal carriage rate of 3GCr and FQr E. coli among the mothers with respect to their age, previous antibiotic exposure, and diarrheal incidence.

Discussion
In this observational study, we found that fecal carriage of antibiotic-resistant E. coli amongst children, but not amongst mothers, was associated with exposure to arsenic via contaminated drinking water. Third-generation cephalosporin (3GC) and fluoroquinolones (FQs) are the two most commonly prescribed/used antibiotic classes for the treatment of bacterial infections in the hospital and community settings in Bangladesh [45,46]. The high rate (94%) of colonization of children with ARO (3GCr/FQr/both) in the arsenic-contaminated area is alarming given the increased risks of adverse health outcomes from ARO carriage in children. Specifically, the gut microbiome during childhood has a long-lasting effect on overall health in

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms adulthood. Moreover, gut colonization with ARO increases the risk of drug-resistant infections, including urinary tract and bloodstream infections and neonatal meningitis [47][48][49]. Additionally, gut colonization with ARO represents a gene pool for antimicrobial resistance genes that can be exchanged with other human pathogens in the gut.

Fig 6. Heatmaps of resistome and virulence genes of E. coli isolates.
The distribution of A) antibiotic resistance genes, and B) virulence genes retrieved from whole genome sequencing data of 30 E. coli isolated from water, mother stool, and child stool samples from the Hajiganj (high arsenic exposure) and Matlab (low arsenic exposure) areas. The heatmap shows genes' presence (black box) or absence (white box). Gene names are listed on the right. E. coli isolate IDs are listed at the bottom. RH and RM in isolate ID indicate if the isolate was obtained from the Hajiganj and Matlab areas, respectively. WU, MS, and CS in isolate ID indicate isolates from drinking water, mother stool, and child stool, respectively.

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms The mothers in both regions (Hajiganj and Matlab) had high rates of fecal ARO carriage. Households in Matlab are exposed to a low level of arsenic through drinking water because of the installation of deep tube wells, which were installed in the area during 2001-2006 [50]. Before 2001, arsenic contamination in wells in Matlab was likely similar to those observed now in Hajiganj. The average age of mothers enrolled in this study was 25 years, so they were likely exposed to high arsenic levels during their childhood, which might have a role in shaping their gut microbiomes including a higher carriage of ARO. Besides, diet could be an important driver for chronic exposure to arsenic among the adult population in the area. Although mothers from Matlab households do not consume arsenic-contaminated water, they might be exposed to arsenic via the food chain, especially through the staple food rice and leafy vegetables which are irrigated with arsenic-contaminated water generally extracted from shallow groundwater aquifers. It has been reported that the relative contribution of dietary arsenic sources becomes more important when there is a lower risk of arsenic exposure via drinking water [51]. A previous study in Bangladesh reported that rice and leafy vegetables can absorb the highest amounts of arsenic depending on the cooking practices [52]. In our study, we did not focus on the dietary habits or food preparation practices among the households, and we also did not analyze the food samples for arsenic. Therefore, further studies are required to identify the other sources of arsenic exposure in areas where arsenic-free drinking water is accessible to the community.
A higher proportion of E. coli isolates from drinking water samples in Hajiganj were resistant to different classes of antibiotics than isolates from water samples in Matlab. Bacteria exposed to arsenic-contaminated groundwater may induce or select bacterial adaptations to one or more antibiotics through co-selection of resistance. This is in agreement with previous studies that organisms isolated from extreme environmental conditions such as industrial or agricultural wastewater have shown arsenic-mediated antibiotic resistance in vitro [15,42,53]. In a related study, 6 h arsenic exposure [As(III) of 0.2-1 mg/L] in drinking water increased bacterial resistance to cephalosporin, tetracycline, and erythromycin as well as increased the relative abundance of antibiotic resistance genes in a bacterial community [42]. In this study, we found that the co-resistance to arsenic and antibiotics was more prevalent among 3GCr E. coli in high arsenic areas compared to low arsenic areas. In particular, arsenic-resistant isolates were more likely to be resistant to β-lactam antibiotics including 3GC suggesting that the use of these antibiotics needs to be evaluated carefully in areas with high arsenic contamination. We also found that arsenic-resistant bacteria were less likely to be resistant to fluoroquinolone, chloramphenicol, and tetracycline antibiotics. Notably, the finding about tetracycline stands in contrast to previous in vitro studies showing bacterial exposure to arsenic promotes bacterial resistance to the tetracycline [15,42]. The difference in our study may be that tetracycline resistance prevalence is very high in both study areas, resulting in insufficient variability between sites to observe a significant difference.
This study finds that antibiotic-resistant E. coli are also resistant to arsenic and carry arsenic resistance genes. This imposes a greater public health risk because these bacteria can survive in water with high arsenic concentrations, which increases antibiotic resistance through selection for co-resistance and promotes antibiotic resistance gene transmission to other organisms via horizontal gene transmission. Notably, none of the isolates in our study carry arsA, a required gene when resistance to arsenic is due to arsRDABC operon. This indicates the presence of a simpler arsRBC operon in these isolates, which is also supported by a previous study [34]. Isolates that possess arsRDABC operon are more resistant to arsenic than those expressing arsRBC operon [54]. The absence of arsB in some isolates corresponds to the loss of arsenic resistance. From the WGS analysis, three isolates (one from Hajiganj and two from Matlab) had incomplete ars operon (missing arsR and arsB) and were sensitive to arsenic. Interestingly, all three arsenic-sensitive isolates belong to ST131 and phylogroup B2. This finding is similar to a previous report by Sutterlin et al where they found that the majority of arsenic-sensitive E. coli isolates belonged to phylogroup B2 [40]. Plasmid-specific arsR and arsD genes were present in some isolates, indicating that some of the isolates carry these genes in the plasmids, as has been previously shown [55,56]. Colocation of arsenic and antibiotic resistance on plasmids among isolates in this study may explain the presence of dual resistance and would provide a mechanism for horizontal gene transfer of multiple resistances to other bacteria.
Analysis of isolates for E. coli diarrheagenic genes revealed that only a few isolates possessed genes related to EPEC and EAEC. Further, these organisms were not phylogenetically clustered based on location (high versus low arsenic exposure). However, from the WGS analysis of 30 E. coli isolates, we observed a higher number of virulence genes in isolates from the low arsenic area than in the high arsenic area. While the virulence mechanism is different from the resistance mechanism for both antibiotics and heavy metals, this difference in diversity may be due to the fitness cost of the isolates in high arsenic areas. Further studies are needed to understand the impact of heavy metal exposure on bacterial virulence properties. Several studies have reported a higher incidence of diarrheal diseases among children in arsenic-endemic areas [33,[57][58][59]. However, whether it is due to the higher prevalence of diarrheal pathogens in the area or due to the vulnerability of the population to diarrhea with respect to their arsenic exposures, hygiene, sanitation, or access to clean drinking water supply is not well explained.
A primary limitation of this study is that the presence of arsenic in groundwater is spatially correlated, so the study participants with and without arsenic exposure could not be randomly chosen. The selection of households focused on two neighboring upazillas, which despite similar lifestyles and food habits, are still distinct neighborhoods, and as such there may be neighborhood-level differences beyond arsenic in water that contribute to the observed differences. Additionally, the fecal carriage of AMR may be driven by other sources of arsenic, including the food chain. Future studies may consider measuring the concentration of arsenic in urine samples of mothers and children from selected households before sampling to determine if arsenic concentrations in water are related to differences in arsenic burden. In addition, analyses were based on one isolate from each sample to compare the characteristics of organisms between the two communities, which does not necessarily fully represent the bacterial community diversity in samples. However, given the findings from this observational study, further studies can be pursued on the impact of arsenic in drinking water on antibiotic resistance carriage, such as randomized controlled trials potentially involving a stepped wedge intervention design focused on reducing arsenic in drinking water.
Our study suggests a plausible link between arsenic exposure and carriage of ARO among children in the study communities, which is an emerging dimension of the arsenic problem. This issue further complicates the existing risks to public health due to both arsenic exposure and AMR infections. Developing a sustainable solution to reduce arsenic exposures in Bangladesh would not only counter the associated health risks due to arsenic toxicity but may also reduce the burden of AMR in the community. Overall, this study highlights the urgent need to reduce arsenic exposures in these communities.

Ethics statement
The study (PR-16086) involving human participants was reviewed and approved by the research and the ethical review committees of the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b). The review committees also monitored the progress of this

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms study. Written informed consent to participate in this study was provided by the mothers themselves and for their children by the participants' legal guardians.

Study design, site selection, and sample collection
Two rural upazillas (sub-districts)-(Hajiganj and Matlab)-in the Chandpur district of Bangladesh were selected for this study. Hajiganj has a history of high arsenic concentrations in drinking water, while the arsenic concentrations in drinking water in Matlab are generally low. These two populations presented the opportunity to study the effect of arsenic exposure on gut colonization with AMR bacteria. A total of 100 households that included 50 households in each upazilla were conveniently selected based on the following criteria, 1) having children of <1 year of age, and 2) members are exposed to high (>100 μg/L) versus low (<20 μg/L) arsenic concentration. A preliminary arsenic concentration in the tube well water of each household was tested with a rapid arsenic detection screening kit (Arsenic Low Range Test Kit, Hach, Loveland, CO, United States). A survey was completed by the mothers in each household. Stool samples from mothers and children and in-house drinking water samples were collected for microbiological analysis. The details of participants' enrollment and sample collection were described previously [60].

Determination of arsenic concentrations in water samples
Water samples (100 ml) from each household were sent to Eawag laboratory, Switzerland in sterile plastic bottles. Samples were analyzed with ICP-MS (Agilent-7500cx) for the quantification of arsenic compounds as described previously [61].

Enumeration and isolation of E. coli
Enumeration and isolation of total E. coli, third-generation cephalosporin-resistant (3GCr) E. coli, and fluoroquinolone-resistant (FQr) E. coli in each water and stool samples were done using the membrane filtration and drop plate methods, respectively, according to the procedures described previously [60]. For both water and stool samples, at least one isolated colony from each sample was confirmed as E. coli by API-20E (bioMérieux, France) and stored in glycerol broth at -80˚C for further characterization.

Antimicrobial susceptibility test
Antimicrobial susceptibility of E. coli (one isolate per sample) against 16 commercially available antibiotics (Oxoid Ltd., United Kingdom) was determined by standard disc diffusion technique following the Clinical and Laboratory Standards Institute (CLSI) guidelines and the procedure described previously [7]. E. coli ATCC 25922 was used as a control strain. An isolate resistant to at least one agent of three or more classes of antibiotics was considered multidrugresistant (MDR) [62]. The presence of extended-spectrum β-lactamase (ESBL) in 3GCr E. coli was determined by the combination-disc diffusion (CDD) test following the CLSI guidelines [63]. A multiple antibiotic resistance (MAR) index was calculated based on the number of antibiotics one isolate was resistant to divided by the total number of antibiotics tested as described previously [64].

Test for susceptibility to arsenic
The minimum inhibitory concentrations (MICs) to sodium arsenite [As(III), NaAsO 2 , Sigma] were determined by agar dilution assay performed according to the methods described previously [65]. Briefly, 10 μl of bacterial suspension (10 8 CFU/ml) from the overnight bacterial

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms culture was applied as a spot, in triplicate, on trypticase soy agar (TSA) plates supplemented with different concentrations of As(III) ranging from 0.5-11.0 mM. The plates were observed for MIC of arsenite salt after overnight incubation at 37˚C. The MIC was read as the lowest concentration with no visible growth. E. coli ATCC 25922 was used as the control strain. The isolates were considered resistant to arsenic if the As(III) MIC value was �5 mM (~1024 mg/L) [66].

PCR for ESBL and diarrheagenic genes among E. coli isolates
Boiled DNA extracted from E. coli isolates was used for the PCR assays. Primers specific for four major ESBL genes, bla CTX-M-1 , bla SHV , bla TEM, and bla OXA-1 were tested by PCR using primer sequences and PCR conditions as described previously [67]. Multiplex PCR was used for the detection of enteropathogenic E. coli (EPEC), enterotoxigenic E. coli (ETEC), enteroaggregative E. coli (EAEC), enteroinvasive E. coli (EIEC), and Shiga toxin-producing E. coli (STEC) according to the procedure described previously [9].

PCR for arsenic resistance genes
All E. coli isolates were tested for the presence of five major arsenic resistance genes, arsA, arsB, arsC, arsD and arsR. The primer sets and nucleotide sequences are presented in S2 Table. PCR reaction for arsA, arsB, and arsC was done according to the methods described previously [68]. Two different sets of primers were used to detect arsR; the K12_arsR primer set was used to detect chromosomally encoded arsR and the R46_arsR primer set was used to detect plasmidencoded arsR. In addition, plasmid-encoded arsD was detected with another set of primers.

Statistical analyses
Data were entered in SPSS 20.0 (IBM Inc., Chicago, IL, United States). Data cleaning, statistical analyses, and graphical presentations were done in Stata 13.0 (College Station, TX, United

PLOS PATHOGENS
Effects of arsenic on the fecal carriage of antibiotic-resistant organisms States), GraphPad Prism 9 (La Jolla, CA, United States), and R-3.4.2 (R Core Team, 2014). Chi-squared tests were used for the contingency tables where applicable (e.g., comparing the prevalence of E. coli, 3GCr E. coli, FQr E. coli, or arsenic resistance genes in each type of sample between two areas). The abundance of E. coli, 3GCr E. coli, and FQr E. coli in water and stool samples was calculated by log 10 transformed colony counts (CFU/100 ml water or CFU/g stool) and the comparison between the two areas was done using the non-parametric Mann-Whitney test. Due to the high number of negative samples, only the positive samples were used in the analysis. The association between demographic characteristics of children and mothers in low versus high arsenic areas with a prevalence of fecal carriage of 3GCr E. coli and FQr E. coli was calculated using Chi-squared tests (Fisher's exact test was used if the expected frequency was <5). Familywise error rates accounted for multiple comparisons were adjusted using the Holm-Bonferroni test [78]. The odds ratios and their 95% confidence intervals (CIs) were estimated using the logistic regression with antibiotic resistance as a binary outcome and arsenic resistance as a binary exposure variable. A forest plot was used to present and compare the odds ratios among different antibiotics. Data were analyzed by R-package glm v4.2.1. Statistical significance was determined using p <0.05 for all tests.