Potential pathogens and antimicrobial resistance genes in household environments: a study of soil floors and cow dung in rural Bangladesh

ABSTRACT

In low- and middle-income countries, living in homes with soil floors and animal cohabitation may expose children to fecal organisms, increasing the risk of enteric and antimicrobial-resistant infections. Our objective was to understand whether cow cohabitation in homes with soil floors in rural Bangladesh contributed to the presence and diversity of potential pathogens and antimicrobial resistance genes (ARGs) in the home. In 10 randomly selected households in rural Sirajganj District, we sampled floor soil and cow dung, which is commonly used as sealant in soil floors. We extracted DNA and performed shotgun metagenomic sequencing to explore potential pathogens and ARGs in each sample type. We detected 7 potential pathogens in soil only, 38 pathogens in cow dung only, and 182 pathogens in both soil and cow dung. Cow dung exhibited modestly higher potential pathogen genus richness compared to soil floors (Wilcoxon signed-rank test, P = 0.002). Using Bray-Curtis dissimilarity, potential pathogen species community composition differed between floors and cow dung (permutational multivariate analysis of variance, P < 0.001). All soil floor and cow dung samples contained ARGs; detected ARGs confer resistance to antibiotic classes including sulfonamides, rifamycin, aminoglycosides, lincosamides, and tetracycline. Paired floor and cow dung samples shared ARGs against rifamycin and glycopeptides, but otherwise, there was little overlap in resistomes between sample types. Our findings contribute to the growing literature on household soil and domestic animals as potentially important contributors to disease transmission and as reservoirs of antimicrobial resistance in low-income country settings.

IMPORTANCE

In low-income countries, inadequate housing materials and animal cohabitation can lead to fecal contamination of rural homes. Contaminated soil floors are difficult to clean and may harbor organisms causing illness and antibiotic resistance, especially in young children, who frequently ingest soil. We sequenced soil floor and cow dung samples from households in Sirajganj district, Bangladesh, and identified potential pathogens and antibiotic resistance genes. We detected 182 potential pathogens in both soil and cow dung; organisms present in both sample types at the highest relative abundances were Escherichia coli, Klebsiella pneumoniae, Salmonella enterica, and Pseudomonas aeruginosa. Antibiotic resistance genes were found in all samples. In cow dung, the most common genes conferred resistance to the antibiotics lincosamide, rifamycin, cephamycin, tetracycline, and multiple antibiotics. In soil floors, the most common genes conferred resistance to rifamycin, sulfonamides, and multiple antibiotics. Household soil and cow dung may be important reservoirs of pathogens and antimicrobial resistance in low-income country settings with high levels of animal cohabitation compared to settings with finished household floors and minimal animal cohabitation.

INTRODUCTION

In low- and middle-income countries (LMICs), inadequate housing remains common (1, 2) and is associated with disease and mortality (3). Rural households frequently cohabitate with domestic animals in LMICs, and animal husbandry is a critical source of income and nutrition (4). Yet, animals also contribute to fecal contamination of rural households (5, 6). Studies using avian and ruminant microbial source tracking markers found that animals contribute to Escherichia coli—a widely used environmental indicator of human fecal contamination—in soil in households in Bangladesh (7). Young children frequently touch and ingest soil, which may be contaminated with human or animal feces, within household premises (8–10). A prior study in Bangladesh estimated that mouthing of child hands, direct soil ingestion, and direct feces ingestion were leading contributors to child E. coli ingestion among children under 1 year in household settings (9).

Young children’s exposure to soil and animal feces can facilitate disease transmission in the home. Household soil floors are a largely overlooked reservoir for soil-transmitted helminths (STH), Shigella, pathogenic E. coli (11–13), and possibly other pathogens. Some studies have detected levels of E. coli and STH in household soil floors that exceed those in samples taken from latrine floors (12, 14, 15). Contact between young children and domestic poultry and livestock is associated with an increased risk of diarrhea (16). Exposure to household fecal contamination via household soil, stored drinking water, child hands, and fomites is associated with increased risk of diarrhea (17), enteric pathogen infections (18), child growth faltering (17, 18), and antimicrobial resistance (19).

Soil harbors diverse bacteria with naturally occurring antibiotic resistance that can exchange genes or plasmids with human and animal pathogens in fecally contaminated soils, making such soils a potential reservoir of emerging antimicrobial resistance (20–22). While it has been suggested that a fraction of transmission events between soil and human pathogens are antimicrobial resistance genes (ARGs) (20), drivers of this transfer require a deeper understanding of selective pressures (23, 24). Prior studies have detected extended-spectrum beta-lactamase (ESBL)-producing E. coli in household soil in rural Bangladesh (25) and multidrug-resistant E. coli in household yard soil in Tanzania (26). A genomic study of E. coli in rural Bangladesh found multiple virulence and antibiotic resistance genes in household soil and yard soil, and phylogenetic analyses suggested that E. coli in soil was likely from diverse human and animal sources (19).

Studies have also found evidence of horizontal gene transfer of antimicrobial resistance genes between bacteria in animals and humans within household settings in LMICs, but it is unknown how frequently this occurs (27). Unhygienic animal husbandry practices (e.g., handling cow dung with bare hands), sharing of household spaces with domestic animals, poor management of animal waste and carcasses, and inadequate hygiene while caring for domestic animals contribute to zoonotic and antimicrobial-resistant (AMR) pathogen transmission in LMICs (28, 29). Moreover, inadequate access to veterinary resources contributes to the misuse of antimicrobials for prophylaxis or as feed additives in low-income settings (30). Domestic animals may also become infected with or colonized by human pathogens containing genes that confer AMR through contact with human feces in homes with inadequate sanitation (31). A recent review of studies investigating the contribution of animals to AMR in humans reported mixed results (32); they concluded that this question remains poorly understood in LMICs because many prior studies were either conducted in high-income settings with low levels of human-animal contact or did not use appropriate methods to capture transmission dynamics.

In rural, low-income communities where cattle rearing is common, cows often cohabitate closely with humans, and cow dung may contribute to household pathogen contamination. Cow dung is commonly used as fertilizer, cooking fuel, and as a coating for floors and household walls in household settings in LMICs and in Bangladesh (28, 33–35), yet the extent to which household contamination with cow dung contributes to zoonotic or AMR pathogen transmission is unknown. Cow dung commonly contains human pathogens such as Salmonella spp., Campylobacter spp., Listeria monocytogenes, Yersinia enterocolitica, E. coli, Cryptosporidium parvum, and Giardia lamblia (36). A prior study in Bangladesh found that the presence of cow dung in household courtyards and detection of a molecular marker of cow feces on mothers’ hands were associated with the presence of pathogenic E. coli and Giardia on mothers’ hands (6). Carbapenem resistance genes (37) and AMR E. coli and Salmonella spp.—WHO high-priority pathogens for development of AMR (38)—have been detected in cow dung samples (39). Studies in Bangladesh have also found high levels of carbapenem resistance genes in household cattle dung samples (37).

Our objective was to understand whether cow cohabitation in homes with soil floors in rural Bangladesh contributed to the presence of potential pathogens and ARGs in the household setting. We hypothesized that (i) cow dung and soil floors would contain potential human pathogens and ARGs and (ii) cow dung and soil floors from the same households would have overlapping microbiomes and resistomes. We conducted this exploratory pilot study using shotgun metagenomic sequencing of paired soil floor and cow dung samples from households in rural Sirajganj district, Bangladesh, where cattle rearing is common, and cows frequently cohabitate inside homes with soil floors. Prior studies have mostly investigated microbiomes or resistomes of outdoor soil and animal feces in LMICs (40–43). Studies of indoor household soil have focused on single bacteria (e.g., E. coli, Campylobacter jejuni) in soil, human, and animal feces, with some employing whole genome sequencing to assess antibiotic resistance genes in E. coli (19, 25, 44–46). One study in Bangladesh examined microbial communities in human and animal stool samples across urban and rural settings using 16S rRNA and long-read sequencing (47). Here, our use of metagenomic analysis of paired soil floor and cow dung samples in rural Bangladeshi households allowed us to capture a wide range of microbes and antimicrobial resistance genes beyond those associated with E. coli.

RESULTS

We enrolled 10 households from Sthal union, Chauhali upazila, Sirajganj Bangladesh. Households were eligible for enrollment if they had a soil floor, a child under the age of 2 years, available cow dung for sampling, and no self-reported cases of anthrax among their domestic animals or household members. The mean number of household members was 6 (range 4–8), and most homes were approximately 300 square feet (Table 1). Households typically had access to a pit latrine and tubewell within their compound. In addition to keeping cows, all but one household owned sheep or goats, and all but one owned chickens, ducks, or pigeons. The mean number of animals owned by each household was 4 cattle, 4 sheep or goats, and 17 chickens, ducks, or pigeons. Seven households kept their cows tied up outside during the day, while the rest kept them in a different structure in the compound (e.g., a cattle shed). At night, seven households kept their cows inside a household while the rest kept them in a different structure in the compound or tied up outside (Table S1). Four of 10 households use cow dung for household purposes. At the time of the survey, wet, dry, or processed cow dung was visible in the courtyard of eight households and on the main area of indoor household floors in two households.

TABLE 1.

Characteristics of study households

Characteristic	Mean (range) or N (%)
Number of household members	6.3 (4, 8)
Number of children under 5 in the household	1.3 (1, 3)
Number of animals owned
Cattle	3.6 (1, 8)
Goat or sheep	3.5 (0, 7)
Chickens, ducks, or pigeons	16.8 (0, 30)
Location of cattle during the daytime
In a different house in the compound	3 (30%)
Tied up outside	7 (70%)
Location of cattle at nighttime
Free inside the home with no barrier	2 (20%)
Tied up inside the home with barrier	5 (50%)
In a different house in the compound	1 (10%)
Tied up outside	2 (20%)
What is done after cow defecates
Remove it from the house	1 (10%)
Remove it from the house and then clean the floor	7 (70%)
Pile it inside the house	2 (20%)
Household uses cow dung for cooking, fertilizer, etc.	4 (40%)
Cow dung visible in the courtyard	8 (80%)
Cow dung visible on the main area of the household floor	2 (20%)

Microbial composition

We performed shotgun metagenomic sequencing and used Kraken/Bracken for taxonomical analyses on paired floor soil and cow dung samples from the same households. Sequencing yielded a total of 152.94 million paired reads of DNA from the 10 cow dung samples (7.18–28.4 million reads per sample) and 33.28 million reads from the 10 soil samples (0.58–11.01 million reads per sample) (Table S2). After quality filtering for all samples and host filtering for cow dung samples, the average reads remaining per sample were 96.3 million in cow dung (4.9–13 million reads per sample; 63% retained on average) and 17.7 million in soil (0.3–6 million reads per sample; 53% retained on average) that were used as the input for contig assembly and metagenomic analysis.

Soil and cow dung samples exhibited a large number of microbes, with an average of 1,935 genera and 6,799 species detected per sample in cow dung and 1,057 genera and 3,837 species detected per sample in soil floors. The genera with the highest relative abundance were Bacteroides (mean relative abundance = 6.5%), Clostridium (2.8%), Prevotella (2.5%), Faecalibacterium (2.4%), and Blautia (2.2%) in cow dung and Janibacter (8.2%), Nocardioides (5.5%), Brachybacterium (4.7%), Streptomyces (4.0%), and Serinicoccus (4.0%) in soil.

In order to focus on microbes with pathogenic potential, we restricted microbial composition analyses to microbes with known pathogenicity in humans and re-estimated relative abundance among only potential pathogens (48). Across these potential pathogens, the most abundant genera across samples were Clostridium (mean relative abundance = 13.01%), Prevotella (11.92%), Bacillus (9.31%), Streptococcus (6.7%), and Pseudomonas (5.56%) in cow dung samples and Pseudomonas (14.58%), Corynebacterium (8.26%), Mycolicibacterium (7.22%), Acinetobacter (6.96%), and Escherichia (6.92%) in soil floors. The most common potential pathogen species in cow dung were Escherichia coli (mean relative abundance = 13.99%, including pathogenic and/or non-pathogenic strains), Staphylococcus aureus (7.7%), Prevotella melaninogenica (5.57%), Salmonella enterica (3.79%), and Clostridium perfringens (3.75%); in soil, the most common potential pathogen species were E. coli (20.76%), Pseudomonas aeruginosa (5.96%), Klebsiella pneumoniae (5.91%), Stenotrophomonas maltophilia (5.8%), and Pseudomonas putida (5.31%) (Fig. 1). We detected seven potential pathogens in soil only, 38 pathogens in cow dung only, and 182 pathogens in both soil and cow dung. The only potential pathogen with a relative abundance >5% in both soil floors and cow dung from the same households was E. coli (N = 7 households) (Fig. 2; Fig. S1). Table S3 lists all potential pathogen species, and the full list of taxa detected by sample type is included in the Supplement.

Fig 1.

Sample-level relative abundance of non-host reads for potential pathogens by metagenomic next-generation sequencing analysis at the (a) genus level and (b) species level. Includes the top 30 species by average relative abundance across all samples, with all other genera or species labeled as “Other.” C1–C10 refer to cow dung samples, and S1–S10 refer to floor soil samples; each number corresponds to a different household (e.g., C1 and S1 are from household 1).

Fig 2.

Heatmap of filtered, non-host reads for potential pathogens detected in samples of cow dung (listed as C1–C10) and floor soil (listed as S1–S10). Tile colors indicate the relative abundance of each species within each sample. Gray tiles indicate that a species was not detected. Includes taxa with an average relative abundance across all samples of at least 0.5%. The heatmap displays hierarchical clustering of rows using Euclidean distance and Ward’s minimum variance method.

Diversity of potential pathogens within samples

We determined the alpha-diversity of potential pathogens using measures of microbial genus richness and evenness within each sample type using the richness attribute, Pielou’s evenness index (49), Shannon index (50), and Simpson index (51). Cow dung samples exhibited modestly higher levels of potential pathogen species and genus richness (Wilcoxon signed-rank test, P = 0.002 for species, P = 0.006 for genera) compared to soil samples (Fig. 3; Fig. S1). There was a wider range of within-sample potential pathogen diversity in soil samples compared to cow dung samples by diversity metrics that included evenness.

Fig 3.

Comparisons of (a) species richness, (b) species evenness, (c) Shannon diversity indices, and (d) Simpson diversity indices for potential pathogen species by sample type. Includes 10 household-paired cow dung and soil floor samples. Indices were compared between sample types using the Wilcoxon signed-rank test.

Diversity of potential pathogens between samples

To compare the diversity of potential pathogens between samples, we calculated Bray-Curtis dissimilarity and performed principal coordinates analysis. Potential pathogen species community composition differed between floor and cow dung samples (pairwise permutational multivariate analysis of variance, R² = 0.37, P < 0.001) but not between households (R² = 0.37, P = 0.933) or between households with and without visible animal feces on the floor inside the home (R² = 0.04, P = 0.59) (Fig. S2).

Antimicrobial resistance

Next, we detected ARGs using the Chan Zuckerberg ID (CZID) pipeline. Analyses revealed diverse ARG profiles with substantial variation between households and sample types. The most common ARGs we detected conferred resistance to rifamycin, sulfonamide, aminoglycoside, or multiple drug classes in soil floors and tetracycline, cephamycin, lincosamide, rifamycin, or multiple drug classes in cow dung (Fig. 4 and 5a). Only ARGs that confer resistance to rifamycin, tetracycline, and glycopeptide antibiotics were found in both soil floors and cow dung in multiple households. There was a larger number of distinct ARGs in soil floors than in cow dung samples. Genes associated with resistance to multiple classes of antibiotics were common in both sample types, particularly in soil. ARGs in soil most commonly conferred resistance through antibiotic target alteration or antibiotic efflux; the most common mechanism for ARGs in cow dung was antibiotic target alteration followed by antibiotic target protection (Fig. 5b). Nine of 10 soil floors and all cow dung samples contained at least one ARG in the highest quartile of risk to human health (e.g., sul1, tet(Q), mexF, ermF, cfxA2) (Fig. S3) (52). Both sample types contained high-risk ARGs from multiple drug classes (Fig. S4).

Fig 4.

Heatmap of antibiotic resistance genes detected in cow dung and soil samples. Tile colors indicate the read coverage breadth. Includes genes with read coverage breadth >10% and >5 reads mapped or contig coverage breadth >10%. Right annotation indicates the drug class to which the ARG confers resistance. Colors indicate the highest alignment confidence based on contig match quality (blue) or reads (green). “Perfect” contig matches identically matched reference sequences in the Comprehensive Antibiotic Resistance Database. “Strict” contig matches were those that matched previously unknown variants of known ARGs, including secondary screening for key mutations. “Nudged” contig matches had at least 95% identity to known AMR genes and were matched using a percent identity threshold not taking alignment length into account.

Fig 5.

Relative abundance of antibiotic resistance genes detected in cow dung and soil samples by (a) drug class and (b) antibiotic resistance mechanism. Both panels include genes with read coverage breadth >10% or contig coverage breadth >10% and >5 reads mapped.

DISCUSSION

In this exploratory study, we found that household soil floors and cow dung in rural Bangladesh contained diverse microbial communities, including numerous potential human pathogens. While our analysis does not allow us to infer transmission from cow dung to soil floors, our findings suggest that the presence of cow feces in domestic spaces may contribute to microbial contamination of household surfaces. Antimicrobial resistance genes against multiple drug classes were prevalent in both sample types, and nearly all soil floor and cow dung samples contained ARGs that a prior study showed were associated with increased risk to human health (52). There were few shared ARGs present in paired samples from the same households, suggesting that cows and soil floors had limited overlap in their resistomes. However, our small sample size may have limited our ability to establish a link between these sample types. Given children’s high levels of soil and animal contact in this setting (10), our findings suggest that household soil floors and cow dung may be important reservoirs of diverse pathogens and antimicrobial resistance. It is also possible that cows’ exposure to human fecal contamination contributed to colonization of cattle with ARGs.

E. coli and S. enterica were present in both sample types in higher relative abundances than other identified potential pathogens, suggesting that soil and cow dung may be a household source of enteric infections with these pathogens. However, our results must be interpreted with caution as it remains possible that taxonomical misclassification occurred. Additionally, many E. coli strains are commensal; while our limited sequencing depth did not provide high enough coverage of the E. coli genome to allow for pathogenic E. coli strain identification, a separate analysis of this study population found that 8% of E. coli isolates were pathogenic (53). Other studies in Bangladesh have detected a similar prevalence of pathogenic E. coli in household soil samples (19, 25). While we are not aware of any prior studies in LMICs that have detected Salmonella spp. in household floors, studies in the USA have identified Salmonella spp. in soil near household entrances and in vacuum cleaners in homes with infected infants (54, 55). In cattle, S. enterica is a facultative pathogen that can cause illness (e.g., enteric infection, reproductive loss); additionally, certain serotypes, such as S. enterica serotype Dublin, can result in lifelong asymptomatic carriage with intermittent shedding in cattle stool and severe illness if transmitted to humans (56, 57).

We also detected potential pathogens in soil floors and cow dung that can cause illness in individuals with lower immunity (K. pneumoniae, P. aeruginosa, Stenotrophomonas maltophilia, Clostridium difficile, Staphylococcus aureus). Possible routes of exposure for these pathogens include direct contact with contaminated surfaces (S. aureus, P. aeruginosa, K. pneumoniae), ingestion of contaminated soil or dung particles (C. difficile), and inhalation of aerosolized particles (S. maltophilia, P. aeruginosa) (58–62). The presence of pathogens in household environments has been documented in LMICs and high-income countries, though studies in LMICs are limited. A study in Malawi detected extended-spectrum beta-lactamase K. pneumoniae at low levels on household floors (63), while research in Ghana identified some of the same microbes we detected in household dust, including Acinetobacter baumannii, Bacillus cereus, and Enterobacter cloacae (64). C. difficile was present in all cow dung samples, consistent with other studies which have found that it can colonize healthy cattle and be transmitted zoonotically (65). It was only detected in two soil samples, consistent with a prior study that detected it in approximately one-third of household soil samples in rural Zimbabwe (66). In high-income countries, household surfaces can also be highly contaminated with potential pathogens; E. coli, K. pneumoniae, P. aeruginosa, S. aureus, and S. maltophilia have been found on sink and shower drains, floors, and surfaces (67–70). Additionally, studies in high-income countries have found that household dust contains E. coli, Pseudomonas, Acinetobacter, Enterobacter, Enterococcus, Bacillus, and Staphylococcus (71). The health implications of household contamination with these pathogens remain unclear, and it is not known whether the levels of contamination commonly seen in LMICs or high-income countries contribute to infections (72). Our metagenomic analysis did not allow us to confirm the presence or quantify the abundance of potential pathogens; assessing the risk of infection under typical LMIC household exposures and quantitative microbial risk assessment is an important area of future research.

Many potential pathogens were present in soil floors and cow dung samples from the same households, which may reflect contamination of floors with cow feces or contamination of floors and cow dung by another source, such as human feces or feces of other domestic animals. Additionally, cow dung is frequently used to seal soil floors in this setting; it is not known how long pathogens in cow dung used to seal soil floors survive, but survival is likely to depend on temperature, humidity, rainfall, and soil composition (73, 74). Our cross-sectional sample and use of metagenomic sequencing did not allow us to establish the source of microbes in soil floors. Prior studies in Bangladesh have detected ruminant fecal markers in household soil and hand rinses in households that both owned or did not own ruminants (7, 75). Some of the potential pathogens found in soil floors are common in soil or other environmental niches exposed to human activity, so their presence may not imply fecal contamination by cow dung (e.g., E. coli, K. pneumoniae, B. cereus, P. aeruginosa) (76–80). Other potential pathogens are commonly found in the gastrointestinal tract of warm-blooded animals (74, 81, 82), such as E. coli, K. pneumoniae, and S. enterica, so their presence in soil may suggest that floors were contaminated with feces of humans or other animals, such as chickens or goats. B. cereus and P. aeruginosa also cause mastitis in cattle (83, 84), and it is possible that they are shed in cow dung during infection. Future research using microbial source tracking and phylogenetic analyses with longitudinal samples could elucidate the contribution of cow feces to household soil microbiota.

There was limited overlap in ARGs in paired soil floor and cow dung samples, and ARGs in each sample type provided resistance to different drug classes. These findings may imply that cow fecal contamination did not result in transmission of ARGs to soil, but the methods we used in this study did not allow us to formally assess transmission. Additionally, ARGs in soil may result from natural rather than anthropogenic processes. Some ARGs we detected in household soil have been detected in pristine soils in Tibet and thus may not reflect fecal pollution (e.g., vanRO, rpoB2, rpoB, rphA, muxB, mexF, mexK, mexW) (85). Our finding that the dominant mechanism of resistance in soil ARGs was through efflux pumps is consistent with the prior literature (86). These ARGs often confer resistance to multiple antibiotics; some have been found in pristine soils, and it is believed that they originally did not evolve to confer resistance to human antibiotics (87). Our results differ from a prior study that linked ARGs in E. coli in household soil in Bangladesh to human and animal sources using whole genome sequencing (19). Other studies have identified livestock manure as a primary source of ARGs in agricultural soils (88). Possible explanations for our finding of limited overlap in resistomes between cow dung and soil also include that (i) sequencing depth was limited, so we failed to detect overlapping ARGs, (ii) transferred DNA degraded over time, (iii) ARGs were present below detection limits, (iv) soil bacteria harbor their own distinct resistomes shaped by local selective pressures, (v) soil ARGs may primarily originate from other sources (e.g., human or chicken feces), or (vi) ARGs in cow dung may not reside on mobile genetic elements capable of transfer between bacteria (89).

Our analysis revealed ARGs in environmental samples that align with findings of prior studies on ARGs in environmental samples and local antibiotic usage patterns. The most common ARGs we detected in cow dung confer resistance to tetracyclines, rifamycins, lincosamides, and multiple antibiotics; the most common ARGs we detected in soil floors confer resistance to these drug classes as well as sulfonamides. Evidence was strongest for ARGs against sulfonamides in soil samples and lincosamides and tetracyclines in cow dung, and glycopeptides in both sample types. A systematic review of ARGs in agricultural soils identified ARGs against tetracyclines, sulfonamides, and aminoglycosides (64). ARGs encoding resistance to rifamycin were relatively common in both sample types. Studies in dairy farms have detected Salmonella spp. and Listeria resistant to rifamycins (90). Additionally, many environmental bacteria are naturally resistant to rifamycin, such as Amycolatopsis, which was present in nearly all cow dung and soil samples. However, rifamycins have not been detected in prior studies of ARGs in environmental reservoirs (91). ARGs we detected conferred resistance to antibiotics that are used by farmers in some rural communities in Bangladesh. In a survey in Mymensingh, Bangladesh, farmers commonly treated animals with tetracyclines, sulfonamides, and aminoglycosides (92). Lincosamides and rifamycins are frequently used to treat mastitis in cows (90, 93). Tetracycline is commonly used to treat gram-positive and gram-negative bacteria in animals and humans and to promote growth in livestock (94).

A strength of using metagenomic sequencing in environmental samples is that it can reveal more ARGs than culture-based approaches since many bacteria in environmental reservoirs cannot be cultured (89). However, culture-based approaches are required to assess functional resistance through bacterial growth inhibition. A separate analysis of this study population found cefotaxime-resistant E. coli in 71% of household floors, and all isolates produced ESBL and were multidrug resistant (53). PCR analysis found that the majority of isolates contained the bla_CTX-M gene, and some samples contained bla_TEM and bla_SHV genes. This analysis complements the prior study to show that household floors and cow dung contain a wide range of additional ARGs beyond those that inhibit beta-lactams; it also remains possible that our sequencing depth limited us from detecting important ARGs.

Our study also has some limitations. First, because metagenomic sequencing did not include field or sample processing controls, there are limitations in determining background contamination potentially introduced during the sample collection and library preparation process, which may lead to false positives. There are also limitations in the methodology of short-read shotgun metagenomic sequencing. It is difficult to determine the sensitivity of metagenomic next-generation sequencing (mNGS) to pick up all present microbes and ARGs equally in a sample as well as clearly distinguish contaminants from sample-associated microbes. Similarly, there are some concerns that taxonomic classification with Kraken could lead to misclassification since the approach may favor incorrect species matches over no match. However, we believe that our use of assembled contigs and probabilistic reclassification with Bracken increases the accuracy of our classifications under these methods. Nonetheless, it is important to note that, here, we present the results of mNGS work as exploratory research, and these results need to be further validated with techniques like PCR and culture-based methods. Second, our choice and use of the pathogen filter at the genus and species level may not be as inclusive of all known human pathogens or granular enough to decipher pathogenicity. For example, within a single microbe genera and species, there may be some strains that are non-pathogenic while others are pathogenic (e.g., E. coli). Third, we only investigated pathogens in cows, but other studies have found that chickens are important for AMR in similar settings (47). We also did not collect data on antibiotic use for cows, chickens, and household members. Fourth, we only detected pathogens using DNA, so our study excluded RNA viruses. Taxonomic classification did include DNA viruses, but none were retained after filtering for potential pathogens. Because we used a cross-sectional design, we were not able to rigorously investigate the directionality of pathogen or ARG transmission between cows and soil. Additionally, as this was an exploratory pilot study, we did not include a control group (e.g., households with soil floors and no cows). Finally, our sample size was small, which limits the statistical precision and generalizability of our findings.

Our findings contribute to the growing literature on household soil and domestic animals as potentially important contributors to disease transmission and as reservoirs of antimicrobial resistance in low-income country settings. Overall, our finding that household soil floors harbored diverse potential pathogens and ARGs underscores the need for housing upgrades that provide hygienic household surfaces and animal management practices that minimize human contact with animal feces and vice versa in low-income settings. Future interventions to reduce infections and AMR in similar household settings may consider focusing on reducing human exposure to soil, including indoor soil floors, and cow dung.

MATERIALS AND METHODS

Sample collection

This study was conducted as a pilot study as part of the CRADLE trial (95). The pilot study enrolled households in Sthal union of Chauhali sub-district in Sirajganj district, Bangladesh. Sthal is located in a rural area on a sand bar within the Jamuna River. It is highly susceptible to flooding and erosion. The community residing in Sthal is primarily composed of agricultural workers, and cattle rearing is common, and the majority of homes have soil floors. Field staff enrolled 10 households. Households were eligible for enrollment if they had a child under 2 years of age, a floor fully made of soil and not fully covered with a mat, carpet, or jute sack, and if cow dung was available for sampling. Because anthrax outbreaks have occurred in the area prior to the study, for the safety of the study staff, we further restricted to households with no cases of anthrax among domestic animals or household members. Cow dung and floor soil were collected from each household.

To collect floor soil, field staff placed a bleach-sterilized 50 cm × 50 cm metal stencil on the floor next to the head of the bed where the child under 2 years of age slept. Using a sterile scoop, they scraped the soil inside the stencil once vertically and then once horizontally, with the goal of collecting 20 g of soil. Soil was placed in sterile Whirlpak bags. Field staff collected fresh cow dung from a defecation event since dawn the same day. They prioritized collecting cow dung from the same room as the floor samples. If this was not possible, they collected it from another room in the house, and if that was not possible, they collected it from the compound courtyard or the edges of the compound. Cow dung was collected using a sterile stool collection tube with a sterile spoon. Floor scrapes and cow dung were placed in a cooler with ice and transported to the International Centre for Diarrhoeal Disease Research, Bangladesh laboratory in Dhaka for analyses. All samples were stored at 4°C overnight and transferred to a −80°C freezer the following morning to be stored until DNA extraction.

DNA extraction

DNA was extracted from 250 mg aliquots of cow dung using the QIAamp PowerFecal Pro DNA Kit (Cat# 51804) and following the manufacturer’s instructions with one minor modification: instead of the recommended 100 µL elution buffer, 70 µL of buffer was used for the elution to achieve a higher DNA concentration. DNA was extracted from 10 g aliquots of floor scrapes using the DNeasy PowerMax Soil Kit (Cat# 12988-10) following the manufacturer’s instructions. The kit was chosen because of its ability to extract DNA from a large mass of soil and its design specifically tailored to soil samples with potentially low nucleic yield. DNA quantification was performed using a Qubit 4.0 Fluorometer. The initial 5 mL volume of DNA elution was concentrated to 200 µL using a 5M NaCl solution as suggested by the manufacturer (Table S4) to enable successful sequencing for shotgun metagenomics. The extracted DNA quality was evaluated for suitability for subsequent shotgun sequencing using both Nanodrop and Qubit measurements. A Nanodrop spectrophotometer was used to assess DNA purity, where a 260/280 ratio reading of 1.8 indicated high quality and minimal contamination. Moreover, the 260/230 ratio reading was measured at 2.0, confirming minimal interference from substances such as carbohydrates or organic compounds.

Library preparation and sequencing

A total of 300 ng DNA was used to prepare libraries by utilizing the Illumina DNA Prep Reagent Kit (Cat#20018705) and an automated liquid handler (epMotion 5075) according to the Illumina DNA Prep Reference Guide (1000000025416). The prepared DNA libraries underwent paired-end (2 × 150 bp) sequencing using the Illumina NextSeq 550 platform (Illumina, San Diego, CA, USA).

Read quality control and contig assembly

The CZID platform (http://czid.org) was used for quality control and contiguous sequence (contig) assembly of raw reads (96). Briefly, after demultiplexing, sequencing reads were processed and analyzed using the default CZID bioinformatics pipeline version 8.3.1. The pipeline performs multiple quality filtration steps on raw reads that include removal of sequence adapters, low-quality bases, low-quality reads, low complexity reads, short reads (<35 bp), and duplicated reads. The pipeline also excludes reads that align to the human genome. For cow dung samples, we included an additional filter to exclude any reads that aligned to the cow host genome by creating a custom reference made up of all Bos taurus (cow) genome assemblies under National Center for Biotechnology Information (NCBI) taxid 9913. As part of the CZID pipeline, each sample was randomly subsampled down to 1 million paired-end reads (2 million total reads) for input into SPAdes (97) to generate de novo assemblies of post-filtered short-read sequences into longer contig sequences.

Taxonomic classification

We conducted taxonomic classification using a previously published workflow (98). We completed taxonomic classification of assembled contigs using Kraken v.2.1.3 (99) with the “PlusPF” index release from 1 December 2024 (k2_pluspf_20240112), which combines various reference sequences from the NCBI Reference Sequence Database that include archaea, bacteria, viral, plasmid, human, protozoa, and fungi genomes. Kraken results were passed through Bracken v.2.9.0 to estimate abundance at every taxonomic level. In some analyses, we restricted Bracken outputs to organisms with known pathogenicity in humans (48).

Antibiotic resistance genes

We used the CZID platform to identify ARGs in both sample types. After read processing and contig assembly, the Resistance Gene Identifier tool (100, 101) was used to match reads and contigs to the Comprehensive Antibiotic Resistance Database (CARD) (102) for AMR gene detection (103). We downloaded report tables from the AMR module and filtered to ARGs with read coverage breadth >10% and >5 reads mapped or contig coverage breadth >10% (103). Because short reads may not adequately differentiate between highly homologous alleles or genes belonging to the same gene family, in the case when a single sample contained multiple ARGs and one was detected by reads and the other by contigs, we retained the one detected by contigs (103). We generated a heatmap of ARGs within each sample; to reflect alignment confidence, we classified each ARG by a categorical variable indicating contigs match quality (perfect, strict, nudged) or read matches consistent with a prior study using CZID for ARG detection (103). “Perfect” contig matches identically matched reference sequences in CARD (102). “Strict” contig matches were those that matched previously unknown variants of known ARGs with additional screening for key mutations. “Nudged” contig matches had at least 95% identity to known AMR genes and were matched using a percent identity threshold not taking alignment length into account. “Nudged” matches may capture distant homologs but may also have a higher false-positive rate. Heatmaps show the highest confidence level for contig matches for each ARG in each sample.

Statistical analysis

We determined the alpha-diversity within each sample type using the richness attribute, Chao index for species richness (104–106), Pielou’s evenness index (49), Shannon index (50), Simpson index (51), and inverse Simpson index. We tested for differences in alpha-diversity between sample types using the Wilcoxon signed-rank test (107). To assess microbial diversity, we calculated the Bray-Curtis dissimilarity (108) and investigated clustering by sample type, household membership, and the presence of visible animal feces on floors at the time of measurement using principal coordinates analysis (109) and permutational multivariate analysis of variance (110). Alpha- and beta-diversity metrics were calculated using the R package vegan (111).

ETHICS APPROVAL

The parent study was approved by the International Centre for Diarrhoeal Disease Research, Bangladesh Ethical Review Committee (PR-22069) and the Stanford Institutional Review Board (63990). Participants provided written informed consent.

DATA AVAILABILITY

Raw sequencing data are deposited under the NCBI Bioproject PRJNA1130536. Analysis scripts and files are available at https://github.com/kalanir/cradle-pilot-seq.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/aem.00669-25.

Supplemental tables and figures. aem.00669-25-s0001.docx.

Tables S1 to S4 and Fig. S1 to S4.

Supplemental File S1. aem.00669-25-s0002.csv.

Full list of taxa detected by sample type.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.