Highly Specific Sewage-Derived Bacteroides Quantitative PCR Assays Target Sewage-Polluted Waters

Bacteroides are major members of the gut microbiota, and host-specific organisms within this genus have been used extensively to gain information on pollution sources. This study provides a broad view of the population structure of Bacteroides within sewage to contextualize the well-studied HF183 marker for a human-associated Bacteroides. The study also delineates host-specific sequence patterns across multiple hypervariable regions of the 16S rRNA gene to improve our ability to use sequence data to assess water quality. Here, we demonstrate that regions downstream of the HF183 marker are nonspecific but other potential human-associated markers are present. Furthermore, we show the most abundant Bacteroides in sewage is free living, rather than host associated, and specifically found in sewage. Quantitative PCR assays that target organisms specific to sewer pipes offer measures that are independent of the human microbiome for identifying sewage pollution in water.

ABSTRACT

The identification of sewage contamination in water has primarily relied on the detection of human-associated Bacteroides using markers within the V2 region of the 16S rRNA gene. Despite the establishment of multiple assays that target the HF183 cluster (i.e., Bacteroides dorei) and other Bacteroides organisms (e.g., Bacteroides thetaiotaomicron), the potential for more human-associated markers in this genus has not been explored in depth. We examined the Bacteroides population structure in sewage and animal hosts across the V4V5 and V6 hypervariable regions. Using near-full-length cloned sequences, we identified the sequences in the V4V5 and V6 hypervariable regions that are linked to the HF183 marker in the V2 region and found these sequences were present in multiple animals. In addition, the V4V5 and V6 regions contained human fecal marker sequences for organisms that were independent of the HF183 cluster. The most abundant Bacteroides in untreated sewage was not human associated but pipe derived. Two TaqMan quantitative PCR (qPCR) assays targeting the V4V5 and V6 regions of this organism were developed. Validation studies using fecal samples from seven animal hosts (n = 76) and uncontaminated water samples (n = 30) demonstrated the high specificity of the assays for sewage. Freshwater Bacteroides were also identified in uncontaminated water samples, demonstrating that measures of total Bacteroides do not reflect fecal pollution. A comparison of two previously described human Bacteroides assays (HB and HF183/BacR287) in municipal wastewater influent and sewage-contaminated urban water samples revealed identical results, illustrating the assays target the same organism. The detection of sewage-derived Bacteroides provided an independent measure of sewage-impacted waters.

IMPORTANCE Bacteroides are major members of the gut microbiota, and host-specific organisms within this genus have been used extensively to gain information on pollution sources. This study provides a broad view of the population structure of Bacteroides within sewage to contextualize the well-studied HF183 marker for a human-associated Bacteroides. The study also delineates host-specific sequence patterns across multiple hypervariable regions of the 16S rRNA gene to improve our ability to use sequence data to assess water quality. Here, we demonstrate that regions downstream of the HF183 marker are nonspecific but other potential human-associated markers are present. Furthermore, we show the most abundant Bacteroides in sewage is free living, rather than host associated, and specifically found in sewage. Quantitative PCR assays that target organisms specific to sewer pipes offer measures that are independent of the human microbiome for identifying sewage pollution in water.

INTRODUCTION

Human fecal pollution in urban waters from untreated sewage contains pathogenic bacteria, virus, and protozoa that cause gastrointestinal diseases through the ingestion of polluted water (1, 2) or skin, eye, and respiratory infections through direct contact (1). Human sources of fecal pollution are considered a higher health risk than animal sources of fecal pollution (3, 4). The detection of traditional fecal indicator bacteria (FIB), such as fecal coliforms, Escherichia coli, and enterococci (5), does not distinguish human sources from animal sources of fecal pollution, because they commonly occur in all mammalian intestines (6). Many studies have demonstrated a lack of correlation between FIB levels and pathogen occurrence or adverse human health outcomes (7–10), because some sources of fecal pollution do not carry human pathogens.

Microbial source tracking (MST) methods, which rely on the quantification of levels of certain fecal microorganisms that are specific to a host (11), have been used for fecal source identification for a number of years (12). To date, the most characterized microorganisms used in MST belong to the genus Bacteroides within order Bacteroidales (see Table S1 in the supplemental material), one of the most predominant genera in the human gut. The best-studied human Bacteroides marker, the HF183 marker, is found in Bacteroides dorei and its closely related taxa (13) and is located in the V2 hypervariable region of the 16S rRNA gene. This marker was first reported by Bernhard and Field (14) as a PCR assay (i.e., HF183F/Bac708R) (Table S1) (14–16).

Because most human-associated Bacteroides markers have been developed using clone libraries that target Bacteroidales using the Bac708R primer (13), quantitative PCR (qPCR) assays have been limited to the V2-V4 hypervariable regions. Assays and their average specificities include HF183/SSHBac-R (91.1%) (17–21), HF183/BFDrev (76.8%) (22, 23), HB (90.9%) (24, 25), HF183/BacR287 (91.2%) (21, 23, 25), BacHum-UCD (77.9%) (15, 18, 20, 21, 26), BacH (92.6%) (15, 21, 27), HuBac (54.5%) (15, 16, 18, 28), Human-Bac1 (44.4%) (15, 29), and BacHuman (81.5%) (30). More detailed information on these assays is found in Table S1, and a primer map is shown in Fig. S1.

In addition to the assays that use the HF183 marker directly as a forward primer (i.e., HF183/SSHBac_R, HF183/BFDrev, HB, and HF183/BacR287), assays that use primers or probes that overlap the HF183 marker (Fig. S1), such as the BacHum-UCD and BacH assays, also reported low-level animal cross-reactivity, further demonstrating the human specificity of the HF183 marker. Human Bacteroides fecal marker PCR/qPCR assays have also been developed for the 16S rRNA gene and genomic sequences of Bacteroides thetaiotaomicron (Table S1) (22, 31, 32), another predominant species in human feces that typically shows up more often in human feces than in animal sources (31, 33). Overall, there is no one assay that is exclusively human specific, and animal source cross-reactions were reported for all the PCR/qPCR assays mentioned above, such as cat, dog, pig, chicken, turkey, cow, and deer (Table S1).

The goal of this study was to explore the potential of genus Bacteroides for MST, in addition to the widely applied HF183 marker, to expand methods for sewage detection and quantification. The characterization of the population structure using hypervariable regions beyond V2, and delineation of the linkage patterns among markers in different regions, may reveal additional host-preferred and host-specific Bacteroides organisms and help couple community sequencing data to marker assays. To explore the host specificity of this genus, we compared the population structure of Bacteroides in 27 sewage and 151 animal fecal samples using next-generation sequencing (NGS) across multiple variable regions. We also explored the human Bacteroides V2, V4V5, and V6 region sequence linkages and specificities by analyzing V2-V9 region sewage clone libraries. Several human-associated Bacteroides markers that were not related to the HF183 marker, including one from the V4V5 region and one from the V6 region, were identified from NGS data. We also identified a sewage-associated Bacteroides that appears to be specifically propagated in urban sewer systems and developed two TaqMan qPCR assays targeting the V4V5 region and the V6 region.

RESULTS

Bacteroides population structures in sewage, animal hosts, and freshwater samples.

We applied oligotyping to V6 region sequences of Bacteroides from 27 sewage influent samples and 151 animal fecal samples. In total, 1.48 × 10⁷ Bacteroides reads (97.66% of total reads) were analyzed, including 1.96 × 10⁶ reads from sewage samples and 1.29 × 10⁷ reads from animal fecal samples. The oligotype (n = 1,730) distribution in each sample is shown in Fig. 1. Eighty-two oligotypes were exclusively found in sewage, and of these, 30 were among the 100 most abundant sewage oligotypes. The sewage oligotype patterns were consistent between U.S. and Spain sewage samples and were distinguishable from animal hosts (Fig. 1). The animal and sewage oligotype profiles were dissimilar in individual host groups (n = 8, adonis R² = 0.419, P = 0.001) and in sewage compared to those of a pooled animal group (adonis R² = 0.119, P = 0.001). A Bray-Curtis dissimilarity-based hierarchical cluster analysis of Bacteroides oligotypes demonstrated animal and sewage samples clustered by source, and sewage was the most distant sample group compared to all other animal groups (see Fig. S2). In addition, certain oligotypes were associated with specific hosts (Fig. 1 and S2). For example, 80 oligotypes were found only in cows, 11 were only in deer, 5 were only in dogs, and 4 were only in pigs, indicating that genus Bacteroides is a good target for animal fecal markers.

FIG 1.

Oligotype patterns of the V6 region sequences of the Bacteroides 16S rRNA gene in sewage and seven animal hosts. Samples are grouped by host types. Colors correspond to different sequences in the profile, with the bar height representing the oligotype relative abundance.

We also examined freshwater samples using the Indicspecies package to identify potential freshwater Bacteroides sequences based on the relative abundances of unique V6 sequences in freshwater samples (n = 35) compared with those in sewage and animal samples (n = 178). Three unique Bacteroides V6 sequences were found only in freshwater samples (mean ± standard deviation [SD] relative abundance, 4.7% ± 9.3% of all Bacteroides), and 27 unique sequences were found in freshwater with comparatively low occurrence in sewage (relative abundance 37.4% ± 32.1% of all Bacteroides compared with 1.5% ± 0.74% in sewage) and with no occurrence in animal samples. BLAST results against the NCBI nucleotide database showed no identical matches between a human fecal source and the three freshwater-specific sequences; for the 27 “freshwater-preferred” sequences, only two were found to have identical matches with a human stool source, and another two were found to match with bioreactors using farm animal waste (e.g., cow and pig). This indicates that there is a potential for Bacteroides populations to occur in the freshwater environment in the absence of fecal contamination.

Identification of V4V5 and V6 regions downstream of the HF183 human Bacteroides marker.

We utilized our sewage clone libraries to examine the specific marker sequences in the V4V5 and V6 regions of 16S rRNA gene that were downstream of the HF183 marker sequence. A total of 136 clones matching the HF183 marker (41% of sequences) were found in library 1, which targeted human Bacteroides by using the BacH_f primer to make the library. There was one primary V4V5 and one primary V6 sequence downstream of the HF183 organism (Fig. 2A1). Only 3% of sequences in library 2 (representing total Bacteroides from sewage) had the HF183 marker (Fig. 2B1), indicating that the HF183 marker cluster is a small fraction of Bacteroides in sewage. All HF183-positive clones in library 2 had the BacH_f primer site, supporting that library 1 was inclusive of HF183 cluster organisms.

FIG 2.

Associations of the V2, V4V5, and V6 regions of sewage Bacteroides clones. (A1) Associations of the three regions in HF183-matched clones in library 1. (A2) Associations in non-HF183 clones in library 1. (B1) Associations in HF183-matched clones in library 2. (B2) Associations in non-HF183-clones in library 2. The deep/light blue circles represent V2 region, black/gray circles represent V4V5 region, and red/pink circles represent V6 region. Circle sizes are directly proportional to the sequence read numbers (except the non-HF183-matched V2 region in library 2, which is smaller than the proportional area). The unique type numbers in V4V5 and V6 regions are annotated at the bottom of A2 and B2. Numbers within parentheses indicate clone numbers. Clones that have no NGS matches are not included.

We used the NGS data set of animal fecal samples to examine the host distribution of the primary sequences downstream of the HF183 marker (designated V4V5-4 and V6-4 according to their rank of abundance in sewage samples in corresponding NGS data sets) and found they occurred in multiple animals. The V4V5-4 sequence occurred in 40% of the samples, including cat, dog, cow, and deer, and the V6-4 sequence was found in 16.8% of the samples, including cat, dog, cow, pig, chicken, deer, raccoon, and rabbit, indicating the regions downstream of the HF183 marker are not specific to humans (see Fig. S3). We tested a subset of these samples for the HF183 marker by qPCR in cases where DNA material was available. Overall, 2 of 13 samples containing the V4V5-4 sequence were positive by the HF183/BacR287 assay. For available samples containing the V6-4 sequences, only one of three samples was positive by the HF183/BacR287 assay. These results support that the downstream region is not specific, as opposed to these animals carrying an HF183-positive organism.

Potential human and sewage markers in Bacteroides V4V5 and V6 regions that are not associated with the HF183 marker cluster.

We aimed to identify additional human- or sewage-associated Bacteroides markers in the V4V5 and V6 regions so that they could be used in PCR applications but, more importantly, could also be used as markers in sequencing data sets, since these regions are commonly sequenced. We applied the Indicspecies permutation test and identified markers from the V4V5 region and V6 region that were >90% sewage specific and sensitive for sewage. Within these, there were nearly 20-fold more V6 region markers than V4V5 markers that were 100% specific and sensitive to sewage (see Fig. S4). These results may be due to the higher variability in the V6 region, which provides more resolution and therefore more unique human- or sewage-associated sequences than the V4V5 region. Although the V4V5 region had fewer markers, there were two that had 100% specificity and sensitivity (V4V5-1 and V4V5-7), both of which did not appear in human gut microbiome data sets, suggesting they were resident organisms within the sewer pipes. There were seven markers identified in the V6 region, with only one of these associated with human feces. The most abundant sewer pipe-associated markers fell within a clade of Bacteroides graminisolvens (see Fig. S5). Human-associated and sewer pipe-associated markers and their specificities for the V4V5 region are listed in Table S2, and markers in the V6 region are found in Table S3.

Assay development and sensitivity for sewage detection.

TaqMan qPCR assays were developed to target the most abundant Bacteroides in sewage, which were designated the BacV4V5-1 assay and BacV6-21 assay (Table 1). We tested 40 U.S. sewage influent samples, including 20 from Milwaukee and 20 from ten other U.S. cities, using the two assays and compared the results with those from the HB and HF183/BacR287 assays (Fig. 3). All four assays showed 100% sensitivity in sewage samples. In Milwaukee sewage samples, the BacV6-21 assay showed approximately the same magnitude of copy numbers (CN) (4.9 × 10⁷ ± 5.8 × 10⁷ CN/100 ml [mean ± SD]) as the HB assay (5.8 × 10⁷ ± 3.3 × 10⁷ CN/100 ml) and the HF183/BacR287 assay (5.1 × 10⁷ ± 2.6 × 10⁷ CN/100 ml), but was found to have greater fluctuation. The BacV4V5-1 marker was approximately 4-fold higher than the V6-21 marker, with CN equal to 1.9 × 10⁸ ± 2.2 × 10⁸ per 100 ml. In other U.S. cities, similar sewage sensitivities were detected for BacV4V5-1 and BacV6-21 assays, with the BacV6-21 assay showing 5.5 × 10⁷ ± 6.67 × 10⁷ CN/100 ml and the BacV4V5-1 assay showing 1.9 × 10⁸ ± 2.3 × 10⁸ CN/100 ml. The V4V5-1 marker mirrored the V6-21 marker fluctuation, suggesting they target the same organism (Pearson’s r = 0.931, P < 2.2e−16). Likewise, the HB assay and the HF183/BacR287 assay were tightly coupled (Pearson’s r = 0.990, P < 2.2e−16). The V4V5-1 and V6-21 markers were not correlated to either the HB or the HF183/BacR287 marker, with the Pearson’s r values ranging from −0.083 to −0.061.

TABLE 1.

BacV4V5-1 and BacV6-21 marker assays

Assay marker	Primer or probe name and sequence (5′→3′)
	Forward primer	Probe	Reverse primer
V4V5-1	Bac573f, AAGGGAGCGTAGGTTGACATA	Bac599p, FAM-CAGCTGTGAAAGTTTACGGCTC-NFQ-MGB	Bac673r, CGCCCACCTCTTGTACACT
V6-21	Bac989f, GCTTGAATTGCAGAGGAATA	Bac1010p, FAM-AGTTGAAAGATTATGGCCGCA-NFQ-MGB	Bac1162r, GCAGTCTCACTAGAGTCCTCAG

FIG 3.

Comparison of the BacV4V5-1, BacV6-21, HB, and HF183/BacR287 assay copy numbers (CN) in sewage samples. (A) Results from four assays showing CN in 20 sewage samples from Jones Island (JI) and South Shore (SS) wastewater treatment plants, Milwaukee, WI. (B) Results from four assays showing CN in 20 sewage samples from ten other U.S. cities, each tested at two different time points.

The four Bacteroides assays were also tested using freshwater samples that had no known evidence of human fecal pollution (n = 30). The HB, HF183/BacR287, and BacV6-21 assays all showed negative results. The BacV4V5-1 assay, however, showed low CN in two lake/harbor samples (200 ± 6 CN/100 ml [mean ± SD]) and six beach samples (180 ± 111 CN/100 ml). All qPCR results for the four assays are shown in Data Set S1.

Assay validation in animal fecal samples.

We validated the newly designed assays using samples from a total of 76 animals in the formats of individual and pooled samples, including from cat, dog, pig, cow, deer, gull, and chicken (Table 2). The BacV4V5-1 and BacV6-1 assays showed higher specificity than the two human-associated assays. The BacV4V5-1 assay gave a very low signal (5.2 CN/ng DNA) in one pig sample (pig pool 3) at 1 ng · μl⁻¹, and was negative at 0.1 ng · μl⁻¹ and 0.01 ng · μl⁻¹ DNA template levels. The BacV6-21 assay was negative for all animals at all three dilutions of DNA template. In contrast, the HB and HF183/BacR287 assays showed sporadic cross-reactivity with animals. The HB assay cross-reacted with one dog pool sample, whereas the HF183/BacR287 assay was negative for this sample (25). Results for all three dilutions of DNA are detailed in Data Set S1.

TABLE 2.

Animal validation results of the Bacteroides assays

Animal	Total no. (no. of pools containing two samples)	Results for:
		BacV4V5-1		BacV6-21		HBa		HF183/BacR287a
		No. positive	Average CN per ng DNA/average CN per g feces	No. positive	Average CN per ng DNA/average CN per g feces	No. positive	Average CN per ng DNA/average CN per g feces	No. positive	Average CN per ng DNA/average CN per g feces
Cat	13 (1)	0	0	0	0	1	8/115,000	1	5/77,000
Cow	10 (2)	0	0	0	0	0	0	0	0
Deer	11 (1)	0	0	0	0	3b ,c	406/404,000	3b ,c	364/362,000
Dog	13 (2)	0	0	0	0	2b	375/3,330,000	0	0
Pig	22 (2)	2b ,c	5/84,700	0	0	0	0	0	0
Chicken	4	0	0	0	0	0	0	0	0
Gull	4	0	0	0	0	0	0	0	0

^aPartial results for the HB and HF183/BacR287 assay validation were generated in a previous study (25).

^bA pool was positive in each of these animal groups and was counted as two animals being positive.

^cThe positive pooled samples were also run in format of individuals at 1 ng · μl⁻¹ DNA material level (see Data Set S1 in the supplemental material for details).

Sensitivity of BacV4V5-1, BacV6-21, HB, and HF183/BacR287 assays for environmental water samples.

We tested the BacV4V5-1, BacV6-21, HB, and HF183/BacR287 assays using 20 sewage-contaminated local river water samples and 13 known agricultural contaminated local river water samples. Overall, the four assays were significantly correlated for these environmental water samples (Table 3). The BacV4V5-1 and BacV6-21 assay CN results showed very similar fluctuation patterns and were more highly correlated with each other than with the HB or HF183/BacR287 assay (Table 3). The BacV4V5-1 marker CN was 4.0- ± 1.4-fold (mean ± SD) higher than the BacV6-21 marker CN, which corresponded to levels in the U.S. sewage samples that were tested. The HB and HF183/BacR287 assays showed identical CN fluctuation patterns in sewage-contaminated river samples (see Fig. S6A) and in agricultural contaminated samples containing low levels of sewage (Fig. S6B) and were highly correlated with each other, indicating the equivalency of these two HF183 marker-based assays. In addition, all four assays clearly distinguished human contamination from that of ruminants in agricultural contaminated river samples, with none of the results from the assays proportionally increasing with increasing ruminant contamination (Fig. S6). Detailed CN data are shown in Data Set S1.

TABLE 3.

Pearson’s correlation of the four Bacteroides assays in 20 sewage-contaminated and 13 agricultural contaminated water samples

Assay	Pearson’s r (P value) for:
	BacV4V5-1	BacV6-21	HB	HF183/BacR287
BacV4V5-1	1.000
BacV6-21	0.995 (<2.2 × 10−16)	1.000
HB	0.842 (7.9 × 10−10)	0.817 (6.7 × 10−9)	1.000
HF183/BacR287	0.824 (3.8 × 10−9)	0.792 (3.9 × 10−8)	0.995 (<2.2 × 10−16)	1.000

DISCUSSION

Potential of genus Bacteroides to contain host and sewage markers.

The identification of human fecal pollution provides evidence to assess public health risks caused by waterborne diseases. The fecal anaerobic microorganism Bacteroides has been utilized as a target for human fecal source detection since the early 2000s, when the specificity of the HF183 cluster was identified (12). Our study further explores the host specificity patterns of this genus among 27 sewage and 151 animal fecal samples across seven hosts by using deep-sequencing data. We demonstrated the host-specific nature of Bacteroides populations, consistent with previous studies using the V4V5 and V6 variable regions (34) and the V2 region (18, 28). The oligotyping results from previous studies and this study suggest that the V6 region from genus Bacteroides could be used as markers for certain animals, such as cows and deer, since specific patterns were evident within these hosts. For example, dairy cow and beef cattle had dissimilar Bacteroides oligotypes, which may be caused by dietary differences (35, 36) (Fig. 1; see also Fig. S2 in the supplemental material). With the high variability among cattle, the development of more restrictive host animal fecal markers could be useful; for example, specific markers targeting dairy cows or cattle on forage diets common to certain regions may be more feasible than employing a “universal” cattle marker.

Multiple markers within the V4V5 and V6 regions were identified as being specific to sewage. Many of these did not match human microbiome organisms, but appeared to originate within the sewer pipes, and included the most abundant Bacteroides in sewage. Bacteroides in mammalian guts is responsible for the breakdown of complex polysaccharides (37–40). In addition, studies have been focused on free-living Bacteroides species, which also have the ability to degrade complex organic matter, such as polysaccharides (41–43). The sewer pipe-derived Bacteroides organism represented by the V4V5-1 and V6-2 (and V6-21) markers closely matched B. graminisolvens based on near-full-length clone sequences (Fig. S5). This organism was isolated from a methanogenic reactor at a cattle farm, where it was implicated in breakdown of hemicellulose (41), and has been detected in a microbial fuel cell reactor, where it performed similar functions (i.e., degrading carbohydrates) (44). Just as Bacteroides has coevolved and been selected for in the human gut (45), it appears that Bacteroides organisms in the urban sewer infrastructure may have been selected for or evolved in sewer pipes as a result of the nutrition available to them from human waste, where they provide further breakdown of material not completely utilized in the gut. Most notable is the ubiquitous occurrence of identical V4V5 and V6 marker sequences in all of the cities studied. It is unknown if these organisms were originally deposited as minor members of the human gut microbiome or if they arose from an environmental source. Given the short transit time in some of the systems studied (i.e., 6 to 24 h) (46), coupled with the high abundance patterns in relation to what is found in human fecal material, the sewer pipe-derived Bacteroides organisms appear to be residents in the system.

We designed assays targeting the most abundant Bacteroides represented by the V4V5-1 marker. The most common marker for this organism in the V6 region was V6-2 (Fig. 2B). However, this sequence was also found in animals. Therefore, we targeted a smaller subpopulation for qPCR assays, represented by the V6-21 marker for the V6 region assay. The sewer pipe-associated marker assays strengthen the capability to detect human sources of pollution, because they are sewer derived and have essentially no cross-reactivity with animal sources, unlike gut-derived organisms, for which the distinguishing members of the community are more often host-preferred organisms versus those that are strictly host specific (6). Furthermore, sewer pipe-associated markers may not be subjected to differences in the human microbiome in different regions, as is observed with some of the human-derived markers (47, 48). Further testing of urban sewer systems worldwide is needed to determine the applicability of the markers in areas where the HF183 marker is low or absent. Importantly, since this organism appears to be free living rather than host associated, further validation studies of uncontaminated water are needed to determine if this organism is exclusively found in sewer systems and similar environments (manure detention ponds, anaerobic digesters, etc.).

In addition, we demonstrated the presence of Bacteroides in a freshwater environment, which differed from Bacteroides in sewage and in the seven animal fecal sources. Bacteroides in freshwater has previously been reported on Cladophora mats (49, 50), which is consistent with the organism’s ability to breakdown complex polysaccharides. In general, there was a single dominant sequence type in an apparently uncontaminated sample, and different samples had different sequence types (Data Set S2, Tab 2), indicating the freshwater Bacteroides population may be very diverse and specific to the location. Freshwater Bacteroides may be detected when using universal Bacteroides marker assays (18), causing false-positive results for fecal pollution detection. In addition, high levels of these organisms may interfere with Bacteroides assays that employ closely related primer sequences.

The HF183 assays and the sewer-associated Bacteroides assays target two independent Bacteroides organisms but are overall correlated.

The high correlation between the two HF183-based assays (Table 3) indicated that they amplify the same Bacteroides organism. In our animal validation results, the HF183/BacR287 assay showed better specificity (93.2%) than the HB assay (90.5%) because of cross-reactivity with certain dog samples in the latter assay. Overall, these two assays showed nearly identical sensitivity patterns among sewage and sewage- and agriculture-contaminated environmental water samples, demonstrating they are interchangeable for the purpose of human fecal source detection. However, their application needs to be considered cautiously when employing the HB marker if dog waste is suspected, and specific testing using a canine marker or verification using a second human marker should be considered.

The BacV4V5-1 and BacV6-21 markers (targeting a sewer pipe-derived Bacteroides) had consistent ratios in sewage and sewage-contaminated environmental water samples (i.e., the CN was approximately 4.0- ± 1.0-fold higher in the BacV4V5-1 assay than in the BacV6-21 assay) and were highly correlated in environmental waters. The linkage of the V4V5-1 and V6-21 markers in clone libraries (Fig. 2B2) verified that these two assays target the same Bacteroides organism. In water samples where sewage was present, all four assays were correlated, demonstrating that they all detect sewage similarly.

Next-generation sequencing revealed potential cross-reactions of human fecal marker with animals.

Having access to a large V4V5 NGS data set allowed us to examine other established human-associated Bacteroides assays targeting the V4 region. We compared the HumanBac-1 (29) and HuBac (28) assays, which both target the V4 region of the Bacteroides 16S rRNA gene, with our V4V5 NGS data set; exact primer and probe matches in both assays were found in cat, dog, pig, cow, deer, and rabbit, suggesting true animal cross-reactions occur, which explains the comparatively low human specificity of these assays (Table S1).

Cross-reaction of human fecal markers with animals can be influenced by multiple complex factors, such as similarities in gut microbial communities as a result of dietary factors (25, 35, 51) and possible animal ingestion of human waste (52). We have previously noted that employing markers from two different bacterial groups, such as Bacteroides and Lachnospiraceae, can increase confidence in results where cross-reactivity is suspected (25). For a well-designed fecal marker qPCR assay (e.g., one optimized for avoiding dimers, hairpin structures, annealing temperature, etc.), NGS could not only be used to verify the assay’s host specificity but also identify closely related sequences that might interfere. Deep sequencing has also been proven to be valuable for the identification of host-associated markers on the scale of the whole microbial community without the effort of constructing a sequence clone library. However, linking different regions to the same organism is difficult without continuous more-extended sequence data, since some variable regions appear to be less discriminatory and found in multiple host types. Therefore, sequencing databases for common regions of 16S rRNA genes in sewage and animal fecal samples from a wide geographical range (e.g., across the United States) with key host information (e.g., animal diet and cohabitation) could help to verify the applicability of markers and interpret site-specific data. Data like these could be shared between research laboratories and would be extremely useful in assay validation in silico, therefore providing substantial evidence of specificity and sensitivity (25).

Combining NGS and qPCR for water quality assessments.

qPCR is indispensable for rapidly quantifying sources of fecal pollution such as human or cattle waste. However, most contamination scenarios are complex, especially in urban environments where there may be sewage contamination mixed with nonpoint sources from storm water that add a significant fecal indicator bacterium burden (24). In addition, there are known sensitivity and specificity issues with individual fecal bacterium markers; the microbiomes of many animals have not been characterized, and so cross-reactivity has not been completely characterized (53). NGS data create a high-resolution inventory of organisms present, and with falling sequencing costs, NGS may be feasible to employ in the future for directly characterizing fecal pollution sources. Computational or machine learning approaches such as Source Tracker (53) or random forest (54) can use sequence abundance patterns of the whole community, or of taxonomic groups, to identify pollution signals within a water sample. These methods rely on signatures of sequences that include their relative abundance patterns within the community, and sequences shared between sources generally do not also share overall relative abundance patterns within the signature (54). Furthermore, fecal bacterium sequences within these data sets that do not match a characterized source could be used to indicate extraneous sources that may be contributing fecal indicator bacteria but are not considered a significant human health risk (i.e., bird or pet waste and urban wildlife). Anchoring the relative abundance derived from sequencing with qPCR for host-associated markers will enable quantification. As the complexity of fecal pollution signals is unraveled, combining NGS with qPCR methods for source tracking may become common metrics in the future for assessing microbial water quality.

MATERIALS AND METHODS

Sample collection and DNA extraction.

Influent sewage samples used for qPCR in this study were from Jones Island (JI) and South Shore (SS) wastewater treatment plants (WWTPs) in Milwaukee, WI (n = 20), along with samples from ten other U.S. cities representing geographical regions of the United States that were sampled in two different seasons over a year (n = 20) (55). Sewage-contaminated river water samples (n = 20) were collected during a 2016 Milwaukee combined sewer overflow (CSO) event. Agricultural contaminated water samples were collected from the Milwaukee River (n = 13) after rains in the spring and early summer in 2014 and 2015; these samples also had evidence of sewage contamination but at 3 to 4 orders of magnitude lower than ruminant contamination as determined using a ruminant marker (56). Freshwater samples that had no evidence of human fecal contamination, i.e., had zero or extremely low colony counts of FIB and were negative in HB and human Lachnospiraceae qPCR assays (25), were collected from Lake Michigan (n = 20) and Milwaukee area beaches (n = 10).

A total number of 76 animal fecal samples, including from 22 pigs, 13 dogs, 12 cats, 11 deer, 10 cows, 4 gulls, and 4 chickens, were collected for qPCR assay validation. Among these animal fecal samples, 46 were extracted in a previous study (25) but rediluted for qPCR experiments in this study. Fecal sample processing and storage were as described previously (25).

All sewage, animal fecal, and environmental water sample details, including their associated studies and qPCR results, are listed in Data Set S1 in the supplemental material.

NGS data used for oligotyping, clone comparisons, and marker identification.

To examine the overall population structure of Bacteroides populations, sequence data generated from two previous studies (25, 34) of the V6 region 16S rRNA gene from 27 sewage samples and 151 animal fecal samples, including hosts of cat, dog, pig, cow, deer, raccoon, and chicken, were analyzed using oligotyping (57). All raw sequences were trimmed using cutadapt software (58) and assembled using PEAR (59) software. Sequences were then classified using GAST (60) with a comparison to SILVA reference database version 132 to parse out Bacteroides sequences. Oligotyping was run using parameters -s (the minimum number of samples with an oligotype present) equal to 9 (5% of total sample), -M (the minimum substantive abundance) equal to 85, and -c (number of base locations) equal to 33. The output of the oligotype count matrix was plotted using ggplot2 package (61) in R (version 3.5.1) (62). The statistical analysis of sewage and animal oligotypes was performed with the adonis function in the vegan package (63) in R.

For clone comparisons and marker identification (described below), V4V5 and V6 sequence data sets from previous studies (25, 34, 55) were obtained from the Visualization and Analysis of Microbial Population Structures platform (VAMPS; https://vamps2.mbl.edu) (64) with reference to SILVA database version 119. A taxbyseq file, which described whole community unique sequences, taxonomy, and abundance in each sample, was used. The total number of sequences for each sample was normalized to the median total sequence count for all samples (V4V5 region data set, 89,341; V6 region data set, 741,189). Singletons were removed to form the whole community NGS data sets. The genus Bacteroides data were then extracted. The samples, their usage in this study, the associated studies, and SRA study accession numbers are listed in Data Set S2, Tab 1.

Sewage clone libraries.

Two sewage clone libraries were generated using four sewage influent samples collected from different U.S. cities (Milwaukee, Palo Alto, Laramie, and Key West) in August 2012 (55). The first clone library (library 1) was constructed using a human Bacteroides group forward primer (BacH_f) (27) and a universal 16S rRNA gene reverse primer (1492R); the BacH_f primer was chosen to form human Bacteroides amplicons that were long enough to cover the V2 region. The second clone library (library 2) was generated using the universal 8F primer and a new reverse primer, designated 1030R (5′-CCACCTTCCTCACATCTTACGA-3′), which was designed to broadly target Bacteroides. The Probe Match function in the Ribosomal Database Project (RDP) (65) demonstrated that the 1030R primer matched 34,100 of 35,602 Bacteroides sequences. The PCR products were cloned into the pCR2.1 vector using the TOPO TA Cloning kit (Invitrogen, Carlsbad, CA), and plasmids were extracted as previously detailed (46). Sanger sequencing was performed with M13F, 331F, and M13R primers using the ABI BigDye Terminator kit (Applied Biosystems, Foster City, CA) on an ABI 3730xl DNA Analyzer (Applied Biosystems, Foster City, CA) (46). In all, 332 sequences were generated in library 1 and 375 sequences were generated in library 2. Nearly 64.6% of the Bacteroides sequence types in the V4V5 data set and 35.5% in the V6 NGS data set were represented in library 2.

Linkage of the HF183 marker representing the V2 region with the V4V5 region and the V6 region of Bacteroides.

The sewage clone libraries were compared with the HF183 marker sequence to identify clones containing this marker, and then the corresponding unique V4V5 and V6 sequence types in the clones were identified in the NGS data sets; both comparisons were performed using BLAST+ (66). The V4V5 and V6 sequence types for each HF183 clone were compiled in Excel using the VLOOKUP function.

Freshwater Bacteroides population identification.

We used freshwater samples with no or low levels of fecal pollution (n = 35) that were previously sequenced for the V6 region to identify environmental Bacteroides (Data Set S2, Tab 1). These were compared to the sewage and animal fecal samples used in oligotyping. The “uncontaminated” samples were collected under baseflow conditions (i.e., no rain in the previous 48 h) from Lake Michigan nearshore and offshore surface water (n = 6) as grab samples (67) and from the Milwaukee, Kinnickinnic, and Menomonee Rivers (n = 29) using automated Teledyne ISCO 3700 full-size portable sequential samplers (56). To identify freshwater-preferred Bacteroides sequences, the R package Indicspecies (68) was applied with a setting of 999 permutations. These V6 region Bacteroides sequences are listed in Data Set S2, Tab 2.

Bacteroides marker identification.

We used sewage, sewage-contaminated water, and animal fecal samples previously sequenced for the V4V5 region (25, 55) and V6 region (25, 34) for marker identification (Data Set S2, Tab 1). For the V6 region, 22,006 unique Bacteroides sequences were present from 40 sewage and sewage-contaminated water samples and 156 animal fecal samples; for the V4V5 region, 22,104 unique Bacteroides sequences were present from 195 sewage and 60 animal samples (see Data Set S2, Tab 3 and Tab 4 for the 100 most abundant Bacteroides V4V5 and V6 region sequences in sewage). These unique Bacteroides sequences were named according to their abundance ranks in sewage samples in the data set, e.g., V4V5-1 and V6-1 are the most abundant V4V5 and V6 unique Bacteroides sequences in sewage samples, respectively.

To identify sewage-associated Bacteroides markers in the V4V5 and V6 regions, we used a subset of samples from the NGS data sets where both V4V5 and V6 regions were sequenced, which included 16 sewage and 51 animal fecal samples. These data were analyzed using R package Indicspecies (68), and the numbers of indicators for each region that were >90% host specific and sensitive were compared. To identify candidate sewage-associated Bacteroides markers, candidates were chosen by the criteria that they were >90% sensitive and 100% specific from Indicspecies results. To identify the probable source of these marker candidates (i.e., whether they are human derived or likely residents of the sewer pipes), we compared these marker candidate sequences with the National Center for Biotechnology Information (NCBI) nucleotide database and published V3V5 (69–71), V4V6 (72), V6 (73), and V6V8 (74) region human stool sequences using BLAST+. The most abundant Bacteroides in sewage was specific to sewage but did not appear to be of fecal origin. This organism was chosen for qPCR assay development, with the corresponding markers identified as V4V5-1 (V4V5 region) and V6-21 (V6 region). Candidate markers and their specificity, the probable source, and the sequence are shown in Tables S2 and S3.

Phylogenetic placement of sewer pipe-associated markers.

Near-full-length Bacteroides clones containing the matched V4V5 and/or the V6 markers identified by Indicspecies were used to construct a maximum likelihood tree in MEGA7 (75), based on Kimura 2 parameters (76) with gamma distribution and invariant sites (K2 + G + I) with bootstrapping for 1,000 replications.

Design of sewage-specific Bacteroides 16S rRNA gene fecal marker assays.

Primers and probes were designed based on 16S rRNA gene sequence alignment using animal fecal and sewage samples and visualized in MegAlign Pro program in DNASTAR software (version Lasergene 12). The marker sequences, a B. dorei 16S rRNA gene reference sequence (GenBank accession number AB242142) and sewage clone library sequences containing the V4V5-1 and V6-21 marker sequences, were included in the alignment. In addition, published near-full-length animal fecal Bacteroides clone sequences were included in the alignment of sequences from pigs (77), dogs (78), cows (36), chickens (79, 80), and mice (81) to discriminate from possible animal sources in the assay design. Primers and probes were named according to their base pair locations when aligned to an E. coli reference sequence (GenBank accession number J01859), with a comparison of universal 16S rRNA gene primers. Details are shown in Table 1. The amplicons of the two assays and their reference clone GenBank accession numbers are listed in Table S4.

qPCR experiments.

The qPCR reaction conditions, volumes, and methods for establishing the standard curve and testing inhibitions were followed as described by Templar et al. (24). Each run included a sewage-positive control and a no DNA control. The annealing temperatures were optimized by running a gradient qPCR for 1:100 (vol/vol) diluted sewage DNA (n = 4) from 59°C to 64°C. The same sewage samples were then tested in different dilution ratios at different annealing temperatures to make sure no amplification efficiency was lost (25). The amplification program included one cycle at 50°C for 2 min, followed by one cycle at 95°C for 10 min, and then 40 cycles of 95°C for 15 s followed by 1 min at 64°C for the BacV4V5-1 assay and 60°C for the BacV6-21 assay.

For assay validation to test for cross-reactivity, cat, dog, pig, cow, and deer fecal samples were tested in the formats of individual samples (i.e., from a single animal) and pooled samples (i.e., from two single animals of the same type). Pooled samples were retested individually unless there was insufficient material. Gull and chicken fecal samples were tested only as individuals. Each animal fecal sample was tested at DNA template concentrations of 1 ng · μl⁻¹, 0.1 ng · μl⁻¹, and 0.01 ng · μl⁻¹, and the animal qPCR results were converted to the units of copy number (CN) per 1 ng of input DNA, CN per 0.1 ng of input DNA, and CN per 0.01 ng of input DNA. For sewage samples, DNA templates were diluted 1:100 (vol/vol). For environment water samples, DNA templates were tested without dilution. All the sewage and environmental water results were expressed as CN/100 ml filtrated sample. Amplification after cycle 35 was considered negative for all samples. A subset of 40 samples, including sewage, animal feces, and environmental water samples, were tested for inhibition using salmon sperm DNA as the internal control as previously described (24). No inhibition was observed in these samples. Statistical analysis of qPCR assay correlations was performed using cor and cor.test functions in R. The qPCR assay slopes, y intercepts, r², and efficiency values are shown in Table S5.

Accession number(s).

The partial 16S rRNA gene sequences of the sewage clone libraries were deposited in the NCBI GenBank database. The library 1 sequences were deposited under accession numbers MH515295 to MH515584 and MH515940 to MH515981, and library 2 sequences were deposited under accession numbers MH515585 to MH515939 and MH515982 to MH516001. All V4V5 region NGS sequences of sewage and animal samples were from BioProjects PRJNA261344 (55) and PRJNA433408 (25). The V6 region NGS sequences of sewage and animal were from NCBI Sequence Read Archive (SRA) SRP041262 (34) and BioProject PRJNA433407 (25); V6 region NGS sequences of baseflow lake samples were from SRA SRP056973 (67), and the baseflow river sample sequences were deposited in NCBI SRA SRP168560.