Occurrence and fate of potential pathogenic bacteria as revealed by pyrosequencing in a full-scale membrane bioreactor treating restaurant wastewater

Abstract

One of the primary concerns on wastewater reuse is the presence of pathogenic bacteria. Considering that indicator bacteria might only offer limited information, we applied high-throughput pyrosequencing in this study to reveal bacterial pathogen diversity in a full-scale membrane bioreactor (MBR) treating restaurant wastewater. The results showed that fecal indicator bacteria could provide a rough estimation rather than an accurate characterization of the potential pathogenic bacteria in wastewaters, particularly from non-fecal sources. In general, MBR treatment showed good removal of potential pathogenic bacteria. The bacterial count of Arcobacter was decreased by nearly seven orders of magnitude, from (8.35 ± 0.87) × 10⁷ to <10 counts per mL, and Aeromonas, Enterobacter, Enterococcus, and Pseudomonas were not detected in the treated wastewater. The most dominant potential pathogens in activated sludge and treated wastewater were affiliated to the genera of Legionella, Clostridium and Mycobacterium. Species-specific comparison showed that only a small portion (0.0–1.6%) of the corresponding sequences had identities of > 99% to the neighbor pathogenic species, including Arcobacter butzleri and Arcobacter cryaerophilus. This study, therefore, provides insights into the occurrence and fate of potential bacterial pathogens in restaurant wastewater treatment and reclamation using MBRs.

---

1. Introduction

During the last decade, the catering industry has experienced an explosive growth in China and the business turnover doubled from 2006 (≈$161 billion per year) to 2011 (≈$332 billion per year).¹ Wastewater streams discharged from restaurants are generally characterized by high contents of oil and grease (O & G), suspended solid (SS) and detergent.^2,3 High O & G is a tremendous burden on the municipal wastewater systems because these organic substances usually tend to clump together, causing corrosion of drainage pipelines under anaerobic conditions. Therefore, appropriate treatment of restaurant wastewater is necessary in order to reduce the adverse impacts of its discharge.⁴ As an option, the membrane bioreactor (MBR) is a fascinating and promising technology, which presents distinctive advantages, such as high volumetric organic loading, small environmental footprint, and sound separation of emulsions that contain oil droplets with diameters less than 20 μm.⁵ MBRs also offer the opportunity to spare the expenditure of wastewater treatment since their superior effluent is more suitable for on-site reuse (e.g., for flushing toilets) in the restaurants.

On the condition that treated wastewater utilization is expected, contaminant removal should be sufficient to meet stringent regulatory standards, because of public health concerns.⁶ In MBRs, bacteria play an important role in the biochemical process, consuming nutrients and organic matter. To date, numerous studies have been conducted to improve the efficiency of the biochemical process, but most of them failed to attach importance to the potential hazard of the bacteria accordingly. One of the primary concerns about bacteria is the community of pathogenic bacteria originating from the excrement of disease-carrying humans and animals or other sources.⁷ In many public places (e.g., general merchandise stores and restaurants), the outbreak of gastroenteritis or other infections due to access of reused water could actually be masked by the background levels of assumed sources, such as food-borne and community-based infections.⁸ Microbial assessment of pathogenic bacteria in treated wastewater is thereby important in view of consequent health risks.

Historically, fecal indicator bacteria, including total and fecal coliforms and enterococci, have been widely used as a monitoring tool to predict the presence of potential bacterial, viral and protozoan pathogens.⁹ The major drawback of fecal indicator bacteria arises from their poor correlation with pathogens, especially those from non-fecal sources.^8,10 Moreover, membranes have a size-selective retention of different bacteria, and the abundance of pathogenic bacteria in the permeate could be underestimated or overestimated when referenced to certain indicator bacteria.¹¹ In recent years, real-time qPCR assays have been proposed and these assays are now used in many diagnostic and reference laboratories for the detection of pathogenic bacteria in clinical fluids.^8,12,13 Compared to indicator bacteria methods, the qPCR assay enables quantitative and highly specific detection, which could target the 16S ribosomal RNA, encoding genes or housekeeping genes of actual pathogens.¹² Nevertheless, the application of this technology is still hindered due to its limited throughput capacity. In environmental samples, bacterial pathogen diversity can be extremely high, as reflected by the more than thirty phylogenetic genera and thousands of strains. Clueless one-by-one detection is definitely time-consuming, and might also miss a potential infectious risk. Therefore, efficient solutions are urgently required for elucidating bacterial pathogen diversity in MBRs and assessing the full microbial risk of treated wastewater reuse in public places.

In this study, 454 high-throughput pyrosequencing was used to investigate the occurrence and fate of potential pathogenic bacteria in a full-scale MBR treating restaurant wastewater. Pyrosequencing is a high-throughput analytical method that generates a large number of DNA reads through a massive parallel sequencing-by-synthesis approach, and this technology can provide an adequate resolution of the microbial diversity of different environmental samples.^7,14–16 In the present work, 258 438 reads of the hypervariable V1∼V3 regions of the bacterial 16S rRNA gene were obtained. Sequence subsets with capacities of 10 000 and 100 were generated from the maternal gene libraries by a semi-random extraction method, and comprehensive comparison of these datasets was then carried out. Bacterial pathogen diversity was analyzed at genus level using the Ribosomal Database Project (RDP) Classifier.¹⁷ Alignment of the corresponding sequences to the known pathogen was further conducted by phylogenetic analysis.

2. Materials and methods

2.1. Sample collection and pyrosequencing

Sewage and sludge samples for pyrosequencing were taken from a full-scale MBR. The reactor, as schematically shown in Fig. S1 of the ESI,† was located in a general merchandise store (31.3°N 121.4°E) in Shanghai, China and has been in operation for over 6 years. The influent wastewater of the MBR includes (1) fresh food processing (FFP) wastewater, (2) restaurant wastewater generated from restaurants serving Chinese, Japanese and Western style food, (3) toilet flushing wastewater, (4) greywater from office regions and washing basins, and (5) car washing wastewater. The raw wastewater passed through screens, a dissolved air flotation tank and an aerobic MBR tank. The treated wastewater was temporarily stored in an effluent tank and finally reused for toilet flushing, lawn watering and car washing. The MBR tank had an effective volume of 60 m³. 600 poly(vinylidene fluoride) flat-sheet membrane modules (Zizheng Environm Technol Co. Ltd., Shanghai, China) with a mean pore size of 0.20 μm were installed in the tank. Details of the MBR setup, and characteristics of the influent and treated wastewater are summarized in the ESI (Section I, Fig. S1 and Table S1 in the ESI†).

Influent wastewater, activated sludge and treated wastewater samples, termed as A1, A2 and A3 samples, were taken from the inlet pipe, aerobic tank and outlet pipe of the MBR, respectively (see Fig. S1†). After DNA extraction and PCR amplification (see Section II of the ESI†), amplicons from A1, A2 and A3 were mixed at equal concentration, and the mixture was used for pyrosequencing on a Roche 454 FLX Titanium platform at Majorbio Bio-Pharm Technology Co., Ltd (Shanghai, China).

2.2. Read quality control and subset construction

After pyrosequencing, 258 438 raw reads (0.1 G) were obtained according to the unique match to the barcodes (Table 1). The results were deposited into the NCBI short reads archive database (accession number: SRA169387). To improve the validity of subsequent data processing, Qiime (version 1.17 http://qiime.org/) was applied to (1) check the completeness of the 3′ end of primers and adaptors; (2) remove reads containing an ambiguous base (‘N’) or homologous run that was longer than 10 nucleotides; (3) enable sliding window test of quality scores (-w 50 and -s 20); and (4) remove reads shorter than 200 bps.¹⁸ Barcodes and primers were also stripped from the resulting sequences, and finally pyrosequencing produced 24 962 (A1), 113 131 (A2) and 54 525 (A3) high-quality V1–V3 tags of the 16S rRNA gene with an average length of 462 bp (Table 1).

Table 1 Statistical summary for pyrosequencing and microbial diversity analysis

Sample ID	Raw reads	High-quality reads	Assigned reads^a	OTU	Chao	Shannon	Fo
a Assigned reads are the reads that match the OTU in each sample. Some high-quality reads may not match any OTU for these reasons: (1) the read is chimeric, and (2) the read that has a singleton sequence is discarded.b A1, A2 and A3 represent pyrosequencing results of influent wastewater, activated sludge and treated wastewater samples; B1, B2 and B3 represent the subsets with 10 000 reads extracted from A1, A2 and A3; C1, C2 and C3 represent the subsets with 100 reads extracted from B1, B2 and B3.c n.a. indicates the value is not available.
A1^b	34 949	24 962	19 411	897	1323	3.70	0.92
B1	n.a.^c	10 000	8162	864	1886	3.88	0.89
C1	n.a.	100	89	40	164	2.90	0.64
A2	158 938	113 131	63 243	1712	2026	4.85	0.90
B2	n.a.	10 000	6163	1132	2664	5.04	0.81
C2	n.a.	100	63	46	187	3.65	0.42
A3	64 551	54 525	36 644	1362	1670	5.20	0.89
B3	n.a.	10 000	7132	1063	2092	5.34	0.83
C3	n.a.	100	75	59	190	3.95	0.37

---

For a comprehensive understanding of the impacts of sequencing depth, subsets with capacities of 10 000 and 100 were generated from the high-quality maternal sets of A1∼A3 by a semi-random extraction method. Initially, the sub.sample command of the MOTHUR program (http://www.mothur.org/wiki/Sub.sample) was used for A1∼A3 to create 30 subsets comprised of 10 000 sequences, i.e., B1_i = {x_j | x_j ∈ A1, j = 1–1000} (i = 1–10), B2_i = {x_j | x_j ∈ A2, j = 1–1000} (i = 1–10) and B3_i = {x_j | x_j ∈ A3, j = 1–1000} (i = 1–10). Principal coordinates analysis (PCoA) with the Bray–Curtis index (R package, http://www.r-project.org/) was then performed to evaluate the relationship between A1∼A3 and B1_i∼B3_i (i = 1–10), and the subsets with the highest homology were retained and specified as B1∼B3 (Fig. S2 in the ESI†). Afterwards, a similar procedure was applied to create 30 subsets containing 100 sequences from B1∼B3, and the subsets with the highest homology with B1∼B3 were specified as C1∼C3. Despite the debate that semi-random extraction is reliable enough compared to independent sequencing, this method is similar to pyrosequencing run in reverse; and in practice the final gene libraries (e.g., A1∼A3) can be obtained based on the deficient datasets (e.g., B1∼B3) by further sequencing of the amplicons. Nevertheless, evaluation of pathogenic bacteria diversity was mainly based on the original pyrosequencing results.

2.3. Phylogenetic classification and biodiversity analysis

Clustering of the high-quality reads into operational taxonomic units (OTUs) was performed using UPARSE pipeline (vsesion 7.1, http://drive5.com/uparse/).¹⁹ Briefly, abundance-sorted reads of the nine datasets (A1∼A3, B1∼B3 and C1∼C3) were clustered by setting a minimum identity of 97%, and the uchime_ref command was used to filter out chimeras. The abundances of OTUs in each dataset were obtained by searching the reads as a query set against the OTU representative sequences. For the cluster files, alpha-diversity and rarefaction curves were generated in MOTHUR for each sample (version v.1.30.1, http://www.mothur.org). Functional organization indices (F_o) were calculated according to the standard method reported by Marzorati et al.²⁰ Representative sequences from each OTU were assigned down to the phylum and genus level using the RDP Classifier with a set confidence threshold of 80% (https://rdp.cme.msu.edu/classifier/classifier.jsp, 16S rRNA training set 10).¹⁷

Venn diagrams with shared and unique OTUs were utilized to depict the similarity and difference between microbial communities. A pairwise statistical comparison of taxonomy at phylum level between maternal sets and subsets was carried out using STAMP.²¹ Biological relevance between samples at the genus level was evaluated using linear regression of SigmaPlot software (version 12.5, Systat Software, Inc., U.S.). Furthermore, the LDA Effect Size (LEfSe) algorithm was introduced herein to identify taxa that characterize the differences among the three environmental samples.²² A1∼A3, B1∼B3 and C1∼C3 were grouped according to the source (e.g., influent wastewater, activated sludge or treated wastewater sample), and each sample was first normalized to the sum of the values of 0.05 M. The parameters for data processing were set as follows: ‘alpha value for the factorial Kruskal–Wallis test among classes’ = 0.05, ‘threshold on the logarithmic LDA score for discriminative features’ = 2.7, and ‘set the strategy for multi-class analysis’ = all-against-all.

Alignment of microbial communities to pathogenic genera was first evaluated using the taxonomic results of the RDP Classifier. The lists of known pathogenic genera summarized by Ye and Zhang and Bibby et al. were used as in ref. 7 and 23. Representative sequences from OTUs that were assigned as Arcobacter, Clostridium, Legionella, and Mycobacterium were further separated for phylogenetic analysis at the species level. The 16S rRNA gene of known pathogens and non-pathogens from the four genera were achieved from the NCBI Genbank (Table S2 of the ESI†), and merged with the corresponding sequences of this study into a fasta file. ClustalW was used for aligning and bootstrapping of the phylogenetic tree, which was then viewed, edited and published with MEGA 6.²⁴ Default settings were used. Furthermore, bacteria assigned to the families of Enterobacteriaceae and Enterococcaceae were regarded as the representative fecal indicators in this study.

2.4. Quantification of bacterial biomass using flow cytometry (FCM)

Bacteria biomass in wastewater and sludge samples was quantified using a flow cytometer. Initially, influent wastewater, activated sludge and treated wastewater (A1, A2 and A3) were diluted to 1 : 20 (v/v), 1 : 500 (v/v) and 1 : 1 (v/v) using 0.22 μm filtered phosphate-buffered-saline solution (0.84, pH = 7) to achieve optimal concentrations of bacteria for FCM analysis. Then the diluted mixtures (A1, A2 and A3) were subjected to ultrasonication treatment at power densities of 25, 80 and 0 kJ L⁻¹, respectively. After filtration with 10 μm filters, samples were stained with SYBR Green I at a ratio of 100 : 1, incubated for 15 min in the dark at room temperature and finally processed by the flow cytometer (BD Accuri™ C6, U.S.). Each sample was tested in triplicate and the total bacterial counts of A1, A2 and A3 were (2.31 ± 0.24) × 10⁸, (7.06 ± 0.30) × 10⁹ and (3.35 ± 0.82) × 10⁴ counts per mL, respectively.

3. Results

3.1. Diversity and similarity analysis of microbial communities

By performing the alignment at an α of 0.03 using UPARSE pipeline, 897, 1712 and 1362 OTUs were obtained from A1, A2 and A3 (Table 1). At a degraded and uniform library size of 10 000, the Chao1 richness estimators of the three samples were 1886, 2664 and 2092, and the Shannon diversity indices were 3.88, 5.04 and 5.34, respectively. Alpha-diversity analysis suggested that the bacterial community from the influent wastewater sample had the lowest microbial richness and diversity. Moreover, the F_o values of the three samples were 0.89–0.92. It could be inferred that all the microbial communities were highly functionally organized.^14,20 Pairwise comparison using Venn analysis showed that the similarity of A2–A3 was the highest, followed by that of A1–A2 and that of A1–A3 (Fig. 1). Notably, A2 and A3 had 707 shared OTUs that contained 65.0% and 77.8% of the reads, respectively. In contrast, only very few reads (0.5% and 0.1%) were classified into the OTUs that were shared by A1 and A3, which indicated that MBR treatment introduced a profound influence on the structure of the microbial community in wastewater.

Fig. 1 Similarity analysis of the microbial communities (A1, A2 and A3) based on the clustering results at 3% distance cutoff. The numbers in the black circles represent the number of OTUs that are present in the core OTUs shared by the three samples. The numbers in the blue circles represent the OTUs shared by two samples. The numbers in the red circles represent the unique OTUs observed in only one sample. Percentages listed beside the branches indicate the percentages of reads of each sample assigned into the nearby OTUs groups.

3.2. Impacts of sequencing depth

Generally, the microbial communities of environmental samples are highly diverse, and in this study rarefaction curves showed that new bacterial phylotypes continued to emerge even after 60 000 reads were sampled (Fig. S3 in the ESI†). Addressing an appropriate sequencing depth is crucial for high-throughput pyrosequencing to detect pathogenic bacteria at low abundance; an increased depth significantly increases the sequencing and processing cost, while a small library size could only provide insufficient resolution. Since 10 000 and 100 library sizes are always considered in pyrosequencing and conventional molecular biology studies, sequence subsets with corresponding capacities were generated from the maternal gene libraries by the semi-random extraction method.

Table 1 indicates that insufficient resolution reduced the accuracy of alpha-diversity analysis. For instance, at the sequencing depth of 100, only 40, 46 and 59 OTUs were predicted for the whole microbial communities in influent wastewater, activated sludge and treated wastewater, respectively. Furthermore, we compared the taxonomic results of the nine datasets at phylum and genus levels (Fig. 2). In total, 25 phyla were classified at the threshold of 80%. Proteobacteria was the most dominant phylum, accounting for 53.1–68.3% of the total communities, respectively. Pairwise comparison using STAMP shows that there is no significant dissimilarity of taxonomic results between Aj and Cj (j = 1, 2, 3) at phylum level (Fig. 2a, c and e). However, the reliability was significantly reduced at the genus level. Most taxa of A1∼A3 could not be predicted by the taxonomic results of C1∼C3 in a 95% predication band. In contrast, linear regression showed that except for a few categories, B1∼B3 supplied a credible characterization of the microbial communities of A1∼A3 at the genus level (Fig. 2b, d and f). The results suggested that compared to low-throughput sequencing methods, pyrosequencing could provide a more valid estimation of the population structure of diverse communities, especially at terminal taxonomic levels (e.g., genus level). Increasing the library size from 10 000 to 100 000, however, did not improve the taxonomic results as expected, probably due to the abundance of singletons at high sequencing depth, which were always discarded after de-noising (Table 1).

Fig. 2 Pairwise comparison of biological relevance of (a) A1 and C1 at phylum level, (b) A1, B1 and C1 at genus level, (c) A2 and C2 at phylum level, (d) A2, B2 and C2 at genus level, (e) A3 and C3 at phylum and (f) A3, B3 and C3 at genus level. Taxonomic results based on OTU clustering at a 3% distance were compared using STAMP at phylum level. A corrected P-value lower than 0.05 is significant. Correlations of assignment results in each of the three samples were carried out at genus level. The horizontal and vertical axes in each subfigure (b, d and f) indicate the numbers of the corresponding genus sequences. The red lines represent the 95% prediction bands of linear regression.

3.3. Detection and characterization of the potential pathogenic bacteria

During MBR treatment, the structure of the microbial community in the wastewater changed in response to the environmental selective pressures, and taxa were differently enriched in different samples (Fig. S4 of the ESI†). Fig. 3, according to the alignment to the lists of known pathogenic genera,^7,23 shows the eleven genera of potential pathogenic bacteria found in the three samples. It could be noticed that only Arcobacter and Clostridium were ubiquitous in all the samples. In A1, Arcobacter was the most abundant genera, accounting for about 40% of the population. Except for Clostridium, the other potential pathogens, including Aeromonas, Enterobacter, Enterococcus and Treponema, were present at very low abundances (0.026–0.031%). The number of sequences assigned to potential pathogenic bacteria was significantly decreased in the activated sludge sample (Table S3 of the ESI†). For example, Arcobacter were underrepresented in A2, with a nearly two orders of magnitude difference in abundance compared to that found in A1. A similar decay was found for Aeromonas, Enterobacter, Enterococcus and Treponema, at even lower abundances. Instead, a well-known source of infection, Legionella,²⁵ became abundant among the potential pathogens. Furthermore, membrane retention induced a selective pressure on bacterial pathogen diversity. Several widely-reported pathogenic bacteria, including Aeromonas, Enterobacter, Enterococcus, and Pseudomonas,^7,8,23,26 were not found in A3, and the sequences assigned to potential pathogenic genera only accounted for 0.3% of the total population (Table S3†). However, a gram-positive genus, Clostridium, was found to be the most abundant among the potential pathogenic phylotypes. Since no significant difference was noted by LEfSe analysis for this taxon (Fig. S4†), it was possible to infer that Clostridium were more resistant to the treatment of the MBR.

Fig. 3 Relative abundances of potential pathogenic genera in influent wastewater (A1), activated sludge (A2), treated wastewater (A3), subsets with 10 000 reads (B1, B2 and B3) and subsets with 100 reads (C1, C2 and C3). Relative abundance is defined as the percentage of a pathogenic genus in the total population. The different colors represent the percentages of sequences in the corresponding confidence ranges.

In this study, the depth of pyrosequencing also had a significant influence on the detection and characterization of potential pathogenic bacteria. For B1∼B3, the dominant pathogenic genera could be well identified, while those with low abundances (e.g., Enterobacter, Enterococcus and Vibrio) were neglected (Fig. 3). It is worth noting that sequencing failed to reveal the majority of potential pathognic bacteria in environmental samples when the library size was reduced to 100. Specifically, no pathogenic bacteria were detected in C2 and Mycobacterium were obviously overestimated in C3 (Fig. 3).

Overall, the corresponding sequences had good alignment with the potential pathogenic genera by using the RDP Classifier, and most had a bootstrap confidence over 80% (Fig. 3 and Table S3†). Since species-specific comparison with known pathogenic bacteria could give a more accurate estimation of the potential pathogens in the samples, representative sequences of concerned OTUs assigned into Arcobacter, Clostridium, Legionella and Mycobacterium genera (Fig. 3) were retrieved from the datasets. Phylogenetic analysis was then conducted by building a library with representative 16S rRNA gene sequences of pathogenic and non-pathogenic bacterial species. As shown in Fig. 4, 15 OTUs out of the total (37 OTUs) had an identity of over 95% with neighbor pathogens, including 8 OTUs assigned into Arcobacter, 3 OTUs into Clostridium and 4 OTUs into Mycobacterium (Table S4 of the ESI†). In the Arcobacter genus, the most abundant taxon OTU2964, accounting for 32.9% of the population in A1, showed a low alignment with known pathogens. Without additional information, it could not be concluded whether these sequences referred to nonpathogenic strains, because variants might be also associated with disease but not yet identified.²³ Of particular importance is that only two strains (OTU2202 and OTU2091) were recognized as potential pathogenic species by species-specific comparison. It could be deduced that phylogenetic analysis at the genus level might lead to an overestimation of the pathogenic bacteria in environmental samples.

Since fecal indicator bacteria are still widely used to predict the presence of bacterial, viral and protozoan pathogens,⁹ the abundances of typical indicators were further evaluated based on the taxonomic results in the present work. Because there is no full taxonomic definition of fecal indicators yet, bacteria assigned to the families of Enterobacteriaceae and Enterococcaceae were regarded as the representative fecal indicators herein. As shown in Table 2, 41 sequences from A1 were classified, which contributed 0.21% of the dataset. In A2, only 10 sequences got a valid match from the RDP Classifier, including 6 sequences assigned into Enterobacteriaceae and 4 sequences into Enterococcaceae. Probably due to the sound separation of the 0.20 μm poly(vinylidene fluoride) membranes, no Enterobacteriaceae or Enterococcaceae were detected in the treated wastewater (A3) in this study.

Table 2 Summary of sequences assigned to Enterobacteriaceae and Enterococcaceae^a

	A1		A2		A3
	Number of sequences	r^b, %	Number of sequences	r, %	Number of sequences	r, %
a based on the taxonomic results of the RDP Classifier.b r indicates the relative abundance of sequences.
Enterobacteriaceae	36	0.185	6	0.009	0	0
Enterococcaceae	5	0.026	4	0.006	0	0
Total	41	0.211	10	0.015	0	0

---

4. Discussion

In this study, a group of predominant potential pathogens, Arcobacter, were differently abundant in the influent wastewater compared to other samples (Fig. 3 and S4†). The genus Arcobacter, belonging to the RNA Superfamily VI of Proteobacteria, was proposed in 1991, and the International Commission on Microbiological Specification for Foods has considered Arcobacter to be one of the most frequently notified food-borne infectious agents.²⁷ Full understanding of its occurrence and fate during wastewater reclamation is, thereby, very important, especially for a rapid and accurate diagnosis of the infection source of outbreaks (e.g., acute enteric disease) in public places. In the present work, the RDP Classifier indicated that 21 OTUs from A1 were classified into Arcobacter genus, which accounted for 36.5% of the total population. The microbial composition might be a typical pathogenic characteristic of restaurant wastewater, because raw and undercooked meat and poultry products have been recognized as sources of Arcobacter.²⁸ Overall, the results showed that the hybrid MBR system presented a good removal of Arcobacter; ∼87% of the influent Arcobacter were eliminated in the activated sludge and <10 counts per mL were detected in the treated wastewater (Section III of the ESI†). Furthermore, phylogenic analysis suggested that the genus-specific comparison could result in an overestimation of pathogenic bacteria, since more than 90% of Arcobacter had highest homology with a free-living nitrogen-fixing bacterium, A. nitrofigilis. Notably, 2 strains (OTU2091 and OTU2202) had an identity over 99% with known pathogenic Arcobacter species (e.g., A. butzleri), which were estimated to be at concentrations of (1.40 ± 0.15) × 10⁶, (6.81 ± 0.29) × 10⁶ and ∼1 counts per mL in influent wastewater, activated sludge and treated wastewater, respectively (Section III of the ESI†).

From the view of pathology, it is of great concern to focus on the pathogenic bacteria emerging in the aerobic tank of the MBR because not only do these tolerant microorganisms have a competitive advantage with the biomarkers involved in contaminant degradation (e.g., Zoogloea and Dechloromonas, as shown in Fig. S4†) but also aerosols containing pathogens could be generated from the aeration tank and further transported and dispersed by wind. A genus of gram-negative coccobacilli, Legionella, was well recognized in the activated sludge sample (A2). It has been reported that Legionella prefer to inhabit man-made aquatic environments where the water temperature is higher than ambient temperature, and that the growth of Legionella spp. can be aided by co-existing micro-organisms (e.g., protozoa).^25,29 Although this bacterial genus was enriched in activated sludge, the Legionella found herein seemed non-pathogenic; all the representative sequences had a low alignment (87–92%) with the foremost pathogenic species L. pneumophila, L. longbeachae, L. micdadei and L. bozemanii. Fig. 3 indicates that MBR removal of Legionella from restaurant wastewater was mainly attributed to membrane retention, which could efficiently eliminate the hosts (e.g., amoebae) in the treated wastewater. Moreover, a recent study on bacterial pathogen diversity in biosolids (digested sludge) using pyrosequencing has revealed that most of the pathogenic sequences belonged to the genera of Mycobacterium and Clostridium.²³ In the present work, our results showed that despite low relative abundances, all sequences belonging to the Mycobacterium genus had more than 95% similarity to a ‘freak’ pathogenic species, M. abscessus. The gene order phylogeny of M. abscessus groups the organisms into rapid and slow-growers.^30,31 M. abscessus is closer to the non-pathogens in terms of its growth characteristics and is placed away from the pathogens (Fig. 4), which could lead to taxonomic bias based on 16S rRNA gene pyrosequencing. As a result, virulence assays that target the functional genes are further required to support the relevant conclusions.

Fig. 4 Phylogenic tree of concerned OTUs from A1∼A3 that were assigned into potential pathogenic genera (Arcobacter, Clostridium, Legionella and Mycobacterium). Representative sequences from these OTUs were reachieved for alignment and phylogenic analysis. The number of the OTU (e.g., OTU448) only indicates the logical order in OTU clustering. OTUs with satisified identities to neighbor pathogens are bolded and marked with * (95–99%) or ** (>99%). Bootstrap values are calculated by 1000 repetitions, and values >50% are given. indicate the non-pathogenic species, the pathogenic species and the vague species.

This study also reinvigorates the debate that the indicator bacteria are inefficient in representing the potential pathogenic bacteria from non-fecal sources. The relative abundance of fecal indicator bacteria did not show a good relationship with that of potential pathogenic species, though provided a rough evaluation on the occurrence of potential pathogens in the restaurant wastewater (Tables 2 and S4†). Notably, Enterobacteriaceae and Enterococcaceae were not detected in A3 but 97 of the total 36 644 sequences were classified into the pathogenic genera, including 25 sequences with identities of >95% with M. abscessus, C. difficile and C. botulinum. Furthermore, the Clostridium genus was an important group in A3, which was resistant to the MBR treatment (Fig. 3). In this study, the ambiguously defined taxon contained Clostridium cluster sensu stricto (Clostridiaceae 1), Clostridium cluster IV (Ruminococcaceae) and Clostridium cluster XI (Peptostreptococcaceae). Three strains from Clostridium cluster XI and Clostridium cluster sensu stricto had a good phylogenetic alignment with C. difficile and C. botulinum, respectively (Fig. 4 and Table S4†). Since Clostridia (spores) are highly resistant to chlorination,³² disinfection efficiency could be easily overestimated when referenced to the elimination of intolerant indicator bacteria (e.g., Enterobacteriaceae).

In this study, 454 pyrosequencing was introduced for a comprehensive understanding of bacterial pathogens in restaurant wastewater. Compared to conventional culture-based methods and qPCR assays, this technology is high-throughput for mining potential pathogenic bacteria in environmental samples, which avoids the mis-estimation of pathogens by using a certain group of indicator bacteria. Molecular biology methods that offer ≈100 tags could only provide rough information on the structure of microbial communities at the phylum level (Fig. 2). By contrast, the 10 000-sequence datasets were generally valid in forecasting individuals within microbial communities, but overrepresentation and underrepresentation were still noted regarding the highly likely bacterial pathogens (Table S4†). For accurately reaping the rare strains, exponential growth of the library size (1–2 orders of magnitude) might be unwise since a large number of singletons were generated at a sequencing depth of 30 000–150 000 (Table 1). DNA fragment pretreatment (e.g., the use of multiple genus level PCR primers) should therefore be considered in pathogenic studies. Furthermore, 16S rRNA gene pyrosequencing provided the opportunity to discover the important strains that have not been cultured yet (e.g., OTU2964). Short-gun metagenomic and metatranscriptomic sequencing could be used to predict their functions. Phylogenic analysis of the concerned pathogenic and non-pathogenic bacterial species herein gave a more accurate evaluation of the abundance and diversity of bacterial pathogens. Virulence of relevant communities could be further analyzed using qPCR or microarrays that target the functional genes. Overall, the present work showed that restaurant wastewater is suitable for reclamation using MBR technology. Pathogenic bacteria were efficiently removed in the hybrid systems, and the membrane filtration process retained the communities that were resistant to biological treatment. The tolerant bacterial pathogens in treated wastewater revealed by pyrosequencing provide insights into the selection of specific tertiary treatment and also proper disinfection methods.

5. Conclusions

In the present work, high-throughput pyrosequencing was used to characterize the potential pathogenic bacteria in a full-scale MBR treating restaurant wastewater. The results indicated that the influent pathogenic community might be highly diverse and that 39.2% of the population was assigned to pathogenic genera. Overall, MBR treatment showed good removal of Aeromonas, Arcobacter, Enterobacter, Enterococcus and Treponema, and in the treated wastewater the bacterial count of Arcobacter was decreased to <10 counts per mL. The most dominant potential pathogens in activated sludge and treated wastewater were affiliated to the genera of Legionella, Clostridium and Mycobacterium. Nevertheless, species-specific comparison showed that only a small portion (0.0–1.6%) of the corresponding sequences had identities of >99% to the neighbor pathogenic species, suggesting that phylogenetic analysis at the genus level might lead to an overestimation of the potential pathogens. This study provided insights into assessing pathogenic bacteria risk in wastewater purification and reclamation.

Acknowledgements

The work is financially supported by National Natural Science Foundation of China (51422811), Shanghai Rising-Star Program (14QA1403800) and the Shanghai Science & Technology Commission Program (13231202002).

References

1. Chinese catering information center, http://www.canyin.com/a/zixunzhongxin/xingyexinwen/20120724/991.html (accessed April 18, 2014).
2. X. Zhu, Z. Wang and Z. Wu, Process Biochem., 2011, 46, 1001–1009.
3. B. Yang, G. Chen and G. Chen, Sep. Purif. Technol., 2012, 88, 184–190.
4. J.-x. Kang, L. Lu, W. Zhan, B. Li, D.-S. Li, Y.-Z. Ren and D.-q. Liu, J. Hazard. Mater., 2011, 186, 849–854.
5. S. Judd, Trends Biotechnol., 2008, 26, 109–116.
6. M. E. Verbyla, S. M. Oakley and J. R. Mihelcic, Environ. Sci. Technol., 2013, 47, 3598–3605.
7. L. Ye and T. Zhang, Environ. Sci. Technol., 2011, 45, 7173–7179.
8. W. Ahmed, H. Brandes, P. Gyawali, J. P. S. Sidhu and S. Toze, Water Res., 2014, 53, 361–369.
9. O. Savichtcheva and S. Okabe, Water Res., 2006, 40, 2463–2476.
10. W. Ahmed, S. Sawant, F. Huygens, A. Goonetilleke and T. Gardner, Water Res., 2009, 43, 4918–4928.
11. K. Zhang and K. Farahbakhsh, Water Res., 2007, 41, 2816–2824.
12. P. Bourhy, S. Bremont, F. Zinini, C. Giry and M. Picardeau, J. Clin. Microbiol., 2011, 49, 2154–2160.
13. D.-Y. Lee, H. Lauder, H. Cruwys, P. Falletta and L. A. Beaudette, Sci. Total Environ., 2008, 398, 203–211.
14. J. Ma, Z. Wang, Y. Yang, X. Mei and Z. Wu, Water Res., 2013, 47, 859–869.
15. T. Zhang, M. F. Shao and L. Ye, ISME J., 2012, 6, 1137–1147.
16. A. S. Laufer, J. P. Metlay, J. F. Gent, K. P. Fennie, Y. Kong and M. M. Pettigrew, mBio, 2011, 2, e00245.
17. Q. Wang, G. M. Garrity, J. M. Tiedje and J. R. Cole, Appl. Environ. Microbiol., 2007, 73, 5261–5267.
18. J. G. Caporaso, J. Kuczynski, J. Stombaugh, K. Bittinger, F. D. Bushman, E. K. Costello, N. Fierer, A. G. Pena, J. K. Goodrich, J. I. Gordon, G. A. Huttley, S. T. Kelley, D. Knights, J. E. Koenig, R. E. Ley, C. A. Lozupone, D. McDonald, B. D. Muegge, M. Pirrung, J. Reeder, J. R. Sevinsky, P. J. Tumbaugh, W. A. Walters, J. Widmann, T. Yatsunenko, J. Zaneveld and R. Knight, Nat. Methods, 2010, 7, 335–336.
19. R. C. Edgar, Nat. Methods, 2013, 10, 996–998.
20. M. Marzorati, L. Wittebolle, N. Boon, D. Daffonchio and W. Verstraete, Environ. Microbiol., 2008, 10, 1571–1581.
21. D. H. Parks and R. G. Beiko, Bioinformatics, 2010, 26, 715–721.
22. N. Segata, J. Izard, L. Waldron, D. Gevers, L. Miropolsky, W. S. Garrett and C. Huttenhower, Genome Biol., 2011, 12, R60.
23. K. Bibby, E. Viau and J. Peccia, Water Res., 2010, 44, 4252–4260.
24. K. Tamura, G. Stecher, D. Peterson, A. Filipski and S. Kumar, Mol. Biol. Evol., 2013, 30, 2725–2729.
25. J. M. Blatny, B. A. P. Reif, G. Skogan, O. Andreassen, E. A. Hoiby, E. Ask, V. Waagen, D. Aanonsen, I. S. Aaberge and D. A. Caugant, Environ. Sci. Technol., 2008, 42, 7360–7367.
26. H. de Man, M. Bouwknegt, E. van Heijnsbergen, E. J. T. M. Leenen, F. van Knapen and A. M. de Roda Husman, Water Res., 2014, 54, 254–261.
27. I. Gonzalez, T. Garcia, S. Fernandez and R. Martin, Food Anal. Methods, 2012, 5, 956–968.
28. M. Mor-Mur and J. Yuste, Food Bioprocess Technol., 2010, 3, 24–35.
29. H. Y. Lau and N. J. Ashbolt, J. Appl. Microbiol., 2009, 107, 368–378.
30. H. Medjahed, J.-L. Gaillard and J.-M. Reyrat, Trends Microbiol., 2010, 18, 117–123.
31. A. N. Prasanna and S. Mehra, PLoS One, 2013, 8, e71248.
32. T. Karpova, P. Pekonen, R. Gramstad, U. Ojstedt, S. Laborda, H. Heinonen-Tanski, A. Chavez and B. Jimenez, Water Sci. Technol., 2013, 68, 2090–2096.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c4ra10220g

---