Microbial composition and seasonal dynamics at domain and phylum levels
The seasonality of microbial communities in different ecosystems like ocean (Barberán et al., 2011), lake (Pajdak-Stos and Fialkowska, 2012) and soil (Venter et al., 2004) have been well documented, with little attention being paid to the AS system (Kim et al., 2013), in which the seasonal dynamics of microbial communities greatly affects the performance and stability of pollutants removal.
As shown in Table 1, microbial communities in AS were predominated by Bacteria, with abundances between 85.6% and 93.0%, followed by Eukaryota (0.73–7.3%) and Archaea (0.11–0.40%). Paired t-test indicated that there was a significant (P-value < 0.05) difference in domain distribution between winter and summer samples. The abundance of Bacteria in summer was 87.8 (± 2.8)%, which was lower than that in winter [91.0 (± 1.6)%]. Similar results were observed for Archaea, with winter abundance doubling that of summer. Differently, the average abundance of Eukaryota in summer [5.0 ± (2.3)%] was much higher than that in winter [1.92 ± (0.90)%]. Previous studies have widely demonstrated that bacteria-eating Eukaryota, like Rotifera and Nematoda, in AS could reduce biomass production via predation of the microorganisms, improving the performance of WWTPs (Yiannakopoulou et al., 2009). As most of the Eukaryota found in Sha Tin AS were affiliated with Rotifera and Nematoda, the great variation of Bacteria and Eukaryota during summer and winter could be associated with significant change of quantities of predator and prey in the AS system, which could affect the performance of WWTPs.
Table 1. Domain distribution of the sequences in the eight data sets derived from AS samples
| ||Summer abundance (%a)||Winter abundance (%a)||P-valuesb|
At phylum level, a total of 40 phyla were found in all eight samples, and 22 of them were identified as major phyla (top 15 in each sample). Among the major phyla, 16 were from Bacteria, 5 from Eukaryota and 1 from Archaea (Fig. 1). Similar to previous findings on AS using cloning (Snaidr et al., 1997), microarray (Xia et al., 2010) and 454 pyrosequencing (Zhang et al., 2011), metagenomic data sets in this study found that Proteobacteria was the most abundant phylum (38–43%, averaging at 40.8%) in all 4-year AS samples, and the subdominant phyla were Actinobacteria (averaging at 21.9%), Bacteroidetes (10.6%), Chloroflexi (8.7%) and Firmicutes (4.8%) (Fig. 1, Table S3). Notably, most of the Eukaryotal populations belonged to Rotifera (1.3%) and Nematoda (0.7%), and their occurrence in the AS is beneficial to the reduction of the biomass production by predation of the microorganisms.
Figure 1. The relative abundances of major phyla in the eight Sha Tin AS samples. The number above each bar represents the abundance ratio of each phylum in summer to winter samples. The phylum names in black, red and green represent phyla that were affiliated to Bacteria, Eukaryota and Archaea, respectively.
Download figure to PowerPoint
Seasonal variation patterns were observed at phylum level, as shown in Fig. 1. On the one hand, abundance of Actinobacteria (phylum) was usually higher in winter 26.0 ± (3.1)% than in summer 17.9 ± (6.9)% (Table S3), with abundance ratio of summer to winter (P1-ratio) of 0.7. Similar seasonal dynamics was also observed for two other bacterial phyla, that is Verrucomicrobia and Thermotogae (with P1-ratios of 0.6 and 0.1), two eukaryotic phyla, namely Arthropoda and Gastrotricha (with P1-ratios of 0.5 and 0.5), and one archaeal phylum Euryarchaeota (with P1-ratio of 0.6), although their abundances were relatively lower (Table S3). On the other hand, another four bacterial phyla, that is Nitrospirae, Cyanobacteria, Acidobacteria and Spirochaetes (with P1-ratios of 3.3, 2.4, 2.0 and 2.0), and three eukaryotic phyla, namely Rotifera, Nematoda and Streptophyta (with P1-ratios of 4.9, 7.5 and 3.6), showed an opposite variation dynamics, with abundances in summer 1.0–6.5 times higher than those in winter.
Grouping of the eight AS samples
The similarity patterns of the eight AS samples were evaluated at class and family levels through two independent methods: principal coordinate analysis (PCoA) and cluster analysis (CA), based on Bray–Curtis distance. We hypothesized that temperature exerts the most significant effect on the differentials in microbial structures between summer and winter samples.
As shown in Fig. 2A and B, PCoA bases on abundances of classes (A) and families (B) revealed that the microbial communities in the eight AS samples could be clustered into four subgroups, i.e. Group I: samples collected in summer of 2007 and 2008 (AS07-7 and AS 08-7); Group II: samples collected in winter of 2008 and 2009 (AS08-1 and AS09-1); Group III: samples collected in summer of 2009 and 2010 (AS09-7 and AS10-7); and Group IV: samples collected in winter of 2010 and 2011 (AS10-1 and AS11-1). CA of the eight AS samples also showed the same seasonal grouping patterns using benchmarks of 0.87 and 0.78 (as indicated by the blue dotted lines in Fig. 2C and D). As demonstrated by the CA and PCoA, sludge samples collected in summer or winter were certainly similar to each other, possibly due to the seasonal temperature difference, as well as other environmental or operational parameters, such as salinity and sludge retention time.
Figure 2. Principal coordinate analysis (PCoA) and cluster analysis (CA) of eight activated sludge (AS) samples at class and family levels. PCoA (A and B) was conducted using the Bray–Curtis distance and a transformation exponent of 2 (as recommended). Similar grouping patterns were adopted as used in the cluster analysis. CA (C and D) was performed using unweighted pair group mean averages as algorithm and the Bray–Curtis distance for similarity measurement. The blue and red dotted lines show the similarity cut-off levels to cluster the eight Sha Tin AS samples.
Download figure to PowerPoint
Apart from the above seasonal patterns, the eight AS samples were divided into two clusters using a benchmark of 0.81 for class level and 0.73 for family level respectively (as indicated by the red dotted lines in Fig. 2C and D). Cluster I was composed of four samples (Groups I and II) collected in the first 2 years; and Cluster II comprised four samples (Groups III and IV) collected in the last 2 years, revealing the existence of dual grouping patterns that were beyond reasonable explanation merely by seasonality (summer and winter). Similarly, PCoA plots (Fig. 2A and B) also supported the existence of another grouping pattern beyond summer and winter, considering the fact that samples collected in the last 2 years (Groups III and IV) tended to cluster closer to each other (than to samples in Group I or Group II) regardless of their opposite seasonal characteristics as summer and winter samples. Noteworthy, the grouping patterns of AS samples displayed by CA and PCoA were quite similar to the mixed variation patterns of chemical oxygen (COD), total Kjeldahl nitrogen (TKN-N) and total phosphate (TP) concentrations (Table S1, Appendix S1), possibly revealing an intrinsic correlation between the AS microbial structure and the available nutrients for microbial growth.
Seasonal genus dynamics and genus–environment relationship
Comparative analysis revealed the core and distinct genera harboured in the AS of Sha Tin WWTP over a period of 4 years. As shown in Table S5, 100 out of the total 643 assigned genera were shared by all eight AS samples, occupying 79.4% of the classified sequences at genus level, while 202 rare genera that only appeared in one sample accounted for merely 1.2% of total assigned sequences, indicating the existence of considerable amount of rare species in AS.
The major genera (top 20 in each sample) were selected (a total of 43 genera for all eight samples) and compared with their abundances in other samples, as shown in Fig. S2 and Table S4. Judging from the averaged abundance of each genus in all samples, Mycobacterium (8.9 ± 2.4%) and Nitrospira (6.1 ± 4.2%) were found as the two most abundant genera in AS of Sha Tin WWTP, followed by Planctomyces (5.4 ± 2.2%) and Caldilinea (4.3 ± 1.1%). The other abundant genera (> 1.0%) included one genus of phosphate-accumulating organisms (PAO), Tetrasphaera (2.4 ± 2.0%); one hydrolyser-related genus, Lewinella (1.7 ± 0.7%); one marine nematode, Diplolaimella (2.3 ± 3.3%); two denitrifying bacteria, Rhodobacter (2.1 ± 0.5%) and Paracoccus (1.4 ± 0.6%); four fermentative or photo-fermentative bacteria that could utilize a variety of organic substrates, including Streptococcus (1.3 ± 0.5%), Bifidobacterium (1.1 ± 0.7%), Rhodobacter and Rhodobium (1.0 ± 0.4%); four bulking- and foaming-related genera, i.e. Candidatus Microthrix (4.0 ± 1.5%), Gordonia (2.0 ± 2.5%), Nocardioides (1.5 ± 0.7%) and Tetrasphaera (Guo and Zhang, 2012a); plus five not well-described genera, i.e. Iami, Amaricoccus, Caldithrix, Pirellula and Haliea (Fig. S2 and Table S4). Other less abundant genera (< 1.0%) includes another four widely reported denitrifying-related genera, that is Azoarcus, Zoogloea, Thauera and Hyphomicrobium, and one well-known ammonia-oxidizing genus, Nitrosomonas, which plays particularly significant roles in oxidizing ammonia into nitrite during nitrification process in WWTPs (Raponi et al., 2004).
The variations of AS communities over the 4-year sampling period were examined based on two-way analysis of variance. As shown in Fig. 3A and B, 33 genera displayed significantly changed [P-value < 0.05, false discovery rate (FDR) < 0.3] abundances over the 4-year sampling period. Among these genera, 12 (names in blue) manifested significantly changed abundances with seasonal alternations from summer to winter (P1, Fig. 3A); 15 (names in purple) had significantly changed abundances from the first 2 years to the last 2 years (P2, Fig. 3B); and 6 (names in green) displayed significantly changed abundances across both P1 and P2. To unravel the underlying influential factors responsible for the complex variation patterns observed in the AS communities, those genera with significantly changed abundances across P1, P2 and P1P2 were extracted (as shown in Fig. 3A and D), and the relationships between their abundances and the operational conditions/wastewater characteristics of the AS process were explored by CCA using 4-year monitoring data in Sha Tin WWTP (Table S1) and the past software (Hammer et al., 2001).
Figure 3. Boxplots showing all genera (A, B) with significantly changed abundances across Pattern 1 (P1) and/or Pattern 2 (P2) based on two-way analysis of variance, and genus-conditional triplot (C) displaying variations of these significantly changed genera with respect to the environmental variables, based on canonical correspondence analysis (CCA). P1: summer vs winter samples; P2: first-2-years vs last-2-years samples. For subfigures A and B, the genus names in bold blue, purple and green represent those genera with significantly changed abundances across P1, P2 and P1P2 (both P1 and P2) respectively. For subfigure C, eigenvalues of horizontal and vertical axes equal to percentage variances of 49.2% and 32.1% respectively; each genus is represented by a coloured point (blue for P1, purple for P2, and black for both P1 and P2), accompanied by the genus name. Environmental variables are indicated by thick green lines with variable names (in red) at the end.
Download figure to PowerPoint
AS shown in Fig. 3C, all 12 P1-affiliated genera (the blue points) were distributed either along or very close to temperature and salinity lines (or their extension lines). These strong correlations of P1-affiliated genera with either temperature or salinity indicated that temperature and salinity were the major variables that led to the significantly different abundances of these populations from summer to winter, with those abundant in summer being either positively correlated with temperature or negatively correlated with salinity, or even both. This positive correlation of abundances of NOB like Nitrospira and hydrolysers like Lewinella with temperature partially explains the decreased nitrification and hydrolysis activities at lower temperature, leading to the incomplete nitrification and hydrolysis problems that commonly occur within full-scale biological treatment units during winter season. On the contrary, those P1-affiliated genera located on negative direction of horizontal axis were positively correlated with salinity, and simultaneously negatively correlated with temperature, with P1-ratio between 0.2 and 0.8 (Table S4). Among them, two genera, that is Tetrasphaera and Nocardioides, may be associated with bulking and foaming problems that commonly occurred in the AS of many WWTPs (Guo and Zhang, 2012a); other three genera, namely Bifidobacterium, Dorea and Ruminococcus, have been regarded as human faecal bacteria (Qin et al., 2010), which mainly originated from the sea water used for toilet flushing.
Moreover, Fig. 3C showed that all P2-affiliated genera (the purple points) were distributed between the lines (or their extension lines) of two clusters of influential factors: i.e. cluster (I) dissolved oxygen (DO), mean cell retention time and sludge retention time (SRT); and cluster (II) COD, mixed liquor suspended solids, TKN-N and TP, perhaps implicating the combined influences of multivariables that led to the significantly changed abundances across P2 in P2-affiliated genera. It could also be found that P2-affiliated genera were independent from the influences of temperature and salinity, considering the fact that these genera located somewhere around the vertical directions of lines of temperature and salinity. Moreover, those P2-affiliated genera located on positive direction of vertical axis, such as Clostidium, Pirellula, Amaricoccus and Pseudorhodobacter, had significantly higher abundance in the first 2 years than the last 2 years, with ratios of averaged abundances from first 2 years to last 2 years (P2-ratio) between 1.4 and 5.6. However, for those P2-affiliated genera located on the negative direction of vertical axis (e.g. Flavobacterium, Flexibacter, Haliea and Aeromonas), their abundances in the first 2 years were usually lower than the last 2 years, with P2-ratios between 0.03 and 0.4.
Particularly, six genera (the black points in Fig. 3C), including Nitrosomonas, Kosmotoga, Tetrasphaera and etc., displayed dual variation patterns characterized by significantly changed abundances across both P1 and P2, indicating the coexistence of multiple variables shaping the abundances of these genera in AS.
Global gene functional profiles and seasonal dynamics in AS
The overall functional profiles were predicted for the eight AS metagenomic data sets using the SEED subsystem (Overbeek et al., 2005) in mg-rast at the e-value cut-off of 10−5 (Meyer et al., 2008). For each data set, 41.1–51.4% of the 17 001 280–18 608 516 tags contained predicted proteins assigned to known functions at Level 1. This was comparable to the percentage of annotated sequences (40%) in a previous study on AS using pyrosequencing (Sanapareddy et al., 2009). The most dominant functional categories were those involved with clustering-based subsystems (15.6 ± 0.1%), carbohydrates (10.2 ± 0.2%), protein metabolism (8.8 ± 0.2%), and amino acids and derivatives (8.6 ± 0.1%) (Table S6), suggesting their significant roles in microbial communities of AS. These dominant functional categories evident in our AS metagenomes were also highly represented in metagenomic surveys in other environments like grassland soil (Costello et al., 2009), marine environment (Chaffron et al., 2010), freshwater (Pandit et al., 2009) and enhanced biological phosphorus removal (Horner-Devine et al., 2007), indicating high similarity of Level 1 functional categories among these ecosystems. Moreover, another two categories, namely ‘nitrogen metabolism’ and ‘phosphorus metabolism’, merely accounted for 1.38 ± 0.07% and 0.90 ± 0.02% of all the predicted proteins that were assigned to known functional categories, although they are of particular importance for biological nitrogen and phosphorus removal from wastewater.
Due to the difference in environmental and operational parameters (e.g. temperature, salinity, DO, SRT), significantly different abundances of functional genes were expected to occur in summer and winter. However, it seemed that differences in the 28 Level 1 functional categories between summer and winter were not significant, considering the fact that relative standard deviations of those functional categories in all AS samples were no more than 9.7% (averaging at 3.2%), as shown in Table S6. Further comparisons at subsystems Level 2 and Level 4 (Fig. 4) showed that except for a small number of low-abundance functional categories at Level 4, the majority of functional categories showed no significant difference (using twofold as the cut-off) in abundance between summer and winter. This implicated that compared with its microbial compositions, the functional categories in AS were much more stable. There are several possible explanations for this: (i) high proportion of functional genes related to fundamental metabolism are shared in summer and winter, (ii) gradients of influential factors are not strong enough to make significant changes in abundances of functional genes, and/or (iii) influential factors exert discrepant effects on abundances of microorganisms carrying similar functional genes.
Figure 4. The signal intensities (reads number) of functional categories from summer (X-axis) and winter (Y-axis) samples. The results are based on subsystem annotation at Level 2 (A) and Level 4 (B). The differentially detected genes were identified as signal intensity difference of ≥ 2 folds, indicated by the two diagonal lines. Each point in the figure represents one functional category at Level 2 or Level 4 determined by subsystem implemented in mg-rast. The signal intensities were indicated by the number of sequences that were assigned into each category.
Download figure to PowerPoint
Comparison with previous studies on AS
This is the first systematic metagenomic profiling of seasonal dynamics of AS communities by conducting Illumina HTS. Our analysis demonstrates that some core genera, especially those belonging to functionally important groups in AS (such as nitrifying bacteria, bulking and foaming bacteria, denitrifying bacteria, fermentative bacteria, hydrolyser, PAO, etc.), were shared by all the eight samples, but they had quite different abundances in summer and winter seasons. Meanwhile, functional analysis based on SEED subsystems reveals that the functional categories in AS showed no significant difference between summer and winter.
Previously, most studies on the diversity of AS microbial communities mainly rely on analysis of a single AS sample using clone library analysis (Sasaki et al., 1994), microarray (Xia et al., 2010) or 454 pyrosequencing (Kwon et al., 2010), which targeted specific genes (such as 16S rRNA) for exploring the microbial diversity without observing their functions within AS. For the first time, microbial communities in AS were correlated with 11 environmental and operational parameters in a full-scale WWTP by metagenomic technique and CCA analysis, which provides a comprehensive understanding of the influences of different variables on the microbial component and dynamics of functionally significant microorganisms. Our findings indicate that beside temperature and salinity, other variables (such as SRT, DO, salinity, etc.) also play significant roles in shaping the overall AS community structure. However, for two variables with opposite effects on abundances of the same species, it is not easy to distinguish whether the species is more positively affected by one variable, or more affected by the negative effect of the other. For example, judged from the CCA plots, temperature and salinity exert almost contrary effects on AS microbes, although the effect of temperature is more significant. Unfortunately, it is hard to distinguish the effects of different variables on microbial community in a full-scale WWTP since a lot of uncontrollable or even undetectable influential factors are involved in such a pollutant-removing process. Further studies, probably laboratory-scale ones designed with much simpler and more controllable experimental conditions, are needed to clarify the effect and significance of each variable on the AS microbial community dynamics as well as stability of the process.
Several technical limitations may affect our results. First, the length of tags obtained in this study was short, about 167 bps in average, although taxonomic classification of this length by LCA in megan indicates that 100 bp is long enough to identify a species, and 200 bp might constitute an optimal trade-off between the rate of under-prediction and the production cost of such reads (Huson et al., 2007). Moreover, in this work a sequencing depth of 5G is still not deep enough to accurately explore the rare species within AS owing to the limited number (12 624–15 473 for each sample) of the identified 16/18S rRNA gene tags, although metagenomic sequencing is neither low throughput nor PCR-based. Finally, biases are also likely to be introduced during the DNA extraction, although the optimal extraction kit (after comparison with several other kits) has been used, and Illumina HTS has been proved to possess good repeatability.