The role of inoculum and reactor configuration for microbial community composition and dynamics in mainstream partial nitritation anammox reactors

Abstract Implementation of partial nitritation anammox (PNA) in the mainstream (municipal wastewater treatment) is still under investigation. Microbial community structure and reactor type can influence the performance of PNA reactor; yet, little is known about the role of the community composition of the inoculum and the reactor configuration under mainstream conditions. Therefore, this study investigated the community structure of inocula of different origin and their consecutive community dynamics in four different lab‐scale PNA reactors with 16S rRNA gene amplicon sequencing. These reactors were operated for almost 1 year and subjected to realistic seasonal temperature fluctuations as in moderate climate regions, that is, from 20°C in summer to 10°C in winter. The sequencing analysis revealed that the bacterial community in the reactors comprised: (1) a nitrifying community (consisting of anaerobic ammonium‐oxidizing bacteria (AnAOB), ammonia‐oxidizing bacteria (AOB), and nitrite‐oxidizing bacteria (NOB)); (2) different heterotrophic denitrifying bacteria and other putative heterotrophic bacteria (HB). The nitrifying community was the same in all four reactors at the genus level, although the biomasses were of different origin. Community dynamics revealed a stable community in the moving bed biofilm reactors (MBBR) in contrast to the sequencing batch reactors (SBR) at the genus level. Moreover, the reactor design seemed to influence the community dynamics, and reactor operation significantly influenced the overall community composition. The MBBR seems to be the reactor type of choice for mainstream wastewater treatment.


| INTRODUCTION
Partial nitritation anammox (PNA) is the combination of anaerobic ammonium oxidation (Anammox) with partial nitrification. Implementation of PNA for mainstream (municipal) wastewater treatment (WWT) can result in a paradigm shift for municipal WWT, as it holds promise to lower energy demands, enables better use of the organic carbon, and saves costs for excess sludge management Siegrist, Salzgeber, Eugster, & Joss, 2008). However, its applicability to the mainstream is still unclear and under investigation.
PNA is a complex biotechnological application where controlled partial nitrification (nitritation) through aerobic ammonia-oxidizing bacteria (AOB) is required to provide an ideal ratio of 1.32 g NO -2 -N/g NH + 4 -N, for anammox activity under anoxic conditions (Strous, Kuenen, & Jetten, 1999). Maintaining these conditions is challenging due to an intrinsic competitive microbial environment. To improve process understanding for better reactor performances and process stability, a link to microbial ecology is essential (Rittmann, 2006).
Several studies have investigated the microbial community in PNA systems. The key members of this microbial community are AOB and anaerobic ammonium-oxidizing bacteria (AnAOB), and nitrite-oxidizing bacteria (NOB) Persson et al., 2014;. Some studies also reported putative heterotrophic denitrifying bacteria (HB) in PNA systems based on clone libraries Langone et al., 2014;.
Despite the importance of HB for the PNA process-positive (Desloover et al., 2011;Wang et al., 2010) or negative (Bürgmann, Jenni, Vazquez, & Udert, 2011;Lackner, Terada, & Smets, 2008)-characterization of the heterotrophic bacteria has been very limited, until recently. Only a few studies have provided detailed insight into the community composition of the PNA systems using next-generation sequencing platforms (Chu et al., 2015;Costa et al., 2014;Pellicer-Nácher et al., 2010;Pereira et al., 2014;Speth, in 't Zandt, Guerrero-Cruz, Dutilh, & Jetten, 2016). These studies reported two important aspects: (1) a vastly diverse community composition in PNA systems, (2) the need to further investigate the whole community composition and structure in PNA systems, to shed more light on the role of each community member for removing nitrogen in PNA systems, directly or indirectly. However, these studies were performed on systems which were not operated under mainstream conditions, rather as anammox enrichment reactors or sidestream PNA reactors. Thus, knowledge about whole microbial community structure of PNA systems under mainstream conditions remains to be elucidated.
For start-up of sidestream PNA systems, it is common practice to inoculate with biomass from existing PNA systems due to the low specific growth rates of AnAOB. Such a start-up strategy is even more important for mainstream PNA systems due to low temperatures and low nitrogen concentrations and thus even lower specific activities and growth rates of both AnAOB and AOB (Hendrickx et al., 2012).
However, previous studies have reported that the PNA reactors contain only one dominant AnAOB (Hu et al., 2010;Park, Rosenthal, Jezek, et al. 2010), indicating that specific environmental conditions might support one AnAOB over another. For example, anammox reactors with continuously high nitrite concentrations and COD: N ratios reported a predominance of the genus Ca. Brocadia over other AnAOB phenotypes (Jenni, Vlaeminck, Morgenroth, & Udert, 2014;Laureni et al., 2015). There is, however, no clear consensus in previous studies on what drives the development of one particular anammox strain over another. Such variations also extend to the occurrence of different genera of AOB. Therefore, selection of the right inoculum for a mainstream PNA system might be crucial due to variations in the dominant AnAOB and AOB. The choice of the right inoculum might extend to other community members too, as they are also significant (Speth et al., 2016).
All known AnAOB genera have been identified in different PNA system configurations (Egli et al., 2001;Park, Rosenthal, Jezek, et al. 2010;van der Star et al., 2007;Wang et al., 2010). A recent study highlighted the need for a deep taxonomic resolution, that is, down to genus or species level.
Fast start-up and stable performance of mainstream PNA systems thus relies on the inoculum composition and the reactor configuration. 16S rRNA gene amplicon sequencing provides such a resolution for comprehensive community determination, although it also has its pitfalls (Albertsen, Karst, Ziegler, Kirkegaard, & Nielsen, 2015).
The question remains how different the biomass composition in different full-scale PNA systems actually is, and thus, how important it is to choose the "right" inoculum considering that the microbial community structure might be distinct for different wastewater compositions, plant configurations, and operating conditions. Also, the subsequent impact of a temperature decrease and the adjusted implementation of PNA in the mainstreams are still unclear.
In our previous study , we compared process performance of four different PNA lab-scale reactors operated under mainstream conditions. The biomass originated from four different full-scale PNA reactors at three different locations. We also monitored the abundances of AnAOB, AOB, and NOB based on the qPCR analysis. The qPCR results suggested the presence of other abundant microbial members. Also, the process performance data revealed significant differences in performance among the four lab-reactors .
These findings raised further concerns on: (1) the analysis of the whole microbial community and its composition, (2) differences in the community composition among the inocula and the responsibility for differences in reactor performances. Therefore, in this study, we focused on answering the questions based on findings from other studies as well as our previous work. The overall aim of this study was twofold: (1) a comprehensive comparison of the microbial community composition of the biomasses from four full-scale sidestream PNA systems which were used as inocula in the four lab-scale reactors operated under mainstream conditions; (2) in-depth monitoring of the species abundance distribution and their dynamics in these lab-scale reactors, subjected to a gradual temperature decrease and low nitrogen concentrations, and determine the correlation between the previously reported reactor performances with the response of the whole community to the change in conditions.

| Reactor setup and operation
Our model systems, to gain comprehensive insight into the composition and dynamics of the total microbial community of the four different PNA biomasses, were four 10 l lab-scale reactors -two sequencing batch reactors (SBRs), one with suspended biomass (SBR1) and one with granular biomass (SBR2); two moving bed biofilm reactors (MBBRs), one with BiofilmChip™ M carrier material (MBBR1) and one with K3 ® carrier material (MBBR2), (both AnoxKaldnes AB, Lund, Sweden) which were operated as described in Gilbert et al. (2015).
These reactors were of particular interest, as the biomass in these reactors originated from four different full-scale sidestream PNA which also differ in their biomass enrichment methods (Table 1), and their biomass is used to inoculate other PNA reactors in Europe. All four labreactors were operated in parallel and fed with synthetic wastewater with 50 mg-N l −1 ammonium (see also Suppl. Information). The reactors were equipped with online sensors for dissolved oxygen (DO), temperature, pH, conductivity, and ammonium and nitrate (Endress+Hauser, Germany). All reactors were controlled based on ammonium effluent concentrations which were set to 6-8 mg-N l −1 by adjusting the influent flow rate. The pH was maintained at 7.3 ± 0.3, DO and biomass concentrations are provided in Table 1. Reactor performances and microbial population dynamics were monitored over a period of 45 weeks. The applied temperature profile is shown in Fig. S1 and was as follows: reactor operation started at week 0 and during phase I, the reactors were operated at 20°C (15 weeks); in phase II, the temperature was gradually decreased by 0.5°C every 7 days down to a temperature of 10°C (weeks 16-35); phase III covered operation at constant 10°C (week 36-45).

| Sampling and amplicon library preparation
Biomass samples were collected from the reactors during operation and stored at −80°C for further analysis (see also Suppl. Information).
Extracted genomic DNA was used for library preparation employing a procedure adapted from Caporaso et al. (2011). Using 5 ng/μl of DNA, template PCR amplification (see also Suppl. Information) was performed in duplicate using hypervariable region V4 primers (targeting 253 bp partial 16S rRNA gene sequence) fused with barcodes and adapters for the Illumina MiSeq platform. The duplicate library reactions were pooled together, and the libraries were cleaned using Agencourt AMPure XP beads (Beckman Coulter) with the standard protocol. The quality of the library was checked using TapeStation and D1K ScreenTape (Agilent Technologies Inc., USA).

| Sequencing and sequence analysis
The multiplexed amplicon library was sequenced on a MiSeq System (Illumina) using MiSeq Reagent Kit v3 (Illumina). Samples from the libraries were analyzed by a quick quality check to evaluate overall data quality (check the quality score, the occurrence of ambiguous bases and k-mer abundances). Afterward, the libraries were processed to (1) truncate read 1 to 250 bp length and discard read 2; (2) screen for PhiX contamination; (3) format metadata for QIIME; and (4) remove unique reads and normalize libraries to an even depth of 10,000 reads per sample.
OTU (operational taxonomical unit) picking and classification was done using QIIME v. 1.7 (quantitative insight into microbial ecology) (Caporaso et al., 2010) with its default settings. Overall, the de novo clustering of OTUs was done with 97% identity, corresponding to species level. The sequences were then classified using the RDP (Ribosomal Database Project) classifier (80% confidence threshold) based on the taxonomy in the Greengenes database (97% confidence threshold, version 13.5 May 2013) (McDonald et al., 2012).
Additionally, NCBI BLAST database was used to determine the closest bacterial relatives (having 100% sequence homology) of the reported OTUs. OTU representative sequences have been submitted to the GenBank under the accession numbers KY226724 -KY228337.

| Data analysis
The sequencing data was analyzed using R to gain insight into the overall community structure dynamics over time in all four reactors T A B L E 1 Characterization of the inocula and reactor operation data (averages over the entire 45 weeks)

| Characterization of the inocula-microbial community composition
The biomasses that were used to inoculate the four lab-scale reactors, without any cross-inoculation during the 45 weeks of operation, originated from different full-scale sidestream facilities treating centrates from sludge dewatering units. The operating conditions at these sites differed with influent ammonium concentrations of 500-1,000 mg-N l −1 , effluent ammonium concentrations around 50-100 mg-N l −1 , and operating temperatures between 22 and 34°C, caused by the installation in the sidestream (after anaerobic digestion) .
The characteristics of the biomasses are summarized in Table 1.
The biomasses varied in their physical structure (type) as a result of the reactor type: two inocula were from sequencing batch reactors, suspended (SBR1) and granulated (SBR2) biomasses, respectively, characterized by different particle size distributions. The other two inocula came on carrier materials with biofilms of maximum thicknesses of 2 mm (MBBR1) and 10 mm (MBBR2), respectively. For clarity, in this study, the total microbial community was divided into a nitrifying microbial community (NMC, typically studied community) and a heterotrophic microbial community (HMC   (Table S2) and PCA analysis (See Fig.   S3 for details).
The results clearly showed that the predominant NMC members and several HMC members were present in all the inocula, which was not expected considering the different settings (reactor type, wastewater composition) at each site. Similar to our result characterizing the inocula, other studies also reported low fractions of AOB and NOB in PNA reactors (Almstrand et al., 2014;Kindaichi et al., 2007). The relative abundance of Ca. Brocadia was relatively high in all four inocula, but it was the HMC that constituted the major biomass fraction of each inoculum (Figure 1a) although the COD/N ratios were <2 in the systems of origin . Our results are in congruence with previous studies about the coexistence of HB in anammox systems (Costa et al., 2014;Garcia Costas et al., 2012;Kindaichi et al., 2007;Langone et al., 2014;Laureni et al., 2015;Ni, Ruscalleda, & Smets, 2012;Speth et al., 2016). In a study (Ni et al., 2012), the presence of (residual) organic content including soluble microbial products (SMP) in the side-and mainstream wastewaters was reported to support the HMC in PNA systems. Interestingly, all biomasses fostered the same AOB and NOB, sharing a single common OTU related to Nitrosomonas europaea-eutropha and Nitrospira, respectively. These observations can be explained based on the aeration strategies employed in PNA systems-limited oxygen, and therefore, lower transient nitrite availability. These conditions are reported to support Nitrosomonas europaea-eutropha (Li et al., 2009) and Nitrospira (Park, Rosenthal, Jezek, et al. 2010

| Reactor performance
Detailed performance analyses of all four reactors have been reported previously . Here, we summarize the operational performance of these reactors and determine reasons for the observed performance change in conjunction with the response of the whole microbial community to mainstream conditions. With the start of reactor operation (week 0), the temperature was set to 20°C and only ammonium and nutrients were fed to the lab-reactors, hence removing any external carbon source, compared to the conditions at the full-scale facilities (a few hundred mg/L organic carbon measured as COD). These reactors were operated with organic carbon-free influent to reduce the potential contribution of denitrification to PNA and to be able to attribute changes mainly to the temperature gradient. All four lab-reactors were exposed to the same temperature profile over three defined phases (Fig. S1)-phase I (weeks 0-15), start-up at 20°C; phase II (weeks 16-35), temperature gradient (20°C to 10°C); and phase III (weeks 36-45) operation at 10°C. Table 1 provides the main operating parameters. Over the course of this study, biomass concentrations were quite stable. However, the concentrations differed significantly between the reactors, with 0.8 and 2.0 g-TSS l −1 for the SBRs, and 5.7 and 9.3 g-TSS l −1 for the MBBRs (Table 1). Deviations in operational parameters (HRT and DO) were based on the biomass characteristics, that is, different thicknesses and total amounts.
Reactor performances are summarized in Figure 2 with additional details in S. Table 1. Initial nitrogen loadings (week 0) were highest in the MBBRs (around 200 g-N m −3 d −1 ), followed by SBR2 (approx. 80 g-N m −3 d −1 ) and SBR1 (40 g-N m −3 d −1 ), which was mainly attributed to the different biomass concentrations. The ammonium removal was above 90% and similar in all four reactors. By the end of phase I, the nitrogen loading was reduced to maintain the effluent concentration. The biomass-specific conversion rates were between 5 and 8 g-N kg-TSS −1 d −1 . During the temperature decrease (phase II), the turnover rate reduced by a factor of 2, as was expected for a temperature decrease down to 10°C. In phase III, ammonium removal remained rather stable.
Nitrite and nitrate production significantly differed between the reactors. During phase II, a significant nitrite accumulation occurred in SBR1 and SBR2 (from almost 0% (week 15) to 62% in SBR1 and 54% in SBR2 (week 35)), with a simultaneous decrease in nitrate production to almost nothing (3%-6% in week 35) ( Figure 2, Table S1). Nitrite accumulation in MBBR1 also increased significantly from 0% to 30% in phase III. Only MBBR2 did not accumulate any nitrite during the temperature decrease. In phase III, nitrite production remained at high levels except for MBBR2 (<5% of the converted ammonium). Nitrate production decreased in all reactors during the temperature gradient from 20°C to 10°C from 25%-40% to less than 15%. The reactor performances correlated well with biomass structure and aggregate/ biofilm thickness. MBBR2 with a biofilm of up to 10 mm thickness performed best (least nitrite and nitrate accumulation), whereas the reactors with thinner biofilms/smaller aggregates struggled with nitrite accumulation.

| Microbial community dynamics
The results of the reactor operation suggest that the systems were influenced differently by the temperature gradient, although influent composition and operating conditions were the same in all four reactors. The question was now whether the microbial community composition could explain the differences in behavior. Therefore, a constrained correspondence analysis (CCA) was performed as a first step to ascertain whether the temperature change affected the community composition over time. This analysis revealed that the microbial community composition in the MBBRs was relatively stable, whereas it was highly dynamic in the SBRs (Figure 3). The response of the community composition showed a high correlation (ANOVA (analysis of variance), p < .001) to the temperature decrease and nitrite accumulation (Figure 2, Figure 3). The influence of the temperature change on community structure was more significant for the SBRs compared to the MBBRs, especially in SBR1, which exhibited a continuous instability during the entire operation, corresponding to lower efficiency ( Figure 2, and Supplementary information for details).

| Nitrifying microbial community
The correlation of the community dynamics with the nitrite buildup suggested a significant influence of the temperature decrease on the Nitrifying microbial community (NMC). However, nitrite buildup can originate from loss of AnAOB activity as well as from loss of NOB activity or a relative increase in AOB activity; thus, causing overproduction

| Heterotrophic microbial community
Another reason for the differences in reactor performances might stem from the large Heterotrophic microbial community (HMC) in all reactors. Diversity and richness within the HMC at the end of phase I increased in all reactors except SBR1 compared to the inocula biomasses (Fig. S6, Table S2). Afterward, the differentiation in community dynamics among the SBRs and MBBRs was analogous to the NMC (Fig.   S2, Fig. S6). The community diversity in the MBBRs was relatively stable compared to the SBRs. The relative abundance profiles of the most dominating OTUs associated with different phyla within the HMC differed in the four reactors ( Figure 1b, Table S3). They belonged to the genera Rubrivivax, Rhodoplanes, Dok59, Dokdonella, Thermomonas, and some unclassified members of the phyla Chloroflexi, Chlorobi, Actinobacteria, and Acidobacteria. Few OTUs in the MBBRs and SBRs behaved similarly in the different phases (Table S3). During phase II, the relative abundance of some OTUs increased significantly in the SBRs (Figure 1b, Fig. S7

| Interrelationship between microbial community, reactor type, and reactor performance
This study clearly showed that the temporal community dynamics were linked to the reactor type and driven by a temperature decrease.
The communities shifted in all reactors from the initial inocula ( Fig.   S5), particularly the abundance of the nitrifying community reduced, attributed to the selective pressure imposed by changes in the operational conditions. Although the inocula of all four lab-reactors har-  (Dosta et al., 2008;Hu et al., 2013 (Lotti et al., 2015). They reported that the AnAOB activity in biofilms was less influenced than in suspended biomass. We found a strong correlation between the NMC composition and the change in temperature, the nitrite accumulation (i.e., performance and stability), and the reactor type in our long-term study. Interestingly, the observed NMC dynamics occurred only within the framework of the initially detected OTUs. The magnitude of the effect of the temperature decrease was linked to the thickness of the microbial aggregates. In our study, SBR1 (with suspended biomass) responded first by a decrease in AnAOB population and sudden nitrite accumulation, followed by SBR2 (granular biomass). AOB were producing nitrite, but its consumption was not counterbalanced by either of the microbial members responsible for the nitrite sink. AnAOB were influenced by the temperature decrease rather than the competition for nitrite imposed by NOB and denitrifiers, which is often attributed to the failure of PNA systems (Joss et al., 2011;Lackner et al., 2015). Moreover, NOB also seemed to be negatively influenced by the temperature decrease. Although nitrite was available, no increase in their population was observed. Huang, Gedalanga, Asvapathanagul, and Olson (2010) also reported negative impact of low temperature on Nitrospira (also detected in our study). The accumulation of NO 2 - (Figure 2) clearly indicated that AOB activity and competition for nitrite were not responsible for the poor reactor performance below 13°C.
The absence of organic carbon in the influent of the lab-reactors had no significant influence on the composition of the HMC. Thus, in consensus with previous studies (Ni et al., 2011(Ni et al., , 2012 Thermomonas fusca, which can thrive at 10°C (Mergaert, Cnockaert, & Swings, 2003;Langone et al., 2014)) other members of the HMC were negatively affected. This observation is supported by previous studies, that lower temperature has a negative impact on denitrifiers (Lu, Chandran, & Stensel, 2014). Additionally, the reactor configuration was also relevant for the dynamics of the HMC. The MBBRs had a rather stable total community composition compared to the SBRs.
Similar results were obtained in a study by Laureni et al. (2016), who reported higher community dynamics in suspended biomass compared to biofilm carriers in PNA reactors operated at 15°C. Such community behavior seemed to be caused by the lack of structural preservation which was available in the MBBRs.

| CONCLUSIONS
The detailed microbial community study performed on biomasses