Different Engineering Designs Have Profoundly Different Impacts on the Microbiome and Nitrifying Bacterial Populations in Municipal Wastewater Treatment Bioreactors

ABSTRACT

Numerous wastewater treatment processes are designed by engineers to achieve specific treatment goals. However, the impact of these different process designs on bacterial community composition is poorly understood. In this study, 24 different municipal wastewater treatment facilities (37 bioreactors) with various system designs were analyzed by sequencing of PCR-amplified 16S rRNA gene fragments. Although a core microbiome was observed in all of the bioreactors, the overall microbial community composition (analysis of molecular variance; P = 0.001) as well as that of a specific population of Nitrosomonas spp. (P = 0.04) was significantly different between A/O (anaerobic/aerobic) systems and conventional activated sludge (CAS) systems. Community α-diversity (number of observed operational taxonomic units [OTUs] and Shannon diversity index) was also significantly higher in A/O systems than in CAS systems (Wilcoxon; P < 2 × 10⁻¹⁶). In addition, wastewater bioreactors with short mean cell residence time (<2 days) had very low community α-diversity and fewer nitrifying bacteria compared to those of other system designs. Nitrospira spp. (0.71%) and Nitrotoga spp. (0.41%) were the most prominent nitrite-oxidizing bacteria (NOB); because these two genera were rarely prominent at the same time, these populations appeared to be functionally redundant. Weak evidence (AOB:NOB « 2; substantial quantities of Nitrospira sublineage II) was also obtained suggesting that complete ammonia oxidation by a single organism was occurring in system designs known to impose stringent nutrient limitation. This research demonstrates that design decisions made by wastewater treatment engineers significantly affect the microbiome of wastewater treatment bioreactors.

IMPORTANCE Municipal wastewater treatment facilities rely on the application of numerous “activated sludge” process designs to achieve site-specific treatment goals. A plethora of microbiome studies on municipal wastewater treatment bioreactors have been performed previously; however, the role of process design on the municipal wastewater treatment microbiome is poorly understood. In fact, wastewater treatment engineers have attempted to control the microbiome of wastewater bioreactors for decades without sufficient empirical evidence to support their design paradigms. Our research demonstrates that engineering decisions with respect to system design have a significant impact on the microbiome of wastewater treatment bioreactors.

INTRODUCTION

The activated sludge process is a general term used to describe virtually all wastewater treatment bioprocesses involving dense suspensions of microbial biomass. Activated sludge processes can be designed to mineralize labile organic carbon (typically measured as 5-day biochemical oxygen demand [BOD₅]), oxidize ammonia, reduce nitrite/nitrate, and perform numerous other biotransformations, such as enhanced biological phosphorus removal. Several different process designs have been developed to achieve the desired treatment goals; these designs can vary substantially depending on wastewater type (e.g., municipal versus industrial), wastewater characteristics (especially nutrient concentrations), regulatory requirements, and receiving water body characteristics (which nutrients are limiting?). More recently, activated sludge process design has also focused on reducing the energy required for treatment (1) and simultaneously recovering resources (2–6) while still maintaining treatment quality.

Different activated sludge designs used for treating municipal wastewater almost certainly impose different selective pressures that alter microbial community composition and physiological function. Activated sludge bioreactors are well known to support complex, diverse, and dynamic microbial communities (7, 8), although recent results obtained by sequencing of PCR-amplified 16S rRNA gene fragments suggest that there is a core municipal wastewater treatment microbiome (7, 9–11). A few keystone populations are particularly pertinent for municipal wastewater treatment performance, including beneficial organisms like nitrifying bacteria (12) and polyphosphate-accumulating organisms (13) as well as nuisance organisms such as organisms that cause bulking (14, 15), organisms that cause foaming (15), and glycogen-accumulating organisms (16).

Even though there is a substantial body of knowledge regarding the bacterial communities in municipal wastewater treatment bioreactors, there is little understanding of how process design affects bacterial community composition in these bioreactors (17). For example, the conventional activated sludge (CAS) process consists of a single aerated bioreactor coupled to a quiescent settling chamber (known as a secondary clarifier); this process is suitable for reducing BOD₅ and for ammonia removal via nitrification. In contrast, the A/O process consists of an anaerobic bioreactor in series with an aerobic bioreactor, followed by a secondary clarifier. This process is intended to achieve enhanced biological phosphorus removal in addition to reducing BOD₅ and ammonia. Numerous other process designs are also used to achieve different treatment goals.

The objective of this study is to test the hypothesis that different system designs profoundly affect microbial community composition and, in particular, the types and quantities of nitrifying bacteria supported by activated sludge bioreactors. In this study, therefore, we tracked the bacterial community composition of 24 different municipal wastewater treatment facilities (eight different system designs) for 12 months. These facilities encompassed numerous process designs, including a high-rate and high-purity oxygen (HR-HPO) system, an adsorption/bio-oxidation (A/B) system, 7 A/O systems, 10 CAS systems, a trickling filter followed by a conventional activated sludge (TF-CAS) system, a Johannesburg (JHB) system, 2 membrane bioreactor (MBR) systems, and a sequencing batch reactor (SBR) system. Bacterial community composition was tracked by sequencing of PCR-amplified 16S rRNA gene fragments (V4 region). Microbial community composition, α-diversity metrics, and the abundance and composition of nitrifying bacteria were compared between these full-scale bioreactors.

RESULTS

Community composition.

Samples were collected approximately monthly from 24 different wastewater treatment facilities for 1 year; because 4 facilities had parallel treatment processes and because numerous facilities had multiple bioreactors in series, these 24 wastewater treatment facilities involved 37 different bioreactors. Characteristics of the facilities as well as the average operational parameters are summarized in Table S1. A total of 430 bioreactor samples were collected and used as the template for sequence analysis of PCR-amplified 16S rRNA gene fragments. An average of 23,744 (standard deviation [SD] = 9,904) quality sequences were obtained per sample. This analysis generated 2,003 ± 307 operational taxonomic units (OTUs; 97% sequence similarity) and 350 ± 50 genus-level phylotypes per sample.

Although temporal variation was observed in the microbial community composition of all bioreactors, substantial similarity was observed in the average genus-level community composition (Fig. 1A). Seventeen phylotypes had greater than 1% overall relative abundances, and 16 of them were detected in all of the bioreactors. This suggests that wastewater treatment bioreactors support a core microbiome as has been suggested previously (7, 10, 11). Principal coordinate analysis (PCoA) of the community Bray-Curtis dissimilarities, however, suggested that bioreactor design had a significant effect on microbial community composition (Fig. 1B; see Fig. S1 in the supplemental material for phylotype-based analysis). Specifically, analysis of molecular variance (AMOVA) demonstrated that the microbial communities in the A/O facilities (n = 7) were significantly different from the microbial communities in the CAS facilities (n = 10) (AMOVA F = 1.21, P = 0.001) when averaged dissimilarity values were used for the facility with multiple bioreactors. Although there was insufficient replication (n < 3) of the other facility types to make statistically based comparisons, substantial differences in microbial community composition were also observed for these facility designs.

FIG 1.

(A) The abundance of 16 phylotypes that were found in all bioreactors. Results shown are the average community composition of monthly samples collected from July 2017 to July 2018. The key shows phylotypes as best classified. (B) Principal coordinate analysis (PCoA) of community dissimilarities between the bioreactors (constructed with operational taxonomic unit [OTU]-based community profile). Bray-Curtis dissimilarities between the bioreactors were calculated every month and averaged to determine community dissimilarities. Labels are used to identify the bioreactors from which the samples collected.

Wastewater treatment process design also significantly affected the α-diversity of the bacterial communities enriched wastewater treatment bioreactors (Fig. 2; see Fig. S2 for phylotype-based analysis). Statistical analysis confirmed that the number of observed OTUs in the A/O systems (median = 2,155) was significantly higher than that in the CAS systems (median = 1,972) (Wilcoxon Z = 8.5, P = 1.8 × 10⁻¹⁷). Similarly, the A/O systems had a Shannon diversity index significantly higher than that of the CAS systems (Wilcoxon Z = 8.2, P = 1.9 × 10⁻¹⁶). In addition, Tukey’s honestly significant difference (HSD) tests showed that the HR-HPO, A/B(high rate [HR]), CAS2(1), and CAS2(2) had α-diversity (number of observed OTUs) significantly lower than that of all of the other bioreactors (Table S2). A/O systems did not have significantly different richness from each other; in contrast, the α-diversity in CAS systems varied substantially (Table S2).

FIG 2.

The number of observed operational taxonomic units (OTUs) and the Shannon diversity index for each bioreactor. Symbols represent arithmetic means of samples collected from July 2017 to July 2018; error bars show one standard deviation of the mean.

Nitrifying bacteria.

Substantial quantities of ammonia-oxidizing bacteria (AOB) were detected within all wastewater treatment bioreactor communities (Fig. 3). Nitrosomonas spp. (0.35% of sequences) were the most prominent ammonia-oxidizing bacteria, although small populations of both Nitrosospira spp. (0.0009%) and Nitrosococcus spp. (0.0007%) were also detected. Many bioreactors exhibited substantial seasonal variation (>1-log difference between maximum and minimum) in the abundance of Nitrosomonas spp., including A/O2, A/O5, A/O7(2,3,4), CAS2, CAS5, MBR1, and MBR2 as well as the lone SBR, A/B, TF-CAS, and HR-HPO systems. Similarly, many of the treatment designs supported seasonally stable populations of Nitrosomonas spp., including the other A/O and CAS systems as well as all four bioreactors of the lone JHB system. A/O3, A/O7(1), and JHB system showed a very stable Nitrosomonas abundance (<0.5-log difference between maximum and minimum).

FIG 3.

Temporal changes in Nitrosomonas abundance in each bioreactor. Data represent the log₁₀-transformed percentage of sequences identified as Nitrosomonas spp. compared to the total number of sequences. Samples in which Nitrosomonas spp. were not detected were given a value equal to one-half of the detection limit and shown as open symbols.

Nitrospira spp. (0.71%) and Nitrotoga spp. (0.41%) were the most prominent nitrite-oxidizing bacteria (NOB) in the wastewater treatment bioreactor communities (Fig. 4), although a small number of Nitrobacter spp. (0.00008%) were also detected. NOB composition exhibited substantial variation by system design, with many bioreactors dominated by Nitrospira spp. (Fig. 4A), and other bioreactors were dominated by Nitrotoga spp. (Fig. 4B); the NOB in some bioreactors appeared to alternate between Nitrospira spp. and Nitrotoga spp. Nitrosomonas abundance was positively correlated with Nitrospira (Spearman’s rho [r_s] = 0.26, P = 3.8 × 10⁻⁸), Nitrotoga (r_s = 0.17, P = 4 × 10⁻⁴), and total NOB (r_s = 0.31, P = 2 × 10⁻¹¹) abundances. In contrast, Nitrospira and Nitrotoga abundances were negatively correlated with each other (r_s = −0.57, P = 2 × 10⁻³⁸). Although the abundances of Nitrospira spp. and Nitrotoga spp. fluctuated substantially in many of the bioreactors, the total NOB abundance was much more stable (Fig. 4C). In addition, the total population of NOB in the SBR systems and A/O7 exhibited a substantial decline during the winter months, which was similar to the decline in Nitrosomonas spp. in these systems.

FIG 4.

Temporal changes in (A) Nitrospira abundance, (B) Nitrotoga abundance, and (C) nitrite-oxidizing bacterium (NOB) abundance. Data represent the log₁₀-transformed percentages of sequences identified as Nitrospira spp. and/or Nitrotoga spp. compared to the total number of sequences. Samples in which NOB were not detected were given a value equal to one-half of the detection limit and shown as open symbols. The keys in panel C also apply to panels A and B.

System design did not have a statistically significant effect on the abundances of either AOB or NOB (Fig. 5). Nitrosomonas abundances in the A/O systems (median = 0.39% of total community) and in the CAS systems (median = 0.31% of total community) were not statistically different (Wilcoxon Z = 1.5, P = 0.14) (Fig. 5A). Other system designs did appear to affect the quantities of AOB; the HR-HPO, A/B(HR), and MBR2 had Nitrosomonas abundance significantly lower than that of all of the other bioreactors (Table S3). Similarly, total NOB abundance in the A/O systems (median = 0.80%) was not significantly different from that in the CAS systems (median = 0.79%; Wilcoxon Z = 0.99, P = 0.32) (Fig. 5B). In addition, the A/B(HR) bioreactor exhibited abundance of both AOB and NOB significantly lower than that of the A/B(low rate [LR]) bioreactor, even though they are operated in series (Table S4). Similarly, the HR-HPO system had low abundances of both AOB and NOB, consistent with a well-known design strategy that predicts that short mean cell residence times would result in lower quantities of nitrifying bacteria (17).

FIG 5.

The average abundance of (A) ammonia-oxidizing bacteria (AOB) and (B) nitrite-oxidizing bacteria (NOB) in each bioreactor. Data represent the arithmetic means of the log₁₀-transformed percentages of sequences identified as AOB or NOB compared to the total number of sequences. Open symbols indicate that some values were below the detection limit; results that were below the detection limit were assigned a value equal to half of the detection limit for computing the arithmetic mean. Error bars show one standard deviation of the log₁₀-transformed abundance values.

System design, however, affected the composition of Nitrosomonas spp. and Nitrospira spp. in the wastewater treatment bioreactors (Fig. 6). The most prominent Nitrosomonas spp. OTU in the A/O systems was typically different from the most prominent Nitrosomonas spp. OTU in the CAS systems; statistical analysis confirmed that Nitrosomonas spp. in the A/O system were different from the Nitrosomonas spp. in the CAS systems (AMOVA F = 1.95, P = 0.04; Fig. 6A). In contrast, the composition of Nitrospira spp. OTUs in the A/O system was not statistically different from that of the Nitrospira spp. OTUs in the CAS systems (AMOVA F = 1.13, P = 0.32; Fig. 6B). Substantial differences, however, in the Nitrospira spp. OTU composition were observed in the other system designs (Fig. 6C). Specifically, most of the A/O systems and many CAS systems were dominated by Nitrospira sublineage I OTUs, whereas the A/B(LR), A/O1, TF-CAS, CAS3, CAS8, CAS10, and MBR2 had substantial quantities of Nitrospira sublineage II OTUs.

FIG 6.

Principal coordinate analysis (PCoA) of the average Bray-Curtis community dissimilarities constructed with (A) 59 operational taxonomic units (OTUs) classified as Nitrosomonas spp. and (B) 69 OTUs classified as Nitrospira spp. (C) The average of the relative abundance of different sublineages of Nitrospira spp. in each bioreactor.

Thermodynamic calculations predict that for canonical nitrification, the quantity of AOB should be double that of the NOB (18, 19). A one-sided t test confirmed that all bioreactors except for the HR-HPO (most samples had nondetectable quantities of either AOB or NOB) and CAS9 systems had an AOB:NOB ratio smaller than 2 (P < 0.004) (Fig. 7; see Fig. S3 for temporal data). The lowest ratio of AOB:NOB was observed with the MBR2 system, which, in fact, had nondetectable quantities of Nitrosomonas spp. In addition, the only known organisms capable of complete ammonia oxidation are Nitrospira sublineage II (20, 21). Most of the A/O systems and CAS systems were dominated by Nitrospira sublineage I, although the A/B(LR), TF-CAS, CAS8, CAS10, and MBR2 had substantial quantities of Nitrospira sublineage II (Fig. 6C).

FIG 7.

The average of the ratio (log₁₀-transformed) between the abundance of the ammonia-oxidizing bacteria (AOB) and nitrite-oxidizing bacteria (NOB). The dashed line indicates an AOB:NOB ratio of 2; asterisks indicate that AOB:NOB ratio was not significantly smaller than 2 (one-sided t test P > 0.05). Open symbols indicate that averages included at least one sample that was below detection (see Fig. S3 for full temporal data); when AOB, NOB, or both were below detection, a numerical value of −3.5, 1.5, or 0, respectively, was used to compute the average. Error bars show one standard deviation of the mean of the log₁₀-transformed ratio of AOB:NOB.

DISCUSSION

Wastewater treatment engineers have developed numerous process designs for treating municipal wastewater that impose markedly different conditions to theoretically manipulate bacterial community composition to achieve a specific treatment goal (17). Although wastewater treatment bioreactors generally perform as intended by the design engineer (e.g., systems designed to perform enhanced biological phosphorus removal (EBPR) effectively remove phosphorus), the effects of different engineering decisions on bacterial community composition are not yet fully understood. This study demonstrated that the A/O design produces a microbial community composition significantly different from that of the CAS design, consistent with the a priori engineering goal. In addition, the A/O designs supported microbial communities more diverse than those supported by the CAS designs, which should be of practical importance because communities with higher α-diversity also have greater metabolic diversity and activity, greater resistance to perturbation (22), and greater performance stability (23), although the connection between higher diversity and better stability has been questioned (24).

System design also appears to have a profound impact on the abundance of nitrifying bacteria in wastewater treatment bioreactors. Historically, wastewater treatment engineers have attempted to control the abundance of nitrifying bacteria via a kinetic-based approach (17), which suggested that bioreactors with short mean cell residence times (<2 days) would support lower quantities of nitrifying bacteria than would bioreactors with higher mean cell residence times (>6 days). Our results are consistent with this paradigm. The A/B(HR) and the HR-HPO both supported significantly fewer nitrifying bacteria than did the other system designs. Similarly, the relative abundance of nitrifying bacteria was statistically similar between the A/O systems and the CAS systems, as these systems are operated at similar mean cell residence times. Substantial seasonal differences were also observed in the abundance of nitrifying bacteria in some systems, with the loss of nitrifying activity during cold weather months being commonly observed (17). However, some of the treatment systems in our study sustained abundant quantities of nitrifying bacteria throughout the calendar year, suggesting that different designs and/or operational strategies can be more effective at maintaining year-round nitrification (25).

System design appears to have a muddled impact on the composition of nitrifying bacteria in wastewater treatment bioreactors. Our study identified Nitrosomonas spp. as the most prominent AOB, similar to numerous previous studies (26), although other studies have also detected Nitrosospira spp. in wastewater treatment bioreactors (12). In contrast, the different wastewater treatment systems supported abundant quantities of Nitrospira spp. and/or Nitrotoga spp. as the most prominent NOB. Our results suggest that Nitrospira spp. and Nitrotoga spp. are functionally redundant during municipal wastewater treatment because the correlation between the quantity of AOB and the sum of Nitrospira spp. and Nitrotoga spp. was stronger than the correlations between AOB and either Nitrospira spp. or Nitrotoga spp. on an individual basis.

Interestingly, there was no consistency among the CAS systems and A/O systems as far as the specific population of NOB that was observed. Some systems were dominated by Nitrospira spp., some systems were dominated by Nitrotoga spp., and some systems exhibited oscillating populations of Nitrospira spp. and Nitrotoga spp. Previously published research is similarly inconsistent regarding which population of NOB is most prominent. For example, Nitrotoga spp. were the most abundant NOB in numerous Danish activated sludge plants (10), while Nitrobacter spp. and Nitrospira spp. were the most abundant NOB from a study conducted in the Netherlands (19). It is possible that operating conditions are more pertinent than system design in the ecology of NOB; a slightly acidic pH (5.7 to 6.8), lower operating temperature, and elevated nitrite loading have been reported to favor the growth of Nitrotoga spp. over that of Nitrospira spp. (27, 28).

In contrast, certain bioreactor designs appear to strongly favor individual organisms capable of complete ammonia oxidation. Nitrospira spp. lineage II was detected in bioreactors (A/B[LR], TF-CAS, and MBR systems); these bioreactors are typically associated with stringent nutrient limitation, which is a theoretical prerequisite for complete ammonia oxidation activity (29). From a wastewater treatment design perspective, stringent nutrient limitation is imposed either by using biofilm systems (TF-CAS) or by operating with very high mean cell residence times (A/B[LR], MBRs). Curiously, substantial differences were observed in the AOB:NOB ratio in the two MBR systems. MBRs are well known to impose a stringent nutrient limitation (30–32), so we would expect MBRs to be likely candidates to support complete ammonia oxidation activity. MBR1 had an AOB:NOB ratio similar to that of most of the other wastewater treatment bioreactors; however, its mean cell residence time was approximately 22 days, which is somewhat similar to that of the other wastewater treatment bioreactors. Consistent with our expectation, MBR2 supported substantial quantities of Nitrospira sublineage II but low or nondetectable quantities of conventional AOB; MBR2 had a mean cell residence time of 70 to 85 days.

Our research also suggests that other wastewater bioreactor designs (A/B, JHB, HR-HPO, SBR, MBR) might also support microbial community compositions specific to these designs. We were unable, however, to obtain samples from a sufficient number of replicate designs to reach statistically based conclusions. There are additional limitations of our study that are due to the nature of field-based research; multiple facilities were compared based on the system design (CAS systems versus A/O systems), but there were other confounding variables that might affect microbial community composition. We would expect that, for example, facility location, influent wastewater composition, bioreactor size, and site-specific operational strategies would also significantly affect bacterial community composition in full-scale wastewater treatment bioreactors. A strength of our study, however, is that we studied a large number of full-scale wastewater treatment bioreactors. This allowed us to avoid study-to-study and researcher-to-researcher variations caused by using different methods for sample collection, DNA extraction, PCR, sequence clustering, and diversity/distance metrics (33).

In this study, microbial community composition was investigated at numerous facilities with various system designs. Our results suggest that A/O systems and CAS systems supported statistically different microbial communities. A/O systems also had community richness and Shannon diversity indices higher than those of CAS designs. Systems with short mean cell residence times (<2 days) had very low community richness and low abundance of nitrifying bacteria. Nitrosomonas spp. were the most prominent AOB in all of the systems that supported substantial quantities of nitrifying bacteria. Nitrospira spp. (sublineages I and II) and Nitrotoga spp. were the most prominent NOB; these two genera appeared to be functionally redundant. A few systems supported substantial populations of Nitrospira sublineage II, the only organisms known at this time to be capable of performing complete ammonia oxidation. Our research demonstrates that engineering decisions with respect to system design have a significant impact on the microbiome of wastewater treatment bioreactors.

MATERIALS AND METHODS

Wastewater treatment plant descriptions.

Twenty-four wastewater treatment facilities in the midwestern United States were selected for this study (see Table S1 and Fig. S4 in the supplemental material). The HR-HPO system was designed to achieve rapid removal of labile organic carbon but circumvent nitrification to save on aeration costs. The A/B system was designed to achieve rapid adsorption of organic nutrients in the first reactor followed by an aerated reactor with a long retention time that was intended to mineralize more recalcitrant organic compounds and to perform nitrification (34). The A/O systems were designed to simultaneously remove biodegradable organic carbon, oxidize ammonia, and achieve enhanced biological phosphorus removal. The CAS systems were designed to achieve mineralization of organic carbon and to perform nitrification. The goal of the TF-CAS system was to inexpensively remove labile organic carbon using a trickling filter and then subsequently use a CAS bioreactor to further reduce organic carbon concentrations and to oxidize ammonia to nitrate. The JHB system was designed to oxidize organic nutrients while simultaneously achieving nitrification, denitrification, and enhanced biological phosphorus removal. The MBR systems were designed to achieve mineralization of organic carbon and to perform nitrification. Finally, the SBRs were designed to simultaneously achieve organic carbon removal and perform enhanced biological phosphorus removal.

Bioreactor names are designated based on their system design, plus a number if replicate systems with the same design were investigated (e.g., from CAS1 to CAS10). When samples were collected from multiple bioreactors operated under markedly different conditions within a single treatment system, it was indicated in parentheses, such as A/O4 (aerobic) and A/O4 (anaerobic). Some facilities had parallel treatment trains; when samples were collected from parallel trains, it was indicated with numbers in parentheses, such as CAS2(1) and CAS2(2). In total, there were 37 bioreactors studied from 24 facilities. Typical wastewater characteristics and treatment performance values are summarized in Table S1. All of the wastewater treatment facilities that participated in this study were able to continuously meet their discharge requirements throughout the study period (although regulations typically allow higher effluent ammonia level during cold weather periods).

Sample collection and DNA extraction.

Samples were collected monthly from July 2017 to July 2018. Samples were collected at each facility by operators on the same week and kept frozen in sterile vials until processed. For DNA extraction, bioreactor samples were fully thawed and 0.1 ml of biomass was mixed with 0.9 ml of lysis buffer (5% sodium dodecyl sulfate, 120 mM sodium phosphate buffer [pH 8]). Mixed samples then underwent three freeze/thaw cycles followed by a 90-min incubation at 70°C. DNA was then purified using the FastDNA kit (MP Biomedicals, Santa Ana, CA, USA) per manufacturer’s instructions.

Amplicon sequencing.

From each sample, amplicons of the V4 region of the 16S rRNA gene were generated with primer set Meta_V4_515F and Meta_V4_806R; amplicons were subsequently sequenced with Illumina MiSeq. Amplification and sequencing were performed at University of Minnesota Genomics Center. The primer sequences, amplification conditions, and sequencing procedures have been described previously (35).

DNA sequence analysis.

For sequence analysis, mothur was used (36), with a protocol adopted from the MiSeq standard operating procedures (SOP) (37). Paired-end reads were merged, and then sequences with one or more ambiguous bases or with a homopolymer longer than 8 nucleotides were removed. Sequences were then aligned to SILVA version 132 reference file. Aligned sequences were filtered to remove the columns that contained only gaps, and sequences with more than 300 bases or fewer than 280 bases were discarded. Chimeric sequences were also identified and discarded using Vsearch (38). Sequences were classified with the Bayesian classifier using the MiDAS reference file (39). Sequences that did not classify as bacterial or archaeal were discarded (i.e., unknown, chloroplast, mitochondria, or Eukaryota). Quality sequences were subsampled down to 4,434 sequences per sample, discarding three samples that had fewer than 4,434 quality sequences. For OTU-based analysis, the sequences were clustered into OTUs at 97% similarity using the OptiClust method (40). For phylotype-based analysis, sequences were binned into phylotypes according to their genus-level taxonomic classification (36).

Community calculations and statistical analysis.

Spearman’s rank correlation was used to determine if two parameters had a monotonic relationship. Wilcoxon rank sum tests were used to determine if a metric for one group was significantly greater than or less than that for another group. Tukey’s HSD was performed to determine whether there were pairwise differences between bioreactors. AMOVA was performed to determine if there were significant differences in the microbial community composition between bioreactors (41–43). Spearman’s rank correlation and Wilcoxon rank sum tests were performed using MATLAB R2017b (MathWorks, Natick, MA, USA), Tukey’s HSD was performed with JMP Pro 14 (SAS Institute, Cary, NC, USA), and AMOVA was completed using mothur.

Data availability.

The sequences of this study have been deposited in NCBI Sequence Read Archive under accession number PRJNA558356.