Increased Replication Rates of Dissimilatory Nitrogen-Reducing Bacteria Leads to Decreased Anammox Reactor Performance

Anaerobic ammonium oxidation (anammox) is a biological process employed to remove reactive nitrogen from wastewater. While a substantial body of literature describes the performance of anammox bioreactors under various operational conditions and perturbations, few studies have resolved the metabolic roles of community members. Here, we use metagenomics to study the microbial community within a laboratory-scale anammox bioreactor from inoculation, through performance destabilizations, to stable steady-state. Metabolic analyses reveal that dissimilatory nitrogen reduction to ammonium (DNRA) is the primary nitrogen removal pathway that competes with anammox in the bioreactor. Increased replication rates of bacteria capable of DNRA leads to out-competition of annamox bacteria, which is the key source of fixed carbon, and the loss of reactor performance. Ultimately, our findings underline the importance of metabolic interdependencies related to carbon and nitrogen-cycling within anammox bioreactors and highlight the potentially detrimental effects of bacteria that are otherwise considered to be core community members.

bioreactors under various operational conditions and perturbations, few studies have resolved the 23 metabolic roles of community members. Here, we use metagenomics to study the microbial community 24 within a laboratory-scale anammox bioreactor from inoculation, through performance destabilizations, to 25 stable steady-state. Metabolic analyses reveal that dissimilatory nitrogen reduction to ammonium 26 (DNRA) is the primary nitrogen removal pathway that competes with anammox in the bioreactor. 27 Increased replication rates of bacteria capable of DNRA leads to out-competition of annamox bacteria, 28 which is the key source of fixed carbon, and the loss of reactor performance. Ultimately, our findings 29 underline the importance of metabolic interdependencies related to carbon and nitrogen-cycling within 30 anammox bioreactors and highlight the potentially detrimental effects of bacteria that are otherwise 31 considered to be core community members.

Main Text 36
Anammox bacteria obtain energy from the conversion of ammonium and nitrite into molecular 37 nitrogen gas 1 . The only currently known bacteria to catalyze this process are Planctomycetes 2,3 , none of 38 which have yet been isolated 3,4 . In practice, anammox bacteria are employed in combination with the 39 partial nitritation (PN) process to remove ammonium from wastewaters or anaerobic digestor though side-40 streams. First, in PN, approximately half of the ammonium in solution is aerobically oxidized to nitrite. 41 Second, in anammox, both ammonium and nitrite are anaerobically converted to N2 5,6 . PN/anammox is 42 beneficial because it consumes 60% less energy, produces 90% less biomass, and emits a significantly 43 smaller volume of greenhouse gases than conventional nitrogen removal by nitrification and 44 denitrification processes 7 . 45 Even though over 100 full-scale PN/anammox processes have been installed across the globe at 46 municipal and industrial wastewater treatment plants 8 anammox bacteria have very low growth rates 47 within engineered environments and are easily inhibited by a variety of factors, including fluctuating 48 substrate and metabolite concentrations 9,10 . Furthermore, recovery from an inhibition event can take up to 49 six months, which is unacceptably long for municipalities who must meet strict nitrogen discharge 50 limits 11 . These problems are compounded by what is currently only a cursory understanding of the 51 Bioreactor performance. The performance of a laboratory-scale anaerobic membrane bioreactor 77 (described in methods) was tracked for 440 days from initial inoculation, through several performance 78 crashes, to stable and robust anammox activity ( Figure 1). Performance was quantified in a variety of 79 ways, including by its nitrogen removal rate (NRR, g-N L -1 d -1 ). Bioreactor performance generally 80 improved over the first 103 days of operation. At this point, the hydraulic residence time was reduced 81 from 48 to 12 hours and influent concentrations were reduced to maintain a stable loading rate. 82 Additional biomass from a nearby pilot-scale PN/anammox process was added on Day 145 and reactor 83 performance improved enabling influent ammonium and nitrite concentrations to be steadily increased 84 until the NRR approached 2 g-N L -1 d -1 . On Day 189 the bioreactor experienced a technical malfunction 85 and subsequent performance crash, identified by a rapid decrease in the NRR and the effluent quality. On 86 Day 203, the bioreactor was again amended with a concentrated stock of biomass and the NRR quickly 87 recovered. Influent ammonium and nitrite concentrations were again increased until the NRR reached 2 g-88 The bioreactor subsequently maintained steady performance for approximately 75 days, until Day 90 288, when effluent concentrations of ammonium and nitrite unexpectedly began to increase and nitrate 91 concentrations disproportionately decreased. Seven days later, the NRR rapidly plummeted and had since 92 no technical malfunctions had occurred this indicated that a destabilized microbial community may have 93 been responsible for the performance crash. At that time, the cause of the performance decline was not 94 understood, so the bioreactor was not re-seeded with biomass. After 50 days of limited performance, 95 concentrations of copper, iron, molybdenum, and zinc in the bioreactor influent were increased 21-24 and 96 the NRR rapidly recovered. Stable and robust bioreactor performance was subsequently maintained. From all samples, 337 genomes were binned, 244 of which are estimated to be >70% complete. 105 The genomes were further dereplicated across the six time-points into clusters at 95% average nucleotide 106 identity (ANI). This resulted in 127 representative and unique genomes (Table 1), which were used for all 107 downstream analyses. Mapping showed an average read recruitment of 76% to representative genomes 108 ( provide strong evidence to support a core anammox community ( Figure 2). The relative abundances of 117 bacteria from the dominant phyla across these three reactors are fairly similar: in each reactor the 118 anammox, along with Chloroflexi, Ignavibacteria, and Proteobacteria bacteria, compose >70% of the 119 community ( Figure 2B). 120 Due to the significantly larger genome yield and time-series analysis in this study, our 121 metagenomes had more genomes in common with each of the other reactors than the other reactors shared 122 between themselves. Nevertheless, three genomes were identified from bacteria that are closely related 123 across all three reactors: Brocadia (responsible for anammox), an unclassified Chloroflexi, and an 124 unclassified Ignavibacteria. All three of these genomes were present in our reactor during stable operation 125 on D437, and two of them (Brocadia and an Ignavibacterium) are among the ten most abundant genomes 126 at that time. In total, 21 genomes from our reactor are closely related to those from at least one of the two 127 other reactors, 17 of which are present at D437 (Supplemental Table 2 16S rRNA gene sequencing efforts, and these 38 bacteria accounted for the majority of the bioreactor 140 microbial community (Figure 3). 141 The Brocadia genus accounted for a small fraction of the bacteria in the inoculating biomass. 142 Consistent with previous research of a combination PN/anammox bioreactor with oxygen amendment, 143 members of the phyla Acidobacteria, Bacteroidetes, Ignavibacteriae, and Proteobacteria were also 144 present. During the first 100 days of bioreactor operation, Brocadia increased in relative abundance. Its 145 replication rate at D82 is high (Supplemental Table 3), which corroborates with its overall enrichment in 146 the community. Following the reactor malperformance and biomass amendment on Day 147, the 147 bioreactor became dominated by a bacterium represented by a single genome of the phylum Bacteroidetes 148 (order Sphingobacteriales). The bacterium's calculated replication rate was low on D166, and over the 149 next 100 days its relative abundance steadily declined. In contrast, the Brocadia replication rate was 150 extremely high on D166 allowing it to once again dominate the microbial community. Brocadia remained 151 dominant until Day 290, when the relative abundances of several Chloroflexi (most notably, one from the 152 class Anaerolineae) and an Ignavibacteria dramatically increased. Shortly after this shift, the bioreactor 153 experienced an unexplained period of performance decline and subsequent performance crash. During 154 this period the Brocadia replication rated dramatically declined, while the Chloroflexi replication rate 155 increased (Supplemental Table 3). These shifts in replication rates six days before a response in relative 156 abundance profiles and 12 days before a response in NRR are consistent with an instability in population 157 dynamics having directly impacted the reactor performance. 158 The relative abundances of Brocadia and the Chloroflexi, as well as their replication rates, 159 remained fairly constant over the next 44 days. After the influent media trace metal concentrations were but is highly reduced in other times ( Figure 4C). Group B dominates D328, while maintaining a similar 179 abundance in all other time points. Group C is mostly unique to D82 although a few of its members 180 remain in the reactor after the crash at low abundance. Group D bacteria show little change up to D284 181 (except a spike at D82), after which they increase in abundance. 182 It is interesting to note that the nascent anammox community is different from that of the 183 destabilized and the mature anammox communities. Because the nascent anammox community was 184 supplemented by a source inoculant biomass amendment, we cannot resolve a linear trajectory for the 185 microbial community between the initial and final states. B and D groups, while distinct, share many 186 similarities, and the majority of the genomes associated with group B were still present in the reactor on 187

Microgenomates (CPR bacteria). Group ɣ is composed entirely of Gram (-) bacteria (including Brocadia), 209
Group δ is composed of Ignavibacteria and Bacteroidetes (other members of these phyla were clustered in 210 Group ɣ). Only the Ignavibacteria of Group δ are associated with the AA group, so further analysis only 211 included those. Group ε was composed of Proteobacteria. 212 Based on the KEGG module clustering, we reconstructed the representative metabolisms of the 213 groups ( Figure 6). We used a module completeness threshold of 67% per genome, and considered it 214 representative if it was complete in >50% of its members. Group δ is not represented since it diverged 215 from group ɣ by auxotrophies in several modules ( Figure 5A, red rectangle). The Brocadia metabolism is 216 shown in Supplemental Figure 3. 217 While module completeness was used for most of the analyses, in several cases it was not 218 sufficient (e.g., overlap between modules, no module for path). In the cases of oxidative phosphorylation, 219 fermentation, carbon fixation, several amino acid synthesis pathways, and nitrogen metabolism we 220 analyzed gene presence manually. With the exception of two CPR, all of the genomes in the reactor contained genes encoding 228 assimilation of ammonia into glutamate ( Figure 7A). More than half (49) of the bacteria could reduce 229 nitrate, and the same number could further reduce nitrite to nitrogen monoxide (NO), however only 26 230 bacteria could do both steps. The most common gene encoding for nitrate reduction is narGH; niK is 231 more common than nirS (36 and 19 occurrences, respectively). The remaining steps of denitrification are 232 encoded in a smaller number of genomes. The nrxAB gene was only identified in two genomes, one of 233 which was Brocadia 234 One-step DNRA is identified in 22 genomes, predominantly with nrfAH. While ammonia 235 assimilation and nitrate reduction are fairly similar in the AA and SA bacteria, DNRA is more common in 236 AA and denitrification beyond nitrite in the SA genomes ( Figure 7C). 237 Bacteria could improve reactor performance if they remove nitrate (nitrate reducers) and excess 238 nitrite, but they could be detrimental if they compete with anammox for nitrite (DNRA and denitrification 239 from nitrite). To check for changes in the abundance of these groups, we classified bacteria by the 240 presence of genes encoding for DNRA, denitrification or nitrate reduction ( Figure 7B). Some bacteria 241 classified as denitrifiers or DNRA also encode nitrate reduction. A few genomes in the D0 sample 242 encoded both denitrification and DNRA, but their abundances were negligible. The anammox bacterium 243 has genes required for DNRA but, given the overall reactor performance, was expected to be primarily 244 performing anammox for energy generation. DNRA could potentially be used by the anammox bacteria 245 for detoxification by cycling potentially toxic excess nitrite back to ammonium where it could then 246 participate in the anammox reactions 18,25 . 247 In the inoculant source community, the nitrate reducers were the most dominant group (38%), 248 with similar amounts of denitrifiers and DNRA (26% and 25% respectively). The abundance of anammox 249 was consistent with the reactor performance ( Figure 1). The denitrifying group of bacteria decreased in 250 relative abundance to 8% at around D284. On the other hand, bacteria capable of DNRA were relatively 251 abundant throughout the reactor start up. Most notably, these bacteria dominated the reactor during its 252 destabilization, reaching 48% at D328, compared to 23% for the anammox bacteria. An increase of 253 bacteria capable of DNRA is consistent with the reactor performance data which showed a decline in the 254 amount of ammonium consumed relative to overall reactor performance. At this time period the following 255 four DNRA bacteria were highly abundant: (anamox2_sub_Ignavibacterium_album_33_16_curated, 256 anamox1_Bacteria_56_37_curated, LAC_NA06_sub_Chloroflexi_61_22_curated, and 257 LAC_NA06_sub_Chloroflexi_59_14). Three of the four are group B bacteria, and one is group D. All 258 four bacteria show an increase in relative abundance between D284 and D328. Three of the four also had 259 increased replication rates just before the onset of the crash, as mentioned above. The two other abundant 260 bacteria (apart from the anammox bacterium) are LAC_NA07_Bacteria_70_305_curated (nitrate reducer) 261 and LAC_NA07_Burkholderiales_70_312_curated (denitrifier). The former is among the most abundant 262 when the community is not SA dominated, while the latter is always one of three most abundant bacteria. 263 These 7 bacteria constitute >75% of the community at D328. fixing carbon via the Wood-Ljungdahl pathway using energy from the anammox pathway. All other 268 bacteria had genes for reduction of nitrogen compounds. To confirm that these bacteria are likely 269 autotrophs, we checked for genes conferring the ability to use inorganic electron donors. Three of these 270 bacteria had no potential electron donor and therefore were classified as heterotrophs. The remainder had 271 genes for oxidizing sulfide or hydrogen, and were classified as potential autotrophs. Of these nitrate 272 reducing bacteria (n = 8), only one was relatively abundant after D166, and increased in abundance 273 between D284 and D328. 274 LAC_NA07_Burkholderiales_70_312_curated, can fix carbon by the Calvin cycle, is a 275 denitrifier, and can possibly oxidize sulfide to sulfite (dsrAB are present in the genome). This bacterium 276 is among the most abundant at all time-points; it increased significantly in abundance between D284 and 277 D328 and the increase continued to D437. However, the replication rate of the bacterium decreased from 278 D166 onwards (Supplemental Table 3), so it is not likely competing with or destabilizing the anammox 279 bacterium. 280 281 Electron transfer. Apart from nitrogen reduction, another common anaerobic respiration pathway was 282 acetate fermentation (genes detected in 60% of the genomes). This process was much more common in 283 AA (69%) bacteria than in SA (51%) bacteria. Ni-Fe Hydrogenase was present in 31% of the genomes, 284 but was most common among the Chloroflexi of group α (87% and 48% of all occurrences of 285 hydrogenases found) ( Figure 6B). 286 The majority of bacteria in the reactor are potentially facultative aerobes (58%). All have high 287 affinity complex IV, which differed between AA and SA bacteria. In the AA bacteria, the bd type is 288 found in all aerobic members of group α (one also has a cbb3 type) and the Ignavibacteria, and the cbb3 289 type occurs mostly in Proteobacteria. For the SA bacteria, the cbb3 type is found in 24/25 aerobes and the 290 bd-type is only found in 6/25 (only in one bacterium it is the sole variant). Complex III, which is also 291 essential to aerobic respiration, was only found in 14 Proteobacteria, one Actinobacterium, and one 292 Chloroflexi. It is possible that other bacteria have an alternative Complex III 30 that cannot be found by 293 current KEGG annotations. Complexes I/II are found in nearly all of the bacteria, except CPR. Only five 294 bacteria lack the F-type ATPase; two have the V-type ATPase instead. 295 296 Central carbon metabolism. It is likely that nearly all bacteria (98%) can oxidize sugar by glycolysis 297 ( Figure 6A and E), while fewer bacteria (69%) have the pentose phosphate pathway (PPP). Acetyl-CoA 298 could be synthesized from pyruvate (90% general, 98% AA, and 81% SA), or by beta-oxidation (49% 299 general, 57% AA, and 43% SA). The majority of bacteria had the full TCA cycle (84%, or 88% after 300 excluding CPR). A possible major carbon source for the bacteria in the reactor are amino acids (aa.), with 301 95% being able to incorporate aa. into their central carbon metabolism. The most common aa. (aspartate) 302 can be converted into oxaloacetate and fed into the TCA cycle. Three aa. (serine, alanine, and cysteine) 303 can be converted into pyruvate. Of these, only cysteine is unidirectional, so aa., as a carbon source, cannot 304 be ascertained. Group α has additional genes that support a reliance on proteins for their metabolism 305 ( Figure 6B). They also have a set of peptidases, as well as multiple transporters covering all forms of aa., 306 peptides, and polyamines. 307 Some metabolic groups can use aa. as precursors for synthesis of other metabolites. Glutamate 308 and histidine can be converted to PRPP, and with glutamine to pyrimidines ( Figure 6A). Groups ɣ and ε 309 can use aspartate to synthesize NAD + , and glutamine to synthesis IMP ( Figure 6C). NAD + and IMP When combining all of the above data, we found that groups ɣ and ε both had mutualistic 345 associations with Brocadia ( Figure 8). Group ε potentially provides more metabolites to Brocadia than it 346 receives whereas groups ɑ and  are seem to gain more from Brocadia than they provide. Interestingly, 347 four members of group α and one member of group  were identified as the possible cause of the 348

destabilization. 349
By the end of the experiment (D437) when reactor performance had stabilized, members of group 350 α are the second most abundant group after Brocadia The ten most abundant bacteria at this point 351 included four members of group  and three members of group ε. Comparing these relative abundances 352 to bacterial abundances during lowest reactor performance (D328) we find that Brocadia and group ε are 353 reduced in abundance by about 50%, while groups α and  are increased by 70% and 100% respectively. In this study we present an in-depth analysis of the development of an anammox community from 357 seed to stable state (through several perturbations) in an anaerobic membrane bioreactor. By combining 358 several methodologies, we were able to gain important insights into the dynamics and interactions of 359 more than 100 species in the reactor community. 360 Previous studies have discussed a potential core anammox community 12-16 . With the exception of 361 very few studies, all such work has been conducted with single gene markers. Our metagenomic analysis 362 of an anammox community is the largest to-date and thus expands the ability to test this hypothesis. Our 363 results support the existence of a core community, while identifying factors that differentiate 364 communities. The high similarity between bacteria originating from three distinct anammox reactors 18,24 365 strongly suggests a global core anammox microbial community. In the construction of the phylogenetic 366 tree we used >3000 reference genomes originating from diverse environments. Even with the sheer 367 number and diversity of sources, the anammox community formed distinct clades at the species level. 368 More than half of the bacteria did not have species level relatives, and an additional 26% only had a 369 relative found in our anammox reactors or those from previous studies 18,24 . Together, nearly 80% of the 370 bacterial are unique to anammox reactors so that it is clear that the anammox reactor selects for a unique 371 set of bacteria. Parameters that increased the differences between communities were the species of the 372 anammox bacterium and the reactor configuration. Since both parameters relate to the same reactor 18 , we 373 cannot conclude which would have a stronger effect. 374 We identified several potential bacterial destabilizers of the anammox process. Analysis of 375 replication rate days prior to the destabilization event revealed that these bacteria increased their 376 replication rate, while Brocadia nearly ceased replication. These results imply a causative nature to the 377 change. Genes conferring DNRA capability were detected in these bacteria, which would allow them to 378 compete with Brocadia for nitrite. This supposition is consistent with the reactor performance which 379 exhibited decreased nitrogen removal and increased ammonium in the effluent during this period. The 380 dominating bacteria during reactor malperformance were heterotrophs. In full-scale anammox reactors, 381 where influent organic carbon is essentially ubiquitous, heterotrophic dominance could continue 382 indefinitely without some sort of active countermeasure. Therefore, future research should target the 383 inhibition of potential destabilizing heterotrophs. 384 A broader investigation of metabolic interdependencies within the community sheds light on the 385 stability of the anammox community. Brocadia is the source of organic material in the community, but 386 obtains essential metabolites from some members, especially Proteobacteria. This forms a basis for a 387 mutual symbiotic relationship. On the other hand, Chloroflexi, the largest group of bacteria besides 388 Brocadia, receive numerous metabolites while apparently providing few in return. They are characterized 389 by an array of extracellular proteases and amylases, likely used to breakdown the matrix formed by 390 Brocadia. Chloroflexi as a group are most associated with anammox bacteria, and form a large fraction of 391 the core community. They also account for the majority of the destabilizing bacteria. Together the results 392 point to a parasitic symbiosis. Further investigation into these relations is warranted. 393 While anammox generates sufficient organic carbon to support the growth of its co-occurring 394 heterotrophic microorganisms the tipping point between stable and unstable operation and the factors that 395 control it have not been fully identified. Input changes may be able to restore anammox activity but this is 396 just an empirical solution. Our findings improve the understanding of nitrogen-cycling within an 397 anammox bioreactor and advance the comprehensive control of this promising technology. However 398 further work is needed to elucidate the precise mechanisms that control community interactions.  All bins from both automatic and manual binning tools were input into DASTool 41 to iterate 455 through bins from all binning tools and choose the optimal set of bins. checkM was run to analyze 456 genome completeness 42 . The scaffold-to-bin file created by DASTool was uploaded back to ggKbase and 457 all scaffolds were rebinned to match the DASTool output. Each of the new bins were manually inspected 458 and scaffolds that were suspected to be falsely binned were removed. 459 After we inspected the first round of binning, we decided to improve the high coverage bins, by 460 subsampling the read file, followed by the same SOP as above 43 . I addition, refinement of the Brocadia 461 Genome bins was done with ESOMs 44 (Supplemental methods). 462 463 Post binning analysis. Unique representative genomes were determined by the dereplication tool, dRep 45 , 464 using a 95% threshold for species level clustering. Within each cluster, the representative genome was 465 chosen based on their completeness, length, N50, contamination, and strain heterogeneity. In several 466 clusters with higher heterogeneity, a second strain was chosen (Supplemental Table 1). The strain 467 threshold was set at 2% difference (but lower than 5%). 468 All the representative and strain genomes were curated by correcting scaffolding errors 469 introduced by idba_ud, using the ra2.py program 38 . Following curation, the genomes were processed 470 again for gene calling and annotation (see above for details). Analysis of replication rates at different 471 time-points was performed with the iRep program 46 using the default parameters. For the analysis of phylogenetic distance between different anammox community members, we 492 used the APE package 52 in R 53,54 to extract the distance matrix. Species level distance was set at 5% of the 493 longest measured distance on the tree. Subsequent sequence processing and data analysis were performed in-house using MOTHUR 506 v.1.39.5, following the MiSeq SOP 55,56 . In summary, sequences were demultiplexed, merged, trimmed, 507 and quality filtered. Unique sequences were aligned against the SILVA 16S rRNA gene reference 508 alignment database 57 . Sequences that did not align to the position of the forward primer were discarded. 509 Chimeras were detected and removed. Remaining sequences were clustered into operational taxonomic 510 units (OTUs) within a 97% similarity threshold using the Phylip-formatted distance matrix. 511 Representative sequences from each OTU were assigned taxonomic identities from the SILVA gene 512 reference alignment database 57 . Sequences that were not classified as bacteria were removed. Remaining KO annotations that were questionable were removed from analysis. 546 From the KO list we created a presence absence matrix (Jaccard index), and clustered using the 547 Complete method. From module completeness we created a Euclidean distance matrix, followed by 548 clustering with the ward.D method. Based on module completeness clustering we assigned genomes to 549 metabolic groups ɑ-ε. 550 For each metabolic group a representative metabolic map was created. A module completeness 551 greater than 67% in at least half of the group members, was considered as representative of the group. 552 Once the modules were selected they were drawn and connected based on metabolic KEGG maps. 553 Additional reaction, complexes and transporters were added according to KO presence (e.g a.a synthesis, 554 oxidative phosphorylation complexes, flagellar motor, etc.). 555 For nitrogen metabolism all relevant KOs were examined. For the purpose of this study, nitrate 556 reduction was consider as a separate path than denitrification/DNRA, since it could be the first step in 557 both, using the same enzymes. Denitrifying bacteria were considered as bacteria capable of full 558 conversion of nitrite to N2. DNRA bacteria were considered as bacteria capable of conversion of nitrite to 559 ammonium using the nrfAH enzymes. No partial nitrogen process in considered for this paper, although it 560 is present, according to per step analysis. performing anammox monitored over a period of 440 days. The influent did not contain nitrate, so 716 influent nitrate is not plotted. The nitrogen removal rate (NRR), is plotted against the secondary y-axis. 717 Sampling time points for metagenomes are indicated with purple stars below the x-axis. sequencing results falling within three days of each other have been merged. "Unmatched" includes the 763 OTUs and genomes that were not able to be matched across the two sequencing platforms. The similar 764 relative abundance profiles at shared time points across metagenomic and 16S rRNA gene sequencing 765 platforms (highlighted in the black boxes) provided us with the confidence to extrapolate high-resolution 766 relative abundance profiles of our representative genomes from our 16S rRNA gene sequencing efforts. latter is not presented here since it shares all paths with group ɣ and only differs by auxotrophies. (A) 808 Metabolic map of paths that are common to all bacteria in the reactor (except Microgenomates and 809 Brocadia sp.). The vast majority of bacteria in the reactor are heterotrophs, capable of carbohydrate-based 810 metabolism (glycolysis, pentose phosphate pathway) and amino acid-based metabolism. Some bacteria 811 can respire oxygen, but can also ferment (acetate/alanine). (B) Paths unique to group ɑ. The bacteria have 812 genes for hydrogen oxidation, supporting anaerobic growth, as well as genes for oxidative 813 phosphorylation with cytochrome BD complex. These bacteria have a cassette of extracellular proteases 814 and decarboxylases, paired with a wide array of transporters. The bacteria are also potentially capable of 815 synthesizing long chain isoprenoids. (C) Paths found in Gram (-) bacteria (groups ɣ, δ, and ε). Most paths 816 are related to fatty acid and lipid synthesis. Several important precursors (chorismate and IMP) can 817 potentially be synthesized by these bacteria. Motility is also a common feature in these bacteria (via a 818 flagellar motor) (D) Unique paths of group ε (Proteobacteria). This group has the potential to synthesize 819 multiple vitamins and cofactors (biotin, pyridoxal, glutathione, etc.), as well as several aa. (tyrosine, 820 phenylalanine, proline). Another unique feature is the multiple secretion systems present in the bacteria.

821
(E) Metabolic profile of CPR bacteria (Microgenomates). The bacteria are obligate anaerobes that ferment 822 pyruvate. They can only utilize carbohydrates as carbon source. Some of the bacteria in this group might 823 also be able to synthesize long chain isoprenoids, in the same path as group ɑ. path was assigned to each genome for the purpose of this analysis. Since nitrate reduction is also 833 considered a first step in denitrification and DNRA is was assigned only when other paths were not 834 present. Bacteria with no complete metabolic path are depicted in light grey. Anammox is the dominant 835 nitrogen metabolic path at Days 82, 284, and 437. This matches the reactor performance monitoring 836 (Figure 1). At times when the source community is predominant (Days 0 and 166), nitrogen reduction is 837 the most common metabolic path, followed by DNRA. During the period of reactor destabilization (Day were assigned according to absence of ability to synthesis a metabolite and connect to all groups that do 849 have the ability (meaning there is redundancy in arrows). The arrowhead points at the group that receives 850 the metabolite. The width of the arrow is proportional to the ratio of metabolites of a given type that are 851 provided; amino acids -20 metabolites; Peptides -deduced from proteases and transporters ( Figure 6B); 852 Vitamins/Co-factors -10 metabolites; Lipids/Fatty acids -7 metabolites. The size of each group is 853 proportional to their relative abundance at Day 437. Group β is not shown since the assumption is that its 854 members obtain all of their nutrients and metabolites from a their host. Overall, groups ɑ and ẟ receive 855 the most metabolites and group ε the least. Group ẟ has the highest number of aa. synthesis auxotrophies 856 and can potentially acquire these from many other community members. Group ε has only a single 857 auxotrophy in vitamin/Co-factor synthesis while most other groups have multiple auxotrophies (group ɑ 858 capable of only a single metabolite). Brocadia sp. is the only bacterium capable of vitamin B12 synthesis. 859