Long-term multi-meta-omics resolves the ecophysiological controls of seasonal N2O emissions

16 17 The potent greenhouse gas n itrous oxide (N 2O) originates primarily from natural and 18 engineered microbiomes. Emission seasonality is widely reported while the underlying 19 metabolic controls remain largely unresolved, hindering effective mitigation. We use 20 biological wastewater treatment as tractable model ecosystem over nearly two years. Long -21 term metagenomic-resolved metaproteomics is combined with ex situ kinetic and full-scale 22 operational characterization. By leveraging the evidence independently obtained at multiple 23 ecophysiological levels, from individual genetic potential to actual metabolism and emergent 24 community phenotype, the cascade of environmental and operational triggers driving N 2O 25 emissions is resolved. We explain the dynamics in nitrite accumulation with the kinetic 26 unbalance between ammonia and nitrite oxidi sers, and identify nitrifier denitrification as the 27 prime N2O-producing pathway. The dissolved O2 emerged as the key actionable parameter for 28 emission control. This work exemplifies the yet-to-be-realized potential of multi -meta-omics 29 approaches for the mechanistic understanding and ecological engineering of microbiomes, 30 ultimately advancing sustainable biotechnological developments. 31 32

microbial ecophysiology; microbial communities; nitrous oxide; wastewater 33 treatment; multi-meta-omics 34 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 3

35 36 The yearly anthropogenic emissions of nitrous oxide (N2O), currently the third most important 37 greenhouse gas, are projected to increase by 50% in the coming 50 years if no mitigation 38 strategies are employed 1. N2O is mainly pro duced by m icrobial communities in natural, 39 managed and engineered ecosystems 2. Yet, the mechanisms governing biological N 2O 40 emissions remain largely unknown . The main challenge lies in the coexistence of nitrogen -41 converting guilds in complex microbiomes, each emitting N2O under a range of complementary 42 conditions that alternate or overlap in most ecosystems (e.g. alternating oxic-anoxic conditions 43 in wastewater treatment plants 3 and sea sediments 4; substrate concentration gradients in 44 oceans 5, soils 6 and wastewater treatment biofilms 7). In general, high ammonium (NH4+) and 45 oxygen (O 2) concentration s stimulate N 2O production through hydroxylamine (NH 2OH) 46 oxidation by ammoni a-oxidising bacteria (AOB) , while h igh nitrite (NO 2-) and low O2 47 concentrations enhance the nitrifier denitrification pathway 8 (Fig. 1A). High NO2- and high O2 48 concentrations result in N 2O accumulation from imbalanced denitrification by heterotrophic 49 denitrifying bacteria (DEN) 8 (Fig. 1A). Seemingly ubiquitous is the strong seasonality of N2O 50 emissions in many natural and managed environments, such as oceans 9,10, soils 11–13, lakes 14,15 51 and rivers 16, and engineered systems such as wastewater treatment plants 17–24 (WWTPs, 52 summarized in Table S1). This indicates that seasonally-impacted macroscopic factors directly 53 influence biological N 2O turnover. Yet, s tudying the interactions between environmental 54 conditions, complex microbiome dynamics and N 2O emissions, and capturing the underlying 55 ecological principles is inherently challenging. To this end , we use biological wastewater 56 treatment as a more tractable model ecosystem, as the N 2O seasonality is well -represented, 57 while other variables (e.g. aeration, biomass concentration ) are controlled or extensively 58 monitored 25. 59 60 Most WWTPs emit the majority of their yearly N2O during a winter or spring peak lasting 3-4 61 months, with simultaneous NO 2- accumulation 17,21–24,26 (Table S1). Similarly, higher N 2O 62 emissions during colder seasons are widely reported for oceans 10, soils 12,13, and lakes 14. Low 63 or increasing temperatures have been hypothesized as the underlying causes for the seasonal 64 N2O emissions, but a clear correlation is often missing 10,13,14,18,19,27,28. The immediate effect of 65 diverse environmental and process parameters on the N2O production rates of AOB and DEN 66 largely explain the short-term N2O dynamics in WWTPs 3,29 and natural environments 5,6,30,31, 67 but fail to describe the widely observed seasonality. Emblematic is the reported higher N 2O 68 production by AOB at high temperatures 32, while most seasonal emissions occur in winter. 69 Broadly applied c orrelation analyses between N2O and environmental and operational 70 parameters have proved insufficient to explain seasonal emissions in WWTPs 18,24,33, oceans 71 9,10, soils 11–13 and freshwater systems 14–16. Despite the evident central microbial role in N2O 72 conversions, most studies do not take potential seasonal dynamics of the microbiome’s 73 metabolism into account, likely overlooking key mechanisms linking environmental triggers 74 and emissions. A delay between triggers , metabolic adaptations and emergent phenotype is 75 expected in slow-growing natural and WWTP communities 28. Only few studies investigated 76 microbial dynamics during seasonal nitrogen oxides peaks in WWTPs with seemingly 77 contradicting results . Seasonal NO 2- and N 2O accumulation events have been attributed to 78 decreased nitrite-oxidising bacteria ( NOB) 16S rRNA gene abundances 19,23 and increased 79 difference between AOB and NOB activity 17,22, while in other instances no seasonal 80 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 4 fluctuations were observed in the nitrifying community 34. To date , the operational and 81 metabolic mechanisms controlling seasonal N2O emissions remain largely unknown, hindering 82 effective mitigation. 83 84 We combine long-term metagenomic-resolved metaproteomic analyses with ex situ kinetic and 85 full-scale process characterizations to address the mechanistic gap in seasonal N2O emissions. 86 The cascade of environmental and operational triggers underlying N2O emissions is resolved 87 by l everaging the evidence obtained at multiple ecophysiological levels , from individual 88 genetic potential to actual metabolism and emergent community phenotype. We identify 89 nitrifier denitrification as the prime N 2O-producing pathway, and the dissolved O 2 as the 90 central operational parameter to minimize emissions . This work exemplifies the yet-to-be-91 realized potential of multi-meta-omics approaches to inform ecologically-driven strategies for 92 the management and engineering of microbiomes, ultimately advancing sustainable 93 biotechnological developments. 94 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 5

95 96 Signature metabolite accumulation profiles 97 98 The ecophysiological response of N 2O-emitting complex microbial communities to seasonal 99 environmental and operational dynamics was studied using the Amsterdam -West wastewater 100 treatment plant (WWTP) as model ecosystem (Fig. 1A -B). The monitoring and sampling 101 period lasted eighteen months and covered two highly comparable N2O emission peaks ( Fig. 102 2). The peaks occurred during periods with low water temperatures, namely Feb – May 2021 103 and Nov 2021 – Mar 2022, and were preceded by the sequential accumulation of NH 4+, O2, 104 and NO2- (Figs. 2 and S2). The same trend was followed in the five years prior to this study 105 (data not shown). Central to the plant operation is the control of the dissolved O 2 (DO) 106 concentration as a function of the residual NH4+ concentration in the aerated compartment. To 107 counteract the temperature -induced nitrification rate reduction, and consequent NH 4+ 108 concentration increase, the weekly average DO concentration was increased from 1 up to 109 almost 3 mg O 2·L (Fig. 2). In spite of this , O 2 remained the rate-limiting substrate for 110 nitrification during low temperature periods with high N2O emissions, as evidenced by a lower 111 O2/NH4+ ratio in the aerated compartment compared to warmer periods with l ow N 2O (Fig. 112 S3). Following the increase in DO , the average NO2- concentration in the pooled effluent 113 rapidly increased up to 1.1 mg N·L -1. Finally, N2O started to accumulate, reaching maximum 114 daily rates of 110 (1 st peak) and 101 kg N·d -1 (2nd peak) (Figs. 2 and S2). The delay between 115 the maximum DO concentration and the maximum N2O emission rate ranged between six and 116 seven weeks for both peaks (Fig. 2), consistent with the imposed average sludge retention time 117 of 11-15 days . Statistically, NO 2- strongly correlated with the O 2 concentration (Pearson 118 correlation coefficient of 0.8), and N2O with NO2- (correlation coefficient 0.7), while they only 119 weakly correlated with all other parameters including the temperature (Fig. S4 and Table S2). 120 121 Figure 1. Schematic representation of the nitrogen cycle, experimental approach and obtained datasets. (A) Nitrogen 122 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 6 conversions in the biological nitrogen removal process and respective enzyme complexes. Ammonia-oxidising bacteria (AOB) 123 aerobically oxidise ammonium (NH4+) to hydroxylamine (NH2OH) with the ammonium monooxygenase (AMO), NH2OH to 124 nitric oxide (NO) with the hydroxylamine oxidoreductase (HAO), and NO to nitrite (NO2-) with a yet unknown enzyme. AOB 125 can biologically produce N2O through the oxidation of NH2OH with cytochrome P460 (cyt P460) or through the reduction of 126 NO – produced from NH 2OH oxidation or nitrifier denitrification (NO 2- reduction with the nitrite reductase NIR) – with the 127 nitric oxide reductase (NOR) (dotted arrows). Nitrite-oxidising bacteria (NOB) aerobically oxidise NO2- to nitrate (NO3-) with 128 the nitrite oxidoreductase (NXR). Normally under anoxic conditions, denitrifying bacteria (DEN) reduce NO 3- to NO2- with 129 the membrane-bound or periplasmic nitrate reductase (NAR, NAP), NO2- to NO with NIR, NO to N2O with NOR and N2O to 130 N2 with the nitrous oxide reductase (NOS). Some DEN perform only some steps of the denitrification pathway while others 131 perform the entire pathway. (B) Overview of the methodological approach adopted in this study for the eighteen-months 132 characterization of a full -scale WWTP to resolve the microbial mechanisms underlying seasonal N 2O emissions. Sludge 133 samples were used for metagenomics (6 samples), metaproteomics (12 samples) and ex situ activity tests at 20 oC (26 samples). 134 (Created with BioRender.com.) 135 136 Figure 2. Performance of the wastewater treatment plant (WWTP) monitored during nearly two years (Oct 2020 – Jul 137 2022). Weekly average parameters at the WWTP , from back to front (light green to dark blue): concentration of NH4+ and 138 dissolved O2 in the nitrification compartment (left axis), pooled effluent NO2- concentrations (right axis), N2O emission rates 139 measured in the off-gas from all reactor compartments (right axis). The water temperature inside the reactor is represented on 140 the right axis (symbols). All metabolites were measured in a single biological nutrient removal lane of the WWTP, except the 141 effluent NO2- (seven lanes pooled together). Occasional sharp NH 4+ peaks were caused by outliers on rainy days (Fig. S2). 142 The scheme above the plot represents the sampling time points for metagenomic (DNA), metaproteomic (protein) and ex situ 143 activity tests (bioreactor). 144 145 Maximum nitrogen metabolites conversion rates 146 147 To quantify seasonal changes in the microbiome metabolic potential, we estimated every 148 second week the maximum oxidation and reduction rates of the main nitrification (i.e. NH4+ 149 and NO 2-) and denitrification (i.e. NO3-, NO 2- and N 2O) intermediates, respectively. T he 150 maximum NH4+ oxidation rate almost always exceeded the NO2- oxidation rate, with their 151 difference being the highest during the seasonal full-scale metabolite accumulation peaks (Fig. 152 S5). No clear seasonality emerged in the NO3-, NO2-, and N2O maximum reduction rates, and 153 the N2O reduction capacity was 1.4 to 2.1-fold higher than all other nitrifying and denitrifying 154 rates (Fig. S5). 155 156 Genome-resolved taxonomic diversity 157 158 The WWTP metagenome was sequenced at six time points to follow the dynamics in microbial 159 composition and functional potential, and to serve as database for the metaproteomic analysis 160 (Fig. 1B). Combined short-read (two samples; average 147 million reads per sample) and long-161 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 7 read DNA sequencing (five samples , one of which also sequenced with short -reads; average 162 4.3 million reads per sample) resulted in 143 Gbp data, after quality filtering and trimming. A 163 total of 349 high-quality metagenome-assembled genomes (HQ MAGs, ≥ 90% completeness 164 and ≤ 5% contamination) (Fig. 3, Supplementary Data 1 ) were obtained . The 89 MAGs 165 generated from the five long -read samples were dereplicated with the HQ MAGs from 166 Singleton et al. 35 at 95% average nucleotide identity of open reading frames to increase the 167 genome-resolved read coverage. From the final 349 HQ MAGs, 44 were unique to our dataset, 168 268 were unique to the dataset of Singleton et al. 35, and 37 overlapped between both datasets 169 (Fig. S6). Overall, the HQ MAGs covered 31 phyla and 272 different genera, and included two 170 archaeal species (only bacterial MAGs are represented in Fig. 3). The full 16S rRNA gene was 171 identified in 347 (99.4%) MAGs. The relative abundance of the individual MAGs showed no 172 marked seasonal trend and little variation over the six time points (Fig. S7 and Supplementary 173 Data 1). We therefore discuss the average of their relative abundance among all samples. The 174 two most abundant MAGs belonged to the Ca. Microthrix (4.0%) and Nitrospira (2.7%) genera 175 (Fig. 3). All other MAGs had an average relative abundance lower than 1% . The majority of 176 the non-nitrifying MAGs contain ed at least one denitrification gene (DEN, 304 ) (Fig. 3, 177 Supplementary Data 2). 51 MAGs had the genetic potential to perform dissimilatory nitrite 178 reduction to ammonium (DNRA, containing the nrfAH genes), 46 of these also had at least one 179 denitrification gene (Fig. S14, Supplementary Data 2 ). Seven MAGs harboured the amoABC 180 genes (AOB) and eight harboured the nxrAB genes (NOB), most of these also had at least one 181 denitrification gene, mainly nir and nor encoding the NO 2- and NO reductases, respectively 182 (Fig. S 14, Supplementary Data 2 ). Neither complete ammonia-oxidising (comammox) nor 183 anaerobic ammonia-oxidising (anammox) MAGs were found in the metagenomes. 184 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 8 185 Figure 3. Phylogenetic tree of the 347 bacterial high-quality MAGs extracted from activated sludge (the only two archaeal 186 MAGs are not represented) . From the inner to the outer circle: (i) circular phylogenetic tree with the identification of key 187 activated sludge genera Nitrosomonas, Nitrospira, Ca. Accumulibacter and Ca. Microthrix; (ii) identification of ammoni a-188 oxidising bacteria (AOB, containing amoABC genes, dark blue), nitrite-oxidising bacteria (NOB, containing nxrAB genes, light 189 blue) and denitrifying organisms (DEN, non-AOB and non-NOB MAGs harbouring at least one denitrification gene, yellow) . 190 Some of the AOB and NOB MAGs also contained one or more denitrification genes (Supplementary Data 2); (iii) average DNA 191 relative abundance of each MAG in the community; (iv) average protein relative abundance of each MAG in the community; 192 (v) identification of the six most abundant phyla. 193 194 195 Metaproteomic-based functional profile 196 197 The dynamics in protein expression of the entire microbial community across twelve samples 198 was assessed by shotgun metaproteomics. We used the protein expression as proxy for active 199 metabolisms and to estimate the protein-based relative abundance of each MAG. In total, 3868 200 unique protein groups were detected, and 1884 had at least two unique peptides (accounting 201 for 44 ± 1% of the total mass-normalized spectral counts) . 1105 of the identified proteins 202 (accounting for 68 ± 1% of the two unique peptides filtered normalized spectral counts) 203 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 9 uniquely matched with a single protein predicted in the metagenome (including all MAGs and 204 unbinned sequences). The remaining 779 proteins (accounting for 32 ± 1% of the two unique 205 peptides filtered normalized spectral counts) matched multiple highly similar proteins and 206 could not be linked to a single MAG, yet could be functionally and taxonomically annotated at 207 the genus level . Out of the 349 HQ MAGs, proteins from 143 MAGs (101 genera) were 208 detected (Supplementary Data 1). The HQ MAGs covered 39 ± 1 % of the total protein pool, 209 higher than the 28 ± 4 % coverage of the total community DNA (Fig . 4A). On average, t he 210 relative abundance of key activated slud ge taxa (e.g. Ca. Microthrix, Ca. Accumulibacter, 211 Nitrosomonas and Nitrospira,) differed up to 20 -fold between the metagenomic and 212 metaproteomic approaches (Fig. S12). For example, the AOB:NOB ratio was 0.1 in the 213 metagenome and 3.6 in the metaproteome (discussed in Supplementary Section 6). 214 Taxonomically, the diversity was greatest within the DEN guild (proteins from 124 MAGs 215 were detected) with no clear dominant MAG (Fig . 4B). Owing to th is high diversity, many 216 DEN organisms were present in too low abundance to be recovered as MAGs even at the 217 already high sequencing depth employed here (20 -25 Gbp per sample) . Consequently, DNA 218 sequences from many DEN remained in the unbinned portion of the metagenomes, resulting in 219 the majority of the detected denitrification enzymes , namely nitrate, nitrite and nitrous oxide 220 reductases being assigned to the unbinned fraction ( Fig. S15). Proteins from a ll seven AOB 221 and four NOB MAGs were detected in the metaproteome. The AOB consisted entirely of 222 Nitrosomonas MAGs, and were dominated by one MAG (Fig. 4B). NOB were dominated by 223 a Nitrospira and a Chloroflexota MAG belonging to the Promineofilaceae family (Fig. 4B), 224 but the alpha- and beta -subunits of the nitrite oxidoreductase (Nxr A and NxrB ) w ere only 225 expressed by Nitrospira and Ca. Nitrotoga (Fig. S15). Almost all detected nitrifying enzymes 226 belonged entirely to the recovered MAGs, highlighting the nearly full coverage of the active 227 nitrifying community by the MAGs (Fig. S15). Throughout the monitoring period, the relative 228 proteomic abundance of DEN hardly fluctuated, and t he AOB and NOB guilds fluctuated 229 similarly over time (Fig. 4C). The maximum guild-specific fold change in the proteome was 230 1.1 (DEN), 1.8 (AOB) and 2.5 (NOB). Overall, there were no major shifts in the MAG -based 231 composition of each guild, at both DNA and protein level ( Figs. S8-S10), and there were no 232 significant correlations between protein-level taxa abundance and WWTP performance (Table 233 S4). 234 235 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 10 236 Figure 4. MAG-based functional guild distribution in the metagenomes and metaproteomes of the activated sludge. (A) 237 Average relative abundance of denitrifying bacteria (DEN, non-AOB and -NOB MAGs containing at least one denitrification 238 gene, yellow), nitrite-oxidising bacteria (NOB, containing nxrAB genes, light blue), ammonia-oxidising bacteria (AOB, 239 containing amoABC genes, dark blue), other metagenome-assembled genomes (dark grey) and unbinned sequences (light grey) 240 in the total metagenome (DNA) and metaproteome (Protein) of the activated sludge. Some of the AOB and NOB MAGs also 241 contained one or more denitrification gene s (Supplementary Data 2). The error bars represent fluctuations within six (DNA) 242 and twelve (protein) activated sludge samples taken throughout eighteen months. (B) MAG-based composition of the DEN, 243 NOB and AOB guilds. The most abundant genera in the DEN ( Ca. Accumulibacter and Ca. Competibacter), NOB 244 (unidentified Promineofilaceae genus, Ca. Nitrotoga, Nitrospira ) and AOB (Nitrosomonas) guilds are highlighted. (C) 245 Temporal fluctuations in the relative protein abundance of the DEN (yellow), NOB (light blue) and AOB (dark blue) guilds. 246 The error bars represent standard deviations between technical duplicates and are all smaller than the symbols. 247 248 249 Unbalanced nitrification drives seasonal nitrite accumulation 250 251 The net accumulation and potential emission of any nitrogen intermediate results from the 252 unbalance between its production and consum ption rates . Nitrite, a central metabolite 253 exchanged between AOB, NOB and DEN (Fig. 1A), always accumulated prior to the N 2O 254 peaks (Fig. 2) . To understand the NO 2- flux balance dynamics , we focused on the DNA, 255 expressed proteins and ex situ activity ratios of NO2--producing and -consuming guilds. At all 256 levels (genomic, proteomic and kinetic), the DEN guild did not display significant seasonal 257 dynamics (Figs. 4 C, S8 and S19). Contrastingly, the (un)balance between AOB (NO2- 258 producer) and NOB (NO2- consumer) fluctuated the most during the monitored period . The 259 ratio between the total abundances of AOB and NOB, both at DNA and protein level, was up 260 to 3-fold higher during periods of high effluent NO 2- concentrations, compared to the rest of 261 the year (Fig. 5A-B). At individual protein level, including MAG and unbinned proteins, the 262 ratio between the expression of the key NH4+-consuming enzyme (represented by the beta -263 subunit of the ammonia monooxygenase – AmoB) and NO2--producing enzyme (represented 264 by the hydroxylamine oxidoreductase – Hao) of AOB relative to the catalytic subunit of the 265 NO2- oxidoreductase of NOB ( NxrA) were also higher (Fig. 5C, Supplementary Data 3 ). 266 Consistently, the ratio between the maximum NH4+ and NO2- oxidation activities was larger 267 during high NO2- concentration periods (Fig. 5D). 268 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 11 269 Figure 5. Genomic, proteomic and maximum activity fluctuations of AOB and NOB in activated sludge during periods 270 of high and low nitrite accumulation. Left axes: (A) Ratio between the total relative DNA abundance of ammonia- (AOB) 271 and nitrite-oxidising bacteria NOB (circles). (B) Ratio between the total relative protein abundance of AOB and NOB (circles). 272 (C) Ratios between the relative abundance of NO2--producing and -consuming enzymes of AOB and NOB, respectively: beta-273 subunit of the ammonia monooxygenase (AmoB) divided by the catalytic subunit of nitrite oxidoreductase (NxrA) (diamonds); 274 and hydroxylamine oxidoreductase (Hao) divided by NxrA (x4, circles). The enzyme abundances include the proteins 275 belonging to the MAGs and the unbinned fraction. The error bars in all protein ratios were propagated from standard deviations 276 of technical duplicates and some are smaller than the symbols . The respective enzymatic conversions are represented on the 277 right. (D) Ratio between the maximum ex situ NH4+ and NO2- oxidation rates measured at 20 ˚C (circles). Right axes: (A-D) 278 Weekly average NO2- concentration in the effluent (seven parallel lanes pooled together, grey area). 279 280 Overexpressed nitrifier denitrification during N2O accumulation 281 In analogy to nitrite, we used ratios between the relative abundance of enzymes directly or 282 indirectly producing and consuming N2O as proxy for the N2O flux balance. The total enzyme 283 abundances include MAG and unbinned protein abundances (Supplementary Data 3). The 284 seasonally accumulated NO2- can be reduced to N2O by both AOB and DEN, sequentially using 285 the Cu- (NirK) or cd1-type (NirS) NO 2- reductases and the nitric oxide reductase (Fig. 1A). 286 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 12 Here, NirK and NirS were exclusively expressed by nitrifiers and DEN, respectively (Fig. S15). 287 Four Nitrosomonas (AOB) and one Nitrospira MAG (NOB) accounted for most of the NirK 288 expression (75% and 17%, respectively) (Fig. S15). Within the nitrifying community, t he 289 relative abundance of NirK over the key AOB enzymes AmoB and Hao was the highest during 290 periods of high NO2- and N2O accumulation (Fig. 6A). The ratio of total relative abundance of 291 NirK over the competing NO2--oxidising NxrA (NOB) and NO 2--reducing NirS (DEN) 292 followed a similar trend (Fig. 6B). NosZ is the only known N2O-reducing enzyme, and the ratio 293 NirK/NosZ clearly reflected the seasonal dynamics , being higher during seasonal peaks (Fig. 294 6C). Similarly, yet to a significantly lower extent, also the ratio between the hydroxylamine 295 (NH2OH) producing AmoB and consuming Hao and CytP 460 (Fig. S18), and the ratio 296 NirS/NosZ (Fig. S19C) displayed some seasonality. The here employed protein extraction 297 protocol does not allow for the quantification of membrane-bound proteins, such as the nitric 298 oxide reductases, which were therefore not included in the discussion. 299 300 Figure 6. NirK overexpression relative to other nitrogen enzymes during periods of high NO2- concentrations and N2O 301 emissions. Left axes (symbols): (A) NirK vs. other AOB enzymes. Ratio between the total relative abundance of NO2--302 consuming NirK and the other key AOB enzymes Hao (circles) and AmoB (triangles). (B) NirK vs. competing NO 2- 303 consuming enzymes. Ratio between the total relative abundance of NO2--consuming NirK and the NO 2- competing NxrA 304 (circles, NOB) and NirS (x50, triangles, DEN). (C) NirK in N2O balance. Ratio between the total relative abundance of NirK 305 (producing the N 2O precursor NO) and the only known enzymatic N2O-sink N2O reductase (NosZ) (circles). The enzyme 306 abundances include the proteins belonging to the MAGs and the unbinned fraction. The error bars in the protein ratios were 307 propagated from standard deviations of technical duplicates. All enzymatic conversions are schematically represented on the 308 right. NirK is expressed by both AOB and NOB, but the activity and function of the enzyme in NOB are yet unknown. Right 309 axes: (A-C) Weekly average NO2- concentration in the effluent of the WWTP (seven parallel lanes pooled together, grey area) 310 and N2O emission rates measured in the off-gas from all the reactor compartments in one lane at the WWTP (grey line). 311 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 13

312 313 We postulate that the seasonal accumulation of NO2- and subsequent emissions of the potent 314 greenhouse gas N2O at a full -scale WWTP are related to fluctuations in the balance of key 315 nitrogen-converting populations, rather than their individual abundance or activity. No major 316 changes in the DNA and protein composition, nor significant correlations with plant 317 performance, were observed throughout eighteen months of operation. This is consistent with 318 previous metagenomic and 16S rRNA gene amplicon sequencing reports in WWTPs 36–39. The 319 microbiome was dominated by a taxonomically diverse DEN community ( 74% of the binned 320 community proteome), in line with most genomic and transcriptional analyses of conventional 321 WWTPs 19,40,41. While the high DEN abundance may have masked fluctuations at guild level, 322 the absence of significant changes at the activity and individual protein level further supports 323 the DEN stability. Instead, the DNA and protein abundance s of the nitrifying community, 324 dominated by one AOB and two NOB MAGs, fluctuated over time, yet not consistently with 325 the observed nitrogen oxides accumulation dynamics. This aligns with most studies reporting 326 limited to no correlation between AOB and NOB 16S rRNA gene abundances and seasonal 327 nitrification failures 34, or AOB and NOB conversion rates and N2O production 22. Only few 328 studies observed a correlation between increased N2O emissions and increased relative AOB 329 abundances (16S) 42, AOB ex situ activities 43, or decreased NOB abundances (16S) 19,23. Yet 330 evidence remains sparce and seemingly conflicting, ultimately hindering mechanistic 331 generalizations. This lack of general consensus resides in the fundamental dependency between 332 metabolite dynamics and the trade-off between their production and consumption rates (i.e. the 333 balance between the producing and consuming guilds), rather than their individual magnitudes. 334 335 Against a relatively stable DEN community, featuring a fairly constant nitrite production and 336 reduction potential, we identified the unbalance between AOB (NO2- producer) and NOB (NO2- 337 consumer) as the primary cause for seasonal nitrite accumulation. During the nitrite peaks 338 preceding the N2O ones, a higher ratio of AOB over NOB was observed at genomic, proteomic 339 and kinetic levels. To date, only Bae et al.22 quantitatively linked N2O emissions with increased 340 AOB/NOB ex situ activity ratios in an otherwise stable nitrifying community based on 16S 341 rRNA gene sequencing. Gruber et al.23 observed stable AOB but lower NOB and filamentous 342 bacteria 16S rRNA gene abundances during winter N2O emission s, and hypothesized a 343 selective NOB washout due to compromised floc integrity. Here, the fluctuations in the sludge 344 settleability (representing floc integrity) and in the DNA and protein abundances of Ca. 345 Microthrix (filamentous bacteria) did not follow the full-scale metabolite profiles, nor the NOB 346 abundance or the AOB/NOB ratio (Figs. S2, S11 and Table S2). The known higher sensitivity 347 of NOB to the toxic free ammonia and nitrous acid compared to AOB 44–46 has also been 348 suggested as potential cause for nitrite accumulation 45. However, in our case, the estimated 349 concentration of free ammonia (0.03 mg N·L -1) and nitrous acid (0.001 mg N·L -1) were far 350 below the NOB toxicity thresholds (Tables S6-S7) 44–47. Instead, we argue that the unbalanced 351 AOB/NOB ratio results from a cascade of separate environmental and operational perturbations 352 differentially impacting their respective growth rates (Fig. 7). The decrease in temperature 353 reduces both AOB and NOB growth rates, and may alone promote the selective washout of the 354 slower-growing NOB (as estimated in this work and consistent with literature values; Table 355 S8, Fig. S21). In addition, reduced AOB rates lead to the accumulation of ammonium, with the 356 operationally undesired worsening of effluent quality. In response, most WWTPs increase the 357 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 14 operational dissolved O 2 set point to promote nitrification. The increased availability of 358 ammonium selectively favours AOB, while, in principle, the increase in dissolved O2 positively 359 impacts the growth rate of both AOB and NOB. However, the reported lower AOB apparent 360 affinity for O2 in activated sludge 48–51 likely favours AOB over NOB, further enhancing the 361 initial differential temperature impact on their growth rates. Ultimately, nitrite accumulation is 362 the result of the progressive relative enrichment of AOB over NOB. To test our hypothesis, we 363 developed and implemented a simple mathematical model based on the experimentally 364 estimated kinetic parameters and literature-derived stoichiometric parameters (Tables S9-S12). 365 The model reproduced all observed seasonal metabolite s peaks induced by decreasing 366 temperatures and consequent increase in ammonium and operational dissolved O 2. The 367 simulations also captured the progressive relative biomass increase of AOB over NOB (Fig. 368 S23). These results strongly indicate that the sequential seasonal nitrogen oxides peaks result 369 from a cascade of distinguishable events, where temperature is the initial trigger but not the 370 sole direct cause, as commonly hypothesized . The absence of a single parameter correlating 371 with nitrite and subsequent N 2O emissions likely explains the difficulties of past studies to 372 identify direct correlations 18,24,33. Importantly, the dissolved O2 concentration emerged as the 373 central operational parameter to act upon, and we posit that the AOB/NOB unbalance may be 374 largely prevented by anticipating in time, i.e. before measurable NH 4+ accumulation, the 375 operational O2 increase. 376 377 The last metabolite to accumulate along the reconstructed ecophysiology cascade is N2O. High 378 nitrite concentrations are well -known to lead to N 2O emissions through both nitrifier and 379 heterotrophic denitrification 3, yet the dominant pathway underlying seasonal N 2O emissions 380 remains unclear 21,24,43. We use the nitrite reductase s (NirK and NirS) as proxy for N 2O 381 production, and their genome-resolved taxonomy to differentiate between nitrifier and 382 heterotrophic denitrification. Considering the fast turnover of NO 8, the use of Nir allows to 383 overcome the challenges in detecting the membrane-bound hydrophobic nitric oxide reductase 384 in metaproteomic analyses 52,53. Unbalanced heterotrophic denitrification was excluded as the 385 main N2O producing pathway during the seasonal emissions owing to t he relatively constant 386 ratio between NirS and NosZ, both exclusively expressed by DEN, and their rates. The nitrite 387 reductase NirK was exclusively expressed by nitrifiers, primarily by AOB, so it was used as 388 proxy for nitrifier denitrification. NOB Nitrospira contributed to about one fifth of the total 389 detected NirK, but its activity and function remain unknown 54–56. A marked increase in the 390 ratio of NirK over other AOB enzymes (AmoB and Hao) and the competing NO2--consuming 391 enzymes (NxrA from NOB and NirS from DEN) was observed during the seasonal nitrogen 392 oxide peaks. The higher expression of NirK was likely induced by the seasonally increased 393 nitrite concentrations 57,58, and suggests an increased relative nitrite flux towards nitrifier 394 denitrification rather than nitrite oxidation or heterotrophic nitrite reduction. Emissions also 395 coincided with periods in which O 2 was identified as the metabolically limiting substrate for 396 AOB (i.e. lower O2/NH4+ ratios compared to the rest of the year), likely forcing AOB to resort 397 to nitrifier denitrification as additional electron sink 59,60. The observed slight imbalance 398 between hydroxylamine-producing AmoB and -consuming Hao and Cyt P460 makes it here 399 tempting to speculate that hydroxylamine accumula ted as a result of the kinetic O 2 limitation 400 59, further supporting an electron unbalance in the AOB metabolism. To date, only one report 401 suggested a correlation between N 2O emissions in WWTPs and nirK gene transcripts 402 abundance, quantified by RT-qPCR 61. Y et, the nirK transcripts were not taxonomically 403 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 15 classified and were assumed to entirely belong to heterotrophic denitrifiers 61. All other studies 404 discussing seasonal N2O emissions in WWTPs infer the main N2O-producing pathways based 405 on metabolite profiles, and a general consensus is still lacking 18,20,21,24,43 (Table S1). Gruber et 406 al. 62 suggest heterotrophic denitrification as the main N 2O-producing pathway in a WWTP 407 using natural isotopic signatures, but seasonal dynamics were not captured. More importantly, 408 the isotopic signature s of N 2O produced through nitrifier and heterotrophic denitrification 409 largely overlap, challenging the possibility to univocally distinguish the two pathways 62,63. For 410 the same reason , 15N/18O tracer methods also did not yield conclusive results 63. Instead, by 411 integrating metagenomic-guided metaproteomics with kinetic analyses and full -scale 412 operational data we provide independent evidence on multiple ecophysiological levels 413 identifying nitrifier denitrification as the prime N 2O-producing pathway during seasonal 414 emissions. More broadly, our results demonstrate the untapped potential of multi-meta-omics 415 integration in biotechnological developments to resolve the complexity and advance the 416 engineering of the underlying microbiomes. 417 418 Figure 7. Schematic representation of the proposed ecophysiological cascade underlying seasonal N 2O emissions in 419 WWTPs. A decrease in temperature causes lower growth rates of ammonia - (AOB) and nitrite -oxidising bacteria (NOB), 420 promoting ammonium accumulation and a selective washout of the slower-growing NOB; the resulting increased ammonium 421 concentrations stimulate the growth of AOB and induce the process control to increase the operational dissolved O 2 422 concentration; the increased O 2 concentrations increase the growth rates of both AOB and NOB, but may selectively benefit 423 AOB with a lower apparent affinity for O 2. The resulting increased AOB/NOB ratio cause s the accumulation of nitrite and 424 consequent stimulation of nitrifier denitrification by AOB, as observed in the overexpression of the Cu -type nitrite reductase 425 (NirK). The ammonium, nitrite and N 2O concentration increases are a result of changes in the microbial community 426 metabolism, while the increase in O2 concentration is the only manually controlled parameter in the cascade. 427 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 16

428 429 WWTP operation. The covered Amsterdam -West WWTP has the daily capacity to treat 430 200,000 m 3 municipal wastewater under dry weather conditions (1 million population 431 equivalents). After fine screening and primary sedimentation, carbon, phosphorus and nitrogen 432 are biologically removed in a modified University of Cape Town configuration in seven 433 independent parallel cylindric plug -flow activated sludge tanks (Fig. S1). Nutrient removal 434 occurred in four compartments: anaerobic (biological phosphorus removal), anoxic 435 (denitrification), facultative (aerated when additional nitrification capacity was required), and 436 aerobic (nitrification) (Fig. S1). The setpoint for the dissolved O2 concentration in the aerobic 437 and facultative zones was set as function of the measured NH 4+ concentration in the aerated 438 compartment. The average sludge retention time (SRT) was 11-15 days and was controlled to 439 maintain an average total suspended solids of 4.2 g·L-1. N2O was measured in the combined 440 gas exhaust of all compartments (anaerobic + anoxic + facultative + aerobic) of a single lane 441 using an Rosemount TM X-STREAM gas analyser (Emerson). NH4+, NO 3- and N 2O were 442 measured in a single biological nutrient removal lane of the WWTP, NO2- was measured in the 443 pooled effluent of seven lanes. 444 445 Ex situ batch activity tests with full-scale activated sludge. The maximum nitrification and 446 denitrification activities of the activated sludge were measured every two weeks between 447 January 2021 and May 2022. For consistency, the sludge sampling, handling and storage, and 448 the activity tests were always performed in the same manner. Samples were collected from the 449 aerated compartment of the monitored full-scale activated sludge reactor and stored in two litre 450 glass bottles in the fridge for a maximum of four hours. The sludge was transported under cold 451 conditions (never reaching a temperature above 10 oC) and immediately placed in a 3 L 452 jacketed glass bioreactor with a 2 L working volume (Applikon, Getinge). The sludge was 453 made anoxic by sparging with N2 for 1 h at 0.5 L·min-1 (after which the bioreactor was sealed) 454 and was incubated overnight with 50 mg N·L-1 NaNO3 to consume the internal carbon storages. 455 During overnight storage and subsequent activity tests, the sludge was stirred at 750 rpm by 456 two six -blade turbines, the temperature was maintained at 20 ± 1 oC using a cryostat bath 457 (Lauda), and the pH was automatically maintained at 7.0 ± 0.1 by 1 M HCl and 1 M NaOH 458 with two peristaltic pumps (Watson Marlow) controlled by an in -Control process controller 459 (Applikon, Getinge). The pH and dissolved oxygen were continuously monitored with probes 460 (Applikon AppliSens, Getinge). Influent gas flows were controlled by mass -flow controllers 461 (Brooks). After overnight incubation with NO3-, the sludge was activated by adding a spike of 462 NaNO3 (5 mg N·L -1) and a mixture of organic carbon (acetate, pyruvate, glucose, 37.5 mg 463 COD·L-1 each). The batch activity tests were sequentially performed on the same day in the 464 following order: N2O, NO2- and NO3- reduction (denitrification), and NH4+ and NO2- oxidation 465 (nitrification) (Table 2). Before each batch, the depletion of the previous nitrogen compound 466 was ensured. Substrates were added to the bioreactor with a syringe and needle through a rubber 467 septum, marking the start of the batches. The batches’ progress was monitored with NO 2- and 468 NO3- MQuant® colorimetric test strips (Merck). 469 Nitrogen compounds were added at 12 mg N·L -1, in the form of N 2O (sparging 1.5% N 2O + 470 98.5% N2 at 0.5 L·min-1 during 15-20 min), NaNO2 (1.2 mL), NaNO3 (1.2 mL) and NH4HCO3 471 (1.2 mL) from concentrated stocks. The proportion bicarbonate to nitrogen was kept the same 472 for the two nitrification batches by supplying 0.9 mM NaHCO 3 to the NO 2- oxidation batch. 473 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 17 The organic carbon compounds were added at the start of the denitrification batches (each 75 474 mg COD·L-1, at least 2-fold higher than stoichiometrically needed) from anoxic concentrated 475 stock solutions: sodium acetate (C 2H3NaO2, 3 mL), sodium pyruvate (C 3H3NaO3, 3 mL) and 476 glucose (C6H12O6, 3 mL). The concentration of pyruvate was 4 -fold lower in the batch tests 477 from January until mid-August 2021, but this had no effect on the measured activities. Before 478 each denitrification test, anoxic conditions were ensured by sparging N 2 at 0.5 L·min-1 for 20 479 min, after which the reactor was sealed off from the exterior. The transition from anoxic to oxic 480 conditions was achieved by sparging air at 0.5 L·min-1 for at least 1 h. During each nitrification 481 test, oxic conditions (> 70% air saturation) were ensured by continuously sparging air at 0.5 482 L·min-1. When necessary, foam formation was reduced with a few drops of six times diluted 483 antifoam C 391 emulsion (Merck Life Science NV). For supernatant analysis, samples were 484 taken every 3, 5, 10 or 15 min (depending on the length of the batches), and immediately 485 filtered with a 0.45 µm PVDF Millex syringe filter (Merck) and placed on ice. The samples 486 were stored at 4 oC until analysis on the following day. 487 488 Table 1. Order and details of the nitrification and denitrification activity tests performed on a single day, every second 489 week. The denitrification tests (N 2O, NO2- and NO3- reduction) were performed under anoxic conditions, with a mixture of 490 organic carbon compounds as electron donor. Prior to each denitrification batch the broth was sparged with N2 during 20 min 491 to ensure anoxic conditions and remove intermediate nitrogenous gases. The nitrification tests (NH4+ and NO2- oxidation) were 492 performed with O 2 as electron acceptor, under continuous aeration. Between the denitrification and nitrification batches, the 493 broth was made oxic by sparging air for 60 min. Each nitrogen compound was added at a final concentration of 12 mg N·L-1. 494 Batch Electron donor Electron acceptor Length (min) Sparging Conditions N2O reduction (DEN) Acetate, pyruvate, glucose N2O 24 - 105 Off Anoxic NO2- reduction (DEN) Acetate, pyruvate, glucose NO2- 25 - >150 Off NO3- reduction (DEN) Acetate, pyruvate, glucose NO3- 35 - >150 Off NH4+ oxidation (AOB) NH4+ O2 30 - >150 Air Oxic NO2- oxidation (NOB) NO2- O2 45 - >150 Air 495 Analytical methods. The concentrations of NH 4+, NO2- and NO3- in the filtered supernatant 496 were spectrophotometrically measured on the day following the batches, using the Gallery TM 497 Discrete Analyzer (Thermo Fisher Scientific) or cuvette test kits (LCK339, LCK342 and 498 LCK304, Hach Lange). When measuring NO 3- with the cuvette test kits, the samples were 499 diluted 1:1 with 20 g·L-1 sulfamic acid to remove NO2- as interference. The volatile suspended 500 solids concentration (ash content subtracted from the dried biomass), measured in triplicate, 501 was taken as proxy for the biomass concentration. Immediately upon arrival, 3x 25 mL of 502 sludge was centrifuged at 4200 rpm for 20 min, the pellet was resuspended in 15 mL MilliQ 503 water, dried at 105 oC (24 h) and burned at 550 oC (2 h). The concentrations of O 2, CO2 and 504 N2O in the condenser -dried reactor off-gas were monitored by a Rosemount NGA 2000 off -505 gas analyser (Emerson). The dissolved N2O concentrations were monitored and recorded every 506 minute with a standard N 2O-R microsensor (customized concentration range 0.4 – 2 mM, 507 Unisense) and a picoammeter PA2000 (Unisense). The dissolved N 2O concentrations were 508 calculated using the average of all calibrations performed 1-2 days before every batch series. 509 510 Calculations activity tests. The maximum NO 2- and NO 3- reduction and NH 4+ and NO 2- 511 oxidation rates were obtained through linear regression of the substrate concentration profiles 512 over time. The slope was determined using at least four concentration points in the linear range. 513 The maximum N2O reduction rate was calculated in Spyder IDE v5.1.5 using Python v3.9.12 514 and the NumPy v1.21.5 64, SciPy v1.7.3 65 and Pandas v1.4.2 66 packages, taking into account 515 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 18 the gas -liquid transfer between the reactor broth and headspace throughout the batch test 516 (Supplementary Section 13). A system of ordinary differential equations (ODEs), representing 517 the liquid and headspace mass balances, was defined to describe the gas -liquid transfer over 518 time: 519 dcN2O,liq dt = rN2O − kLa∙ (cN2O,liq − cN2O,gas ∙ KH,N2O∙R∙T p ) (eq. 1) 520 dcN2O,gas dt = 𝑉𝑙𝑖𝑞 𝑉𝑔𝑎𝑠 𝑘𝐿𝑎 ∙ (cN2O,liq − cN2O,gas) (eq. 2) 521 With cN2O,liq and cN2O,gas the N2O concentration in the liquid and headspace, rN2O the unknown 522 N2O consumption rate, kLa the experimentally determined volumetric mass transfer coefficient 523 (5 h-1), KH,N2O the Henry coefficient (27.05 mM/atm), R the ideal gas constant (8.206 x 10 -5 524 L·atm·K-1·mmol), T the temperature, p the pressure, and Vliq and Vgas the broth and headspace 525 volumes. The rates were obtained by fitting the model to the experimental data, i.e. by 526 minimizing the sum of squared errors between the experimentally measured and calculated (eq. 527 1-2) N2O concentrations (see code in Supplementary Section 13). 528 529 DNA extraction, library preparation and sequencing. Samples of 2 mL were taken 530 immediately after cold transport of the sludge, and centrifuged at 16,200 x g for 5 min at 4 oC 531 to separate the biomass from the supernatant. The biomass pellets were stored at -80 oC until 532 DNA extraction. The DNA of the 12 Nov 2020, 9 Jun, 16 Dec 2021 and 11 May 2022 samples 533 was extracted with the DNeasy PowerSoil Pro Kit (Qiagen). The manufacturer’s instructions 534 were followed, with exception of these steps: approximately 50 mg biomass was resuspended 535 in the CD1 solution by vortexing before transferring to the PowerBead tube; 3 x 40 s bead -536 beating (Beadbeater -24, Biospec) was alternated with 2 min incubation on ice; tubes were 537 gently inverted instead of vortexed to prevent DNA shearing 35. The DNA of the 20 Jan and 3 538 Mar 2021 samples (1/3 pellet) w as extracted with the DNeasy UltraClean Microbial Kit 539 (Qiagen) following the manufacturer’s instructions. The DNA concentration and quality were 540 assessed with the Qubit 4 Fluorometer (Thermo Fisher Scientific) and the BioTek Synergy 541 HTX multimode microplate reader (Agilent), respectively. 542 The samples of 12 Nov 2020 (np1), 9 Jun (np2), 16 Dec 2021 (np3) and 11 May 2022 (np4) 543 were prepared for long -read sequencing using the Ligation Sequencing Kit V14 (Oxford 544 Nanopore Technologies Ltd), the NEBNext ® Companion Module for Oxford Nanopore 545 Technologies® Ligation Sequencing (New England BioLabs) and UltraPure TM BSA (50 546 mg/mL) (Thermo Fisher Scientific). The incubations in the Hula mixer were replaced with 547 slow manual inversions, all resuspensions were performed by flicking the tube, and the last 548 room temperature incubation step was performed a 37 oC to improve the recovery of long DNA 549 fragments. Four MinION R10.4 flow cells (Oxford Nanopore Technologies), one for each 550 sample, were used to sequence on a MinION for 89-96 h in accurate mode (260 bps), yielding 551 21-29 Gbp per sample. The sample of 20 Jan 2021 (np1.5) was prepared with a Ligation 552 Sequencing Kit V12 and sequenced on a GridION with MinION R9.4 flow cells (Oxford 553 Nanopore Technologies), generating 11.2 Gbp. Short-read sequencing was also performed on 554 the samples of 20 Jan (il1) and 3 Mar 2021 (il2) on a n Illumina NovaSeq 6000 platform by 555 Novogene Ltd. (UK), resulting in over 20 Gbp (per sample) of 150 bp paired-end reads with a 556 350 bp insert. 557 558 Processing of metagenomic data and MAG recovery. After sequencing, the DNA data was 559 processed to obtain metagenome -assembled genomes (MAGs). The final set of MAGs was 560 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 19 obtained from the five nanopore-sequenced samples (np1-4 and np1.5). The Illumina reads (il1 561 and il2) were solely used for differential coverage binning and to estimate the relative 562 abundance of each MAG on the respective dates. The raw long reads were basecalled in super-563 accurate mode with the “dna_r10.4.1_450bps_sup.cfg” configuration file and --564 do_read_splitting option using Guppy v6.4.2 (np1 -4) or with “dna_r9.4.1_450bps_sup.cfg” 565 using Guppy v5.0.7 (Oxford Nanopore Technologies) (np1.5). The duplex reads of np1-4 were 566 filtered using pairs_from_summary and filter_pairs from Duplex tools v0.2.19 (Oxford 567 Nanopore Technologies). The duplex reads were basecalled using the duplex basecaller of 568 Guppy and merged with the remaining simplex reads using SeqKit v2.3.0 67. The reads were 569 filtered, trimmed and inspected with NanoFilt v2.8.0 68 (options -q 10 -l 200), Porechop v0.2.4 570 (https://github.com/rrwick/Porechop) and NanoPlot v1.41.0 68. The Illumina reads were 571 filtered and trimmed using Trimmomatic v0.39 69 with the options LEADING:3 TRAILING:3 572 SLIDINGWINDOW:4:15 MINLEN:35 HEADCROP:5 . The kmer algorithm of Nonpareil 573 v3.401 70 estimated a diversity coverage of 69.9% (il1) and 71.3% (il2) for the trimmed 574 Illumina reads. 575 The long reads were individually assembled and pairwise co -assembled (np1-np2, np2-np3, 576 np3-np4) with Flye v2.9.1 71 in --meta mode. The reads were mapped on the assembly with 577 Minimap2 v2.24 72. The individual assemblies were polished with Racon v1.4.3 578 (https://github.com/isovic/racon) and two times with Medaka v1.5.0 579 (https://github.com/nanoporetech/medaka). The reads from all samples were mapped to each 580 assembly using Minimap2, the alignments were converted from SAM to BAM and sorted with 581 SAMtools v1.10 73 and the contig coverage in each sample was calculated with 582 jgi_summarize_bam_contig_depths 74. The differential coverages were used for automatic 583 binning of each assembly with MetaBAT2 v2.15 74, MaxBin2 v2.2.7 75 and CONCOCT v1.1.0 584 76, setting the minimum contig length at 2000 bp. The outputs were combined into an optimized 585 set of non -redundant bins with DAS Tool v1.1.3 77, which used Prodigal v2.6.3 78 and 586 DIAMOND v2.0.8 79. The bins obtained from all assemblies (np1, np1.5, np2, np3, np4, np1 -587 np2, np2 -np3, np3 -np4) were dereplicated with the 1083 HQ MAGs from Singleton et al. 588 (2021) 35 at 95% average nucleotide identity of open reading frames using dRep v3.2.2 80 with 589 the options -comp 70 -con 10 -sa 0.95 --S_algorithm gANI. 590 Bin completeness and contamination was assessed with the lineage_wf workflow of CheckM 591 v1.1.3 81. The relative abundance of the bins in each sample (np1, np2, np3, np4, il1, il2) was 592 determined with CoverM v0.6.1 (https://github.com/wwood/CoverM), using the options --593

relative_abundance mean --min-read-percent-identity 95 --min-read-aligned-594 percentage 50. Bins with completeness 5% or with zero abundance in 595 all samples were discarded, resulting in a non-redundant set of 349 HQ MAGs. The HQ MAGs 596 were taxonomically classified using the classify_wf mode of GTDB -Tk v2.3.0 82 and the 597 GTDB release 207 83 (gtdbtk_r207_v2_data.tar.gz). The presence of 16S rRNA genes was 598 verified with barrnap v0.9 ( https://github.com/tseemann/barrnap). A bacterial phylogenetic 599 tree was made with FastTree v2.1.11 84 using the multiple sequence alignment generated with 600 the identify and align modes of GTDB -Tk, adjusted with the TreeTools v1.10.0 85 package in 601 RStudio v22.0.3 86 with R v4.2.2 87 and visualized with iTol v6.8.2 88. 602 603 Gene prediction and f unctional annotation. Genes were predicted in all assemblies using 604 Prodigal v2.6.3 78 with the -p meta option. The gene sequences were concatenated and 605 duplicates were removed using grep and rmdup from SeqKit v2.3.1 67, resulting in a unique set 606 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 20 of genes covering all metagenomic samples. The predicted genes were functionally annotated 607 with the annotate pipeline of EnrichM v0.6.5 ( https://github.com/geronimp/enrichM), using 608 DIAMOND v2.0.8 79 and HMMER v3.2.1 (http://hmmer.org/) and the EnrichM v10 database, 609 including a KO-annotated UniRef100 2018_11 89 DIAMOND database and HMM libraries of 610 the KEGG 88.2 90, PFAM 32.0 91, and TIGRFAMs 15.0 92 databases. In general, the genes of 611 interest from the nitrogen cycle were identified by their KO identifier (Table S3). Cytochrome 612 P460 was identified through its PFAM identifier PF16694. The genes encoding the alpha- and 613 beta-subunit of the cytoplasmic nitrate reductase ( narG and narH) and the nitrite 614 oxidoreductase ( nxrA and nxrB) have the same KO identifier, so these were distinguished 615 through a phylogenetic analysis using the graft command of GraftM 93 and the respective 616 packages (7.70.nxrA_narG and 7.69.nxrB). If the alpha-subunit was classified as narG or nxrA, 617 the putative beta-subunit located in the same contig was manually annotated. The unclassified 618 sequences were left with the narGH annotation. The nxrAB genes from the Ca. Nitrotoga MAG 619 (NOB) could not be distinguished with GraftM, but were confirmed with a BLAST on UniProt 620 94. Similarly, the alpha -subunit of the ammonia monooxygenase gene ( amoA) was 621 distinguished from the methane monooxygenase gene ( pmoA) using the 20170316_pmoA 622 package of GraftM. Unidentified sequences remained annotated as pmoA. The beta - and 623 gamma-subunits located in the same contig as amoA were manually annotated as amoB and 624 amoC. Distinction between t he quinol -dependent nitric oxide reductase (qNor, encoded by 625 norZ) and the alpha subunit of the cytochrome c-dependent reductase (cNor, encoded by norB), 626 was made by identifying the fused quinol oxidase domain on the N -terminal of norZ 95. A 627 multiple sequence alignment was performed between putative NorB and NorZ protein 628 sequences found in the metagenomes (K04561), and reference sequences of NorB 629 (Pseudomonas stutzeri, P98008) and NorZ (Cupriavidus necator, Q0JYR9), extracted from 630 UniProtKB 94, using Clustal Omega v1.2.4 96. The alignment was visualized and analysed, and 631 the quinol oxidase domain was identified with Jalview v2.11.3.2. The distinction between clade 632 I and II nitrous oxide reductase, respectively TAT- and Sec-dependent, was made by combining 633 the TIGRFAM annotation of EnrichM and the phylogenetic analysis of GraftM with the 634 7.45.nosZ package. The sequences not classified as either clade I or II remained annotated as 635 unclassified nosZ. Data processing was performed using RStudio v22.0.3 86 with R v4.2.2 87, 636 and the plyr v1.8.8 98, tidyverse v2.0.0 99 , readxl v1.4.2 100, data.table v1.15.0 101, aplot v0.2.2 637 102 and reshape2 v1.4.4 103 packages. 638 639 Protein extraction. Biomass samples were taken and stored as detailed in the DNA extraction 640 section. Proteins were extracted from 12 samples, as previously described 104. Briefly, around 641 60 mg of the biomass pellet were homogenised in three cycles of vortexing and ice incubation 642 with glass beads (150 – 212 µm, Sigma Aldrich), 50 mM TEAB buffer 1% (w/w) NaDOC and 643 B-PER reagent (Thermo Scientific). Proteins in the supernatant were precipitated with 1:4 644 trichloroacetic acid (Sigma Aldrich). The pellet was washed and disrupted with acetone two 645 times and re -dissolved in 200 mM ammonium bicarbonate with 6 M Urea (Sigma Aldrich). 646 Human serum albumin (0.1 µg, Sigma Aldrich) was added to all samples to control the 647 digestion efficiency. The mixture was reduced with 10 mM dithiothreitol (Sigma Aldrich) at 648 37 °C for 60 min, and alkylated with 20 mM iodoacetamide (Sigma Aldrich) in the dark for 30 649 min. Samples were diluted with 100 mM ammonium bicarbonate to obtain a urea concentration 650 lower than 1 M. Protein digestion occurred overnight at 37 oC and 300 rpm with 1.5 µg 651 sequencing grade trypsin (Promega). 0.5 pmol of the Pierce TM Peptide Retention Time 652 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 21 Calibration mix (Thermo Scientific) was added to all samples to control the chromatographic 653 performance. Solid phase extraction was performed with an Oasis HLB 96-well µElution Plate 654 (2 mg sorbent per well, 30 µm, Waters) and a vacuum pump. The columns were conditioned 655 with MeOH, equilibrated with water two times, loaded with the peptide samples, washed with 656 two rounds of 5% MeOH and sequentially eluted with 2% formic acid in 80% MeOH and 1 657 mM ammonium bicarbonate in 80% MeOH. The samples were dried in a centrifuge 658 Concentrator plus (Eppendorf) at 45 oC and stored at -20 oC until analysis. 659 660 Shotgun metaproteomics. Peptide samples were dissolved in 20 µL of 3% acetonitrile and 661 0.01% trifluoroacetic acid, incubated at room temperature for 30 min and vortexed thoroughly. 662 The protein concentration was measured at 280 nm wavelength with a NanoDrop ND -1000 663 spectrophotometer (Thermo Scientific) and s amples were diluted to a concentration of 0.5 664 mg/mL. Shotgun metaproteomics was performed as previously described 104, with a 665 randomized sample order. Briefly, approximately 0.5 µg protein digest was analysed using a 666 nano-liquid-chromatography system consisting of an EASY nano-LC 1200, equipped with an 667 Acclaim PepMap RSLC RP C18 separation column (50 μm x 150 mm, 2 μm, Cat. No. 164568), 668 and a QE plus Orbitrap mass spectrometer (Thermo Fisher Scientific). The flow rate was 669 maintained at 350 nL/min over a linear gradient from 5% to 25% solvent B over 90 min, from 670 25% to 55% over 60 min, followed by back equilibration to starting conditions. Solvent A was 671 a 0.1% formic acid solution in water (FA), and solvent B consisted of 80% ACN in water and 672 0.1% FA. The Orbitrap was operated in data dependent acquisition (DDA) mode acquiring 673 peptide signals from 385–1250 m/z at 70 K resolution in full MS mode with a maximum ion 674 injection time (IT) of 75 ms and an automatic gain control (AGC) target of 3E6. The top 10 675 precursors were selected for MS/MS analysis and subjected to fragmentation using higher -676 energy collisional dissociation (HCD) at a normalised collision energy of 28. MS/MS scans 677 were acquired at 17.5 K resolution with AGC target of 2E5 and IT of 75 ms, 1.2 m/z isolation 678 width. The protein reference sequence database was generated through whole metagenome 679 sequencing of the microbial samples, which included all metagenome -assembled genomes 680 (MAGs) and unique unbinned sequences from all samples. The raw mass spectrometric data 681 from each sample were analysed against this database using PEAKS Studio X (Bioinformatics 682 Solutions Inc.) in a two -round database search process. The initial round was conducted 683 without considering variable modifications and missed cleavages. Subsequently, the focused 684 database was further searched, allowing for a 20 ppm parent ion and a 0.02 m/z fragment ion 685 mass error tolerance, up to 3 missed cleavages, and iodoacetamide as a fixed modification, with 686 methionine oxidation and N/Q deamidation as variable modifications. 687 688 Metaproteomic data analysis. Peptide spectrum matches were filtered against 5% false 689 discovery rates (FDR) and protein identifications with ≥2 unique peptide sequences were 690 considered significant. The human serum albumin added as internal process control was filtered 691 out. Proteins were grouped according to their unique protein group identification. The peptide 692 spectral counts were divided by their molar mass for normalisation and technical duplicates 693 were averaged. The relative abundance of each protein in a certain sample was determined by 694 dividing the respective normalized spectral counts by the sum of normalized spectral counts of 695 all proteins detected in that sample. The total relative abundance of each MAG in the 696 metaproteome was calculated by summing the relative abundance of all proteins belonging to 697 that MAG. The same was performed to calculate the total relative abundance of functionally 698 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 22 identical proteins. Some functionally identical proteins belonging to different MAGs from the 699 same genus could not be distinguished because of their high similarity. Therefore, the se 700 proteins were grouped by their functional annotation and genus for the data analysis. Proteins 701 that simultaneously matched unbinned sequences and one or more MAGs from a certain genus, 702 were classified as belonging to that genus. The catalytic subunits of the nitrogen -converting 703 enzymes of interest were used as representative of that protein during data analysis, with 704 exception of the ammonia monooxygenase (AMO). The catalytic alpha -subunit (AmoA) is 705 located in the cell membrane 105, and is thus hydrophobic, so it is not well detected in the 706 proteomic analysis (Fig. S15). The beta-subunit (AmoB), only partially in the membrane, was 707 detected in much higher amounts so it was here used as proxy for AMO. In any case, the results 708 were similar for AmoA and AmoB (Fig. S17). Data processing was performed using RStudio 709 v22.0.3 86 with R v4.2.2 87, and the plyr v1.8.8 98, tidyverse v2.0.0 99 , readxl v1.4.2 100, 710 data.table v1.15.0 101, aplot v0.2.2 102, reshape2 v1.4.4 103 and matrixStats v1.2.0 106 packages. 711 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 23

712 713 1. IPCC. Climate Change 2014: Synthesis Report. (2014). 714 2. Tian, H. et al. A comprehensive quantification of global nitrous oxide sources and sinks. 715 Nature 586, 248–256 (2020). 716 3. Vasilaki, V., Massara, T. M., Stanchev, P., Fatone, F. & Katsou, E. A decade of nitrous oxide 717 (N2O) monitoring in full-scale wastewater treatment processes: A critical review. Water Res. 718 161, 392–412 (2019). 719 4. Marchant, H. K. et al. Denitrifying community in coastal sediments performs aerobic and 720 anaerobic respiration simultaneously. ISME J. 11, 1799–1812 (2017). 721 5. Freing, A., Wallace, D. W. R. & Bange, H. W. Global oceanic production of nitrous oxide. 722 Philos. Trans. R. Soc. B 367, 1245–1255 (2012). 723 6. Butterbach-Bahl, K., Baggs, E. M., Dannenmann, M., Kiese, R. & Zechmeister-Boltenstern, S. 724 Nitrous oxide emissions from soils: How well do we understand the processes and their 725 controls? Philos. Trans. R. Soc. B Biol. Sci. 368, (2013). 726 7. Sabba, F., Picioreanu, C., Boltz, J. P. & Nerenberg, R. Predicting N2O emissions from 727 nitrifying and denitrifying biofilms: A modeling study. Water Sci. Technol. 75, 530–538 728 (2017). 729 8. Schreiber, F., Wunderlin, P., Udert, K. M. & Wells, G. F. Nitric oxide and nitrous oxide 730 turnover in natural and engineered microbial communities: Biological pathways, chemical 731 reactions, and novel technologies. Front. Microbiol. 3, 1–24 (2012). 732 9. Yang, S. et al. Global reconstruction reduces the uncertainty of oceanic nitrous oxide 733 emissions and reveals a vigorous seasonal cycle. Proc. Natl. Acad. Sci. U. S. A. 117, (2020). 734 10. Wan, X. S., Lin, H., Ward, B. B., Kao, S. J. & Dai, M. Significant Seasonal N2O Dynamics 735 Revealed by Multi-Year Observations in the Northern South China Sea. Global Biogeochem. 736 Cycles 36, (2022). 737 11. Bremner, J. M., Robbins, S. G. & Blackmer, A. M. Seasonal variability in emission of nitrous 738 oxide from soil. Geophys. Res. Lett. 7, 641–644 (1980). 739 12. Wagner-Riddle, C. et al. Globally important nitrous oxide emissions from croplands induced 740 by freeze-thaw cycles. Nat. Geosci. 10, 279–283 (2017). 741 13. Wang, X. et al. Hot moment of N2O emissions in seasonally frozen peatlands. ISME J. 17, 742 792–802 (2023). 743 14. Kortelainen, P. et al. Lakes as nitrous oxide sources in the boreal landscape. Glob. Chang. 744 Biol. 26, 1432–1445 (2020). 745 15. Liang, X. et al. Nitrification Regulates the Spatiotemporal Variability of N2O Emissions in a 746 Eutrophic Lake. Environ. Sci. Technol. 56, 17430–17442 (2022). 747 16. Beaulieu, J. J., Shuster, W. D. & Rebholz, J. A. Nitrous oxide emissions from a large, 748 impounded river: The Ohio river. Environ. Sci. Technol. 44, 7527–7533 (2010). 749 17. Brotto, A. C. et al. Factors controlling nitrous oxide emissions from a full-scale activated 750 sludge system in the tropics. Environ. Sci. Pollut. Res. 22, 11840–11849 (2015). 751 18. Chen, X. et al. Assessment of Full-Scale N2O Emission Characteristics and Testing of Control 752 Concepts in an Activated Sludge Wastewater Treatment Plant with Alternating Aerobic and 753 Anoxic Phases. Environ. Sci. Technol. 53, 12485–12494 (2019). 754 19. Valk, L. et al. Exploring the microbial influence on seasonal nitrous oxide concentration in a 755 full-scale wastewater treatment plant using metagenome assembled genomes. Water Res. 219, 756 118563 (2022). 757 20. Sieranen, M. et al. Seasonality of nitrous oxide emissions at six full-scale wastewater 758 treatment plants. Water Sci. Technol. 89, 603–612 (2023). 759 21. Daelman, M. R. J., van Voorthuizen, E. M., van Dongen, U. G. J. M., Volcke, E. I. P. & van 760 Loosdrecht, M. C. M. Seasonal and diurnal variability of N2O emissions from a full-scale 761 municipal wastewater treatment plant. Sci. Total Environ. 536, 1–11 (2015). 762 22. Bae, W. Bin et al. Temporal triggers of N2O emissions during cyclical and seasonal variations 763 of a full-scale sequencing batch reactor treating municipal wastewater. Sci. Total Environ. 797, 764 149093 (2021). 765 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 24 23. Gruber, W. et al. Linking seasonal N2O emissions and nitrification failures to microbial 766 dynamics in a SBR wastewater treatment plant. Water Res. X 11, 100098 (2021). 767 24. Gruber, W. et al. Estimation of countrywide N2O emissions from wastewater treatment in 768 Switzerland using long-term monitoring data. Water Res. X 13, (2021). 769 25. Daims, H., Taylor, M. W. & Wagner, M. Wastewater treatment: a model system for microbial 770 ecology. Trends Biotechnol. 24, 483–489 (2006). 771 26. Gruber, W. et al. N2O emission in full-scale wastewater treatment: Proposing a refined 772 monitoring strategy. Sci. Total Environ. 699, 134157 (2020). 773 27. Kosonen, H. et al. Nitrous Oxide Production at a Fully Covered Wastewater Treatment Plant: 774

of a Long-Term Online Monitoring Campaign. Environ. Sci. Technol. 50, 5547–5554 775 (2016). 776 28. Daelman, M. R. J., Van Voorthuizen, E. M., Van Dongen, L. G. J. M., Volcke, E. I. P. & Van 777 Loosdrecht, M. C. M. Methane and nitrous oxide emissions from municipal wastewater 778 treatment - Results from a long-term study. Water Sci. Technol. 67, 2350–2355 (2013). 779 29. Chen, H., Zeng, L., Wang, D., Zhou, Y. & Yang, X. Recent advances in nitrous oxide 780 production and mitigation in wastewater treatment. Water Res. 184, 116168 (2020). 781 30. Baggs, E. M. Soil microbial sources of nitrous oxide: Recent advances in knowledge, 782 emerging challenges and future direction. Curr. Opin. Environ. Sustain. 3, 321–327 (2011). 783 31. Thomson, A. J., Giannopoulos, G., Pretty, J., Baggs, E. M. & Richardson, D. J. Biological 784 sources and sinks of nitrous oxide and strategies to mitigate emissions. Philos. Trans. R. Soc. B 785 Biol. Sci. 367, 1157–1168 (2012). 786 32. Pijuan, M. & Zhao, Y. Quantification and Modelling of Fugitive Greenhouse Gas Emissions 787 from Urban Water Systems. Quantification and Modelling of Fugitive Greenhouse Gas 788 Emissions from Urban Water Systems (2022). doi:10.2166/9781789060461. 789 33. Vasilaki, V., Volcke, E. I. P., Nandi, A. K., van Loosdrecht, M. C. M. & Katsou, E. Relating 790 N2O emissions during biological nitrogen removal with operating conditions using 791 multivariate statistical techniques. Water Res. 140, 387–402 (2018). 792 34. Johnston, J., LaPara, T. & Behrens, S. Composition and Dynamics of the Activated Sludge 793 Microbiome during Seasonal Nitrification Failure. Sci. Rep. 9, 1–15 (2019). 794 35. Singleton, C. M. et al. Connecting structure to function with the recovery of over 1000 high-795 quality metagenome-assembled genomes from activated sludge using long-read sequencing. 796 Nat. Commun. 12, 1–13 (2021). 797 36. Wang, Y. et al. Successional dynamics and alternative stable states in a saline activated sludge 798 microbial community over 9 years. Microbiome 9, 1–15 (2021). 799 37. Ju, F., Guo, F., Ye, L., Xia, Y. & Zhang, T. Metagenomic analysis on seasonal microbial 800 variations of activated sludge from a full-scale wastewater treatment plant over 4 years. 801 Environ. Microbiol. Rep. 6, 80–89 (2014). 802 38. Saunders, A. M., Albertsen, M., Vollertsen, J. & Nielsen, P. H. The activated sludge 803 ecosystem contains a core community of abundant organisms. ISME J. 10, 11–20 (2016). 804 39. Herold, M. et al. Integration of time-series meta-omics data reveals how microbial ecosystems 805 respond to disturbance. Nat. Commun. 11, (2020). 806 40. Yu, K. & Zhang, T. Metagenomic and metatranscriptomic analysis of microbial community 807 structure and gene expression of activated sludge. PLoS One 7, (2012). 808 41. Lu, H., Chandran, K. & Stensel, D. Microbial ecology of denitrification in biological 809 wastewater treatment. Water Res. 64, 237–254 (2014). 810 42. Castellano-Hinojosa, A., Maza-Márquez, P., Melero-Rubio, Y., González-López, J. & 811 Rodelas, B. Linking nitrous oxide emissions to population dynamics of nitrifying and 812 denitrifying prokaryotes in four full-scale wastewater treatment plants. Chemosphere 200, 57–813 66 (2018). 814 43. Song, M. J. et al. Identification of primary effecters of N2O emissions from full-scale 815 biological nitrogen removal systems using random forest approach. Water Res. 184, 116144 816 (2020). 817 44. Duan, H. et al. Improving wastewater management using free nitrous acid (FNA). Water Res. 818 171, (2020). 819 45. Zhou, Y., Oehmen, A., Lim, M., Vadivelu, V. & Ng, W. J. The role of nitrite and free nitrous 820 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 25 acid (FNA) in wastewater treatment plants. Water Res. 45, 4672–4682 (2011). 821 46. Park, S. & Bae, W. Modeling kinetics of ammonium oxidation and nitrite oxidation under 822 simultaneous inhibition by free ammonia and free nitrous acid. Process Biochem. 44, 631–640 823 (2009). 824 47. Anthonisen, A. C., Srinath, E. G., Loehr, R. C. & Prakasam, T. B. S. Inhibition of nitrification 825 and nitrous acid compounds. J. Water Pollut. Control Fed. 48, 835–852 (1976). 826 48. Cao, Y. et al. The influence of dissolved oxygen on partial nitritation/ anammox performance 827 and microbial community of the 200,000 m3/d activated sludge process at the Changi water 828 reclamation plant (2011 to 2016). Water Sci. Technol. 78, 634–643 (2018). 829 49. Law, Y. et al. High dissolved oxygen selection against nitrospira sublineage i in full-scale 830 activated sludge. Environ. Sci. Technol. 53, 8157–8166 (2019). 831 50. Picioreanu, C., Pérez, J. & van Loosdrecht, M. C. M. Impact of cell cluster size on apparent 832 half-saturation coefficients for oxygen in nitrifying sludge and biofilms. Water Res. 106, 371–833 382 (2016). 834 51. Regmi, P. et al. Control of aeration, aerobic SRT and COD input for mainstream 835 nitritation/denitritation. Water Res. 57, 162–171 (2014). 836 52. Olaya-Abril, A. et al. Exploring the denitrification proteome of Paracoccus denitrificans 837 PD1222. Front. Microbiol. 9, 1–16 (2018). 838 53. Zorz, J. K., Kozlowski, J. A., Stein, L. Y., Strous, M. & Kleiner, M. Comparative proteomics 839 of three species of ammonia-oxidizing bacteria. Front. Microbiol. 9, 1–15 (2018). 840 54. Mundinger, A. B., Lawson, C. E., Jetten, M. S. M., Koch, H. & Lücker, S. Cultivation and 841 transcriptional analysis of a canonical nitrospira under stable growth conditions. Front. 842 Microbiol. 10, 1–15 (2019). 843 55. Daims, H., Lücker, S. & Wagner, M. A New Perspective on Microbes Formerly Known as 844 Nitrite-Oxidizing Bacteria. Trends Microbiol. 24, 699–712 (2016). 845 56. Starkenburg, S. R. et al. Genome sequence of the chemolithoautotrophic nitrite-oxidizing 846 bacterium Nitrobacter winogradskyi Nb-255. Appl. Environ. Microbiol. 72, 2050–2063 (2006). 847 57. Beaumont, H. J. E., Lens, S. I., Reijnders, W. N. M., Westerhoff, H. V. & Van Spanning, R. J. 848 M. Expression of nitrite reductase in Nitrosomonas europaea involves NsrR, a novel nitrite-849 sensitive transcription repressor. Mol. Microbiol. 54, 148–158 (2004). 850 58. Yu, R. & Chandran, K. Strategies of nitrosomonas europaea 19718 to counter low dissolved 851 oxygen and high nitrite concentrations. BMC Microbiol. 10, (2010). 852 59. Soler-Jofra, A., Schmidtchen, L., Olmo, L., van Loosdrecht, M. C. M. & Pérez, J. Short and 853 long term continuous hydroxylamine feeding in a granular sludge partial nitritation reactor. 854 Water Res. 209, (2022). 855 60. Perez-Garcia, O., Villas-Boas, S. G., Swift, S., Chandran, K. & Singhal, N. Clarifying the 856 regulation of NO/N2O production in Nitrosomonas europaea during anoxic-oxic transition via 857 flux balance analysis of a metabolic network model. Water Res. 60, 267–277 (2014). 858 61. Vieira, A., Galinha, C. F., Oehmen, A. & Carvalho, G. The link between nitrous oxide 859 emissions, microbial community profile and function from three full-scale WWTPs. Sci. Total 860 Environ. 651, 2460–2472 (2019). 861 62. Gruber, W. et al. Tracing N2O formation in full-scale wastewater treatment with natural 862 abundance isotopes indicates control by organic substrate and process settings. Water Res. X 863 15, (2022). 864 63. Duan, H., Ye, L., Erler, D., Ni, B. & Yuan, Z. Quantifying nitrous oxide production pathways 865 in wastewater treatment systems using isotope technology e A critical review. Water Res. 122, 866 96–113 (2017). 867 64. Harris, C. R. et al. Array programming with NumPy. Nature 585, 357–362 (2020). 868 65. Virtanen, P. et al. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat. 869

17, 261–272 (2020). 870 66. McKinney, W. pandas: a Foundational Python Library for Data Analysis and Statistics. Python 871 High Perform. Sci. Comput. 1–9 (2011). 872 67. Shen, W., Le, S., Li, Y. & Hu, F. SeqKit: A cross-platform and ultrafast toolkit for FASTA/Q 873 file manipulation. PLoS One 11, 1–10 (2016). 874 68. De Coster, W., D’Hert, S., Schultz, D. T., Cruts, M. & Van Broeckhoven, C. NanoPack: 875 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 26 Visualizing and processing long-read sequencing data. Bioinformatics 34, 2666–2669 (2018). 876 69. Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: A flexible trimmer for Illumina 877 sequence data. Bioinformatics 30, 2114–2120 (2014). 878 70. Rodriguez-R, L. M. & Konstantinidis, K. T. Nonpareil: A redundancy-based approach to 879 assess the level of coverage in metagenomic datasets. Bioinformatics 30, 629–635 (2014). 880 71. Kolmogorov, M. et al. metaFlye: scalable long-read metagenome assembly using repeat 881 graphs. Nat. Methods 17, 1103–1110 (2020). 882 72. Li, H. Minimap2 : pairwise alignment for nucleotide sequences. 34, 3094–3100 (2018). 883 73. Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–884 2079 (2009). 885 74. Kang, D. D. et al. MetaBAT 2: An adaptive binning algorithm for robust and efficient genome 886 reconstruction from metagenome assemblies. PeerJ 2019, 1–13 (2019). 887 75. Wu, Y. W., Simmons, B. A. & Singer, S. W. MaxBin 2.0: An automated binning algorithm to 888 recover genomes from multiple metagenomic datasets. Bioinformatics 32, 605–607 (2016). 889 76. Alneberg, J. et al. Binning metagenomic contigs by coverage and composition. Nat. Methods 890 11, 1144–1146 (2014). 891 77. Sieber, C. M. K. et al. Recovery of genomes from metagenomes via a dereplication, 892 aggregation and scoring strategy. Nat. Microbiol. 3, 836–843 (2018). 893 78. Hyatt, D. et al. Prodigal: Prokaryotic gene recognition and translation initiation site 894 identification. BMC Bioinformatics 11, (2010). 895 79. Buchfink, B., Xie, C. & Huson, D. H. Fast and sensitive protein alignment using DIAMOND. 896 Nat. Methods 12, 59–60 (2014). 897 80. Olm, M. R., Brown, C. T., Brooks, B. & Banfield, J. F. DRep: A tool for fast and accurate 898 genomic comparisons that enables improved genome recovery from metagenomes through de-899 replication. ISME J. 11, 2864–2868 (2017). 900 81. Parks, D. H., Imelfort, M., Skennerton, C. T., Hugenholtz, P. & Tyson, G. W. CheckM: 901 Assessing the quality of microbial genomes recovered from isolates, single cells, and 902 metagenomes. Genome Res. 25, 1043–1055 (2015). 903 82. Chaumeil, P. A., Mussig, A. J., Hugenholtz, P. & Parks, D. H. GTDB-Tk v2: memory friendly 904 classification with the genome taxonomy database. Bioinformatics 38, 5315–5316 (2022). 905 83. Parks, D. H. et al. GTDB: An ongoing census of bacterial and archaeal diversity through a 906 phylogenetically consistent, rank normalized and complete genome-based taxonomy. Nucleic 907 Acids Res. 50, D785–D794 (2022). 908 84. Price, M. N., Dehal, P. S. & Arkin, A. P. FastTree 2 - Approximately maximum-likelihood 909 trees for large alignments. PLoS One 5, (2010). 910 85. Smith, M. TreeTools: create, modify and analyse phylogenetic trees. Comprehensive R 911 Archive Network at https://doi.org/10.5281/zenodo.3522725 (2019). 912 86. RStudio Team. RStudio: Integrated Development Environment for R. RStudio, PBC, Boston, 913 MA http://www.rstudio.com/ (2021). 914 87. R Core Team. R: A language and environment for statistical computing. R Foundation for 915 Statistical Computing, Vienna, Austria https://www.r-project.org/ (2022). 916 88. Letunic, I. & Bork, P. Interactive tree of life (iTOL) v5: An online tool for phylogenetic tree 917 display and annotation. Nucleic Acids Res. 49, W293–W296 (2021). 918 89. Suzek, B. E., Wang, Y., Huang, H., McGarvey, P. B. & Wu, C. H. UniRef clusters: A 919 comprehensive and scalable alternative for improving sequence similarity searches. 920 Bioinformatics 31, 926–932 (2015). 921 90. Kanehisa, M., Sato, Y., Kawashima, M., Furumichi, M. & Tanabe, M. KEGG as a reference 922 resource for gene and protein annotation. Nucleic Acids Res. 44, D457–D462 (2016). 923 91. Mistry, J. et al. Pfam: The protein families database in 2021. Nucleic Acids Res. 49, D412–924 D419 (2021). 925 92. Haft, D. H. et al. TIGRFAMs and genome properties in 2013. Nucleic Acids Res. 41, 387–395 926 (2013). 927 93. Boyd, J. A., Woodcroft, B. J. & Tyson, G. W. GraftM: a tool for scalable, phylogenetically 928 informed classification of genes within metagenomes. Nucleic Acids Res. 46, E59 (2018). 929 94. The UniProt Consortium. UniProt: the Universal Protein Knowledgebase in 2023. Nucleic 930 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 27 Acids Res. 51, 523–531 (2023). 931 95. Zumft, W. G. Nitric oxide reductases of prokaryotes with emphasis on the respiratory, heme-932 copper oxidase type. J. Inorg. Biochem. 99, 194–215 (2005). 933 96. Madeira, F. et al. Search and sequence analysis tools services from EMBL-EBI in 2022. 934 Nucleic Acids Res. 50, W276–W279 (2022). 935 97. Waterhouse, A. M., Procter, J. B., Martin, D. M. A., Clamp, M. & Barton, G. J. Jalview 936 Version 2-A multiple sequence alignment editor and analysis workbench. Bioinformatics 25, 937 1189–1191 (2009). 938 98. Wickham, H. The split-apply-combine strategy for data analysis. J. Stat. Softw. 40, 1–29 939 (2011). 940 99. Wickham, H. et al. Welcome to the Tidyverse. J. Open Source Softw. 4, 1686 (2019). 941 100. Wickham, H. & Bryan, J. readxl: Read Excel Files. https://cran.r-project.org/package=readxl 942 (2023). 943 101. Barrett, T. et al. data.table: Extension of ‘data.frame’. at https://rdatatable.gitlab.io/data.table 944 (2024). 945 102. Yu, G. aplot for decorating a plot with associated information. at https://github.com/YuLab-946 SMU/aplot (2023). 947 103. Wickham, H. Reshaping data with the reshape package. J. Stat. Softw. 21, 1–20 (2007). 948 104. Kleikamp, H. B. C. et al. Metaproteomics, metagenomics and 16S rRNA sequencing provide 949 different perspectives on the aerobic granular sludge microbiome. Water Res. 246, 120700 950 (2023). 951 105. Musiani, F., Broll, V., Evangelisti, E. & Ciurli, S. The model structure of the 952 copper‑dependent ammonia monooxygenase. J. Biol. Inorg. Chem. 25, 995–1007 (2020). 953 106. Bengtsson, H. et al. matrixStats: Functions that Apply to Rows and Columns of Matrices (and 954 to Vectors). at https://github.com/HenrikBengtsson/matrixStats (2023). 955 956 957 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint 28

958 We deeply thank Dirk Geerts (TU Delft) for valuable support with the bioreactor, Carol de 959 Ram and Dita Heikens (TU Delft) for help with the protein extraction protocol, Waternet for 960 providing the activated sludge (Hidde Schijfsma, Alex Veltman and Adrien Azé for sampling), 961 Mads Albertsen and team at Aalborg University for sequencing one of the samples with Oxford 962 Nanopore R9.4 during the “Hands -on metagenomics using Oxford Nanopore DNA 963 sequencing” course, and Alexandra Deeke (Waterschap de Dommel), Cora Uijterlinde 964 (STOWA), Inge Pistorius, Robert Kras, Floris de Heer and Maurice Ramaker (Waterschap Aa 965 en Maas), Maaike Hoekstra (HHNK), Mariska Ronteltap (Hoogheemraadschap van Delfland), 966 the Dutch Community of Practice for N2O, Adriano Joss (Eawag), and Wenzel Gruber (upwater 967 AG) for insightful discussions. 968 969 Funding 970 This work was financed by Stichting Toegepast Onderzoek Waterbeheer (STOWA; 971 JG191217009/732.750/CU), Hoogheemraadschap Hollands Noorderkwartier (HHNK; 972 20.0787440) and Waterschap de Dommel (Z62737/U131154). ML was partially supported by 973 a Veni grant from the Dutch Research Council (NWO; project number VI.Veni.192.252). 974 975 Competing interest statement 976 The authors declare no competing interests. 977 978 Data availability 979 Raw DNA reads were deposited in the NCBI Sequence Read Archive and the 54 high -quality 980 MAGs were deposited in Genbank under BioProject PRJNA1082082 . The raw mass 981 spectrometry proteomics data acquired in this project have been deposited in the 982 ProteomeXchange Consortium database under the dataset identifier PXD051095. The HQ 983 MAGs accession numbers, quality and abundance can be found in Supplementary Data 1, and 984 the gene presence and protein abundance in the MAGs and the unbinned sequences can be 985 found in Supplementary Data 2 and 3, respectively. The Python codes used to calculate the 986 maximum N 2O rates and to simulate the biological nitrogen removal process are in 987 Supplementary Information. 988 989 Author contributions 990 NR, MvL and ML conceptualized the study. NR, MP, MvD, CHM, MvL and ML designed the 991 experiments. NR, MvD and CHM performed the maximum activity measurements. MZ 992 collected and pre -processed the WWTP data. NR performed the Nanopore sequencing, 993 implemented the bioinformatics pipeline and performed metagenomic analysis with input from 994 TA. NR performed the protein extraction with input from MP. MP performed the shotgun 995 metaproteomics. NR wrote the Python and RStudio scripts for data analysis and NR, MvD, 996 CHM analysed the data with inputs from MZ, MvL and ML. NR wrote the draft manuscript 997 and created the visuals with strong inputs from ML and contributions from all co -authors. All 998 authors reviewed and approved the final manuscript. 999 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 18, 2024. ; https://doi.org/10.1101/2024.04.17.589950doi: bioRxiv preprint