Temporal dynamics and ecophysiology of thermophilic composting analyzed through metagenome-assembled genomes

Background: Thermophilic composting is a semi-engineered process carried out by diverse microbial communities. Composting is an environment friendly way of degrading biomass; its study may help uncover important biomass-degrading organisms and key enzymes. DNA sequence-based previous studies have presented a general description of the microbial-molecular features of composting, but they have lacked more specic information on the key organisms that are active during the process and their genomes. Methods: We present an analysis of metagenome-assembled genomes (MAGs) obtained from time-series samples of a thermophilic composting process in the São Paulo Zoological Park (Brazil). Our results are based on a careful analysis of MAG gene content and on metabolic modeling of their interactions. Results: We recovered 60 MAGs from sequencing datasets from two separate composting cells. Phylogenetic analysis shows that 47 of these MAGs represent novel taxa at the genus or higher levels. We have analyzed the gene repertoire of these MAGs in terms of lignocellulose degradation, secondary metabolite production, antibiotic resistance genes, denitrication genes, sulfur metabolism, hydrogen metabolism, and oxygen metabolism. For one of the composting cells we also had metatranscriptome data, which allowed a deeper analysis of 49 MAGs. This analysis showed the presence of three distinct clusters of MAGs with varying activity during the 99-day composting process. The interaction model pointed to Sphaerobacter thermophilus and Thermobispora bispora as key players in the process, as well as other bacteria that are novel. Our results also show the importance of coadjuvant bacteria and of microbial functions related to ecient bioenergetic processes during biomass conversion, such as N2O reduction and hydrogenases. A novel acidobacteria MAG encodes N2O reductase hallmark genes (nosZD). Strong metabolic dependencies predicted between MAGs revealed that cross-feeding in composting can be determined by complementary functions found in the genomes of producers and consumers, supporting the Black Queen hypothesis for co-evolutionary interactions. Conclusions: This study reveals for the rst time the key bacterial players in thermophilic composting and provides a model of their dynamic metabolic interactions. These ndings pave the way for more rational composting procedures and provide information that could help the development of novel biomass-degrading technologies.


Introduction
Thermophilic composting is a semi-engineered process carried out by microbial communities able to thrive in those environments [1][2][3]. Composting microbes present a remarkable metabolic exibility and are able to break down complex compounds such as lignocellulosic biomass [3,4]. Therefore, the composting microbiome is a valuable microbial resource with potential for contributing in a number of industrial applications [5,6]. In spite of all this potential, knowledge on how to control and explore those microbes and their functions remains encrypted within their genomes and the multiple combinations of metabolic pathways that they can activate [7][8][9].
So far, research on composting microbes has focused mainly on enriched cultures [9][10][11] or taxonomic biodiversity assessments based on 16S rRNA gene amplicon sequencing data [12,13]. Enriched cultures have lower diversity richness due to cultivation bias and the limited number ecological niches [9].
Community pro ling based on 16S rRNA gene amplicon data may reveal insights about microbial composition and the dynamics of ecological succession [14], however those methods cannot assess microbial functional diversity and metabolic activity. Lignocellulosic biomass is composed by different biopolymers: cellulose (25-55%), hemicellulose (19-40%) and lignin (18-35%), and smaller fractions of pectin and minerals [15]. Therefore, a diverse set of enzymes is required for effective sacchari cation (i.e., depolymerization of lignocellulose compounds into monosaccharide components) [16]. Microbial dynamics during lignocellulose breakdown seems to be heavily dependent on syntrophic interactions [9]. The microbial populations need to share the burden of enzymatic production; and sharing metabolites reduces the negative feedback effect of intermediate metabolite accumulation (e.g., intracellular enzymatic competition) [17]. Syntrophic interactions happen in the presence of opportunistic microbes in biomass degrading systems; these are microbes that do not express or very often lack the required enzymes for biomass degradation, but constitute one important portion in microbial consortia obtained from enrichment cultures (sometimes referred to as 'sugar cheaters') [9,18].
Shotgun metagenomic sequencing has helped reveal the diversity of microbial communities in environmental samples [19][20][21]. New methods and computational tools allow the recovery of metagenome-assembled genomes (MAGs) from complex environments [22][23][24][25][26]. The availability of MAGs from organisms in a given environment provides detailed taxonomical and functional diversity information, and therefore has the potential to allow a better understanding of the ecological context and the arsenal of enzymes encoded by microbes living in that environment [25,27,28].
In order to control and fully explore the composting system as a microbial source for biotechnological solutions, it is important to consider the ecological basis of the composting microbiome. Therefore, here we present an analysis of MAGs obtained from time-series samples of a thermophilic composting process [3]. Our aim was obtain a more detailed view of the microbial populations active in a composting process, advancing on our previous work [3].
We used the collection of genomes recovered to build a framework for the molecular-microbial temporal dynamic in the composting process, from which we were able to infer the major ecophysiological patterns of its microbiome. Using this framework as a reference, we built genome-scale metabolic models for predicting syntrophic interactions and the more frequently exchanged compounds.
For ease of reading, in the text we sometimes we refer to MAGs when we actually mean the bacteria from which these genomes were obtained.

Results
MAGs recovered from composting A total of 11 MAGs from the ZC3 dataset and 49 from the ZC4 dataset were recovered. (Supplementary  Table S1). All of our MAGs meet the medium-quality requirement (> 50% completeness and < 10% contamination) of the MIMAG standard [29] (with one exception: ZC4RG07 had 10.87% contamination); 34 MAGs meet the high-quality requirement (> 90% completeness and < 5% contamination). The average number of contigs in these MAGs is 363.75, with a minimum of 15 and a maximum of 1,655. Thirteen MAGs could be assigned to species for which there is at least one genome publicly available; this allowed us to compare our MAGs to publicly available genomes of the same species (Supplementary Table S2a). In all cases the two-way ANI measure was above 98% and the GGDC DDH estimate (formula 2) was at least 87.3%, strongly suggesting that the assignments are correct and that the recovered genomes are of high quality. These 13 MAGs correspond to 11 species that have been found in environments related to biomass degradation, with the possible exception of M. hassiacum, which has been isolated in human patients. All these species have been reported as thermophilic bacteria (Supplementary Table S3).

Taxonomic assignments
The 60 recovered MAGs could be assigned to six different phyla: Acidobacteria, Actinobacteria, Bacteroidetes, Chloro exi, Firmicutes and Proteobacteria (Supplementary Table S1). At the order level there is remarkable diversity: 32 different orders are represented. The most frequent order was Limnochordales (six MAGs). Most of the MAGs seem to be novel: there are seven MAGs for which no order could be assigned, 14 MAGs for which no family could be assigned, 14 MAGs for which no genus could be assigned, and 12 MAGs for which no species could be assigned. This counting does not take into account GTDB-tk assignments to taxa that are not currently accepted (such as family WCHB1-69 within the order bacteroidales).

Presence of MAGs in other datasets
We compared the 60 MAGs with genomes recovered from our previous studies [3,11] and between both composting cells (Supplementary Table S2b). We observed that a few genomes can be said to be present (using the same methodology for MAG species assignment) in different samples, although the majority was found in only one sample. The recovery of the "same" genome from different samples lends additional con dence to our genome recovery process. It is probable that there may be more occurrences of the same MAGs in these samples than here reported, but our sequencing coverage and stringent genome recovery criteria may have prevented us from recovering additional already-observed MAGs from all samples. For functional analysis, we have focused on ZC4 samples only, since for them alone do we have   metatranscriptome data. ZC4 is composed of nine time-series samples (days 1, 3, 7, 15, 30, 64, 67, 78 and  99). A comparison of variation in relative abundance of transcripts over time showed that all MAGs here analyzed were transcriptionally active (Supplementary Table S4) The composting process is clearly very complex, both in terms of its microbiota and in terms of the varied subprocesses that occur over the approximately three months during which composting takes place. For the functional analysis that follows we have focused on functions that we deemed relevant for composting microbial systems. Those that we chose to study in detail are: lignocellulose degradation; denitri cation; sulfur metabolism; hydrogen metabolism; and oxygen metabolism. Secondary metabolite production and antibiotic resistance genes were also included given their role in microbial interactions.
As stated above, we have strong evidence that each MAG here analyzed was transcriptionally active. We checked the expression of CDSs related to lignocellulose degradation, and determined that all CAZymes mentioned here are being expressed in the composting process (Fig. 2).
Transcripts coding for resistance genes were more abundant in the early (D01 and D03) and nal (D78 and D99) days (Supplementary Table S9).

Aerobic and Anaerobic respiration strategies
The analysis of oxygen metabolism indicates that nearly all bacteria from which MAGs were obtained are aerobes (Supplementary Table S10). The oxidases detected were all active and transcript abundance variation over time shows a slight decrease in D7, with an increase following the turning procedure  Table S4).
Sulfate reduction via cysteine desulfurase (cysCN) was detected as a widespread and active function among MAGs (Supplementary Table S10). Such mechanism of sulfate reduction is part of the assimilatory sulfate reduction pathway by which sulfate is incorporated into cysteine. Evidence for dissimilatory sul te reductase function (dsrAB and apsA) was not detected in the genomes, which is evidence that the respiratory sulfate reduction was not the main strategy for anaerobic respiration employed by the bacteria in this composting process.
Active denitri cation genes were detected in several MAGs (Supplementary Table S10). Eighteen of them presented the nitrite respiration gene nirK. The variant nirS was not detected in the MAGs. ZC4RG13 (s__Rhodothermus marinus),, ZC4RG22 (o__Luteitaleales), ZC4RG26 (s__Sphaerobacter thermophilus),, and ZC4RG29 (f__Cyclobacteriaceae) encode the complete denitri cation pathway (i.e., nitrous-oxide reductase pathway, nosDZ). ZC4RG29 lacks the nirK gene and has evidence for nitrite reductase using the nirB gene. Nitrous-oxide reductase genes (nosZD) were also active during the composting process. The variation in abundance of transcripts of these denitri cation genes increases over time, with a peak starting in D7 ( Supplementary Fig. S2).

Chemolithotrophic metabolism
We found evidence for chemolithotrophic metabolism based on MAG genes related to the oxidation of inorganic sulfur compounds (Supplementary Table S10). Nearly all MAGs have genes from the sulfur oxidation pathway via sulfur dioxygenase. Some of the MAGs were found to be more versatile and had genes annotated with other functions associated with the oxidation of sulfur compounds. ZC4RG20 (c__Gammaproteobacteria), for instance, represents a bacterial population that showed transcripts associated with sulfur dioxygenase, sul de oxidation (sqr), and thiosulfate oxidation (soxC), including transcripts associated with carbon xation via rubisco activation, supporting a chemolithoautotrophic growth. The Sox system, which is able to oxidize sul te and sulfone group in thiosulfate, was found in MAGs ZC4RG25 (f__Hyphomicrobiaceae), ZC4RG31 (f__Hyphomicrobiaceae), ZC4RG33 (g__Aquamicrobium),, and ZC4RG42 (o__Betaproteobacteria), although soxC was apparently lacking in all of them. Sul de oxidation (sqr) to thiosulfate was detected also in ZC4RG25. These observations highlight that members of the bacterial populations in the composting microbiome were capable of harvesting energy by oxidizing inorganic sulfur compounds. CDSs associated with nitri cation genes (amo and hao) were not detected in the MAGs, indicating that this trait was of minor or no relevance for the composting microbiome.
Several hydrogenases were found to be present and expressed (Supplementary Table S10

Co-occurrence of MAGs
Using the 49 ZC4 MAGs, we inferred correlation patterns using their variation in abundance (metagenomic datasets) and in activity (metatranscriptomic datasets) during the composting process.
The correlation patterns derived from the abundance pro le resulted in a graph composed by 40 nodes and 107 interactions, and four clusters (Fig. 5a). The correlation patterns derived from the activity pro le resulted in a graph composed by 43 nodes and 76 interactions, and three clusters (Fig. 5b).
We observed a high concordance between both correlation analyses. Fifty-ve interactions observed in the graph based on metagenomic datasets (Fig. 5a) are also present in the graph based on metatranscriptomic datasets (Fig. 5b). The correlations based on the activity of MAGs give us predictions on who are the key microbial players in the composting process as well as when and with whom they interact. In what follows we describe the main features of each activity cluster, highlighting high number of transcripts for selected cluster member MAGs on particular days (Supplementary Table S11).
Cluster 1: Seven MAGs form this cluster. Transcripts from members of this cluster are more abundant on initial days (D01 and D03) with a slight later increase on D64 (immediately after turning), followed by another increase on D99. Cluster members ZC4RG02 (g__Pseudoxanthomonas),, ZC4RG04 (T. bispora),, and ZC4RG28 (f__Streptosporangiaceae) presented the highest number of transcripts in the initial days.
Cluster 2: This cluster is composed of 10 MAGs, all Firmicutes and mostly abundant and active between D3 and D15, followed by a peak on D64. Many transcripts from ZC4RG12 (g__Caldicoprobacter) and ZC4RG32 (s__Caldicoprobacter faecalis) related to lignocellulose degradation were identi ed, especially on D3, D7, and D15.

Metabolic dependencies based on genome-scale models
Based on the results obtained with the co-occurrence analysis (Fig. 5b) and the activity of relevant functional genes (Supplementary Tables S7, S9, S10, S11, Fig. 2, and Supplementary Fig. S2), we identi ed MAGs according to their importance in the different stages of composting and the main functions associated with them (Fig. 6). We used this model in turn to assess the metabolic dependencies between these MAGs based on genome-scale models (Supplementary Fig. S3). The results revealed strong dependencies between some of the MAGs, as they received a maximum dependency score (Table 1). According to the models obtained, the most frequent compounds involved in the interactions between MAGs are hypoxanthine, H + , uracil, and phosphate. ZC4RG26 (s__Sphaerobacter thermophilus) had the highest level of dependency from other MAGs, predicted to be a metabolite receiver of 11 compounds, followed by ZC4RG08 (s__Pseudomonas thermotolerans) and ZC4RG20 (c__Gammaproteobacteria), predicted to receive six and ve compounds, respectively. ZC4RG04 (s__Thermobispora bispora) and ZC4RG22 (o__Luteitaleales) were predicted to have the highest number of interactions as metabolite producers. They were both predicted to be donors of seven compounds, followed by ZC4RG28 (f__Streptosporangiaceae), predicted as donor of six compounds. The set of keystone MAGs during the nal composting stage (D99) presented higher possibilities for strong metabolite dependencies between MAGs compared to the other stages (Table 1).
In order to test if the predicted metabolic interactions are likely to be speci c of the bacteria strains that were able to thrive in the composting microbiome, we chose three of the MAGs assigned to known species that were more frequently involved in the metabolic interactions model (ZC4RG26, ZC4RG08, ZC4RG04) and replaced them by the respective reference genomes from GenBank according to the taxonomic assignment based on GTDB: Sphaerobacter thermophilus (RefSeq: GCF_000024985.1), Pseudomonas thermotolerans (RefSeq GCF_000364625.1), and Thermobispora bispora (RefSeq GCF_000092645.1). The results show that several metabolic exchange interactions were lost (Fig. 7). For instance, in the models built using the RefSeq genomes, S. thermophilus and T. bispora were not able to share metabolites with each other. On the other hand, the RefSeq P. thermotolerans was able to share palmitate with other nodes, while ZC4RG08 (s_P. thermotolerans) was not (Fig. 7).

Discussion
Here we present detailed analyses of metagenome-assembled genomes (MAGs) from a composting process. A total of 60 environmental genomes of composting bacteria were obtained, of which 47 are potentially new bacterial species.
Lignocellulose breakdown depends on the e ciency of sacchari cation, which is often considered a bottleneck in the composting process [9]. Therefore, in addition to an analysis of MAGs and genes directly associated with biomass degradation, we also investigated other functions involved in microbial interactions and energy metabolism. We also carried out a co-occurrence analysis of MAGs based on their activity pro les during the process. Based on all these results, we propose here a framework for the molecular-microbial temporal dynamic in the composting process that we have studied (Fig. 6).
MAGs from clusters 1 and 2 (Fig. 5b) are the main constituents of the composting stage characterized by days 1, 3 (composting start), and 64 (recapitulation of composting start after the turning procedure) (Fig.  6). These MAGs represent bacterial populations expressing genes such as ARGs and classi ed with functions related to secondary metabolite production. These activities could be explained by intense competition between microorganisms. Indeed, these stages have high microbial diversity [3]. According to this interpretation, secondary metabolite production and ARG expression would be the consequence of the arms-shields race taking place in the composting microbial community [40][41][42]. Among the MAGs in these clusters, ZC4RG03 (g__Calditerricola) and ZC4RG07 (g__Bacillus) are the major producers of secondary metabolites. We hypothesize that these MAGs represent important bacterial players able to produce compounds with selective antimicrobial activity against pathogenic and opportunistic competitor bacteria; at the same time they seem able to consume easily degradable compounds, which we assume are particularly abundant during the initial composting stages (or right after the turning procedure). Niche protection through antagonistic competition has been observed in microbial systems associated with the human and animal gut [43][44][45][46].
According to our framework (Fig. 6), MAGs from cluster 2 are primarily active between days 1 and 15 (and on D64) (Supplementary Table S4). All of them are Firmicutes encoding hydrogenases. H 2 -oxidizing bacteria can use the molecular hydrogen produced during fermentative conversion of organic compounds [47], which possibly justi es the relevance of MAGs with hydrogenase activity in this stage. Hydrogenases participate in the mechanism that allows bacteria to store metabolic energy as an electrochemical potential across the membrane via the proton-motive force [48]. H 2 metabolism coupled with CO 2 as carbon source, which allows autotrophic growth via the acetyl coenzyme A (i.e., the Wood pathway) [49], is a metabolic strategy employed by acetogenic bacteria. These bacteria are strict anaerobes, most of which are also capable of heterotrophic growth [48].
In this context, Firmicutes MAGs ZC4RG12, 32, and 34 (Supplementary Table S1) (all belonging to cluster 2) would be the putative acetogens in the composting microbiome (Supplementary Table S10). Acetogens are ubiquitous in nature, including in thermophilic environments, and often constitute a fundamental group in the digestive tract of animals [48][49][50][51][52]. The latter could be the main source of acetogens for the composting microbiome analyzed here, given that a considerable portion of the composting material is made of feces from Zoo animals [3]. It has been reported that acetogens can outcompete methanogens and sulfate reducers given their higher metabolic exibility [48,53]. This could explain why these other two groups of bacteria were not detected among the recovered MAGs (Supplementary Table S10).
After day 15 we conjecture that easily degradable organic nutrients become more scarce [3], allowing other populations that can degrade more recalcitrant material to become dominant. Our results suggest that these populations are represented primarily by ZC4RG13 (s__R. marinus),, ZC4RG22 (o__Luteitaleales), and ZC4RG29 (f__Cyclobacteriaceae), all from cluster 3 (Fig. 6). Around this time oxygen probably becomes more limited and denitri cation processes come into play (Supplementary Fig.  S2). Accordingly, the above MAGs, including ZC4RG26 (s__S. thermophilus),, express CDSs annotated as nitrous-oxide reductase (including nosZ and nosD), which is the last step in denitri cation, being therefore able to perform the complete pathway (Supplementary Table S10). Anaerobic respiration using nitrous oxide (N 2 O) is a widespread trait in prokaryotes, however not all denitri ers encode this nal step in denitri cation [54]. Here we were able to detect four MAGs encoding nitrous-oxide reductase (Supplementary Table S10). One of them is an Acidobacteria (ZC4RG22) (also from cluster 3), and to our knowledge this is the rst report of a nitrous-oxide reductase in this clade [54]. The ability to utilize nitrous oxide for respiration might be crucial to improve e ciency of nitrogen use in the composting process. In this context, it is worth mentioning that N 2 O is a potent greenhouse gas, and microbial conversion of N 2 O to N 2 is the unique sink known for N 2 O in the biosphere [55].
The following stages in the composting process (represented by days 30, 78 and 99) are mainly dominated by MAGs from cluster 3 that perform lignocellulose degradation, denitri cation, and sulfur oxidation. In these stages there is a decrease in the overall phylogenetic microbial diversity [3]. Nevertheless, cluster 3 contains the largest and most diverse group of MAGs. We hypothesize that in these stages most nutrients derive from recalcitrant material (e.g. lignin). Examples of MAGs from cluster 3 are ZC4RG13 (s__R. marinus),, ZC4RG08 (P. thermotolerans),, and ZC4RG20 (c__Gammaproteobacteria). ZC4RG20 has a high potential of AAs compared to other MAGs, and ZC4RG08 (s__P. thermotolerans) is the only one to have genes annotated as belonging to the AA1 family. AA1 enzymes that have been experimentally studied are multicopper oxidases that use diphenols and related substances as donors, with oxygen as acceptor, and known for their role in the enzymatic conversion of recalcitrant polysaccharides such as lignin [56].
By applying metabolic modeling methods, we were able to predict metabolic dependencies between MAGs (Supplementary Fig. S3). These results suggest that metabolic interactions in composting can be determined by complementary functions found in the genomes of producers and consumers (Supplementary Fig. S3 and Fig. 7). According to the Black Queen hypothesis, in order to increase tness, one microbial population may lose genes related to a function when that function is already provided by another microbial population in the community. Therefore, the genomic differences between closely related strains are likely to be driven by local adaptation and coevolutionary interactions [57,58], which could explain why composting bacteria can present different metabolic dependencies compared with closely related strains that were isolated from other environments (Fig. 7).
Hypoxanthine and uracil are among the frequently exchanged compounds between keystone players (Table 1). It is known that many microbial groups, including members of the phylum Firmicutes, are auxotrophic for both purines and pyrimidines and rely on the salvage pathway for growth [59].
Dependencies on H + exchange can be associated with hydrogenase activity (Supplementary Table S10) and proton ux across membranes, which is related to ATP synthesis, pH homeostasis, and maintenance of solute gradients [60]. A metabolite transported into the extracellular environment as a waste product by one bacterium is often used by neighboring bacteria [9,58]. This could also explain the O 2 exchange ux predictions (Table 1). Oxygen can be a product derived from reactive oxygen species (ROS) detoxi cation systems, such as those encoded by chlorate-reducing bacteria via chlorite dismutase [61] (Supplementary  Table S10). Due to intense redox activity, ROS-detoxi cation is a vital function in the composting microbiome, as observed also by the overall pro le of dominant functions detected in the metatranscriptomic dataset (Supplementary Table S12). Another important pathway detected in the activity pro le is related to the phosphate starvation response (Supplementary Table S12), which is consistent with the frequency of metabolic dependencies based on phosphate exchange between MAGs (Table 1).

Conclusions
The results obtained here expand our knowledge of the taxonomic and functional diversity of composting bacteria. Based on our results we propose a model that predicts the key players in the composting process at each of the major stages. This model includes predicted metabolic interactions among the bacterial populations that we have identi ed, therefore providing unprecedented level of detail for microbial-molecular processes in composting.
Our results emphasize the importance of functions involved in e cient bioenergetic processes during biomass conversion, such as N 2 O reduction, sulfur oxidation, and hydrogenases. Taken together, our results contribute to future research aiming at the engineering of e cient biomass-degrading microbiomes.

MAGs recovery from composting metagenomic data
The composting metagenomic datasets on which this study is based have been described previously [3].
Brie y, the samples come from the composting facility at the State Zoological Park in the city of São Paulo, Brazil. Two composting cells were sampled: one called ZC3 and the other ZC4. For both, composting lasted 99 days. For ZC3, samples come from days 1, 30, 64, 78, and 99, and for ZC4 they come from days 1, 3, 7, 15, 30, 64, 67, 78, and 99). A turning procedure was performed on day 65 for ZC3 and on day 63 for ZC4. DNA shotgun sequencing was done for all samples, and metatranscriptomic sequencing was done for ZC4 samples.
Shotgun metagenomic reads from all samples were ltered and soft-trimmed (Q ≥ 12) using BBDuk from the BBTools package (https://jgi.doe.gov/data-and-tools/bbtools/). Reads with length shorter than 80 bp were removed and the remaining reads were de novo assembled using Spades (k-mer = 21,33,55,77,99,113,121,127,-meta) [62]. To obtain MAGs, the following steps were carried out, for each composting cell (ZC3 and ZC4) ( Supplementary Fig. S4): 1) reads from all samples were assembled and the contigs binned with Metabat2 (31). 2) reads from individual samples were assembled and the corresponding contigs were binned using Metabat2. After these two steps, only bins with completeness at least 50% and contamination at most 10% were kept for further processing, based on results from CheckM [22]; 3) bins from individual samples (step 2) were compared with each other using Mash [63], which allowed us to establish when the "same" bin occurred on different days (Mash distance at most 0.05); 4) for each "distinct" bin determined in step 3 its reads were reassembled and the results rebinned with Metabat2 and MyCC [64]; 5) bins from step (1) and those from step (4) were compared, again using Mash; 6) the MAGs selected for additional analyses were those distinct bins with best completeness and contamination results (when there was more than one bin for the same MAG), provided completeness was at least 80% and contamination at most 11% ( Supplementary Fig. S4 and Table S13).
Taxonomic assignment MAG taxonomic assignment was based on GTDB [65,66]. For those assignments that reached the species level, we carried out further comparisons with reference genomes of those species (whenever available, with complete genomes). These comparisons were done with the ANI tool (http://enveomics.ce.gatech.edu/ani/) and with GGDC [67]. In the text, we refer to MAGs by their identi ers, providing in parenthesis the GTDB classi cation, which uses the pre xes p, c, o, f, g, and s for phylum, class, order, family, genus, and species, respectively.

Functional Annotation
MAGs were annotated using the NCBI Prokaryotic Genome Annotation Pipeline [68]. Their protein-coding gene sequences were compared against the Clusters of Orthologous Groups (COGs) [69] database using rpsblast+ [70], with a cut-off e-value of at most 10 -5 . COG categories were assigned to the best hits with cdd2cog script (https://github.com/aleimba/bac-genomics-scripts/tree/master/cdd2cog).. The amino acid sequences of the predicted coding sequences (CDSs) were classi ed using the dbCAN HMM-based database for carbohydrate-active enzymes (CAZymes) [56]. CAZymes are categorized in different classes and families, including key enzymes for lignocellulose degradation such as glycoside hydrolases (GHs) and auxiliary activities (AAs), and the following complementary enzymes: glycosyltransferases (GTs), polysaccharide lyases (PLs), carbohydrate esterases (CEs) and carbohydrate-binding modules (CBMs).
The CDSs were also classi ed using a set of HMMs for detecting other metabolic pathways of interest [71], such as genes involved in denitri cation, sulfur metabolism, hydrogen metabolism, and oxygen metabolism. Evidence supporting MAGs with strictly anaerobic metabolism was obtained based on the consistency between the results provided by TRAITAR and CDS classi cation as oxidases [72]. The global pro le of functions in the metatranscriptomic dataset was determined using FMAP [73] (Supplementary   Table S12).
We analyzed the presence of antibiotic resistance (AR) in our MAGs by comparing protein-coding sequences against the CARD database (April 2019) [74]. We ltered out all results below 70% identity and 85% sequence coverage.
We used antiSMASH (v.5.0) to nd gene clusters involved in the biosynthesis of secondary metabolites [75].

Abundance and activity pro les of MAGs
The abundance pro le of MAGs across the metagenomic datasets was obtained using the function quant_bins provided by metaWRAP followed by normalization based on TPM (transcripts per kilobase million) [76], which translates to genome copies per million reads in our context. Similarly, relative abundance of expressed CDSs was obtained by determining metatranscriptome reads that mapped to CDSs using BEDTools [77], followed by normalization based on TPM. We use the term transcripts to refer to MAG CDSs to which metatranscriptome reads could be mapped.
Co-occurrence of MAGs based on their abundance and activity pro les In order to identify patterns of co-occurring bacteria represented by MAGs in the ZC4 datasets, we performed correlation analysis based on relative abundance of MAGs and their transcripts, as described.

Metabolic interaction models
Based on MAG co-occurrence patterns and the relative abundance of transcripts with annotation related to biomass degradation, denitri cation, sulfur metabolism, hydrogen metabolism, and oxygen metabolism, we de ned subsets of MAGs according to their importance in the different stages of the composting process. For each subset we built a metabolic interaction model using SMETANA [80], based on genome-scale metabolic reconstructions that were obtained from les annotated in the PATRIC platform [81] (at the closest taxonomic level possible). The results were submitted to KBase [82] in order to run the Build Metabolic Model function, including the default option gap lling, which relies on the ModelSEED Biochemistry Database [83]. With this method, metabolic genes were mapped onto biochemical reactions, and this information was integrated with information on stoichiometry reactions, subcellular localization, biomass composition, and estimation of thermodynamic feasibility, in order to produce a detailed stoichiometric model of metabolism at the genome scale. Metabolic dependency score calculated by SMETANA is normalized to a range between 0 and 1, meaning complete independency and complete dependency, respectively [80]     Antibiotic resistance genes (ARGs) pro le in ZC4 MAGs.  Schematic representation of keystone microbial players according to their importance in the different stages of composting. MAGs are represented as roughly circular numbered shapes, and their colors re ect the cluster they were assigned to (Fig 5b and Supplementary Table S4). Lignocellulose breakdown and relevant active functions are represented across the stages by the various symbols, and the irregular background shapes connect MAGs that express the same functions. The turning procedure was performed on day 63, therefore day 64 is considered the recapitulation of composting start, based on the patterns of microbial functions and activity that we observed in the present study and in our previous work [3].

Figure 7
Major metabolic dependencies between genomes of known bacteria species in the composting process studied. ZC4RG04, ZC4RG08, and ZC4RG26 were identi ed as the known bacteria species presenting major levels of metabolic dependencies ( Table 1). The gure represents all the strong metabolic interactions (i.e., maximum dependency score) predicted for these MAGs (a). In panel (b) these MAGs were replaced by their closest relatives in the NCBI RefSeq database, and the consequent predicted changes in metabolic interactions are denoted with fading colors and by dotted arrows.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.