Deciphering a Marine Bone-Degrading Microbiome Reveals a Complex Community Effort

Bones are an underexploited, yet potentially profitable feedstock for biotechnological advances and value chains, due to the sheer amounts of residues produced by the modern meat and poultry processing industry. In this metagenomic study, we decipher the microbial pathways and enzymes that we postulate to be involved in bone degradation in the marine environment.

linkages. Finally, in the digestion of the cholesterol-containing marrow, cholesterol oxidases may be involved.
To date, only a few studies have been published that focus on microbial communities to understand the necessary complex interactions in bone degradation, mainly relying on 16S rRNA gene sequencing data (30)(31)(32)(33) and one metagenomic study of a whale fall (34). We here provide a first comprehensive overview and identify putative key functions involved in bone degradation of the marine bone microbiome retrieved from deployed bone material, including microbial communities from the gutless worm Osedax mucofloris and free-living microbial assemblages developing on the bone surface.

RESULTS
Recovery of artificially deployed bone for bone microbiome metagenomic analysis. Turkey and bovine bones were deployed at 69 m depths in Byfjorden, a fjord outside Bergen, Norway. After 9 months of incubation, underwater images taken by a remotely operated vehicle (ROV) showed microbial colonization of the bone surfaces (see Fig. S1A in the supplemental material). Osedax mucofloris worms were observed, especially on the joints, and in some cases forming small colonies of several individuals inside a single cavity under the external bone tissue. Although not the subject of this study, a larger diversity of invertebrate fauna including Ophryotrocha, Vigtorniella, and Capitella worms were also observed. Dense microbial mats developed asymmetrically with preference for the joint adjacent sections (epiphysis), which also appeared in aquarium settings (Fig. S1B).
Metagenome-assembled genomes (MAGs) from the marine bone microbiome display taxonomic diversity and novelty. Fifty-nine high-quality MAGs ( Fig. 1) (.90% completion and ,10% redundancy) were extracted from the coassembled metagenomes (see Table S3 for MAG sequence statistics). The MAGs span 11 phyla, 14 classes, 19 orders, and at least 23 families. About 63% of the MAGs (37/59) possess genomic novelty as determined by their relative evolutionary divergence (RED) (36) to their closest common ancestor (Table S2). One MAG could be identified only up to phylum level, seven to class level, seven to order level, 18 up to family level, and four up to genus level. The taxonomy of most MAGs was fully resolved based on 120 marker genes. The three best represented phyla were Proteobacteria (22 MAGs), Campylobacterota (14 MAGs), and Bacteroidota (8 MAGs). However, the percental distribution of the most abundant  11.12.2017 classes differs between the two metagenome sets. The OB-MAGs were dominated by the classes Gammaproteobacteria (27%), Campylobacteria (27%), and Alphaproteobacteria (20%), while the BB-MAGs were mainly affiliated with Gammaproteobacteria (30%), Campylobacteria (23%), and Bacteroidia (16%). Sulfur cycling in the marine bone microbiome. All MAGs were investigated using the multigenomic entropy-based score pipeline (MEBS) (37) for their ability to utilize sulfur as energy source via the abundance of selected marker genes for related pathways (Fig. 2). Sulfur cycling is of relevance to bone degradation due to the generation of free protons by sulfur and sulfur compound oxidation processes (thiotrophy), which leads to an acidification.
MAGs affiliated with the order Campylobacterales encode almost complete Sox (soxXYZABCD) and Sor (sorABDE) enzyme systems for sulfur oxidation. Furthermore, Gammaproteobacteria affiliated with Beggiatoaceae and other unclassified Gammaproteobacteria are all potentially capable of thiotrophy via utilization of reduced sulfur compounds FIG 1 Phylogenomic maximum likelihood tree of the obtained 59 high-quality MAGs from Osedaxassociated bone microbiome (OB) and bone surface-associated biofilm (BB). Bootstrap values greater than 0.5 are displayed as circles on the branches. The five most common bacterial classes are colored (green, Bacteroidia; light purple, Desulfuromonadia; purple, Campylobacteria; orange, Alphaproteobacteria; and blue, Gammaproteobacteria). Order-level identifications are listed, and MAGs for which only class-level identification could be inferred are marked with an asterisk. as electron donors (flavocytochrome c sulfide dehydrogenase [(fccB] and adenosine-59-phospho-sulfate reductase [aprAB]) and partial predicted Sox sulfur/thiosulfate oxidation pathway. MAGs identified as Desulfuromonadia, Desulfobacteria, Desulfobulbia, and Desulfovibrionia possess marker genes for sulfur reduction (qmoABC, hydACD, sreABC) and lack Sox and Sor pathway genes, all of which belong to the new proposed phylum of Deltaproteobacteria (38). In all Gammaproteobacteria (except BB32), Desulfobacteria, Desulfobulbia, and Desulfovibrionia, genes for dissimilatory sulfite reductase (dsrABC) are present. Müller et al. (2015) described that gammaproteobacterial dsrAB-type genes are commonly involved in oxidative reactions, whereas dsrAB in Desulfobacterota are reductivetype dsrAB (39). All MAGs contain at least partial pathways for dissimilatory tetrathionate reduction (ttrABC), thiosulfate disproportionation (phsABC and rhodanase), and dimethylsulfide (DMS) degradation (ddhABC) and contain also genes for sulfoacetaldehyde degradation (isfD, xsc, and safD). In addition, the phenotypic trait of H 2 S production was identified in 10 MAGs (Traitar analysis [40]), two of which were Marinifilaceae, two Krumholzibacteria, two Sulfurospirillum, one Spirochaetaceae, two Desulfobacteraceae, and one Pseudodesulfovibrio.
The anticipated thiotrophy has the potential to contribute massively to the acidification of the environment via the oxidation of reduced sulfur compounds leading to production of sulfuric acid (41). This requires a close interaction between sulfur-reducing bacteria (SRB) producing hydrogen sulfide and sulfur-oxidizing bacteria (SOB) utilizing the hydrogen sulfide, while releasing protons (Fig. 3). The Traitar analysis identified 10 MAGs potentially able to produce hydrogen sulfide, including known SRB like Desulfobacteraceae, Pseudodesulfovibrio, and others like Sulfurospirillum (42)(43)(44). The bone microbiome is especially enriched in taxa containing known SOB, like the large filamentous bacteria Beggiatoales (5 MAGs) (45) and Campylobacterales (10 MAGs) (46). Furthermore, one MAG identified as Desulfobulbaceae was found in the boneassociated metagenomes. Members of this group are known to be able to perform sulfur oxidation and sulfur reduction (47,48).
Acidification by carbonic anhydrases. Carbonic anhydrases were identified in 51 of 59 MAGs. Nineteen out of 94 carbonic anhydrases contained a signal peptide for  (37), and MAGs were phylogenetically grouped according to the GTDB-Tk pipeline (36). The gray scale represents the completeness of a given pathway or multienzyme system shown in the heatmap for each MAG. The OB MAGs are highlighted in blue. The "Sox system" is constituted by the soxXYZABCD genes, and the "Sor system" consists of sorABDE. extracellular export (16 MAGs). Fifteen were predicted to contain a Sec signal peptide (SPI) and four to encode lipoprotein (SPII) signal peptides. Four out of five Beggiatoales MAGs were predicted to contain carbonic anhydrases with an SPI signal peptide (BB2, BB3, and BB20) or an SPII signal peptide (BB16). The remaining three SPII signal peptides were found in carbonic anhydrases from Campylobacterales (BB8, BB10, and BB11). Interestingly, BB8 contains at least three carbonic anhydrases, one with an SPI, one with an SPII, and one where a signal peptide was not predicted. Three SPI signal peptides were found in carbonic anhydrases from unclassified gammaproteobacteria (BB4, BB34, and BB36). The remaining SPI including carbonic anhydrases were found in five Campylobacterales MAGs (BB14, BB26 [three carbonic anhydrases and two containing SPI signal peptides], BB30, BB41, and OB7) and one Desulfobulbaceae MAG (BB13). Based on phylogenetic relationship to known carbonic anhydrases described in the work of Capasso and Supuran (49), 18 out of 19 carbonic anhydrases belong to the a-carbonic anhydrase family and one to the b-family, with no g-family carbonic anhydrases found (Fig. S2).
Gelatin hydrolysis. With respect to microbial bone degradation, the phenotypic feature of gelatin hydrolysis was analyzed using the Traitar software (40), which provides genome-informed phenotype predictions. Twenty-two MAGs showed capacity for gelatin hydrolysis (19 MAGs in the bone surface community [BB] and three in the Osedax-associated communities [OB]). With gelatin being a primarily bone collagen-derived compound, we consider gelatin hydrolysis a key trait for the microbial community studied here. All eight Bacteroidia-affiliated MAGs possess the gelatin hydrolysis trait, as do seven Gammaproteobacteria MAGs, one tentative Planctomycetota MAG, one Spirochaetia MAG, two Krumholzibacteria MAGs, one Thiovulaceae MAG, one Geopsychrobacteraceae MAG, and one Fermentibacteria MAG (Fig. 3).
Collagen degradation. We investigated the genomic context of each M9 collagenase for potential links to metabolic pathways, such as proline utilization (Fig. S3). Colwellia MAG BB5 possessed an approximately 21-kbp-long gene cluster presumably devoted to collagen utilization, which is unique in the data set and in the public databases. The functional cluster spans at least 15 different genes (Fig. 5A), featuring a secreted Zn-dependent M9 collagenase, a secreted peptidyl-prolyl cis-trans isomerase (cyclophilin-type PPIase [peptidyl-prolyl cis-trans isomerase]), a secreted unknown protein, and an unknown Zn/Fe chelating domain-containing protein. Additionally, one putative transporter (major facilitator superfamily [MFS] family), a TonB-dependent receptor, and several genes involved in the catabolism of proline and hydroxyproline, e.g., prolyl-aminopeptidase YpdF, intracellular peptidyl-prolyl cis-trans isomerase (rotamase), pyrroline reductase, hydroxyproline dipeptidase, 4-hydroxyproline epimerase, and others. Moreover, genes involved in transcription regulation such as PutR and the stringent starvation proteins A and B were identified in the same putative gene cluster.
Specificity of the bone microbiome. To investigate the potential specificity of the bone microbiome in respect to its taxonomic composition and function, the generated MAGs were compared to 832 seawater MAGs generated from the Tara Oceans data sets, comprising 93 metagenomes from various locations (64). The bone microbiome was rebinned to retrieve all MAGs with .70% completion to match the Tara Oceans threshold for high-quality MAGs, resulting in 86 MAGs. The taxonomic class abundances were compared between the two data sets, and Gammaproteobacteria and Bacteroidia appeared in similar abundances (bone microbiome, 23.26% and 15.12%, and Tara Oceans, 23.56% and 13.34%, respectively), but other than those high taxonomic ranks, the data sets were different. The bone microbiome is dominated by Campylobacteria, accounting for 26.74% of the MAGs, whereas only 0.12% Campylobacteria MAGs are found in the Tara Oceans data set. In contrast, Alphaproteobacteria are better represented in the Tara Oceans data set than in the bone microbiome, 21.88% to 5.81%, respectively. The bone microbiome furthermore also contains a number of bacterial classes that are represented at low levels (1 to 5% of the MAGs), which are either not represented in the Tara Oceans data set or represented only at minuscule levels as low as 0.1 to 0.5% (Fig. S4A). The functional repertoire was compared via screening the Tara Oceans data set with the previously generated HMM profiles for enzymes potentially involved in bone degradation and calculated as ratio of enzymes per MAG. The ratios for 9 out of 12 profiled enzyme families were higher in the bone microbiome, ratios for a-Nacetylgalactosaminidases (COG0383) and cholesterol oxidases (COG2303) were higher in the Tara Oceans data set, and the ratio for S1 peptidases was equal between the two MAG sets (Fig. S4B).

DISCUSSION
In this study, 59 high-quality MAGs were reconstructed from microbes colonizing bone surfaces and from symbionts of the bone-eating worm Osedax mucofloris. Metabolic reconstruction revealed a complex, diverse, and specialized community. Our MAGs span at least 23 bacterial families and uncover a large potential for taxonomic novelty (over 50% according to genome-based taxonomy) from species up to class level in the bone microbiome. Interestingly, only genomes of Gram-negative bacteria were reconstructed, and despite Gram-positive bacteria being widespread in the marine environment, they make up only minor portions of the metagenomes (4 to 5% of the reads affiliated with Actinobacteria and Firmicutes in the Osedax metagenomes and 5 to 9% in the biofilm metagenomes, respectively) (65). This is remarkable since they are known to carry out potentially relevant metabolic processes (thiotrophy, sulfidogenesis) (66,67), are capable of dealing with low-pH conditions which are likely encountered during bone dissolution (68), and possess high capacity for the secretion of hydrolytic enzymes (69). Despite this underrepresentation of Gram-positive taxa, this study reveals the existence of a specialized bone-degrading microbiome in the marine environment and starts to explore the enzymatic activities involved in the complete demineralization of bone material. The bone microbiome is different from seawater communities and from other specialized habitats (recolonized volcanic eruption site) in its microbial composition as well as functional makeup (see Fig. S4 in the supplemental material).
The role of Osedax endosymbionts in bone utilization. Two distinct bacterial endosymbiont genomes belonging to the order Oceanospirillales have previously been sequenced, but their role in bone degradation in the marine environment remained unclear (18). The bacterial fraction of the here-sequenced Osedax mucofloris metagenome is made up of 7% to 22% Oceanospirillales-affiliated reads, whereas the bone surface metagenome contains only 1% to 4% reads of this order according to the performed Kaiju analysis. This relative difference confirms that the methodological approach to minimize cross-contamination was successful and that the OB-MAGs affiliated with Oceanospirillales likely represent the symbiotic community of Osedax mucofloris worms. Two MAGs belonging to the Oceanospirillales were identified in the Osedax-associated metagenome, belonging to the genera Neptunomonas (OB1) and Amphritea (OB2). Both genera are known to have an aerobic organotrophic metabolism and are also able to thrive as free-living bacteria (70). In fact, the scarce representation of Oceanospirillales in the bone surface has been reported before (30,31) and contrasts with their character as common dwellers in the marine environment (71,72). Although we cannot rule out that Oceanospirillales preferentially colonize bone surfaces in earlier or later stages, their preference for a symbiont life supports the notion of a casual and facultative association with Osedax worms, triggered by the common benefit from a sudden nutrient pulse as previously hypothesized (18,21).
The degradative functions within the bone microbiome. (i) Acidification via a closed sulfur biogeochemical cycle. Free-living microbial communities must deal with similar challenges as the Osedax holobiont to access the nutrient-rich, collagenmade organic bone matrix and eventually the lipid-rich bone marrow by dissolving the hydroxyapatite. The association in large specialized microbial consortia may be a beneficial strategy for achieving this task. We hypothesize that sulfur-driven geomicrobiology (sulfate/thiosulfate/tetrathionate reduction and sulfide/sulfur/thiosulfate oxidation) is the major factor responsible for bone dissolution in the marine environment by free-living bacterial communities. Campylobacterales are one of the most abundant bacterial orders in the here-investigated metagenomes, in both the Osedax-associated metagenomes (OB) and the bone surface biofilms (BB). Campylobacterales represent the most abundant group in terms of absolute read number, although it is the second largest taxon with reconstructed MAGs (Fig. 1). Members of the Campylobacterales have previously been found to be associated with Osedax, albeit not as endosymbionts (21). The majority of retrieved Campylobacterales MAGs (14 in total) belong to different families of aerobic and facultative anaerobic (nitrate, manganese) sulfur-oxidizing bacteria (73, 74) (Thiovulaceae, Sulfurovaceae, and Arcobacteraceae). Other aerobic/facultatively anaerobic (nitrate) sulfur-oxidizing bacteria are also well represented in the order Beggiatoales (Gammaproteobacteria, 5 MAGs). Beggiatoa-like bacterial mats are commonly associated with whale falls (75), indicating a potential indifference regarding the bone type they dwell on. Sulfide oxidation produces elemental sulfur or sulfate (41,45) while releasing protons and thereby causing a drop in pH. This acidification mechanism has been linked to bone demineralization. The dissolution of the hydroxyapatite mineral exposes the organic matrix to enzymatic degradation (30,31). Besides thiotrophy, which seems to be a major acid-producing mechanism in the microbial community, other mechanisms might also contribute significantly. In this respect, a number of carbonic anhydrases (CAs) were annotated, which are normally housekeeping genes involved in internal pH homeostasis and other processes (76) but known to play a role in environmental acidification by Osedax (15). Here, the CAs were found to contain signal peptides for extracellular export (19 out of 94) and therefore could also be involved in acidification. Interestingly, 18 out of 19 identified and potentially secreted CAs belong to the a-CA family and only one member of the b-CA family was found (Fig. S2). The a-CA family is found only in Gram-negative bacteria, which is also the case here, and it is evolutionarily the youngest of the three bacterial CA families (49).
Besides a large number of SOB, eight MAGs related to SRB were identified that are affiliated with the families Desulfobulbaceae (also SOB), Desulfobacteraceae, Geopsychrobacteraceae, and Desulfovibrionaceae. Moreover, they are prevalently associated with the free-living community attached to the bone surface in this study. Sulfate, tetrathionate, or thiosulfate can serve as electron acceptors and/or donors, and gene markers for all pathways are present in the genomes (Fig. 2). Microbial sulfidogenesis on the bone surface or the surrounding sediments can feed the thiotrophic community and therefore accelerate the demineralization process. The generated sulfide is known to quickly react with iron, blackening the bone surfaces with insoluble iron sulfide (77). In our incubation experiments, blackening occurs preferentially on the epiphysis, which is also where complex white/pink microbial mats are forming over time (Fig. S1B). However, from our analysis SRB seem unable to degrade large complex molecules. This is supported by the lack of bone-degrading enzymes here investigated, such as S8/S53 peptidases, mannosidases, sialidases, fucosidases, and a-galactosidases. SRB are likely dependent on the generation of simple organic compounds produced as metabolites by fermenters or aerobic organotrophic bacteria of the wider bone microbiome. The bone dissolution driven by sulfur geomicrobiology relies on other specialized members of the community to degrade the organic matrix and to fuel the acid generation.
(ii) Degradation of organic compounds via peptidases, glucosidases, and oxidases. Once the inorganic hydroxyapatite is removed, an array of different enzymes is required to digest the various organic bone components. Bacteroidia appear to be especially remarkable in this respect and represent the third most abundant taxon. Eight high-quality MAGs could be reconstructed, seven of them from the bone surface metagenome. Bacteroidia, and especially the family of Flavobacteriaceae, are known to be versatile degraders of polysaccharides like agar (78), chitin (79), ulvan (80), alginate (81), carrageen (82), and cellulose and xylanose (83) and polypeptides like elastin (84), spongin (85), and others. The recently described Marinifilaceae family (86) includes isolates that are reported to present xylanase activity (87). Despite the discrepancy between abundance and reconstructed genomes, the Bacteroidia MAGs appear to be the most versatile order of the investigated MAGs in respect to their richness in bone-degrading enzymes (Fig. 4), and all were predicted to possess the gelatin hydrolysis trait (Fig. 3). They were also the only MAGs containing sialidases (COG4409) and a-galactosidases (Pfam16499) (Fig. 4). Due to this taxon-specific trait and their presence being limited to the bone surface-associated microbiome, we hypothesize that Bacteroidia play a pivotal and specialized role in the free-living community via the degradation of specific organic bone components.
Differential microbial colonization of the spongy cancellous bone tissue over the cortical compact bone has also been observed in the terrestrial environment and has been related to easier access to the red marrow (88), although a priming effect linked to the differential composition of the bone cannot be ruled out. Complex microbial mats form preferentially on the epiphysis of the long bones, and this area is normally covered with hyaline cartilage (89) which was not removed before deployment. Cartilage is a related connective tissue made of nonfibrous type II collagen and a sulfated-proteoglycan matrix rich in N-acetylgalactosamine and glucuronic acid residues. This would explain the abundance of a-galactosidases, N-acetylglucosaminidase, and glucuronidases. Moreover, other groups such as Kiritimatiellales (PVC superphylum) are known marine anaerobic saccharolytic microbes specialized in degrading sulfated polymers that we find in this environment (90).
(iii) Collagen degradation by Gammaproteobacteria. Peptidases and especially M9 collagenases are of special interest for the degradation of the proteinogenic compounds within bone, as they are able to degrade collagen, the main source of carbon in this environment. The class Gammaproteobacteria is comparatively enriched in these enzymes, and it is the best represented class in the data set, with 17 MAGs. Of particular interest are the MAGs affiliated with the order Enterobacterales (two MAGs of the families Kangiellaceae and one Alteromonadaceae). They possess the gelatin hydrolysis trait (Fig. 3, MAGs BB5, BB44, and OB12), have a high number of S1 and U32 peptidases, and are the only MAGs with M9 collagenases. The Colwellia MAG BB5 is particularly remarkable as it contains an entire gene cluster dedicated to collagen utilization (Fig. 5A). The collagen degradation gene cluster comprises at least 15 different genes, including an M9 collagenase, a PepQ proline dipeptidase, an aminopeptidase YpdF, several transporters, epimerase, isomerases, and others. The gene cluster encodes nearly the entire pathway necessary to unwind and hydrolyze triple-helical collagen, transport and uptake of collagen oligopeptides into the cell, and utilization of its main components, mainly hydroxyproline and proline, for energy production via the tricarboxylic acid (TCA) cycle and/or the urea cycle or for polyamine biosynthesis (Fig. 5B). Accessory genes for an alternative catabolic route of proline to glutamate are located elsewhere in the genome (P5CDH). This kind of functional condensation for collagen utilization has not been described before in Colwellia or elsewhere. Interestingly, Colwellia bacteria are also one partner in a dual extracellular symbiosis with sulfur-oxidizing bacteria in the mussel Terua sp. 'Guadelope', retrieved from a whale fall in the Antilles arc, and are supposedly involved in the utilization and uptake of bone components (91). A cluster of functionally related genes was found in the publicly available genomes of Colwellia piezophila and Colwellia psychrerythraea. However, the gene cluster described for MAG BB5 contains several supplementary features, like a rotamase, a 1-pyrroline-4-hydroxy-2carboxylate deaminase, and a Zn/Fe binding domain protein potentially attributed to collagen utilization, which are absent in the published genomes (Fig. 6). Moreover, the gene cluster contains regulatory elements like the PutR regulator and stringent starvation proteins known to be activated under acid stress or amino acid starvation FIG 7 Hypothesis of the interplay in the marine bone microbiome and degradome. Sulfur-oxidizing bacteria (SOB, shown with a halo) convert elemental sulfur and H 2 S into sulfate and protons that lead to an acidification and therefore bone demineralization. Sulfate-reducing (SRB, green) and sulfurdisproportioning bacteria produce H 2 S from sulfate. Enterobacterales and others, especially Gammaproteobacteria, secrete collagenases to degrade collagen. Bacteroidia and other bacteria secret glycosidases and other enzymes to hydrolyze the organic bone components (glycosides, esters, lipids). This exemplifies a bone demineralization loop that fuels itself as long as sulfur is available and degrades the organic bone components in the process. conditions in Escherichia coli (92). This supports our hypothesis that other members of the microbial community need to dissolve the bone calcium phosphate via acid secretion, before collagen and other organic bone compounds can be accessed. In summary, the publicly available gene clusters lack regulatory elements to switch on the collagen utilization pathway under "bone-degrading"/acidified conditions and are missing key enzymes to exploit collagen's key components proline (rotamase missing to transverse D-proline to L-proline) and hydroxyproline (pyrroline-4-hydroxy-2-carboxylate deaminase missing that breaks down 1-pyrroline-4-hydroxy-2-carboxylate to alpha-ketoglutarate semialdehyde).
(iv) Bone degradation-a complex microbial community effort. The marine bone microbiome is a complex assemblage of various bacterial classes that requires the synergistic action of many different interwoven enzymatic reactions to access the recalcitrant bone material for its nutritional resources. Based on metagenomic predictions, we envision the following scenario of these complex processes (Fig. 7). The primary requirement in utilizing organic bone compounds is likely the dissolution of mineralized calcium phosphate (hydroxyapatite) by acidification, which can potentially be performed via proton release by a versatile community of sulfur-oxidizing (SOB) Gammaproteobacteria (mainly Beggiatoa-like), Campylobacterales (Sulfurimonas, Sulfurospirillum, Sulfurovum), Desulfobulbales, and Alphaproteobacteria. This acidification via thiotrophy may be fueled by sulfur-reducing bacteria (SRB), like Desulfobacteraceae, Geopsychrobacteraceae, and Pseudodesulfovibrio, creating a sulfur biogeochemical loop between SRB and SOB. Once the organic compounds (collagen, fatty acids, proteins, and peptidoglycans) are accessible, the Bacteroidia (Flavobacteriaceae and Marinifilaceae) and Gammaproteobacteria (Alteromonadaceae and Kangiellaceae) may become the main protagonists. These Bacteroidia are especially rich in bone-degrading enzymes, but importantly, the Gammaproteobacteria are the only members identified with M9 collagenases, and one genome identified as Colwellia contains an entire gene cluster dedicated to collagen degradation (Fig. 5). Here, we disentangled the potential functional roles of specialized members of the bone-degrading microbial community, which together make bone-derived nutrients accessible-not only to themselves but also to generalists within the bone microbiome. We posit that Flavobacteriales and Enterobacterales are the most promising candidates for novel enzyme discovery, as they display the most versatile sets of bone-degrading enzymes.

MATERIALS AND METHODS
Sample collection. Bone material after manual meat deboning was kindly provided by a local slaughterhouse operated by Norilia (Norway). Since deboning does not completely eliminate the animal tissue attached to the bone, some remains were still present. Therefore, in order to avoid bacterial colonization and decomposition, all bone material was kept at 220°C until deployment. Four sets of turkey thigh bones and one bovine lower leg bone were placed in a crab trap and deposited at the bottom of Byfjorden (60.397093N; 5.301293E) close to Bergen, Norway, at a depth of 69 m and approximately 150 m offshore in May 2016, incubated for 9 months, and retrieved using a small ROV ( Table 1). The material was transported to the lab in Styrofoam boxes either for processing within 2 h (bone surfaces) or for prolonged incubation in seawater aquaria and subsequent dissection of Osedax worms. The meatless bone surfaces were scraped with a sterile scalpel for microorganisms, and Osedax mucofloris specimens were extracted from the bone using sterile scissors and forceps. Their root tissue was dissected from the body, rinsed in sterile 70% (vol/vol) seawater, preserved in storage solution (700 g/liter ammonium sulfate, 20 mM sodium citrate, and 25 mM EDTA, pH 5.2), and stored at 270°C until further processing.
DNA extraction and sequencing. DNA was extracted from 10-to 50-mg samples of either scraped biofilm or Osedax root tissue, using the Qiagen AllPrep DNA/RNA minikit according to the manufacturer's instructions with cell lysis by a bead beating step in Lysing Matrix E tubes (MP Biomedicals) in a FastPrep homogenizer (MP Biomedicals) with a single cycle of 30 s at a speed of 5,500 rpm. The obtained DNA was quantified and quality controlled using a NanoDrop2000 (ThermoFisher Scientific) and a Qubit fluorometer 3.0 (ThermoFisher Scientific). The obtained DNA concentrations ranged from 11.9 ng/ml to 166 ng/ml according to Qubit readings. The DNA libraries were prepared using the Nextera DNA library prep kit according to the manufacturer's guidelines. Fifty nanograms of DNA was used for the preparations. In brief, the DNA was fragmented by the Nextera transposome at 55°C for 5 min and barcoded adapters were added in a 5-cycle PCR amplification. The resulting libraries, including ;140 bp of adapter, had an average fragment size of 436 bp (6112 bp) and were sequenced on an Illumina HiSeq4000 platform (150-bp paired-end reads) at the Institute of Clinical Molecular Biology (IKMB), Kiel University, Germany.
Metagenomic read profiling. Illumina raw reads were quality trimmed and adapters were removed with Trimmomatic version 0.36 (93). The quality-filtered reads were used individually and combined with respect to their sample source (either Osedax-associated or bone surface biofilms) to profile the taxonomic origin of the reads with Kaiju (35).
Metagenomic assembly, binning, taxonomic identification, ORF prediction, and annotation. For each sample type (Osedax mucofloris and bone surface biofilm communities), the quality-filtered metagenomic reads (Trimmomatic version 0.36) were coassembled with SPAdes v3.12 (94) for kmers 21, 33, 55, 77, and 99, with the metaSPAdes-assembler option enabled (see Table S1 in the supplemental material for read counts). Binning was conducted on the resulting assemblies using the MetaWRAP pipeline (version 1.0.1) (95). This pipeline combines initially three different binning methods, CONCOCT (96), MaxBin2.0 (97), and metaBAT2 (98), to generate MAGs. In the next step, all MAGs are combined in different MAG sets and a bin_refinement module is utilized to choose the best version of each MAG according to the desired minimum completion and maximum contamination levels. We considered only high-quality MAGs with .90% completeness and ,10% redundancy for further analyses. CheckM was used for quality assessment of the assembled genomes (99), and GTDB-Tk version 0.1.3 (36) was used for taxonomic identification, coupled with an estimate of relative evolutionary divergence (RED) to their next common ancestor. RED is a normalization method to assign taxonomic ranks according to lineage-specific rates of evolution, based on branch lengths and internal nodes in relation to the last common ancestor calculated by GTDB-Tk. Open reading frames (ORFs) of the obtained MAGs were predicted with Prodigal version 2.6.3 (100). Predicted ORFs were annotated using eggNOG-mapper v1 (101) with eggNOG orthology data version 4.5 (102). Additionally, the MAGs were annotated and metabolic models were calculated using the RAST (rapid annotation using subsystem technology) server (103,104). The MAGs were further investigated for the presence or absence of major metabolic pathways and phenotypic microbial traits based on their genomic sequences using MEBS (multigenomic entropy-based score, version 1.2) (37) and Traitar (40). MEBS is a software package used here to detect genes related to sulfur metabolism; this was done by providing protein fasta files of the high-quality MAGs that are annotated by MEBS with InterProScan (105) and are then searched with HMM profiles for genes related to sulfur metabolism. The sulfur metabolism-related genes investigated by MEBS are based primarily on the MetaCyc database (106). Traitar is a software package that can predict 67 phenotypic traits from a genome sequence. In brief, the analysis is based on known phenotypic traits of 234 bacterial species and infers from their genome Pfam families that are either present or absent in a specific trait. In this paper, the traits gelatin hydrolysis and H 2 S production are of interest and these are based on the presence of 70 and 43 and absence of 51 and 22 Pfam families, respectively. Phylogenomic trees were calculated with FastTree (107) as maximum likelihood trees and visualized with iTOL (108,109), and heatmaps were visualized with Heatmapper (110). Gene cluster maps were drawn with Gene Graphics (111). Signal peptides were predicted with the SignalP-5.0 server using nucleotide sequences to predict the presence of Sec/SPI, Tat/SPI, and Sec/SPII signal peptides in a given sequence (112).
Tara Oceans comparison. This comparison is based on the work of Delmont et al. (2018), who binned 93 metagenomes generated from the Tara Oceans project (64). The metagenomes represent 61 surface water samples and 32 samples from the deep chlorophyll maximum layer of the water column. They generated 957 nonredundant high-quality bacterial, archaeal, and eukaryotic genomes. The 957 MAGs were here reanalyzed with GTDB-Tk and the generated HMM profiles as previously described. Eight hundred thirty-two MAGs were identified as of bacterial origin and included in the comparison. The Tara Oceans MAGs were generated with a quality threshold of .70% completion; therefore, the bone metagenomes were rebinned (.70% completion, ,10% redundancy) for better comparison to avoid bias due to the higher threshold used for functional analysis in other parts of this paper.
The 59 high-quality MAGs analyzed in this study were deposited in the NCBI database as well, as part of the BioProject ID PRJNA606180; the BioSample accession numbers are SAMN16086327 to SAMN16086385 (biofilm MAGs 1 to 44, SAMN16086327 to SAMN16086370, and Osedax MAGs 1 to 15, SAMN16086371 to SAMN16086385).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. FIG S1, TIF

ACKNOWLEDGMENTS
This work is dedicated to the memory of Prof. Hans Tore Rapp and his effort on characterizing Osedax mucofloris.
We acknowledge financial support from ERA-NET Marine Biotechnology (GA no.: 604814) funded under the FP7 ERA-NET scheme and nationally managed from the German Federal Ministry of Education and Research and the Norwegian Research Council. M.F. acknowledges the support of the grants PCIN-2017-078 (within the Marine Biotechnology ERA-NET) and BIO2017-85522-R from the Ministerio de Economía y Competitividad, Ministerio de Ciencia, Innovación y Universidades (MCIU), Agencia Estatal de Investigación (AEI), and Fondo Europeo de Desarrollo Regional (FEDER) and Competitiveness. Additional funding was received from the Norwegian Biodiversity Information Centre (PA 809116 knr. .
We thank Hans T. Kleivdal for early developments of the concept. We thank Norilia AS for supplying bone residue material for the field work and ROV AS for providing underwater services during the deployment and sampling campaign. We acknowledge Kira S.