Illuminating Key Microbial Players and Metabolic Processes Involved in the Remineralization of Particulate Organic Carbon in the Ocean’s Twilight Zone by Metaproteomics

ABSTRACT

The twilight zone (from the base of the euphotic zone to the depth of 1,000 m) is the major area of particulate organic carbon (POC) remineralization in the ocean, and heterotrophic microbes contribute to more than 70% of the estimated remineralization. However, little is known about the microbial community and metabolic activity directly associated with POC remineralization in this chronically understudied realm. Here, we characterized the microbial community proteomes of POC samples collected from the twilight zone of three contrasting sites in the Northwest Pacific Ocean using a metaproteomic approach. The particle-attached bacteria from Alteromonadales, Rhodobacterales, and Enterobacterales were the primary POC remineralizers. Hydrolytic enzymes, including proteases and hydrolases, that degrade proteinaceous components and polysaccharides, the main constituents of POC, were abundant and taxonomically associated with these bacterial groups. Furthermore, identification of diverse species-specific transporters and metabolic enzymes implied niche specialization for nutrient acquisition among these bacterial groups. Temperature was the main environmental factor driving the active bacterial groups and metabolic processes, and Enterobacterales replaced Alteromonadales as the predominant group under low temperature. This study provides insight into the key bacteria and metabolic processes involved in POC remineralization, and niche complementarity and species substitution among bacterial groups are critical for efficient POC remineralization in the twilight zone.

IMPORTANCE The ocean’s twilight zone is a critical zone where more than 70% of the sinking particulate organic carbon (POC) is remineralized. Therefore, the twilight zone determines the size of biological carbon storage in the ocean and regulates the global climate. Prokaryotes are major players that govern remineralization of POC in this region. However, knowledge of microbial community structure and metabolic activity is still lacking. This study unveiled microbial communities and metabolic activities of POC samples collected from the twilight zone of three contrasting environments in the Northwest Pacific Ocean using a metaproteomic approach. Alteromonadales, Rhodobacterales, and Enterobacterales were the major remineralizers of POC. They excreted diverse species-specific hydrolytic enzymes to split POC into solubilized POC or dissolved organic carbon. Temperature played a crucial role in regulating the community composition and metabolism. Furthermore, niche complementarity or species substitution among bacterial groups guaranteed the efficient remineralization of POC in the twilight zone.

INTRODUCTION

Particulate organic carbon (POC), the core component of the biological pump, mediates carbon cycling and sequestration in the ocean (1). Most POC is ultimately remineralized to CO₂ while sinking through the water column (2), and more than 70% of this sinking POC in the oligotrophic ocean is remineralized in the twilight zone (3, 4). Consequently, the twilight zone determines the magnitude of the biological carbon storage in the ocean, thereby helping to regulate the global climate (5).

In the last 10 to 20 years, concerns over the biological remineralization of POC by resident biota within the twilight zone have been raised (6–8). Prokaryotes are the major players that govern the remineralization of POC in the twilight zone, and are thought to be responsible for 70 to 92% of the estimated remineralization (3, 9). The rate of sinking POC converted into CO₂ by heterotrophic organisms controls the oceanic carbon storage (5). Rapid biological consumption and remineralization of carbon reduces the sequestration efficiency in the twilight zone (2). However, information regarding the prokaryotic communities involved in the remineralization of POC in the twilight zone, such as their taxonomic composition, community structure, and metabolic activity, is still limited. Metagenomic and metatranscriptomic studies have revealed the microbial diversity and genetic potential in the mesopelagic zone, and have indicated that microbial groups with specific metabolic capabilities might be involved in POC remineralization (10–16). However, evidence directly linking these microbial groups to POC remineralization in the twilight zone is scarce.

Studies have reported higher hydrolytic activity in marine-snow-associated bacteria compared with free-living bacteria (17, 18), indicating the dominant role of particle-attached bacteria in the remineralization of POC. A recent study also demonstrates that particle-attached bacteria possess specific extracellular hydrolytic enzymes that allow the release of carbon trapped in POC, and expression levels and the properties of these enzymes control the rate of microbial breakdown of POC (19). These studies suggest a direct correlation between the presence of extracellular hydrolytic enzymes and metabolic activity. However, studies on microbial enzymes associated with POC remineralization are limited, mainly owing to a lack of suitable methodology for in-depth evaluation of microbial metabolic functions. Metaproteomics can directly assess the active microbial groups and their in situ metabolic activities compared with metatranscriptomics, which provides information on expressed functional capacity of communities, elucidating various processes associated with materials and energy metabolism (20–23). Nevertheless, obtaining sufficient POC for metaproteomic analysis remains challenging owing to its extremely low concentration in the twilight zone (6).

In the present study, we focused on bacteria tightly associated with POC, considering that they act as the primary consumer of POC compared with free-living bacteria (19). We applied a large-volume in situ water transfer system to enrich POC (size range: 0.7 to 200 μm) in the upper twilight zone of three contrasting environments (sites K2, B1, and B9) in the Northwest Pacific Ocean (Fig. S1 and Table S1 in the supplemental material); the protein profiles of POC were characterized using a metaproteomic approach (Fig. S2). The K2 site is influenced by the Kuroshio Current, and is characterized by warm water and low-level nutrients. Site B1 is situated at the edge of the Oyashio Current, which features cold water and high concentrations of nutrients. The B9 site is in the convergence region of the Kuroshio and Oyashio currents. However, variations in POC concentration along the water column were similar among these three sites, and the minimum value was observed at a depth of 200 m (Table S1). Our results unveiled the key microbial players and metabolic pathways directly associated with POC remineralization. Alteromonadales, Rhodobacterales, and Enterobacterales were the main remineralizers, while niche complementarity and species substitution among these bacterial groups drove the efficient remineralization of POC in the upper twilight zone.

RESULTS AND DISCUSSION

Key microbial players involved in POC remineralization.

The POC samples collected from the deep chlorophyll maximum (DCM), 100, 200, and 500 m layers of sites K2, B1, and B9, were subjected to metaproteomic analysis and each sample had two biological replicates and two technical replicates (Fig. S2). A three-step search strategy (24) (Fig. S3) against the global Ocean Microbial Reference Gene Catalog (OM-RGC) yielded high qualitative and quantitative protein coverage (Fig. S4A), and almost all the samples exhibited good biological and technical reproducibility (Fig. S4B and C). An average of 7,177 proteins per sample were identified with an outcome of 24,967 nonredundant proteins with quantitative information from all samples. This protein recovery from natural marine environments greatly exceeded that found in previous studies (20, 22, 25), and 97.28% of the proteins identified were assigned to putative taxonomic origins. Cluster analysis of peptides and proteins indicated that samples from the euphotic and twilight zones were separated from each other (Fig. S4B and C). Furthermore, a significant difference in microbial community structure and function was observed between these two zones (Fig. S4D and E). In the present study, focus was placed on the microbial community and metabolic activity in the upper twilight zone (200-m and 500-m layers), the main region of POC remineralization.

In the upper twilight zone, 66.77% of proteins were assigned to bacteria, 28.95% to eukaryotes, 1.33% to archaea, and 0.23% to viruses (Fig. S5). The low contribution of viral proteins might be caused by low viral biomass in the POC and/or by high passing rate through the GF/F membrane. Because of the high proportion of prokaryotes and the essential role of these microorganisms in POC remineralization (3), this study was primarily centered on the prokaryotic community. Although prokaryotic community structure differed significantly between layers and sites (Fig. 1A), the abundances of most KEGG pathways were similar (Fig. S6), indicative of relatively stable functions and processes in the twilight zone, even in contrasting environments. The relative abundance of KEGG pathways had a lower variability than that of the microbial population, as indicated by their coefficient of variation (0.28 versus 0.60). Proteobacteria (73.47%, the average value of all twilight samples) were the predominant contributors, while Alphaproteobacteria and Gammaproteobacteria were the major contributors at the class level, but with opposing depth distributions. At the order level, Alteromonadales was the dominant group in the K2-500 m (27.61%) and B9-200 m (27.81%) layers, while Alteromonas (48.85%) and Pseudoalteromonas (32.4%) were the dominant genera (Fig. S7A). The abundance of Rhodobacterales exhibited trends similar to those of Alteromonadales, and this is consistent with previous studies (10, 11, 26). However, Rhodobacterales presented much more diversely than Alteromonadales, consisting of 94 genera, and the composition differed significantly among the three sites (Fig. S7B). Roseobacter and Sulfitobacter were the dominant genera, accounting for 11% and 24% of the detected bacteria, respectively. The highly diverse ensemble of species of Rhodobacterales provided strong abilities to respond effectively to environmental changes (27). Interestingly, Enterobacterales, an order of the Gammaproteobacteria, replaced Alteromonadales as the dominant group in cold-water environments, comprising 67.59% of the identified bacteria in the B1-200 m layer, 29.65% in the B1-500 m layer, and 20.04% in the B9-500 m layer. However, the distribution was similar at the genus level among the three sites, even though the proportion of members of the Enterobacterales was different (Fig. S7C), and Escherichia was the dominant genus (24.72%). Members of marine Enterobacteria are detected not only in the fecal pellets of marine mammals and tissues of other animals, but also are isolated from surface or deep waters (28, 29). Therefore, it is not surprising to detect enterobacterial proteins in POC due to diverse sources of POC in the twilight zone. Overall, Alteromonadales, Rhodobacterales, and Enterobacterales accounted for nearly 50% of the prokaryotic community, implying that they play important roles in the twilight zone.

FIG 1.

(a) Taxonomic composition of prokaryote communities at depths of 200 m and 500 m from sites K2, B1, and B9 based on the metaproteomic analysis. Prokaryotic taxa with >1% relative abundance on average are displayed and named in the format, phylum_class_order. (b) The interaction network within the prokaryotic community based on Spearman’s correlation (P < 0.05). The red and green lines represent the positive and negative relationships, respectively. The circle size represents the abundance of the prokaryotic community. (c) Interactive chord diagram visualizing the relationship between prokaryotic communities and functional ontology (COG functions) based on metaproteomes collected at depths of 200 m and 500 m from sites of K2, B1, and B9. The number represents prokaryotic communities and the alphabet represents functional ontology (COG functions). The three dominant bacterial groups of Rhodobacterales, Alteromonadales, and Enterobacterales highlighted in red font are indicated by red arrows in panels a and c.

The network analysis revealed that Alteromonadales, Rhodobacterales, and Enterobacterales were located at the central nodes and were directly connected with numerous other bacterial groups (Fig. 1B, Table S2), suggesting their important roles in the microbial community. Furthermore, Alteromonadales and Rhodobacterales exhibited a similar interaction pattern with other bacteria. Most bacteria had a positive relationship with Alteromonadales and Rhodobacterales, indicating that these bacteria, such as Maribacter, might be cross-feeders. A recent study demonstrates that the Alteromonadales and Rhodobacterales are the primary degraders and able to cross-feed when grown with secondary consumers (19). In contrast, Enterobacterales had a negative relationship with most bacterial groups. In addition, the correlation analysis between the microbial community and COG functions revealed that Alteromonadales, Rhodobacterales, and Enterobacterales performed the major functions (47.75%) among the microbial communities (Fig. 1C), especially those related to the transport and metabolism of amino acids and carbon. These results suggest that these three groups were the major players in the POC-associated microbial community of the twilight zone. Studies on both particle-attached and free-living bacteria in different regions (30, 31), as well as those on particulate-derived experimental isolates (19), have demonstrated that many members of these three bacterial groups are particle attached. Furthermore, Alteromonadales and Rhodobacterales are the main degraders of transparent exopolymer particles (26) and chitins (19), both of which are major constituents of POC. A recent study shows that Alteromonadales play a key role in the mineralization and biogeochemical transformation of sinking particulate organic matter in the deep ocean (32). Combined, the present metaproteomic results indicated that particle-attached members from Alteromonadales, Rhodobacterales, and Enterobacterales are the primary POC remineralizers in the twilight zone.

Essential metabolic processes involved in the remineralization of POC.

The composition of POC is complex and generally includes living and dead cells, excretory products, fecal pellets, and marine snow (5). Extracellular microbial enzymes are thought to be responsible for the degradation and transformation of this organic material as POC sinks through the water column (33). In the present study, proteases (6.14%) and hydrolases (2.46%) from a variety of bacterial groups were relatively highly represented in the KEGG orthology analysis of protein function (Fig. S8), indicative of their important roles in POC degradation. Furthermore, most proteases and hydrolases (e.g., protease IV, protease S8, pectinlyase) were predicted as extracellular and outer membrane proteins, and were highly diverse, based on the predicted substrates including proteins and polysaccharides (Table S3). Although we did not conduct the enzymatic activity assay, the high activity of extracellular hydrolases and proteases has been reported in attached and isolated bacteria collected from marine snow (17). These results demonstrate that proteases and hydrolases play essential roles in degrading proteinaceous components and polysaccharides, which are the primary constituents of POC.

The taxonomic distribution of protease and hydrolase was mainly attributed to Alteromonadales, Rhodobacterales, and Enterobacterales (57.73% of proteases and 37.22% of hydrolases; Fig. S9). However, the diversity and abundance of these enzymes differed significantly among the three groups. Alteromonadales presented the greatest protease and hydrolase diversity and abundance (Fig. 2), indicative of a strong ability to split POC into a wide range of substrates or materials, including glucose and complex polymeric compounds such as alginate and agarose, which likely provide carbon sources for Alteromonadales. Therefore, the results of the present study demonstrated that Alteromonadales were the “hot spots” favoring motile copiotrophic bacteria, which is consistent with a previous study (34). A high diversity and abundance of proteases and hydrolases were also detected in Rhodobacterales, but the predicted substrates differed from those of Alteromonadales, indicating that these two bacterial groups have different nutrient requirements (Fig. 2). Notably, only one hydrolase was detected in Enterobacterales, even though diverse proteases were identified, indicating that proteases are the main POC degraders in Enterobacterales. The difference in diversity and abundance of proteases and hydrolases among the three bacterial groups might be caused by the different compositions of sinking POC. The present results indicated that the sinking POC from the three sites was mainly composed of phytoplankton; however, the community structure differed (Fig. S10). Cyanobacteria were the dominant phytoplanktons in K2, while Dinophyceae were the dominant phytoplanktons in B1. The phytoplankton community structure in B9 was more diverse, presenting a high abundance of Cyanobacteria, Haptophyceae, and Bacillariophyta. Correspondingly, a high abundance of α-L-fucosidase was detected in Alteromonadales in K2 and B9 because Cyanobacteria can produce a relatively high percentage of fucose (35). These results indicated that Alteromonadales, Rhodobacterales, and Enterobacterales secrete specific proteases and hydrolases to degrade POC of different biological origins.

FIG 2.

Vertical distribution and relative abundance of transporters, proteases, and hydrolases in Alteromonadales, Rhodobacterales, and Enterobacterales. The transporter, protease, and hydrolase proteins were grouped according to the predicted substrate specificity of the substrate-binding proteins. Samples are named in the format of site_depth;, for example, K2_200 indicates the sampling location at a depth of 200 m at the K2 site. AA, amino acid; ABC, ATP-binding cassette; AHL, N-acyl homoserine lactone; BCAAs, branched chain amino acids; DLH, dienelactone hydrolase; HAGH, hydroxyacylglutathione hydrolase; FAA, fatty acid amide; GTP, guanosine triohosphte; NagA, N-acetylglucosamine-6-phosphate deacetylase; NCA, N-carbamoyl-L-amino-acid; NCP, N-carbamoylputrescine; PRA-CH, phosphoribosyl-AMP cyclohdryrolase; SADH, succinylarginine dihydrolase; TBDTs, TonB-dependent transporters; TRAP, tripartite ATP-independent periplasmic; TTT, tripartite tricarboxylate.

Interestingly, diverse and abundant transporters involved in organic carbon transport were detected in the three bacterial groups, including the ATP-binding cassette (ABC) transporter, TonB-dependent receptor (TBDT), and tripartite ATP-independent periplasmic (TRAP) and tripartite tricarboxylate (TTT) transporters, but the expression patterns of these proteins were species specific (Fig. 2). Notably, fewer transporters were identified in Enterobacterales compared with the other two bacterial groups. The abundance of ABC and TRAP transporters was relatively higher in Rhodobacterales than in Alteromonadales, whereas that of TBDT transporters was lower. Rhodobacterales exhibited the highest expression of ABC transporters, which may be attributable to their involvement in the uptake of amino acids and carbohydrates, the primary metabolic products of POC (36). Alteromonadales expressed TBDT transporters for biopolymer uptake, suggesting that Alteromonadales actively absorbs high-molecular-weight organic matter (37). These results indicated a niche differentiation on POC between Rhodobacterales and Alteromonadales, which may be responsible for maintaining the high POC remineralization efficiency in the twilight zone.

The processes of carbohydrate and amino acid metabolisms were present in relatively high abundance among all pathways in Alteromonadales (Fig. S11). Proteins involved in the pathways of glycolysis, propanoate metabolism, pyruvate metabolism, purine metabolism, and pyrimidine metabolism were abundant. Similar to Alteromonadales, Rhodobacterales were also involved in the fundamental metabolic processes in the twilight zone, such as aerobic respiration, anaerobic fermentation, sulfur oxidation, autotrophic carbon fixation, nitrogen fixation, and hydrogen production. However, the pathways of carbon and nitrogen metabolism in Rhodobacterales differed from those in Alteromonadales. The pathways of C5-branched dibasic acid metabolism and butanoate metabolism were relatively enriched in Rhodobacterales, indicating that Rhodobacterales and Alteromonadales utilized different carbon and nitrogen sources. There is some evidence that Rhodobacterales can directly utilize sulfate-containing compounds (38). The metabolic processes were similar between Enterobacterales and Alteromonadales in terms of types and expression patterns, especially the carbon metabolism pathways. However, some differences existed between the two groups. The glycolytic pathway was the most enriched in the carbon metabolism category in Enterobacterales, implying that sugar might be its main source of carbon. The abundance of proteins involved in the nitrogen metabolic pathway was lower in Enterobacterales than in Alteromonadales, reflecting the poor ability of Enterobacterales to use organic nitrogen when compared with Alteromonadales. Nonetheless, the concentration of particulate organic nitrogen in the B1-200 m layer was relatively high, while the concentration of POC remained at a normal level (Table S1). This result further supports that nitrogen metabolism in Enterobacterales was lower than that in Alteromonadales. In addition, proteins involved in glycolysis, oxidative phosphorylation, and two essential cellular respiration processes were highly expressed; this was consistent with the abundance of transporters, proteases, and hydrolases, and was indicative of the active breakdown, transport, and respiration of organic carbon in the three bacterial groups.

Environmental influence on POC remineralization in the twilight zone.

Environmental factors are known to influence marine microbial communities and their metabolic activity (39). The distance-corrected dissimilarities in taxonomic and functional community composition were correlated with environmental factors at the three sites. Overall, temperature had the strongest correlation with both taxonomic and functional composition in the twilight zone (Fig. 3), while salinity influenced the taxonomic composition. Furthermore, the correlations between each bacterial group and specific environmental factors indicated that temperature had a significant effect on some bacterial groups, such as Enterobacterales and Rhodobacterales (Fig. S12A). In addition, several COG functions were also correlated with temperature (Fig. S12B). Notably, Enterobacterales was one of the few groups that presented a negative correlation with temperature. The abundance of cold shock proteins in all samples also showed a negative correlation with temperature (Fig. 4A), and the majority of them were detected from Enterobacterales (Fig. 4B), particularly at the B1 site located at the edge of the Oyashio current characterized by cold waters. A previous study has demonstrated that there exist environmentally adapted species/strains of Enterobacterales and some of them can outcompete their typical enteric counterparts at low temperatures (40). These results correspond with the patterns in the microbial community, where the dominant bacterial group shifted from Alteromonadales in high-temperature layers (K2-200 m, K2-500 m, and B9-200 m) to Enterobacterales in low-temperature layers (B1-200 m, B1-500 m, and B9-500 m) (Fig. 1A). Nevertheless, the flux of PN in B1-200 m was relatively high, while the flux of POC maintained a normal situation (Fig. S13). This result also supported that the nitrogen metabolism in Enterobacterales was lower than that in Alteromonadales (Fig. S11), indicating that the Enterobacterales had a poor ability to use organic nitrogen. Taken together, our results indicated that temperature was the main environmental factor modulating the composition and metabolic activity of the microbial groups involved in the remineralization of POC in the twilight zone.

FIG 3.

Pairwise comparisons of environmental factors are shown with a color gradient denoting the Spearman’s correlation coefficients. The taxonomic (based on two independent methods: metaproteomics and 16S rRNA) and functional (based on COGs) community composition was correlated to each environmental factor using the partial (geographic distance-corrected) Mantel test. The edge width corresponds to the Mantel’s r statistic for the corresponding distance correlations, and the edge color denotes the statistical significance based on 9,999 permutations.

FIG 4.

(a) The vertical distribution and relative abundance of heat and cold shock proteins present in metaproteomes collected at depths of 200 m and 500 m from the sites K2, B1, and B9. The black dots represent seawater temperatures at different sampling layers. (b) Taxonomic distribution of cold shock proteins. The relative abundances of these proteins from each taxon were averaged for all samples in the pie chart.

It is worth noting that we collected two biological replicates from each layer due to the challenge of sampling POC from the twilight zone and the poor sea conditions during the survey period. Consequently, in order to make up this deficiency, we applied the Mantel test to assess the environmental influence on microbial community based on all data from 12 samples, considering the principle of the Mantel test, and the results were acceptable and in line with previous studies. Of course, sufficient biological replicates should be guaranteed if sea conditions permit to minimize uncertainty and enhance the statistical power in a future study. Meanwhile, other alternatives, including more sampling sites across a large geographical scale to increase the sample number or the combination of metaproteomics and metagenomics to authenticate each other, should also be considered.

In summary, the results of this study highlight the primary microbial organisms and essential metabolic processes involved in the remineralization of POC in the twilight zone. Particle-attached members from the Alteromonadales, Rhodobacterales, and Enterobacterales were the key players governing the remineralization of POC in the twilight zone. These bacteria secrete specific hydrolytic enzymes to either split POC into solubilized POC or degrade POC to dissolve organic carbon, thereby providing diverse substrates for various metabolic activities (Fig. 5). Interestingly, Alteromonadales and Rhodobacterales employed complementary mechanisms for substrate utilization, leading to efficient POC degradation. Although data for microbial community function in water columns of other oceans is limited, global metagenomic studies have shown that Alteromonadales and Rhodobacterales are the dominant bacterial groups in the mesopelagic ocean (10, 11). This suggests that Alteromonadales and Rhodobacterales are the main drivers of POC remineralization in the ocean’s twilight zone. This hypothesis needs further examination in different oceanic regions. Temperature was the primary environmental factor influencing POC remineralization, through modulation of the composition and metabolic activity of the active microbial groups. In cold environments, Enterobacterales replaced Alteromonadales and dominated the remineralization of POC (Fig. 5). The results indicate that climate change could lead to alterations in microbial community structure and metabolic activity, which may influence POC remineralization in the twilight zone, and subsequently affect ocean carbon cycling and global climate. Investigation of microbial metabolism throughout the twilight zone in different oceans, combined with enzymatic activity measurements and metabonomic analysis of some key bacterial groups isolated from the twilight zone, are necessary for a comprehensive understanding of POC remineralization in the world’s oceans.

FIG 5.

Schematic depiction of the key microbial players and essential enzymes involved in the remineralization of POC in the twilight zone. The relative abundances of the main processes/pathways are presented in yellow for each order type, which were calculated by comparing within their own group’s ratio. The relative abundance of different groups of membrane transporters is also presented, summing 100% for each group type. POC, particulate organic carbon; CM, carbohydrate metabolism; AAM, amino acids metabolism; NCM, nitrogen compound metabolism; MM, methane/methanol metabolism; OT, other transporter; TFO, transporter for organic substrate; TFI, transporter for inorganic substrate.

MATERIALS AND METHODS

Cruise information and sampling strategy.

A multidisciplinary cruise was conducted in the Northwest Pacific Ocean from 30 March to 10 May 2015, aboard the RV DongFangHong 2. Samples of POC for metaproteomic analysis were collected from three sites: K2 (134°00′ E, 25°00′ N), B1 (147°00′ E, 38°00′ N), and B9 (147°00′ E, 30°00′ N) (Fig. S1). Due to very low concentrations of protein in POC from the twilight zone (48), large volumes of seawater (∼500 to 1,000 liters) are required for concentrating sufficient protein for metaproteomic analysis, which is time consuming, laborious, and expensive. Considering the limitation of sampling time, poor sea conditions during the survey period, as well as the minimum requirements for metaproteomic study, we collected two biological replicates from each layer using the large volume water transfer system (McLane Research Laboratories, Inc., USA). During the filtration, water samples were prefiltered through a 200-μm mesh filter and retained on a GF/F membrane (142 mm in diameter, Millipore Corporation, MA, USA). The GF/F membranes containing POC were stored at −80°C until use. The temperature, salinity, depth, chlorophyll fluorescence, and dissolved oxygen values at each site were retrieved from a conductivity-temperature-depth (CTD) probe and rosette sampling system (Sea Bird Electronics, USA). Water samples for analyzing inorganic nutrients (NO₃⁻, NO₂⁻, SiO₃²⁻, and PO₄³⁻) were filtered through a 0.22-μm pore size polycarbonate membrane and immediately analyzed on board using an AA3 automatic continuous flow analyzer (Seal AA3, Norderstedt, Germany). The analytical precision of NO₂⁻, NO₃⁻, PO₄³⁻, and SiO₃⁻ measurements was 0.1, 0.1, 0.2, and 0.05 μM, respectively. POC and particulate organic nitrogen were analyzed using a Thermo Finnigan Flash EA1112 elemental analyzer (Thermo Electron Corp., Waltham, MA, USA), with acetanilide as the calibration standard.

Protein extraction, separation, and liquid chromatography-electrospray ionization-tandem mass spectrometry (LC-ESI-MS/MS) analysis.

In order to obtain a better recovery of proteins from the POC, we optimized the protein extraction method before metaproteomics analysis and the combination of particle breakdown with a homogenizer and cell lysis with TRIzol reagent proved the best way to efficiently liberate proteins from the POC. Briefly, the filter membranes were thawed and cut into chips (2 × 2 mm) using a sterilized razorblade. The membrane chips (per gram) were suspended in 20 ml of TRIzol reagent (Ambion-Life Technologies, USA), and disrupted in a FastPrep-24 homogenizer (MP Biomedicals, Santa Ana, CA, USA) three times each for 20 s. Subsequently, the mixtures were centrifuged at 20,000 × g for 30 min after incubation at 25°C for 1 h. The supernatants were then used for protein extraction following the TRIzol reagent manufacturer’s protocol. A rehydration buffer containing 7 M urea, 2 M thiourea, and 4% (wt/vol) CHAPS (3-[(3-cholamidopropyl)-dimethylammonio]-1-propanesulfonate) was added to dissolve the protein pellets. The samples were then centrifuged at 14,000 × g to remove cell debris and the protein concentrations in the supernatants were determined using a 2D Quant Kit (GE Healthcare, USA). A buffer exchange was performed with 100 mM NH₄HCO₃ using a 10-kDa Amicon filter (Millipore, MA, USA). The purified protein samples (100 g) were subjected to overnight trypsin digestion and separated on C₁₈ (Sigma-Aldrich, USA) and strong cation exchange solid-phase extraction columns. The peptides were analyzed on an UltiMate 3000 UHPLC system (Thermo Scientific, Waltham, MA, USA) equipped with a 300 μm i.d. × 0.5 cm C₁₈ TRAP column (μ-Precolumn, Thermo Scientific, Waltham, MA, USA) and a 75 μm i.d. × 25 cm analytical column packed in-house with 3 μm C₁₈ particles, and coupled to a Q Exactive HF hybrid quadrupole-Orbitrap mass spectrometer (Thermo Fisher. Scientific, Waltham, MA, USA). The MS parameters were set as follows: spray voltage 2 kV, capillary temperature 320°C, positive mode, scan range of 350 to 1,500 m/z, loop count 30, NCE 26, MS resolution 120,000, MS/MS HCD scan resolution 30,000, and dynamic exclusion duration 30 s. Each fraction was injected twice for confident identification.

Protein identification and bioinformatics analysis.

The MS raw data of each sample were merged and formatted to MGF files using Proteome Discoverer (ver. 1.3.0.339; Thermo Fisher Scientific, San Jose, CA). The MS/MS spectra were searched against the global Ocean Microbial Reference Gene Catalog (OM-RGC), the current largest environmental database of the global ocean, including samples from the mesopelagic regions as well as the investigated region in this study (39). A three-step iterative search was applied for protein identification and quantification using the MetaPro-IQ approach, which has proved to be the most suitable workflow to produce highly efficient protein identification when matched metagenome sequencing is not available and the currently close-to-complete and huge OM-RGC database is used instead (41) (Fig. S3). Briefly, the first- and second-step database searching against the OM-RGC database were first performed to generate a reduced database that contained all possible proteins derived from peptide-spectrum matches (PSMs) for all samples using X! Tandem software (2017.2.1 version). The reduced database containing the resulting protein lists was then imported into MaxQuant software (1.6.1.0 version) for protein quantification (24, 42). In the first- and second-step database searches, the tandem search was performed with up to one miss-cleavages (trypsin/P), carbamidomethylation of cysteine as a fixed modification, and oxidation of methionine as a potential modification. A fragment ion tolerance of 20 ppm and a parent ion tolerance of 0.05 Da were applied. All matched protein sequences for the first-step search were extracted as the sample-specific database (S_all-macth.fasta). The X! Tandem outputs of the target-decoy database search (step 2) were summarized with an in-house software to generate an identified protein list at a PSM false discovery rate (FDR) cutoff of 0.01. The resulting protein list for all samples was then combined, and duplicates were removed to generate a “combined nonredundant database” for protein quantification with MaxQuant. Except that a commonly used precursor mass tolerance was set to 20 ppm for the first search and 7 ppm for the main search, similar peptide identification parameters with X! Tandem database searches were used for MaxQuant analysis (parameter files were uploaded together with the raw data and result files to ProteomeXchange). To reduce the probability of false peptide identification, the FDR was set to less than 1% at both PSM and protein levels. In addition, the label-free quantification (LFQ) algorithm was used for quantification. Both razor and unique peptides were selected for protein quantification, and the minimum ratio count was set as 1. An alignment retention time window of 20 min and match time window of 5 min were applied to match the same accurate masses between different runs. High-confidence proteins matching at least two PSM and one unique peptide were selected for further analysis.

Proteins identified by the same set or a subset of peptides were clustered together as one protein group. Leading proteins (defined as the top-ranked protein in a group, where ranking is based on the number of peptide sequences, the number of PSMs, and the sequence coverage) of the protein group were selected for further taxonomic and functional analysis (41).

Taxonomic and functional annotation of proteins.

For taxonomic annotation, the identified sequences from the OM-RGC database were species specific and other sequences were annotated against the NCBI nonredundant protein database (NCBInr 2017-9) using BLASTP. The BLASTP search results were loaded into MEGAN (43) and the taxonomic assignment was performed using the lowest common ancestor algorithm (bit score > 80) (44). For functional analysis, the identified protein sequences were aligned against (i) the Clusters of Orthologous Groups (COGs) database (ftp://ftp.ncbi.nih.gov/pub/COG/COG2014/data) (45), (ii) the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (46), and (iii) the Gene Ontology (GO) database with BLASTP, using default parameters (E value cutoff 0.00001). Because of each layer having two biological and technical replicate samples, the proteins identified in two or more replicate samples were selected for further analysis and the relative abundances of proteins were averaged among the replicate samples. For the analysis of proteome-based community structure, the relative abundances of each microbial group were calculated by summing their total protein abundances and then dividing by the sum of all protein abundances in a sample (community-level analysis). For function comparison, the relative abundance of each protein and metabolic KEGG pathway was calculated by summing the related protein abundances and then dividing by the sum of all protein abundances in a species (species-level analysis) (47).

Correlation of distance matrices between metaproteomes and environmental factors.

Pairwise distances between twilight zone samples were computed on the basis of the following: (i) relative abundance of taxonomic groups (16S rRNA operational taxonomic units [OTUs] and metaproteomic abundance) and protein function composition (at the COG level), and the compositional data; and (ii) in situ measurements of the physicochemical data and the environmental data. Environmental data were transformed into z-scores before calculating the distances. Euclidean distances were used for compositional and environmental data, while Haversine distances were used for physicochemical data. Based on these distance matrices, Partial Mantel correlations between compositional and environmental data given geographic distance (9,999 permutations) were computed using the vegan R software package. Partial Mantel tests were also performed between species richness and temperature and latitude (39).

Data availability.

The mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium via the PRIDE partner repository, www.ebi.ac.uk/pride/archive/ (identifier PXD014630).