The ‘DeepDOM’ cruise (KN210-04) took place on the R/V Knorr from March 25 to May 9, 2013, beginning in Montevideo, Uruguay and ending in Bridgetown, Barbados. In total, twenty-three stations were sampled, with water column samples retrieved from the surface (5 m) to bottom waters (4000 m) along an 8482 km cruise track in the South Atlantic Ocean ( Figure 1 ). We particularly focus on a 2765 km stretch of this cruise track, represented by Stations 7, 15 and 23, because these were the stations where taxonomic, metagenomic and metatranscriptomic samples were concurrently collected. Samples span both the South Atlantic Subtropical Gyre (Station 7) and the equatorial region, represented by Station 15 ( Figure 1 ). All samples were collected in the open ocean, a minimum of 300 km off the coast of South America, with Station 23 exhibiting influence from the Amazon River plume. Down the water column, samples were collected at the surface (5 m), deep chlorophyll maximum (50-130 m), transition zone (250 m), oxygen minimum zone (~500 m), Antarctic Intermediate Water (AAIW, typically sampled at ~750-850 m), and North Atlantic Deep Water (NADW, typically sampled at ~2500 m) masses. The latitude and longitude for all sites and depths sampled, and the availability of taxonomic, metagenomic and metatranscriptomic data is summarized in Supplementary Table 1 .

In situ water column measurements of conductivity, temperature and depth were collected with a CTD rosette using a SPE 9+ CTD system with a depth limit of 6000 m, and equipped with sensors for oxygen, fluorescence, turbidity, and photosynthetically available radiation (PAR). The in situ water column profiles plotted in this study were from the casts used to generate taxonomic profiles. Samples for major nutrients, including nitrate ( N O 3 − ) , nitrite ( N O 2 − ) ammonium ( N H 4 + ) , phosphate ( P O 4 3 − ) , silicate ( S i O 4 4 − ) , total nitrogen (TN), and total organic carbon (TOC) were taken directly from 10L Niskin bottles. Samples for nutrient analysis were generally taken at the same depths where taxonomic data was obtained ( Supplementary Table 1 ) using polycarbonate bottles, and frozen at -20°C until analysis could be completed at Oregon State University. N O 3 − + N O 2 − ,   N H 4 + ,   P O 4 3 − , and S i O 4 4 − were all analyzed on a hybrid Technicon AutoAnalyzer and Alpkem RFA300 system, following protocols modified from Gordon et al. (1994). The estimated precision for each element was as follows: P O 4 3 − , ± 0.008 µM; N O 3 − +   N O 2 −   ±   0.15   µ M ; N O 2 − , ± 0.01 µM; S i O 4 4 − ±   0.3   µ M . To measure the concentration of TN and TOC, 40mL seawater was preserved via acidification to pH 3 with concentrated trace-metal grade hydrochloric acid (HCl). These were then stored in combusted glass vials at 4°C until analysis on a Shimadzu TOC-VCSH total organic carbon analyzer coupled to a TNM-1 analyzer at the Woods Hole Oceanographic Institution. Blanks consisting of MilliQ water, and standard curves made with potassium hydrogen phthalate and potassium nitrate were interspersed into sample runs. The coefficient of variability between replicate injections was < 1%.

Collection and processing of dissolved (< 0.2 um) and particulate (> 0.2 um) metabolites was done as per the methods outlined in Johnson et al. (in revision). Briefly, water (4 L) was directly collected from Niskin bottles into polytetrafluoroethylene (PTFE) or polycarbonate (PC) bottles, and then filtered through a previously combusted 0.7 µm GF/F filter (Whatman) and 0.2 µm filter (Omnipore, EMD Millipore) using a peristaltic pump. The resulting filtrate was then acidified with 4 mL of 12 M HCl (~pH 2-3). Dissolved organic molecules (< 0.2 um) were extracted from the filtrate using solid phase extraction (SPE) modified styrene-divinylbenzene polymer (Agilent Bond Elut PPL) (Dittmar et al., 2008; Longnecker, 2015). Extracts were stored at -20°C until analysis. Process blanks were created using shipboard and laboratory Milli-Q water. GF/F Filter (0.7 um) and Omnipore filter (0.2 um) membranes, collectively representing the particulate metabolite fraction, were extracted within 48 h prior to mass spectrometry analysis and processing occurred as per Johnson et al. (in revision). In brief, samples were run on a Phenomenex C18 column for LC-MS/MS analysis (Synergi Fusion, 2.1 x 150 mm, 4 µm) coupled via heated electrospray ionization (ESI) to a triple quadrupole mass spectrometer (Thermo Scientific TSQ Vantage) operated under selected reaction monitoring mode (SRM) (Kido Soule et al., 2015). Quantification and confirmation SRM transitions were monitored for each analyte. Before each batch of samples was run, the column was conditioned with 5 injections of the pooled samples. A pooled QC sample was then run after every ten samples. XCalibur RAW files generated by the mass spectrometer were converted to mzML files using msConvert (Chambers et al., 2012). MAVEN (Melamud et al., 2010; Clasquin et al., 2012) was used to select and integrate peaks and quality assessed as per Johnson et al. (in revision). Both particulate and dissolved metabolite concentrations (in pM) were normalized to sample volume.

At each sampling depth, 2 L of seawater was collected directly from Niskin bottles into polycarbonate bottles for 16S rRNA gene sequencing. These samples were then filtered through a 0.2 µm PVDF Sterivex filter (EMD Millipore) without a pre-filter. Larger volume samples (20 L and 8 L) were collected for metagenomic and metatranscriptomic analyses, respectively, into polycarbonate carboys. These larger volume water samples were filtered through a 2.7 µm GF/D pre-filter followed by filtration through a 0.2 µm PVDF Sterivex filter (EMD Millipore). Sample filtration was done using a peristaltic pump (Masterflex), equipped with Tygon tubing, set to < 50 mL/min to avoid rupturing cells during filtration. Duplicate Sterivex filters were collected per depth for metagenomic and metatranscriptomic analyses. Filtration of samples for metatranscriptomic analyses was done as soon as possible after collection at room temperature benchtop aboard the R/V Knorr, taking approximately 20-25 minutes per sample. Residual seawater was removed by extrusion and 1.8 mL of sucrose lysis buffer or RNAlater were added to samples destined for genomic DNA and total RNA extraction, respectively. All filters were then frozen at -80°C until extraction. All plasticware and tubing was acid washed (using 2% HCl) and MilliQ water rinsed prior to use and in between different samples. In addition to the seawater, three additional samples from a sediment net trap were collected as described in Durkin et al. (2016) from Stations 7 and 23. Briefly, these sediment traps consisted of a large, vertically hanging 2 m diameter conical net with a cod end which were deployed at ~150 m depth, below the surface mixed layer. This net trap served to capture fresh sinking particulate material, and was deployed for 23 h at Station 7 and ~29 h at Station 23. From the trap, a total of 160 mL of seawater was filtered at Station 7, and between 210-220 mL at Station 23.

Genomic DNA was extracted from Sterivex filters using a chemical lysis method, as described in Zaikova et al. (2010). Briefly, filters were first thawed on ice and cells were lysed with lysozyme and then incubated with rotation at 37°C. Afterward, proteinase K, 10% SDS and RnaseA were added to the filters, with rotating incubation at 55°C. The lysate was then extracted with phenol: chloroform:isoamyl alcohol (IAA) (25:24:1), followed by chloroform: IAA (24:1). The subsequent aqueous layer was then concentrated with an Amicon Ultra (10 Kda, Millipore Sigma, MA, USA) filter cartridge in a final volume of 200 µL of TE buffer. The quality and size of the extracted genomic DNA was verified by gel electrophoresis on a 0.8% agarose run overnight at 16 V. DNA concentration was quantified using PicoGreen (Thermofisher) following the vendor’s protocol. As an additional step to validate the ability to build amplicon libraries, PCR was carried out using primers 515F-Y and 926R encompassing the bacterial and archaeal V4-V5 hypervariable region (Parada et al., 2015).

Total RNA was extracted from Sterivex filters, in duplicate, beginning with the extrusion of the RNAlater, followed by rinsing with Ringer’s solution (Sigma). Samples were then placed in a rotisserie, allowing rotating incubation to take place at 37°C for 20 min. Following a modified mirVana kit extraction protocol, which includes a phenol:chloroform:IAA (25:24:1) extraction, the rest of the manufacturer’s protocol was performed as described in Hawley et al. (2017). Samples were concentrated to a final volume of 150-400 µL by centrifugation. DNA was subsequently removed using the TURBO DNA-free kit and total RNA was purified using the Rneasy MiniELute Cleanup Kit. The quality of purified RNA was verified on the Bioanalyzer using an RNA nano Analysis Kit (Agilent Technologies) to validate RNA integrity and sample quantification before cDNA library production and sequencing (see below). For both genomic DNA and total RNA, those samples demonstrating amplification at the expected band size and also passing amplification quality control were stored at -80°C until sequencing could occur.

Libraries for 16S rRNA gene amplicon, metagenomic and metatranscriptomic sequencing were constructed at the DOE Joint Genome Institute (JGI) (Berkeley, CA) and paired ends were sequenced on an Illumina Miseq (for 16S rRNA gene) and HiSeq (for the metagenomic and metatranscriptomic sequencing) platforms.

16S rRNA gene amplicon sequences were processed using JGI’s iTagger pipeline v2.0 (Tremblay et al., 2015) and annotated using the SILVA database (Pruesse et al., 2007). The metagenomic and metatranscriptomic datasets underwent quality control at JGI as described in Hawley et al. (2017), prior to assembly using Velvet. Metapathways v3.0, a modular open-source bioinformatics pipeline (Konwar et al., 2013) integrating automated gene finding and pathway prediction with Pathway Tools (Karp et al., 2002) and MetaCyc v21.5 (Caspi et al., 2013), was used for identification of open reading frames (ORFs), functional annotations and reconstruction of metabolic pathways present in both the metagenome and metatranscriptome datasets. Metapathways uses the Pathologic algorithm in Pathway Tools (Karp et al., 2002) to map functional annotations on the MetaCyc database of highly curated pathways and enzymes representing all domains of life (Caspi et al., 2013). MetaCyc emphasizes core microbial metabolism including biogeochemically relevant pathways. For both the metagenome and metatranscriptome datasets, we used metacount to create the unit of Reads Per Kilobase mapped per Million (RPKM), which takes into account gene length and total reads in a sample as a proxy for gene abundance (Konwar et al., 2015). In our analyses below we use both the MetaCyc assignments of individual pathways (hereafter base pathways) predicted in our metagenomic and metatranscriptomic datasets, and the grouping of these assignments into broader functional classes based on their biological functions and on the metabolites produced or consumed as denoted by the MetaCyc classification hierarchy (Caspi et al., 2013).

Principal Component Analyses (PCA) was performed on the physical and chemical water column parameters, with the data scaled centered on the mean. Abundance values for 16S rRNA gene amplicon sequences were generated using the USEARCH I clustering algorithm, at the 97% identity level. 16S rRNA gene abundance values and pathway abundances in RPKM were Hellinger transformed prior to multivariate statistics (i.e., hierarchical cluster analysis, non-metric multi-dimensional scaling analysis (NMDS)), and correlation analyses to enhance inter-sample comparisons. Indicator Species Analysis (ISA) were performed on the raw, non-transformed datasets. We provide the raw, non-transformed data tables for all of the molecular data as Supplementary Data sets (16S rRNA gene sequence data: Supplementary Dataset 1 , metagenomic RPKM table: Supplementary Dataset 2 , metatranscriptomic RPM table: Supplementary Dataset 3 ). All analyses were conducted in R (version 1.1.456) using the vegan (Oksanen et al., 2019), factoextra (Kassambara & Mundt, 2017), indicspecies (De Caceres & Legendre, 2009), labsdv (Roberts, 2019), cooccur (Griffith et al., 2016) and ggplot2 (Wickham, 2016) packages.

For multivariate analyses (NMDS, PCA) analyses, samples from each station were pooled according to specific depth zones in the water column: “surface” (5 - 150 m) versus “deep” water (250 - 4000 m). Classifications are based on 0.1% of surface PAR, where PAR was generally equal to 0.1% of surface levels at 150 m. This division is sensible because besides light, oxygen and nutrient availability, which both exhibit strong photic zone patterms, are the strongest selectors for the structuring of microbial diversity down the water column (DeLong et al., 2006). Analyses beyond the multivariate ordinations were conducted using classifications at particular depths (surface (5 m), DCM, 250 m, 500 m, bottom (4000 m- 5000 m) and water mass (NADW & AAIW) for greater resolution into how potential and expressed pathways changed through the water column. Visual groupings observed in multivariate ordinations were assessed for significance using the vegan ANOSIM function, on Bray Curtis dissimilarity matrices for biological data and Euclidean matrices for water column parameters, at 999 permutations. To determine key pathways driving the observed NMDS ordination patterns, indicator species analysis was used to identify indicator metatranscriptomic pathways, which were defined as those that were unique and abundant at a particular station/depth, by having an indicator value > 85 and a p value < 0.05 (Dufrêne and Legendre, 1997).

The top 20 microorganisms (defined as those with raw counts of ≥ 27,500 summed across the entire dataset, or those representing 8% of the entire dataset) used in the heatmaps were aggregated at the order level. The MetaCyc v21.5 Level 2 functional classes of the top 20 metagenomic (defined as those with RPKM values ≥ 3,600 summed across the entire dataset) and top 20 metatranscriptomic (defined as those with RPKM values ≥ 29,000 summed across the entire dataset) pathways were used for both heatmaps and cluster heatmaps, as they provided a broad overview of both potential and expressed metabolic pathways involved at specific depths. We show that aggregating our pathways into the Level 2 classes illustrates consistent patterns as those found by examining individually the top 20 most abundant base pathways in both the metagenomic ( Supplemental Figure 1A ) and metatranscriptomic ( Supplemental Figure 2A ) datasets. We acknowledge that in the Metapathways approach, which works within the Pathways Tools framework (Karp et al., 2010), when an ortholog and its associated RPKM is identified as belonging to a particular pathway, then that pathway, including any variants which the ortholog also belongs to, is retained (Konwar et al., 2015). Given the high number of pathway variants associated with the TCA cycle, we show the distribution of each TCA variant in each sample for both the metagenomic ( Supplementary Figure 1B ) and metatranscriptomic datasets ( Supplementary Figure 2B ). To show the relationships between microbial orders, metagenomic and metatranscriptomic pathways in our heatmaps, these orders and pathways were respectively clustered using the “complete” algorithm. To identify specific pathways that changed at depth, and those associated with indicator microorganisms, heatmaps were constructed using individual (base) pathways as identified in MetaCyc v21.5.

Indicator organisms, which are microorganisms that are unique and abundant at a particular site (Dufrêne & Legendre, 1997) were defined herein as those OTUs associated with an indicator value > 95 and a p value < 0.05. Correlations between microbial indicators and metatranscriptomic pathways were computed using Spearman’s rho. Spearman’s rank correlation was used here because it does not require linear relationships between parameters. Both the x (metatranscriptomic pathways) and y axes (indicator microorganisms) of bubble plots demonstrating correlations between indicator organisms and select metatranscriptomic pathways, were also clustered using the “complete” algorithm to demonstrate their inter-relatedness. To validate the associations between indicator taxa and metatranscriptomic pathways, we explored taxonomic assignments provided by the least common ancestor algorithm in MetaPathways for ORFs resolved in the metatranscriptomic dataset (Konwar et al., 2015). Here, we identified each open reading frame (ORF) belonging to metatranscriptomic pathways having a positive correlation of 0.5 or greater with indicator taxa, limiting our analysis to those ORFs present at an abundance > 10 RPKM. We then searched these taxonomic assignments to see if any indicator taxa were identified as being associated with these ORFs. This process allowed us to link the metatranscriptomic pathways identified by the correlational analysis more directly to indicator taxa.

We also used co-occurrence networks based on a probabilistic model (Veech, 2013) to determine pairwise occurrence patterns between abundance of particulate metabolite AMP (adenosine monophosphate) and specific metatranscriptomic pathways related to respiration at Station 23, as the abundance of AMP was found to be elevated at this station (Johnson et al., in revision). Finally, the Mantel test was used to assess the correlation between the 16S rRNA gene, metagenomic and metatranscriptomic datasets, respectively, with water column and metabolite data, using Kendall’s tau. Kendall’s tau was selected because it is a non-parametric test based on ranked data and is known to be more statistically robust than the Spearman correlation (Croux and Dehon, 2010). All reported p values were computed using 999 permutations. To validate Mantel associations between 16S rRNA gene amplicon sequences and particulate metabolites, we ran a Spearman correlation between indicator microorganisms, defined above, and metabolite data. Since indicator organisms are defined as driving the differentiation in community structuring, specific associations between metabolites and indicator organisms can explain the overall influence metabolites may have on the total microbial community structure.

Profiles of water column parameters (temperature, salinity, dissolved oxygen, macronutrients ( N O 3 − + N O 2 − ,   P O 4 3 − ,   T O C ,   S i O 4 4 − ,   N H 4 + ) , and select particulate metabolites (adenosine monophosphate (AMP), dimethylsulfoniopropionate (DMSP) and methylthioadenosine (MTA)) are shown in Figure 2A . We chose to highlight these metabolites because they were most strongly and significantly correlated (r ≥ 0.45, p <0.05) with the taxonomic dataset ( Supplementary Table 2 ), out of the full suite of targeted metabolites (Johnson et al., in revision). Of note, the temperature and salinity at Station 15 was generally lower in comparison to Stations 7 and 23. However, at the surface of Station 23, the salinity was markedly lower than at any other point on the cruise track, likely influenced by input from the Amazon River plume (Johnson et al., in revision). Dissolved oxygen concentrations remained >85 µmol/kg throughout the water column and followed expected open water column profiles, being high at the surface, decreasing (from 250 µmol/kg to 150 µmol/kg) at 500 m depth, and increasing beyond ~1500 m to ≥ 250 µmol/kg. Concentrations of macronutrients and total organic carbon (TOC) also followed expected water column profiles. Macronutrients were low at the surface, and generally increased with depth. TOC concentrations were higher (70-90 μM) at the surface, and decreased with depth, with the lowest concentration being 45 μM, at Station 15 at 5000 m. Silicate concentrations were similar across all stations; however, lower concentrations (~40 μM) were observed below 500 m at Station 23, compared to stations 7 and 15 (> 100 μM). The concentrations of particulate metabolites (i.e., AMP, DMSP and MTA) were highest in the surface waters and generally decreased with depth. Concentrations of AMP and MTA were particularly high (45 pM and 1 pM, respectively) in the surface waters of Station 23, with both being most concentrated at the DCM of Station 15. In comparison, the highest concentration of DMSP (4000 pM) was observed in the surface waters of Station 7.

A PCA ordination of all the physical parameters and chemical concentrations ( Figure 2A ) revealed that the water column delineated into two significantly distinct environments (ANOSIM, p < 0.05, R = 0.590) separating into the “surface” (5-150 m) and “deep” (250-4000 m) zones ( Figure 2B ). Hereafter, we use these surface and deep zones as well as finer scale separation down the water column defining different water masses/types (i.e., surface, DCM, 250 m, the oxygen minimum layer, the AAIW layer, the NADW layer, and the bottom waters) to describe our results. No spatial differentiation between the three stations (7, 15, and 23) was observed in this analysis (ANOSIM, P=0.100, R= 0.0966; Figure 2C ). Thus, based on this limited spatial resolution, it appears that our resolved physical parameters and chemical concentrations did not form distinct environmental conditions across latitude.

Microbial communities separated according to water column depth (surface and deep zones, as well as finer scale depth separation), forming significantly distinct ANOSIM, p < 0.05, R=0.810 communities across a wide range of oceanic geography, spanning all 23 stations of the cruise track ( Figure 3 ). A stacked bar plot showing the relative abundance of all phyla across depth and station in Supplementary Figure 9 . For the three stations (7, 15, and 23) where taxonomic, metagenomic and metatranscriptomic data were concurrently collected, the metagenomic (potential) pathways significantly separated ANOSIM, p < 0.05, R= 0.790 down the water column ( Figure 4A ), and significantly distinct metatranscriptomic (expressed) pathways (ANOSIM, p < 0.05, R=440) were also observed ( Figure 4B ) to be associated with the surface (0-150 m) and deeper zones (250-4000 m) of the water column.

The total number of pathways recovered for the metagenomic dataset was 881 versus the 724 pathways resolved for the metatranscriptomic dataset ( Supplementary Table 3 ). The metagenomic and metatranscriptomic surface samples both generally contained the largest number of pathways as well as the highest percentage of unique pathways. However, a large proportion (87%) of the metagenomic pathways were shared amongst the two zones (surface and deep) whereas a slightly smaller, but still large, fraction (79%) of overall metatranscriptomic pathways were shared between these regions. Indicator analysis revealed that the metatranscriptomic pathways responsible for the observed differentiation between the surface and deep ocean were those related to photosynthetic processes (i.e., seleno-amino acid synthesis & the Calvin-Benson-Bassham cycle, Figure 5 ) in the surface ocean, and recalcitrant carbon cycling (i.e., ethylbenzene degradation, Figure 5 ) in deeper waters.

The most abundant taxonomic orders (i.e., representing the top 20 microorganisms), resolved by 16S rRNA gene amplicon sequencing, varied according to major depth zones/water mass down the water column ( Figure 6A ). Cyanobacteria (comprising ~ 40% of all orders found in the photic zone) and SAR11 (comprising ~ 25% of all orders found in the photic zone), were the most prevalent groups in the surface and DCM regions of the ocean. Members of SAR324, comprising ~ 8% of all orders found at depths below the DCM, and Marinimicrobia, a microbial dark matter phylum formerly known as Marine Group A and SAR406 (Allers et al., 2013; Wright et al., 2014; Hawley et al., 2017), comprising~12% of all phyla found at depths below ~ 150 m, were more prevalent immediately below the DCM and in the deeper ocean. Thaumarchaeota Marine group 1 (Swan et al., 2014) were most prevalent in the AAIW, and Alteromonadales was predominant in the bottom waters of the western Atlantic Ocean ( Figure 6A ).

The most abundant metagenomic pathways corresponded to specific core functional processes related to cellular housekeeping (i.e., biosynthesis including nucleic acid production and amino acid production (i.e., arginine & methionine synthesis), and cellular metabolism (i.e., tRNA charging, TCA cycle pathways, biosynthesis) and remained highly abundant throughout the water column ( Figure 6B and Supplementary Figure 1 ).

The majority of the top 20 metatranscriptomic pathways were also generally present in low proportion (< 50%) at all depths; and also exhibited differential expression across the water column ( Figure 6C ). Pathways related to photosynthesis and electron transfer ( Figure 6C and Supplementary Figure 2 ), were found to be most prevalent (> 50%) in the surface, DCM, and surface/AAIW respectively. Specifically, amino acid synthesis and pathways related to energy metabolism involving NAD/NADH phosphorylation and dephosphorylation processes were prevalent in the surface ocean ( Figure 6C and Supplementary Figure 2 ). Degradation pathways, such as degradation of C1 compounds (i.e., methanol oxidation) and pathways related to the TCA cycle dominated in the bottom regions (AAIW, NADW, bottom) of the water column ( Figure 6C and Supplementary Figure 2 ).

To determine the relationships between metatranscriptomic pathways and the microbial taxa found at depth, we calculated Spearman correlations between taxonomic indicator OTUs in specific compartments of the water column and all metatranscriptomic pathways. We resolved the differences in microbial community expression at particular depths (surface, DCM, 250 m, oxygen minimum zone, & bottom waters) and between water masses (i.e., the AAIW and NADW). Top microbial indicator OTUs (i.e., those having been assigned a significant indicator value ≥ 95) were associated with the DCM, NADW and bottom waters ( Figure 7 and Supplementary Figure 3 ).

Indicators for the DCM included microorganisms related to Flavobacteriales, SAR11, Bradymonadales and Legionalles orders ( Figure 7A ), where Flavobacteriales was most strongly and positively correlated with pathways relating to aldehyde degradation, specifically with pathways related to sulfoacetaldehyde catabolism (Spearman correlation > 0.8) and was negatively correlated with both carboxylate and carbohydrate degradation ( Supplementary Figure 3A ). Within the DCM, SAR11, Bradymonadales and Legionalles were most positively correlated with siderophore synthesis, alcohol degradation, hormone and aromatic compound degradation, and most negatively correlated with amino acid and nucleotide degradation ( Figure 7A ). Specific pathways positively correlating with Bradymonadales and Legionalles included ethanol, chlorinated phenol compounds and aromatic amine degradation. SAR11 was positively correlated with similar degradation pathways, in addition to those associated with degrading phytoplankton-related by-products such as DMSP and phytol ( Figure 7A ). We note that Cyanobacteria are also key members of both the DCM and surface waters and are indicators for these zones at an indicator value of 90, but are not shown on Figure 7 , which is limited to a higher IV value (IV > 95) to aid data visualization.

Indicators for the NADW included members related to phylum Proteobacteria & Phycisphaerae, along with orders belonging to Rhodospirillales and Verrucomicrobiales ( Figure 7B ). Pathways most strongly correlated with these microorganisms included those related to C1 compounds (most notably, the oxidation of formaldehyde compounds), chlorinated compound degradation and chemoautotrophic energy metabolisms (specifically pathways relating to hydrogen, ammonia, nitrate and nitrite oxidation), and carboxylate degradation ( Figure 7B ). Alphaproteobacterial indicators were most positively correlated with pathways related to amino acid biosynthesis (especially tyrosine and alanine production), and cell structure biosynthesis, particularly with regards to sporopollenin and peptidoglycan production ( Figure 7B and Supplementary Figure 3B ).

Taxonomic indicators for the bottom waters of the western Atlantic Ocean were related to members of SAR324, SAR202, Flavobacteriales, Oceanospirillales, Sphingobacteriales, and to phyla Chloroflexi and Proteobacteria, along with Archaea related to Thermoplasmatales and phylum Woesarchaeota ( Figure 7C ). Amongst these, organisms related to Chlorobi, Sphingobacteriales, SAR324, Oceanospirillales and Proteobacteria shared similar correlations to various pathways, such as a strong positive correlation to pathways related to mercury detoxification, nucleotide, and polyamine biosynthesis and general degradation pathways related to akylnitronate, glucose, chitin and nicotine break down ( Figure 7C ). These microorganisms also had a strong negative correlation with pathways involved in fatty acid & lipid degradation, such as with oleolate oxidation and triacylglycerol degradation ( Supplementary Figure 3C ). Other notable pathway relationships with bottom water indicator organisms included the strong positive correlation between Thermoplasmatales and secondary metabolite biosynthesis, such as salidroside and canavanine biosynthesis ( Figure 7C ). The indicator related to phylum Chloroflexi was also found to strongly and positively correlate with similar pathways that also correlated with Thermoplasmatales, for example, secondary metabolite biosynthesis and methanogenesis using trimethylamine substrates ( Figure 7C ).

Strong positive associations (Spearman correlation > 0.5) between indicator taxa and metatranscriptomic pathways were further verified by determining the taxa assigned using the Metapathways least common ancestor algorithm with the functional ORFs resolved in the predicted metatranscriptomic pathways ( Supplementary Table 4 ). In most instances, we were only able to resolve microbial taxonomy for abundant groups from the metatranscriptomic dataset for a particular water column compartment (i.e., SAR11 in the DCM; Rhodospirillales in the NADW and Gammaproteobacteria in bottom waters). However, these microbial taxa mirrored some of the associations observed in our correlational analysis. For example, in the DCM, a functionally annotated ORF coding for an aldehyde dehydrogenase ( Supplementary Table 4 ) was determined to be taxonomically related to SAR11. Given that the DCM indicator SAR11 positively correlated with alcohol degradation ( Figure 7A ), the aldehyde dehydrogenase, an ORF resolved from this pathway, may be a component of its alcohol degradation capabilities. Thus, this analysis was useful in corroborating the relationship determined using our correlational approach. Similarly, in the NADW, a functionally annotated ORF coding for methanol dehydrogenase subunit 1 ( Supplementary Table 4 ), which is a component of C1 compound (methanol) degradation, was associated with Rhodospirillales sp. and Alphaproteobacteria ( Figure 7B ). In our correlational analysis, C1 compound metabolism was a pathway class with which both Alphaproteobacterial and Rhodospirillales indicators positively correlated ( Figure 7B ). Lastly, in the bottom waters, an ORF encoding a nucleoside diphosphate kinase ( Supplementary Table 4 ), was found to be associated with Gammaproteobacteria, and thus may validate the correlation between the indicator Gammaproteobacteria Oceanospirillales with nucleotide biosynthesis ( Figure 7C ). Taken together, metatranscriptomic pathway analysis with associated indicator taxa reinforce the utility of using a pathway-centric approach for identifying potential key biogeochemical processes driving matter transformation within specific water column compartments and geographies.

We used a Venn diagram to visually assess the distribution of metabolic pathways by latitude ( Figure 8A ). In total, 62% of metatranscriptomic pathways were shared across all three stations, while Station 7 contained the highest number of pathways (802) and Station 23 the lowest (595) ( Supplementary Table 5 ). Station 7 also contained the most unique pathways (11%) compared to Station 15 (3.8%) and Station 23 (1.9%), respectively ( Figure 8A ). In contrast to the metatranscriptome, the abundance of metagenomic pathways was relatively consistent between stations. There was an average number of 896 pathways present and 80% of the pathways were shared across all three stations ( Supplementary Table 5 ). Normalizing the RPKM values, by dividing total RPKM counts with the overall number of resolved pathways, reveals the same trend as the unnormalized datasets ( Supplementary Table 5 ).

The top 20 base metatranscriptomic pathways (MetaCyc assignments of individual pathways) appeared to be differentially expressed across all three stations though all pathways occurred in low proportion (< 50%, Figure 8B ). Station 23 appeared to contain greater expression of pathways related to respiration (particularly aerobic respiration, ( Supplementary Figures 4, 5 ), some of which were unique pathways not observed at the other stations, such as pathways related to oxidative stress (i.e., ectoine synthesis and thioredoxin biosynthesis). Station 15 contained 6% more pathways related to photosynthesis compared to Stations 7 and 23. Station 15 also contained 8.5% more biosynthesis pathways (including specifically amino acid biosynthesis, for example, cysteine and seleno amino acid biosynthesis), compared to other stations, whereas Station 7 contained a greater number (8% more) of pathways related to the TCA cycle (specifically, TCA cycle via the oxoglutarate ferrodoxin oxidation pathway), and energy metabolism (i.e., the methylaspartate cycle, Figure 8B and Supplementary Figure 5 ).

To evaluate possible drivers of microbial community and metabolic pathway composition, we conducted Mantel tests between our physical, chemical, and metabolite (dissolved and particulate) measurements with each of the high throughput sequencing datasets (16S rRNA gene amplicon, metagenomic and metatranscriptomic data). This analysis revealed that microbial community taxonomic structure is strongly and significantly (r >0.5, p <0.05) influenced by physical water column properties like temperature, density (i.e., related to water mass) and concentrations of TOC and major nutrients ( N O 3 − +   N O 2 − ,   P O 4 3 − , TN, and S i O 4 4 − ) ( Figure 9 ). Taxonomic structure was also strongly and significantly influenced by concentrations of particulate (combined 0.7 µm + 0.2 µm filter membranes) metabolites such as adenosine monophosphate (AMP), dimethylsulfoniopropionate (DMSP) as well as methylthioadenosine (MTA) ( Figure 9 ). In contrast, the structure of pathways resolved in the metagenome were not strongly and/or significantly influenced by any of the metabolite parameters and many of the environmental parameters (excepting N O 3 − +   N O 2 − ,   P O 4 3 − , and TN), except for S i O 4 4 − ( Figure 9 ). Metatranscriptomic pathway structure, however, was significantly driven by the same metabolites driving microbial taxonomic structure, although not as strongly (0.2 ≤ r ≤ 0.5, Figure 9 ). In terms of other oceanographic properties, expressed pathways were weakly, but significantly (r < 0.25, p <0.05) influenced by temperature and nutrient ( P O 4 3 − ,   a n d   N O 3 − + N O 2 − ) concentrations ( Figure 9 ). Both metatranscriptomic pathways and taxonomic community structure were also significantly influenced, albeit weakly (r ≤ 0.40), by other particulate metabolites, including certain amino acids (phenylalanine, tryptophan, glutamic acid), and nucleosides (adenosine and guanine, Supplementary Table 2 ). However, none of the high throughput sequencing datasets were significantly impacted by any of the dissolved (<0.2 µm filtrate) metabolites measured ( Supplementary Table 2 ).

Overall, microbial communities appear to cluster into significantly distinct assemblages that can be divided into two broad marine regions: the surface (5 - 150 m) and deep (250 - 4000 m) ocean ( Figure 3 ), as shown by others previously (DeLong et al., 2006; Zinger et al., 2011; Friedline et al., 2012; Milici et al., 2016a; Fu et al., 2019). Here, we observe that different microbial taxa flourish as a result of the unique conditions defining the surface, DCM, oxygen minimum and deep waters. For example, microbial orders related to Cyanobacteria and SAR11 were most prevalent in the photic zones of the ocean ( Figure 6A ), consistent with many previous findings in tropical and mid-latitude oceans (i.e., Zubkov et al., 1998; Jing et al., 2013). In contrast, members of Thaumarchaeota, SAR324 and sinking particle associated Alteromonadales (Boeuf et al., 2019) dominated the bathypelagic ocean (> 1000 m: Figure 6A ). Thaumarchaeota are among the most abundant microbial groups in the global mesopelagic and deep ocean (Swan et al., 2014), and are known ammonia oxidizers (Pester et al., 2011). They generally increase in abundance at greater depths (Amano-Sato et al., 2013), possess both shallow and deep-water ecotypes (Hatzenpichler, 2012; Santoro et al., 2017) and are especially prevalent in the open ocean where low N H 3 / N H 4 + fluxes prevail (Santoro et al., 2010). Many clades associated with the deep ocean have potential to contribute to dark ocean carbon cycling since they are flexible chemolithotrophs capable of degrading a range of substrates, including a variety of C1 and sulfur-containing compounds (Wright et al., 1997; Swan et al., 2011; Sheik et al., 2013). The Deltaproteobacteria SAR324 are also often found at depth and may be associated with sinking particles (Swan et al., 2011). Heterotrophic bacteria, such as the Gammaproteobacteria Alteromonadales, are abundant in the open ocean and have also been found to remain active on sinking particles as they transition down the water column (Boeuf et al., 2019). Alteromonadales are also believed to be central conduits for nutrient and DOC cycling, contributing to organic matter turnover at depth (Pedler et al., 2014).

Many of the microbial metabolic pathways recovered in the metagenome were found ubiquitously throughout the water column ( Figure 6B ). The larger proportion (87%) of shared metagenomic pathways between the surface and deep ocean is consistent with the presence of a persistent core of functional pathways used for cellular housekeeping and central metabolic processes that are found in nearly all microorganisms, regardless of in situ environmental conditions (Furhman, 2009; Ferreira et al., 2014; Hanson et al., 2014). For example, the most abundant pathway observed in the metagenomic dataset was the TCA cycle, or variants thereof, which are central to energy metabolism and carbon processing for a variety of aerobes and anaerobes (Huynen et al., 1999; Makarova & Koonin, 2003). Yet, even within the metagenome, it is still possible for a unique set of pathways to define the surface and deep ocean (Ferreira et al., 2014), such as photosynthetic pathways observed in the photic zone, and energy metabolism pathways (methylaspartate cycles) in the deeper zones of the ocean ( Figure 6B and Supplementary Figure 1 ). These specific differentiations drive the observed clustering patterns in our NMDS ordination ( Figure 4A ) beyond broad compartments like the “surface” and “deep” ocean. Functional redundancy refers to the fact that multiple, co-existing, yet taxonomically distinct, organisms are able to carry out the same metabolic function (Louca et al., 2018). Ultimately, this leads to the observation of a stable functional composition within a metagenome and/or a metatranscriptome despite a highly variable taxonomic composition (Louca et al., 2018). Here, such redundancy is manifested in the fact that diverse taxonomic groups can perform photosynthesis in the surface ocean as well as similar metabolisms to generate energy throughout the water column ( Figure 6 and Supplementary Figure 8 ). In the metatranscriptome, we observe this functional redundancy, for example, in the DCM, where photosynthetic pathways were positively correlated to the same degree for SAR11, Legionellales and Bradymonadales, likely as a consequence of these organisms feeding off of photosynthetic byproducts ( Supplementary Figure 3A ). Similarly, in the deeper regions of the ocean, diverse bottom-dwelling microbial indicator groups (Thermoplasmatales, SAR202 and Flavobacteriales) were all positively correlated with carboxylate degradation ( Supplementary Figure 3B ).

The greater variation in metatranscriptomic pathways down the water column ( Figure 6C ) emphasizes the diverse metabolic responses found at depth extremes (i.e., surface, deep waters, DeLong et al., 2006; Friedline et al., 2012). For example, photosynthetic pathways (i.e., light reactions and the Calvin-Benson-Bassham cycle, Supplementary Figure 2 ) and pathways corresponding with energy metabolism and electron transfer were, unsurprisingly, most prevalent in the photic zones, although electron transfer was notably less prevalent at the oxygen minimum zone ( Figure 6C and Supplementary Figure 2 ). Further, the presence of surface indicator pathways related to seleno-amino acid production ( Figure 7 ) is likely due to the uptake of dissolved selenium by phytoplankton (Cutter and Bruland, 1984), which is used in the production of proteins required for cell viability in many algal species (Araie & Shiraiwa, 2009). In contrast, metatranscriptomic pathways related to the degradation of recalcitrant carbon and C1 compounds and to the TCA cycle (or variants thereof) were more abundant in the deep ocean ( Figure 6C and Supplementary Figure 2 ). This may suggest that either there are fresh sources of detritus at lower depths replete with low molecular weight carbon (Kirchman, 2018) or a prevalence of microbially produced methane by deep sea sedimentary archaea (Colwell et al., 2008). Additionally, these recalcitrant carbon and C1 degradation pathways may also be linked to methylotrophic breakdown of both particulate and dissolved organic matter at depth (McCarren et al., 2010; Fang et al., 2014).

Probing the metatranscriptomic dataset more deeply, we observe different correlational relationships between taxonomic indicators for distinct water column compartments and expressed pathways ( Figure 7 ). In the DCM, SAR11 and Bradymonadales, a novel Deltaproteobacteria group (Mu et al., 2020) with versatile predation strategies, were strongly and positively correlated with alcohol and aromatic compound degradation pathways ( Figure 7A and Supplementary Figure 3A ), consistent with photic marine waters being richer in phytoplankton and cyanobacterial derived carbohydrate by-products (Mühlenbruch et al., 2018). This is exemplified by the strong positive correlation of SAR11 with DMSP and phytol degradation ( Figure 7A ), by-products of marine phytoplankton osmoprotectant production and a chlorophyll a component, respectively (Yoch, 2002; Salih and Celikbicak, 2012; Archer et al., 2018). Additionally, strong positive correlations between DCM indicators and pathways related to chlorinated phenol, aromatic amine and ethanol compound degradation suggest that the metabolic capacity to cope with organic pollutants is potentially present in the western Atlantic Ocean (Ghosal et al., 2016; Ghattas et al., 2017).

In the deep ocean, we observed that different indicator taxa and expressed pathways were present in the NADW compared to the bottom waters ( Figures 7B , 7C and Supplementary Figures 3B, C ), suggesting that these water masses have distinct microbial communities with specific metabolic processes (Shi et al., 2011; Lekunberri et al., 2013; Guerrero-Feijoo et al., 2016; Fu et al., 2019). SAR202, SAR324, and certain orders of Archaea have been found in the deep ocean below 1500 m (Vetriani et al., 1999; Morris et al., 2004; Sheik et al., 2013; Wu et al., 2013), whereas Rhodospirillales are especially common at intermediate depths (~500 m) and below 1500 m in the North Atlantic (Guerrero-Feijoo et al., 2016). In particular, the positive correlation between the Bottom indicator group SAR202 (an abundant aphotic zone heterotroph from the phylum Chloroflexi; Mehrshad et al., 2018) and pathways related to the oxidation of recalcitrant carboxylate compounds (i.e., oxobutanoate compounds; Figure 7C and Supplementary Figure 3C ) is consistent with the presence of multiple genes encoding carboxylic acid degradation previously found in abundance in the deep ocean (Landry et al., 2017). Recent work analyzing the SAR202 genome and its DOM degrading capacity has suggested a role in the breakdown of refractory DOM persistent in the deep ocean (Landry et al., 2017; Saw et al., 2020). Overall, recalcitrant DOM degrading capabilities appear more prevalent in deep-dwelling marine microorganisms (Sheik et al., 2013), likely supporting their growth in the bathypelagic ocean (Landry et al., 2017; Kirchman, 2018). Like SAR202, SAR324 is ubiquitously present in the global dark ocean, with previous work illustrating that its metabolic versatility likely plays a role in its widespread distribution (Sheik et al., 2013). Here, we observe a strong positive correlation between SAR324 and lipid and alcohol degradation pathways, such as oleate beta oxidation and ethylene glycol and acetone degradation ( Supplementary Figure 3C ). This correlation may be indicative of specific metabolic versatility in SAR324 with regards to fatty acid metabolism, as oleate and beta oxidation involve the degradation of unsaturated fatty acids, producing two acetyl CoA molecules (Nie et al., 2008), which SAR324 may funnel into the TCA cycle for energy production (Sheik et al., 2013).

In contrast, we observed that though bottom-dwelling microbial indicator groups like Thermoplasmatales and Chloroflexi did not demonstrate similar patterns in terms of recalcitrant carbon degradation and/or fatty acid degradation ( Figure 7 ) as the NADW indicator groups, the bottom-dwelling indicators did show a strong positive correlation (r>0.5) with a specific methanogenesis pathway (methylotrophic methanogenesis from trimethylamine, Figure 7C ). Thermoplasmatales, as a member of the phylum Euryarchaeota are likely capable of methylotrophic methanogenesis, and members of the phylum Chloroflexi, such as Anaerolineaceae, have been previously shown to contain genes involved in methanogenesis (Paul et al., 2012; Liang et al., 2015). Given the higher oxygen concentrations found in the bathypelagic layers of the ocean ( Figure 2A ), methanogenesis may be occurring within the lower oxygen micro-environments of marine snow particles (Steiner et al., 2020). Branching of Thermoplasmatales and Chloroflexi away from other indicator taxa in Supplementary Figure 3C indicates that they may occupy a different metabolic niche compared to other indicator taxa for the bottom waters and may reflect different metabolisms required for particle-associated vs planktonic lifestyles (Zhang et al., 2015; Milici et al., 2017).

Examining taxonomic and functional patterns latitudinally across our transect revealed divergent patterns where microbial community composition showed no variation (data not shown) but predicted metabolic pathways partitioned across latitude. This geographical homogeneity in microbial community composition is similar to that observed elsewhere (Zinger et al., 2013; Salazar et al., 2016; Djurhuus et al., 2017; Ibarbalz et al., 2019), and is likely a result of our cruise track spanning the South Atlantic subtropical gyre and spatially expansive deep-water masses (i.e., AAIW, NADW). Gyres have been shown to facilitate the transport and mixing of microbial communities (Milici et al., 2016b), contributing to the relative uniformity observed latitudinally in microbial community distribution (Fu et al., 2019). Different deep-water masses have been shown to have characteristic temperature and salinity signatures and may serve to transport microbial communities across oceanic basins, contributing to homogenization of latitudinal differences in microbial community composition (Galand et al., 2010; Friedline et al., 2012; Pernice et al., 2016; Djurhuus et al., 2017).

In contrast, the predicted expressed pathways from the metatranscriptome did differ between latitudinally distinct stations (7, 15, and 23) ( Figure 8 ), which we hypothesize is driven by different nutrient conditions as a result of riverine input or upwelling. Station 23 is close to the Amazon River plume, which spans thousands of kilometers due to freshwater input from the world’s largest river (Araujo et al., 2017; Varona et al., 2019). This riverine input delivers large amounts of terrestrial carbon and nutrients to the surrounding ocean, altering metabolic activities of impacted marine microbial communities (Satinsky et al., 2017). The high expression and relative abundance of aerobic respiration pathways at Station 23 ( Supplementary Figure 4 and Supplementary Figure 5C ) is consistent with increased metabolic activity associated with a plentiful flux of organic carbon and nutrients (Satinsky et al., 2017). Additionally, the presence of ectoine synthesis pathways at Station 23, absent in the other stations ( Supplementary Figure 5 ), suggests a response to osmotic stress. The presence of both aerobic metabolic pathways and ectoine synthesis may allow microorganisms to better respond to chemical fluctuations and/or take advantage of the carbon influx from the Amazon River at shallower depths (Richter et al., 2019). To a lesser extent, compared to the Amazon River plume, equatorial upwelling at other locations along our cruise track (i.e., Station 15), can also mix the water column and lead to elevated concentrations of nutrients and metabolites at the ocean surface (Wang et al., 2017; Johnson et al., in revision). Indeed, the high relative abundance of pathways associated with aerobic respiration at Station 15 ( Supplementary Figure 5B ) is consistent with an expected higher metabolic activity associated with sources of organic carbon and nutrients from below the nutricline. Further, an increased relative abundance of predicted pathways related to photosynthesis at Station 15 (i.e., seleno amino acid biosynthesis) compared to Stations 23 and 7 ( Figure 8B and Supplementary Figure 4 ) may also indicate a nutritive environment conducive to enhanced photosynthetic activity in the surface ocean (Richardson & Bendsten, 2017).

The differences in the metatranscriptome between stations is also evident in the distribution of particulate (combined 0.7 µm and 0.2 µm fractions) metabolites generally indicative of greater microbial abundances and metabolic activity (Johnson et al., in revision). Metabolites, the small molecule intermediary products of metabolic reactions catalyzed by cellular enzymes, can provide a snapshot of the microbial community’s metabolic response to fluctuations in environmental conditions across geography (Johnson et al., in revision). Here we observed that the elevated presence of certain metabolites (i.e., AMP and MTA) are closely correlated with increased microbial abundance at more nutrient rich sites (Johnson et al., in revision). AMP is a major contributor to the marine phosphorous pool and other ubiquitous cellular processes, such as RNA synthesis, regulation of purine breakdown (Liechti and Goldberg, 2011), nucleotide synthesis and interconversion between the energy-carrying molecules ADP and ATP (Hardie, 2018). Its universal presence in many cellular processes thus makes it a general indicator for enhanced metabolic activity. At Station 23, elevated AMP (Johnson et al., in revision) coincides with an increased presence of broad classes of metatranscriptomic pathways related to respiration ( Figure 8B ). However, we note that no statistically significant pairwise co-occurrence patterns between specific metatranscriptomic pathways related to respiration (i.e., alternative oxidase pathways, oxoisovalerate decarboxylation, chlorate reduction) and AMP were found, likely due to the ubiquity of AMP across cellular processes. An increase in the abundance of particulate MTA was also observed at Station 15 but not at Station 7, located in the oligotrophic South Atlantic subtropical gyre (Johnson et al., in revision). In distinct oceanic regions, levels of MTA and riboflavin, as well as 4-aminobenzoic acid (PABA) and pantothenic acid (vitamin B₅), have been shown to closely correlate with abundant microbial community members, and by extension, microbial metatranscriptomic pathways (Johnson et al., in revision). For example, pico/nanoeukaryotes in the North Atlantic Ocean have been found to be linked to an elevated concentration of MTA and riboflavin as these organisms contribute significantly to the overall organic sulfur and nitrogen pool in this region (Johnson et al., in revision). Conversely, pantothenic acid and PABA have been found to be more prevalent in the surface ocean at the equator (in this study represented by Station 15) and further south in the Atlantic (here, represented by Station 7) (Jonhson et al., in revision). The elevated concentrations of these metabolites are likely due to the abundance of Cyanobacteria at these stations (Johnson et al., in revision). Cyanobacteria have been found to frequently release organic carbon in the form of carboxylic acids, such as PABA (Fiore et al., 2015), which heterotrophic organisms found in the surface ocean can then utilize. Previously, Becker et al. (2019) noted that the cyanobacteria Procholorococcus are able to support SAR11’s organic carbon requirement by virtue of the DOM exudates they produce. Similarly, in this study we observed a strong correlation between the DCM indicator organism SAR11 and degradation pathways ( Figure 7A ). Finally, levels of guanine, an essential constituent of DNA often correlated with cell abundance, were found to be elevated in the particulate fraction at the surface waters of Station 7, Station 15 and Station 23 (Johnson et al., in revision). Correspondingly, metatranscriptomic pathways related to guanine metabolism (guanine salvage) were also elevated in the same surface waters of Stations 7, and 23 ( Supplementary Figure 7 ).

Understanding how the environment selects for and impacts microbial community structure and function is integral to the study of marine microbial ecology. Such integrated analyses shed light on how the unique conditions in water column structure shape microbial community structure. Our correlated dissimilarities of total taxonomic and functional composition with our water column physical, chemical, and particulate metabolic parameters ( Figure 9 ) aims to broadly identify the environmental and metabolic drivers influencing the community structure and functional gene expression captured by our datasets. We also use targeted metabolites, including vitamins, amino acids and other biosynthetic intermediaries to bridge the microbial metatranscriptome and, hopefully, the dynamic microbial community response, to environmental conditions down the water column.

The cycling of nutrients is most dynamic in the photic zone, where cyanobacteria, phytoplankton and zooplankton continuously produce and break-down a variety of marine metabolites (both dissolved and particulate, acting as intermediaries for metabolic reactions), polysaccharides and inorganic molecules (Meyer et al., 2016). Particulate break-down products amalgamate and sink, thus transported to lower depths, where the more recalcitrant particles may eventually settle. In this way, recalcitrant metabolites may influence the diversity of deep-water microorganisms, by selecting for those able to metabolize these particular energy sources (Meyer et al., 2016). Due to the continual release and sinking of particulate break-down products in the water column, the particulate metabolite profile can also vary directly in association with changes in abundant microbial taxa, particularly in waters above 200 m (Johnson et al., in revision). Taxonomic profiles observed in this study were consistent with previous reports in other oceanic locales (DeLong et al., 2006, Furhman et al., 2015; Sunagawa et al., 2015), with microbial community composition being strongly and significantly influenced by physical water column properties (e.g., temperature, depth), and also concentrations of chemical species, like TOC, major nutrients ( S i O 4 4 − ,   N O 3 − ,   P O 4 3 − ) and particulate metabolites such as DMSP, AMP and MTA ( Figure 9 ). Similar to the microbial community, the overall diversity in the functional (metatranscriptomic) dataset was impacted by the availability of nutrients, as found by other studies (McCarren et al., 2010; Gifford et al., 2016; Salazar et al., 2019), though with weaker correlation ( Figure 9 ). The metabolome represents the functional activity of cells which are produced in response to a variety of environmental cues ranging from nutrient availability (Kujawinski et al., 2017) and oxidative stress (Jones et al., 2019) to temperature changes (Johnson et al., in revision). Like the taxonomic community composition, the metabolic pathways resolved in this dataset representing these activities also showed a significant correlation with particulate metabolites (DMSP, MTA and AMP) ( Figure 9 ), thus illustrating the power of metatranscriptomic data to provide a snapshot of microbial activity under specific environmental conditions (Aguiar-Pulido et al., 2016).

The strong and significant correlation between particulate metabolites and both the taxonomic and functional datasets ( Figure 9 ) is likely because marine metabolites are produced as specific microorganisms respond to changes in their surrounding environment (de Carvalho and Fernandes, 2010). For example, DMSP, a by-product of marine phytoplankton osmoprotectant production (Yoch, 2002; Archer et al., 2018), is used as an energy source by microbial communities in the photic zone once it is released into the dissolved organic matter pool (Kiene et al., 2000). A Spearman correlation of this metabolite with indicator taxa showed that particulate DMSP is strongly associated with our indicator microbial organisms for the DCM, like Bradymonadales, SAR11 and Legionellales ( Supplementary Figure 6 ). DMSP assimilation and degradation, through both biotic and abiotic processes, fuels a variety of diverse carbon and sulfur pathways (i.e., oxidation of sulfur for energy or incorporation of sulfur into select amino acids, Moran et al., 2012). This metabolite has been found to support ~13% of microbial carbon degradation processes as it is a readily available substrate in surface waters (Kiene et al., 2000). DMSP is also an important source of reduced sulfur for SAR11 strains (Liu et al., 2018), with its degradation by-products, dimethyl sulfide and methanethiol, having been implicated in elevated greenhouse gas contributions to the atmosphere (Liu et al., 2018). DMSP’s ability to feed into multiple carbon and sulfur pathways may therefore explain why there is a strong and significant association between this metabolite, microbial community structure and, also, the metabolic pathways expressed in the microbial metatranscriptome ( Figure 9 ). Similarly, since MTA is a toxic by-product originating from sulfur salvage and/or polyamine synthesis, which is then expelled into the extracellular environment (North et al., 2017), we expect that this metabolite will follow similar trends as DMSP ( Figure 9 ). Finally, we observed a strong significant association between AMP, microbial community composition and the microbial metatranscriptome ( Figure 9 ). As noted above, AMP’s ubiquity across cellular processes makes it a broad indicator for higher metabolic activity. Here, we see this fact reflected by AMP correlating with abundant surface microbial community members (i.e., Cyanobacteria, Figure 6A ) at different stations (i.e., DCM of Station 15 and the surface waters of Station 23, Johnson et al., in revision).

Other particulate metabolites, such as amino acids and purines ( Supplementary Table 2 ), were also found to have some correlation with the composition of microbial communities and metatranscriptomic pathways, although the strength of the correlation was less than that for AMP, DMSP, and MTA ( Supplementary Table 2 ). All of these metabolites ( Supplementary Table 2 ) form an important part of the marine organic carbon pool and are often rapidly cycled as they are ubiquitously required components of life (Moran et al., 2016, Johnson et al., in revision). Some of these metabolites (i.e., guanine, AMP, and most amino acids targeted in the study) demonstrate an almost universal presence across the ocean, in terms of both depth and latitude, and tend to associate with areas characterized by higher microbial biomass, like surface waters (Johnson et al., in revision). However, the low strength of correlation between each of these individual metabolites and both the microbial community composition and expressed functional pathways points to the fact that often specific metabolites cannot fully explain the variation present in taxonomic and functional datasets. This may be the result of the presence of an underlying core of omnipresent metabolic pathways, similar to observed in the metagenomic dataset, whose consistent distribution across the water column is ultimately driving the distribution metabolites targeted in this study. Amino acids, such as tryptophan, phenylalanine, isoleucine and leucine, along with purines such as guanine and adenosine, are often associated with deeper marine waters. These molecules can travel down the water column as a component of marine snow (Johnson et al., 2020). Some, like in the case of phenylalanine and tryptophan, are also likely produced at depth by endogenous microbial communities (Dauwe et al., 1999). Indeed, in this study, we observed a strong correlation between guanine, phenylalanine, and tryptophan with bottom-dwelling indicator microorganisms ( Supplementary Figure 7 ). However, this correlation may be influenced by the fact that these specific metabolites were some of the only deep-water metabolites identified above the detection limit (Johnson et al., in revision).

In contrast to the taxonomic community composition and metatranscriptomic datasets, the metagenomic pathways correlated strongly and significantly with only one water column property, S i O 4 4 − concentrations ( Figure 9 ). This correlation may be due to a diatom bloom occurring at the time of sampling at Station 23 (Howard et al., 2017) or to increased Si export from the Amazon River (Gomes et al., 2018). The finding that most of the microbial metagenome is shared across the water column emphasizes that the metagenome of marine microorganisms, based on metabolic pathways predicted here, has a high degree of homogeneity with “all functions are everywhere, but the environment selects” which genes are expressed (Raes et al., 2011). As found elsewhere, the microbial metagenome is primarily composed of pathways such as housekeeping genes (Furhman, 2009; Ferreira et al., 2014) that are present irrespective of the environmental conditions outside of the cell. In the future, however, further sampling should be undertaken in order to validate the associations shown here between the marine environment and the microbial metagenome.

Finally, our analysis did not identify any significant associations between dissolved metabolites, which include an array of vitamins, nucleic acids, osmolytes, amino acids and metabolic intermediates (i.e., biotin, caffeine, 4-aminobenzoic acid (PABA), p-hydroxybenzoic acid (PHBA), pantothenic acid, phenylalanine, tryptophan), and taxonomic community composition, metagenomic pathways or metatranscriptomic pathways. This finding is consistent with prior conclusions that these dissolved metabolites are likely rapidly consumed and cycled as they are released into the ocean (Moran et al., 2016). Thus, their generally low and/or sporadic concentrations make it challenging to tie their distribution to the observed variation in the microbial taxonomic community or the total potential and functional gene pools. One exception to this trend was with regards to dissolved riboflavin, which was significantly correlated with the metatranscriptomic dataset ( Supplementary Table 2 ). Recent studies have found that dissolved distributions of this metabolite can be controlled by abundant levels of photosynthetic microorganisms, like Procholorococcus and Synechococcus (Johnson et al., in revision). Patterns of abundance in these organisms at different oceanic depths across the transect ( Figure 6 ) may thus drive the significant correlation between dissolved riboflavin and expressed functional pathways.