Main

The biological carbon pump annually supplies 5–10 Gt of carbon (C) to the oceans’ interior1,2, driving long-term sequestration and providing energy for deep-sea ecosystems. The pump efficiency is set initially by myriad transformations of newly formed particles in the euphotic zone (including aggregation and grazing), followed by pronounced vertical flux attenuation via mesopelagic processes that ultimately influence atmospheric carbon dioxide (CO2) concentrations3. The vertical particulate organic carbon (POC) flux attenuation is typically measured by multiple-depth radionuclide sampling, sediment-trap deployments and/or optical particle measurements. They provide estimates of POC flux attenuation often expressed as a power-law exponent or b value6. The b value is widely used to intercompare and understand flux attenuation patterns across different provinces7,8,10. It is also used to explore the role of factors including oxygen11, temperature7, primary productivity12 or floristics8,10 in driving global patterns of POC flux attenuation.

Despite over three decades of research, there is no consensus on the controls of POC flux attenuation; opposing latitudinal trends in b values have emerged from global syntheses8,10, observations7 and model analyses13,14 (Extended Data Table 1). Such disagreement is unsurprising as POC flux attenuation is the cumulative signature of distinct transformations from zooplankton flux-feeding to particle-attached microbial degradation15. A better understanding of the factors that control the biological carbon pump therefore requires a separation of the processes that jointly set POC flux attenuation. Teasing these roles apart can provide insights (here, directly for microbes and indirectly for zooplankton) into their relative magnitude, vertical trends, controls and degree of influence in setting regional patterns in particle flux attenuation. Such a mechanistic approach will improve model projections16,17 needed to resolve the contribution of microbes and zooplankton to current and projected future changes in the functioning of the biological carbon pump3,18.

Particle degradation by attached bacteria, and other microbes (viruses, archaea and microbial eukaryotes including microzooplankton), is pivotal in POC flux attenuation through colonization and transformation of sinking particles17,19 (Extended Data Fig. 1). Wide-ranging terms, including particle remineralization, recycling, solubilization and respiration are employed for this microbial degradation process (Extended Data Fig. 1). Here we use microbial remineralization (MR) throughout to represent the microbially mediated component of POC flux attenuation. We present the mesopelagic MR of sinking microbially colonized particles, derived from in situ experiments using the C-RESPIRE (that is, Carbon RESPIRE rather than TM-RESPIRE20) dual particle interceptor and incubator9 (Extended Data Fig. 2) to explore controls on MR magnitude with depth and across regions, along with its percentage contribution to flux attenuation in the upper mesopelagic (that is, approximately 100–300 m depth), where POC flux is highly attenuated1.

Deconstructing the Martin curve

The drivers of the Martin curve were decoded by assessing the contribution of MR to POC flux attenuation. The fate of particles intercepted by C-RESPIRE differs fundamentally from sediment traps. Conventionally, with traps, sinking POC is attenuated by MR, zooplankton flux-feeding and diel vertical migration (DVM) until particles are intercepted (Fig. 1a) and instantly preserved. This method, over multiple depths, enables estimation of b values but offers few insights into the relative importance of MR, flux-feeding or DVM in setting b. In contrast, C-RESPIRE intercepts settling particles that have undergone the joint influence of these processes in overlying waters, then incubates them (Fig. 1a). Particle interception arrests the settling particles (see later) but their capture is required to obtain enough particles to measure MR. Such interception, in situ, is less intrusive than previous attempts to collect, subsample and/or isolate particles from the sinking assemblage to assess the importance of MR in lab-based21,22 and in situ23 experiments. C-RESPIRE isolates particles within minutes from zooplankton-mediated transformations (Extended Data Fig. 2), and thus in a subsequent multi-day incubation provides only particle transformations driven by MR, based on optode-based microbial respiration rates and a mesopelagic respiratory quotient (RQ; see Methods for caveats). By also measuring POC not transformed by microbes during the incubation (termed the residual POC flux), C-RESPIRE indirectly estimates the cumulative POC flux (that is, MR + residual POC flux), like that measured in sediment traps (Fig. 1b). This residual flux also yields indirect insights into the relative importance of zooplankton-mediated processes (that is, zooplankton were excluded from C-RESPIRE by design) in particle flux attenuation (that is, flux-feeding + DVM).

Fig. 1: Mesopelagic transformations of downward POC flux at multiple depths investigated across contrasting regions.
figure 1

a, Schematic of the cumulative transformations of settling particles (denoted by solid vertical bars) due to zooplankton flux-feeding (FF), DVM and MR before particle interception by C-RESPIRE during the initial collection phase at each of three depths (Extended Data Fig. 2). MR (in blue) represents the subsequent incubation phase of C-RESPIRE in which only MR acts on the intercepted particles to decrease POC. b, Deconstruction of the main drivers of the POC flux attenuation. MR (blue areas) is as described in a and is inferred from measured O2 consumption and a fixed RQ (Methods). Dissolved organic C accumulation rates during incubation were low (representing on average 21 ± 16% of MR; Extended Data Table 4), supporting a close coupling between solubilization and MR. The residual POC flux (open circles) corresponds to the (intercepted) POC measured at the end of the multi-day in situ incubation. Cumulative POC flux (filled circles) is reconstructed using the sum of the residual POC and MR (that is, residual POC flux + MR) and should reflect a Martin curve, represented by the solid black line (Methods). c, Locations of C-RESPIRE deployments overlaid on a map of satellite-derived net primary productivity (NPP) climatology (2003–2018) (obtained from the NASA Ocean Color website and the CAFE algorithm). Green, SG; brown, BEN; red, SAZ; orange, PAPA; blue, MED; purple, SPSG.

C-RESPIREs were deployed across regions from highly productive (South Georgia (SG) and Benguela (BEN)) to low-productivity, high nutrient–low chlorophyll (HNLC; Subantarctic Zone (SAZ) and northeast subarctic Pacific (PAPA)) and oligotrophic (Mediterranean (MED) and South Pacific Subtropical Gyre (SPSG)) regimes (Fig. 1c). Some regions had different mesopelagic temperatures, O2 concentrations (Extended Data Table 2) and phytoplankton assemblages (Extended Data Table 3). These environmental characteristics have been identified by others as potential drivers of mesopelagic POC flux attenuation (Extended Data Table 1). The diversity of both productivity regimes and phytoplankton assemblages reflects a 30-fold range in POC flux (Fig. 2c). Over the approximately 200-m depth range sampled, cumulative POC flux followed the canonical Martin curve at 5 of the 6 sites (POC flux increased with depth at SAZ). SAZ has a ‘composite’ water column, a HNLC subpolar layer overlying subtropical mesopelagic waters, which may explain the anomalous pattern24. The cumulative POC fluxes reconstructed from C-RESPIRE are Martin curves (Fig. 2c) comparable to those measured using sediment traps6, providing confidence in the validity of our reconstruction approach.

Fig. 2: Deconstructing POC flux attenuation.
figure 2

ac, Microbially mediated component of the POC flux attenuation, MR (a), the residual POC flux (b) and the cumulative POC flux (c) as a function of depth at the six study sites investigated (same colour code as for Fig. 1c). Note the different flux scales for SG and BEN and for all sites in a. Owing to ship logistics, incubation times varied by threefold between sites (Extended Data Table 2). The coloured curves correspond to the best fit of a power-law model (not fitted for MED and SPSG where only two depths below the PPZ were sampled), and error bars were derived using uncertainty-propagation laws (Methods). These Martin curves are presented only to demonstrate that the cumulative POC flux approach reproduces a power-law attenuation trend. A fixed RQ of 0.70 (that is, for plankton-based particles such as marine snow45) was used to calculate MR. The coloured-shaded rectangles in a represent the range of values of MR obtained based on the upper (that is, 0.89 for phytoplankton cells44) and lower (that is, 0.50 for methane oxidation within particles, the theoretical minimum for RQ46) bounds of the range of RQ values used in the sensitivity analysis (Methods). The beige-shaded areas represent the PPZ where new particles are produced.

The next step in deconstructing the Martin curve is to intercompare vertical trends in MR with the residual POC flux. To do so, we assume constant sinking velocities over the approximately 200-m depth range sampled (Methods). At most sites, there is no evidence of a Martin-type relationship for MR (Fig. 2a), and across all sites the C requirement of particle-attached microbes should readily be met as the median contribution of MR to the cumulative POC flux attenuation ranges from 7% to 29% (Methods), and importantly never exceeds 35% regardless of using bounds of 0.89–0.50 for RQs for unmodified to heavily modified particles (Fig. 2 caption and Extended Data Fig. 3a). Although this trend in MR is consistent with the conclusions from lab-based and field studies21,22,23,25, such wide-ranging MRs, across six provinces, has not been previously observed. Within our sites, the magnitude of MR changed little with depth. Thus, even though C-RESPIRE halts particle sinking (Extended Data Fig. 2) to conduct multi-day incubations (during which particles, if not intercepted, could sink a further approximately 100–200 m), C-RESPIRE provides a representative estimate of MR across the upper mesopelagic (Methods). In contrast, this MR contribution ranged fourfold across provinces, probably reflecting regionally different patterns in particle flux composition and planktonic assemblages4 (see later).

The residual POC flux, probably representing POC inaccessible to microbes, follows a power-law decrease at most sites and dominates the cumulative flux (80 ± 13%); thus, it largely reflects trends for cumulative POC flux attenuation (Fig. 2b). As flux-feeding and DVM are the other main drivers of particle flux attenuation, vertical trends in the residual POC flux are probably driven by zooplankton-mediated processes in waters overlying each C-RESPIRE (Fig. 1a), assuming that particle-attached microbes within C-RESPIRE had remineralized all available POC by the end of multi-day incubations (Extended Data Fig. 1 and Methods). The Martin-like shape of the residual POC flux probably derives from pronounced vertical distributions of resident26 and vertically migrating27,28 flux-feeders in the epipelagic that modulate particle transformations. In summary, the different-shaped profiles of MR-mediated (approximately constant), residual and cumulative POC fluxes (power-law decreases; Fig. 2) indicates that the relative influence of MR on the cumulative POC flux will increase with depth assuming constant sinking velocities (Extended Data Fig. 4).

Patterns in C-specific remineralization

To explore changes in the influence of MR on POC flux attenuation (Fig. 2), we normalized MR to incubation time and cumulative POC flux (Methods). The derived C-specific remineralization rate (Cremin) ranged from 0.02 d−1 to 0.43 d−1 across sites and depths (Fig. 3a). The upper bound (0.43 d−1) is several-fold higher than that in previous studies21,23, which subsampled the settling particle assemblage. Our range, when reformulated as Cremin, demonstrates that the particle flux supplied enough C (Extended Data Fig. 1 caption) to sustain particle-attached microbial needs in the upper mesopelagic, an unresolved issue highlighted previously4,29 (but see ref. 30). This 20-fold range in Cremin may be controlled by both site-specific and depth-dependent factors. Comparison of Cremin across sites reveals no clear relationship with the magnitude of cumulative POC flux (based on the shallowest C-RESPIRE). Such a relationship is probably confounded by differing biogeochemical characteristics across the sites including: depth of the primary production zone (PPZ); proportion of large diatoms; lability of particles (spring bloom (SG) versus highly transformed (SPSG)); biogenic (SAZ) versus lithogenic (MED); and resulting seasonality of particle properties. These multi-faceted characteristics preclude robust statistical analysis, for example, large diatoms contributed to the setting POC flux at SG (Extended Data Table 3) where high fluxes were little attenuated by microbes (Fig. 2). In contrast, changes in Cremin with depth reveal more distinct trends.

Fig. 3: Site-specific patterns of C-specific MR.
figure 3

MR was normalized to the cumulative POC flux and incubation time. a,b, Cremin as a function of depth (a) and ambient temperature (b)31 at each study site shown in Fig. 1c. The median Cremin rate ranged from 0.08 d−1 (SG) to 0.23 d−1 (SPSG). The metric Cremin enables inspection of the expected relationship (based on Fig. 2) between MR-mediated and cumulative POC fluxes—an increase in Cremin with depth (Extended Data Fig. 4). The coloured-shaded rectangles represent the range of values of Cremin obtained based on the upper and lower bounds of the range of RQ values used in the sensitivity analysis, as presented in Fig. 2. The beige-shaded areas in a represent the PPZ, as presented in Fig. 2. The dashed lines represent the best fit of the linear model. Error bars were derived using uncertainty-propagation laws (Methods).

Vertical trends in Cremin differ between regions (Fig. 3a). At SG, SAZ, PAPA and MED, Cremin increased with depth, as expected (Extended Data Fig. 4) considering the approximately constant MR-mediated versus decreasing cumulative POC fluxes (Fig. 2). However, decreasing Cremin with depth was observed at sites with larger vertical temperature gradients (BEN (R2 = 0.29) and SPSG (R2 = 0.43)). When plotted against temperature31 (Fig. 3b), depth-dependent Cremin patterns generally become more conspicuous, suggesting a demarcation in these distinctive trends at sites with vertical gradients of more than 2 °C (R2 = 0.80 and R2 = 0.39 at BEN and SPSG, respectively).

Modes of control on C remin

Mesopelagic temperatures may influence the marked transition in the Cremin versus depth relationship between sites (Fig. 3). Temperature is a key driver of microbial community composition32, metabolic activity33 and organic matter decomposition34,35. The latter two properties are reflected by thermal performance curves (that is, growth or respiration versus temperature36,37,38) of isolates of marine microbes. Temperature has been proposed as the driver of particle flux attenuation in the palaeo-oceanographic39 and contemporary7 oceans. Temperature may control Cremin at subtropical sites (BEN and SPSG; Fig. 3b) with pronounced mesopelagic gradients (Extended Data Fig. 5). A similar Cremin pattern was reported at Bermuda (subtropical Atlantic), using a C-RESPIRE prototype40. This raises the question of whether temperature also exerts control on Cremin at mid- and high-latitude sites where MR is relatively invariant with depth (excluding outliers at BEN and MED; Fig. 2a) and Cremin increases (Fig. 3).

Mid- and high-latitude sites have upper mesopelagic gradients of less than 2 °C (Fig. 3b). Although no thermal performance curves exist for mesopelagic microbes (that is, under in situ conditions), this physiology versus temperature relationship36,37,38 helps explore the implications of different temperature gradients on Cremin. Each temperature–Cremin relationship (Fig. 3b) can be recast using the characteristic shape of a thermal performance curve (Fig. 4a). Although it is not known where on a curve these temperature–Cremin data would lie, a marked change in Cremin with temperature (straddling slope and plateau) is more likely for a 5 °C gradient than a less than 2 °C gradient (Fig. 4a). Thus, at mid- and high-latitude sites, the influence of temperature on Cremin may be diminished, enabling additional factors to set depth-dependent patterns in Cremin. Next, we examine the regional characteristics at these mid- and high-latitude sites that may define the functioning of their biological carbon pumps.

Fig. 4: Putative evidence for multiple factors controlling distinctive vertical patterns in Cremin.
figure 4

a, Particle-attached microbial communities may encounter large (SPSG site, purple envelope) versus small (MED site, blue envelope) changes in temperature, as they settle through the water column, differentially altering their physiological performance, as projected by the thermal performance curve relationship, such as in MR. b, Sinking particles are characterized by a continuum in biochemical lability (that is, altering substrates) partially controlling the RQ41. c, The thermal sensitivity of soil BGE (analogous to bacterial carbon-use efficiency) is both strain dependent and substrate dependent49,50 as represented by the dashed arrow linking b and c. d, In the upper mesopelagic, microzooplankton and mesozooplankton (and also physical disaggregation) fragment large aggregates into smaller slower-sinking particles with large surface area:volume ratios55,56 that may promote microbial colonization and enhanced MR.

Drivers of microbial remineralization

To explore the reasons for the two distinct trends in Cremin (Fig. 3), there is insufficient knowledge of the drivers of MR based on the current parameterization of marine models of MR17. We draw on advances from soil microbiology (joint role of bacterial strain and biochemical substrate on physiology), marine pelagic physiology (thermal performance curves) and mesopelagic microbial ecology (altered community structure with depth) to explore potential explanations.

At higher-latitude sites with small temperature gradients, depth-dependent increases in Cremin (Fig. 3) may be disproportionately influenced by vertical gradients in particle characteristics (for example, substrate availability), ecological shifts (community) and/or physiological changes (exo-enzymes to match substrate changes) of particle-attached microbes41. For example, the joint and complex influence of temperature and substrate on marine bacterial physiological performance is long established42,43. There is indirect evidence of depth-dependent changes in RQ based on a comparison of quotients from unmodified directly sinking phytoplankton cells (RQ = 0.89 (ref. 44)), marine snow (RQ = 0.70 (ref. 45)) to heavily modified particles (RQ = 0.50 (ref. 46)) that suggest that the number of particle transformations as they sink results in altered particle substrate quality with depth (Fig. 4b).

Microbial soil studies show that enzymatic decomposition of biochemically recalcitrant organic matter requires high activation energy, and is more sensitive to changes in temperature than decomposition of more labile organic matter47,48. Soil microbiology studies suggest that the thermal sensitivity of the bacterial growth efficiency (BGE; analogous to carbon-use efficiency) is both substrate dependent49 (Fig. 4b) and strain dependent49,50 (Fig. 4c). For soil microbes, across multiple substrate and strain permutations49,50, there are combinations in which the BGE versus temperature relationship is positive, that is, BGE decreases (for example, respiration increases more than growth) at lower temperatures. If this relationship holds for oceanic regions, this would increase MR with depth (Fig. 4c). These scenarios are consistent with increasing Cremin with depth at some higher-latitude sites (Fig. 3b) and offer the most compelling explanation currently available. In addition to the likelihood of depth dependency of substrate quality41, there is evidence of changes in microbial community structure and function with depth50,51 that together could provide such substrate and strain combinations49,50 leading to a constancy in MR (Fig. 2a) in contrast to the power-law decrease in cumulative POC flux, and hence higher Cremin at depth (Fig. 4c).

Potential linkages between depth-dependent changes in substrates and microbial community composition are evident from recent mesopelagic studies. Pronounced mesopelagic physico-chemical gradients influence particle-attached microbes and are linked to changes in the composition and function of microbes in the mesopelagic51,52. Others have shown that microscale community ecology of particle-attached bacteria can partially control carbon turnover rates53,54. Bacterial community structure and function (and hence BGEs) may also be influenced by depth-dependent changes in particle characteristics, including sinking rates and surface area:volume ratios driven by mesopelagic particle fragmentation55. Fragmentation has been linked to the concept of microbial gardening by microzooplankton56 and mesozooplankton57 with the exposure of fragmented ‘fresh’ surfaces enabling microbial colonization and potential changes to MR (Fig. 4d). In the absence of a dominant driver such as temperature, at higher-latitude sites (Fig. 3b), we suggest that the most likely additional processes to influence increases in Cremin are microbial community structure and function and their interplay with substrates, and potentially the small changes in temperature.

Our main finding of distinct trends in Cremin versus depth (increasing versus decreasing), across sites representing wide-ranging environmental, planktonic assemblages and productivity regimes (Fig. 1c), differs from studies proposing temperature as the dominant driver of POC flux attenuation (that is, by MR, flux-feeding and DVM; Fig. 1a) globally7,39 (Extended Data Table 1). A previous study39 predominately used sediment cores from low-latitude sites, probably influencing their conclusion on the role of temperature on POC flux attenuation. In another study7, sites were either high latitude or low latitude with no intermediates, and can be clustered into high- versus low-latitude trends such that at high latitudes there is no relationship between b values and temperature (their Fig. 2a). Thus, their findings7 may not differ markedly from our two distinct trends. Globally, there is a high/mid- to low-latitude transition from small to large mesopelagic temperature gradients (Extended Data Fig. 5)—as reported here—with regions characterized by small temperature gradients comprising two-thirds of the ocean. Model projections suggest that these mesopelagic trends are unlikely to change in the future58,59, pointing to the continued importance of factors other than temperature in driving POC flux attenuation across mid- and high-latitude regions.

Conclusions

Our study reveals that when POC supply (intercepted sinking particles) and MR (oxygen optode time series and an RQ) are measured concurrently, there is a surfeit of C to sustain the needs of particle-attached microbes. The contribution of MR to the cumulative POC flux attenuation in the upper mesopelagic is less than 30% at all sites, indicative of a lesser but albeit influential role. This trend is consistent with the relative invariance of MR versus depth profiles, which differ from the Martin curve for cumulative POC fluxes. The dominant influence of zooplankton in shaping this curve, that we infer from indirect evidence, probably derives from the strong epipelagic influence of flux-feeders in modulating particle transformations55,57. However, MR may increase in relative importance as an attenuation pathway with depth due to the flat shape of the MR profile relative to that of the rapidly attenuating Martin curve26,60.

Temperature may set Cremin but only at low-latitude sites with high vertical temperature gradients, precluding the use of a global relationship to parameterize microbially mediated POC flux attenuation. At higher-latitude sites, multiple causality is likely and there may be important lessons from soil microbiology on concepts such as BGE that link microbial ecology and particle biochemistry, and from lab-based particle experimental approaches61. This deconstruction of the Martin curve reveals a diverse range of regional biological carbon pumps across the ocean in which microbes have differing influences. This diversity requires more complex representation of POC flux attenuation in biogeochemical models that demarcate the roles of zooplankton flux-feeders and particle-attached microbes. This decoding of the biological carbon pump will provide avenues of enquiry for microbial modellers and experimentalists.

Methods

Site characteristics

High-productivity regions

Regions downstream of South Georgia (SG) and in the Benguela Current offshore of Namibia (BEN) were sampled using C-RESPIRE within the framework of the UK COMICS (Controls over Ocean Mesopelagic Interior Carbon Storage) project62 during research cruises DY086 and DY090 onboard the RSS Discovery. At SG (52.7° S, 40.2° W), 6 individual C-RESPIRE incubations to provide measurements of the microbially mediated component of POC flux attenuation (MR) were performed during November to December 2017. These naturally iron-fertilized waters downstream of SG are characterized by an intense phytoplankton bloom dominated by large diatoms63, sustaining high downwards C export fluxes64. Relatively low in situ temperatures, ranging from 0.8 °C to 1.8 °C, were observed at the SG site (Extended Data Table 2). In the BEN region, stations BS (21.6° S, 9.4° E) and BN (18.2° S, 11.0° E) were sampled in May to June 2018 to provide 7 individual C-RESPIRE measurements of MR. Intense coastal upwelling occurs in this locale, supplying nutrients that support high pelagic biological productivity and downwards C export. High rates of mesopelagic respiration are responsible for the formation of a subsurface low-oxygen zone65. During COMICS 2, intermittent low-oxygen events (25–60 µmol kg−1) were observed between about 100-m and 440-m depth, with a persistent low-oxygen band between 300 m and 400 m (ref. 66), and C-RESPIRE incubations were performed at dissolved oxygen concentrations as low as 45 µmol kg−1 at 235-m depth at the BN station (Extended Data Table 2).

HNLC regions

The SAZ site (46.9° S, 142.2° E) is located in the Subantarctic Zone southwest of Tasmania. It corresponds to the location of the Australian Southern Ocean Time Series (SOTS) site67, a long-term particle flux time-series observatory operating at 1,000-m, 2,000-m and 3,800-m depth68. This locale often has both low dissolved silicate and iron concentrations over the annual cycle, along with a phytoplankton community dominated by haptophytes, flagellates and small diatoms69. Deployments of C-RESPIRE were performed from the R/V Investigator in March (2019), that is, about 2–3 months after the peak of phytoplankton productivity (December) and near the end of the downwards export period. Ocean Station Papa (PAPA) in the northeast subarctic Pacific was sampled during the US EXPORTS NP campaign onboard the R/V Roger Revelle in late summer (August to September) 201870,71. Owing to low dissolved iron supply, this HNLC area is characterized by a modest primary productivity signature (March to April) dominated by diatoms and coccolithophores72, that is reflected by low–medium fluxes of the biological carbon pump73. For technical reasons, only data obtained during the last deployment (Epoch 3) of the EXPORTS NP voyage are available.

Oligotrophic regions

MR at two sites within the Mediterranean Sea (MED) were obtained using the C-RESPIRE deployed during the French PEACETIME cruise74 (https://doi.org/10.17600/17000300) onboard the R/V Pourquoi Pas? in May to June 2017. This enclosed basin is characterized by a west-to-east gradient of increasing oligotrophy, and by intense atmospheric deposition of desert dust from the Sahara75. C-RESPIREs were deployed in the western Algerian basin (37.9° N, 2.9° E) and Ionian Sea (35.5° N, 19.8° E). The South Pacific Subtropical Gyre (SPSG) is characterized by the most oligotrophic waters of the global ocean76. The edge of the gyre (20.4° S, 166.5° W) has a subsurface chlorophyll maxima at about 150-m depth, and was sampled for MR in November 2019 onboard the R/V L’Atalante during the TONGA cruise77 (https://doi.org/10.17600/18000884).

The C-RESPIRE dual particle interceptor and incubator

Principles of operation and sampling microbial communities

The C-RESPIRE particle interceptor and incubator is designed to non-intrusively intercept sinking particles, concentrate them within a 1.45 l volume of (unpoisoned) filtered seawater, and then incubate them at in situ pressure and temperature (Extended Data Fig. 2). Reference 9 provides a detailed discussion on the potential artefacts associated with the C-RESPIRE method, such as the implications of halting sinking particles, particles sitting on the base plate of the instrument during incubations potentially reducing degradation rates61, and a demonstration of the minor contribution of free-living bacteria to microbial respiration rates. C-RESPIRE consists of a titanium cylinder with a collection area of 0.0124 m2. At its entrance, a baffle limits the turbulence and resuspension of the sinking particles that ultimately settle into an indented rotating sphere (IRS). Pre-programmed rotations (typically every 5–10 min) of the IRS transfer particles into an inner chamber (1.45 l) equipped with an oxygen optode. The IRS ensures the exclusion of mesozooplankton and isolation of the inner chamber when not rotating (Extended Data Fig. 2). The roughly 1- to 3-day particle collection phase (Extended Data Table 2) promotes the accumulation of sinking particles in the chamber, along with their attached microbial consortia, including heterotrophic bacteria and microzooplankton78,79. This accumulation of sinking particles ensures that the remineralization signature most closely corresponds to that of the attached microbial assemblage, as opposed to free-living heterotrophic bacteria potentially present in the <0.2-µm filtered seawater initially used to fill the inner chamber.

Deployment of the free-drifting C-RESPIRE mooring array

C-RESPIREs were deployed below the PPZ depth (except at the SPSG site where one C-RESPIRE was deployed within the PPZ; Fig. 2), that is, the depth where the in vivo chlorophyll fluorescence drops to 10% of its maximum value80. Before deployment, the C-RESPIREs were gently filled with <0.2-µm filtered seawater collected at each deployment depth. A lagged delay in particle interception of about 3 h was pre-programmed to avoid collection of sinking particles above the programmed depth. Two or three C-RESPIREs were deployed on a free-drifting surface-tethered array comprising wave-damping floats and/or 25-m-long bungee cord to reduce the vertical motion of the mooring line. Temperature and pressure sensors (RBR) were used to confirm the deployment depth and assess the vertical motions of the mooring during the deployment. Deployment duration ranged from about 2 days to 5 days and was adapted to weather conditions, ship logistics and other activities onboard.

Sampling procedure and data collection

Collection (TCOLL) and incubation (TINCUB) times were chosen as a function of the deployment duration and the estimated particle flux magnitude in the region at the time of the year of each voyage (Extended Data Table 2). After recovery of the array of two or three C-RESPIREs, seawater above the IRS was siphoned off and the inner chamber was sampled from each C-RESPIRE. Samples were stored in the dark at 4 °C until further processing. These steps were performed within <1 h of recovery of the C-RESPIREs. The absence of mesozooplankton in the inner chamber was confirmed by visual inspection. The 1.45-l sample was split into 8 equal fractions using a custom particle splitter similar to that reported in ref. 81. POC-designated subsamples were filtered through pre-combusted 13-mm- or 25-mm-diameter GF/F or QMA filters, and stored at −20 °C until processing and analysis. POC samples were acidified overnight with 2 M HCl to remove inorganic C, and then dried at 60 °C for 2 days. Samples were then analysed on a CHN analyser (Thermo Finnigan EA 1112 Series Flash Elemental Analyser).

Remineralization rate

Oxygen time series

The inner chamber of each C-RESPIRE was equipped with an oxygen optode (Aanderaa 3830 or 4831 series). Dissolved oxygen concentrations were corrected for pressure, salinity and temperature according to the Aanderaa Operating Manual and ref. 82. Optodes were post-calibrated at the Mediterranean Institute of Oceanography (Marseille, France) or deployed on a CTD (conductivity, temperature and depth) instrument to cross-compare with dissolved oxygen concentrations obtained with the Winkler method (O2 range = 230–330 µmol kg−1, slope = 0.94 ± 0.01, R2 = 0.98, P < 0.001). The total amount of dissolved oxygen consumed during incubation (ΔO2, mmolO2) was obtained by using the equation:

$${\Delta {\rm{O}}}_{2}=({{\rm{O}}}_{2{\rm{PRE}}\text{-}{\rm{INCUB}}}-{{\rm{O}}}_{2{\rm{POST}}\text{-}{\rm{INCUB}}})\times 1.45$$

where O2 PRE-INCUB and O2 POST-INCUB are dissolved oxygen concentrations (mmolO2 l−1) measured at the start and end of the incubation, respectively, and 1.45 is the volume of the inner chamber (l). Care was taken to attain dissolved oxygen concentrations close to ambient concentrations before the start of the incubation phase. However, it is noted that owing to the strong variability in dissolved oxygen concentrations in the Benguela area66, for two C-RESPIRE measurements, pre-incubation dissolved oxygen concentrations diverge by more than 15% from ambient dissolved oxygen concentrations before the incubation with CTD cast(s) (BN31 deployment; Extended Data Table 2).

Respiratory quotient

A respiratory quotient (RQ = ΔDIC/–ΔO2, where DIC is dissolved inorganic carbon) is used to convert an oxygen-based to a carbon-based remineralization rate (see below). RQ varies depending on the stoichiometry and oxidation state of the substrate, and the metabolic pathway used83. As a consequence, a large range of RQs has been reported from incubations and field measurements84. Assuming complete oxidation, the average composition of phytoplankton cells results in an RQ of 0.89 (ref. 44). The lower bound of RQs is the theoretical minimum of 0.50 for methane oxidation46. This range of published values of RQs for different particle types straddle individual phytoplankton cells (that is, no particle transformations) to heavily transformed particles. Within these bounds, ref. 45 reported from two independent methods a global estimate of RQ in the surface ocean of about 0.70 with small spatial variability. Over longer time periods, a Corg:O2 conversion factor of 0.69 (that is, 117/170) representative of the stoichiometric ratio of particle remineralization has been obtained, for the mesopelagic realm deeper than 400 m, across the South Atlantic, Indian and Pacific basin85. These RQ values are broadly representative of plankton-based particles such as marine snow and mesopelagic conditions (that is, differing substrate quality, microbial communities). Consequently, we selected an RQ of 0.70—reported by refs. 45,85 as an interim value—to convert the oxygen-based remineralization rate to C.

In addition, we measured microbially mediated RQ values during the SOLACE voyage (Southern Ocean Large Area Carbon Export; 4 December 2020 to 15 January 2021) at three study sites straddling an approximately 9 °C mesopelagic temperature range: one subantarctic site (SAZ), and two sites south of the Polar Front (PF1 and PF2; Extended Data Table 5). At these sites, surface-tethered particle interceptor traps (PITs) were deployed for 3 days at 3 depths between about 180 m and 280 m. PIT tubes containing no preservative were used for intercepting ‘fresh’ sinking particles. After recovery of the PITs, tubes were transferred to a temperature-control lab (set to ambient mixed-layer surface ocean temperature at each site). The overlying waters were gently siphoned off and particles were collected from each tube and were immediately placed in 4-ml glass sample vials (Merck, Z256064), each containing PreSens O2 (SF-RPSu4) and CO2 (SF-CD2R) sensors and ambient (filtered) seawater from each depth. Calibration of the sensors was conducted in accordance with the manufacturer’s instructions. The O2 sensors were standardized using a two-point calibration (0%, 100% saturated air) and the CO2 sensors were calibrated using a 6-point calibration with specific gas mixes (0%, 0.04%, 0.06%, 0.08%, 1%, 5%; BOC Linde gas). Glass vials containing particles and seawater were placed in individual wells of a 24-well flat-bottom plate (Corning CLS3527). Each plate contained a total of 16 samples from unpreserved PITs, and 4 negative control vials (filtered seawater collected from the same site and depth). The sample vials were incubated for 12 h at in situ temperature, with readings taken for O2 and CO2 concentrations using the VisiSens TD (PreSens). The readings were corrected for site and depth temperature and salinity following the manufacturer’s instructions. Subsequently, all samples were processed for preservation using 0.5% glutaraldehyde for further analysis and microscopy. If zooplankton were found within a sample, that sample was removed from the subsequent analysis, focusing only on mesopelagic microbes. Sample images were processed using PreSens IDL evaluation software. Images obtained over a 12-h period were used to determine particle RQs. The amount of O2 consumed (molO2) and CO2 produced (molCO2) during incubation was obtained using the following equation:

$${\Delta {\rm{O}}}_{2}=(({{\rm{O}}}_{2{\rm{PRE}}\text{-}{\rm{INCUB}}}-{{\rm{O}}}_{2{\rm{POST}}\text{-}{\rm{INCUB}}})-({{\rm{O}}}_{2{\rm{CONTROL}}{\rm{PRE}}\text{-}{\rm{INCUB}}}-{{\rm{O}}}_{2{\rm{CONTROL}}{\rm{POST}}\text{-}{\rm{INCUB}}}))$$
$${\Delta {\rm{CO}}}_{2}=(({{\rm{CO}}}_{2{\rm{PRE}}\text{-}{\rm{INCUB}}}-{{\rm{CO}}}_{2{\rm{POST}}\text{-}{\rm{INCUB}}})-({{\rm{CO}}}_{2{\rm{CONTROL}}{\rm{PRE}}\text{-}{\rm{INCUB}}}-{{\rm{CO}}}_{2{\rm{CONTROL}}{\rm{POST}}\text{-}{\rm{INCUB}}}))$$

At the three sites and depths, the total alkalinity versus salinity relationship was used to calculate the total alkalinity from the Global Ocean Data Analysis Project (GLODAP) hydrographic section data86. This dataset was also used to estimate the change in the partial pressure of CO2 (\({p}_{{{\rm{CO}}}_{2}}\)). For each salinity, temperature and \({p}_{{{\rm{CO}}}_{2}}\) value at each depth across the three sites, a Matlab routine was used to calculate a Revelle factor (R)—the incremental change in \({p}_{{{\rm{CO}}}_{2}}\) divided by the incremental change in DIC, so \(R=[\Delta {p}_{{{\rm{C}}{\rm{O}}}_{2}}/{p}_{{{\rm{C}}{\rm{O}}}_{2}}]\div[\Delta {\rm{D}}{\rm{I}}{\rm{C}}/{\rm{D}}{\rm{I}}{\rm{C}}]\). R was calculated from the total alkalinity values (salinity), \({p}_{{{\rm{CO}}}_{2}}\) comes from a typical value in the field near the site, and DIC is calculated with \({p}_{{{\rm{CO}}}_{2}}\) and the total alkalinity in CO2SYS routine (CO2SYSv3 for MATLAB). The values obtained were used to calculate the apparent RQ of each sample vial, as follows:

$${\rm{aRQ}}=\Delta {\rm{DIC}}/{-\Delta {\rm{O}}}_{2}$$

At the shallowest depth, aRQ values ranged from 0.77 to 0.88 (Extended Data Table 5), consistent with the RQ values of 0.70 from global analyses45,85 and 0.89 for a substrate similar to the average composition of planktonic material44. These aRQ values generally decreased with depth but never reached the theoretical minimum of 0.5 (for methane oxidation46) except for one outlier at the deepest depth at the PF1 site. The release of DOC during remineralization (Extended Data Table 4) and the potential for subsequent partial oxidation of this DOC pool by microbes (leading to a decrease in O2 but not a corresponding increase in CO2 or DIC) could potentially lead to RQs <0.5. These depth-dependent changes in aRQ can be linked to the remineralization of labile particles by microbes and the biochemical sequence of degradation that occurs as particles sink, becoming less labile and more recalcitrant. It is noted that apparent RQs may change depending on incubation time (that is, decrease further during longer incubations; F.K., unpublished data), which may be linked to the solubilization of DOC and its eventual use as a microbial substrate.

RQ sensitivity analysis

To test how the choice of an RQ value will influence the outcome of the results presented in Figs. 2 and 3, we conducted a sensitivity analysis using a range of fixed and depth-varying RQ values. Fixed RQ values were chosen to reflect sinking particles that have been degraded by multiple particle transformations as they settle through the mesopelagic realm: an upper bound of 0.89 (for phytoplankton cells44), a mid-point of 0.70 (for plankton-based particles such as marine snow45) and a lower bound of 0.50 (that is, the theoretical minimum for methane46 formed within marine particles such as faecal pellets87).

Depth dependency for RQ was parameterized using our mesopelagic aRQ values measured during the 2020–2021 SOLACE voyage (Extended Data Table 5). To take into account differences in C-RESPIRE sampling depths relative to the base of the PPZ (zPPZ) between sites (Fig. 2 and Extended Data Table 5), sampling depths were standardized to zPPZ. On the basis of the aRQ values measured during SOLACE and assuming an RQ value of 0.89 (ref. 44) at zPPZ (and above), the following relationship was obtained:

$${{\rm{RQ}}}_{Z}=-0.21\,{10}^{-2}\times (z-{z}_{{\rm{PPZ}}})+0.89$$

Applying this relationship to our dataset (that is, z – zPPZ between 0 m and 181 m), we obtained a depth-varying RQ (RQZ where z represents the depth) ranging between 0.51 and 0.89 (Extended Data Fig. 3e) with an average value of 0.73 ± 0.12. The contribution of MR to the cumulative POC flux attenuation and vertical profiles in Cremin obtained using the tested range of fixed and depth-varying RQ values are presented in Extended Data Fig. 3. This sensitivity analysis reveals that the contribution of MR to the cumulative POC flux attenuation (Extended Data Fig. 3a) and the depth-dependent increases in Cremin (Extended Data Fig. 3b–e) are largely insensitive to the choice of an RQ, such that the influence of MR never exceeds 35%. This insensitivity is mainly due to the dominance of the residual POC flux over MR at most sites.

Deconstruction of the POC flux

Here we expand on how the POC flux was teased apart into three components: MR, the residual flux (FRESIDUAL) and the cumulative flux (FCUMUL). Conceptually, settling particles undergo continuous transformation from attached microbes and discrete transformations by zooplankton15. Upon interception by C-RESPIRE, particles are predominantly transformed by attached bacteria. However, if these sinking particles had not been halted by the C-RESPIRE, they would have continued to have been modified by both zooplankton and particle-attached microbes. By measuring MR in C-RESPIRE, it is possible to first derive an estimate of the POC that would have been available to zooplankton to transform (termed FRESIDUAL) had the particles continued to sink. Next, by combining MR and FRESIDUAL we can estimate FCUMUL. Below we outline both the equations used and also several of the assumptions that have to be made for this deconstruction.

Microbial contribution

The microbially mediated component of the POC flux attenuation (MR, mmolC m−2 d−1), was calculated from ΔO2 and a fixed RQ of 0.70 (see previous section for justification):

$${\rm{MR}}={\Delta {\rm{O}}}_{2}\times {\rm{RQ}}/0.0124/{T}_{{\rm{COLL}}},$$

where 0.0124 is the collection area (m2) and TCOLL is the collection time (d) of the C-RESPIRE. We acknowledge that MR (and hence FRESIDUAL) is partly dependent on the incubation time, which varied between sites due to ship logistical constraints (Extended Data Table 2). The O2 consumption rate was typically constant over the first approximately 30–48 h of incubations and then the rate often decreased, indicative of the onset of substrate limitation (Extended Data Fig. 1). Hence, our estimates of MR will probably reflect the contribution of MR to the cumulative POC flux attenuation (that is, MR/FCUMUL) at the deployment depth. However, this interpretation—of the onset of substrate limitation—must be reconciled with the lack of change in MR observed with depth (Fig. 2a); substrate limitation at a shallower depth should result in a decrease in MR at deeper depths unless there is a change in the particle-attached microbial assemblage that results in the utilization of other substrates19.

On the basis of the assumption that MR becomes substrate limited after more than 30 h of incubation (Extended Data Fig. 1), MR could have been underestimated (and hence FRESIDUAL slightly overestimated) by about 25% for the 1-day incubations (such as at BEN), but as this underestimation would apply to measurements at each depth the overall trends will not alter. In addition, the dominance of FRESIDUAL over MR at most sites (Fig. 2) means that FRESIDUAL would have been overestimated by a much smaller percentage, and hence this trend would not alter despite the uncertainties regarding shorter incubations in relation to substrate limitation.

Residual POC flux

This flux (FRESIDUAL, mmolC m−2 d−1) was obtained using the following equation:

$${F}_{{\rm{RESIDUAL}}}={\left[{\rm{POC}}\right]}_{{\rm{POST}}\text{-}{\rm{INCUB}}}\times 1.45/0.0124/{T}_{{\rm{COLL}}},$$

where [POC]POST-INCUB is the concentration of POC remaining at the end of the multi-day incubation (mmol l−1) and 1.45 is the volume of the incubation chamber (l). FRESIDUAL is interpreted here as the POC that is not available to the microbes due to the onset of substrate limitation (Extended Data Fig. 1). However, on shorter incubations, such as at BEN, it is possible that a small proportion of this residual POC could be available for particle-attached microbes.

Cumulative POC flux

The cumulative POC flux (FCUMUL, mmolC m−2 d−1) is reconstructed using the sum of FRESIDUAL with MR (FCUMUL = FRESIDUAL + MR). FCUMUL is comparable to the flux measured using conventional sediment traps and can be contrasted with published values and/or concomitant sediment-trap measurements. At PAPA, FCUMUL was twofold to threefold higher than POC fluxes simultaneously measured at the same depths71 (but note, different collection periods and aspect ratio of the traps versus C-RESPIRE, which both influence the magnitude of the POC flux), but within the range of previous measurements at this site88,89. In MED, FCUMUL obtained at 195-m depth was consistent with the range of previous estimates (0.4–3.0 mmol m−2 d−1)90, but about fourfold higher than fluxes simultaneously measured at 250-m depth using a PPS5 conical sediment trap91. This discrepancy in flux estimates probably reflects the power-law decrease of the POC flux expected between 195-m and 250-m depth, but also the lower collection efficiency of conical traps compared with cylindrical traps92 such as the C-RESPIRE.

POC flux attenuation

The rate of POC flux attenuation (Fig. 2) was obtained using the power-law function or ‘Martin curve’6:

$${F}_{Z}={F}_{Z0}\times {\left(z/{z}_{0}\right)}^{-b},$$

where z and z0 are the depths of the flux, FZ0 is the POC flux at the reference depth (the shallowest flux in this study), and b is the attenuation coefficient. The b value is highly sensitive to the choice of the reference depth15 and the number of data points per profile93, and is usually estimated over the first >500 m of the water column. In this study that focused on the upper mesopelagic and deployed C-RESPIREs within a narrow depth range (about 200 m), the Martin curves presented in Fig. 2 are only informative.

Carbon-specific remineralization rate

The C-specific remineralization rate, Cremin (d−1), is calculated using the equation:

$${{\rm{C}}}_{{\rm{remin}}}={\Delta {\rm{O}}}_{2}\times {\rm{RQ}}/\left({\Delta {\rm{O}}}_{2}\times {\rm{RQ}}+{\left[{\rm{POC}}\right]}_{{\rm{POST}}\text{-}{\rm{INCUB}}}\times 1.45\right)/{T}_{{\rm{INCUB}}}$$

Relationships between Cremin and depth or temperature were assessed via a linear model fit (Fig. 3). Considering the large range of dissolved oxygen concentrations observed at BEN (about 45–220 µmol kg−1) and the potential role of oxygen in controlling Cremin (refs. 11,84), the relationship between Cremin and dissolved oxygen concentration was also assessed. At BEN, a lower correlation was obtained between Cremin and dissolved oxygen concentrations (R2 = 0.60, P = 0.15) relative to the correlation between Cremin and temperature (R2 = 0.80, P < 0.01).

Error propagation

The errors in MR, FRESIDUAL, FCUMUL and Cremin were calculated by propagating the uncertainties from subsampling the assemblage of particles collected at the end of the C-RESPIRE incubations using a splitter (that introduces an approximately 10% variability), POC analysis, O2 measurements and the 20% error from assuming a constant RQ83.

Justification for assuming constant particle sinking rates for the upper mesopelagic

In this study, owing to a lack of published observations of particle sinking rates over the upper mesopelagic, we assume constant sinking speed, implying that the POC loss per day versus POC loss per metre relationship remains unchanged with depth. This assumption allows us to discuss the relative constancy of MR and its relative importance as an attenuation pathway with depth. Changes in sinking velocities with depth (in the water column above each C-RESPIRE) are inherently taken into account for particles intercepted by C-RESPIREs deployed at different depths (Fig. 1a). There is evidence of an increase of the average sinking velocity with depth over the mesopelagic (that is, about 100 m to 1,000 m)94,95. However, the applicability of this trend to the upper mesopelagic (in our study about 100 m to 300 m depth) where aggregation, solubilization and fragmentation can influence sinking velocity in opposing ways (for example, fragmentation slows particle settling55) is unknown.