Reconciliation of the carbon budget in the ocean’s twilight zone

Photosynthesis in the surface ocean produces approximately 100 gigatonnes of organic carbon per year, of which 5 to 15 per cent is exported to the deep ocean. The rate at which the sinking carbon is converted into carbon dioxide by heterotrophic organisms at depth is important in controlling oceanic carbon storage. It remains uncertain, however, to what extent surface ocean carbon supply meets the demand of water-column biota; the discrepancy between known carbon sources and sinks is as much as two orders of magnitude. Here we present field measurements, respiration rate estimates and a steady-state model that allow us to balance carbon sources and sinks to within observational uncertainties at the Porcupine Abyssal Plain site in the eastern North Atlantic Ocean. We find that prokaryotes are responsible for 70 to 92 per cent of the estimated remineralization in the twilight zone (depths of 50 to 1,000 metres) despite the fact that much of the organic carbon is exported in the form of large, fast-sinking particles accessible to larger zooplankton. We suggest that this occurs because zooplankton fragment and ingest half of the fast-sinking particles, of which more than 30 per cent may be released as suspended and slowly sinking matter, stimulating the deep-ocean microbial loop. The synergy between microbes and zooplankton in the twilight zone is important to our understanding of the processes controlling the oceanic carbon sink.

The global carbon cycle is affected by biological processes in the oceans, which export carbon from surface waters in form of organic matter and store it at depth; a process called the 'biological carbon pump'.Most of the exported organic carbon is processed by the water column biota, which ultimately converts it into CO2 via respiration (remineralization).Variations in the resulting decrease in organic flux with depth 9 can, according to models, lead to changes in atmospheric CO 2 of up to 200 ppm 3 , indicating a strong coupling between biological activity in the ocean interior and oceanic storage of CO 2 .
A key constraint in the analysis of carbon fluxes in the twilight zone is that, at steady state, the attenuation of particulate organic carbon (POC) flux with depth should be balanced by community metabolism.Published estimates of POC flux attenuation with depth are, however, up to 2 orders of magnitude lower than corresponding estimates of heterotrophic metabolism [4][5][6][7] .This discrepancy indicates that either estimates of POC flux and/or community metabolism are unreliable, or that additional, unaccounted for, sources of organic carbon to the twilight zone exist 8 .
We compiled a comprehensive carbon budget of the twilight zone based on an extensive programme of field measurements at the Porcupine Abyssal Plain site (PAP; Extended Data Fig. 1a) in July/August 2009.This site is located in the transition region between the subtropical and subpolar gyres of the North Atlantic 10 .Mixed layer depth remained constant at ~50 m throughout the study period.This depth was subsequently used as the upper boundary of the twilight zone, following the need to normalize export measurements to dynamic upper boundaries for the twilight zone 11 .
Organic carbon sources to the twilight zone include (1) sinking particles, (2)   downward mixing of dissolved organic carbon (DOC), (3) lateral advection of organic matter from the continental shelf, (4) active transport via the diel vertical migration of zooplankton that feed in the mixed layer during night time and rest at depth during the day, and (5) chemolithoautotrophy (prokaryotic growth using dissolved inorganic carbon and chemical energy sources).
The downward flux of sinking particles was measured using simultaneous 48-h deployments of free-drifting, neutrally-buoyant sediment traps 12  at 50, 150, 300, 450   and 600 m (Extended Data Table 1).Satellite chlorophyll imagery and horizontal velocities (obtained using a 150-kHz Vessel-Mounted Acoustic Doppler Current Profiler) confirmed that all of the traps were advected along the edge of an anticyclonic eddy for 50 km before surfacing within 3.5 km of each other.Measured POC flux at 50 m (84±8 mg C m -2 d -1 ) was close to independently derived estimates using 234 Th budgets and studies of collected marine snow particles (99±41 and 146±26 mg C m -2 d -1 , respectively) 13 .Flux attenuation with depth followed the Martin curve (F = F 100 (z/100) b ) 9 with b = -0.70(p<0.01,R 2 =0.95, n=5) and was consistent with observations in the Pacific Ocean (b: -0.50 to -1.38; Fig. 1a) 9,14 .Downward POC flux was extrapolated to 1000 m using b = -0.70.The total loss of POC within the twilight zone was 74±9 mg C m -2 d -1 .
DOC input to the twilight zone was estimated to be 15 (0.4-30) mg DOC m -2 d -1 based on the ratio between DOC concentrations and apparent oxygen utilization 15 , and on DOC gradients coupled to turbulent diffusivity measured from previous work at the study site 16 (Methods; Extended Data Fig. 2).DOC was estimated to supply 17% of total export in agreement with previous estimates of 9-20% across the North Atlantic basin 17 .Organic matter input via lateral advection was assumed to be negligible based on analyses of back-trajectories (derived from satellite-derived near-surface velocities over 3 months) of the water masses arriving at the PAP site during the study period, which suggested that the water had not passed over the continental slope (Extended Data Fig. 1b).The final source of DOC, excretion at depth by active flux, was estimated using net samples of zooplankton biomass and allometric equations 6,18 , giving a supply of 3 mg C m -2 d -1 .Defecation and mortality at depth present further sources of organic carbon to the twilight zone, but these were excluded from the budget due to large uncertainties associated with their estimation.Finally, chemolithoautotrophy has been suggested to be a significant source of organic matter in the deep ocean 19 , but without strong evidence that this poorly understood process could provide a major contribution at our study site, we chose to exclude it from our carbon budget.
The remineralization of organic carbon by zooplankton and prokaryotes was estimated from zooplankton biomass and prokaryotic activity.It is crucial to note that in a steady state system, such as we assume this to be, organic carbon is lost from the system only by export or by remineralization.We focus entirely on community respiration as a measure of remineralization, a fundamental advance over previous methods to derive budgets (Methods).
Zooplankton respiration was estimated by applying allometric relationships 6 to biomass measurements derived from net samples collected vertically every 80 m, twice during both day and night, using the ARIES net-system fitted with 200-µm cod-ends (Extended Data Table 1, Extended Data Fig. 3).These allometric relationships are wellconstrained 6 , however they are based on epipelagic zooplankton and our calculated respiration rates for the lower mesopelagic are therefore likely overestimates of the true rates 20 .Zooplankton resident in the twilight zone, mostly detritivorous copepods (Oithona, Oncaea) and carnivorous chaetognaths, had combined respiration rates of 15.2 and 12.7 mg C m -2 d -1 (50-1,000 m) respectively during the two deployment periods (Fig. 1b).Migrating zooplankton (determined as the difference between day and night biomasses) were excluded from these estimates because we assume that they ingest sufficient carbon during grazing at the surface to satisfy their diagnosed respiration rates at depth (Methods).The organic carbon they respire within the twilight zone is thus imported by diel vertical migration.Prokaryotic heterotrophic production (PHP) was determined using bioassay-isotopedilution techniques using 3 H-leucine tracer 21 .Leucine incorporation rates were 41.7±21.2nmol Leu m -3 d -1 at 150 m and 6.6±4.1 nmol Leu m -3 d -1 at 500-750 m (Fig. 1c), similar to previous estimates in the Northeast Atlantic Ocean (37.7 and 7.5 nmol Leu m -3 d -1 , respectively) 19 .Integrated leucine incorporation based on a power-law fit was 14.5 µmol Leu m -2 d -1 (interquartile range: 13.2-16.1 µmol Leu m -2 d -1 , p<0.001, R 2 =0.86, n=37).This fit was chosen on the assumption that bacterial activity follows the supply of organic carbon 22 , although we lack data from between 50-150 m to confirm this fit.The uncertainty in this interpolation possibly leads to a misestimate of integrated leucine incorporation.Integrated leucine incorporation was converted into respiration using leucine-to-carbon conversion factors (0.44±0.27 kg C mol -1 Leu) and growth efficiencies (interquartile range: 0.04-0.12)specific to the twilight zone derived from thorough literature surveys (Methods; Extended Data Fig. 4).The uncertainty in this calculation was estimated using bootstrap analysis with 100,000 simulations.The final estimate for integrated (50-1,000 m) prokaryotic respiration was 71 mg C m -2 d -1 (interquartile range: 35-152 mg C m -2 d -1 ).
Our study is the first to successfully reconcile the various elements of the carbon budget in the twilight zone of the ocean.This was possible because we (1) considered a dynamic upper boundary for the twilight zone (the base of the mixed layer), (2)   excluded vertical migrators from the estimate of zooplankton respiration in the twilight zone, and (3) compared respiration rather than carbon demand to net organic carbon supply.Depth resolved estimates of supply and consumption (Extended Data Fig. 5) show an excess of supply in the upper twilight zone (50-150 m) and a deficit in the lower twilight zone (150-1,000 m).We suggest that this may be caused by a subtle vertical change in ecosystem structure with depth 23,24 or an unaccounted for vertical transfer of organic carbon between the upper and lower twilight zones.
The suggestion that prokaryotes dominate community respiration seems counterintuitive given that organic carbon supply to the twilight zone is dominated by sinking particles that are accessible to larger (>200 µm) zooplankton.We therefore hypothesise that a major role of zooplankton in the twilight zone is to mechanically degrade particulate material 25 into slow-sinking particulate matter and dissolved organic material that is subsequently remineralized by microbes (prokaryotes and their consumers).
In order to explore whether this conceptual picture is consistent with our current understanding of twilight zone ecology, and to provide a full quantitative picture of the twilight zone carbon cycle, we used a simple steady-state model of the twilight zone carbon cycle 26 .The model traces the turnover and remineralization of sinking POC via three pathways: colonization and solubilisation of detritus by attached microbes, production of free-living microbes following loss of solubilisation products during particle degradation, and consumption by detritivorous zooplankton (Method; Extended Data Fig. 6a).The model was modified to include vertical mixing of DOC and active transport as carbon inputs to the twilight zone and to represent POC in both sinking and suspended forms, the latter produced via zooplankton 'sloppy feeding' 27 .Inputs of carbon to the twilight zone were the measured values given in Table 1.
Modelled respiration rates matched field data well, with 84% of the CO 2 being produced by microbes (prokaryotes and prokaryote consumers) and only 16% by zooplankton (detritivores and carnivores) (Fig. 2).The model further suggests that microzooplankton respiration, which had not been measured during the study, plays a small role in the overall budget contributing only 5 mg C m -2 d -1 .Attached prokaryotes processed half of the POC flux with the remaining half being processed by detritivorous zooplankton, which released 30% of it as suspended POC, thereby confirming our hypothesis.The relative roles of zooplankton and prokaryotes for processing and respiring sinking POC are robust to changing model parameter values (Methods; Extended Data Fig. 7).Moreover, it is consistent with the general perception that detritivores are sloppy feeders that ingest <40% of processed particles, causing the bulk part of fast-sinking POC to break up into slow-or non-sinking POC and DOC 25 .This pool of suspended organic matter stimulates the microbial loop 28 in the twilight zone and ultimately fuels the respiration of prokaryotes 6,26 .
Our results highlight a synergy between zooplankton and microbes in the twilight zone where both play significant roles in processing the organic carbon flux and, subsequently, in controlling the strength of the oceanic carbon sink.Large uncertainties remain, however, particularly with regard to estimating prokaryotic activity.A better understanding of prokaryotic metabolism throughout the twilight zone combined with process studies focusing on the upper twilight zone are necessary to fully understand the biological carbon pump.

Methods summary
We conducted an extensive programme of field measurements at the Porcupine Abyssal Plain site (49°0' N, 16°5' W) from 8 Jul -13 Aug 2009 aboard RRS Discovery.Sinking material was collected for 48 hours using free-drifting, neutrally-buoyant PELAGRA sediment traps 12 .Samples were screened to remove swimmers, split into aliquots, filtered onto precombusted GF/F filters, fumed with sulphurous acid and analysed for POC.DOC input was estimated from data collected near the PAP site during Jun/Oct 2005 (http://www.bodc.ac.uk/).The slope of the correlation between measured DOC and apparent oxygen utilization was compared to the theoretical slope (C org /-O 2 = 117/170), giving the relative contribution of DOC to heterotrophic respiration.A lower estimate was calculated using turbulent diffusivity measurements at the PAP site 16 coupled with the aforementioned DOC profiles.Samples for zooplankton biomass profiles (0-1,000 m at 80-m intervals) were preserved in formaldehyde, sizefractionated, identified and enumerated.1-50 individuals from each group at each depth and size fraction were analysed for dry weight.Zooplankton respiration (μg C individual -1 h -1 ) was estimated as a function of body mass (mg dry weight ind -1 ) and temperature (°C) 6 .DOC excretion at depth was assumed to be equivalent to 31% of respiration by migrating zooplankton 18 .Leucine incorporation rates were estimated on samples (n=37) recovered from depth using a Conductivity, Temperature, and Depth (CTD) rosette sampler.Both time-course experiments and concentration-series bioassays were carried out.Respectively, 3 H-leucine was added at 10-20 nM and 0.025-0.5 nM final concentration and incubated in the dark at in situ temperatures for 4-8 hours and 0.5-2 hours.Samples were filtered onto 0.2-μm polycarbonate filters, washed with deionised water, and their radioactivity measured.Input fluxes and respiration rates (mg C m -2 d -1 ) are based on measurements at the PAP site, North Atlantic.Numbers in brackets refer to lower and upper estimates (see text).
Community respiration (%) was estimated by combining highest and lowest estimates.1).The fraction of grazed POC that is lost to suspended POC due to sloppy feeding by detritivores (λ H ) was varied between 0.1-0.5 (standard value 0.3).PGE (ω fl ) was assigned values of 0.04 (a,b), 0.08 (c,d), and 0.12 (e,f).Red areas show the estimated range based on field data.

Extended Data Figure 8| Twilight zone carbon budgets based on 'carbon demand'.
Budgets were compiled by comparing loss of particulate organic carbon (ΔPOC; black) to 'carbon demand' (ingestion) by zooplankton (dark grey) and prokaryotes (light grey) in the North Atlantic ('PAP'; this study) and at two stations in the Pacific ('ALOHA' and 'K2') 6 .The imbalance of these budgets contrast with our final budget (Fig. 1d

Particulate flux measurements
Sinking flux of particulate organic carbon was measured at five depths (51, 184, 312,   446 and 589 m) concurrently, using free-drifting, neutrally-buoyant traps called PELAGRA (Particle Export measurement using a LAGRAngian trap) 12 .Sample cups for each trap were filled with filtered seawater of 5 ppt excess salinity and sufficient chloroform to give a saturated solution.Traps were deployed with sample cups closed; after a 24 hour period to reach and stabilise at the programmed depth, the cups opened and collected sinking material for 48 hours, before closing immediately prior to ascent to the surface.From each trap, two sample cups were combined, screened through a 350-µm mesh to remove swimmers, and split equally into eight aliquots for different analyses.POC-designated splits were filtered at sea through one or more pre-combusted (450°C, 12 hours), 25-mm-diameter glass fibre filters, stored frozen (-20°C), then later fumed with 100 mL concentrated sulphurous acid for 48 hours, dried (60°C, 24 hours) and pelleted in pre-combusted aluminium foil.Analysis was carried out using a Thermo Finnigan Flash EA1112 elemental analyser with acetanilide as the calibration standard.

DOC input
The contribution of DOC to sustaining interior heterotrophic respiration was calculated following Doval and Hansell 15 by assuming that the utilization of primary elements in the twilight zone generally follows the Redfield ratio.We used previous data collected near the PAP site during Atlantic Meridional Transect (AMT) cruises 16 and 17 in June and October 2005 (http://www.bodc.ac.uk/).DOC shows the characteristic surface enhancement (up to 70 µM) with a reduction to 55 µM at 300 m 29 .The profiles show little variability implying that the supply of DOC is rather constant with season (Extended Data Fig. 2a).Extended Data Fig. 2b shows the regression between apparent oxygen utilization (AOU) and DOC and the theoretical relationship which would occur if all AOU were due to DOC degradation (C org :O 2 ratio of 117:170).The ratio between the two gradients is 0.377 suggesting that DOC explains 38% of the respiration between 57-300 m.This is consistent with an estimated contribution of DOC respiration to total AOU of 18-47% in the upper 500 m across the South Pacific and Indian Ocean 15 .Assuming that sinking POC flux attenuation between 57 and 300 m (49 mg C m -2 d -1 ) made up the remaining 62%, DOC contributed 30 mg C m -2 d -1 to the carbon flux.
An alternative calculation uses turbulent diffusivity measurements at the PAP site 16 coupled with the aforementioned DOC profiles.The flux of DOC into the twilight zone (F DOC ) can be calculated as where k is the diffusivity (10 -5 -10 -4 m 2 s -1 ) 16 , Δz is the depth interval (57-300 m), and ΔDOC is the concentration gradient across this depth interval (10 µM).The estimated export of DOC into the twilight zone via turbulent mixing was 0.4-4 mg C m -2 d -1 .This process does not include DOC fluxes out of the mixed layer from mesoscale processes, and the true DOC export is likely to be closer to the first estimate of 30 mg C m -2 d -1 .
Based on the two estimates for DOC export, we applied a conservative value of 15 mg C m -2 d -1 for the construction of the twilight zone carbon budget at the PAP site.

Lateral advection
Surface ocean currents derived from satellite altimeter and scatterometer data were downloaded from the NOAA OSCAR website (http://www.oscar.noaa.gov/).The obtained currents encompass both the geostrophic and wind-driven (Ekman) motion and are available at 1/3-degree, 5-day resolution.Particles were tracked back in time for 3 months from the initial deployment date of the PELAGRA and ARIES instruments.

Distinction between respiration and carbon demand
The construction of an ecosystem carbon budget is dependent on the definition of input and output terms.If the input is defined as the net supply of organic carbon (the flux entering the twilight zone less that exiting at the base), then the analogous output is the removal of organic carbon via conversion to inorganic carbon during respiration.
Respiration differs from the frequently used 'carbon demand' 4-7,30,31 as the latter is quantified as either 'ingestion' or 'ingestion minus egestion' and therefore an unconstrained quantity.Consider a zooplankton grazer: At steady state, its carbon demand (i.e.ingestion) is balanced by the sum of biomass production (growth and reproduction), excretion, respiration and faecal production 32 .Except for respiration, these processes all produce organic matter that becomes available as food for other heterotrophic organisms such as carnivores or detritivores.In other words, organic carbon is retained and recycled in the system and any one carbon atom may be recycled many times with carbon demand exceeding (being unconstrained by) carbon supply 33, 34 .
In contrast, each carbon atom within organic matter can only be respired once, ending its journey in the food web, such that, at steady state, respiration equals carbon supply.
A similar phenomenon exists when (incorrectly, as has often been the case) comparing bacterial carbon demand with primary production, the correct ratio being bacterial respiration to primary production 34,35 .
To illustrate the impact that making the distinction between respiration and carbon demand has on the calculation of the twilight zone carbon budget, we calculated carbon demand from our data following Steinberg et al. 6 over a similar depth range (150 -1,000 m) to allow direct comparability.We then compared our estimates to their observations from the North Pacific.Prokaryotic carbon demand (PCD) was calculated as where PHP is prokaryotic heterotrophic production measured using tritiated leucine, and PGE is a prokaryotic growth efficiency of 0.15 (as Steinberg et al. 6 ).Zooplankton carbon demand (ZCD) was estimated as where ZR is the allometrically determined respiration rate 36-38 , NGE is the net growth efficiency (0.5 as Steinberg et al. 6 ), and AE is the absorption efficiency (0.6 as Steinberg et al. 6 ).
Our modified budget for the North Atlantic is qualitatively similar to the observations from the oligotrophic subtropical (station 'ALOHA') and mesotrophic subarctic (station 'K2') Pacific 6 (Extended Data Fig. 8).In all cases, the sum of prokaryotic and zooplankton carbon demands exceeds the supply of carbon to the system by a factor of 8-10.This contrasts with the balanced carbon budget we originally calculated at the PAP site.Three key aspects of our original data analyses (use of respiration rather than carbon demand, exclusion of vertical migrators from respiration estimates, and the use of a depth range of 50-1,000 m for the twilight zone) are critical for balancing the twilight zone carbon budget.

Zooplankton collection and preparation
Four vertical high-resolution profiles of zooplankton biomass and abundance were collected in association with the sediment trap deployments: one at daytime and one at night-time at both the beginning and end of the observational period (Table 1).
Excretion at depth via the active flux was estimated by assuming that DOC excretion by migrating zooplankton is equivalent to 31% of their respiration 18 .

Ingestion by vertically migrating zooplankton
Typical vertical migration patterns were observed during both deployments (Extended Data Fig. 3c,d) with large copepods and euphausiids dominating the migrating zooplankton.We assume that at depth these organisms respire material which they have ingested at the surface, and test this assumption using the equation where I ML is the total ingested carbon in the mixed layer (mg POC m -2 d -1 ), F ind is the average clearance rate (mL individual -1 d -1 ), c POC is the concentration of POC in the mixed layer (97 mg POC m -3 ) 13 , n ML represents the number of zooplankton in the mixed layer (7170 and 10370 individuals m -2 during the two deployment periods, respectively), and t is the time that migratory zooplankton spend in the mixed layer each night according to ADCP back scatter profiles (9 h/24 h).Using reported clearance rates of 72-432 mL d -1 individual -1 for Calanus 32,39,40 and 360 -2,400 mL d -1 individual -1 for euphausiids 41,42 , total ingestion rates ranged from 18 to 905 mg C m -2 d -1 .
The daily respiration rates of migratory zooplankton (estimated as for resident zooplankton) were 8 mg C m -2 d -1 , significantly lower than the calculated ingestion rates.This suggests that migrating zooplankton were able to ingest sufficient organic carbon in the mixed layer to satisfy their respiration, as well as other physiological processes such as growth, egestion and excretion.It is noteworthy that the strong coupling between diel vertical migration and environmental variables means that migration patterns and associated carbon cycling may change in response to climate change 43 .

Prokaryotic leucine incorporation
Incorporation rates of radiolabeled leucine 21 were measured following two protocols: time-course experiments 44 and concentration-series bioassays 45,46 .For the time-course experiments, samples were taken from four depths at four stations in association with the trap deployments (Extended Data Table 1).L-[3,4, samples in total, were incubated in 2 mL capped screw top sterile polypropylene microcentrifuge tubes in the dark at in situ temperatures.One of the samples for each concentration was fixed at 0.5, 1, 1.5 and 2 hours, respectively, by adding paraformaldehyde (PFA) to 1% final concentration.
All sample particulate material was harvested onto 25-mm-diameter 0.2-µm polycarbonate filters soaked in unlabelled leucine to reduce background sorption.Filters were washed twice with 4 mL of deionised water (Milli-Q system, Millipore).
Radioactivity retained on filters was measured as disintegrations per minute (DPM) using a liquid scintillation counter (Tri-Carb 3100, Perkin Elmer).Turnover time and estimates of leucine incorporation rate at ambient concentrations from the concentration-series bioassays were calculated following Zubkov et al. 46 .Leucine incorporation rates by the two methods agreed well, and there appeared to be little spatial and temporal variability in the twilight zone.All data was therefore pooled for the calculation of prokaryotic respiration.

Prokaryotic respiration
The estimation of prokaryotic respiration (PR) based on leucine incorporation rates requires two factors: where LeuCF is the leucine-to-carbon conversion factor, and PGE is the prokaryotic growth efficiency.We reviewed all PGEs and LeuCFs determined for the twilight zone (Extended Data Fig. 4a,b) and estimated prokaryotic respiration (and error margins) using bootstrap analysis with 100,000 simulations (Extended Data Fig. 4c).
The simulations were computed as follows: Integrated leucine incorporation rates were determined based on the measured leucine incorporation rates at our site.A power-law distribution was fitted to the bootstrap sample (p<0.001,R 2 =0.86, n=37), interpolated (50-1,000 m), and summed to get the integrated incorporation rate.The resulting leucine incorporation rates had a median of 14.5 µmol Leu m -2 d -1 (interquartile range: 13.2 -16.1 µmol Leu m -2 d -1 ).LeuCFs for the simulation were randomly sampled (with replacement) from all reported LeuCFs for the twilight zone (n=21) 47-50 .The mean LeuCF used in the simulation was 0.44 kg C mol Leu -1 (interquartile range: 0.26 -0.59 kg C mol Leu -1 ).Finally, PGEs were randomly sampled (with replacement) from all reported PGEs for the twilight zone of the North Atlantic (n=26) 48,51-54 .PGEs ranged from 0.001 -0.24 with a median of 0.08 (interquartile range: 0.04 -0.12).
The final estimate of prokaryotic respiration is very sensitive to the interpolation method as well as the two conversion factors (LeuCF and PGE).Our study lacks measurements of leucine incorporation rates from the region between the mixed layer depth (50 m) and 150 m, which is the area where most of the POC is remineralized.To arrive at an integrated estimate for leucine incorporation, we chose to interpolate the available leucine incorporation rates using a power-law function as we assume that prokaryotic production in this region is driven by the supply of organic carbon 22 , which is best described by a power-law function 9 .The choice of interpolation method introduces additional, large uncertainties in our estimate potentially leading to a misestimate of integrated leucine incorporation.We recommend that future studies should avoid this uncertainty by increasing sampling effort in this critical region.

Food web model
The food web analysis (Extended Data Fig. 6a) is based on the steady-state model of First, carbon input to the twilight zone now includes both sinking detritus and DOC, the latter representing both vertical mixing and active transport via migratory zooplankton.
Second, detritus is divided between sinking and suspended forms (AT10 included only the former).It was assumed by AT10 that zooplankton losses due to sloppy feeding are as DOC.It may however be the case that, particularly for copepods feeding on detritus, much of this loss is as fragmentation (so-called coprorhexy 25 ) leading to the generation of small non-sinking particles.
In the new version of the model detritus is therefore divided between sinking material (D1), with inputs as export from the surface ocean and as faecal pellet production from detritivores and carnivores, and suspended detritus (D2) which is derived from coprorhexy by detritivores and carnivores, and as faecal pellet production by microzooplankton ('prokaryote consumers').D1 is consumed by both detritivorous zooplankton and attached prokaryotes, as in AT10, whereas D2 is acted on only by the prokaryotes.
The model was reparameterised as follows (see Extended Data Table 2 for list of parameters).Parameter ψ B , the partitioning of detritus consumption by attached prokaryotes and detritivores, is poorly known and was estimated as 0.75 (75% prokaryotes, 25% zooplankton) by AT10 based on the data of Steinberg et al. 6 .Using the data from the PAP site, we were better able to constrain this parameter and use a value of ψ B = 0.5 (see sensitivity analysis below).Of the POC (D1 and D2) acted on by attached prokaryotes, 50% is solubilised due to the action of hydrolytic enzymes and released as DOC 26 (parameter α).PGEs for free-living and attached prokaryotes (ω fl , ω att ) were set to 0.08 and 0.24 respectively, the former based on the literature review presented above and the latter from AT10.Release of DOC as excretion by prokaryote consumers, detritivores and carnivores was set at 5% of processed prey items (Φ V , Φ H , Φ Z = 0.05) 26 .The corresponding fraction allocated to D2 via sloppy feeding was set as λ H = 0.30 for detritivores (based on Fig. 2 of Lampitt et al. 25 ), thereby assuming that a large fraction of processed food is released as non-pellet POC, with a value of λ Z = 0.15 for carnivores.Sloppy feeding losses by prokaryote consumers were assumed to be zero (λ V = 0) because they ingest their prey whole.Absorption efficiencies (also commonly known as assimilation efficiencies) were assigned values of β H = 0.60, β Z = 0.66 and β v = 0.72 [26,55] .The fraction of prey items that is absorbed across the gut is β(1-Φ)(1-λ), this material being utilised with net production efficiencies for detritivores, carnivores and bacterivores (κ H and κ Z and κ V ) of 0.39, 0.39 and 0.44 [26] .Finally, parameter ζ, the fraction of attached prokaryotes consumed by detritivores (rather than prokaryote consumers) was assigned a value of 0.24 [26] .
Anderson and Tang

Sensitivity analysis
An analysis of the steady state solution of the model is presented in Extended Data respectively).The main detritus source is export of sinking particles from the surface ocean, supplemented by in situ faecal pellet production by detritivores and carnivores.
Although detritivores and attached prokaryotes each utilise 50% of D1 (parameter ψ B ), it is the attached prokaryotes which undertake the majority of POC utilisation overall (57% versus 43%) because they are the sole consumers of D2.Finally, the largest DOC source in the model is solubilisation of detritus by attached prokaryotes (31.5 mg C m -2 d -1 ); greater than the input from the surface ocean (18 mg C m -2 d -1 ).Utilisation of DOC is exclusively by free-living prokaryotes.Overall, the results highlight the significant roles played by both zooplankton and prokaryotes in the carbon cycle of the twilight zone, the former primarily as recyclers and the latter as a carbon sink (Fig. 2).
The robustness of the model results and conclusions with respect to chosen parameter values were investigated by undertaking sensitivity analyses.Parameter ψ B (the fraction of D1 acted on by attached prokaryotes, the remainder by detritivorous zooplankton) was varied between 0.1 and 0.9 (standard value 0.5), λ H (the loss to D2 by sloppy feeding by detritivores) between 0.1 and 0.5 (standard value 0. if less processed D1 is allocated to D2 (in which case more is respired), but not to any great extent.For example, decreasing suspended losses (parameter λ H ) from 30% to 10% means that the required ZR (to match the data) is achieved with ψ B = 0.63, i.e.
detritivores processing 37% of D1.We conclude that the model predictions are robust with respect to a mid-range value of ψ B , e.g.0.5.
PGE is notoriously low in the twilight zone of the ocean 48,51-54 .Visual inspection of Extended Data Fig. 7 shows that predicted ZR and PR are remarkably insensitive to ω fl .For example, decreasing ω fl to 0.04 (half the standard value) meant that predicted ZR (for ψ B = 0.5 and λ H = 0.3) decreased from 14.3 to 14.0 mg C m -2 d -1 and PR increased from 72.5 to 73.7 mg C m -2 d -1 .The relative insensitivity is easy to explain in that prokaryotes are the main sink for carbon and so decreasing PGE just strengthens this.Likewise, increasing ω fl to 0.12 has only a minor impact on model results (Extended Data Fig. 7e,f).Predicted ZR for ψ B = 0.5 and λ H = 0.3 increases to 14.6 mg C m -2 d -1 as carbon transfer to higher trophic levels is increased, while PR decreases to 71.3 mg C m -2 d -1 .
Overall, the results are robust to changes in PGE, as well as changes in detritivore sloppy feeding losses (λ H ). Model solutions indicate that a mid-range value of parameter ψ B , in the region of 0.5, is required in order to match the observational data and thus confirms the overall conclusion of the synergistic role of zooplankton and prokaryotes in carbon cycling in the twilight zone of the ocean.

Figure 1|
Figure 1| Sinks and sources of organic carbon to the twilight zone.a, Particulate organic carbon flux (POC; black dots) below the mixed layer (shaded area) at the PAP site during 3-6 th Aug 2009 fitted to the Martin equation (F z = F 100 (z/100) b ; solid line).The observed attenuation is consistent with rates observed in the Pacific (grey area, dotted lines) 9,14 .Error bars show analytical error (s.d.).b, c, Depth profiles of respiration by non-migratory zooplankton (ZR; b) and leucine incorporation (µmol Leu m -3 d -1 ) by prokaryotes (power-law fit and interquartile range; p<0.001,R 2 =0.86, n=37; c).d, The sum of net organic carbon supply (ΔOC; light grey) of particulate (POC) and dissolved organic matter (DOC) and active flux (marked with an asterisk) matches respiration by non-migratory zooplankton (ZR; dark grey) and prokaryotes (PR; mid grey).Error bars represent upper and lower estimates (see text and Table1).

Figure 2|
Figure 2| Predicted carbon cycle in the twilight zone.Organic carbon is supplied to the twilight zone in particulate (POC) and dissolved form (DOC; vertical mixing + active transport) (green arrows).POC is processed by detritivores (50%) or attached prokaryotes (50%) and recycled in the twilight zone until eventually remineralized (red arrows) whereby prokaryotes dominate respiration (79%).Observed rates are in green and red boxes.Internal net flows (mg C m -2 d -1 ) derived from a numeric model are represented as arrows (line width in scale).Fluxes indicated with an asterisk (*) are for microzooplankton (prokaryote consumers), which are not included in the measured estimates.
) based on respiration.Error bars show analytical errors for POC flux and upper and lower estimates for carbon demands based on a range of conversion factors (see methods by Steinberg et al. 6 for details).Extended Data Table 1| Deployment details.Deployments of neutrally-buoyant sediment traps (PELAGRA) and the plankton sampler (ARIES) at the PAP site in August 2009.Sampling of sediment traps commenced 24 hours after deployment time and lasted for 48 hours.Presented times relate to deployment and recovery of the traps.Depths for PELAGRA deployments are mean depths.Extended Data Table 2| Model parameters and default values.Note, all parameters are dimensionless.
(N)]leucine (specific activity 115.4 Ci mmol -1 , Perkin Elmer) was added to give a final concentration of 20 and 10 nM in triplicate 20-mL and 40-mL samples from the mixed layer and upper twilight zone (0-150 m) and lower twilight zone (>150 m), respectively.Respective samples were incubated for 4 and 8 hours in sterile Falcon vials in the dark at in situ temperatures.The samples were fixed with formaldehyde (2% final concentration).For the concentration-series bioassays, 2-L water samples were collected from different depths throughout the cruise.L-[4,5-3 H]leucine (specific activity 5.26 TBq mmol -1 , Hartmann Analytic GmbH) was added in a range of six final leucine concentrations from 0.025 to 0.5 nM.Four samples (1.6 mL each) for each added concentration, i.e. 24

Anderson and Tang 26 (
hereafter AT10).The starting point of AT10 is POC input to the twilight zone via sinking detritus.The biological utilisation and subsequent respiration of this carbon is then traced via three pathways: (1) colonisation, solubilisation and production by attached prokaryotes, (2) production of free-living prokaryotes fuelled by DOC generated as a product of solubilisation, and (3) consumption by detritivorous zooplankton.We use a new version of this model which maintains these pathways, but with two adjustments.

Fig. 6b- d .
Fig. 6b-d.Inputs of carbon to the twilight zone, namely POC (74 mg C m -2 d -1 ) and DOC (18 mg C m -2 d -1 ) are balanced by community respiration, which is the sum of attached and free-living prokaryotes (23.9 and 48.6 mg C m -2 d -1 respectively), detritivores, carnivores and prokaryote consumers (11.0, 3.3 and 5.1 mg C m -2 d -1 3) and ω fl (PGE for freeliving prokaryotes) was assigned values of 0.04, 0.08 (standard) and 0.12.The resulting predictions for respiration by zooplankton and prokaryotes are shown in Extended Data Fig. 7; the point of interest is the parameter ranges which are consistent with measured estimates of respiration, i.e., ZR = 14 mg C m -2 d -1 and PR = 71 mg C m -2 d -1 .ZR predicted by the model, excluding respiration by microzooplankton (prokaryote consumers), is 14.3 mg C m -2 d -1 using parameter settings ψ B = 0.5 and λ H = 0.3.Extended Data Fig. 7c,d (PGE = 0.08) show that the best solution for ZR (14.3 mg C m -2 d -1 ) is achieved with ψ B = 0.5, i.e. with detritivores and attached bacteria processing half each of D1, and with detritivores releasing 30% of their half as suspended POC (D2).The required zooplankton contribution to processing sinking POC (D1) decreases

Table 1| Carbon budget for the twilight zone (50-1000 m).
26derived steady state equations and constructed the model in a Microsoft Excel spreadsheet.The modifications to the model here (direct DOC input, detritus divided into D1 and D2) make a steady state solution difficult and so we instead constructed two versions of the model in R, the first a Monte Carlo version and the second a dynamic version that is run to steady-state (we show results for the latter, which is deterministic).The R codes are available on request from Thomas Anderson.