Substantial loss of isoprene in the surface ocean due to chemical and biological consumption

Isoprene contributes to the formation of ozone and secondary organic aerosol in the atmosphere, and thus influences cloud albedo and climate. Isoprene is ubiquitous in the surface open ocean where it is produced by phytoplankton, however emissions from the global ocean are poorly constrained, in part due to a lack of knowledge of oceanic sink or degradation terms. Here, we present analyses of ship-based seawater incubation experiments with samples from the Mediterranean, Atlantic, tropical Pacific and circum-Antarctic and Subantarctic oceans to determine chemical and biological isoprene consumption in the surface ocean. We find the total isoprene loss to be comprised of a constant chemical loss rate of 0.05 ± 0.01 d−1 and a biological consumption rate that varied between 0 and 0.59 d−1 (median 0.03 d−1) and was correlated with chlorophyll-a concentration. We suggest that isoprene consumption rates in the surface ocean are of similar magnitude or greater than ventilation rates to the atmosphere, especially in chlorophyll-a rich waters. Isoprene loss through chemical and biological consumption in the surface ocean is comparable or greater than loss via outgassing to the atmosphere, according to seawater incubation experiments conducted across contrasting regions of the global ocean.

I soprene (2-methyl-1,3-butadiene) emissions by terrestrial and marine life are altogether of a magnitude similar to the sum of natural and anthropogenic emissions of methane 1,2 , ca. 500 TgC year −1 . Owing to its reactivity and short lifetime in the atmosphere 3 (minutes to hours), isoprene impacts atmospheric chemistry by forming tropospheric ozone, modifying the oxidation behaviour of other organic compounds, and contributing to secondary organic aerosols 4,5 . Even though the oceans emit much less isoprene than vegetated land, the potential of biogenic aerosols to influence cloud albedo and lifetimes, hence climate, is large over the vast oceans remote from anthropogenic sources 6 .
On land, isoprene is produced and released mainly by trees and shrubs 7 . In the ocean, isoprene is produced primarily by phytoplankton 8 and also by seaweeds 9 . Whilst in vascular plants isoprene production is related to rapid alleviation of thermal and oxidative stress and chemical signalling 10,11 , the ecophysiological functions of isoprene biosynthesis in phytoplankton are unknown, yet a similar antioxidant role has been speculated 12 . In any case, isoprene is ubiquitous in the surface ocean, where it occurs at concentrations mostly within the 1-100 nmol m −3 range 13,14 .
Estimations of the global ocean emission of isoprene have been attempted either by top-down (balancing modelled emissions to atmospheric observations) or bottom-up (modelling oceanic isoprene concentration and air-sea flux) approaches, and they diverge by one or two orders of magnitude 14 (maximum range: 0.1-12 TgC year −1 ). In general, top-down estimates are much higher, which implies that atmospheric measurements, as well as knowledge of the atmospheric processes, are insufficient to properly constrain the top-down models, and/or the bottom-up studies underestimate the net isoprene production. An extra isoprene source through photoproduction by surfactants in the sea surface microlayer was invoked and experimentally demonstrated 15 but later esteemed not enough to resolve the large discrepancy 16 . In any case, the difficulties in constraining the global marine isoprene emission have evidenced that knowledge of the magnitude, drivers, distribution, and dynamics of isoprene cycling processes is still poor due to lack of measurements and depends too much on a number of assumptions and laboratorybased studies 11,14,17 .
It is thought that not all the isoprene produced by phytoplankton escapes to the atmosphere because part of it is degraded in seawater, but the actual proportion is unknown. Chemical oxidation is taken for granted 18 because of isoprene's high reactivity 19 , but it has never been measured. Likewise, the occurrence of isoprene-degrading bacteria in seawater has been demonstrated 20,21 and a significant microbial sink has been suggested [22][23][24] , but it has not been experimentally confirmed, let alone measured, in natural conditions including natural concentrations.
We conducted seawater incubations with the aim to determine if isoprene was chemically and biologically consumed in the surface ocean. Detailed time courses with coastal seawater informed on the kinetics of isoprene loss, and these kinetics were used to calculate loss rates from incubations conducted during four oceanographic expeditions across the Mediterranean Sea and the Atlantic Ocean, in the Tropical Pacific, and in Antarctic and Subantarctic waters. The obtained loss rate constants were compared with rate constants of air-sea flux and vertical mixing, and their variability across samples was examined by comparison with biological and environmental variables, with the aim to propose a predictive model that fills a major gap in the assessment of isoprene turnover in the surface ocean.

Results and discussion
Evidence for biological and chemical isoprene consumption in coastal seawater. The time course of isoprene concentration in coastal seawater samples incubated in closed glass bottles at the in situ temperature and in the dark demonstrated sustained loss for at least 45 h (Fig. 1a). Enclosure without headspace prevented isoprene loss by ventilation, and darkness was assumed to arrest all or most of the biological production 25 and any photochemical production 15 or degradation. Thus, the measured loss was considered the result of microbial degradation and chemical oxidation. In most cases an exponential function fitted better the decay than a linear function (Supplementary Table 1), indicating firstorder (concentration-dependent) kinetics for isoprene loss.
Incubation of microorganism-devoid (filtered through 0.2 µm) coastal seawater sampled next to seaweeds showed an isoprene loss (0.12 d −1 ) that was half the loss in non-filtered water (0.20 d −1 ; Fig. 1b 26 . Besides, should dissolved 27 BrPOs from seaweeds or outer-membrane-bound 28 BrPOs from phytoplankton occur, they would have reacted with added H 2 O 2 to produce hypobromous acid (HOBr), a strong oxidant 29 that would further remove isoprene. Indeed, the addition of BrPO consumed isoprene because it produced HOBr by reaction with the naturally occurring H 2 O 2 . Confirming this interpretation, large HOBr production by simultaneous addition of BrPO and H 2 O 2 caused complete isoprene removal in less than 4 h (Fig. 1b). Therefore, the results shown in Fig. 1b indicate that isoprene is reactive to pervasive H 2 O 2 either directly or through the formation of enzymatically derived HOBr. All in all, first-order total isoprene loss ( Fig. 1a) is expected to depend on photochemically-produced oxidants 30 Table 1 and Supplementary  Table 3). Unfiltered seawater samples from the surface ocean were incubated in glass bottles for 24 h, at the in situ temperature and in the dark, and first-order loss rate constants were determined from initial and final isoprene concentrations (see Methods). Note that loss was determined under the assumption that isoprene production was arrested in the dark 25 . There is published evidence that residual isoprene production may occur in the dark 33 , but in our incubations, it was insufficient to counteract loss. Thus, isoprene losses caused by processes other than ventilation may have been underestimated.
Loss rate constants (k loss = k bio + k chem ) varied over an order of magnitude, ranging 0.03-0.64 d −1 with a median of 0.08 d −1 (Table 1). They did not show any significant relationship to sea surface temperature (SST) (Supplementary Fig. 1) but showed proportionality to the chla concentration ( Fig. 3a) that was best described by the following linear regression equation: k loss ¼ 0:10 ð ± 0:01Þ x ½chla þ 0:05 ð ± 0:01Þ ð 1Þ The fact that the variability of k loss is largely driven by [chla] suggests that the variable term (0.10 × [chla]) corresponds to microbiota-dependent consumption (k bio ), which in our experiments gave values between 0 and 0.59 d −1 , with a median of 0.03 d −1 . These are the first experimental estimates of their kind and, hence, there are no other data to compare to. With a lack of experimental data, a pioneering modelling study 18 proposed the use of a fixed k bio at 0.06 d −1 ; more recently 23 , though, the need for a variable k bio spanning at least between 0.01 and 0.1 d −1 was invoked to balance observed concentrations in situ with predictions of the production term from phytoplankton culture data once the ventilation and chemical losses were accounted for.
Our experimental results indicate that such variable k bio indeed exists and spans even a broader range. The most complete model of the global oceanic isoprene cycle to date 17 also performed the best simulations with a variable k bio . This was computed proportional to the simulated [chla], with a proportionality coefficient of 0.054, i.e. roughly half the coefficient we obtained by linear regression of observations (0.10). Part of the k bio (or variable k loss ) is to be attributed to degradation or utilisation by heterotrophic bacteria. A pioneering study 20 demonstrated the potential for bacterial consumption after isoprene additions at concentrations at least four orders of magnitude higher than natural concentrations. This has been accompanied by sparse but solid evidence 20,21,34 for the presence in marine waters of isoprene-degrading bacteria belonging mainly to the phylum Actinobacteria. Two more recent studies 24,35 suggested that members of the ubiquitous SAR11, the most abundant bacterial clade in the ocean, can also consume isoprene, but this was mainly based on indirect evidence and requires confirmation. Our k loss did not show any significant correlation with the total bacterial abundance ( Table 1). It must be noted, though, that bacterial abundance does not necessarily parallel heterotrophic bacterial activity, less so the activity of specific phyla, whereas a general trend of higher bacterial activity with higher [chla] is commonly observed 36 . Besides, phytoplanktonderived oxidants like the aforementioned H 2 O 2 and HOBr may have also contributed to the dependence of isoprene loss on [chla]. Circumstantial evidence in one study 37 suggested that the cosmopolitan cyanobacterium Synechococcus might consume isoprene; it is worth noting that Synechococcus harbours membrane-bound BrPO 38 and may, thus, consume isoprene as a side-process of combatting oxidative stress caused by H 2 O 2 . If confirmed, this could have contributed to the correlation between k loss and [chla]. However, the three highest k loss of our experimental series were measured in waters colder than 14°C where Synechococcus occurred at very low biomass 39,40 . Therefore, these cyanobacteria cannot be invoked as responsible for the high k loss paralleling high [chla], and a large proportion of the k bio term of k loss must correspond to degradation by heterotrophic bacteria 34 as well as to reaction with biogenic oxidants from phytoplankton.
We attribute the intercept of Eq. (1) to a less variable loss by microbiota-independent chemical oxidation 18 , k chem . In remarkable support to this, the value of the intercept, 0.05 ± 0.01 d −1 , coincides with the k chem commonly prescribed in models hitherto 17,18,41 , which was calculated from reaction rate constants and estimated steady-state concentrations of photochemicallyproduced OH· and 1 O 2 in the surface ocean.
Despite the limited number of experiments, the fact that they cover a wide range of contrasting oceanic regions and conditions confers to Eq. (1) the potential to be used in numerical models of marine isoprene cycling, replacing the fixed term for microbial consumption 18,41 . The k loss vs.
[chla] relationship here proposed can also be used to predict k loss from remote sensing chla measurements (chla sat ). It must be noted, however, that the algorithms used to obtain [chla sat ] from satellite spectral data and to compare among sensors, are validated against HPLC-measured chla 42 , not against the fluorometric chla that was used in Eq. (1). To convert fluorometric to satellite chla concentrations we used a relationship obtained with a global compilation of in situ fluorometric measurements and their match-ups from SeaWiFS and MODIS Aqua sensors: 43 Substitution in Eq. (1) results in: which is our recommended equation for k loss prediction from satellite chla. Note that only the variable term (k bio ) changes from Eq. (1), while the intercept (k chem ) is maintained at 0.05 d −1 .
Comparison of isoprene sinks and total turnover time. The change of isoprene concentration ([iso]) in the surface mixed layer over time can be described as the budget of sources and sinks: where k prod , k vent and k mix are the rate constants of isoprene production, ventilation to the atmosphere and vertical downward mixing by turbulent diffusion, respectively. We calculated k vent from our sampling sites over a period of 24 h (Table 1). Ventilation has been considered the main isoprene sink from the upper mixed layer of the ocean 18 . In our sampling sites, k loss was 0.4 to 10 times the k vent (median factor: 1.2). That is, loss through microbial + chemical consumption was of the same order as ventilation, sometimes considerably faster. Vertical mixing, k mix , was estimated to be one order of magnitude lower than the other process rates (Table 1), and in all cases but one it was calculated or assumed not to be a loss term but an import term into the mixed layer, because vertical profiles generally show maximum isoprene concentrations below the mixed layer and turbulent diffusion causes upward transport 14,17 . Altogether, the microbial, chemical, ventilation, and, where relevant, mixing losses resulted in total turnover times (1/(k loss + k vent + k mix )) of isoprene between 1.4 and 16 days, median 5 days (Table 1).   Isoprene production. Assuming steady-state for isoprene concentrations over 24 h (Supplementary Fig. 2), i.e. Δ[iso]/Δt = 0 in Eq. (4), the sum of the daily rate constants of all sinks (k loss + k vent ) equals the rate constant of isoprene production (k prod ), with k mix adding to either side depending on whether it is an import to or an export from the mixed layer (Table 1). Note that k prod was the highest coinciding with higher [chla]. This is consistent with a recent study 44 where measurement of the net biological isoprene production (i.e. productionconsumption rates) across seasons in the open ocean was attempted; net production rates increased in May, coinciding with a large increase in [chla] and phytoplankton cell abundance.
The product of k prod by the isoprene concentration gives the daily isoprene production rate, which can be normalised by dividing it by the chla concentration. In our study, this specific isoprene production rate varied between 1 and 38 nmol (mg chla) −1 d −1 (Table 1) i.e. in the higher end of the laboratory data range. This is not unexpected, since measurements in monoculture experiments are typically conducted before reaching nutrient limitation, below light saturation and in the absence of UV radiation, to mention three stressors commonly occurring in the surface open ocean. If isoprene biosynthesis and release is enhanced by any of these stressors, as is the case in vascular plants 7,10 , then monoculture-derived results will easily render underestimates of isoprene production in the open ocean. Production by heterotrophic bacteria 46 could have also contributed to increase apparent specific isoprene production rates, but the occurrence and importance of this process in the marine environment is unknown.
When plotted against the SST, which was also the temperature of the incubations, specific isoprene production rates increased exponentially between −0.8 and 23°C and dropped drastically at higher SST (Fig. 3b). Several studies with phytoplankton monocultures have reported positive dependence of specific isoprene production rates on temperature 45,47-50 . One of these studies 45 described that the increase with temperature reaches an optimum for production that varies among phytoplankton strains and with light intensity, but falls around 23-26°C. The most detailed study 47 was conducted with a Prochlorococcus strain; remarkably, the shape of the specific production rate vs. temperature curve for this cyanobacterium strain was almost identical to that of Fig. 3b, with an exponential increase until 23°C and a drop thereafter. This is the canonical curve type of enzymatic activities, but the thermal behaviour of the enzymes for isoprene synthesis in marine unicellular algae has not yet been characterised 12 .
Revising the magnitude and players of the marine isoprene cycle. Our results allow redrawing the isoprene cycle in the surface mixed layer of the ocean. Figure 4 sketches the magnitude of the rate constants for production and sinks presented in Table 1, averaged according to a chla concentration threshold: the blue and green arrows correspond to the experiments in waters with [chla] lower and higher than 0.4 mg m −3 , respectively. Isoprene production in productive (chla-richer) waters is faster than in oligotrophic (chla-poorer) waters. Vertical mixing is assumed to majorly constitute an input into the mixed layer, yet very small. Photochemical production and emission from surfactants 15 in the surface microlayer of productive waters is depicted as uncertain. Among sinks, the microbiota-dependent consumption is much faster in productive waters; actually, the statistical uncertainty of Eq. (1) and the uneven distribution of incubation results along the [chla] axis hamper resolving k bio in phytoplankton-poor waters (<0.4 mg m −3 ), which represent nearly 80% of the area of the global surface ocean as a monthly average. Here, k loss can be anything between 0.03 and 0.09 d −1 , and therefore k bio will be <0.04 d −1 . Putative purely chemical oxidation is considered invariant irrespective of the chla content; consequently, the combined microbial + chemical loss is much faster in productive waters. The k vent for ventilation to the atmosphere is not significantly different between the two groups because it depends on wind speed and SST, which are both independent of [chla].
Note that this comparison applies to process rate constants k (d −1 ), which represent the velocities at which processes occur and are attributable to biological and environmental agents. The actual process rates (nmol m −3 d −1 ) will result from multiplying each of these k by the isoprene concentration, [iso]. Even though [iso] tends to increase with [chla], there is no such a thing as a globally valid proportionality between the two 13,14,51 (Table 1). All in all, more isoprene is produced in productive waters but more is consumed as well; therefore, predicting the resulting effect on isoprene concentrations and air-sea fluxes is not straightforward.
Concluding remarks. Until now, most of the focus of isoprene cycling studies had been on the production term, considering specific production rates by phytoplankton as though they were constitutive and shaped by phylogeny 41 , with an occasional emphasis on how they are tuned by acclimation to environmental conditions 45,47,50 . Even though teasing apart phylogeny and acclimation at the cross-basin and seasonal scales is not an easy task because species and community succession are interlinked with environmental stressors, our results call for a deeper exploration of the ecophysiological drivers of isoprene biosynthesis by phytoplankton. As a matter of fact, whilst isoprene production is grosso modo related to phytoplankton biomass and primary production (Fig. 4), the resulting isoprene concentration does not necessarily follow indicators of phytoplankton biomass such as chla but it is further influenced by environmental factors such as SST [12][13][14]51 . In spite of the lack of a mechanistic explanation, we conclude that temperature plays an important role in governing chla-normalised isoprene production across regions of the open ocean. While expanding the lab-derived database of specific isoprene production rates across phytoplankton taxa is always desirable, we argue there is a need for in situ measurements under variable natural conditions if we are to reliably predict isoprene production in the ocean.
We also show that the loss terms in the cycle are more complex and variable than believed, with a microbiota-dependent sink that is tightly coupled to production and can dominate over ventilation in chla-rich waters (Fig. 4). Considering all sinks together (ventilation, biological and chemical loss and, on one occasion, vertical mixing), the resulting total turnover times of isoprene in the surface mixed layer of the open ocean are in the order of one or two weeks in oligotrophic waters but can be as short as 1 to 4 days in productive waters. The microorganisms and metabolic mechanisms involved in isoprene biological consumption 34 warrant further investigation because this important sink will be regulated by triggers of microbial speciation and activity, potential co-metabolisms, and microbial mortality by predators and viruses. Our results also indicate that chemical onsumption is more variable than estimated hitherto and has abiotic and biotic terms involving photochemically as well as biologically derived oxidants. All in all, isoprene concentration and emission to the atmosphere can no longer be regarded as controlled only by phytoplankton biomass and functional types, with fixed loss rates dominated by the physicochemical processes (air-sea exchange and oxidation), but rather intimately connected to the variable structure and dynamics of the pelagic microbial food web.

Methods
Sampling and physical measurements. Mediterranean coastal water samples were collected in March and May 2021 at the Blanes Bay Microbial Observatory site 52 , over a 20 m water column, and in August 2021 at a rocky pier of the Barceloneta beach (Barcelona), at 20 cm from patches of the seaweeds Dictyoma dichotoma and Corallina elongata. Surface (0.2 m) seawater was hand-collected from a boat or the rocks, and the temperature was recorded with a calibrated SAIVA/S SD204 CTD sensor. Samples were kept in the dark and taken to the ICM-CSIC lab within 3 h for incubations. The BIOGAPS-Moorea expedition took place in April 2018 at the northern coast of the island of Mo'orea, French Polynesia. Surface (0.2 m) seawater was hand-collected from a boat at two locations: in very shallow waters of the coastal coral reef lagoon, and 3.5 km offshore over a water column depth of 1100 m. Seawater temperature was recorded with an SBE56 sensor (Sea-Bird Sci.) continuously flushed with pumped-in surface seawater. Samples were taken to the Gump Research Station (University California Berkeley) on the island for processing. The HOTMIX cruise 53 traversed the Mediterranean Sea from East to West between 27 April and 29 May 2014 on board the R/V Sarmiento de Gamboa. Seawater was collected with a General Oceanics rosette, equipped with 24 L Niskin bottles. Temperature and salinity were recorded with an SBE911 + CTD system (Sea-Bird Sci.). The TransPEGASO cruise 40 crossed the Atlantic Ocean from North to South on the R/V Hesperides, between 20 October and 21 November 2014. Surface seawater was sampled using the ship's underway pumping system, which had the water intake located 4-5 m below sea level. All the parts of the centrifugal pump (BKMKC-10.11, Tecnium) that were in contact with the fluid were made of polypropylene and glass. Seawater temperature and salinity were recorded continuously via the flow-through thermosalinograph SBE21 SeaCAT (Sea-Bird Sci.). The PEGASO cruise 39 was conducted on board de R/V Hesperides in the regions of Antarctic Peninsula, South Orkney and South Georgia Islands from 2 January to 11 February 2015. Seawater samples were collected from either the underway pumping system intake (same as above) or the uppermost (4 m) bottle of the rosette on SBE911 + CTD casts, which recorded temperature and salinity. In PEGASO, Lagrangian tracking of the surface (15 m) water patch was conducted using WOCE (World Ocean Circulation Experiment) standard drifters provided with an Iridium communication system. Each drifter consisted of a spherical floatable enclosure that contained a GPS and an emitter, from which 10 m cylindrical drogues hung 5 m below the sphere. The drifters sent their position every 30 min, and all ship operations were conducted next to them 54 Fig. 2).
Incubation experiments. For loss kinetics experiments with coastal water (Fig. 1a and Supplementary Table 1), duplicate all-glass bottles (2 L) were filled, leaving no headspace, and incubated in the dark in a water bath kept at the in situ temperature ±0.5°C. Aliquots for duplicate isoprene analysis were taken from both bottles at t 0 and 2-4 further time points, never removing more than 15% of the bottle volume in total. Since opening the bottles caused the loss of the headspace, at each time point isoprene concentrations were corrected for this loss using the volumetric headspace:water ratio and the dependence of the isoprene Henry's law constant on temperature 55 : where H cp is the Henry solubility, T is the incubation temperature in Kelvin, H cc is the dimensionless Henry solubility, [iso a ] and [iso w ] are the isoprene concentrations in air (headspace) and water, respectively, and R is the gas constant.
For the chemical oxidation assay ( Fig. 1b and Supplementary Table 1), coastal water was filtered sequentially through polycarbonate 3 and 0.2 µm pore-size filters, and the filtrate was used to fill four series of four gas-tight glass 30 mL vials. A working bromoperoxidase (BrPO) solution was prepared by dissolving 10 units of Corallina officinalis BrPO (Merck) with 100 µl of HEPES (4-(2-hydroxyethyl)−1piperazineethanesulfonic acid) buffer 0.1 mol L −1 in 900 µl of MilliQ water. One of the series of vials had only filtered water, another was added 10 µmol L −1 of H 2 O 2 (Merck), another was added BrPO to a final concentration of 0.0025 units mL −1 , and another was added both H 2 O 2 and BrPO. Together with a series of non-filtered water, they were all incubated in the dark in a water bath kept at the in situ temperature (24 ± 0.5°C). The initial isoprene concentration was measured from the filtrate before filling the vials, and t 1 , t 2 , t 3 , and t 4 concentrations were measured each sacrificing one of the vials.
For the 11 offshore experiments (Fig. 3, Supplementary Fig. 1 and Supplementary Table 2), duplicate all-glass bottles (0.5 L) were filled, leaving no headspace. One of the bottles was analysed in duplicate for isoprene to set the initial concentration. The other bottle was not opened and dark-incubated for 24 h in a tank with constant flushing of pumped-in surface ocean water, to keep incubation temperature the same as in situ. At the conclusion of the incubation time, isoprene concentration was analysed in sequential duplicates, each with its exact incubation time. The sample PEGASO B3 was incubated for only 9 h. The loss rate constant of isoprene k loss (d −1 ) was calculated as the slope of the natural logarithm of the concentration vs. time, under two assumptions: (a) consumption follows first-order kinetics, as in the case of other trace gases such as dimethyl sulfide 56 and methyl halides 57 ; coastal water experiments showed that first-order was a better approximation than zero-order (linear) kinetics ( Fig. 1a and Supplementary Table 1); (b) isoprene production by phytoplankton, which is linked to photosynthesis 25 , is mostly arrested over the 24 h of incubation in the dark; if biosynthesis resumed for a while in the dark 33 , this would have reduced apparent loss and would have caused underestimation of k loss . The error of k loss was the standard error of the slope of Ln(concentration) vs. time.
Isoprene concentration. Isoprene was measured along with other volatile compounds on a gas chromatography-mass spectrometry system (5975-T LTM GC/ MS, Agilent Technologies). Aliquots of 25 mL were drawn from the glass bottle with a glass syringe with a Teflon tube and filtered through a 25 mm glass fibre filter while introduced into a purge and trap system (Stratum, Tekmar Teledyne). The sample was purged by bubbling with 40 mL min −1 of ultrapure He for 12 min while heated to 30°C. The stripped volatiles were trapped on solid adsorbent (VOCARB 3000) at room temperature and thermally desorbed (250°C) into the GC. Isoprene monitored as m/z 67 in selected ion monitoring mode, had a retention time of 2.4 min in the LTM DB-VRX chromatographic column held at 35°C. The detection limit was 1 pmol L −1 , and the analytical precision was 5%. In HOTMIX, TransPEGASO and PEGASO, calibration was performed by injections of a gaseous mixture of isoprene in N 2 . In BIOGAPS-Moorea and coastal water experiments, a liquid standard solution prepared in cold methanol and subsequently diluted in MilliQ water was used instead.
Isoprene ventilation rate constant. The isoprene ventilation or air-sea exchange fluxes (F vent , in nmol m −2 d −1 ) were calculated as: where [iso w ] is the isoprene concentration in surface seawater, [iso a ] is the isoprene concentration in the air, K H is Henry's Law constant for isoprene, and k AS is the gas exchange velocity (cm h −1 ). Air-side isoprene can be considered near zero and neglected for flux calculations because isoprene is highly reactive in the atmosphere, and it is largely supersaturated in the surface ocean. k AS was estimated using 58 : where U 10 is the wind speed at 10 m (m s −1 ), and Sc is the Schmidt number (nondimensional). On cruises, the wind speed was measured by the ships' meteorological stations and averaged over a period of 24 h, which was the duration of the incubations. In offshore Mo'orea, we recorded wind speed on the boat with a portable Skywatch BL500 micrometeorological station. This instantaneous wind speed was converted to the daily average by applying the factor between instantaneous and daily average wind speeds measured at the Gump Station onshore. Sc was computed 18 as: Sc ¼ 3913:15 þ 162:13 x SST þ 2:67 x SST 2 þ 0:012 x SST 3 where SST is in degrees Celsius (°C). The error of the computed ventilation fluxes is estimated 58 to be 20%. To convert the flux F vent (nmol m −2 d −1 ) into the ventilation rate constant k vent (d −1 ), the flux was divided by the mixed layer depth (Z ML , m) and by the isoprene concentration (nmol m −3 ), assuming that the surface concentration was equal to the average concentration in the mixed layer 14 . During HOTMIX and PEGASO, Z ML was determined from CTD profiles as the depth at which density was 0.125 kg m −3 higher than that at 5 m. In the case of the TransPEGASO cruise and the BIOGAPS-Moorea expedition, where no CTD casts were conducted, we used the geolocalised monthly values from a global climatology 59 .
Isoprene vertical mixing rate constant. In the case of the three PEGASO samples (sampling sites #9-11), k mix was estimated from measured vertical profiles of isoprene concentration and the turbulent diffusion across the pycnocline (K z ). Thus, the vertical mixing flux at the bottom of the ML (F mix , nmol m −2 d −1 ) was calculated as: where a K z = 2.6 m 2 d −1 (or 0.3 cm 2 s −1 ) was considered appropriate yet conservative for the Southern Ocean 60 , Δ[iso] (nmol m −3 ) was the isoprene concentration gradient across the upper pycnocline, and Δz (m) was the distance covered by this gradient. Depending on the location of the concentration maximum, F mix was positive (loss term, export from the ML) or negative (gain term, import into the ML). k mix (d −1 ) was calculated by dividing F mix by the surface isoprene concentration and Z ML (determined from the CTD profiles as above). We note that using a wider range of K z for Southern Ocean locations 61,62 (0.1-1.0 cm 2 s −1 ) would give 0.3-3 times the estimates of the k mix at sampling sites #9-11; however, the contribution of k mix to the calculation of k prod is so small (compared to those of the other sinks) that the effect of the K z range on k prod was only noticeable in #11 (±8% of k prod ). For HOTMIX, TransPEGASO and BIOGAPS-Moorea, k mix could not be estimated from in situ data and a fixed value of −0.005 d −1 was taken from the global integral suggested by a model 17 .
Chlorophyll a concentration. Seawater 250-mL samples were filtered on glass fibre filters, which were extracted with 90% acetone at 4°C in the dark for 24 h. The fluorescence of extracts was measured with a calibrated Turner Designs fluorometer.
Bacterial abundance. Aliquots of 2 mL of the initial sample were fixed with 1% paraformaldehyde plus 0.05% glutaraldehyde and stored frozen at −80°C. The numbers of heterotrophic bacteria were determined by flow cytometry (Cube 8, Partec) after staining with SYBR-Green 63 .

Data availability
All data needed to evaluate the conclusions are available in https://doi.org/10.5281/ zenodo.5794234.