Adaptive carbon export response to warming in the Sargasso Sea

Ocean ecosystem models predict that warming and increased surface ocean stratification will trigger a series of ecosystem events, reducing the biological export of particulate carbon to the ocean interior. We present a nearly three-decade time series from the open ocean that documents a biological response to ocean warming and nutrient reductions wherein particulate carbon export is maintained, counter to expectations. Carbon export is maintained through a combination of phytoplankton community change to favor cyanobacteria with high cellular carbon-to-phosphorus ratios and enhanced shallow phosphorus recycling leading to increased nutrient use efficiency. These results suggest that surface ocean ecosystems may be more responsive and adapt more rapidly to changes in the hydrographic system than is currently envisioned in earth ecosystem models, with positive consequences for ocean carbon uptake.

The integrated DOP pool did not significantly change, increase or decrease, from 2004-2010 (Supplementary Figure 2). Throughout the 2010's decade, DOP inventories decreased significantly (Model 1 Linear Regression, R 2 = 0.10; slope = -0.254 ±0.057 mmol m -2 y -1 ; P < 0.001. Rate of decline in inventory is 2-4% y -1 ). stoichiometric ratios differ in their magnitude between the upper and lower euphotic zones. C:P and N:P ratios are lower in magnitude in the lower euphotic zone, consistent with the notion that vertical nutrient inputs are important and that phytoplankton in these lower depths assimilate the upward flux of nutrients resulting in a less nutrient stressed condition than their counterparts in the upper euphotic zone. Seston stoichiometric ratios for the lower euphotic are given in the main manuscript ( Figure 4 in the primary text), and the upper euphotic zone stoichiometric ratios are presented here (Supplementary Figure 5). C:P ratios in the lower euphotic zone increased significantly throughout the 2010's (Model 1 regression, P <0.001, slope = 14.6 ± 3.0 units y -1 ), but not during the years before. N:P ratios in the lower euphotic zone increased marginally (Model 1 regression, P = 0.09, slope = 0.8 ± 0.5 units y -1 ), but not during the years before. Living Phytoplankton Carbon. We assessed the contribution of phytoplankton carbon to total POC stocks, by calculating the ratio of our flow cytometry phytoplankton carbon estimate (FCM PhytoC) to total POC. Quantification of total phytoplankton POC is described in the summary methods in the primary text. Contributions of phytoplankton POC ranged from 0.11-0.71 and averaged 0.30 ± 0.11. There was no significant change in the contribution before or after 2010.  (Supplementary Figure 7a). In contrast, the relative biomass (>5mm biomass as a fraction of total biomass >0.2mm) did not change significantly (Supplementary Figure 7b).
During summer cruises conducted as part of the Trophic BATS program in 2011 and 2012, crustacean zooplankton bodies and zooplankton and salp faecal pellets produced during onboard incubations were provided by Dr. Stephanie Wilson and analyzed for their C, N, and P content. The POC:POP ratio in salp faecal pellets was substantially greater than the POC:POP ratio in suspended particulate matter and was more similar to the POC:POP ratio in POM captured in 150 m surfaced tethered traps from these cruises (Supplementary Figure 8). This is in contrast to the crustacean zooplankton, which had a POC:POP ratio in their body tissue lower than the averaged POC:POP ratio in suspended matter and produced fecal pellets that were even more enriched in P than body tissue. Trait-based model. The phytoplankton stoichiometry model of Inomura et al. 9 is a conceptually simple mechanistic model that facilitates the accurate computation of phytoplankton C:N:P under different environmental conditions. The input variables are light intensity, growth rate, and the presence/absence of limiting nutrients. The model is based on four empirically supported lines of evidence: (1) a saturating relationship between light intensity and photosynthesis, (2) a linear relationship between the RNA-to-Protein ratio and growth rate, (3) a linear relationship between biosynthetic protein and growth rate, and (4) a constant macromolecular composition of the light-harvesting machinery.
Inomura et al. 9 calibrated the model parameters subject to constraints from published chemostat studies for multiple key marine and freshwater phytoplankton species. We used the model parameter set for the cyanobacteria Synechococcus linearis because cyanobacteria such as Synechococcus and Prochlorococcus are the most abundant phytoplankton types at BATS. However, as this species of Synechococcus is a freshwater species, we separately obtained the parameter related to maximum C:P at the zero-growth rate from a chemostat experiment for marine Synechococcus species WH8102 ( 10 ). Supplementary Figure 9 shows that the traitbased model accurately captures the C:P of P-limited Synechococcus at different growth rates as long as the growth rate is less than 0.8 d -1 . As the typical seasonal maximum growth rate observed at BATS rarely exceeds 0.8 d -1 (Figure 4D), we believe that this will not undermine the model's ability to predict phytoplankton C:P ratios at BATS.
The modeled growth rate estimates did not show a significant trend before or after 2010, and rates in the two periods were not statistically different. We compared the modeled growth rate to growth rate estimates based upon direct observational measurements ( Figure 4D in the primary text). Modeled growth rates were not significantly correlated (Pearson Product Moment Correlation) with either observational growth rate estimate, although the two observational estimates were well correlated with each other (Supplementary Table 1). The modeled growth rate (mean ± Std. Err.: 0.52 ± 0.014 d -1 ) was significantly greater (one-way ANOVA, P<0.05) than both the flow cytometry carbon-based growth rate (0.38 ± 0.015 d -1 ) and the C:Chlorophyllbased growth rate estimate (0.46 ± 0.14 d -1 ). The likely reason for the modeled growth rates being higher than the observation-based growth rates is that the model was developed for Synechococcus, while the observations include carbon associated with a wide diversity of eukaryotic phytoplankton.  11 , we drove the Inomura model with satellite-derived phytoplankton growth rate, Chl:C ratio of phytoplankton biomass, and P limitation [as the difference between satellite-derived SST and cubit root-corrected phosphate depletion temperature (PDT3)] to estimate phytoplankton C:P in the surface layer. In doing so, we made three modifications to the original stoichiometry model by Inomura et al. 9 following the protocol by Tanioka et al. 11 . First, we used depth-integrated POC:Chl-a from the satellite ocean color instead of calculating POC:Chl-a as a function of photon-flux density. Second, we imposed a fixed maximum growth rate of 2 d -1 in calculating C:P, which is equal to the maximum growth rate imposed on the satellite-based estimates of growth rate 12 . Third, we imposed a constant C:P value of 102 under the P-replete condition regardless of the P supply.

Supplementary
We used monthly and area-averaged satellite (MODIS Aqua) derived variables for a 3-by-3 pixel area around BATS to predict phytoplankton C:P in the top 100 m between 2003 and 2020. All satellite-derived input data and estimates of mixed-layer depth are available for download from the Oregon State Ocean Productivity Website [http://sites.science.oregonstate.edu/ocean.productivity/index.php (last access: June 21, 2021)].
The original phytoplankton stoichiometry model codes are publicly available (http://doi.org/10.5281/zenodo.4414338), and the modified version used in this study is available upon request.
Supplementary Figure 9. Model-data comparison of C:P of Synechococcus for various growth rates. Curves: Trait-based model C:P under different light intensities (mol m -2 s -1 ). Circles: laboratory data for C:P of Synechococcus WH8102 grown under P-limited chemostat culture at a photon flux density of 195 mol m -2 s -1 (Garcia et al. 2016). The open blue circle is the experimentally derived maximum C:P of 335 at the zero-growth rate. The dashed line represents C:P at maximum growth rate ( ) at various light intensities; high for higher light intensity.