SUPPLEMENTARY INFORMATION

Transcription

1 doi: /nature12448 Contents Page Supplementary Discussion 2 1. Coccolith model for carbon isotopic fractionation 2 A. Model configuration and evaluation of parameters 2 A1. Equations and fractionation factors 2 A2. Model parameterization of C fluxes 3 A3. Main drivers of variation in coccolith A4: Constraints on cellular uptake of HCO 3 and intracellular 6 HCO 3 allocation 6 B. Simulating coccolith and p in coccolithophores from culture using an inverse model 7 C. Consequences for Cenozoic expression of wide range of sizerelated vital effects 8 D. Oxygen isotopic fractionation in coccoliths 9 2. Coccolith preservation 9 3. Coccolithophore depth habitats 11 Full Methods Sampling and site selection 12 A: Sites U1334 and 1090 (EoceneOligocene transition) 12 B: Site 1264 (OligoceneMiocene transition) 13 C: Sites 999 and 1088 (MiocenePliocene) Age models Coccolith separations Sr/Ca productivity records and SST corrections 15 Supplementary Figures 17 Figure S1: Schematic of a coccolithophore cell showing fluxes and isotopic compositions 17 Figure S2. coccolith and p measured in cultures as a function of dissolved [CO 2 ] 18 Figure S3. Principal drivers of variation in coccolith 19 Figure S4. Active fluxes in inverse model solutions under different assumptions of cellular (cytosol) HCO 3 uptake 20 Figure S5: SST correction of Sr/Ca records 21 Figure S6: SEM images of coccoliths from Site Figure S7: SEM images of coccoliths from Site Figure S8: The effect of overgrowth 24 Figure S9: Map showing the location of all (I)ODP sites 24 Figure S10: Coccolith 18 O versus 13 C for Cenozoic time windows at various sites 25 Figure S11: MioPliocene age models 26 Figure S12: pco 2 values as a function of SST for different [CO 2aq ] scenarios 27 Supplementary Tables 28 Table S1: Notation used in numerical model and Figure S1 28 Table S2: Fractionation factors applied in numerical model 29 Table S3: Species composition of size fractions in endmember samples from Sites 999 and pco 2 data sources and uncertainty estimates (Fig. 3 main text) 31 Supplementary References

2 RESEARCH SUPPLEMENTARY INFORMATION Supplementary Discussion 1. Coccolith model for carbon isotopic fractionation We develop a model for Active Carbon Transport and Isotopes in COccolithophores, termed ACTI CO. The model simulates the active and passive carbon fluxes to and within the cell and their effects on the carbon isotopic fractionation into organic matter and calcite, p and coccolith respectively. We apply this model as an inverse problem to derive the active bicarbonate (HCO 3 ) fluxes required to simulate p and coccolith from three differentsized coccolithophore species grown in culture at variable aqueous CO 2 concentrations, [CO 2 ] 12,15, and the relationship of these fluxes to cell size and [CO 2 ]. The purpose of the model is not to derive the absolute fluxes responsible for isotopic fractionation, which is uncertain due to uncertainty in parameters such as membrane permeability. Rather, we evaluate how the relative fluxes must vary as a function of [CO 2 ] and cell size in order to simulate the observed p and coccolith. To this end, we adapt a model for the 13 C composition of photosynthetically fixed carbon of diatoms 31 by adding a module for the coccolith vesicle, which permits us to simulate coccolith as a function of the passive and active carbon fluxes into the coccolith vesicle and cell, as well as p. The multicompartment cellular approach is similar to that employed in references 32 and 33. A. Model configuration and evaluation of parameters A1. Equations and fractionation factors A schematic of the key cellular fluxes is given in Figure S1 and includes the potential for active HCO 3 and passive diffusive CO 2 fluxes into the cell, into the chloroplast, and into the coccolith vesicle (hereafter CV). When the relevant carbon fluxes to and within the cell are specified, the isotopic composition of calcite in the CV (as well as that of photosynthetically fixed organic carbon) may be obtained by solving the six mass balance equations below for the isotopic composition of CO 2 and HCO 3 in the cytosol, chloroplast, and CV. Annotations of the fluxes are illustrated in Figure S1 and notation detailed in Table S1; for simplicity we retain the same system of notation as in reference 31. (1) Chloroplast CO 2 : (2) Chloroplast HCO 3 : 2

3 RESEARCH (3) Coccolith vesicle CO 2 : (4) Coccolith vesicle HCO 3 : (5) Cytosol CO 2 : (6) Cytosol HCO 3 : The principal fractionations employed are summarised in Table S2 and include: 1) The thermodynamic isotopic fractionation between CO 2 and HCO 3 2) The kinetic fractionation for the hydration of CO 2 and dehydration of HCO 3 where this is catalysed by the enzyme carbonic anhydrase (CA) 3) The fractionation of carbon fixed into organic matter by Rubisco 4) The thermodynamic isotopic fractionation between HCO 3 and calcite Following previous workers 31, we assume that the hydration of CO 2 and dehydration of HCO 3 in the cytosol and chloroplast occur in the presence of CA. For the CV we conduct sensitivity analyses, in some simulations assuming CAkinetic fractionation and in others imposing the thermodynamic fractionation of hydration of CO 2 and dehydration of HCO 3 in the CV. We assume that thermodynamic fractionation establishes the isotopic equilibrium between HCO 3 and CO 2 in the extracellular medium. Because the fractionation between HCO 3 and calcite is small (Table S2), coccolith is closely approximated by coccolith 13 C minus dissolved inorganic carbon 13 C of seawater. As described previously for the twocompartment model 31, all of the terms are linear in the unknowns, enabling the system of equations to be solved with linear algebra. A2. Model parameterization of C fluxes Diffusive CO 2 fluxes across cellular membranes are established from CO 2 concentration and relevant CO 2 transfer coefficients. We assume that membrane permeability to CO 2 is similar to that measured for diatoms 31 and calculate the CO 2 transfer coefficients using the cellular surface area of the coccolithophores, taken from microscope measurements of cell radius 12,34 or by calculating relevant coccolithophore cell radius from measurements of cellular carbon quotas according to regressions established by references 35 or 36, whose estimates differ by less than 15 % for the 3 3

4 RESEARCH SUPPLEMENTARY INFORMATION range of cell sizes modelled here. While there is uncertainty in the absolute value of the cell membrane permeability, reasonable variations in membrane permeability alter the absolute fluxes inferred from the model but not the relative trends in fluxes as a function of [CO 2 ] or cell size, which are the focus of the current study. Appreciable deviations between proximalcell and bulk solution concentrations of CO 2 are recognized for plankton cells or aggregates in the range of 25 to 200 m, but are minor for the nonchainforming small (315 m) coccolithophores considered here in both bloom and steadystate conditions 37. Furthermore, transfer coefficients are derived from measurements of cellular fluxes compared to bulk media CO 2 concentrations, and therefore intrinsically account for such boundary layer effects. Transfer coefficients across chloroplast or CV membranes must be proportional to those across the cell membrane, but smaller due to the lower surface area of the intracellular membranes and, in the case of the chloroplast, further reduced by the evolution of characteristics to reduce CO 2 loss from the pyrenoid or chloroplast. Reference 32 define f as the ratio of transfer coefficient across the chloroplast to that of the entire cell and we retain this approach to define f chl for the chloroplast and add an equivalent f cv for the CV. We estimate a reasonable value for the f chl by applying a series of published coccolithophore p measurements 15,38 in the model of reference 32, which yield f chl estimates of 0.01 to 0.03; here we employ These f chl estimates would be slightly larger if we have overestimated the cell membrane permeability. In initial evaluations we determined that the trends in model fluxes are insensitive to f chl in the range of 0.02 to 0.05, but measured p cannot be simulated for several of the experimental points if we employ f chl of 0.10 or larger. As reference 31 point out, the effective lower permeability of the chloroplast membrane to CO 2 may arise not simply because of the membrane itself but because of strong effective tortuosity in the protein network of the pyrenoid. We follow the convention of reference 31 of simulating this in the twocompartment model by a lower effective chloroplast permeability. We include no active transport of CO 2 across the cell or chloroplast because previous studies document that this is a much less efficient strategy for raising CO 2 concentration in the chloroplast compared to HCO 3 uptake to the chloroplast 31. The model therefore calculates absolute CO 2 fluxes into the chloroplast by multiplying f chl by the total diffusive flux into the cell (i.e. that coming across the cell membrane). Because the CO 2 efflux from the chloroplast and the CV are solved by mass balance, and, as documented by reference 32, depend on the transfer coefficient and concentration of CO 2 in the chloroplast and CV, it is possible to estimate the concentration of CO 2 in the chloroplast and CV for each simulation. 4

5 RESEARCH Because HCO 3 rather than CO 2 is the substrate for calcification, the CV membrane may not have evolved to preclude diffusive CO 2 loss and may have a higher effective permeability to CO 2 than the chloroplast. The CV attains large dimensions in the cell, as final coccoliths have lengths comparable to the cell diameter 34. Due to the tight folding of the CV membrane around the forming coccolith, CV surface area is only slightly less than that of a sphere of equivalent diameter 39,40. This indicates that the surface area of the CV may be on the order of 4060 % of that of the cell, so the f cv could be 0.4 to 0.6; here we employ 0.4 and evaluate sensitivity to less permeable CVs of f cv =0.2 and 0.1. Initial simulations indicate no sensitivity to f cv =0.4 versus f cv =0.6. As described above for the chloroplast, the model therefore calculates absolute CO 2 fluxes into the CV by multiplying f cv by the total diffusive flux into the cell (i.e. that coming across the cell membrane). Previous studies 31 showed that passive HCO 3 fluxes across membranes are negligible; therefore we infer a negligible efflux of HCO 3 out of the chloroplast, CV and cytosol. Active transport of HCO 3 into the cell and into the chloroplast may contribute to carbon concentrating mechanisms for photosynthesis, and active transport of HCO 3 into the CV may contribute to calcification. We adjust the active fluxes of HCO 3 that would be required to simultaneously match observed p and coccolith in cultures. Hydration and dehydration fluxes, as well as effluxes of CO 2 from the cell, chloroplast, and CV, are all solved by mass balance constraints. Following previous workers 12,15,4143, coccolith calcification in the model is described by the reaction: 2HCO 3 + Ca +2 CaCO 3 + CO 2 This reaction results in a minimum requirement of a flux of 2 moles of HCO 3 in the CV for each mole of CaCO 3 produced. The HCO 3 may come from direct transport of HCO 3 into the CV and/or from hydration of CO 2 in the CV (Fig. S1). Through a large array of sensitivity studies, we find that, in the absence of CA in the CV, the rate of CO 2 hydration required to simulate typical coccolith carbon isotopic composition is several orders of magnitude greater than the uncatalysed rate of hydration. However, inferred rates are compatible with CAcatalysed rates measured in diatoms 31. Therefore we conclude that CA is required in the CV. A role for CA in the CV has been inferred from the upregulation of EhCA2 in calcifying Emiliania huxleyi compared to noncalcifying E. huxleyi 44,45. In all simulations, the presence of CA in the CV also necessitates a flux of HCO 3 into the cytosol in order to match observed coccolith. The model is developed uniquely to track the cellular allocation of carbon, and not fluxes of hydrogen (e.g., ph) or calcium, because inverse simulations of the carbon isotopes are not able to provide 5 5

6 RESEARCH SUPPLEMENTARY INFORMATION constraints on these latter fluxes. Therefore the model does not resolve the saturation state in the coccolith vesicle with respect to calcite. A3. Main drivers of variation in coccolith The model reveals that coccolith is strongly influenced by 1) the rate of CV HCO 3 uptake relative to calcification, 2) whether the HCO 3 actively taken up by the CV and chloroplast is sourced from hydration of CO 2 in the cytosol or from direct uptake across the cell membrane, and 3) to a lesser extent, the allocation of HCO 3 between the chloroplast and the CV. Increasing contribution of HCO 3 transported into the CV results in a steep increase in coccolith (i.e. heavier 13 C) (Fig. S3a and d). This is the principal driver of variation in coccolith and the relationship is independent of the source of the HCO 3 in the cytosol. For example, regardless of whether the HCO 3 taken up by the cell is a constant, increasing, or decreasing fraction of active uptake into the CV and chloroplast, the slope of the relationship between the proportion of calcification supported by CV HCO 3 uptake and coccolith is constant. In simulations with a constant contribution of CV HCO 3 to calcification, the source of HCO 3 in the cytosol does influence coccolith. coccolith increases as a greater fraction of the total HCO 3 transported into the chloroplast and CV is derived from direct uptake across the cell membrane rather than from hydration of CO 2 within the cytosol (Fig. S3b). Finally, in simulations with both a constant contribution of CV HCO 3 to calcification and a constant contribution of direct HCO 3 uptake across the cell membrane, a greater allocation of HCO 3 to the CV rather than the chloroplast results in a modest decrease in coccolith. This dependence is obliterated when the increasing allocation of HCO 3 to the CV is accompanied by an increase in the calcite to organic carbon (PIC/POC) ratio (Fig. S3c and e). A4. Constraints on cellular uptake of HCO 3 and intracellular HCO 3 allocation Application of the model in inverse form requires parameterisation for the flux of HCO 3 to the cytosol, in order to derive the other two active fluxes, the flux into the CV and the flux into the chloroplast. The flux of HCO 3 to the cytosol cannot be neglected. First, the presence of CA in the CV necessitates a flux of HCO 3 into the cytosol in order to match the range of observed coccolith. Secondly, an active flux of HCO 3 into the CV at least as large as the calcification flux is required to match observed coccolith. Because it is not possible to independently constrain all three fluxes, we evaluate ways to define the cytosol HCO 3 flux as a function of either or both of the other HCO 3 6

7 RESEARCH fluxes to the chloroplast and the CV. For comparison, we also evaluate a scenario with a cytosol HCO 3 influx that is constant proportion of the average photosynthetic fixation rate for each species. Independent constraints exist to guide our parameterisations of the cytosol HCO 3 flux. Direct cellular HCO 3 uptake is inferred for E. huxleyi from studies of inhibition of bicarbonate transporters 46. In 14 C disequilibrium studies with E. huxleyi 47, uptake of HCO 3 is observed to contribute a decreasing fraction of fixed carbon with increasing [CO 2 ]. To reflect this observation in our model, we evaluate scenarios in which the flux of HCO 3 to the cytosol is a) equal to or proportional to the sum of the HCO 3 transport into the CV and chloroplast, b) equal to or proportional to the HCO 3 flux into the CV, c) equal to the HCO 3 flux into the chloroplast, or d) constant. Calcification likely has a longer evolutionary history of HCO 3 requirement than photosynthesis, since the latter requires significant HCO 3 only at low [CO 2 ]. Therefore, we currently favour the assumption of HCO 3 uptake to the cytosol proportional to HCO 3 uptake to the CV or proportional to uptake into both the CV and chloroplast (options a or b above). Nonetheless, additional culture experiments and analysis of disequilibrium experiments will be required to assess which model most closely approximates the response of these coccolithophores. If one set of assumptions led to appreciably higher total transport fluxes of HCO 3 to match typical coccolith, we might judge it a less favourable hypothesis on the grounds that it would entail higher energetic cost. Yet scenarios a), b), and c) yield similar total transport fluxes of HCO 3. B. Simulating coccolith and p in coccolithophores from culture using an inverse model Measurements of coccolith and p from cultures of Gephyrocapsa oceanica and Coccolithus pelagicus subsp. braarudii 12 as well as cultures of E. huxleyi grown at similar daylight duration and irradiance 15 (Fig. S2) show a dramatic 3 decrease in coccolith of C. braarudii as [CO 2 ] decreases from 60 to 12 M. This contrasts with the more constant composition of coccolith in G. oceanica. The large differences in cell size among these three species lead to major differences in the diffusive CO 2 flux into the cell relative to cellular carbon inventory (Figure S2c). The low surface area to volume ratio of the larger cells leads to a smaller ratio of diffusive CO 2 supply to cellular carbon inventory. At low [CO 2 ] (<20 µm), this diffusive limitation in large cells is accentuated, and is accompanied by increasingly depleted coccolith with decreasing [CO 2 ] (Figure S2b). Intracellular fluxes of HCO 3 are expected to be driven by the extent of diffusive CO 2 limitation. Using the model, we evaluate what active fluxes of HCO 3 are required to match the measurements of coccolith and p from cultures 12,

8 RESEARCH SUPPLEMENTARY INFORMATION Regardless of the parameterisation of the cytosol HCO 3 fluxes (described in A4), all inverse simulations derive an increase in the proportion of calcification supported by HCO 3 transport to the CV with increasing [CO 2 ] (Fig. S4a). In all experiments with [CO 2 ] <40 M, the model indicates that the fraction of calcification supported by HCO 3 transport to the CV is larger in small cells of E. huxleyi and G. oceanica than in larger cells of C. braarudii. The changes in relative contribution of CV influx of HCO 3 to calcification are a dominant contributor to the observed pattern of size and CO 2 related vital effects. Inverse models also consistently reveal that at low [CO 2 ] (<20 M), the ratio of CV HCO 3 transport to cellular diffusive CO 2 transport is much lower for large cells than for small cells (Fig. S4e). The overall pattern of HCO 3 fluxes is suggestive of competition for HCO 3 allocation between the chloroplast and CV, with transport to the chloroplast given priority at low [CO 2 ]. All inverse modelling solutions show that the ratio of chloroplast HCO 3 transport to cellular diffusive CO 2 transport increases exponentially with decreasing [CO 2 ] (Fig. S4b), as diffusive supply is progressively less able to support optimal photosynthetic rates. In addition, the proportion of fixation supported by chloroplast HCO 3 transport increases or remains constant with decreasing [CO 2 ].(Fig. S4c). These relationships reflect the high importance of active (HCO 3 ) carbon concentration at the site of photosynthesis at very low [CO 2 ]. With decreasing [CO 2 ], the ratio of CV HCO 3 transport to chloroplast HCO 3 transport decreases (Fig. S4d). At low 1015 M [CO 2 ], this ratio is only 3040% of the flux ratio at a [CO 2 ] of 60 M. As noted above, the proportion of calcification supported by HCO 3 transport to the CV decreases with decreasing [CO 2 ]. This evidence of competition for HCO 3 allocation between the chloroplast and CV, with transport to the chloroplast given priority at low [CO 2 ], is also supported by the response of calcification and photosynthesis in culture experiments. There, under low CO 2 and HCO 3 availability, HCO 3 was increasingly diverted to the chloroplast for photosynthesis 16. The recent finding of less heavily calcified E. huxleyi during the summer period of higher ph and lower [CO 2 ] in the Bay of Biscay 48 may be a manifestation of this process. Our model provides no evidence that the CO 2 efflux from calcification reduces the need for active HCO 3 transport during photosynthesis 49, consistent with recent models confirming that enhanced CO 2 in the cytosol is not an effective mechanism for enhancing CO 2 at the site of photosynthesis 31,32. C. Consequences for Cenozoic expression of wide range of sizerelated vital effects The model elucidates the main mechanisms for vital effects in coccolith carbon isotopes and infers active carbon fluxes for cultured coccolithophores which are consistent with other evidence. The 8 8

9 RESEARCH late Miocene emergence of sizecorrelated vital effects leaves open an important question: why did the vital effects emerge in a relatively short interval of a few million years, whereas [CO 2 ] is inferred to have declined since at least the latest Eocene? Existing cultures provide no experimental data for the larger species at [CO 2 ] between 19 and 43 M, making it impossible to directly assess if the range of interspecific vital effects follows a linear or nonlinear relationship with [CO 2 ]. However the change in C. braarudii carbon isotopic composition is especially steep between 12 and 19 M, leaving open the possibility that the relationship between coccolith and [CO 2 ] is steepest in this [CO 2 ] range, which in a typical midlatitude ocean photic zone with a temperature of 20 C (Fig. S5) and a DIC concentration of 2050 M would correspond to pco 2 of 375 to 575 ppmv. A number of further adaptations at lower [CO 2 ] could restrict the further depression of coccolith in large species, including a reduction in the PIC/POC ratio which could help stabilise the ratio of CV HCO 3 influx to calcification against continued decline at low [CO 2 ]. At the highest [CO 2 ] of 66 M, culture data exhibit an inversion in vital effects, with larger cells of C. braarudii showing higher coccolith than small G. oceanica. If confirmed in further culture studies, this suggests sensitivity of vital effects at high [CO 2 ] and the small inversion in vital effects in our tropical Eocene record (Fig. 2, Fig. S10e) may potentially be a manifestation of this process. In the model, inversions can be produced by a higher relative CV HCO 3 to calcification ratio in larger cells than small cells, or more similar allocation among small and large cells of HCO 3 to the chloroplast and CV at high [CO 2 ]. D. Oxygen isotopic fractionation in coccoliths In both culture and coccoliths from sediments, the oxygen isotopic ratio is highly correlated with carbon isotopic ratios. This temperatureindependent oxygen isotopic fractionation in biogenic carbonates has been attributed in general to variation in the relative contribution of HCO 3 relative to the isotopically lighter CO 2 3 in the calcification space 50. This mechanism was detailed in the case of coccoliths in terms of the variable contribution of HCO 3 relative to isotopically lighter CO 2 3 in the coccolith vesicle as a function of media ph 17. Our carbon isotopic model suggests that a similar effect may arise because of the varying relative contributions of CO 2 and HCO 3 transported into the CV. 2. Coccolith preservation In order to verify that the late Miocene appearance of vital effects in coccolith calcite was not a result of diagenetic homogenisation in the old part of the Site 1088 and 999 records (versus good 9 9

10 RESEARCH SUPPLEMENTARY INFORMATION preservation in younger sediments where vital effects are present), we assessed coccolith preservation by scanning electron microscope (SEM) and considered any potential overgrowth effects on coccolith geochemistry. At Site 999A, calcareous nannofossil preservation is good to moderate from the top of the section down to core 31X (~1011 Ma), with moderate to poor preservation in cores below this interval 51. In the middle Miocene carbonate crash interval, ~1012 Ma, CaCO 3 contents are nearzero therefore our original sample from ~11 Ma was separated but could not be measured. Following the crash interval, CaCO 3 contents then recover to precrash values by 11.6 Ma 52. SEM images from a Pliocene sample at ~4 Ma and our oldest Miocene sample ~15 Ma (Fig. S6) show that delicate central structures such as grills and bars are preserved in some specimens (D and F in plate 1), and although some etching and fragmentation is visible in the Miocene coccoliths, overgrowth appears to be minor (GK in Plate 1). At Site 1088, calcareous nannofossil preservation is described as good or moderate throughout our entire study interval 53. SEM images (Fig. S7) show wellpreserved coccoliths, with some etching but with intact central structures in many specimens (Fig. S7 e and f). In Miocene samples from site 1088, the presence of entire fossilised coccospheres of Calcidiscus, Coccolithus and Reticulofenestra attests to the good CaCO 3 preservation (Fig. S7 J and K). Minor overgrowth is visible in the central area of some Reticulofenestra specimen (J in Fig. S7), however preservation is comparable in older versus younger samples. In the context of this study, we consider overgrowth of coccoliths (and not etching) to be significant to our interpretations based on isotopic differences between size fractions. We therefore performed some simple calculations to assess the potential effect of overgrowth on our coccolith δ 18 O records. At Site 999, the difference in equilibrium δ 18 O between calcite precipitated at the surface and on the sea floor is ~56, whereas at Site 1088, this surfacedeep equilibrium δ 18 O difference is on the order of ~34. Surfacedeep temperature gradient were calculated using sea surface temperatures (SSTs) as described for Sr/Ca corrections and bottom water temperatures 54, assuming 0.23 change in δ 18 O per degree C 55. Thus, any overgrowth will be significantly heavier than original coccolith calcite, even in the size fractions that are furthest from equilibrium δ 18 O (in samples where vital effects are present). Assuming uniform overgrowth on all coccolith size fractions and an original δ 18 O value of 1, 0.5 and 4 for large coccoliths, small coccoliths and overgrown calcite respectively, a 50% contribution from overgrown calcite to measured δ 18 O would halve the magnitude of an original 1.5 difference between small and large coccoliths. A >90% contribution from overgrown calcite to measured δ 18 O is required in order to completely homogenize an original

11 RESEARCH 1.5 δ 18 O difference (Fig. S8). We note that (1) altering the absolute δ 18 O values of the coccoliths or the overgrown calcite, or (2) changing the magnitude of the δ 18 O difference between small and large coccoliths, has no impact on results relating to the degree of overgrowth required to completely attenuate the original δ 18 O difference. Furthermore, the presence of significant amounts of overgrown calcite in samples would have a dramatic impact on measured Sr/Ca values. The partition coefficient for Sr in abiogenic calcite is more than one order of magnitude lower that the partition coefficient in coccoliths, in abiogenic calcite experiments 56 versus in cultured coccoliths 57. Thus, if abiogenic calcite contributed significantly to our sizeseparated samples before the emergence of vital effects (pre ~7 Ma), Sr/Ca values would be approximately one third of the values of ~23 mmol/mol measured at Sites 1088 and 999. Using the above Sr partition coefficients, we calculate that a 70% contribution from abiogenic calcite (i.e. the minimum overgrowth required to nearhomogenise an original 1.5 vital effect) would lead to a reduction in Sr/Ca from mmol/mol in pristine coccolith calcite to mmol/mol. We observe no stepincrease in Sr/Ca coincident with the appearance of vital effects at either site (Fig. S5). The only way such a shift would escape expression in coccolith Sr/Ca is if a 3fold increase in seawater Sr/Ca also occurred over this time, a highly unlikely scenario given that seawater Sr/Ca reconstructions indicate relatively stable Sr/Ca since the Miocene (reference 58 and references therein). In summary: (1) the gradual appearance of vital effects at Site 999 between ~7 and 5 Ma occurs significantly later that the documented improved preservation in sediments younger than 10 Ma, whereas at Site 1088 preservation is good and comparable throughout the MioPliocene. (2) SEM images show only minor overgrowth in some specimens and calculations of the impact of overgrowth on 18 O imply that an enormous amount of overgrowth (<10% of original calcite contribution to signal) is required to cause the complete homogenisation of coccolith 18 O between small and large size fractions. (3) Sr/Ca values are consistent with Sr partitioning into coccolith calcite rather than abiogenic calcite. Thus we conclude that the absence of isotopic differences in coccoliths from sediments older than ~57 Ma at both sites is not a preservational artefact. 3. Coccolithophore depth habitats Recent SST and thermocline depth reconstructions from the Pacific Ocean over the last 13 million years are interpreted as indicating a gradually shoaling global thermocline over the MioPliocene 28. One consequence of this scenario could be a gradual shift in the mean depth habitat of

12 RESEARCH SUPPLEMENTARY INFORMATION coccolithophores from deeper, more CO 2 rich waters (when the mixed layer is thick) to shallower, CO 2 poor waters (when the mixed layer is thin) throughout the MioPliocene. However, SST reconstructions derived from alkenones (organic compounds produced by coccolithophores) record similar SST trends and amplitude of cooling over the MioPliocene compared to other noncoccolithophorebased proxies, despite significant changes in inferred thermocline depth 28,5961. This suggests that significant secular shoaling of coccolithophore depth habitats over the MioPliocene was unlikely, given that such a change would have resulted in reduced cooling as recorded by alkenones relative to other proxies. Full Methods (also see section 1 for model description) 1. Sampling and site selection See Figure S9 for map of all site locations. A: Sites U1334 and 1090 (EoceneOligocene transition) To assess the possible emergence of coccolithophore vital effects during the EoceneOligocene (E O) transition, we studied sizeseparated coccoliths from samples representing late Eocene and early Oligocene background climate state at two distinct sites: one in the equatorial Pacific Ocean and one in the subantarctic Atlantic Ocean. IODP Site U1334 is located in the equatorial Pacific Ocean ( N, W) with a palaeolatitude estimated at ±2 of the equator. Site U1334 was drilled at a depth of 4799 metres below sea level, with palaeodepth of the ~38 Ma crust estimated at ~3500 m. Upper Eocene sediments are mostly composed of nannofossil chalk, radiolarite and claystone, while Oligocene sediments are primarily made up of nannofossil ooze and chalk. Carbonates are present throughout the EO transition 62. At Site U1334, we selected two endmember samples representative of background climate conditions before and after the EO transition, from m and m respectively (revised corrected metres composite depth), based on bulk sediment δ 18 O stratigraphy (P.A. Wilson, personal communication, 2012). ODP Site 1090 is located in the subantarctic Atlantic Ocean on the Agulhas Ridge ( 'S, 'E, water depth m). Sediments at Site 1090 are mostly composed of nannofossil mud and diatoms, with variable carbonate content. At ODP Site 1090, we generated bulk stable isotope data at 40 cm resolution over the interval 238 mcd to 270 mcd, chosen based on previous studies 63,64, to pinpoint exactly where the EO transition occurs at this site. Subsequently, we selected two samples (268.4 mcd and mcd) from prior to and following the EO transition for coccolith separations and isotope analysis. 12

13 RESEARCH B: Site 1264 (OligoceneMiocene transition) ODP Site 1264 is located in the southeastern Atlantic Ocean ( 'S, 'E) at a water depth of 2507 m. OligoceneMiocene sediments are primarily nannofossil and foraminiferabearing nannofossil ooze and chalk 65. We selected 17 samples for coccolith separation and geochemical analysis covering the OligoceneMiocene (OM) transition over the interval mcd to mcd with 50 cm resolution sampling, based on the benthic δ 18 O stratigraphy of reference 66 (our study interval: 23.8 to 22 Ma). Because no significant change in isotopic range between small and large coccoliths occurred during the OM time series, we express isotope values as mean values of the entire timeseries (Fig. S10) or as average values of the Miocene and Oligocene portions (2223 Ma: 9 data points; Ma: 8 data points) (Fig. 2). C: Sites 999 and 1088 (MiocenePliocene) To obtain a record of the evolution of carbon and oxygen isotopic differences in coccoliths over the late Mioceneearly Pliocene at two distinct sites, we studied sediment samples from lowlatitude ODP Site 999 and midlatitude ODP Site 1088 spanning the interval ~160 Ma. We analysed 12 samples from Caribbean Site 999 between 15 and 4 Ma (~126 mcd to 369 mcd) at approximately 1 Myr resolution, extending a higherresolution dataset previously generated from 3.5 to 2 Ma and the last glacial maximum 13. ODP Site 999 was drilled in the Columbia basin (12 44 N, W) at a modern water depth of ~2830 m. MioPliocene sediments primarily consist of clayey nannofossil sediment with some siliceous microfossils and terrigenous contributions from aeolian (including volcanic ash layers) and riverine (primarily the Magdalena River) sources 51. To obtain a higherlatitude perspective, we also analysed 20 samples from subantarctic ODP Site 1088 over the interval 15 to 1 Ma at ~0.51 Myr resolution (~10 mcd to 231 mcd). Site 1088 was drilled on the Agulhas Ridge, just south of the subtropical front (41 08 S, E) at a modern water depth of ~2082 metres 67. Sediments are carbonaterich (mostly nannofossil ooze) with intermittent siliceous microfossils. 2. Age models All age models are based on sitespecific benthic foraminiferal oxygen isotope (δ 18 O benthic ) stratigraphies where possible, supplemented with biostratigraphic data. The Site 999 age model is based on δ 18 O benthic from 8.4 to 2 Ma (5.32 Ma: reference 68; Ma: reference 69) and calcareous nannofossil biostratigraphy between 16 and 8.4 Ma 70. At Site 1088, the age model is primarily based on detailed calcareous nannofossil biostratigraphy 53,64 and in the late Miocene interval of Site 1088 (8.6 to 5.5 Ma), biostratigraphic ages are refined using a highresolution

14 RESEARCH SUPPLEMENTARY INFORMATION δ 13 C benthic record 71. At Site 999, all biostratigraphic ages applied to nannofossil events are those of reference 72, whereas at Site 1088, ages of reference 72 and other authors are applied to biostratigraphic events 64. Over the late Oligocene to early Miocene interval at Site 1264, we apply the δ 18 O benthic age model of reference 66. For the EO samples from Sites U1334 and 1090, no age model is applied because we investigate background conditions in isolated samples before and after the transition only. The EO transition was identified on metres composite depth (mcd) scales of both sites using bulk δ 18 O stratigraphies (Site U1334: P.A. Wilson el al., unpublished data; Site 1090: C.T. Bolton & H.M. Stoll, unpublished data) and previously published information. Because of the variable resolution and length of available δ 18 O benthic records at the two MioPliocene sites (1088 and 999), we cannot justify putting the sites on a common age model, instead we leave each record on its sitespecific astronomical and/or biostratigraphic age model. Thus, orbitalscale (<100 ka) discrepancies between age models may arise from different orbital tuning targets, however given the coarse resolution of our coccolith sizefraction data (~0.51 Myr) we do not think these potential offsets affect our interpretations. Certain features in the δ 18 O benthic and δ 13 C benthic records are near coincident in time at both sites, despite different age models. For example, the late Miocene δ 13 C benthic minimum is visible at ~7 Ma in both Site 999 and 1088 records, and the inflection point towards a greater rate of increase in δ 18 O benthic, indicative of global cooling and the intensification of northern hemisphere glaciation, occurs in both records at ~43 Ma (Fig. S11). 3. Coccolith separations All bulk samples were first gently disaggregated on a rotating ferris wheel in either ethanol (where siliceous microfossils were present) or weak ammonia solution (where sediments were primarily carbonate) then filtered through a 20 µm mesh. At all sites, four to six narrow coccolith sizefractions were isolated from the <20 µm fraction of bulk sediment samples using protocols described in detail in reference 13. In brief, samplespecific protocols were developed following assessment under the light microscope to best separate the coccoliths present in each sample into distinct size classes using a combination of repeat decanting and microfiltering. At Site 999, a similar size range of coccoliths throughout the Miocene and Pliocene meant that an identical protocol was applied to all samples, originally developed for the Pliocene sample set 13. At Site 1088, changes in coccolith size distribution with depth and a variable contribution of large (>

15 RESEARCH µm) foraminifera fragments and tiny (<2 µm) coccolith fragments meant that separation protocols were tailored to individual or small groups of samples. As a result, in some samples we were able to separate six size classes dominated by coccolith CaCO 3, whereas in others only three or four sizefractions could be efficiently separated (ranging from <2 µm to 1012 µm). Over the OM at Site 1264, five coccolith size classes were separated but only three used (24 µm, 46 µm and 69 µm) because the smallest fraction was dominated by fragments, and the largest size fraction primarily constituted moderately overgrown Discoaster, thus their geochemical composition is likely significantly altered (M. Dedert & H.M. Stoll, personal communication, 2012). At sites 1090 and U1334, EO coccoliths were separated into six size classes ranging from 24 µm to 1220 µm. Here, because of the presence of very large placoliths (1020 µm), the <20 µm fraction was sieved at 10 µm before repeat decanting of the <10 µm fraction to ensure no contamination by large coccoliths in the smaller size classes. Smear slides of each size fraction for each sample were assessed by microscope under crosspolarised light at x1250 magnification, and size fractions found to contain a large proportion of fragments were excluded from the coccolith dataset. Coccolith counts to assess species approximate contribution to percent carbonate in each size fraction were carried out on endmember samples (Table S3). 4. Sr/Ca productivity records and SST corrections In coccolithophores, high growth rates have been demonstrated to lead to reduced carbon isotopic fractionation into organic matter 38, and productivity may also affect isotopic fractionation into coccoliths. To rule out large productivity shifts as a driver of coccolith vital effects, we generated coccolith Sr/Ca records on two size fractions at both Site 1088 and 999. Sr/Ca records were corrected for SST variations, but not for secular changes in seawater Sr/Ca given that the latter appears to have been relatively stable since the Miocene 58. Sr/Ca data were corrected for the influence of SST on Sr partitioning using the equations of reference 57 for G. oceanica and Helicosphaera carteri. We use the G. oceanica equation to correct our small coccolith size fractions, which are dominated by small reticulofenestrid coccoliths (of the same lineage as Gephyrocapsa) at sites 999 and We choose the H. carteri equation to correct our large coccolith size fraction at both sites, although Helicosphaera is not always the dominant genus (see Table S3). This is because H. carteri records the highest Sr/Ca ratios and organic carbon fixation rates in culture and sediment studies 73, therefore represents the most extreme large coccolithophore scenario available. We note that the choice of speciesspecific equation has very little effect on our resultant productivity records. The influence of local SST variations was

16 RESEARCH SUPPLEMENTARY INFORMATION subtracted from measured coccolith Sr/Ca, leaving a residual Sr/Ca record attributed to productivity variations (Fig. S5). For the Site 1088 Sr/Ca correction, we use alkenonederived SSTs from the same samples (K.T. Lawrence, unpublished data). Where the exact sample was not available, we interpolate between nearby samples either side. At Site 999, no published SST record spans our entire interval; we therefore performed sensitivity tests using a maximum and minimum SST scenario (pink lines in Fig. S5a). For our minimum SST scenario, we apply the alkenonebased SST record of ODP Site 1010 in the subtropical eastern Pacific 28. This SST record is continuous between 0 and 13 Ma, therefore we necessarily interpolate values for our two oldest samples (14.4 and 15.8 Ma) assuming a constant rate of SST increase over time using the mean rate of increase between 4 and 13 Ma. As a maximum SST scenario for Site 999, we use the alkenonederived SST record of ODP Site 1241 in the eastern equatorial Pacific 61, corrected for the maximum difference between this record and the alkenonebased SST record of Site where the two records overlap (05 Ma, max. difference: +1.4 C). These sensitivity tests at Site 999, and the correction of Site 1088 Sr/Ca, illustrate that the Sr/Caderived productivity residuals at both sites show similar trends to measured Sr/Ca regardless of which SST correction or species equation is applied (Fig. S5). In other words, only a small proportion of the measured variability in Sr/Ca can be attributed to SST changes at this site. 16

17 RESEARCH Figure S1: Schematic of a coccolithophore cell showing fluxes and isotopic compositions incorporated into our numerical model (see Supplementary Discussion 1). See Table S1 for definitions of symbols and Table S2 for fractionation factors (ε) used. CHL is chloroplast and CV is coccolith vesicle. Labelling of fluxes and isotopic compositions follows the convention of reference 31, with the addition of the subscript v to denote fluxes within and to the CV, such that e, i, x, and v denote respectively external, intracellular (cytosol), chloroplast, and CV, and u and o refer to uptake and outflux, respectively. In the CV we also evaluate whether the thermodynamic fractionation applies to the hydration and dehydration of CO

18 RESEARCH SUPPLEMENTARY INFORMATION Figure S2. a: p and b coccolith measured in cultures of coccolithophores as a function of dissolved [CO 2 ]. G. oceanica (blue) and C. braarudii (red) data from reference 12. E. huxleyi (green) data from reference 15. p precision = 0.5 for all species (error bars shown in a). E. huxleyi coccolith values were determined indirectly from isotopic determinations of total particulate carbon and particulate organic carbon therefore have a larger analytical uncertainty of (error bars plotted) compared to data for G. oceanica and C. braarudii (propogated analytical uncertainty = 0.1, error bars not plotted because same size as data symbols). In all panels, mean values of replicate culture experiments are plotted and these values are used in the model. c: diffusive flux of CO 2 into the cell (calculated as described in Section 1 A2) divided by the cellular particulate organic carbon quota (as reported in references 12 and 15). 18

19 RESEARCH Figure S3: Principal drivers of variation in coccolith including the sources of carbon to the CV (a), the relative importance of uptake of HCO 3 to the cytosol (b), and the relative HCO 3 demand between the CV and chloroplast (c). a) and d) illustrate simulations with appreciable variation in the ratio of the CV HCO 3 influx to calcification, the dominant relationship regulating coccolith. Different colours within each panel represent repeat simulations in which different secondary parameters were altered to test sensitivity of trends. HCO 3 can be sourced either from hydration of CO 2 in the cytosol or from direct active uptake across the cell membrane, and in the ordinate of panel (d) we illustrate diverse possible tendencies in the proportion of intracellular HCO 3 transport (to the CV and chloroplast) which is supplied by direct cytosol HCO 3 uptake. All of these scenarios in (d), with colours matching the simulation in a), show a comparable slope for the relationship between the proportion of calcification supported by CV HCO 3 uptake and coccolith. b) illustrates simulations at constant CV HCO 3 in/calcification ratio but variation in the proportion of active HCO 3 uptake to the chloroplast and CV that is supplied by direct influx of HCO 3 to the cytosol. c) and e) show scenarios at constant CV HCO 3 /calcification ratio and constant cytosol HCO 3 /( CV HCO 3 + chloroplast HCO 3 ) ratio but different partitioning of HCO 3 to the CV and chloroplast. In both panels (c) and (e), blue shows result for a constant PIC:POC ratio and red for a variable PIC:POC ratio. Small variations in coccolith arise when more HCO 3 is allocated to the CV and the PIC/POC ratio is stable, but these are damped when increasing allocation of HCO 3 to the CV is accompanied by an increase in the PIC/POC ratio. 19

20 RESEARCH SUPPLEMENTARY INFORMATION a b c d e HCO 3 in/calcification CV chl. HCO 3 in/cyt. CO 2 in chl. HCO 3 in/fixation CV HCO 3 in/chl. HCO 3 in CV HCO 3 in/cyt. CO 2 in Assume cyt. HCO 3 influx = 0.5* (CV+chl HCO 3 influx) Assume cyt. HCO 3 influ x = CV HCO 3 influx Assume cyt. HCO 3 influ x = chl HCO 3 influx Assume cyt. HCO 3 influx = constant f cyt. HCO 3 in/(cv+chl HCO 3 in) [CO 2 ] (μm) [CO 2 ] (μm) [CO 2 ] (μm) [CO 2 ] (μm) Figure S4: Active fluxes in inverse model solutions made under different assumptions of cellular (cytosol) HCO 3 uptake and otherwise with default model parameters as given in Table S1. Red square symbols for C. braarudii (large cell), blue diamonds for G. oceanica (smallmedium cell), and green triangles for E. huxleyi (small cell). In (d), G. oceanica values are shown on right yaxis and E. huxleyi and C. braarudii on left yaxis. 20

21 RESEARCH Figure S5: SST correction of Sr/Ca records. a: Miocene to Pleistocene SST records from various sites 28,61,74 76 and maximum and minimum estimates for Site 999 (pink solid and dashed lines, see Supplementary Methods for details). b: Coccolith Sr/Ca measured on two size fractions at Site 999. Three data points at 10.3 Ma, 12.5 Ma, and 15.7 Ma (shown on graph), from samples with very low CaCO 3 content, were excluded due to suspected Sr contribution from clays indicated by substantially higher Sr/Ca than expected for coccoliths, as well as high Fe/Ca and Al/Ca compared to other samples; c: Sr/Ca productivity residuals at Site 999 with different SST scenarios (normalised to mean values). d: Coccolith Sr/Ca from two size fractions at Site e: Sr/Ca productivity residuals at Site 1088 (normalised to mean value). 21

22 RESEARCH SUPPLEMENTARY INFORMATION Figure S6: Scanning Electron Microscope images of coccoliths from Site 999. A to F are from a Pliocene sample ~4 Ma. A: small placolith (Umbilicosphaera sp.?); B: Calcidiscus leptoporus and Discoaster species; C: Coccolithus pelagicus (distal view); D: Coccolithus pelagicus (proximal view); E: Helicosphaera carteri var. wallichii (proximal view); F: Pontosphaera sp.; G to K are from a Miocene sample ~15 Ma. G: Helicosphaera carteri (distal view); H: Helicosphaera carteri (proximal view) and Calcidiscus leptoporus; I: Coccolithus pelagicus (proximal view); J: Discoaster deflandrei; K: Reticulofenestra sp. (proximal view). 22

23 RESEARCH Figure S7: Scanning Electron Microscope (SEM) images of coccoliths from Site A to F: Pleistocene sample ~ 0.8 Ma. A: Calcidiscus leptoporus (distal view); B: Coccolithus pelagicus (distal view); C: Helicosphaera carteri (proximal view); D: Helicosphaera carteri (distal view) and small placoliths; E: Pontosphaera sp., F: Syracosphaera pulchra; G to K: Miocene sample ~14.8 Ma. G: Calcidiscus leptoporus (distal view) H: Coccolithus pelagicus (distal view); I: Reticulofenestra sp.; J: Reticulofenestra coccosphere; K: Calcidiscus leptoporus coccosphere. 23

24 RESEARCH SUPPLEMENTARY INFORMATION. Figure S8: The effect of overgrowth on the magnitude of an original O difference between small and large coccoliths. Blue line represents the original isotopic difference between the two sizefractions. Figure S9: Map showing the location of all (I)ODP sites from which fossil coccoliths were separated and analysed. Different colours indicate different time intervals studied (see map legend). 24

25 RESEARCH Figure S10: Crossplots of coccolith 18 O versus 13 C for different Cenozoic time windows at various sites (all size fractions shown here; smallest and largest only shown in Figure 2a of main text). a: Last Glacial Maximum 13 ; b: PlioPleistocene transition 13 ; c: OligoceneMiocene transition. Here we show mean isotope values of the 17point time series because no change in interfraction isotopic differences occurs over time (i.e. maximum difference <0.4 in all samples between 23.7 and 22 Ma for both isotopes); d: Early Oligocene at two distinct sites (U1334 and 1090); e: late Eocene at 2 distinct sites (U1334 and 1090); f: PaleoceneEocene thermal maximum

26 RESEARCH SUPPLEMENTARY INFORMATION Figure S11: Benthic foram 18 O (a) and 13 C (b) records from our MioPliocene sites compared to a global compilation over the past 16 Ma. Compiled benthic foraminifer isotope records are from reference 9 and are separated into Atlantic (black) versus Pacific (blue) sites. Site 999 data (orange/yellow) are from references 68 and 69. Site 1088 data (pink/purple) are from references 71 and 77. All records are smoothed with a 5point running average. Grey bands illustrate synchronous features between records. 26