Decrease in coccolithophore calcification and CO2 since the middle Miocene

Marine algae are instrumental in carbon cycling and atmospheric carbon dioxide (CO2) regulation. One group, coccolithophores, uses carbon to photosynthesize and to calcify, covering their cells with chalk platelets (coccoliths). How ocean acidification influences coccolithophore calcification is strongly debated, and the effects of carbonate chemistry changes in the geological past are poorly understood. This paper relates degree of coccolith calcification to cellular calcification, and presents the first records of size-normalized coccolith thickness spanning the last 14 Myr from tropical oceans. Degree of calcification was highest in the low-pH, high-CO2 Miocene ocean, but decreased significantly between 6 and 4 Myr ago. Based on this and concurrent trends in a new alkenone ɛp record, we propose that decreasing CO2 partly drove the observed trend via reduced cellular bicarbonate allocation to calcification. This trend reversed in the late Pleistocene despite low CO2, suggesting an additional regulator of calcification such as alkalinity.

) and the assumption of constant glacial pCO 2 prior to the ice core record, c: estimated surface alkalinity at ODP Site 668 in the tropical Atlantic based on these two parameters, following calculations described in the Methods section, d: size-normalised coccolith thickness (2-5 μm) and G. sacculifer δ 18 O data at Site NGHP 01-01A, e: size-normalised coccolith thickness (2-5 μm) and G. sacculifer δ 18 O data at Site 925. In d and e, the green bar on the right shows the last glacial (G) to current interglacial (IG) range in G. sacculifer δ 18    The theoretical relationship between PIC to POC ratio and cell size assuming constant coccolith thickness. b to i: PIC/POC ratios vs cell diameter from published culture experiments with different species and strains (note different scales). All cell diameters (except h where cell size was measured by the authors) were calculated from POC per cell measurements in the experiments, using the wellconstrained relationship between cell size and POC from ref. 13  24 coccoliths are imaged under cross-polarised light at three specific orientations and create a composite image so as to eliminate the extinction cross. Taking into account the expected variability in mass for coccoliths of a given size resulting from variable geographical locations, ages and growth conditions in the datasets we compare, there is very good agreement between Noëlaerhabdaceae coccolith mass values for the three birefringence-based methods. Calibrations applied to convert grey level to thickness also differ slightly between methods. A sensitivity test performed on our data to account for the sigmoidal shape of the grey level-thickness relationship in the newest published calibrations showed an increase in absolute thickness values, particularly for the smallest coccoliths, but no change in the temporal thickness trends at either site meaning that our interpretations are not affected.  Illustration of mass balance of carbon, and isotopic composition in coccolithophore cell keyed to the relevant mass balance equations using the ACTI-CO model, as described in reference 28. Fractionation factors are identical to reference 28. CHL indicates chloroplast and CV is coccolith vesicle. Labelling of fluxes and isotopic compositions follows the convention of reference 29 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. Notation is as in reference 29.

Carbon isotopes in alkenones and ε p and [CO 2aq ] calculations
Lipids were extracted from sediments and alkenones subsequently isolated as described in ref. 8. Gas chromatography (GC) was performed with an Agilent 7890A chromatograph equipped with a flame ionisation detector and an Agilent J&W HP-1 fused silica column (19091Z-015, 50 m x 0.32 mm internal diameter) coated with CP Sil5-CB stationary phase (dimethylpolysiloxane equivalent, 0.12 µm film thickness). The oven temperature was programmed from 70 to 130 °C at 20 °C min −1 , to 300 °C (held 25 min) at 4 °C and H 2 was used as carrier gas. Compounds were quantified using a C 36 n-alkane that had been added to the alkenone sub-fraction prior to injection. Compound-specific isotope analyses were performed using a GC-combustion-isotope ratio mass spectrometer (GC-C-IRMS) with an Agilent 7890A GC coupled to a Nu Instruments Perspective IRMS. The GC column and temperature programme were as for GC analysis. The internal standard added to the alkenone sub-fraction was of known isotopic composition to ensure instrument stability 8 . Isotope ratio values are reported as δ values (δ 13 C, ‰) and reproducibility was better than 0.7 ‰. Steph (unpublished). From 6 Ma to 8 Ma, G. sacculifer data are from T. Bickert (unpublished). All other G. sacculifer δ 13 C values were generated for this study (as described above for the other sites) (Supplementary Data 2). For three samples in the older part of the record, G. sacculifer could not be extracted in sufficient numbers for reliable isotopic analysis, hence we estimated G. sacculifer δ 13 C from benthic foraminiferal δ 13 C, using the average δ 13 C gradient between Cibicidoides sp. and G. sacculifer at ODP Site 999 (1.4 ‰). Benthic foraminiferal δ 13 C (Cib.) for ODP Site 999 extends back to 8.4 Ma 25,26 and was extended to older periods using a compilation of North Atlantic sites (ODP/DSDP Sites 553, 558, 563, 607, 608, and 959) taken from ref. 27 (Supplementary Fig. 12). The North Atlantic compilation was adjusted by -0.3‰ to account for the mean offset relative to ODP Site 999 benthic δ 13 C during the interval where the two records overlap (2-8.4 Ma) ( Supplementary Fig. 12). For one sample corresponding to 13.1 Ma, planktic foraminferal δ 13 C shows no gradient compared to benthic δ 13 C. Since there is no independent evidence for homogenization of the water column at this time (e.g. in coccolith assemblages), we infer that planktic foraminifers may be altered by recrystallization in the deep-water or sediment environment. Hence for this sample, we also use a value for G. sacculifer δ 13 C calculated assuming a plankticbenthic gradient of 1.4 ‰. Given the absence of tri-unsaturated alkenones in sediments samples, it was not possible to estimate sea surface temperatures (SSTs) from the U k 37 ' index, and therefore maximum and minimum SST estimates for ODP Site 999 28 were used in ε p calculations to calculate δ 13 C CO2(aq) 8 .
Calculations of aqueous CO 2 concentrations from ε p and the various approaches we use to constrain temporal variation in b, thus isolating the component of variation in ε p driven by [CO 2aq ], are described in the main text. The absolute values of b and [CO 2aq ] are much more poorly constrained than the trends, imparting a greater uncertainty in the absolute values of [CO 2aq ] and atmospheric pCO 2 than in the temporal trend. To correct for cell size changes, given that the trend in Noëlaerhabdaceae coccolith size observed in our sites is similar to that found in other tropical sites 30 Figure 7d (error bars). This formulation describes the influence of cell size on ε p assuming that changing cell size does not entail a change in maximum growth rate. Its use is consistent with the observation that for coccolithophores, the size dependence of growth rates is very small, equivalent to less than a 5 % increase in growth rate for the observed reduction in cell diameter from 4 to 2.7 µm, based on a comparison of coccolithophore growth rates in multiple culture studies 34 .
The potential effect of changing growth rates on ε p and calculated [CO 2aq ] is evaluated via changes in the b value, following previous studies 8 . As indicators of potential variations in growth rates, we examine variations in coccolith Sr/Ca ratios suggested to correlate positively with growth rate and the "b" physiological coefficient 35 , and alkenone accumulation rates in sediment that may serve as a proxy for productivity of alkenone producers, particularly in situations where preservation potential is stable 36 . Given stable seawater Sr ratios over this time period 37 , coccolith Sr/Ca might be expected to track growth rates of alkenone producers. We use Sr/Ca data from ODP Site 999, from the size fraction dominated by coccoliths from the alkenone-producing Noëlaerhabdaceae family 28 . Highest coccolith Sr/Ca occurs from 13-10 Ma, coinciding with a local maximum in alkenone mass accumulation rates (Fig. 7c), although the magnitude of Sr/Ca change in this interval may be amplified somewhat by a higher relative contribution of detrital Sr due to the low CaCO 3 content of sediments during the "carbonate crash". In addition, a peak in alkenone accumulation rate occurs at 8 Ma. The higher resolution alkenone mass accumulation rate data for the last 5 Ma 8 evidences higher frequency variability. From these two indicators, we normalised the variation in each indicator and averaged these normalised variations to establish a composite growth rate curve for the interval 16-6 Ma to estimate temporal variation in b (Fig. 7d). There is as yet no calibration of the magnitude of growth rate change implied by either indicator; we use an amplitude of 25, consistent with that explored in ref. 8 and which is within the large uncertainty in the slope of relationship between Sr/Ca and b value in the Equatorial Pacific 35 . To illustrate the sensitivity of b to this choice, we also show growth rate variations with amplitude 20 % lower and higher (i.e., range of b of 20 or 30; error bars in Figure 7d). For the period studied in ref. 8, where ε p data are not accompanied by Sr/Ca measurements, we do not simulate variations in growth rate. Adding this estimated variation in b to that inferred to result from cell size changes results in a range of b of 87 to 167 (Fig. 7d) Figure 7e and include uncertainty in ε p as illustrated in Figure 7a, as well as uncertainty in estimates of b, as illustrated in Figure 7d. For size corrected data, maximum [CO 2aq ] was calculated using the upper limit of ε p (calculated using SST max, G. sacculifer δ 13 C +1SD, and δ 13 C haptophyte biomass -1SD; see error bars in Fig. 7a), and the upper limit of b; whereas minimum [CO 2aq ] was calculated using the lower limit of ε p (calculated using SST min, G. sacculifer δ 13 C -1SD, and δ 13 C haptophyte biomass +1SD, see error bars in Fig. 7a) and the lower limit of b (orange shading in Fig. 7e). For the size and growth-rate corrected data, the maximum and minimum [CO 2aq ] were calculated analogously but using the b, which incorporates both size and growth rate terms (pink shading in Fig. 7e).

The effect of changing active uptake on [CO 2aq ] estimates
To illustrate the potential impact of changes in active uptake of carbon on ε p and calculated [CO 2aq ], we use the ACTI-CO cell model of carbon fluxes in coccolithophores 28 (Supplementary Figure 13  and Supplementary Table 3). We impose several possible dependencies of active HCO 3 transport to the chloroplast as a function of [CO 2aq ], and for each, solve for the [CO 2aq ] required to match observed ε p at ODP 999. Laboratory culture experiments suggest that active HCO 3 transport to the chloroplast becomes more significant at low [CO 2aq ], with the ratio of chloroplast HCO 3 transport to diffusive CO 2 uptake following a logarithmic dependence 28,38 . The resource-replete laboratory experiments likely give an upper limit of the significance of active uptake, because high light intensity leads to high rates of active uptake 38 and laboratory experiments typically feature much higher light intensities than those that characterise the deep chlorophyll maximum at which Noelaerhabdaceae density is highest in the oligotrophic open ocean ocean 39 . In fact, this more limited influence of active uptake in the deep oligotrophic ocean has been identified in measurements of ε p on alkenones through depth profiles in the water column 40 . A first simulation employs a logarithmic dependence of chloroplast HCO 3 transport on [CO 2aq ], with a slope similar to that observed in cultures but an intercept half that observed in cultures to account for lower light intensity (Fig. 8). A second simulation supplements HCO 3 supply to the chloroplast in terms of reallocation of HCO 3 from calcification to photosynthesis. The HCO 3 available for reallocation to the chloroplast is estimated as the difference between the HCO 3 flux to the coccolith vesicle which would be required to match ε coccolith if coccolith calcification remained constant, from that required to match ε coccolith for the observed situation of reduced calcification per cell surface area in the last 8 Ma (reduced PIC/POC) for the coccoliths less than 5 µm. Both simulations parameterize HCO 3 uptake across the cytosol as 0.5x the sum of HCO 3 transport to the coccolith vesicle and chloroplast, analogous to previously published simulations 28 . Large cells also show evidence of shifts in HCO 3 allocation to calcification: a major decrease in ε coccolith , an indicator of coccolith vesicle HCO 3 influx relative to calcification, occurs after 8 Ma (Fig. 5). This reallocation in large coccolithophores probably had a minimal effect on the alkenone ε p record, because by 8 Ma mean Noëlaerhabdaceae size has already decreased significantly, implying that the majority of alkenones were produced by small cells rather than large cells. However, in both large and small cells, the 10-6 Ma onset of this potential HCO 3 reallocation to the chloroplast is similar to the onset inferred for increased HCO 3 to the chloroplast from a simple dependence of active uptake on CO 2 concentrations described as simulation 1, so would be unlikely to modify the simulations of active uptake.

Calculating constraints on past ocean pH and alkalinity
We assess the constraints on the magnitude of potential surface alkalinity increase which are provided by pH estimates from ref. 3 for the last 2 Ma and pCO 2 values from Vostok ice cores for the last 800 ka 4,41,42 ( Supplementary Fig. 6). We use pH determinations made at ODP Site 668 in the eastern equatorial Atlantic, a region with CO 2 in equilibrium with the atmosphere 3 . Over the last 800 ka, atmospheric CO 2 during glacial maxima consistently declines to 180 ± 10 ppmv 4 , coherent with glacial temperature anomalies in Antarctica. If this 180 ppmv level of pCO 2 were characteristic of glacial maxima during the preceding 1200 kyr, then pH and CO 2 could be used to estimate surface alkalinity at the location of ODP 688 during glacial times. Calculations were performed with the programme CO2sys 43 , using constants of ref. 44. Under these assumptions, surface alkalinity at this site could have increased substantially during glacials since the mid-Pleistocene, from 1800 to 2300 µmol kg -1 , an increase of nearly 30 % over the last 0.6 Ma ( Supplementary Fig. 6c). Among interglacials of the past 800 ka, pCO 2 has varied much more significantly from 240 to 300 ppmv, with higher interglacial pCO 2 in the last 500 ka. Consequently, one cannot make an analogous assumption of uniform interglacial CO 2 prior to the ice core record to estimate long-term interglacial alkalinity changes prior to 800 ka. During the period of direct pCO 2 determinations in ice cores, calculated alkalinity during interglacials is lower than that during glacial periods.
As estimates of surface water pH from B isotopic ratios of monospecific foraminiferal samples are available only discontinuously over the last 15 Ma, as shown in Figure 7f, we employ the more complete records of variation in ocean carbon chemistry are available for [CO 3 -2 ] to estimate long term alkalinity change. Long term variations in surface ocean [CO 3 -2 ] have been derived from coupled records of seawater calcium concentration and the history of the carbonate compensation depth (CCD) 7 . These estimates are subject to some uncertainty due to the potential decoupling of the CCD and the carbonate saturation horizon which is the parameter strictly regulated by [CO 3 -2 ] 45 .
Overall this record suggests increasing [CO 3 -2 ] over the last 15 Ma (Supplementary Fig. 7). The combination of [CO 2aq ] records with these [CO 3 -2 ] reconstructions suggests a relatively stable total alkalinity from about 15 to 2 Ma and a progressively increasing surface ocean pH ( Supplementary  Fig. 7). For discrete intervals between 15 to 12 Ma, available direct pH estimates from planktic foraminifera δ 11 B suggest the potential for more dynamic carbon system variations over timescales of a few million years. Therefore, within the uncertainties of existing data between 15 and 2 Ma, we cannot rule out the possibility of variations in alkalinity that may have contributed to changes in SN coccolith thickness during this interval.