Coccolith thickness and cellular calcification

New culture experiments sampling the diversity of modern Noëlaerhabdaceae coccolithophores show variation in coccolith thickness both among the different species and among different strains of the same species. This variation in thickness correlates strongly with variation in cellular calcification per surface area as well as with changes in calcite/organic carbon, a measure of calcification per cell volume (Fig. 2 and Supplementary Fig. 1). The thickness of an individual coccolith is intimately linked to the degree of calcification of a cell, because it represents a key mechanism by which cells can regulate the amount of biomineral for a given cell volume. Various factors may drive cells to adjust calcite per cell surface area. In this study, we focus on changes that are occurring within narrow size classes. In addition, calcite per cell surface area varies with cell size across the modern diversity of placolith-bearing coccolithophores, where small cells are characterized by thinner coccoliths (Supplementary Fig. 2). This latter effect may be an adaptation to compensate for the higher surface area to volume ratio of small cells that, if calcification per cell surface area were constant across all cell sizes, would impose a much higher biomineral requirement relative to cell volume in small cells. While coccolith mass has been used as an indicator of cellular calcification in Pleistocene and recent sediments15,16,17,18, coccolith mass is driven by changes in cell size as well as degree of calcification. On the other hand, coccolith thickness within narrow size classes, or size-normalized coccolith thickness, represent degree of calcification and are indicators better suited to reconstructing coccolithophore calcification on long timescales over which significant coccolith and cell size changes occur. The range of coccolith thickness variation among cultured Noëlaerhabdaceae strains (Fig. 2) is consistent with previous observations that phenotypic differences in the degree of calcification between species and between strains of the same species tend to be much larger than the phenotypic plasticity of a single strain cultured under varying environmental conditions8,9,20. This may arise if coccolith morphotype or thickness is genetically regulated9. The potential for large intraspecific diversity may reflect the genetic architecture, in that the dominant modern Noëlaerhabdaceae Emiliania huxleyi has a pan-genome composed of core genes plus genes distributed variably amongst strains21.

Figure 2: Noëlaerhabdaceae coccolith morphology in culture. (a) Relationship between coccolith thickness and cellular PIC/POC (particulate inorganic carbon/particulate organic carbon) and cellular PIC/cell SA (surface area) for modern strains of Emiliania huxleyi and Gephyrocapsa grown in laboratory culture (Supplementary Table 1). Symbols are averages for each experiment and lines show the range of values between replicate culture bottles for each experiment. Scanning electron microscope images of coccospheres from the strains with lowest (RCC 1257, b), intermediate (RCC 3370, c) and highest (RCC 1292, d) coccolith thickness. Scale bar, 2 μm (in all images). (e–g) Three-dimensional representations of coccolith thickness for the same strains as coccospheres. The vertical scale shows cumulative thickness from zero at the base; therefore in the central area of Gephyrocapsa coccoliths, the bridge (central area bar) is displaced towards the base of the plane of illustration. Full size image

Decreasing cellular calcification since the late Miocene

Over the past 14 Myr, the Noëlaerhabdaceae have undergone large variations in coccolith size (Fig. 3) and degree of calcification, represented by thickness (Fig. 4). Changes in coccolith thickness are evident in both narrowly restricted size classes, as well as in measurements of size-normalized (SN) thickness and calculated ‘shape factor’, confirming that they are not a direct result of temporal changes in coccolith and cell size (isometric scaling, that is, changes related to proportional changes in size) (Fig. 4). The quantification of thickness was not biased by variable coccolith fragmentation (Supplementary Fig. 3). Scanning electron microscope (SEM) observations confirm that in all samples the original crystal structure of the coccolith remains well defined. Only on some older coccoliths did we identify a small amount of diagenetic overgrowth (small abiogenic crystals formed on the surface of the collar in the central area; Supplementary Data 1) However, the presence of this minor overgrowth does not correspond to an increase in coccolith thickness, except in the oldest 14 Myr old sample at the Indian Ocean Site. Thus, with this exception, the preservation visible under SEM makes it unlikely that middle Miocene Noëlaerhabdaceae coccoliths of a given size were originally thinner and more delicate than those present in our samples. This suggests that either (1) overgrowth was minor enough not to significantly impact mean coccolith thickness, or (2) the calcite that recrystallized on the surface of coccoliths was originally derived from dissolution of primary calcite of these same coccoliths. Between 8 and 3 Myr ago at both sites, Spenolithus and Discoaster nannoliths are abundant. These are typically more susceptible than placolith coccoliths to overgrowth due to their crystal structure, yet SEM images show that these susceptible forms exhibit excellent and constant preservation, providing supporting evidence that diagenetic overgrowth was not more significant when Noëlaerhabdaceae coccoliths showed a higher degree of calcification at 6–8 Myr ago relative to at 3–4 Myr ago (Supplementary Data 1; Supplementary Figs 4 and 5).

Figure 3: Long-term evolution of Noëlaerhabdaceae coccolith size and stable isotope vital effects with climate over the last 14 Myr. (a) Noëlaerhabdaceae coccolith size distributions over time at Sites ODP 925 (grey) and NGHP-01-01A (blue). (b) Climate evolution over the Neogene represented by a benthic foraminiferal δ18O stack (data compiled by ref. 86). (c) Onset of major northern hemisphere glaciation at ∼2.6 Myr ago. (d) The emergence of large scale vital effects in the δ18O and δ13C of coccolith calcite around 7–5 Myr ago (ref. 14; this study). (e) Approximate age ranges of Neogene genera belonging to the Noëlaerhabdaceae family76. Full size image

Figure 4: Changes in Noëlaerhabdaceae coccolith thickness and k s value at two tropical sites since 14 Myr ago. (a–d) Site NGHP-01-01A, and (e–h) ODP Site 925. (a–c,e–g) Thickness data for coccoliths of 2–3, 3–4 and 4–5 μm length. Box–Whisker plots illustrate coccolith thickness data for each sample and size class (box shows median value and upper/lower quartiles, whiskers show maximum and minimum values, outliers >1.5 × the interquartile range are shown as crosses). Also shown are mean values of raw (circles) and SN (diamonds) thickness (Supplementary Data 1). Bar graphs show the relative contribution of different genera to the Noëlaerhabdaceae population in each size class and sample. (d,h) k s values (error bars are ±2 s.e.m.). The shape factor k s , which expresses the fraction of the volume of a cube defined by the length of a coccolith that is composed of biomineral77, was originally proposed to estimate coccolith mass from coccolith length and is similar to coccolith thickness. However, unlike thickness, k s does not account for variations in coccolith circularity. Pink symbols are k s for extant Noëlaerhabdaceae species77. Full size image

The measured coccolith populations exhibit large variability in the morphology and degree of calcification of small coccoliths within and between each sample (Fig. 4; Supplementary Figs 4 and 5; Supplementary Data 1). For example, Gephyrocapsa protohuxleyi, a form close to E. huxleyi but with a central area bridge characteristic of Gephyrocapsa, was present in Pleistocene samples at both sites alongside much more heavily calcified Gephyrocapsa coccoliths. Despite this large diversity in morphology and thickness, there are significant changes in the dominance of more heavily calcified versus more lightly calcified forms over time, as well as the emergence during the early Pliocene of coccoliths thinner and/or with larger central area openings than those found in previous intervals. Coccolith degree of calcification was on average highest between 14 and 6 Myr ago and decreased abruptly in the late Miocene to early Pliocene (6–4 Myr ago) to low values that were maintained during the Pliocene and early Pleistocene (4–1 Myr ago). For the few sample points of the last 1 Myr ago, degree of calcification increased both in the Indian and Atlantic Ocean records relative to this Pliocene minimum (Fig. 4). However, we note that assemblages in our samples <1 Myr are dominated by Gephyrocapsa coccoliths and pre-date the emergence of the less heavily calcified E. huxleyi (see k s values, Fig. 4d,h), which is significant especially in modern high and mid-latitude regions. Large changes in degree of coccolith calcification, including the decrease from 6 to 4 Myr ago and the increase around 1 Myr ago at both sites, occurred within the dominant genus at a given time, and do not coincide with major shifts in the contribution of different genera to the Noëlaerhabdaceae (Fig. 4).

Late Miocene changes in cellular HCO 3 − allocation

Geochemical records of carbon isotopic fractionation into coccolith calcite (ɛ coccolith ) can be used to elucidate the relationship between the observed changes in degree of calcification and the resource allocation of carbon to calcification. Models of cellular carbon fluxes have shown that ɛ coccolith becomes increasingly depleted if the rate of supply of HCO 3 − to the site of calcification (coccolith vesicle) is reduced relative to calcification rate14. Our new records of ɛ coccolith from ODP Site 925 show that large cells begin to decrease the HCO 3 − allocation to calcification at about 8 Myr ago, evidenced by decreasing ɛ coccolith (Fig. 5a,b). This trend occurs shortly after a decrease in mean Noëlaerhabdaceae coccolith size (interpreted as a reduction in mean cell size13) at both sites (Fig. 3) that is also observed in other low-latitude records12,22,23. Reduced HCO 3 − allocation to calcification continues in large cells from 6 to 4 Myr ago, as indicated by decreasing ɛ coccolith during this interval, despite a stable trend in mean coccolith size. Although we cannot resolve changes in the degree of calcification of large coccoliths in this study (see Methods), the ɛ coccolith trend suggests that in large cells, the change in HCO 3 − allocation to calcification was of greater magnitude than any concurrent decrease in calcification that may have occurred. This significant reduced allocation to calcification in large cells drove a divergence in the range of vital effects among small and large coccoliths after 8 Myr ago (Fig. 5a), similar to the results from Caribbean ODP Site 999 (ref. 14).

Figure 5: Coccolith geochemistry and SN thickness trends since the Miocene. (a) δ13C values for size-separated coccoliths from ODP Site 925, normalized to the smallest coccolith size fraction. (b) ɛ coccolith for small (3–5 μm) and large (8–10 μm) coccoliths from Site 925. Shading indicates propagated analytical uncertainty on δ13C measurements. (c) mean SN coccolith thickness for Noëlaerhabdaceae coccoliths of 2–5 μm lengths at ODP Site 925 and Site NGHP-01-01A (coccolith thicknesses are normalized to mean coccolith length within the 2–5-μm size fraction over the whole time series at each site, that is, 3.52 and 3.93 μm, respectively). Full size image

Small coccoliths show evidence for decreased HCO 3 − allocation to calcification only since 6 Myr ago. From 11 to 6 Myr ago, ɛ coccolith and SN coccolith thickness are relatively stable (Fig. 5b,c), suggesting minimal changes in HCO 3 − allocation to calcification. In contrast, between 6 and 1 Myr ago, a near-constant ɛ coccolith indicates a stable ratio of HCO 3 − allocation to the coccolith vesicle relative to calcification rate, despite a large decrease in degree of calcification (Figs 4 and 5c). This implies a decrease in HCO 3 − allocation to calcification of comparable magnitude to the decrease in cellular calcification. In the last 1 Myr, an increase in degree of calcification in the small coccoliths with no change in ɛ coccolith suggests that allocation of HCO 3 − to calcification also increased in parallel.

Relationship between calcification and ocean stratification

Water column stratification influences productivity and production depth in the tropics. Stratification can be inferred from foraminiferal δ18O gradients between the upper mixed layer (Globigerinoides sacculifer) and thermocline (Globorotalia menardii), because these reflect the upper photic zone temperature and salinity gradients24. The temporal evolution of planktic foraminiferal δ18O at Sites ODP 925 and NGHP-01-01A is shown in Fig. 6. Between 3.5 and 2 Myr ago, a deep thermocline at Site 925 is inferred from independent foraminiferal assemblage indicators25, potentially suggesting a deeper coccolithophore depth habitat and lower light levels. Decreased light has been shown to reduce cellular calcification (PIC/SA) twofold by a reduction in photon flux density from 80 to 15 μmol m−2 s−1 in culture26, and low light levels have been proposed to decrease cellular HCO 3 − transport27. However, neither site shows a clear decrease in δ18O gradients at this time (Fig. 6a,b), as would be expected if reduced coccolith calcification from 4 to 1 Myr ago were due to a deepening of the thermocline, resulting in a reduced temperature gradient between the two foraminifer species’ depth habitats. Proxy records suggest high productivity from 10 to 8 Myr ago in the Indian Ocean28 and from 6.6 to 6 Myr ago at ODP Site 925 (ref. 29). Thus, reconstructed changes in water column structure and paleoproductivity do not consistently co-vary with changes in degree of coccolith calcification.

Figure 6: Foraminiferal stable isotope records as indicators of water column structure. δ18O (a,b) and δ13C (c,d) records for surface (G. sacculifer), thermocline (G. menardii) and benthic species at Sites ODP 925 and NGHP-01-01A. Benthic isotope data for our sites are not available, therefore values were interpolated using a global compilation86 separated into ocean basins. Because Neogene Indian Ocean data are sparse and trends are very similar to those in the Pacific, a compilation of Indian and Pacific Ocean data were used to interpolate the benthic values in a and c. For b and d, a compilation of all Atlantic data was used. Note: benthic δ18O is plotted on a different y axis. Shading indicates the gradient between surface and thermocline-dwelling planktic foraminiferal species. Full size image

Calcification and [CO 2aq ] in the Miocene–Pliocene

Carbon isotopic fractionation in phytoplankton during photosynthesis (ɛ p ) varies directly with [CO 2aq ] and has been widely applied as a CO 2 proxy in the Cenozoic. However, limited data exist for the interval of major changes in calcification and HCO 3 − allocation between 14 and 5 Myr ago. In addition, the interpretation of any data is complex because of the expected influence of active HCO 3 − allocation on ɛ p (ref. 14). Our new record of ɛ p extends the published record from ODP Site 999 for the last 5 Myr30 back to 16 Myr ago (Fig. 7a). This extended record reveals a decrease in ɛ p from 16 to 8 Myr ago, an excursion to higher ɛ p values at 7 Myr ago, and then a continued decrease towards the present. The decline in ɛ p could be driven by decreasing [CO 2aq ], increasing cellular growth rates that increase carbon demand relative to supply, or increasing cell sizes that reduce surface area to volume and thus diffusive supply (see ref. 31 and references therein). Following previous workers, [CO 2aq ] is estimated with the formula [CO 2aq ]=b/(ɛ f −ɛ p ), where ɛ f is a constant reflecting the maximum effective photosynthetic fractionation by the cell (25‰), and ‘b’ encompasses factors such as growth rate and cell geometry that modulate the ratio of carbon supply to demand by the cell. First, to estimate temporal variations in b due to cell size, we use previous formulations of the relationship between cell size and b32, together with our record of tropical Noëlaerhabdaceae coccolith size evolution (Fig. 7b), which shows trends similar to those at other tropical sites22,23. The decrease in cell size after 9 Myr ago, compared with the average between 11 and 16 Myr ago, corresponds to a 25% reduction in the b value. Second, we estimate the influence of productivity on b using proxy records from ODP Site 999 of coccolith Sr/Ca and alkenone mass accumulation rates (Fig. 7c). These records confirm that there is no long-term productivity increase, and suggest maxima from 13 to 10 Myr ago and at 8 Myr ago. Calculated b values are shown in Fig. 7d. The resulting estimates of [CO 2aq ] (Fig. 7e) show a trend of continued decline over the past 16 Myr, with the exception of a local maximum at 9.3–10.3 Myr ago resulting from the unusually large cell sizes in the geometry correction. Assuming equilibrium with the atmosphere, these results are similar in trend and magnitude to [CO 2aq ] predicted from the atmospheric pCO 2 curve of ref. 33 derived from inverse modelling of climate data (Fig. 7e). The absolute values of [CO 2aq ] are subject to greater uncertainty than the trend.

Figure 7: ɛ p values and estimates of b and [CO 2aq ] at Caribbean ODP Site 999 and other sites for the last 17 Myr. (a) New ɛ p data (Site 999, red diamonds: SST max, orange diamonds: SST min; error bars show propagated analytical uncertainty on δ13C measurements). Published ɛ p records: Site 999 (ref. 30) (purple crosses), ODP Site 925 (ref. 37) (grey circles), DSDP Sites 588 (ref. 84) (grey triangles, maximum ɛ p ) and 608 (ref. 84) (blue squares). (d) Variations in b (Site 999) inferred to arise from changing cell size (b) and growth rate (c) (see Methods). In c, triangles (this study) and line30 show alkenone MARs and blue shading shows Sr/Ca productivity estimates for small coccolithophores14 (all Site 999). In d, purple crosses (Site 999 (ref. 30)), grey circles (Site 925 (ref. 37)), and orange circles (Site 999, this study) show b values recalculated using our new cell size correction. Red circles (Site 999, this study) show b values calculated with cell size and growth rate corrections. For error calculations, see Methods. (e) [CO 2aq ] calculated using cell size (orange circles), or cell size plus growth rate (red circles), correction and ɛ p values (Site 999, this study). [CO 2aq ] was also recalculated using our cell size correction for the Plio-Pleistocene at Site 999 (ref. 30) (purple crosses) and Site 925 (ref. 37) (grey circles). For all sites, reference b=150. [CO 2aq ] assuming constant b for Site 999 (ref. 30) is also shown (blue crosses). Shading indicates maximum and minimum [CO 2aq ] estimates for all data from Site 999 (see Supplementary Methods). We do not apply our size correction to DSDP Sites 608 and 588 ɛ p data because these sites are at significantly higher latitudes; therefore cell size history may be different compared with the tropical sites studied here. Also shown in e is the [CO 2aq ] expected for the Caribbean site if it were in equilibrium with the atmospheric pCO 2 modelled by ref. 33 (grey line). (f) pH derived from δ11B of planktic foraminifers, for the Plio-Pleistocene30,36 and Miocene53. During the Miocene, ODP Site 999 ɛ p values are similar to values at ODP Site 925 (ref. 37) and higher than values from DSDP Sites 588 and 608 (ref. 85). From 16 to 9 Myr ago, the maximum ɛ p at DSDP Site 608 shows a similar trend to ɛ p at ODP Site 999, albeit with slightly lower absolute values, suggesting that either both sites experienced similar changes in growth rates, or that a global CO 2 component exerted a dominant forcing on both ɛ p records. The temporally variable scatter to low ɛ p values seen in the Site 608 record may result from higher frequency oscillations in growth rates at this site83. The much lower average ɛ p at Site 588 suggest that this site experienced on average higher phytoplankton growth rates and productivity compared to Sites 925, 999 and 608. Full size image

As in previous studies30, our calculations would not account for the likely increase in active carbon uptake for photosynthesis as [CO 2aq ] declined14,34, especially after 8 Myr ago. Because active carbon transport increases the chloroplast uptake of inorganic carbon relative to fixation, it can result in higher ɛ p values than would be predicted from passive diffusive CO 2 uptake alone35. Laboratory culture experiments suggest that active HCO 3 − transport to the chloroplast becomes more significant at low [CO 2aq ]. Simulations with the ACTI-CO model of HCO 3 − transport in coccolithophores14 were used to evaluate the potential impact of changes in active carbon uptake on ɛ p and calculated [CO 2aq ] (Supplementary Methods; Fig. 8). In one set of simulations, we specify a logarithmic dependence of chloroplast HCO 3 − transport/diffusive CO 2 uptake on [CO 2aq ] as observed in culture experiments14,27. Alternatively, if enhancement of HCO 3 − transport to the chloroplast is coupled, in part, to reallocation of HCO 3 − from the coccolith vesicle, as inferred from modelling of cultures14, our new ɛ coccolith and SN coccolith thickness data put additional constraints on the timing of this reallocation. Therefore in a second set of simulations, we specify chloroplast HCO 3 − transport based on HCO 3 − spared from the coccolith vesicle by the reduction in cellular calcite in the last 8 Myr. We then derive the [CO 2aq ] implied by measured ɛ p for the specified parameterization of active HCO 3 − uptake to the chloroplast. The results in both cases indicate a greater amplitude of decline in [CO 2aq ] compared with that reconstructed with standard cell size and growth rate considerations only, from around 17 to 6 μM (Fig. 8).

Figure 8: Simulations of the effect of active HCO 3 − uptake on reconstructed [CO 2aq ]. ACTI-CO model simulation for two potential scenarios of active HCO 3 − uptake to the chloroplast (b), and consequences for [CO 2aq ] implied by alkenone ɛ p measurements (c). A first simulation (unfilled circles) employs a logarithmic dependence of HCO 3 − transport to the chloroplast on [CO 2aq ], similar to that observed in cultures (a). A second simulation (filled circles) supplements HCO 3 − supply to the chloroplast as a function of HCO 3 − spared from the coccolith vesicle by the production of thinner coccoliths (that is, a reduced PIC per cell surface area). Crosses in c show [CO 2aq ] estimated from standard regressions between ɛ p and [CO 2aq ] as shown in Fig. 7e (orange circles). Full size image

Calcification in relation to CO 2 and alkalinity since 1 Myr ago

In the last 1 Myr, climate, the carbon cycle and ocean chemistry underwent significantly higher amplitude variations on glacial–interglacial timescales compared with the preceding 15 Myr. Although our sampling resolution does not capture this higher frequency variability, and does not sample very recent major evolutionary events such as the emergence of E. huxleyi, our results nonetheless suggest in the Pleistocene a reversal of the late Miocene–Pliocene trend of more lightly calcified coccoliths and decreasing HCO 3 − allocation to the coccolith vesicle. Within the last 1 Myr, both of these factors rebound to values typical of the late Miocene (Fig. 5). Records based on boron isotopes36, alkenone δ13C (refs 30, 37), and ice cores38 suggest pCO 2 values below around 280 p.p.m. over the last 2 Myr, so this increase in degree of calcification contrasts with the generally positive correlation of [CO 2aq ] and degree of calcification observed over the preceding interval.

The change in relationship between degree of coccolith calcification and [CO 2aq ] is even more salient when we examine which samples fall in glacial or interglacial ocean states. Planktic foraminiferal δ18O values from samples at Indian Ocean Site NGHP-01-01A and the orbital age model for Atlantic ODP Site 925 indicate that our youngest samples at both sites (about 0.27 Myr ago), with high SN coccolith thickness, coincide with glacial periods (Supplementary Fig. 6). The sample at 0.84 Myr ago from Site NGHP-01-01A with high SN coccolith thickness also falls during a glacial period, whereas the sample at 0.95 Myr ago from the Site 925, with lower SN coccolith thickness, falls in an interglacial. These particular sampling points therefore underscore the nature of a change in the relationship between degree of calcification and pCO 2 , as the samples with thicker coccoliths in a given size class are from glacial periods that coincide with pCO 2 minima in the last 800 kyr (Supplementary Fig. 6).