Coccoliths, calcite plates produced by the marine phytoplankton coccolithophores, have previously shown a large array of carbon and oxygen stable isotope fractionations (termed “vital effects”), correlated to cell size and hypothesized to reflect the varying importance of active carbon acquisition strategies. Culture studies show a reduced range of vital effects between large and small coccolithophores under high CO2, consistent with previous observations of a smaller range of interspecific vital effects in Paleocene coccoliths. We present new fossil data examining coccolithophore vital effects over three key Cenozoic intervals reflecting changing climate and atmospheric partial pressure of CO2 (pCO2). Oxygen and carbon stable isotopes of size-separated coccolith fractions dominated by different species from well preserved Paleocene-Eocene thermal maximum (PETM, ∼56 Ma) samples show reduced interspecific differences within the greenhouse boundary conditions of the PETM. Conversely, isotope data from the Plio-Pleistocene transition (PPT; 3.5–2 Ma) and the last glacial maximum (LGM; ∼22 ka) show persistent vital effects of ∼2‰. PPT and LGM data show a clear positive trend between coccolith (cell) size and isotopic enrichment in coccolith carbonate, as seen in laboratory cultures. On geological timescales, the degree of expression of vital effects in coccoliths appears to be insensitive topCO2 changes over the range ∼350 ppm (Pliocene) to ∼180 ppm (LGM). The modern array of coccolith vital effects arose after the PETM but before the late Pliocene and may reflect the operation of more diverse carbon acquisition strategies in coccolithophores in response to decreasing Cenozoic pCO2.
 Coccolithophorids, unicellular haptophyte algae, play an important role in ocean biogeochemistry because they utilize dissolved inorganic carbon (DIC) in surface waters for both photosynthesis (the ‘soft tissue carbon pump’) and calcification, producing coccoliths (the ‘inorganic carbon pump’). Coccoliths are precipitated intracellularly then extruded through the cell membrane, creating a characteristic carbonate external covering that is preserved in marine sediments [Pienaar, 1994; Brownlee and Taylor, 2004]. The dual role of coccolithophores in the marine carbon cycle, their long geological history (∼225 Ma to present [Janofske, 1992; Bown, 1998; Bown et al., 2004]) and rapid evolutionary turnover [Falkowski et al., 2004] make this extant phytoplankton group ideal for investigating biotic responses to past, present and projected carbon cycle perturbations through geochemical studies.
 Modern algal cells are known to use various carbon concentrating mechanisms (CCMs) to elevate CO2 in the chloroplast and increase photosynthetic rate (reviewed by Giordano et al.  and Reinfelder ). CCMs include active HCO3− and CO2 uptake, or enhancement of diffusive CO2 uptake with extracellular excretion of the enzyme carbonic anhydrase (CA) to catalyze the conversion between HCO3− and CO2 [e.g., Raven and Johnston, 1991; Badger et al., 1998; Tortell, 2000; Rost et al., 2003]. Culture experiments with the coccolithophore Emiliania huxleyi indicate that photosynthesis is not dependent on calcification as a CO2source (i.e., a CCM) under nutrient and light-replete conditions [Leonardos et al., 2009]; however cells may use calcification as an energetically cheap way to generate CO2 for photosynthesis under nutrient or light limitation [Reinfelder, 2011; Schulz et al., 2007; Sciandra et al., 2003]. The need for efficient CCMs is greatest (1) in large cells because they have a smaller surface area to volume ratio, therefore lower diffusive CO2 transport rates into the cell relative to carbon demand, and (2) at low CO2 concentrations which limit diffusive CO2 supply; as has been confirmed by numerous laboratory studies in diverse marine alga [e.g., Giordano et al., 2005; Falkowski and Raven, 2007; Moolna and Rickaby, 2012; Reinfelder, 2011]. We therefore hypothesize, following previous work [Stoll, 2005], that during high pCO2 intervals CCM significance diminishes, possibly translating into a reduced range of stable isotope signatures among different sized coccoliths. In a recent study, the difference in coccolith δ18O and δ13C between a large-celled and a small-celled coccolithophore was reduced at higher culture CO2(aq) [Rickaby et al., 2010] (Figure 1b).
 Long-term reduction inpCO2 over the Cenozoic (65.5 to 0 Ma) [Pagani et al., 2005; Royer, 2006; Zachos et al., 2008] may have forced the evolution of adaptive strategies in phytoplankton. A much smaller range of δ18O and δ13C among large and small coccoliths was observed during the Paleocene-Eocene thermal maximum (PETM, ∼56 Ma) and could reflect more uniform carbon acquisition strategies at higher ambientpCO2 [Stoll, 2005]. Falling pCO2 since the early Oligocene may have exerted evolutionary pressure toward a reduction in coccolithophore cell size within some genera [Henderiks and Pagani, 2008]. In addition, an observed global decrease in mean assemblage coccolith size over the Cenozoic could result from evolution, abundance and extinction patterns of different-sized species [Herrmann and Thierstein, 2012]. To explore coccolith stable isotopic fractionation responses to changing pCO2in the past, we present new stable isotope and productivity records of size-separated fossil coccoliths from (1) a new site of latest Paleocene age, (2) the Plio-Pleistocene climate transition (PPT, 3.5 to 2 Ma), and (3) the last glacial maximum (LGM, ∼22 ka). For the latest Paleocene, we seek to verify the previous inference that reduced interspecific isotopic differences at Ocean Drilling Program (ODP) Site 690 [Stoll, 2005] were not biased by diagenetic homogenisation and/or large coccolithophore assemblage changes by presenting new coccolith geochemical records from Bass River, New Jersey. We also examine in detail the PPT, which marks the end of long-term cooling culminating in major northern hemisphere glaciation (NHG) [e.g.,Shackleton et al., 1984; Flesche Kleiven et al., 2002; Lawrence et al., 2009] and encompasses the last major secular pCO2 decline in Earth's history (∼420 to 280 ppm pCO2 decrease estimated from proxies; Figure 2) [Raymo et al., 1996; Pagani et al., 2010; Seki et al., 2010; Bartoli et al., 2011]. Our new records show clear, significant (>1.5‰) interspecific vital effects during the PPT and LGM and confirm only minimal differences during the PETM.
2. Materials and Methods
2.1. Samples and Site Selection
 We analyzed sediments deposited at Bass River (BR), New Jersey, during the PETM (ODP Site 174AX, 39°36′N, 74°26′W) [Miller et al., 1998] to test whether the absence of significant interspecific isotopic differences reported for PETM-aged coccoliths from Site 690 [Stoll, 2005] is a robust environmental signal. Sampling was between 352 m and 373 m depth at 10–20 cm resolution spanning the onset of the PETM negative carbon isotope excursion (CIE) and at 50–200 cm resolution before and during the CIE. Sedimentation rates during the PETM were rapid (∼10 cm/kyr [Röhl et al., 2007]) and silty claystones deposited in a shallow-shelf environment (well above the carbonate lysocline) dominate Paleogene sediments [Miller et al., 2004; John et al., 2008]. The high clay content of these sediments minimizes carbonate dissolution that is intrinsic to deep-ocean PETM sections [Zachos et al., 2005], making BR an ideal record to test the findings of Stoll .
 To investigate coccolithophore vital effects under lower pCO2conditions, we analyzed PPT and LGM sediments deposited at ODP Site 999 in the Caribbean Sea (12°44′N, 78°44′W; 2830 m water depth). Sedimentation at Site 999 is predominantly pelagic with minor aeolian inputs. Plio-Pleistocene sediments constitute clayey Foraminifera and nannofossil ooze with variable carbonate content (40–70%) [Shipboard Scientific Party, 1997]. We analyzed 22 samples from Site 999A spanning 110.85 to 69.27 m corrected depth. An orbital-resolution age model was generated via manual correlation of Site 999 benthic foraminiferalδ18O [Haug and Tiedemann, 1998] to the LR04 benthic δ18O stack [Lisiecki and Raymo, 2005] using Analyseries [Paillard et al., 1996]; yielding sedimentation rates of ∼2.4–3.6 cm/kyr during our study interval, ∼3.5 to 2 Ma. To capture long-term trends in coccolith geochemistry and to minimize variability arising from glacial-interglacialpCO2 and climate oscillations, which also induce primary productivity fluctuations [e.g., Bolton et al., 2010b], samples were selected from peak interglacials (IG). Samples span marine isotope stages MG7 to 79, with one sample from every or every other IG, depending on sedimentation rate. In the modern ocean, surface waters overlying Site 999 are close to equilibrium with the atmosphere with respect to CO2 [Takahashi et al., 2009], a scenario that probably prevailed throughout the last glacial cycle [Foster, 2008]. Our surface mixed-layer signal will therefore be dominated by large-scale changes inpCO2 rather than local variations caused by, for example, changes in vertical water column mixing. To compare our PPT record with data measured on coccoliths deposited under more recent minimum pCO2 conditions, we analyzed LGM sediments from the same site (ODP 999A 1 H 1 W 100–101 cm; ∼22 ka age [Schmidt et al., 2004]).
2.2. Coccolith Size Separations
 Bulk sediment samples were gently disaggregated for 24 h on a ‘ferris wheel’ rotating carousel (designer: Nick McCave, personal communication, 2000), then sieved through a 20 μm mesh. A split of the <20 μm fraction was used for coccolith size separations, with all steps performed in 2% ammonia solution. Sample-specific repeat decanting protocols followingPaull and Thierstein  and Stoll and Ziveri  were developed to most efficiently separate the species present by size using differences in settling velocities (Figure 3). A high number of repeats of each step ensured greater separation efficiency. The use of material remaining in suspension rather than settled material reduced contamination by large coccoliths in the smaller size fractions, where a few large coccoliths can skew carbonate contributions to a high degree. For each repeat, the supernatant was carefully removed by plastic syringe, filtered onto a 0.45 μm nitrate cellulose filter and resuspended in 2% ammonia. Samples were rinsed 3 times in ultrapure MilliQ water and oven-dried at 50°C for several days.
 The high species diversity and low carbonate content of BR sediments meant that we focused on obtaining size-restricted (rather than monogeneric) coccolith size fractions (Table 1). A long settling step was first applied to remove fragments <1.5 μm. The size fractions 1.5–5 μm, 5–8 μm and 8–20 μm were obtained via settling and subsequent microfiltration (at 8 μm) of the settled material. At Site 999, a similar coccolith size distribution during the LGM and the PPT allowed us to use an identical separation protocol for all samples (Figure 3 and Table 2). In PPT samples, we evaluated the effect of contamination by Foraminifera fragments on the isotopic signature of the larger coccolith size fractions (9–12 μm and to a lesser extent 6–9 μm), by determining δ18O and δ13C of the 20–63 μm fraction.
Table 1. Main Constituents of Coccolith Size Fractions Expressed as Carbonate Contribution at Bass River During the PETM in Six Samples Covering the Interval of Maximum Change in Bulk δ13Ca
PETM Bass River
Percent Carbonate Contribution (%)
Placoliths are grouped into three size categories (see section 3.1 for species' details). Where one group/genus heavily dominates carbonate, values are in bold.
v. small coccolith fragments
V. Small placoliths (mean diameter 1.5 μm)
Small placoliths (mean diameter 2.7 μm)
Medium placoliths (mean diameter 5 μm)
V. Small placoliths (mean diameter 1.5 μm)
Small placoliths (mean diameter 2.7 μm)
Medium placoliths (mean diameter 5 μm)
V. Small placoliths (mean diameter 1.5 μm)
Small placoliths (mean diameter 2.7 μm)
Medium placoliths (mean diameter 5 μm)
Large Foraminifer Fragments
Table 2. Main Constituents of Coccolith Size Fractions Expressed as Carbonate Contribution at Site 999 for PPT Samples 2 (2180 ka), 9 (2660 ka) and 21 (3392 ka) and LGM Samplea
PPT ODP Site 999
Percent Carbonate Contribution (%)
Values are only shown where contribution exceeds 10%. Where one species or genus heavily dominates carbonate, values are in bold. * LGM only.
Reticulofenestra (<3 μm)
Gephyrocapsa + Emiliania (<3 μm)
Reticulofenestra + Pseudoemiliania (<4 μm)
Gephyrocapsa + Emiliania (<4 um)
Reticulofenestra + Pseudoemiliania (<5 μm)
Gephyrocapsa + Emiliania (<5 μm)
Large Gephyrocapsa (>5 μm)
Oolithotus fragilis + Calcidiscus
2.3. Microscope Counts and Percent Carbonate
 All calcareous nannofossil counts were performed on smear slides on a light microscope (LM) under cross-polarized light at x1250 magnification, following the taxonomy ofPerch-Nielsen  and Young . Hereafter, coccoliths belonging to the Noelaerhabdaceae family (Plio-Plesistocene generaReticulofenestra, Pseudoemiliania, Gephyrocapsa and Emiliania) are collectively termed ‘reticulofenestrid coccoliths’ owing to their similar coccolith structure to Reticulofenestra [Young, 1998]. A minimum of 400 coccoliths from at least 10 fields of view were counted (>600 coccoliths in PPT < 20 μm samples because of the dominance of Florisphaera profunda and reticulofenestrid coccoliths). Using relative abundances, mean coccolith lengths (determined by LM) and shape constants (ks values) [Young and Ziveri, 2000], coccolith volume and carbonate contributions were calculated. Where a ks value for the exact species or genus was unavailable, the closest species was used (e.g., Gephyrocapsa oceanica for Reticulofenestra). Although cumulative errors in these calculations are typically large [Young and Ziveri, 2000], carbonate contribution data give a much clearer idea of the origin of the isotopic signals contained in a mixed-species sample than relative abundance data, where equal weight is given to all coccoliths regardless of size or shape.
 For the PETM samples, whole-assemblage calcareous nannofossil abundance data were available [Gibbs et al., 2010; S. J. Gibbs, personal communication, 2012], thus counts were performed on the three size fractions of six samples covering the interval of maximum change in assemblage and bulk carbonate isotope composition (356.95 m to 357.90 m). Abundances were converted to carbonate contributions as described above (Figure S1b and Table 1). For all PPT samples, counts were performed on the <20 μm fraction to determine bulk assemblage composition (Figure S2a). Relative abundances of F. profunda and very small Reticulofenestra sp. (VSR, <3 μm), two groups with similar dissolution susceptibilities [Gibbs et al., 2004], were used in the N ratio (N = VSR/VSR + Fp) as an indicator of coccolithophore productivity [Beaufort et al., 1997; Flores et al., 2000; Bolton et al., 2010a]. To obtain a reliable approximation of species carbonate contribution in each size fraction throughout the time series, counts were performed on all five size fractions for PPT samples 2 (2180 ka), 9 (2660 ka) and 21 (3392 ka) and the LGM sample (Table 2). For the two smallest size fractions, essentially constituting mixtures of only small reticulofenestrid (<4 μm) and F. profundacoccoliths, end-member isotopes were calculated using carbonate contributions (for the PPT, mean values of the counted samples were used; seeTable 2).
2.4. Stable Isotope Analysis
 Stable isotope ratios on BR samples were measured at Woods Hole Oceanographic Institute on a Finnegan MAT 253 dual-inlet isotope ratio mass spectrometer (DI-IRMS) with a Kiel III Carb Device with analytical precision of 0.08‰ forδ18O and 0.03‰ for δ13C (1σ). A higher-resolution sample set between 358.81 m and 356.95 m was later run at the University of California Santa Cruz on a GVI Prism DI-IRMS with an analytical precision of 0.08‰ forδ18O and 0.05‰ for δ13C. Different size fractions from the same sample were always analyzed together, thus potential interlab offsets cannot affect inter-fraction isotopic differences. PPT and LGM stable isotope data were generated on a Nu Instruments Perspective DI-IRMS connected to a NuCarb carbonate preparation system at the University of Oviedo with analytical precision of 0.06‰ forδ18O and 0.05‰ for δ13C (1σ). Duplicate analyses show mean reproducibility of 0.07‰ (δ18O) and 0.05‰ (δ13C) (1σ). 1 mg splits of PPT 20–63 μm fractions were analyzed on a Nu Instruments Horizon continuous flow (CF)-IRMS connected to a gas preparation system at the University of Oviedo with analytical precision of 0.1‰ forδ18O and 0.08‰ for δ13C (1σ). All stable isotopes are reported relative to the Vienna Pee Dee Belemnite (VPDB) standard.
 Sr/Ca in coccoliths was measured to define productivity (growth rate) variations that could affect interpretation of temporal or interspecific trends in stable isotope data. In BR samples (where clays dominate sediments thus reliable data cannot be obtained from bulk sediment analysis) Sr/Ca was measured by ion probe analysis on monospecific populations of 12–20 individually picked coccoliths (see Stoll et al. [2007b] and Stoll and Shimizu  for methods). Populations of small and medium Toweius sp. and Coccolithus pelagicus coccoliths were measured. A subset of the medium Toweius Sr/Ca data appears in Gibbs et al. .
 In PPT samples, Sr/Ca was measured in the small reticulofenestrid (2.5–4 μm) and large Helicosphaera (6–9 μm) coccolith size fractions. Subsamples were cleaned with reducing and ion exchange treatments [see Stoll and Ziveri, 2002] then gently dissolved in acetic acid with an ammonium acetate buffer for 12 h. Calcium content was analyzed on splits of all samples, which were then diluted to constant calcium concentrations for Sr/Ca analysis by inductively coupled plasma optical emission spectroscopy (ICP-OES) on a Thermo ICAP DUO 6300 at the University of Oviedo. Although Sr/Ca in coccoliths varies with growth rate, there is also a known temperature influence [Stoll and Schrag, 2000; Rickaby and Schrag, 2002; Stoll et al., 2002, 2007c]. To isolate the productivity component, variance attributable to sea surface temperature (SST) was subtracted from Sr/Ca. We used the Globigerinoides sacculiferMg/Ca-derived SST record for Site 999 [Groeneveld, 2005], resampled and interpolated to the same depths as our samples, and Sr/Ca-temperature relationships derived forGephyrocapsa oceanica and Helicosphaera carteri [Stoll et al., 2002].
3.1. Coccolith Separation Efficiency
 In BR samples, size-separation efficiency was excellent for medium-sized coccoliths (5–8μm) and modest for small-sized coccoliths (1.5–5μm). The reduction in efficiency observed for small coccoliths occurred because a minor percentage of medium-sized liths (<10% rel. abundance) make a disproportionately large mass contribution to this size fraction (e.g., a 5μm lith weighs 57 pg versus 7 pg for a 2.7 μm lith). The BR 8–20 μm fraction was dominated (>50% carbonate mass) by Foraminifera fragments therefore was not used further in this study (Table 1 and Figure S1b). The small coccolith fraction (1.5–5 μm) predominantly contains small Prinsiaceae coccoliths, mainly Toweius species. A large influx of small (<3 μm) placoliths at the CIE onset alters the relative contribution of small versus larger coccoliths. However, mutual relative abundances of medium and large placoliths (Coccolithus pelagicus, Toweis pertutus, other Toweis sp., and Prinsius sp.), the constituents of our 5–8 μm coccolith size fraction, remain similar (Figure S1a) [Gibbs et al., 2010; S. J. Gibbs, personal communication, 2012]. Thus, our size-restricted BR coccolith fractions have relatively constant species composition despite assemblage changes over the PETM interval.
 Separation efficacy of Site 999 coccoliths was very good, resulting in five distinct size-classes. In three of these, carbonate was dominated (65–90%) by a single species or family (Table 2). For the PPT, the <2.5 μm fraction is dominated by F. profunda coccoliths and the 2.5–4 μm fraction by small reticulofenestrid coccoliths (Reticulofenestra and to a lesser extent Pseudoemiliania). The 4–6 μm fraction contains medium-sized coccoliths of mixed species, including largerReticulofenestra, Pseudoemiliania, Umbilicosphaera, and smaller Helicosphaera coccoliths. The 6–9 μm fraction is dominated by Helicosphaera coccoliths, predominantly H. carteri. In the 9–12 μm fraction, Helicosphaera coccoliths dominate in younger samples, with increasing contribution from Discoaster nannoliths in older samples. Foraminifera fragments contribute to the 9–12 μm fraction throughout, with slightly decreasing abundance downcore. Relatively uniform bulk assemblage composition and size distribution mean that the composition of size fractions is relatively constant from 3.5 to 2 Ma (Table 2 and Figure S2). LGM size fractions are dominated by the same species or genera as PPT samples, with the following exceptions: the 2.5–4 μm fraction is dominated by the genera Gephyrocapsa and Emiliania, and the 9–12 μm fraction is almost entirely composed of dinoflagellate calcispheres of the Thoracosphaeraceae family (Table 2). The high efficiency of size separation achieved for coccoliths from the PETM, PPT and LGM allowed us to generate stable isotope data on a large range of narrow size classes, permitting meaningful interpretation of geochemical data in terms of ancient coccolithophore cell size.
3.2. Stable Isotopes
 PETM coccoliths faithfully record the CIE at BR with a ∼3‰ negative shift in δ18O and δ13C in both size fractions around 359 m, coincident with the CIE onset recorded in bulk and foraminiferal carbonate and with similar values to those measured in bulk carbonate (Figures 4a and 4b). Coccoliths record a similar-sizedδ13C shift as the surface-dwelling ForaminiferaAcarinina sp. and a larger shift than recorded in the benthic Foraminifera Cibicidoides sp. [John et al., 2008]. Critically to this study, coccolith stable isotopes exhibit negligible differences between small (1.5–5 μm) and larger (5–8 μm) size fractions for δ13C (mean difference 0.17‰), with smaller coccoliths in most samples showing a slight δ18O enrichment relative to larger coccoliths (mean difference 0.66‰) (Figures 4a and 4b).
 We isolate temporal changes in interspecific isotopic differences in our Plio-Pleistocene data by removing the imprint of cooling and ice sheet expansion associated with NHG; achieved by subtracting the stable isotopic composition of co-existing planktic Foraminifera (Globigerinoides sacculifer) from our PPT and LGM coccolith isotope data (Figure 5; uncorrected data, Figure S3). Our Plio-Pleistocene coccolith fractions typically exhibit a much greater range of isotopic values compared to our PETM fractions. Data showδ18O and δ13C ranges of ∼1.5 to 2‰ (PPT) and ∼1.3 to 1.5‰ (LGM) between smallest and largest coccolith size fractions, with δ18O and δ13C values increasing with decreasing coccolith size (Figures 5, 6a, and 6e). The δ18O and δ13C ranges between fractions do not show long-term temporal trends over the PPT (Figures 5, 6, and 7c). Two samples around 2.6 Ma contain a reduced array of isotopic values because more negative δ18O and δ13C are recorded by the 2.5–4 μm fraction (Figure 5, pink shading in Figure 7c). LM inspection of this size fraction in all PPT samples indicates an isolated change in species composition, with these two samples containing (1) a greater proportion of Pseudoemiliania ovata (∼4 μm) relative to Reticulofenestra minuta (<3 μm) coccoliths and (2) more dislocated Discoaster arms than other samples. These factors are both consistent with the more negative isotopic signature recorded; therefore we interpret data in this interval to reflect a change in carbonate source for some size fractions rather than a contraction of interspecific differences. The 20–63 μm fraction records more positive δ18O and δ13C than the two largest coccolith fractions (Figures 5 and 6), thus the presence of Foraminifera fragments in these fractions potentially reduces the measured range of isotopic values between small and large coccoliths. Increasing foraminiferal fragment abundance upcore could also mask an increase in the isotopic array between different sized coccoliths driven by more depleted values in the large coccoliths.
 We plot PPT coccolith isotopes with Site 999 Foraminifera isotopes to illustrate the distinct trends in the two calcifier groups (Figure 8; see caption for details). Foraminifera record an inverse correlation between δ18O and δ13C, whereas coccoliths record a positive δ18O-δ13C correlation. The δ18O signature of the 20–63 μm fraction is consistent with a mixed-species planktic Foraminifera origin; however the depletedδ13C in this size fraction suggests additional contribution from large coccoliths (confirmed by LM; samples were not ultrasonicated to remove adhering coccoliths and sieving at 20 μm can be inefficient given the small open area of the mesh). In Figure 9, we illustrate BR coccolith and Foraminifera isotopes as mean values (see caption for details). Like at Site 999, Foraminifera display an inverse δ18O-δ13C correlation. Conversely, small versus larger coccolith size-fractions show very similarδ13C values and a δ18O range of <0.75‰. As expected due to the dominance of coccolith carbonate, bulk sediment records a mean isotopic signature closer to that of coccoliths than Foraminifera (Figure 9).
 PETM coccolith Sr/Ca ranges between 1.8 and 2.8 mmol/mol. Before and after the CIE onset interval (∼355–359 m), coccolith Sr/Ca is variable and high values (elevated productivity) coincide with low sea level as inferred from dinoflagellate cyst assemblages (Spiniferites to Areoligeraratio, S-A ratio) (Figure 4) [Sluijs, 2006]. Where Sr/Ca data exist for small and medium Toweius and large Coccolithus, similar trends and values are observed, with an initial decrease in Sr/Ca at the CIE onset followed by two peaks then decreasing values (Figure 4c). During this early PETM interval (355–358 m), the relationship between coccolith Sr/Ca and the dinocyst S-A ratio deviates from the previous trend in showing high productivity despite high sea levels (Figure 4). BR coccolith stable isotopes and Sr/Ca are uncorrelated (R2 < 0.1).
 PPT Sr/Ca in the small reticulofenestrid and Helicosphaera coccolith fractions varies in the range 2.17–2.53 mmol/mol and 2.09–2.75 mmol/mol respectively, with both time series following similar trends but with greater amplitude variability in Helicosphaera Sr/Ca (Figure 7a, dashed lines). Removal of the SST component from Sr/Ca, leaving a residual record attributed to productivity, does not alter the main trends or amplitude variation in the uncorrected Sr/Ca (Figure 7a, solid lines). Productivity maxima occur at ∼400 ka intervals (Figure 7a). The independent N ratio productivity proxy shows similar trends to Sr/Ca productivity records; however maxima and minima are sometimes offset between floral and geochemical records (Figure 7b). Although certain features in coccolith isotope and productivity records appear coincident (Figure 7), no significant correlations were found (R2 < 0.2).
4.1. Are Coccolith Isotope Variations Due to Depth Habitats?
 Variations in the δ13C of DIC, SST, salinity and CO32−concentration throughout the photic zone are expected to influence the equilibrium isotopic composition of carbonate formed at different depths. Thus, deeper-dwelling calcifiers record the more positiveδ18O of colder waters and a more negative δ13C than upper photic zone (UPZ) species because of remineralisation of 12C-enriched organic matter at depth [e.g.,Fairbanks et al., 1982; Rohling et al., 2004]. Oceanographic studies found that most coccolithophores inhabit the UPZ (0–100 m), with a few specialized species, e.g., F. profunda and Thorosphaera flabellata, living in the deep photic zone (DPZ, >100 m) [Okada and Honjo, 1973; Honjo and Okada, 1974]. Our F. profunda (smallest) coccolith fraction records the heaviest δ18O of all fractions, as predicted by its deep habitat; however it also records the most positive δ13C, suggesting that depth habitat does not exert primary control on its isotopic composition (Figures 5 and S3). Furthermore, the interspecific δ13C range recorded may be reduced as a result of the depth habitat of F. profunda relative to UPZ coccolithophores. At Site 999 and BR, trends in coccolith isotopes are distinct from trends predicted by depth habitat, as illustrated by the isotopic signatures of coexisting Foraminifera [Fairbanks et al., 1982; Rohling et al., 2004; Steph et al., 2009] (Figures 8 and 9). Thus, the trend of increasing isotopic enrichment with decreasing coccolith size at Site 999 appears independent of depth habitat.
4.2. Cell Size Control Over Variable Stable Isotopic Fractionation Since the Pliocene
4.2.1. Processes Linking Cell Size and Variable Isotopic Fractionation
 It has long been recognized that many biogenic carbonates have δ18O and δ13C signatures that differ from an abiogenic calcite precipitated from the same seawater [e.g., McConnaughey, 1989a, 1989b]. In coccolithophores, these non-equilibrium offsets, or vital effects, correlate strongly with cell size in cultures and recent sediments [Ziveri et al., 2003; Stoll et al., 2007a]. In cultures, the magnitude of interspecific vital effects in coccoliths diminishes with increasing CO2 [Rickaby et al., 2010]. Our interpretation of results from the sediment record is based on these empirical correlations between vital effects, cell size and CO2 (Figure 1).
 The processes responsible for vital effects have been explored in several organisms and probably vary among biogenic carbonate producers. Foraminifera exhibit a strong ‘carbonate ion effect’ on both carbon and oxygen isotopic composition, resulting in decreased δ18O and δ13C with increasing media [CO32−] and pH [Spero et al., 1997; Bijma et al., 1999]. Zeebe  proposed that as the proportion of isotopically lighter CO32− relative to isotopically heavier HCO3− increases at higher pH, the δ18O of total dissolved inorganic carbon (ΣCO2) decreases. Thus the δ18O of precipitated CaCO3 will be lighter at higher pH providing calcite is formed from a mixture of the carbonate species in proportion to their relative contribution to ΣCO2 [Zeebe, 1999]. This same mechanism was recently modeled for the coccolithophore Calcidiscus leptoporus, whereby a higher ratio of CO32− to HCO3− delivered to the calcification vesicle results in the formation of δ18O-depleted coccoliths at higher media pH [Ziveri et al., 2012]. For both Foraminifera and coccolithophores, this model is applicable only to δ18O (not δ13C), because oxygen isotopes reach chemical equilibrium in the ΣCO2 system significantly faster than isotopic equilibrium (∼16sec versus ∼10 h at 25°C, pH 8.2, salinity 35 [Zeebe et al., 1999]). As a single C. leptoporus coccolith is formed in ∼1.4 h (Ziveri et al.  calculated from Langer et al. ), CO32− converted from HCO3− within the calcification vesicle will transfer the heavier δ18O fingerprint of HCO3− to the coccolith [Ziveri et al., 2012].
 Other processes probably also operate in coccolithophores, since culture studies with constant pH but changing ΣCO2 resulted in a variable degree of carbon and oxygen isotopic fractionation into coccolith calcite [Rickaby et al., 2010]. One hypothesis is that, under high ΣCO2 culture conditions with constant pH (therefore high [CO32−] and [HCO3−] but constant HCO3−:CO32− ratio), the large species Coccolithus pelagicus ssp. braarudii utilizes more HCO3− relative to CO32− for calcification compared to under low ΣCO2, potentially via manipulation of coccolith-vesicle pH [Rickaby et al., 2010]. This process is inferred to explain the observed heavier δ18O and δ13C of C. pelagicus ssp. braarudii coccoliths at high ΣCO2, one consequence of which is a reduced difference between the δ13C and δ18O of coccoliths precipitated by small (G. oceanica) versus large (C. pelagicus ssp. braarudii) coccolithophores at high ΣCO2 (Figure 1b) [Rickaby et al., 2010].
 The suite of cellular transport and fractionation processes responsible for the correlation between isotopic fractionation in coccolith carbonate and cell size has not yet been quantitatively explored. Cell size is a fundamental control on unicellular algal physiology because it controls rates of diffusive transport of essential dissolved compounds (nutrients, CO2, waste products) into and out of the cell. Thus, small cells with a lower carbon cell quota (approximated to cell volume) relative to surface area and a smaller diffusive boundary layer will be able to replenish resources more rapidly, allowing for higher maximum growth rates [Raven, 1998] and strongly affecting carbon isotopic fractionation into organic matter [Popp et al., 1998].
4.2.2. Is Variable Isotopic Fractionation in PPT Coccoliths Correlated to Cell Size?
 Our protocols separate different sized coccoliths that we infer were produced by coccolithophores of different sizes. A detailed study of coccolith, coccosphere, and cell size among several placolith-bearing spherical genera in the fossil record (Reticulofenestra, Cyclicargolithus and Coccolithus) indicates strong linear relationships between these parameters, consistent with relationships in modern descendent species [Henderiks, 2008]. Thus, we assume that fossil coccolith size is proportional to cell size in related or morphologically similar species (e.g., Umbilicosphaera sibogae var. foliosa and Calcidiscus leptoporus [Young et al., 2004]). In contrast, F. profunda, a poorly understood species with a unique morphology, could be an exception to the trend of increasing isotopic enrichment with decreasing cell size. Morphometric analyses indicate that F. profunda coccospheres are large relative to coccoliths (7–8 μm sphere, 2 μm liths [Quinn et al., 2005]) compared to other species (Reticulofenestra: ∼7 μm sphere, 4 μm liths [Henderiks, 2008]; Helicosphaera carteri: ∼14 μm cell diameter [Ziveri et al., 2003], ∼18 μm sphere and 8–9 μm liths (estimated from Young et al. )). We speculate that the heavy isotopic signature recorded in F. profunda coccoliths could relate to the low [CO32−] and high [CO2(aq)] in this species' deep habitat relative to the shallower habitat of the placoliths. Reduced CO2 limitation in the DPZ could enable F. profunda to meet a greater portion of its carbon quota by CO2 diffusion. If the magnitude of vital effect is influenced by the relative importance of CCMs to carbon acquisition, this could explain the similar isotopic signature in F. profundaand the smallest-celled UPZ coccoliths. Within the PPT placoliths, cell size, rather than phylogeny, appears to regulate isotopic fractionation. The 4–6μm coccolith size-fraction is a mixture of species from diverse lineages yet isotopically it falls between smaller and larger coccoliths, as expected if cell size were the dominant control (Figure 6).
4.2.3. Coherency of Size-Related Isotopic Fractionation in Fossil Coccoliths With Cultures and Sediment Traps
δ18O and δ13C differences in Plio-Pleistocene fractions dominated byF. profunda, small reticulofenestrid and Helicosphaera coccoliths (Figures 5 and 6) are of similar magnitude to those measured in sediment trap material from the Bay of Bengal in size fractions dominated by the same three coccolith groups and interpreted as resulting primarily from size-controlled variable isotopic fractionation [Stoll et al., 2007a]. Similarly, our PPT and LGM data show trends consistent with culture studies, although interspecific isotopic differences for a given coccolith size range are greater in culture [Dudley and Goodney, 1979; Dudley et al., 1980; Ziveri et al., 2003] than in fossil and trap material [Stoll et al., 2007a; this study]. This could be partly attributable to the distinct conditions in a laboratory culture set-up versus an open ocean environment. In nutrient and light replete culture growth media, experiments might show the maximum possible range of vital effects because light or nutrients do not limit the potential extra energy requirements necessary to operate CCMs. In nature, below-optimal nutrient (e.g., phosphate, nitrate or iron) or light levels are common and may limit the extent of CCM operation [Gervais and Riebesell, 2001; Rost et al., 2003; Giordano et al., 2005; Cassar et al., 2006; Raven et al., 2008; Schulz et al., 2007].
4.3. To What Extent Does Productivity Influence Vital Effects?
 While there has been no systematic evaluation of the effect of productivity on stable isotope vital effects in coccolith calcite, high algal growth rate has been shown to reduce carbon isotopic fractionation into organic matter during photosynthesis, presumably by increasing cellular carbon demand relative to diffusive CO2 supply [Bidigare et al., 1997]. Estimated phosphorus concentrations are routinely employed to correct for such growth rate effects when inferring fractionation due to changing CO2 levels [e.g., Bidigare et al., 1997; Pagani et al., 1999; Seki et al., 2010]. The significance of active carbon uptake in coccolithophores [e.g., Nimer et al., 1992; Dong et al., 1993; Rost et al. 2003; Cassar et al., 2006; Trimborn et al., 2007] to the cell budget is probably conditioned by the balance of diffusive CO2 availability and cellular carbon demand. At high growth rates, large cells have a greater relative limitation by diffusive CO2 supply than smaller cells and consequently models predict a greater difference between organic carbon isotopic composition of small and large cells [Rau et al., 1996]. In Bay of Bengal sediment traps, coccoliths produced during the high productivity upwelling or eddy-pumping season exhibit a greater difference between theδ18O and δ13C of larger and smaller size fractions (H. carteri and C. leptoporus versus G. oceanica) [Stoll et al., 2007a]. In cultures with constant carbon chemistry but variable light levels, higher growth rates led to a ∼1‰ greater difference between δ18O in large C. leptoporus and small G. oceanica coccoliths (Figure 1c) [Ziveri et al., 2003].
 Within a given species, coccolith Sr/Ca may be a useful indicator of the growth rate effects on carbon isotopic fractionation into organic carbon [Stoll and Schrag, 2000; Stoll et al., 2002]. Therefore we examined whether any significant changes in Sr/Ca occurred during our study intervals that could possibly attenuate or magnify coccolith vital effects. The PPT at Site 999 is likely to have more stable productivity effects than the PETM at Bass River because of the nature of the climate transition and geographical setting (open-ocean oligotrophic versus dynamic shelf environment). Although there is no significant correlation, some peaks in PPT productivity appear to be synchronous with maxima in interspecific coccolith isotope differences (Figure 7), suggesting some productivity influence on isotopic fractionation into coccoliths.
 During the PETM at Bass River, no large changes in the difference between stable isotopic composition of small and medium coccoliths occur. However, productivity inferred from coccolith Sr/Ca exhibits clear variability apparently related to sea level fluctuations. Prior to the PETM, intervals of low sea level inferred from dinoflagellate assemblages [Sluijs et al., 2007] are characterized by elevated coccolithophore productivity (Figure 4). We interpret this to reflect increased proximity to terrestrial nutrient sources and/or increased turbulent mixing in a shallower water column, injecting nutrient-rich deeper waters into the UPZ. The first part of the PETM (356–358 m) deviates from this trend in showing high productivity despite high sea levels, suggesting significant enhancement of terrestrial nutrient supply overriding the sea level influence. This interpretation is consistent with evidence for increased river discharge to the New Jersey margin during this time including increased abundance of low salinity dinocysts [Sluijs et al., 2007] and magnetofossil data suggesting a tropical river system existed on the mid-Atlantic coastal plain of the USA during the PETM [Kopp et al., 2009]. Changes in productivity at BR appear not to affect the magnitude of inter-fraction coccolithδ18O and δ13C observed, which remain low throughout the record (Figure 4).
4.4. Variable Expression of Coccolith Vital Effects Over the Cenozoic pCO2 Decline
 BR data show only small δ13C and δ18O differences (mean 0.17‰ and 0.66‰ respectively) between small and medium-sized coccoliths (Figures 4 and 9) compared to modern interspecific differences (up to 5‰). These new data from very well-preserved coccoliths agree with previous PETM data from ODP Site 690 [Stoll, 2005]; confirming that diagenetic homogenisation is not the principal driver of the small differences observed between different-sized Palaeocene coccoliths. The slightδ18O enrichment observed in the BR smaller coccolith fraction relative to the larger one is inconsistent with a different depth habitat relative to the larger coccoliths (section 4.1), and may represent a small cell-size related vital effect, given that the mechanisms causing fractionation inδ18O and δ13C in coccoliths may be decoupled in some circumstances. The main constituents of the Bass River PETM size-fractions, coccoliths of the generaToweius, Coccolithus and Prinsius (Figure S1), are all placoliths likely to conform to lith:cell size relationships detailed in section 4.2.2 [Henderiks, 2008]. Therefore the finding that, during the PETM, larger coccolithophores did not consistently produce more isotopically depleted coccoliths as they do during the PPT, LGM, modern ocean and laboratory appears robust, and may be an expression of more uniform carbon acquisition strategies among Palaeocene coccolithophores.
 CCMs in the Haptophyta may have existed since ∼300 Ma [Young et al., 2012]. However, their degree of expression and significance to whole-cell carbon acquisition probably changed as a function of CO2(aq), nutrient availability and other factors over the course of the Earth's history [Tortell, 2000], as occurs over a range of CO2 in laboratory cultures [e.g., Giordano et al., 2005; Falkowski and Raven, 2007; Moolna and Rickaby, 2012; Reinfelder, 2011]. Our data suggest that the array of stable isotope vital effects in fossil coccoliths, potentially one expression of the carbon acquisition strategies employed by ancient cells, is related to past climate and CO2 concentrations during the Cenozoic. This is consistent with culture studies showing reduced CCM activity and reduced δ18O and δ13C differences between small and large coccolithophores with higher ambient carbon availability [Rickaby et al., 2010; Moolna and Rickaby, 2012].
5. Summary and Conclusions
 The data presented here provide two Cenozoic end-members between which the ranges of oxygen and carbon stable isotopic compositions of different-sized coccoliths reached their modern array. New PETM data confirm previously reported reduced interpecific differences in coccolith stable isotopes within ‘greenhouse’ climate boundary conditions at this time, and PPT and LGM coccoliths record persistent 1.5–2‰ vital effects. These temporal trends and the correlation between isotopic fractionation and size in PPT and LGM coccoliths, consistent with culture data, suggest that vital effects may be linked in some way to coccolithophore carbon acquisition strategies and the level of expression of CCMs in response to the changing paleoenvironment (e.g., CO2(aq) concentration, nutrient availability). Coccolith vital effects were insensitive to pCO2 changes over the range inferred for the PPT (∼400 to 280 ppm, corresponding to a change in CO2(aq) of ∼10.5 to 6.5 μmol/kg SW at Site 999). During the PPT, productivity fluctuations may have affected isotopic fractionation into coccoliths to a small degree, in the same sense as previously reported for carbon isotopic fractionation into organic matter. During the PETM at BR, stable isotopic differences between size-fractions remain small despite significant changes in productivity that appear correlated to sea level fluctuations on the New Jersey coastal plain. Experimental data on the effect of nutrient and light availability and growth rate on isotopic fractionation into coccoliths in multiple species is currently lacking and would allow for more accurate assessment of potential factors driving coccolith stable isotopes in the fossil record.
 This work used samples provided by the (Integrated) Ocean Drilling Program (IODP). The IODP is sponsored by the U.S. National Science Foundation and participating countries under management of the Joint Oceanographic Institutions (JOI), Inc. We thank Williams College undergraduate research assistants Alicia Jackson and Danielle Zentner for help processing Bass River samples, and Nobu Shimizu of the Woods Hole Northeast National Ion Probe Facility for assistance with Sr/Ca ion probe determinations. We are grateful to John Morrison for expert help during Site 999 stable isotope analyses, Rindy Ostermann for assistance with stable isotope analyses at WHOI and also to staff at Santa Cruz stable isotope laboratory. We thank Ian Bailey and Kirsten Isensee for discussions and comments on an earlier version of the manuscript, Jeroen Groeneveld for access to unpublished data, and reviewers for comments that helped improve the manuscript. All authors acknowledge European Research Council grant UE-09-ERC-2009-STG-240222-PACE, awarded to HMS. HMS also acknowledges funding from National Science Foundation grant EAR-0628336.