We constructed a high-resolution Mg/Ca record on the planktonic foraminifer Globigerinoides sacculifer in order to explore the change in sea surface temperature (SST) due to the shoaling of the Isthmus of Panama as well as the impact of secondary factors like diagenesis and large salinity fluctuations. The study covers the latest Miocene and the early Pliocene (5.6–3.9 Ma) and was combined with δ18O to isolate changes in sea surface salinity (SSS). Before 4.5 Ma, SSTMg/Ca and SSS show moderate fluctuations, indicating a free exchange of surface ocean water masses between the Pacific and the Atlantic. The increase in δ18O after 4.5 Ma represents increasing salinities in the Caribbean due to the progressive closure of the Panamanian Gateway. The increase in Mg/Ca toward values of maximum 7 mmol/mol suggests that secondary influences have played a significant role. Evidence of crystalline overgrowths on the foraminiferal tests in correlation with aragonite, Sr/Ca, and productivity cyclicities indicates a diagenetic overprint on the foraminiferal tests. Laser ablation inductively coupled plasma–mass spectrometry analyses, however, do not show significantly increased Mg/Ca ratios in the crystalline overgrowths, and neither do calculations based on pore water data conclusively result in significantly elevated Mg/Ca ratios in the crystalline overgrowths. Alternatively, the elevated Mg/Ca ratios might have been caused by salinity as the δ18O record of Site 1000 has been interpreted to represent large fluctuations in SSS, and cultivating experiments have shown an increase in Mg/Ca with increasing salinity. We conclude that the Mg/Ca record <4.5 Ma can only reliably be considered for paleoceanographical purposes when the minimum values, not showing any evidence of secondary influences, are used, resulting in a warming of central Caribbean surface water masses after 4.5 Ma of ∼2°C.
The late Miocene/early Pliocene gradual closure of the Panamanian Gateway effectively changed oceanographic patterns between the Pacific and Atlantic oceans, and thus led to profound changes in global atmospheric circulation and climate. The closure of this ocean gateway was responsible for changing the Pacific, Caribbean, and Atlantic water mass signatures, generating the Western Atlantic Warm Pool (WAWP), and intensifying thermohaline circulation in the North Atlantic via increasing salt and heat transport to high northern latitudes [Maier-Reimer et al., 1990; Tiedemann and Franz, 1997; Haug and Tiedemann, 1998; Billups et al., 1999; Haug et al., 2001]. As a consequence, the intensification of the Gulf Stream system might have had an important role in the initiation of the northern hemisphere glaciation (NHG), which started during the mid Pliocene [Haug and Tiedemann, 1998; Driscoll and Haug, 1998].
Haug et al.  interpreted the divergent pattern of planktonic δ18O records from the Caribbean (Site 999) and the tropical Pacific (Site 851) at 4.6 Ma and the subsequent 0.5‰ increase in δ18O in the Caribbean since ∼4.2 Ma as the establishment of the salinity contrast between Pacific and Caribbean surface waters [Keigwin, 1982]. This hydrographic change was caused by the plate tectonic shoaling of the Panamanian sill to a water depth less than 100 m, which effectively restricted the water exchange between the Pacific and the Atlantic, and therefore increased SSS in the Caribbean by 1-2 salinity units by means of an increased atmospheric transport of water vapor from the Caribbean into the Pacific. Recent results from Caribbean Site 1000 [Steph et al., 2006] show that, in accordance with this general increase in Caribbean SSS, salinity gradients within the Caribbean increased and varied on precessional timescales after 4.5 Ma. Low SSS gradients occurred during minima in northern hemisphere summer insolation and were interpreted to reflect enhanced inflow of low-salinity Pacific surface water into the Caribbean. During maxima in northern hemisphere summer insolation, the inflow of low-salinity Pacific surface waters was reduced. In the northern Caribbean (Site 1000), higher-saline Atlantic water predominated, leading to a pronounced SSS gradient between the southern (Site 999) and northern Caribbean (Site 1000).
The final closure of the Panamanian Gateway was not established until ∼2.7–2.2 Ma, when the major exchange of mammals via the isthmus between North and South America took place [Stehli and Webb, 1985; Webb, 1997], and both nannofossil assemblages and SSTMg/Ca from the Caribbean and the eastern Pacific permanently diverged [Kameo and Sato, 2000; Groeneveld et al., 2006].
1.2. Mg/Ca Paleothermometry
The advantage of measuring Mg/Ca and δ18O on the same biotic carrier is that synchronous estimates of SST and global ice volume can be determined and used to calculate δ18Oseawater, a proxy for SSS, thereby avoiding the bias of seasonality and/or habitat differences that occur when proxy data from different faunal groups are used.
We present a high-resolution Mg/Ca record for ODP Site 1000 from the Caribbean for the time interval from 5.6 Ma to 3.9 Ma for the planktonic foraminifer G. sacculifer. The main objectives of the study were (1) to investigate the development of Caribbean SST with the restriction of the exchange of upper ocean water masses between the Pacific and the Caribbean, due to the progressive closure of the Panamanian Gateway, and (2) to explore the potential influence of secondary effects, i.e., diagenesis and salinity, on the Mg/Ca.
2. Material and Methods
ODP Site 1000 was cored during Leg 165 on the Nicaraguan Rise (Pedro Channel) in the Caribbean Sea (16°33.222′N, 79°52.044′W, Figure 1). The shallow site location of 916 m in comparison with the depth of the Caribbean lysocline at 4000 m [Archer, 1996] implies that carbonate dissolution is negligible. Samples were taken every 10 cm, reflecting a temporal resolution of ∼3 ka per sample. The age model for Site 1000 is based on the δ18O record of the benthic foraminifer Cibicidoides wuellerstorfi [Steph et al., 2006]. Since no composite depth existed for Site 1000, a preliminary age model was constructed by identifying major isotopic stages according to the nomenclature of Shackleton et al. . The preliminary age model was then correlated to the orbitally tuned age model of ODP Site 925/926 from Ceara Rise [Bickert et al., 1997; Tiedemann and Franz, 1997; Shackleton and Hall, 1997]. The final age model revealed the presence of dominant precession cycles in the planktonic δ18O record, which were then used to orbitally tune the age model [Steph et al., 2006].
2.1. Mg/Ca Analysis
Twenty to twenty-five specimens of Globigerinoides sacculifer were selected from the 315–400 μm size fraction. Specimens visibly contaminated by ferromanganese oxides and specimens with a final sac-like chamber were not selected for analysis. After gentle crushing, the foraminiferal tests were cleaned according to the cleaning protocol of Barker et al. . To remove clays, the samples were rinsed 4–6 times with distilled deionized water and twice with methanol (suprapure) with ultrasonical cleaning steps (2–3 minutes) after each rinse. Subsequently, samples were treated with a hot (97°C) oxidizing 1% NaOH/H2O2 solution (10 mL 0.1 M NaOH (analytical grade); 100 μL 30% H2O2 (suprapur)) for 10 minutes to remove organic matter. Every 2.5 minutes, the vials were rapped on the bench top to release any gaseous buildup. After 5 minutes, the samples were placed in an ultrasonic bath for a few seconds in order to maintain contact between reagent and sample. This treatment was repeated after refreshment of the oxidizing solution. Any remaining oxidizing solution was removed by three rinsing steps with distilled deionized water. After transferring the samples into clean vials, a weak acid leach with 250 μL 0.001 M HNO3 (subboiled distilled) was applied with 30 seconds ultrasonic treatment and subsequent two rinses with distilled deionized water. After cleaning, the samples were dissolved in 0.075 M nitric acid (HNO3) (subboiled distilled) and diluted with distilled deionized water up to 3 mL containing 10 ppm of yttrium as an internal standard.
Analyses were run on an ICP-OES (ISA Jobin Yvon, Spex Instruments S.A. GmbH) with polychromator applying yttrium as an internal standard. We selected element lines for analyses which were most intensive and undisturbed (Ca: 317.93 nm; Mg: 279.55 nm; Sr: 407.77 nm; Fe: 238.21 nm; Mn: 257.61 nm; and Y: 371.03 nm). Element detection was performed with photomultipliers, the high tension of which was adapted to each element concentration range. The relative standard deviation for the analysis of magnesium was 0.45% and for calcium 0.15%. The Mg/Ca reproducibility of replicate samples (n = 22) of G. sacculifer is 3.3% (σ = 0.104), giving a temperature error of ±0.7°C. The conversion of foraminiferal Mg/Ca ratios into SSTMg/Ca was carried out by applying the equation of Nürnberg et al.  (SST = ((Log (Mg/Ca)-Log 0.491)/0.033). The total error on paleotemperatures reconstructed by means of Mg/Ca is 1–1.5°C [Nürnberg et al., 2000; Lea et al., 1999].
We used an artificial solution, comparable to our calibration standards, to monitor the quality of our Mg/Ca measurements with a Mg/Ca ratio of 2.90 mmol/mol as in-house standard. The long-term standard deviation of this standard is 0.026 mmol/mol or 0.91%. Interlaboratory comparison was done by measuring a set of standards provided by M. Greaves (University of Cambridge) [Greaves et al., 2005]. On average, our ICP-OES measurements were 0.5% (∼0.02 mmol/mol) lower than those from the Cambridge laboratory, which is within the error range given by Rosenthal et al. .
Several samples, which partly exhibited crystalline overgrowths and which were selected for SEM and laser ablation analysis, were also cleaned by including the reductive cleaning step described by Boyle and Keigwin . Mg/Ca ratios of these samples were systematically lower by 0.37 mmol/mol (range 0.26–0.48 mmol/mol) than those cleaned without the reduction step. This offset is comparable to the offset between the “Mg method” and the “Cd method” as shown by Rosenthal et al. .
Iron and manganese concentrations were monitored to identify contaminant clay particles and manganese-rich carbonate coatings, which might affect the foraminiferal Mg/Ca ratios [Boyle, 1983; Barker et al., 2003]. Samples older than 4.5 Ma show average Fe/Ca ratios of 0.23 mmol/mol (σ = 0.16) and average Mn/Ca ratios of 0.22 mmol/mol (σ = 0.05) (Figure 2). Although these values are higher than the 0.1 mmol/mol Mn/Ca and Fe/Ca ratios given by Barker et al.  for clean, uncontaminated foraminiferal tests, we assume our Mg/Ca analyses unaffected by contamination as no correlation with Mg/Ca exists for both Fe/Ca (r2 = 0.01; Probability (P) = 0.05) and Mn/Ca (r2 = 0.02; P = <0.01) (Figure 2). For samples younger than 4.5 Ma, average Fe/Ca ratios are 0.33 mmol/mol (σ = 0.20) and average Mn/Ca ratios are 0.13 (σ = 0.03), and lack any correlation with Mg/Ca (r2 = 0.02; P = 0.06 for Mg/Ca versus Fe/Ca, r2 = 0.02; P = 0.06 for Mg/Ca versus Mn/Ca) (Figure 2). Samples from which Fe/Ca and Mn/Ca ratios were more than 2σ higher than the average were reanalyzed.
Laser ablation inductively coupled plasma–mass spectrometry (LA-ICP-MS) (ICP-MS, Micromass Platform) was used to measure high-resolution Mg, Sr and Ca chamber wall profiles of selected foraminiferal tests. For the laser ablation measurements we selected cleaned G. sacculifer fragments from four different samples, two samples with crystalline overgrowths (sample 14-5-115 and sample 14-5-125) and two samples without crystalline overgrowths (sample 14-5-95 and sample 14-5-145) (Table 1). The fragments were cleaned according to the protocol described in paragraph 2.1. Multiple laser ablation profiles were ablated in different foraminiferal fragments from the same sample to elucidate in-test heterogeneities. In total, 25 profiles were ablated, 11 from foraminiferal test fragments without crystalline overgrowths, and 14 from fragments with crystalline overgrowths (Table 1).
Table 1. Comparison Between Bulk ICP-OES Measurements and LA-ICP-MS Analyses for Mg/Ca and Sr/Caa
Average results are shown from multiple ablation profiles through cleaned G. sacculifer tests from two contaminated, with crystalline overgrowths, samples (14-5-115 and 14-5-125) and two uncontaminated, without crystalline overgrowths, samples (14-5-95 and 14-5-145). The calculated Mg/Ca ratios derived from the LA-ICP-MS measurements agree, within the method standard deviation of 5–10%, to the whole-test ICP-OES measurements performed on the same samples.
Relative standard deviation is 5–10%
no crystalline overgrowth
no crystalline overgrowth
Using an Excimer 193 nm deep ultra violet laser (Lambda Physik) with Geolas optics [Günther et al., 1997] craters of 80 μm were ablated during the analyses. Energy density at the sample surface was kept at 2 mJ/cm2, shot repetition rate at 8 Hz. Laser ablation of calcite requires ablation at ultra violet wavelengths since higher wavelengths result in uncontrolled cleavage. The ablated material was transported on a continuous He flow and mixed with an Ar make up gas before injection into the Ar plasma of the quadrupole ICP-MS instrument (Micromass Platform). A collision and reaction cell was used to improve results by reducing spectral interferences on the minor isotopes of Ca (42Ca, 43Ca and 44Ca). Interelemental fractionation was insignificant due to the relative low depth to width ratio of the ablation craters produced [Mason and Mank, 2001]. The use of a collision cell also prevented interference of clusters like 12C16O16O. Although the Excimer UV-193 usually does not show any matrix effects, possible matrix effects were nevertheless checked with an in-house matrix matched standard (Utrecht Calcite). Calibration was performed against U.S. National Institute of Standards and Technology SRM 612 glass using the concentration data of Pearce et al. , with Ca as an internal standard. Calcium carbonate is well suited for LA-ICP-MS because Ca can be used as an internal standard at 40 wt%. For quantifying Ca we used mass 44, while monitoring masses 42 and 43 as an internal check. Using Ca as an internal standard allows direct comparison to trace metal to Ca ratios from traditional wet-chemical studies. Magnesium concentrations were calculated using masses 24 and 26. Although the accuracy for the more abundant mass 24 is higher, both isotopes agree within a few percent. For calculating strontium concentrations mass 88 was used. Analyses were performed at the Faculty of Geosciences, Utrecht University, Netherlands.
With initiation of the ablation process preferential ablation of material from surfaces that lie orthogonal to the incident laser beam results in greater ablation from the tops of interpore ridges and spine bases [Eggins et al., 2003]. This means that the ablation crater is not flat but follows the curvature of the foraminiferal test and therefore should reveal elevated element concentrations in a diagenetic coating when present. Further in the ablation process the craters become increasingly flatter bottomed. This results in a smoothing of the signal when the ablation reaches the pores on the other side of the test. The fragments of each sample were measured both from the inner side toward the outer side and the other way around to certify that no smoothing of a possible coating peak occurred.
2.3. Analysis of δ18O
For stable isotope analysis 10 specimens of G. sacculifer (without sac-like final chamber) were picked from the 315–400 μm size fraction. All isotope analyses were run at IFM-GEOMAR (Kiel) on a Finnigan MAT 252 Delta-Plus Advantage Mass Spectrometer equipped with an automated Kiel Carbonate Preparation line. The mass spectrometer is calibrated to NBS 19, and the isotope values are reported in ‰ relative to the VPDB (Vienna PDB) scale. The external reproducibility of in-house carbonate standards was ±0.08‰ (±1σ) for δ18O [Steph et al., 2006].
2.4. Carbonate Mineralogy
Carbonate mineralogy was determined using the Philips PW1710 diffractometer at IFM-Geomar with a cobalt Kσ tube at 40 KV and 35 mA. For the interval 4.1 to 3.8 Ma, one sample per 10 cm was analyzed. Samples older than 4.1 Ma were analyzed with a spacing of 30 cm. The samples were scanned from 25°–40° at a scanning speed of 0.01 steps per second to cover the significant peaks of the various carbonate minerals. The percentage of aragonite and calcite relative to total carbonate content was calculated from peak area ratios using in-house calibration curves. Peak areas were measured using the computer-based integration program MacDiff (R. Petschick, Johann Wolfgang von Goethe-Universität, Frankfurt am Main, Germany, 1999). The standard deviation (1σ) of this method, calculated from replicates, is 1%. The nonlinear relationship between calcite and aragonite was calculated from ratios calibrated from standards measured on the diffractometer.
2.5. Calcareous Nannofossil Analysis
The relative abundance record of the coccolithophorid Florisphaera profunda was used as a proxy to reconstruct variations in primary productivity by means of high-resolution scanning electron microscope (SEM) analysis [Crudeli, 2005]. Sample resolution was the same as for Mg/Ca and δ18O. According to Crudeli and Kinkel , 0.1 g of dry sediment was diluted with carbonate-saturated tap water. The solution was then divided with a rotary splitter device. A final solution with 0.001 g of sediment was filtered with a low-pressure vacuum pump onto a polycarbonate-membrane filter (pore size: 0.40 μm; diameter: 50 mm). A segment of the filter was then mounted on a scanning electron microscope (SEM) stub. SEM analyses were conducted at a magnification of 6000X using a CamScan 44-Serie-2-CS-44 SEM (University of Kiel).
3. Results and Discussion
The range of Mg/Ca ratios in G. sacculifer from Site 1000 can be divided into two time intervals, before and after 4.5 Ma (Figure 3). Before 4.5 Ma, Mg/Ca ratios range between 3.5 and 4.5 mmol/mol, translating into SSTMg/Ca between 26.5°C and 29°C. After 4.5 Ma, the range of Mg/Ca ratios increases to values between 4 and 7 mmol/mol, providing SSTMg/Ca between 28°C and 35°C (Figure 3).
3.1. Free Exchange of Upper Ocean Water Masses Before 4.5 Ma
The latest Miocene to earliest Pliocene time interval (5.6–4.5 Ma) is characterized by the absence of an explicit correlation between Mg/Ca and δ18O (r2 = <0.01; P = 0.39). The δ18O record shows moderate variation, ranging between −1.05‰ and −1.7‰. A pronounced salinity effect on planktonic δ18O can be excluded for this time interval, as the free exchange between Pacific and Caribbean surface water masses prohibited the establishment of salinity gradients [Haug et al., 2001]. Hence the fluctuations in δ18O are interpreted to reflect changes in global ice volume and in SST. Orbital-scale changes in global ice volume affected the late Miocene/early Pliocene δ18O signal by up to 0.15–0.25‰ [Hodell and Warnke, 1991; Kennett and Hodell, 1993; Tiedemann et al., 1994; Shackleton et al., 1995; Mix et al., 1995], while variations in SST were 2–3°C as indicated by the Mg/Ca record.
3.2. Progressive Restriction of Upper Ocean Water Masses After 4.5 Ma
After 4.5 Ma, we observe an overall positive correlation between the Mg/Ca and δ18O records of G. sacculifer (r2 = 0.48; P = <0.0001) (Figure 3). Such a distinct positive correlation was hitherto not described. The Mg/Ca ratios suggest SSTMg/Ca of maximum 35°C and amplitudes of up to 5°C (Figure 3), with a general warming trend of 4°C between 4.5 Ma and 3.9 Ma. δ18O values range between −1.6‰ and 0‰, show increasing amplitudes up to 1.4‰, and show a general trend toward higher values [Steph et al., 2006].
The extreme SSTMg/Ca maxima of up to 35°C and amplitudes of up to 5°C appear to be very unlikely for the early Pliocene. Firstly, modern SSTs at ∼30–50 m water depth, the assumed habitat depth of G. sacculifer [Fairbanks et al., 1982], do not exceed 30–31°C in open ocean conditions [Levitus and Boyer, 1994], and it is unreasonable to expect significantly higher SSTs during the early Pliocene. Maximum SSTs during the course of El Niño, for example, never reach much beyond 29°C [Philander, 1990]. Several mechanisms have been described as contributing to this stability of tropical SSTs like cloud albedo effects, evaporation, characteristics of dry-air pools, and heat transport out of the tropics [Newell, 1979; Ramanathan and Collins, 1991; Waliser and Graham, 1993; Pierrehumbert, 1995]. Secondly, this early Pliocene time interval is known to represent a globally warm and stable climate of unipolar glaciation with a pole-equator temperature gradient which was smaller than today [Cronin et al., 1993; Crowley, 1996]. In addition, studies in the high-latitudinal Atlantic indicated that SSTs might have been warmer by up to 8°C than today, while the tropical SSTs remained the same or were, due to increased heat transport to higher latitudes, even slightly cooler than today [Cronin et al., 1993; Crowley, 1996; Dowsett et al., 1996; Billups et al., 1998; Groeneveld et al., 2006]. Hence amplitudes in SSTMg/Ca as large as 5°C appear to be rather unlikely.
How do we then explain the extremely high SSTMg/Ca, the large SSTMg/Ca amplitude variations, and the positive correlation with δ18O during the time interval 4.5–3.9 Ma? In the following sections we discussed two potential tracts, diagenetic overprint (section 4) and salinity influence (section 5), of explaining the anomalous Mg/Ca signal.
4. Diagenetic Alteration of Foraminiferal Tests After 4.5 Ma
Sediments with high calcium carbonate contents are prone to diagenetic overprinting, dissolution and reprecipitation, especially when the carbonate is present as metastable aragonite or high-Mg calcite [Swart and Guzikowski, 1988; Malone et al., 1990; Munnecke et al., 1997; Melim et al., 2002]. Site 1000 is surrounded by carbonate platforms which can supply large amounts of aragonite to the core location [Sigurdsson et al., 1997]. Since aragonite in sediments is susceptible to dissolution, reprecipitation as inorganic calcite can subsequently occur. After 4.5 Ma alternations between aragonite-rich (up to 60%) and nodular, aragonite-poor layers (<30%) occur, showing precessional cyclicity (Figure 4).
Relative abundance variations in the coccolithophorid Florisphaera profunda at Site 1000 reflect temporal changes in marine productivity (Figure 4), with low abundances during times of higher marine productivity at the surface of the mixed layer [Molfino and McIntyre, 1990; Beaufort et al., 1997; de Garidel-Thoron et al., 2001]. Higher marine productivity enhances nutrient utilization, reduces nutrient flux to the bottom of the photic zone, and hence reduces the relative abundances of F. profunda [Crudeli, 2005]. The abundance variations of F. profunda are positively correlated with the aragonite record (r2 = 0.50) (Figure 4), implying that at times of enhanced marine productivity, more organic matter is decomposed, thereby lowering the pore water pH, favoring aragonite dissolution, and subsequently inducing reprecipitation as inorganic calcite [Dix and Mullins, 1992; Frank and Bernet, 2000].
Additionally, SEM analyses show that after 4.5 Ma the inner surfaces of tests of G. sacculifer are regularly covered by crystalline overgrowths, which are resilient to the cleaning procedures routinely applied before Mg/Ca analyses (Figures 5a–5c). The size of the single crystals amounts to 2–3 μm. Thus the alternating cycles of relative abundance of F. profunda and aragonite in combination with the SEM analyses show that a diagenetic overprint is present in these samples. Whether the diagenetic overprint has a significant influence on foraminiferal δ18O and Mg/Ca is discussed in sections 4.1 and 4.2.
4.1. Assessing the Diagenetic Influence on δ18O
The positive correlation between δ18O, aragonite, and the relative abundance of F. profunda after 4.5 Ma (Figures 3 and 4) implies that δ18O is possibly affected by carbonate diagenesis [Killingley, 1983; Schrag et al., 1995; Pearson et al., 2001]. The precipitation of diagenetic calcite from cold pore water alters the original δ18O values toward higher values. We compared the δ18O record of G. sacculifer of Site 1000 with the one from southwestern Caribbean Site 999 [Haug et al., 2001; Steph et al., 2006] to further assess the diagenetic imprint on precessional periods (Figure 6a). Site 999 (12°44′N, 78°44′W) is located in 2828 m water depth and thus well below the aragonite compensation depth. The isotope curves from this site are used as reference records as diagenetic overprints on the δ18O signal are considered to be negligible. The isotope fluctuations of G. sacculifer at Site 999 reveal no clear response to precession-induced climate variability, the amplitudes between 4.2 and 3.8 Ma are distinctly lower (∼0.6‰ versus 1.3‰ at Site 1000), and the δ18O trend toward higher values is less pronounced (Figure 6a). This suggests that a diagenetic overprint may have altered the δ18O signal at Site 1000. Also, the cyclic variability of the δ18O signal at Site 1000 is negatively correlated with the aragonite content. Crystalline overgrowths on foraminiferal tests are found only at samples where δ18O values are high and aragonite contents are low.
In addition to the δ18O record of G. sacculifer, we measured δ18O records of the deep dwelling planktonic foraminifer Globorotalia crassaformis, and the epibenthic foraminifer Cibicidoides wuellerstorfi at Sites 999 and 1000 [Steph et al., 2006] for the interval from 4.1–3.8 Ma (Figure 6). The comparison of the δ18O records of G. crassaformis and C. wuellerstorfi does not support the presence of a significant diagenetic overprint (Figure 6). G. crassaformis is a planktonic foraminiferal species with a deep calcification depth (250–500 m [Niebler et al., 1999]). Accordingly, δ18O values should increase with increasing habitat depth as temperature decreases. This pattern might be modified by changes in local δ18Owater. In contrast to the mixed layer, regional differences in subsurface water mass signatures (temperature and salinity) between Sites 999 and 1000 should be low at water depths of 250 to 500 m. We do not expect significant influence from the low-salinity Pacific inflow at these water depths, as the Panama sill was shallower than 100 m since 4.2 Ma [Haug et al., 2001]. Accordingly, we would expect similar δ18O values, amplitudes, and temporal variations for the deep dwelling G. crassaformis at both sites if diagenetic overprints were negligible. Indeed, the comparison of δ18O records from G. crassaformis reflects similar δ18O amplitudes (about 0.7‰) and temporal variations within the time interval from 4.1–3.8 Ma (Figure 6b) which are about 0.3‰ higher at Site 999. At Site 1000, the δ18O amplitudes of G. crassaformis are only half the δ18O amplitudes of G. sacculifer and the δ18O variability of G. crassaformis at precessional periodicities is as weak as at Site 999. Assuming a strong diagenetic overprint, the δ18O values should be higher at Site 1000 and the precessional amplitudes should be much higher. This is not the case and argues against significant diagenetic overprint on the δ18O signal. The temporal variability of the benthic δ18O records from Sites 999 and 1000 are similar, following a 41-ka rhythm with comparable amplitudes (Figure 6c). Again, a strong diagenetic overprint should have strengthened the cyclic variability at precessional periodicities which is not observed.
The isotopic composition of the pore water is an additional factor which determines the composition of diagenetic calcite [Killingley, 1983; Schrag et al., 1995]. As pore water measurements of δ18O are not available for Site 1000, we used ODP Site 1006 from the Great Bahama Bank as analogue for Site 1000, as their respective settings are similar [Reuning et al., 2005]. Reuning et al.  used pore water measurements of δ18O and varying recrystallization rates to explore the possible range of δ18O values of diagenetic calcite in early Pliocene samples from Site 1006. They found that the impact of up to 5% recrystallization of the foraminiferal calcite did not have a significant effect (∼0.1‰) on the primary signal. This suggests that, even though the δ18O values of Site 1000 appear to have higher values than expected, this is not supported by the magnitude of the diagenetic overprint. However, if the diagenetic alteration of the tests is more extensive than just the visible crystalline overgrowth [Pearson et al., 2001; Hover et al., 2001], it cannot be excluded that the maximum δ18O values of Site 1000 younger than 4.5 Ma are affected by a diagenetic overprint.
We suggest that a significant diagenetic imprint on the δ18O signal of G. sacculifer, without accompanying effects on the δ18O variability of G. crassaformis, and C.wuellerstorfi, appears to be unlikely. Therefore we argue that, in agreement with Steph et al. , the pronounced precession-related amplitudes within the δ18O record of G. sacculifer mainly reflect changes in oceanography and hence represent large fluctuations in salinity.
4.2. Assessing the Diagenetic Influence on Mg/Ca
The positive correlation between Mg/Ca, aragonite, and the relative abundance of F. profunda (Figure 4) and the presence of crystalline overgrowths on foraminiferal tests (Figures 5a–5c), suggests that diagenesis might be responsible for the anomalously high Mg/Ca ratios of up to 7 mmol/mol. Previous studies from the Little Bahama Bank showed planktonic foraminiferal Mg/Ca ratios between 6 and 12 mmol/mol [Rosenthal et al., 2000; Lear et al., 2003; Reuning et al., 2005]. As these samples originated from carbonate platform sediments from shallow core locations, they were interpreted as being influenced by carbonate diagenesis.
Similarly, as foraminiferal Sr/Ca and Mg/Ca ratios are positively correlated after 4.5 Ma (r2 = 0.58), foraminiferal Sr/Ca ratios appear to increase when sedimentary aragonite is low. The range of Sr/Ca ratios of 1.35–1.58 mmol/mol after 4.5 Ma is higher (∼12%) than the range of Sr/Ca ratios for the interval before 4.5 Ma (1.25–1.40 mmol/mol), suggesting that dissolution of Sr-rich aragonite and subsequent carbonate reprecipitation from Sr-enriched pore waters onto the foraminiferal surfaces took place. This increase in foraminiferal Sr/Ca is too large to be attributed to a change in Sr/Caseawater. A reconstruction by Lear et al.  showed an increase in Sr/Caseawater of less than 5% between the late Miocene and the present. Hence we used LA-ICP-MS to quantify the element concentrations within these diagenetic overgrowths to assess whether these diagenetic overgrowths caused the high element/Ca ratios.
High-resolution elemental profiles across the chamber walls of G. sacculifer performed by LA-ICP-MS [Eggins et al., 2003; Reichart et al., 2003; Hathorne et al., 2003] revealed that Mg/Ca distribution patterns are similar both for specimens from samples with and without crystalline overgrowths (Figure 7). Since the data acquisition of the laser ablation measurements is continuous the first 2–3 data points during ablation can safely be assumed to represent primarily the crystalline overgrowths. The similar Mg/Ca ratios of these points in comparison with the remainder of the foraminiferal test indicate that the crystalline overgrowth is not significantly enriched in Mg and therefore appears not to be responsible for the high bulk Mg/Ca ratios. This, however, does not exclude the possibility of other diagenetic alterations of the bulk Mg/Ca ratios such as partial recrystallization or diffusion of Mg into the tests [Hover et al., 2001].
The similar lack of a significant trace metal enrichment in the observed coatings is also indicated by the profiles for Sr/Ca (Figure 7). Although we would have expected that the crystalline overgrowths are represented by elevated Sr/Ca levels due to the origin from aragonite and the clear correlation between Sr/Ca and aragonite, only slight intratest (up to ∼1 mmol/mol) elevations in Sr/Ca were detected (Figures 7a and 7b). This can possibly be explained such that the absolute increase in foraminiferal Sr/Ca after 4.5 Ma in comparison with the older samples was significant but only ∼12%. We argue that the expected increase in Sr/Ca in the crystalline overgrowths has not been captured by the laser ablation measurements as the relative standard deviation of the laser ablation measurements has the same order of magnitude (∼10%) as the diagenetically induced enrichment.
Pore water data (Sample 15-3-145, 4.202 Ma) are from Lyons et al.  and are used to calculate the composition of the inorganic calcite. “Average maximum of the contaminated Sr/Caforam” refers to the average Sr/Ca ratio of those samples which show crystalline overgrowths. “Average maximum of the uncontaminated Sr/Caforam” refers to the average Sr/Ca ratio of the samples from the older, uncontaminated part of the record, but only those from the maximum values. “Average maximum of the contaminated Mg/Caforam” refers to the average Mg/Ca ratio of those samples which show crystalline overgrowths.
By using different distribution coefficients for Sr and the pore water composition, the possible Sr/Ca ratios of the inorganic calcite are calculated. Using these compositions in combination with the measured Sr/Caforam, the mass percentage of the diagenetic overprint of the bulk analyzed foraminifer is calculated [Lohmann, 1995].
Original Mg/Ca ratios of the contaminated foraminiferal tests. Combining the mass percentage of the crystalline overgrowth with the calculated Mg/Ca ratios of the overgrowth, the Mg/Ca ratios of the original contaminated foraminiferal tests are calculated [Lohmann, 1995]. These calculations provide an indication that the crystalline overgrowth does not necessarily elevate the bulk Mg/Ca ratios.
There are several arguments to use Sr/Ca, instead of Mg/Ca, to calculate the possible composition of the crystalline overgrowths. Firstly, the pore water composition for the studied interval of Site 1000 is mainly determined by selective dissolution of aragonite which is relatively enriched in Sr (1000–10,000 ppm). Secondly, the distribution coefficient of Sr is much better constrained (0.015–0.06) than the distribution coefficient of Mg [Baker et al., 1982; Delaney, 1989; Morse and Bender, 1990] (Table 2c). Several experiments have shown that for the recrystallization of calcite, which is typically a very slow process, the effect of changing precipitation rate yields distribution coefficients for Sr between 0.03 and 0.06 [Katz et al., 1972; Baker et al., 1982; Pingitore and Eastman, 1986]. Thirdly, the range of Sr/Ca ratios in modern planktonic foraminifers is small (1.25–1.40 mmol/mol). Such a range is typically found in samples >4.5 Ma having no crystalline overgrowths (average = 1.36 mmol/mol). We therefore assume that the 12% increase in Sr/Ca in the samples <4.5 Ma, which show crystalline overgrowths (average = 1.55 mmol/mol), is due to aragonite dissolution and subsequent reprecipitation. The assumption that this increase is solely due to diagenesis can be used to estimate the proportion of diagenetic calcite with respect to the initial foraminiferal calcite [Lohmann, 1995; Reuning et al., 2005] (Table 2c). Since the process of recrystallization occurs over long time periods and accordingly over a large range of sediment depths, the conditions under which the recrystallization occurs, like changing temperature or changing pore water concentrations, can change significantly. Distribution coefficients which are determined after pore water chemistry therefore only represent the “last stage” in the process and will yield different coefficients than those which are determined after the composition of inorganic precipitates, either sediment-derived or experimentally derived.
We used the estimation of the amount of inorganically precipitated calcite, based on Sr/Ca, to calculate the possible composition of the primary Mg/Ca ratio of the foraminifer [Lohmann, 1995] (Table 2d). The average Mg/Ca ratio of the affected samples (6.33 mmol/mol) was used as bulk value. Although we cannot completely rule out the possibility of diagenetically induced changes within the test, the SEM analyses only show diagenetic crystals on the inner side of the tests. This makes the estimations using 16.1% diagenetic calcite seem too high (Table 2d). The distribution coefficient for naturally precipitated inorganic calcite tends to be lower than the one for experimentally precipitated calcite [Baker et al., 1982; Delaney, 1989] (Table 2b), suggesting that the estimations based on the experimentally derived distribution coefficient of Mucci and Morse  are possibly not realistic either. These two limitations leave a range of possible Mg/Ca ratios for the primary calcite (highlighted in Table 2d with bold text) which differs only slightly from the analyzed bulk values. Thus the exercise on estimating the composition of the primary Mg/Ca ratio in the foraminiferal tests is not conclusive. The theoretical composition of the inorganic precipitates is potentially able to alter the Mg/Ca ratio both to lower and higher values. As the contribution of the diagenetic influence is not conclusive in causing the anomalously Mg/Ca ratios, and the δ18O record is dominated by changes in salinity (section 4.1) [Steph et al., 2006] we explored the influence of large salinity variations on the Mg/Ca in the next section.
5. Correction of Mg/Ca for the Potential Influence of Salinity
Both the positive relationship between the Mg/Ca and δ18OG. sacculifer records (Figure 3), and the suggested increase in Caribbean SSS after 4.5 Ma [Haug et al., 2001; Steph et al., 2006] basically suggest that the planktonic δ18O record is dominated by variations in SSS. Cultivating experiments with G. sacculifer revealed a positive relationship between foraminiferal Mg/Ca and salinity [Nürnberg et al., 1996], suggesting a 7% to 10% increase in Mg/Ca per salinity unit. Lea et al.  concluded from cultivating experiments with Orbulina universa that a change of 1 salinity unit causes a 4 ± 3% increase in Mg/Ca.
The influence of changes in ice volume on the δ18O record is negligible for the studied time interval. Glacial to interglacial variations in global ice volume during the early Pliocene were only on the order of 0.15–0.25‰ [Hodell and Warnke, 1991; Kennett and Hodell, 1993; Tiedemann et al., 1994; Shackleton et al., 1995; Mix et al., 1995] and occurred on obliquity-related cyclicities instead of the precession-related cyclicities encountered in the critical time interval. The temperature effect cannot be extracted from the δ18O record because the Mg/Ca temperatures are subject of the correction and no other independent temperature estimate is available. Nevertheless, as the influence of temperature and salinity are working in opposite directions on the δ18O, the δ18O and Mg/Ca records show a positive correlation to each other, and as the δ18O record is dominated by variations in SSS [Steph et al., 2006], the variations in δ18O can be assumed to be minimum changes in salinity.
We converted the δ18O fluctuations of up to 1.4‰ for the time interval 4.5–3.9 Ma in salinity variations on the order of maximum three salinity units, when applying Broecker's  rule of thumb: 1‰ δ18O = 2 salinity units (Figure 8). Subsequently, we corrected the Mg/Ca record for the effect of these minimum salinity changes by using a 7% increase in Mg/Ca with 1 unit increase in salinity [Nürnberg et al., 1996]. Salinity variations of up to 3 salinity units deduced from the planktonic δ18O record would therefore result in an overestimation in Mg/Ca of maximum ∼21%, equal to ∼1.5 mmol/mol. Correcting Mg/Ca for this overestimation leads to SSTMg/Ca estimates which are up to 3.5°C lower. After the correction SSTMg/Ca shows maximum amplitudes of ∼4.5°C, and maximum SSTMg/Ca estimates up to 32°C (in comparison to 35°C and 5°C amplitude before the correction). The average SSTMg/Ca increase from 4.5–3.9 Ma is ∼3°C (Figure 8). As this correction is only a first estimation, the absolute values are probably still too high. With the availability of an independent temperature proxy the absolute changes in salinity could be determined and used to correct the Mg/Ca record. A first effort was made by Gussone et al. , who used low-resolution early Pliocene records to isolate the salinity component in δ18O by subtracting temperature estimates based on δ44Ca on G. sacculifer, and subsequently correcting Mg/Ca for the influence of salinity.
In summary, we state that the estimation of a 3°C increase in SSTMg/Ca between 4.5 and 3.9 Ma as a result of the salinity correction can be tested for its paleoceanographical validity by only considering the minimum Mg/Ca values for this interval. The minimum Mg/Ca values correlate with the minimum δ18OG. sacculifer values and therefore reflect ages where SSS did not differ significantly from the period before 4.5 Ma. Therefore these samples are considered not to be affected by the influence of salinity on Mg/Ca. Also, since the minimum Mg/Ca values do neither contain any indication for crystalline overgrowths nor are correlated with those intervals where diagenesis occurred, they can be considered as reflecting a primary temperature signal. Accordingly, the trend in SSTMg/Ca between 4.5 and 3.9 Ma, indeed, shows a warming with ∼2.5°C (Figures 3 and 8), possibly showing the initiation of the Western Atlantic Warm Pool.
We constructed a high-resolution Mg/Ca record on the planktonic foraminifer G. sacculifer in order to explore the change in sea surface temperature (SST) due to the shoaling of the Isthmus of Panama as well as the impact of secondary factors like diagenesis and large salinity fluctuations. The study covers the latest Miocene and the early Pliocene (5.6–3.9 Ma) and was combined with δ18O to isolate changes in sea surface salinity (SSS). Our results imply a relatively uniform SSTMg/Ca pattern of 26.5°C–29°C for the time interval 5.6–4.5 Ma, and moderate fluctuations in δ18O representing small fluctuations in global ice volume, salinity, and SST.
After 4.5 Ma, we show a positive correlation between Mg/Ca and δ18O, high SSTMg/Ca are paralleled by high δ18O values on a precessional timescale. Mg/Ca ratios reach up to ∼7 mmol/mol, which would translate into temperatures of ∼35°C with amplitudes of up to 5°C. Amplitudes in δ18O reach 1.4‰ with a general increasing trend toward higher values. We discussed two possible explanations for the anomalous Mg/Ca ratios, a diagenetic overprint on the foraminiferal tests and the possible influence of large salinity variations on Mg/Ca.
SEM analyses of foraminiferal tests of G. sacculifer showing anomalously high Mg/Ca ratios reveal crystalline overgrowths. Foraminiferal Sr/Ca ratios, carbonate mineralogy, and calcareous nannofossil analysis suggest that diagenetically induced aragonite dissolution and reprecipitation caused these overgrowths. LA-ICP-MS measurements, however, show that the crystalline overgrowths do not contain significantly higher Mg/Ca ratios than the foraminiferal tests. Also, calculations based on pore water data from Site 1000 result in possible Mg/Ca ratios of the crystalline overgrowths which can both be higher and lower than the primary Mg/Ca ratios. Hence it appears that even though a diagenetic overprint is present, it possibly is not responsible for the anomalous Mg/Ca ratios. However, we do acknowledge that our approach does not exclude alternative diagenetic processes like diffusion or (partial) recrystallization, which might have caused alteration throughout the foraminiferal tests.
The large fluctuations on precessional timescales in the δ18O record of Site 1000 correlate well with the diagenetically induced cyclicities in aragonite, Sr/Ca, and productivity. As these precession cycles are not present in the δ18O of G. sacculifer in the southern Caribbean Site 999, it is possible that the δ18O signal of Site 1000 contains a diagenetic overprint. The same comparison, however, of the planktonic foraminifer G. crassaformis and the benthic foraminifer C. wuellerstorfiδ18O records between Site 999 and 1000 shows an absence of precession cycles and comparable values, arguing against a diagenetic overprint on δ18O. Also, the large fluctuations in SSS were previously interpreted paleoceanographically by the precessional alternation between relatively fresh Pacific surface water masses and higher saline Atlantic surface water masses bathing Site 1000 [Steph et al., 2006].
The large fluctuations in SSS as suggested by the δ18O record provide an alternative explanation to the anomalous Mg/Ca ratios after 4.5 Ma [Steph et al., 2006]. Cultivating experiments suggest that salinity exerts an influence on foraminiferal Mg/Ca [Nürnberg et al., 1996; Lea et al., 1999]. In case the δ18O record is assumed to represent minimal changes in salinity, the Mg/Ca record can be corrected for the influence of salinity. The resulting corrected Mg/Ca record shows an increase in Caribbean SSTMg/Ca after 4.5 Ma of ∼3°C.
In summary, we state that the Mg/Ca record of Site 1000 after 4.5 Ma can only reliably be considered for paleoceanographical purposes when the minimum values of the record are converted into SST. These are the samples which do not contain any signs of diagenetic overprint, and represent the δ18O values which do not indicate large variations in SSS in comparison with part of the record before 4.5 Ma.
This research used samples and data provided by the Ocean Drilling Program, which is sponsored by the U.S. National Science Foundation and participating countries under management of Joint Oceanographic Institutions. This study was performed within DFG-Research Unit “Impact of Gateways on Ocean Circulation, Climate, and Evolution” (FOR451/1-1 TP B2). We would like to thank Silvia Koch, Kerrin Wittmaack, Nicole Gau, Jutta Heintze, Kristin Nass, Beate Bader, Ute Schuldt, and Gijs Nobbe for sample preparation and laboratory assistance. We gratefully thank Marcus Regenberg, Joachim Schönfeld, Martin Ziegler, Stephan Steinke, and two anonymous reviewers for their useful comments and the improvement of the manuscript.