From the > 150 µm size fraction, a total of 48 specimens of the benthic foraminifera species Eubuliminella exilis were hand-picked under a binocular microscope for SIMS analysis (Table 1). We acknowledge that the presence of Mn-rich authigenic coatings (e.g. Mn (oxyhydr)oxides and organic matter) can be problematic for trace-elemental analysis of foraminifera (Boyle, 1983; Pena et al., 2005; Klinkhammer et al., 2009). Therefore, we employed a rigorous pre-treatment cleaning technique to remove possible organic contamination following the method of Glock et al. (2012) and avoided potential diagenetic coatings during the SIMS analysis by measuring within the massive centre of the test walls (Fig. 2).

For the SIMS analysis, foraminifera from individual sample depths were rinsed over a 63 µm sieve with Milli-Q water. After this rinsing step, the foraminifera were transferred into vials and sonicated for 20 s. Subsequently, the foraminifera were rinsed with methanol and sonicated again for 1 min. Any residual methanol was then removed with Milli-Q water. An oxidative cleaning step was performed to remove organic matter which consisted of mixing the following reagents: 100 µL 30 % H2O2 to 10 mL of 0.1 M NaOH solution. Three hundred and fifty microlitres of this reagent was added to each individual vial, and the vials were put into a water bath at 92 ∘C for 20 min. Afterwards another 20 s sonic bath was undertaken; the foraminifera samples were rinsed again with Milli-Q water in the 63 µm sieve to remove any residues. For the final step, the specimens were transferred back into their respective vials and 250 µL of 0.001 M HNO3 was added to each vial. The vials were put into a sonic bath for 20 s and finally rinsed one last time with Milli-Q water. After the cleaning procedure, the specimens were checked under a binocular microscope to ensure sufficient cleaning and that the tests remained intact.

Foraminifera specimens were embedded in low-viscosity epoxy resin at JAMSTEC, Japan. The foraminifera were then ground to expose a cross section across the test wall using 16 µm silicon carbide paper at the Department of Geosciences, University of Edinburgh, UK. Resin pieces were mounted into low-viscosity epoxy resin disks (Struers) at the NORDSIM laboratory, Laboratory for Isotope Geology at the Swedish Museum of Natural History, Stockholm, Sweden. The mounts were polished using a Struers Rotopol-2 at 150 rpm for 1 min, first with 3 µm diamond suspension and again with 1 µm diamond suspension. Between each grinding and polishing step, mounts were cleaned with ethanol. Each cross-sectioned foraminifera test was examined under high-power reflected light microscopy to evaluate the quality of the carbonate and to assist in assessing the progress of polishing until the cross sections were clear. Subsequently, the mounts were cleaned in high-purity ethanol and coated in a 20-nm thick, high-purity Au coat.

The reference material used for the SIMS was a polished piece of Oka calcite crystal supplied from GEOMAR, Kiel University, Germany (E. Hathorne, pers. comm). This standard was obtained from a matrix-matched specimen for which Mn / Ca has been reported by solution ICP-MS (Glock et al., 2012). During calibration, the Oka was analysed n=16 times, yielding a high sensitivity with 1 standard deviation repeatability of 1.2 % for Mn concentration.

The Mn / Ca analyses of the test cross sections were performed using a Cameca IMS 1280 ion microprobe at the NORDSIM laboratory at the Swedish Museum of Natural History, Stockholm, Sweden. Analysis used a 16O2- ion beam accelerated at 23 kV impact energy (-13 kV primary beam, +10 kV secondary beam). It is vital to only analyse Mn which is located internally within the original test wall to attain the most representative Mn / Ca for developing it as a redox proxy. Therefore, a 50 µm aperture in the primary column was used to shape a slightly elliptical 5 µm spot on the sample surface, which, together with careful placement, reduced the effects of sample contamination from the test wall outer surface. Prior to each analysis, the analytical location was pre-sputtered for 2 min with the ion beam rastered over a 10 × 10 µm raster to remove the Au coat and any remaining surface contamination. During the initial pre-sputtering, the 44Ca distribution was monitored using the ion imaging system of the instrument and maximised to ensure high-precision beam targeting on the fine foraminiferal test walls. The mass spectrometer was operated at a mass resolution of M/ ΔM ∼ 6000 to resolve the 55Mn peak from nearby molecular interferences. A 400 µm contrast aperture was employed for maximum transmission together with a 60 µm entrance slit, a 2001 µm field aperture restricting the field of view on the sample to an area of ca. 12 × 12 µm at the transfer magnification of ca. 160×, and a 45 eV wide energy window – all of which combined to yield adequately flat-topped peaks on the species of interest. Each analysis comprised of 16 cycles of 44Ca (1 s integration cycle-1) and 55Mn (2 s). Each analysis lasted approximately 9 min. Secondary ions were measured using a low-noise (< 0.01 cps) ion-counting electron multiplier. Multiple analysis points were undertaken upon each individual test of E. exilis (ca. 6–10 targets per individual specimen) starting from the aperture and taking measurements alternating between the outer wall and internal walls. For the best targets, programming was performed manually to ensure that widest chamber walls and “t” junctions were targeted since they provide a wider test wall for the analyses (Fig. 2). Furthermore, at such high spatial resolution and precision, it is easy to visually observe and avoid encrusting prior to selecting analysis targets via the connected screen and avoid measuring secondary calcite or authigenic clays which would otherwise affect measurements. With cautious positioning of the primary beam on the test walls and observations of the element distributions during measurements, such detrital material and potential contaminants were avoided and annulled. Therefore only the elements actually incorporated into the calcitic tests were measured. As an additional prerequisite to this, analyses with Ca values > 500 kcps were classified as being reliable. Mn / Ca was first normalised to those determined in the Oka standard and subsequently converted to the true value in the Oka based on the Glock et al. (2012) value for Mn / Ca of 4920 µmol mol-1.

The advantage of SIMS is that it is non-destructive, and, as the foraminiferal cross sections are preserved within the mounts, they can be stored for further analyses. Mounts are archived at the NORDSIM laboratory.

For FT-ICP-OES, 20–50 specimens per sample depth of E. exilis from the GeoB7926-2 record were selected from samples corresponding to H1, BA and the YD for comparisons with the SIMS data. These three climatic intervals encompassed the only samples where a sufficient number of pristine E. exilis individuals were present. The tests were gently crushed in a 0.5 mL vial, and fragments were transferred into a PTFA filter with 0.45 µm mesh.

For analysis, the filters were connected to a flow-through automated cleaning device (Klinkhammer et al., 2004; Haarman et al., 2011). Automatic cleaning prevents the loss of material which occurs with traditional cleaning, allowing the analysis of very small samples (∼ 20 µg). The flow-through was then connected to an ICP-OES (Agilent Technologies, 700 Series with autosampler (ASX-520 Cetac) and micro-nebulizer). Time-resolved analysis (TRA) was used to analyse the samples at MARUM, University of Bremen, Germany. After an initial rinse (5 min) with buffered Seralpur water to remove clays, the samples were slowly dissolved using an acid ramp formed by mixing of Seralpur with 0.3 M QD HNO3 (no additional oxidative and/or reductive cleaning was performed). Starting with pure Seralpur the acid contribution was stepwise increased every minute to 100 % acid after 30 min. The flow speed of the solution was 250 µL L-1. Mn / Ca of the foraminiferal calcite was determined by identifying the TRA interval which showed a consistent linear relationship between Mn and Ca counts. Potential diagenetic phases like Mn(oxy)hydroxides are avoided this way, as they would have a different slope, i.e. a significant Mn signal along with the absence of a Ca signal. Mn / Ca was then calibrated using the characteristic slope of this linear relationship of a known in-house standard solution analysed on the same acid ramp. The average standard error on the determination of the slope for Mn / Ca was 0.75 %. We analysed an international limestone standard (ECRM752-1) for Mg / Ca to validate the results following the same FT protocol. The average Mg / Ca of the ECRM752-1 (n=4) was 3.76 mmol mol-1, which compares well with the average published value of 3.75 mmol mol-1 (Greaves et al., 2008).

For geochemical bulk analyses, samples were dried and ground in an agate mortar and homogenised. Total dissolution of samples was undertaken using HF and HNO3 following the standard procedures of Gallego-Torres et al., (2007). Mn and Al content for samples corresponding to 35–10 ka was determined by atomic absorption spectrometry, using Re and Rh as internal standards at the Analytical Facilities (Centro de Instrumentación Cientifica) at the University of Granada, Spain. Redox-sensitive Mn was normalised to Al content in order to correct for detrital variations (van der Weijden, 2002). Mn / Al data corresponding to 25–10 ka have previously been published (Gallego-Torres et al., 2014), and here we extend the Mn / Al record to 35 ka.

Overall, Mn / Ca determined by SIMS varied between 2 and 750 µmol mol-1. Average values ranged from 6.5 to 260 µmol mol-1 throughout the record and displayed a decreasing trend down-core (Figs. 3–4, Table 1).

Mn / Ca was lowest in foraminiferal tests from MIS3 and the LGM, with values ranging from 25 to 68 (average 50 µmol mol-1) and from 2 to 225 µmol mol-1 (average 70 µmol mol-1) respectively. Foraminifera from samples derived from H1 and BA exhibited a slightly elevated range of Mn / Ca at 50–380 (average 117 µmol mol-1) and 27–280 µmol mol-1 (average 133 µmol mol-1) respectively. Highest Mn / Ca occurred in foraminiferal tests from the YD (average 175 µmol mol-1) period ranging from 23 to 750 µmol mol-1 (Fig. 3, Table 1).

In order to determine statistically significant differences between samples from different productivity regimes, Kruskal-Wallis tests were run and showed a statistically significant difference between the mean Mn / Ca values (per individual foraminifera specimen) between all five climatic intervals (p=0.003). By testing the mean Mn / Ca of each foraminifera between each climatic interval individually in turn, using post hoc (Mann–Whitney U) tests, significant differences lay between climatic intervals with high- and low-export-productivity regimes (based on diatom accumulation rate). Namely, significant differences in Mn / Ca were evident between the YD interval and the LGM and MIS3 (Table 2).

Mean SIMS-derived Mn / Ca per sample depth compares well with FT-ICP-OES results (Table 1), agreeing with maximum differences of 15–24 µmol mol-1 (Fig. 4). Mn / Ca from the FT-ICP-OES on bulk foraminiferal samples tended to be slightly higher compared to the mean ratios obtained from the SIMS microanalyses for the same sediment depth (for example 140 µmol mol-1 compared to 116 µmol mol-1 during the YD).

Bulk sedimentary Mn / Al showed highest values during MIS3 at 30–24 ka as well as during 19–17.5 ka and the YD. The YD was characterised by a sharp Mn / Al increase at 12.3 ka, coinciding with maximum diatom productivity (Fig. 4). Relatively low Mn values occurred during 35–32 ka, and during the LGM, Mn was below the level of detection (< 0.06 %). Between 32 ka and the onset of the LGM, a progressive increase was observed.

Our results indicate that Mn / Ca in benthic foraminifera might prove to be a valuable proxy for oxygen in the bottom and pore waters. The down-core variability in foraminiferal Mn / Ca at site GeoB7926-2 displays a consistent agreement between the mean SIMS-determined Mn / Ca of each sample depth and the bulk foraminifera Mn / Ca measured by FT-ICP-OES. In general, the agreement in values suggests that the SIMS-determined Mn / Ca is likely to be a true signal within our reconstruction. The slightly higher Mn / Ca determined by FT-ICP-OES in comparison to SIMS-derived Mn / Ca highlights the issue of comparing bulk foraminiferal samples with individual tests comprising only 6–10 analytical targets. Overall, when a sufficient number (minimum weight of 0.1 mg) of benthic foraminiferal specimens are not available in sediment samples for solution-based analyses (in this case from 35 to 18 ka), SIMS has the potential to provide reliable results from a few individuals to compensate for this.

Recent culturing experiments on benthic foraminifera demonstrate that calcification can occur even under anoxic conditions (Nardelli et al., 2014). This is key for the discussion of trace-elemental data derived from the foraminiferal tests, as not only does the timing of the calcification determine the geochemical signature, but it also shows that the signature is recorded in a wide range of oxygen conditions.

When comparing foraminiferal Mn / Ca to previously published sediment measurements of Mn / Al (Fig. 4) for site GeoB7926-2 (Gallego-Torres et al., 2014), in general we do not find a continuous relationship in trends throughout the record, but overall they largely agree on the former oxygen conditions. This is most likely due to diagenetic processes and migration of redox fronts through the sediment which redistributes the bulk Mn after deposition, whereas the foraminiferal tests record the Mn concentration at the time of calcification. In fact, bulk sediment Mn concentrations are often interpreted as being related to diagenetic (post-production) oxidation fronts and less often to the syn-sedimentary environment (e.g. Thomson et al., 1995; de Lange et al., 2008). Thus, two different processes govern Mn fixation in sediment and foraminifera tests, and consequently we can expect an offset between the two signals.