Age models and core sampling
The Northwest Atlantic Ocean samples are the same as those used by ref. 14; therefore, we adopt the same age models and same sampling strategies (Extended Data Table 1). For our Northeast Atlantic cores, we also use previously published age models51,52. Stratigraphic information for each core, including multi-species benthic and planktic foraminiferal δ18O and age constraints, is also shown alongside our multi-proxy temperature estimates in Extended Data Fig. 6.
Multi-species and multi-proxy approach
As no single species of benthic foraminifera is present across all depths between 1.5 km and 5 km, we used a multi-species approach, measuring trace-metal ratios in multiple different taxa. We also took a multi-proxy approach, reconstructing deep-ocean temperatures using independent techniques: Δ 47 and trace-metal ratios, thus providing additional support for the individual proxy reconstructions.
Trace-metal analyses
Monospecific benthic foraminiferal samples consisting of between 5 and 15 individuals (where possible) were picked from the >250-μm sediment size fraction (where foraminiferal abundances were low, we also picked from the >212-μm fraction). Foraminifera were then gently crushed between glass slides, after which approximately one-third of the material was removed for isotopic analyses (these data were previously reported14). The remaining material was then loaded into 500-μl Bio-Rad polypropylene microcentrifuge tubes that had been pre-leached in hot 10% hydrochloric acid and used for trace-metal analysis.
All samples underwent oxidative and reductive cleaning following established methods53, before each individual sample was analysed for a suite of trace and minor elemental ratios on a Thermo Finnigan Element2 magnetic sector inductively coupled plasma mass spectrometer at the University of Colorado, Boulder, as described in ref. 54. Long-term ±1σ precision is 0.5% for Mg/Ca, 0.9% for Li/Ca and 4.2% for B/Ca. Mn/Ca, Al/Ca and Fe/Ca ratios were also measured to screen for potential contamination from detrital material and/or secondary phases. For low-Mg species (for example, Melonis pompilioides), samples with contaminant ratios >0.1 mmol mol−1 were rejected. For high-Mg species, for example, G. affinis, this threshold was scaled accordingly55. However, in cases where contamination indicators were only marginally above threshold and/or Mn/Ca values were elevated, but Mg/Ca was in good agreement with multiple coeval samples with low contaminant ratios, these data were retained (individual measurements are provided in the Source Data).
We also omitted all trace-metal data from the shallow infaunal benthic foraminifera Uvigerina peregrina, despite suggestions that it is less affected by ΔCO 3 2− (ref. 56). This is because the global calibration poorly constrains the relationship between Mg/Ca and temperature (R2 = 0.68)57, particularly at the cold end of the calibration (<5 °C), which is most relevant to the deep ocean. The temperature sensitivity (approximately 0.07 mmol mol−1 °C−1) is also substantially lower than that of other shallow, low-Mg infaunal species and does not align well with our common calibration dataset (Extended Data Fig. 2f). As a result, U. peregrina Mg/Ca yielded implausibly warm glacial temperature estimates (>5 °C) at our core sites, which are not compatible with our clumped isotope temperature estimates from this species.
Trace-metal temperature calibrations
Deep-ocean temperatures were reconstructed using benthic foraminiferal Mg/Ca and species-specific calibrations taking the form: Mg/Ca = s × T + c, where T is the calcification temperature in °C, and s and c are the calibration slope (temperature sensitivity) and intercept, respectively (Fig. 2a,b). One-sigma uncertainties on individual monospecific temperature estimates were propagated, incorporating analytical error (assumed as a relative percentage error on Mg/Ca (see ‘Trace-metal analyses’)) and uncertainties in both the slope and intercept (Extended Data Fig. 4a,c,d).
... continue reading