ALBERT

Description: Samples were processed for GDGTs at the Birmingham Molecular Climatology Laboratory, University of Birmingham. Lipids were extracted from ~10-15 g of homogenized sediment by ultrasonic extraction using dichloromethane (DCM):methanol (3:1). The total lipid extract was fractionated by silica gel chromatography using n-hexane, n-hexane:DCM (2:1), DCM, and methanol to produce four separate fractions, the last of which contained the GDGTs. Procedural blanks were also analyzed to ensure the absence of laboratory contaminants. Samples were filtered using hexane:isopropanol (99:1) through a 0.4 µm PTFE filter (Alltech part 2395), before being dried under a continuous stream of N2. Samples were then sent to the University of Bristol for analysis by LC-APCI-MS. HPLC-APCI-MS analyses were conducted at the National Environmental Isotope Facility, Organic Geochemistry Unit, School of Chemistry, University of Bristol, with a ThermoFisher Scientific Accela Quantum Access triple quadrupole MS in selected ion monitoring (SIM) mode. Normal phase separation was achieved using two ultra-high performance silica columns (Acquity UPLC BEH HILIC columns, 50 mm × ID 2.1 mm × 1.7 µm, 130 Å; Waters) were fitted with a 2.1 mm × 5 mm guard cartridge after Hopmans et al. (2016). The HPLC pump was operated at a flow rate of 200 µL min-1. GDGT determinations were based on single injections. A 15 µL aliquot was injected via an autosampler, with analyte separation performed under a gradient elution. The initial solvent hexane:iso-propanol (IPA) (98.2:1.8 v/v) eluted isocratically for 25 min, followed by an increase in solvent polarity to 3.5 % IPA in 25 min, and then by a sharp increase to 10 % IPA in 30 min (Hopmans et al., 2016). A 45 min washout period was applied between injections, whereby the column was flushed by injecting 25 µL hexane:isopropanol (99:1 v/v). GDGT peaks were integrated manually using Xcalibur software. In-house generated standard solutions were measured daily to assess system performance. One peat standard was run in a sequence for every 10 samples and integrated in the same way as the unknowns. Selected ion monitoring (SIM) was used to monitor abundance of the [M+H] + ion of the different GDGTs instead of full-scan acquisition in order to improve the signal-to-noise ratio and therefore yield higher sensitivity and reproducibility. SIM parameters were set to detect the protonated molecules of isoprenoid and branched GDGTs using the m/z (Schoon et al., 2013). The majority of sediments were found to contain a full range of both isoprenoid and branched GDGTs. Sea surface temperature (SST) estimations from GDGT assemblages are show based on two methodologies: the BAYSPAR Bayesian regression model of Tierney and Tingley (2014, 2015) using the 'analogue' version for deep-time applications; and, the OPTiMAL Gaussian process model of Dunkley Jones et al. (2020). When plotting BAYSPAR SSTs we distinguish samples with BIT indices greater than and less than 0.4, as high BIT can be associated with a small warm bias (Weijers et al., 2006). For the OPTiMAL model we apply its own internal screening criteria that quantifies the extent that fossil GDGT assemblages are non-analogue relative to the modern calibration data, using the Dnearest criteria with a cut-off value of 0.5. All but one pre-NIE GDGT assemblages have Dnearest values that exceed 0.5, whereas eight samples above this level have values less than 0.5.Only OPTiMAL SST data that pass the Dnearest screening criteria are shown.