Individual species were identified and 550 individuals per species were picked from the >250 µm fraction under dissecting microscope for each sample. Approximately 7 mg of picked and identified foraminifera shells were crushed between glass microscope slides and rinsed with MilliQ water. Samples were cleaned prior to δ¹⁵NFB measurement as described in Marcks et al., (2023). Once samples were clean, organic nitrogen was released into solution by acid dissolution of the foraminiferal calcite. Samples were acidified prior to measurement. Nitrate concentrations were measured by chemiluminescence on a Teledyne Instruments (Model 200E) chemiluminescence NO/NOx analyzer. δ15NFB samples, 10 nmol in size, were measured by bacterial conversion of nitrate to nitrous oxide, with measurement of the δ¹⁵N of the nitrous oxide by automated extraction and gas chromatography-isotope ratio mass spectrometry on a Thermo Delta V Plus IRMS. The potassium nitrate reference materials IAEA-N3 and USGS 34 (+4.7 ‰ and 1.8 ‰, respectively) were used to standardize results (Gonfiantini et al., 1995). Note, testing of a subset of 6 samples, each with full procedural triplicates, for a total of 18 samples, showed negligible differences in nitrogen content and δ¹⁵NFB values with and without a reductive cleaning step, and so it was omitted here to avoid unnecessary loss of sample material. Sample replicates and triplicates were analyzed when possible. Full procedural replicates were analyzed for 134 sample splits, representing 66 unique samples, when enough foraminifera were available for duplicate or triplicate analysis. The average standard deviation of procedural replicates is 0.4 ‰. Full operational blanks and amino acid standards (USGS 65 glycine) were measured in each batch. The average standard deviation of glycine standards measured in triplicate is 0.3 ‰. We estimated the δ¹⁵N value of the persulfate blank using a dilution series (5, 7.5, 10, and 20 μM of the glycine standard and the fraction of the blank in standards. We applied a blank correction to each sample based on the calculated mean δ¹⁵N value of all of the persulfate blanks for the dataset and the fraction of the blank in the N content of each sample. Data were subset to exclude N content outliers (>2 s.d. from mean and where the blank was greater than 20 % of the sample N content, with significantly different δ¹⁵N values from other replicates). Full propagated analytical error associated with measurement and blank correction, following Higgins et al., (2009, doi:10.1021/ac8017185 ), was on average 0.6 ‰. Propagating the errors, including not only the procedural replicates and their variance, but the relative size of the blanks, the mean of the calculated blank δ¹⁵N values (5± 10 ‰). The mean value, 0.6 ‰, is used where procedural replicates were limited by sample availability. The age model is based on the benthic oxygen isotope stratigraphy presented by Starr et al. (2021, doi:10.1038/s41586-020-03094-7). This age model for Site 361-U1475 was generated with 12 radiocarbon dates and 33 benthic oxygen isotope tie points which were graphically aligned with a probabilistic stack of 180 globally distributed benthic oxygen isotope records (Starr et al., 2021, doi:10.1038/s41586-020-03094-7).