*3.1. Sediment Core Locations and Methodological Overview*

The present study is based on the weight analyses of foraminifer shells from three sediment cores (Table 1). *G. bulloides* shell weights from the 300–355 µm sieve fraction existed for cores ODP 982 [39] in the North Atlantic and GeoB 8502-2 [40] in the Tropical Atlantic and thus the weight analysis was extended to specimens from core GeoB 1710-3 in the South Atlantic. GeoB 1710-3 is 10.45 m long with an average sedimentation rate of ≈ 5 cm/ky [41] and extends back 245 kyr to Marine Isotope Stage (MIS) 7, yet the present analysis was restricted to the last 200 kyr to match the extent of GeoB 8502-2. Samples from GeoB 1710-3 were taken at a resolution of ~2000 years (10 cm sampling interval) by extracting a slice of material, 1 cm in thickness, which corresponds approximately to an average of 200 years of sedimentation. The coarse fraction of all samples was already available from a previous study [42] and was subsequently dry-sieved into several sieve fractions. Non-fragmented *G. bulloides* shells from the 315–355 µm size fraction were picked for mass analysis. The very narrow size interval (40 µm) should be sufficient to overcome the greater proportion of natural size variability without further normalization [43]. Since any record of shell weight is a composite signal of dissolution superimposed upon the initial shell weight variability, the cores considered here were already assessed for carbonate dissolution using the same method (ODP 982 [17], GeoB 8502-2 [40], GeoB 1710-3 [44]) and their foraminiferal carbonate is reported to be well preserved. The coarse fraction from the studied cores was disaggregated with deionized water and then wet-sieved through a 63 µm mesh. Since all samples underwent the same washing process, any offset due to residual fine debris would be constant among samples. Although treatment solely with water is not a very efficient cleaning method for weighing analyses [45], the examined specimens did not show increased contamination (see Section 4).

**Table 1.** List of the core sites.


The weight analyses revealed two instances where *G. bulloides* shell masses between the studied cores converge and these convergence intervals were termed Shell Mass Convergence Event (SMCE) I and II, during the last (~18.4 ka) and the penultimate (~122.4 ka) deglaciations, respectively. In order to better understand the physiology of the shells during these intervals, the selected specimens from GeoB cores were analyzed by high resolution X-ray microcomputed tomography (XµCT). The tomographic analysis was extended to the shells that mark the Last Glacial Maximum (LGM) in core GeoB 1710-3, which were the heaviest found in all three records during the last 200 kyr. µCT was used to inspect the interior and the internal structure of the foraminiferal tests. Apart from addressing the test's integrity, XµCT allowed for the assessment of the degree to which the recorded masses are the result of interference from shell inclusions or of changes in test thickness. Finally, the µCT analysis led to total shell volume estimates that allowed for the calculation

of volume-normalized shell weights or *G. bulloides* shell densities, presenting a more precise method of eliminating the contribution of shell size to shell weight.

## *3.2. Weight Analysis*

Where available, ideally, 50 (minimum 31) *G. bulloides* shells were weighed in a preweighed aluminum carrier in the Department of Earth Sciences at the University of Oxford using a Sartorius SE 2 ultra-microbalance with a precision of ± 0.1 µg. Replicate weight measurements of specimens from core ODP 982 were performed in the Godwin laboratory at the University of Cambridge using a UMX2 Mettler Toledo ultra-microbalance at the same precision. Average shell masses were calculated by dividing the recorded mass by the total number of foraminifera weighed. Subsequently, for each sample, the average shell weights of batches of five individuals (minimum four, maximum eight) were determined, in order to estimate standard shell mass deviations. As explained above, performing shell weight analyses on a narrow size fraction of foraminifera constrains the ontogenic stage of the specimens to a certain number of chambers, and thus minimizes size-related weight differences [43,46]. The analytical error, estimated by triplicate measurements of 50 random specimens, ranged from 0.04 to 0.06 µg for both balances, which is in accordance with their analytical error.

#### *3.3. Determination of Atlantic Seawater Paleodensity*

The acquired shell weight measurements were converted to paleoseawater densities using Equation (1) that was derived by correlating weight and geochemical analyses of equally sized *G. bulloides* specimens between both 250–315 and 300–355 µm from the Atlantic Ocean [17]:

$$\text{Seawater density} = 0.29(\pm 0.01)^{\circ} \text{ G.}\text{ bullioides shell mass} + 1022.78(\pm 0.11) \tag{1}$$

This equation is based on the hypothesis that foraminifera shell masses can be used as a direct (paleo)seawater density proxy and is considered to describe the ocean density at 100 m depths. When *G. bulloides* shell weights approach zero, like the smallest juvenile tests, the density approaches 1023 kg/m<sup>3</sup> , thus describing well the modern average surface ocean density. The weight-derived ocean densities were also compared to published geochemically derived seawater densities for the penultimate deglaciation from core GeoB 8502-2. The propagated error from the transformation of shell mass to seawater density is ± 0.23 kg/m<sup>3</sup> .

## *3.4. X-ray Micro-Computed Tomography (µCT)*

For X-ray microscopic analysis, in total, 28 specimens were scanned from four samples of cores GeoB 8502-2 and GeoB 1710-3 that correspond to the time intervals of shell mass convergence during the last and penultimate deglaciations (SMCE I and II), and a sample from the last glacial maximum in core GeoB 1710-3, where the highest shell weights of all the studied records were recorded. The dataset was complemented with CT data previously published in Zarkogiannis, et al. [40] that refer to the time-slice of SMCE II in core GeoB 8502-2. On average, seven (min four, max five) specimens were scanned from each of the studied intervals. Each batch of shells was poured into a quartz cylindrical carrier 1 mm in diameter [47]. They were stabilized with diluted tragacanth glue and left to dry prior to scanning. The micro-CT (µCT) scanning was carried out with a Zeiss Xradia 510 Versa at the Maxwell Centre of the University of Cambridge. X-ray source and detector geometry were kept constant throughout the scans. The anode voltage was set at 100 kV, the X-ray tube current was 90 µA, and the exposure time was 2 s at an optical magnification of 4 ×. By processing approximately 1024 images per sample, a scan resolution voxel size of ~1.2 µm<sup>3</sup> was typically achieved using this setup in order to maximize the number of specimens that could be analyzed in a single scan. The images were combined to build a 3D rendering using Avizo software, which was also used for segmentation. The segmentation resulted in the separation of the tomographs into shell area, area occupied by clay infillings (dirt), and internal shell (protoplasm) voids.

Subject to the degree of segmentation, the X-ray microscopic analysis allows for the determination and study of a number of biometric characteristics of the foraminifera shells, such as total shell volume, thus shell density (volume-normalized weight) and calcite (test) volume, and thus test density and calcite (test) surface [40]. The calculation of the total cell volume and the volume of shell calcite allowed for the determination of the percentage that the shell occupies within the cell. The ratio of calcite volume/calcite surface provides a linear unidimensional quantity in length units and can thus serve as a measure of average test thickness. In this study, in addition to shell density, that is, the ratio of shell volume to shell mass, we used the "specific surface area", that is, the ratio of test volume/test surface, as a measure of average test thickness [48] and the test density, that is, the ratio of test volume to shell mass, as an indication of test porosity. Furthermore, by segmenting the area occupied by clay infillings, the degree of contamination in weight measurements was calculated as percent by volume. Links to raw tomographic data can be found in the data availability statement section below.
