The sediment core TN057-06-PC4 (hereafter TN057-6) was retrieved from the Agulhas Ridge (42º 54.8' S, 8º 54.0' E, 3751 m) during the pre-site survey of the Ocean Drilling Program (ODP) Leg 177 at nearly the same position as ODP Site 1090 (42° 54.5′ S, 8° 54.0′ E, at 3,702 m). The high-resolution XRF Fe concentration records from ODP Site 1090 and TN57-06 were used to align the two cores using the software Analyseries (46). The good correlation of the two records after the alignment allows us to directly compare the measurements performed in the two cores with minimal stratigraphic uncertainty (Fig. S1a). This indicates that the 230 Th measurements performed in can be used to reconstruct sedimentary fluxes of different elements and/or organic compounds in both records.For example, in Figure S1c, the high-resolution 230 Th-normalized mass flux (Fig.S1b) was used to calculate iron fluxes both in ODP Site 1090 and TN057-6 ( The age model of our record was generated by graphic correlation of the highresolution 230 Th-normalized Fe flux from ODP Site 1090 to the most recent ice core dust reconstruction from EPICA Dome C (44), using ice core chronology (AICC2012) (47,48) (Fig. S2). This new age model is generally in good agreement with the original ages proposed by Venz and Hodell 2002 (65) using benthic δ 18 O stratigraphy (Fig. S3). The good correlation of ODP Site 1090 Fe flux and EDC dust records down to sub-millennial timescales allows us to directly compare our marine data to other ice core measurements (e.g., atmsopheric pCO 2 ) with minimal stratigraphic uncertainty (Fig. S2). Foraminifera-bound δ 15 N analysisAround 600-800 specimens of G. bulloides and O. universa were picked manually under a dissecting microscope (250-425--6 mg of clean foraminiferal calcite. Foraminifera-bound δ 15 N was measured at Princeton University following the protocol described in (23,32). In the intervals where foraminifera abundances were adequate, samples were measured in duplicate. The reported error 2 bars in Fig. S9 represent the standard deviation estimated from the means of duplicated measurements. The average standard deviation of all the replicated measurements (from foraminifera-bound organic matter oxidation onward) was 0.19‰. Bulk sediment δ 15 N and %Nitrogen analysisThe N content and δ 15 N of the bulk sediment were analyzed using a Thermo FisherSeries 1112 elemental analyzer coupled with a Thermo Fisher Delta V Plus mass spectrometer at ETH Zurich. Between 50 and 100 mg of sediment were used for eachanalysis. An in-house peptone standard, which has been referenced to international reference materials (IAEA-N1 and IAEA-N2), was measured in each sample run and used for the final corrections. The standard deviation for the peptone standard among the different runs was <0.1‰. 230 Th and opal analysisConcentrations of U and Th isotopes were measured by inductively coupled-plasma mass spectrometry after strong acid sediment digestion and chromatographic separation as described by Fleisher and Anderson, 2003 (49). Opal...

Fixed nitrogen (N) is a limiting nutrient for algae in the low-latitude ocean, and its oceanic inventory may have been higher during ice ages, thus helping to lower atmospheric CO2 during those intervals. In organic matter within planktonic foraminifera shells in Caribbean Sea sediments, we found that the 15N/14N ratio from the last ice age is higher than that from the current interglacial, indicating a higher nitrate 15N/14N ratio in the Caribbean thermocline. This change and other species-specific differences are best explained by less N fixation in the Atlantic during the last ice age. The fixation decrease was most likely a response to a known ice age reduction in ocean N loss, and it would have worked to balance the ocean N budget and to curb ice age-interglacial change in the N inventory.