Modeling and proxy studies indicate that a reduction of Atlantic Meridional Overturning Circulation (AMOC) strength profoundly impacts temperatures and salinities in the (sub)tropical Atlantic, especially on subsurface levels. While previous studies focused on prominent periods of AMOC reduction during the last deglaciation, we aim to test whether similar reconfigurations of the subtropical hydrography occurred during the moderate climatic alterations punctuating the last interglacial Marine Isotope Stage (MIS) 5. Here, we present temperature and salinity records from a Florida Straits core by combining δ18O and Mg/Ca analyses on surface (Globigerinoides ruber, white) and deep-dwelling (Globorotalia crassaformis) foraminifera covering MIS 5 in high resolution. The data reveal increasing salinities at intermediate depths during interglacial cooling episodes, decoupled from relatively stable surface conditions. This probably indicates the spatial expansion of saline subtropical gyre waters due to enhanced Ekman downwelling and might also point to a changed density structure and altered geostrophic balance in Florida Straits. Notably, these oceanographic alterations are not consistently occurring during periods of AMOC reduction. The data suggest that the expansion of gyre waters into Florida Straits was impeded by the increasing influence of Antarctic Intermediate Water (AAIW) from MIS 5.5 to ∼107 kyr BP. Afterward, increasingly positive benthic δ13C values imply a recession of AAIW, allowing the temporary expansion of gyre waters into Florida Straits. We argue that the inferred transient subtropical salt accumulation and warm pool expansion might have played a pivotal role in reinvigorating meridional overturning and dampen the severity of interglacial cold phases.
 The penultimate interglacial Marine Isotope Stage (MIS) 5 is known as a period of comparatively stable climate at mid and high latitudes that was only interrupted by short cooling intervals [e.g., Chapman and Shackleton, 1999]. While there is ample evidence for reduced deep Atlantic Meridional Overturning Circulation (AMOC) during those cooling episodes [e.g., Chapman and Shackleton, 1999; Curry and Oppo, 2005], the knowledge of concomitant changes in the subsurface and intermediate water is sparse. Furthermore, most research on MIS 5 focused on the high northern latitudes, which are particularly sensitive to changes in the ocean and climate system [e.g., Bauch and Kandiano, 2007; Bauch et al., 2000; Helmke and Bauch, 2003; McManus et al., 2002]. High-resolution paleotime series from (sub)tropical oceans are less abundant, although proxy as well as modeling studies pointed out that the (sub)tropical ocean with its enormous heat and salt storage capacity plays a significant role in the oceanic and atmospheric circulation system [e.g., Rühlemann et al., 1999; Schmidt et al., 2004, 2006].
 The intention of this study is to reconstruct changes in the vertical upper ocean gradients in the Florida Straits, which is a key area in the AMOC surface path. Variations in the global ocean circulation are expected to have considerable effects on the hydrography of this important oceanic gateway. Modeling studies suggest that profound and complex changes will take place in the tropical and subtropical Atlantic when AMOC strength is reduced [Chang et al., 2008; Wan et al., 2009; Zhang, 2007]. The model results invoke a subsurface warming parallel to surface cooling between 0 and 20°N, but a general cooling in the (sub)tropics north of 20°N [Chang et al., 2008], which was accompanied by a salinity increase in the northern hemisphere tropics [Wan et al., 2010]. In order to test the model predictions, the focus of our study was to reconstruct subsurface temperature and salinity dynamics during interglacial cooling events in a conduit of the AMOC surface flow path, where different water masses from the tropical and subtropical Atlantic are encountered.
 Today, the temperatures of intermediate waters in the Florida Straits range from 12 to 24°C. They are dominated by recirculated subtropical gyre waters of the western North Atlantic (Figure 1a), while surface (>24°C) and deep (<12°C) waters are predominantly of South Atlantic origin [Schmitz and Richardson, 1991]. Subtropical gyre water constitutes a several hundred meter thick pool of warm and saline water (Western North Atlantic Central Water, WNACW) that is the ultimate source of the Subtropical Underwater (SUW). Before entering the southern Florida Straits, the SUW forms a subsurface salinity maximum in the Gulf of Mexico, being vertically less extensive than the WNACW in the subtropical Atlantic (Figure 1c). The piling-up of water masses in the subtropical gyre in combination with the Coriolis force results in a geostrophic flow through Florida Straits [Wüst, 1924] imprinted in the tilted pycnoclines at Florida Straits (Figure 1d). We expect similar conditions as today to have been prevailed during MIS 5. To reconstruct the geostrophic flow in the past, Lynch-Stieglitz et al. [1999a] used benthic δ18O records from both sides of Florida Straits, converted them into density gradients, and calculated the geostrophic transport. This approach showed a quite pronounced variability of the density structure in the Florida Straits during the past 20 kyrs [Lund et al., 2006; Lynch-Stieglitz et al., 1999a, 2009], in particular less tilted pycnoclines during the Last Glacial Maximum, interpreted to represent a significantly reduced glacial Gulf Stream strength [Lynch-Stieglitz et al., 1999a].
 In this study, we present the first surface and intermediate water temperature and salinity (δ18Osw) records from Florida Straits that cover MIS 5 at high temporal resolution. We will demonstrate that a tight but complex coupling prevailed between northern hemisphere atmospheric changes and hydrographic variations in Florida Straits.
2. Material and Methods
2.1. Core Selection and Sampling
 Quaternary sediments in the southern Florida Straits comprise foraminiferal oozes on the slope off Florida Keys, calcareous clays with intercalations of carbonate sand turbidites in the trough center, and patchy veneers of pteropod sands in the northeastern sector between Florida, Cay Sal, and Great Bahama Banks. Winnowing and redeposition is common, and sedimentation rates generally decrease from west to east [Brunner, 1986]. On R/V Sonne Cruise SO164, we discovered a mud patch between 700 and 1100 m depth on the northern slope to the northwest of Cay Sal Bank [Nürnberg et al., 2003]. This patch is fringed by carbonate sands and hardgrounds. Earlier hydrographic investigations revealed that the Florida Current is funneled in this area and occupies the entire cross section of the Strait down to 800 m water depth [Richardson et al., 1969]. Sediments from this mud patch therefore recorded the properties of surface and subsurface water masses of the Florida current and the influence of underlying intermediate water masses during the geological past.
 Piston core SO164-17-2 was recovered during R/V Sonne Cruise SO164 from this mud patch in the southern Florida Straits (24°04.99′N; 80°53.00′W; 954 m water depth; Figure 1). The core is 1172 cm long and comprises a uniform succession of carbonaceous olive gray to gray sandy and silty clay. Sedimentary textures indicating winnowing or redeposition were not recognized despite two layers of fine carbonate sand with a fining upward gradation, which were encountered at 790 and 1021 cm, far below the interval considered in this study. Bulk sediment samples, in particular half-core slices with a thickness of 1.5 cm were taken every 2 cm from the core top to 180 cm, and every 4 cm at depths below. The sample resolution was later refined to 2 cm in the MIS 5 interval down to 480 cm by using 10 cm3 cutoff syringes of 1.5 cm diameter. The samples were freeze dried and washed through a 63 μm screen. The residues were dried at 60°C, weighed, and further subdivided into different size fractions by dry sieving.
2.2. Sample Preparation
 After freeze drying, wet sieving, and fractionation, 30 specimens of Globigerinoides ruber (white variety) and Globorotalia crassaformis were picked from the 315–355 μm fraction as a pooled sample for stable isotope and Mg/Ca analysis. In the case of G. crassaformis, the size fraction for combined δ18O and Mg/Ca analysis was occasionally extended to 355–400 μm in order to achieve the required amount of carbonate for the analyses. In few cases, we also sampled the 250–315 and >400 μm fraction, but because of the size dependence of the Mg/Ca paleothermometer [Elderfield et al., 2002], we used G. crassaformis specimens from these fractions only for stable isotope analyses. Shells were gently crushed between two glass plates to open the chambers, mixed with a brush in order to achieve a homogenized sample and then divided into subsamples constituting a third of the total amount for stable isotope analyses and two thirds for Mg/Ca measurements.
2.3. Stable Isotope Analyses and Radiocarbon Dating
 Subsamples for stable isotope analyses of planktic foraminifera were rinsed three times with ultrapure water and methanol. For the benthic isotope record 1–3 tests of Cibicidoides wuellerstorfi were selected from the 250–400 μm size fraction. In addition, 20 specimens of Globigerinoides sacculifer were picked from the 315–400 μm fraction to establish an initial oxygen isotope stratigraphy for the whole core. Stable isotope samples were run on Finnigan MAT 252 and Thermo Fisher Scientific 253 Mass Spectrometers with automated carbonate preparation devices. Analytical long-term precision (n = >1000 analyses) based on an internal standard referenced to the NBS 19 standard was <0.06‰ for δ18O and <0.03‰ for δ13C. Replicate measurements yielded an external reproducibility of δ18O values of ±0.10‰ (n = 5) for G. ruber (w) and ±0.20‰ (n = 18) for G. crassaformis (1σ values).
 For 14C dating, 695 to 800 tests of G. ruber (w) were picked from the >250 μm size fractions of 9 samples between 4 and 40 cm core depth. Radiocarbon ages were determined via accelerator mass spectrometry (AMS) at the Leibniz-Labor of Kiel University. The precision of the 14C ages ranged from ±25 to +360/−370 years (1σ). Carbon-14 ages younger than 24 ka BP were transferred to calendar years by using the Calib 5.02 program with MARINE04 calibration curve. For older datings, we used the CalPal Online calendar year correction program (CalPal2007_HULU calibration curve). An ocean reservoir correction of 400 yr was applied to all calibrated ages.
2.4. Foraminiferal Mg/Ca Cleaning and Analyses
 For Mg/Ca analysis the crushed subsamples were treated following the cleaning protocol of Barker et al.  without a reductive cleaning step and analyzed using a simultaneous, radially viewing ICP-OES (Ciros CCD SOP, Spectro A.I., Germany) with a cooled cyclonic spraychamber in combination with a microconcentric nebulizer (200 μL/min sample uptake) optimized for best analytical precision and minimized uptake of sample solution. Sample introduction was performed via an autosampler (Spectra A.I.). For Ca, we used the 307.603 nm spectral line, while for Mg the 279.553 nm line was chosen. Matrix effects caused by varying concentrations of Ca were found to be insignificant as long as Ca concentrations are within the range of 20–80 ppm. To correct for machine offsets, we normalized the data relative to the ECRM 752–1 standard applying a Mg/Ca ratio of 3.762 mmol/mol according to Greaves et al.  and corrected for analytical drift using the ECRM 752–1 as an internal consistency standard every 10 samples. Due to the small volume of our sample solutions, we were able to run only a limited amount of duplicate analyses yielding a standard deviation for Mg/Ca of 0.17 mmol/mol (G. ruber (w), n = 12, equals ±0.4°C) and 0.11 mmol/mol (G. crassaformis, n = 23, equals ±0.5°C), respectively.
2.5. Temperature Calculation
2.5.1. Sea Surface Temperatures
 Calculation of sea surface temperatures (SST) from Mg/CaG.ruber was done following the multispecies equation provided by Anand et al. (1), which is based on a sediment trap study in the Sargasso Sea in the western subtropical Atlantic:
We chose this equation as it is based on the same cleaning protocol and has a lower error than the single-species calibration of Anand et al. .
 The average Holocene SSTMg/Ca values of 29.1 ± 0.4°C (see Data Set S1 in the auxiliary material) are warmer than present annual mean SSTs (26.5–27°C [Locarnini et al., 2006]) but are in the range of today's summer temperatures [Locarnini et al., 2006], pointing at a summer bias in the SSTMg/Ca record. MIS 5.5 SSTMg/Ca are 2 to 3°C higher than Holocene SSTMg/Ca (Figure 2d), which is in line with other records from the Gulf of Mexico (MD02–2575 Nürnberg et al. ) and Caribbean (ODP 999 Schmidt et al. ). It should be noted that MIS 5.5 temperatures as high as 32.5°C are on the upper limit of the applied calibration, introducing an additional uncertainty for SSTMg/Ca reconstructions.
2.5.2. Intermediate Water Depth Temperatures
 Intermediate water depth temperatures (IWTMg/Ca) were calculated using the multispecies calibration equation for deep-dwelling foraminifera by Regenberg et al. (2) based on core tops from the tropical Atlantic and Caribbean:
The Holocene IWTMg/Ca was on average 11.3±1.3°C (see Data Set S1 in the auxiliary material), which agrees with the proposed calcification depth of G. crassaformis at ∼400 m [Steph et al., 2009] (Figure 1b). The single-species equation for G. crassaformis by Anand et al.  as well as the multispecies equation by the same authors yield consistently higher temperatures (Holocene mean = 20.2, and 18.7°C, respectively). The equation by Regenberg et al.  is better constrained for deep-dwelling foraminifera, especially G. crassaformis, due to the higher number of samples included (45 samples of G. crassaformis in the work of Regenberg et al.  compared to 17 in the work of Anand et al. ) and the wider range of Mg/Ca ratios covered (1.4–2.6 mmol/mol in the work of Regenberg et al. , which is extended to 3.3 mmol/mol when including G. truncatulinoides, compared to 1.3–2.1 mmol/mol for G. crassaformis in the work of Anand et al. ).
2.6. Calculation of δ18Oivf-sw
 Sea surface and intermediate water depth salinities can be approximated by computing δ18Oivf-sw (ice volume-free seawater). For better comparison with published records [Nürnberg et al., 2008; Schmidt et al., 2004] we used the procedure published by Schmidt et al. , where δ18Oivf-sw is derived from the combination of Mg/Ca temperatures and δ18O measurements applying the δ18O temperature relation of Thunell et al.  and a correction for the global ice-volume signal following Waelbroeck et al. . Although the present-day linear relationship between salinity and δ18Owater is well constraint for the Florida Straits [e.g., Lynch-Stieglitz et al., 1999b] we restrain from calculating salinities for MIS 5 with this regression. The δ18O salinity relation is likely to have changed through time due to different hydrological conditions altering the end-member characteristics of mixing water masses. As discussed later, we expect shifts in (sub)tropical wind fields and concomitant variations in the evaporation/precipitation ratio over the subtropical gyre during MIS 5 that should have caused a different δ18O salinity relationship for the surface ocean over time. At intermediate depths, the relative contribution of different water masses (i.e., the varying influence of AAIW and SUW) is hard to assess, but most likely will have affected the middle-depths δ18O-salinity relationship.
 Error propagation based on the uncertainty of the paleotemperature and δ18Ocalcite to δ18Oseawater conversion equations and the reproducibility of the δ18O and Mg/Ca measurements results in cumulative uncertainties of ±0.26‰ for the δ18Oseawater of G. ruber (w) and ±0.35‰ for G. crassaformis. The amount of the additional error attributed to the ice volume correction with reference to age model uncertainties and the error of the ice volume estimates itself is difficult to assess. However, during the interval between MIS 5.1–5.4, which is considered in detail in this study, the amplitude of the δ18O variations caused by ice volume changes were ∼0.15‰ according to Waelbroeck et al. , which is very low compared to the observed δ18O anomalies in the G. crassaformis record.
 Please note that the data obtained on SO164-17-2 is presented in the auxiliary material (Data Sets S1–S4) and will be made available in the online database PANGAEA (www.pangaea.de).
3. Results and Discussion
 The stratigraphic reach of core SO164-17-2 was constrained with calcareous nannofossils and Globorotalia menardii biozones. A smear slide from the core base showed small Gephyrocapsa and Emiliania huxleyi. Pseudoemiliania lacunosa was not recorded, which infers the nannoplankton zone NN21 with a maximum age of 270 ka. The boundaries of Ericson zones Z, Y, X, W, and V [Ericson and Wollin, 1968] were determined by screening of the size fractions >250 μm for the first and last occurrences of Globorotalia menardii (Figure 3a and Data Set S3). The V/W boundary at 528 cm indicates that the record of the upper part of the core is complete and goes back beyond 165 ka [Brunner, 1984; Peterson et al., 2000]. Ambiguous abundance fluctuations of G. menardii in the lower part of the core inhibit the recognition of older Ericson zone boundaries. A correlation of the planktonic oxygen isotope curve of Globigerinoides sacculifer (Figure 3b and Data Set S4) with records from ODP Site 1002 [Haug et al., 2001; Peterson et al., 2000] and stacked standard records [Martinson et al., 1987] revealed that core SO164-17-2 comprises MIS 1 through 8 and goes back to 271.7 ka.
 The upper section of the core comprising the Holocene and last deglaciation was condensed, i.e., 18.8 kyrs were represented by 32 cm only. The following 2 cm comprise the entire Last Glacial Maximum as constrained by bracketing AMS 14C datings (see Data Set S5 in the auxiliary material). No sharp or oblique contact was visible at this interval which would indicate an unconformity despite an indistinct color change from light gray to light cream. The top of the core is well preserved. If at all, only a few centimeters were missing as indicated by a dark seam of manganese oxides at 2 cm in core SO164-17-2, which was found between 5 and 6 cm sediment depth in box core SO164-17-1 from the same location. Even though the Holocene is apparently complete and well constrained in core SO164-17-2, its condensed succession limits the comparability with the record of MIS 5.
 MIS 5 ranges from 160 to 440 cm. A detailed age model was developed for this interval by tuning the benthic isotope record of Cibicidoides wuellerstorfi to the global benthic isotope stack LR04 [Lisiecki and Raymo, 2005] (Figures 3c and 3d). Sedimentation rates vary between 2.7 and 16 cm/kyr with an average of 5.2 cm/kyrs during MIS 5, with the highest sedimentation rate between 123 and 125 ka at the peak warm stage MIS 5.5. Given that carbonate contents (78.5–87 wt% CaCO3) are relatively stable throughout the record it is likely that the pronounced differences in sedimentation rates are not primarily the effect of productivity changes but point at the highly dynamic nature of the deep currents at Florida Streets affecting sediment accumulation.
3.2. Assessment of Contamination and Diagenetic Effects on the Mg/Ca Ratios
 To identify contamination by clay particles and Mn-Fe carbonate and oxide coatings, which might affect the foraminiferal Mg/Ca ratios [Barker et al., 2003], aluminum, iron and manganese concentrations were monitored. Fe/Ca ratios were commonly below 0.1 mmol/mol and therefore indicating that no contamination by detrital material or Fe-rich overgrowth had taken place [Barker et al., 2003]. Mn/Ca ratios were mostly higher than the ratio of 0.1 mmol/mol given by Barker et al.  (mean: 0.32 ± 0.09 mmol/mol), but Mn concentrations were an order of magnitude smaller than those of Mg. There is no correlation (r2 < 0.01) between Mn/Ca and Mg/Ca for G. ruber (w), which infers that their Mg/Ca ratios are not biased by Fe/Mn coatings. However, a significant correlation of r2 = 0.54 is recognized for G. crassaformis (Figure 4), even though the Mn/Ca ratios are on average lower than those of G. ruber (w). Calculated IWTMg/Ca for MIS 5 were on average ∼14.0°C, thus considerably warmer than the Holocene values (Figure 2f, see also Data Set S1). Mean Mn/Ca values of MIS 5 (0.25 mmol/mol) are higher by 0.15 mmol/mol than during the Holocene, thus, diagenetic overprinting [see Barker et al., 2003; Pena et al., 2005] by Mn-bearing minerals might have caused a bias toward higher temperatures by contributing additional Mg as a contaminant. Mn-Mg carbonates (kutnahorite) overgrowths as a common source for diagenetic contamination [Barker et al., 2003] usually have Mg/Mn ratios of ca. 0.1 mol/mol [Barker et al., 2003; Pena et al., 2005]. As such, a 0.15 mmol/mol increase in Mn/Ca would contribute to a Mg/Ca increase of only ∼0.015 mmol/mol, which is equivalent to a temperature increase by only ∼0.07°C. Even the assumption of an unrealistic Mg:Mn ratio of 1:1 in the diagenetic overgrowths would not affect the overall shape of the Mg/Ca curve (presented as Mg/Ca* in Figure 4; Mg/Ca ratios corrected for Mn/Ca). We thus conclude that diagenetic overprinting is not the primary reason for the MIS 1 versus MIS 5 temperature difference and that MIS5 subsurface temperatures were indeed considerably higher than during the Holocene.
 Al/Ca ratios of >0.1 mmol/mol are commonly used to detect samples contaminated by clay particles that were not sufficiently removed during the cleaning process [e.g., Barker et al., 2003]. Since Al concentrations are in most samples close to the detection limit of the ICP-OES, we chose to discard samples with Al/Ca ratios >0.15 mmol/mol, because a sufficiently reliable reproducibility (RSD <10%) could only be achieved for samples with correspondingly high Al contents.
 Mg/Ca-based temperature reconstructions may yield unrealistic high estimates in case the foraminiferal tests were overgrown by microcrystalline Ca carbonates due to carbonate dissolution and reprecipitation during early diagenesis. Regenberg et al.  particularly noted unusual high Mg/Ca ratios >6 mmol/mol in planktic foraminifera (incl. G. ruber) in some cores from the Caribbean for MIS 5 and earlier stages. The elevated Mg/Ca ratios were found to go along with reduced Sr/Ca (<1.3 mml/mol) and high δ18O values, which were both attributed to the chemical composition of the overgrowths. In our record from core SO164-17-2, we note that Sr/Ca ratios of G. ruber (w) and G. crassaformis are varying in two different ways (Figure 4): G. ruber (w) has “normal” Sr/Ca ratios between ∼1.4 and ∼1.55 mmol/mol also during intervals of Mg/Ca >6 mmol/mol with no similarity to the Mg/Ca record. G. crassaformis, on the other hand, has covarying Sr/Ca and Mg/Ca ratios, with Sr/Ca as low as 1.2 mmol/mol but without the inverse relation reported by Regenberg et al. . In fact, a high temperature sensitivity of Sr/Ca ratios of deep dwelling Globorotalia species has been found in different studies [Cléroux et al., 2008; Elderfield et al., 2000; Mortyn et al., 2005], which would explain the similarity of the Sr/Ca and Mg/Ca records of G. crassaformis. Application of the Sr/Ca temperature equation for Globorotalia spp. put forward by Cléroux et al.  [TSr/Ca = (Sr/Ca-1.052(±0.001))/0.0217(±0.021)] (we refer to equation (2) in the main text of Cléroux et al. , not the formula given in the abstract (C. Cléroux, personal communication, 2010)) yields temperature estimates not significantly different from that of the Mg/Ca-based calibration after Regenberg et al. . However, we will further use TMg/Ca since G. crassaformis was not included in the Sr/Ca temperature calibration data set of Cléroux et al. , and caution has to be taken with the application of this temperature equation prior to the Holocene since the foraminiferal Sr/Ca ratio is prone to influences by other oceanographic parameters such as Sr/Ca changes of the seawater and carbonate saturation [Cléroux et al., 2008; Elderfield et al., 2000; Mortyn et al., 2005].
 The dissimilarity between Sr/Ca variations for G. ruber and G. crassaformis indicates furthermore that diagenetic overgrowth by microcrystalline Ca carbonates apparently does not bias our results as it should affect both species likewise.
 A core top study from the Caribbean indicated that dissolution affecting Mg/Ca ratios in water depths only below ∼2700 m [Regenberg et al., 2006]. As such, we consider the Mg/Ca ratios to be largely unaffected by dissolution due to the relatively shallow water depth (954 m) and the good aragonite preservation throughout core SO164-17-2.
3.3. Temperature and Salinity Variations
 The SO164-17-2 SSTMg/Ca largely follows the summer insolation at 30°N (Figures 2d and 2e), suggesting that climate variability in Florida Straits is predominately controlled by summer insolation and the expansion of the Atlantic Warm Pool in relation to the summer position of the Intertropical Convergence Zone (ITCZ) [Ziegler et al., 2008]. SSTMg/Ca of SO164-17-2 during MIS 1 (not shown) and MIS 5 is generally 1–3°C higher than those from the Gulf of Mexico and the Caribbean (Figure 5; note that the ODP 999 SSTMg/Ca record was adjusted to account for the different cleaning procedures used and the effects of calcite dissolution as described by Nürnberg et al.  and Regenberg et al. ). The offset relative to the ODP 999 temperatures can be explained considering the inferred summer bias of SSTMg/Ca in Florida Straits: Mean Holocene temperatures in the Caribbean are ∼27.7°C [Locarnini et al., 2006; Schmidt et al., 2004] while the mean summer SST in the Florida Straits is in the range of 29–30°C [Locarnini et al., 2006]. The difference of SSTMg/Ca from SO164-17-2 to the also summer-biased SSTMg/Ca from MD02–2575 [Ziegler et al., 2008] reflects the influence of cool and fresh water derived from the Mississippi and other rivers discharging into the northern Gulf of Mexico. These waters lead to a cooling at water depths between ∼20–100 m along the shelf and slope of the northern and eastern Gulf of Mexico [Locarnini et al., 2006]. Integrated for the surface mixed layer, this results in a difference of ∼2°C between the sites MD02–2575 and SO164-17-2, which in the range of the offset between the calculated SSTMg/Ca.
 The SSTMg/Ca increase with the onset of MIS 5.5 occurs very abruptly, with an almost 3°C step toward a short-lived peak with SSTMg/Ca of 32.5°C at 124 ka. The abrupt warming after the end of Termination II is in contrast to the more gradual SST increase observed elsewhere in the Caribbean [Schmidt et al., 2004] and Gulf of Mexico [Nürnberg et al., 2008] (Figure 5). It is unlikely, however, that the unusual abrupt warming in our core is caused by reworking since the benthic foraminiferal δ18O record does not display any irregular feature that deviates from other benthic oxygen isotope curves (compare Figure 3d). In fact, the rapid SSTMg/Ca warming in SO164-17-2 has a striking similarity with the contemporaneous occurrence of G. ruber (w) as an exotic species in core M23414 from the North Atlantic (Figure 2c). It might be suspected that Florida Current and ensuing Gulf Stream strength was particularly enhanced during early MIS 5.5 and facilitated a northward transport of warm and saline water masses entraining G. ruber (w). Interestingly, a comparable increase in G. ruber (w) abundance in the North Atlantic was not observed during the early Holocene [Bauch and Kandiano, 2007], probably as a result of the climatically more favorable conditions during MIS 5.5.
 The δ18Oivf-sw of surface water, approximating salinity, is relatively constant within error bars during most of MIS 5 and ∼1.8‰ (SMOW) lower, indicating less saline conditions than during full glacial MIS 4 and 6 (Figure 2b). After 85 ka, well-expressed fluctuations in the surface and intermediate water δ18Oivf-sw occur, resembling the pattern of Dansgaard/Oeschger-like rapid climate oscillations (Figure 2). However, the sample resolution is partly too low to identify all D/O interstadials and stadials. Compared to other δ18Oivf-sw reconstructions from MIS 5, Florida Straits values are similar to those from the Gulf of Mexico site MD02–2575 and slightly higher by ∼0.5‰ during MIS 5.4 as compared to ODP 999 from the southern Caribbean (Figure 5). Although slightly lower annual mean salinities at ODP 999 than summer salinities in Florida Straits are in accord with the present-day situation [Antonov et al., 2006], a 0.5‰ offset in δ18Oivf-sw is within the error bars of the δ18Oivf-sw calculations.
 Intermediate water δ18Oivf-sw of G. crassaformis is generally higher by 1.5 to 2‰ during glacial stages MIS 4 and 6 than during most of MIS 5, similar to the surface conditions. However, there are two intervals of markedly elevated intermediate water δ18Oivf-sw during MIS 5: a short excursion at the end of MIS 5.4 at ∼107 ka, and a more prominent event starting with a gradual increase of δ18Oivf-sw at ∼94 ka with maximum values throughout MIS 5.2 (from ∼85–95 ka). These periods of elevated intermediate water δ18Oivf-sw are accompanied by high-amplitude IWTMg/Ca oscillations at ∼107 ka and lowered IWTMg/Ca during MIS 5.2. IWTMg/Ca parallels the SSTMg/Ca increase at the beginning of MIS 5.5 but lack the short-lived maximum that is evident in the SSTMg/Ca. Notably, the SSTMg/Ca cooling during MIS 5.4 does not affect IWTMg/Ca. The anomalously high δ18Oivf-sw values of G. crassaformis during MIS 5.2 and at ∼107 ka are also expressed as an increased δ18O gradient between surface and deep dwellers (depicted as Δδ18O in Figure 3f), which is almost as high as during the full glacial intervals of MIS 4 and 6. Following the approach of Lynch-Stieglitz et al. [1999b], an enhanced vertical δ18O gradient represents a more pronounced density gradient between surface and deep water masses, potentially the result of less steeply tilted pycnoclines at the core location (cf. Figure 1d). A decreased pycnocline tilt is then indicative of a reduced vertical shear in the Florida Current and, thus, reduced Gulf Stream strength. A lowered Florida Current strength during cold periods is in accord with the findings of Lynch-Stieglitz et al. [1999a] for the Last Glacial Maximum. One remarkable aspect of the δ18O gradients shown in Figure 3f is that high gradients do not only occur during the cold interglacial phases MIS 5.2, 5.4, 4 and 6, but also during the climatic optimum of MIS 5.5 and the Holocene. Interpreted in terms of density gradients and isopycnal tilt this would indicate reduced Florida Current strength, which is at odds with the assumption of strong throughflow during interglacials and reduced throughflow during glacials [Lynch-Stieglitz et al., 1999a]. However, based on the temperature reconstructions, this offset between δ18O of G. ruber (w) and G. crassaformis during MIS 5.5 seems to be caused by the maximum SSTs recorded by G. ruber (w) and therefore represents primarily a surface signal not related to geostrophic flow. Hence, although the δ18O records might imply changes in the density structure of the Florida Straits, we note that based on just one core from the center of the Florida Straits solid inferences about the tilt of isopycnals and Gulf Stream strength cannot be made and remain speculative unless multispecies planktonic foraminiferal oxygen isotope and Mg/Ca records from both sides of Florida Straits will be available.
 The key issue to be discussed is the origin of the pronounced intermediate depth δ18Oivf-sw increases during MIS 5. It seems plausible that intermediate water salinity increases at ∼107 ka and during MIS 5.2 as a result of the vertical expansion of highly saline WNACW/SUW waters in Florida Straits (schematically depicted in Figure 6c). However, increasing salinities and decreasing temperatures recorded by G. crassaformis during MIS 5.2 and at ∼107 ka appear contradictorily, as enhanced vertical SUW extension and WNACW invasion into Florida Straits should have led to higher IWTMg/Ca (Figure 1b). Intermediate warming is presumably counteracted by widespread wind evaporation–driven surface cooling in the subtropical Atlantic [Chiang et al., 2008], as evidenced by alkenone-derived SSTs in the northern subtropical gyre (Bermuda Rise core MD95–2036, Figure 2g [Lehman et al., 2002]). As subduction of water masses in the subtropical gyre occurs late in winter [Bryden et al., 1996], IWTMg/Ca in the Florida Straits are generally more similar to the SSTalkenone of the gyre region (MD95–2036) and display the same 3–3.5°C amplitude of cooling during MIS 5.2. The proxy evidence is supported by modeling results, which indicate a relative cooling of subsurface water masses in the western (sub)tropical Atlantic north of 20°N as a response to reduced AMOC [Chang et al., 2008].
 The combined cooling and salinity increase recorded by G. crassaformis basically argues against a change in calcification depth of this species as the primary cause of the proxy variations. According to the current water column structure, an upward migration of G. crassaformis to waters with higher salinity would be associated with higher temperatures. Although we cannot exclude that G. crassaformis changed its habitat, ongoing studies suggest that this species is rather conservative in its preferred calcification depth at Florida Straits, also during the last deglacial [Cléroux and Lynch-Stieglitz, 2010].
 Low benthic δ13C values at ∼107 ka and during MIS 5.2 at Ceara Rise site EW9209 [Curry and Oppo, 1997] point to a sluggish AMOC during these intervals (Figure 2i). This also coincides with transient periods of continental ice shield growth [e.g., Chapman and Shackleton, 1999; Lisiecki and Raymo, 2005], as indicated by enhanced portions of ice rafted debris (IRD) in North Atlantic cores [Chapman and Shackleton, 1999; Helmke and Bauch, 2003] (Figure 2j). In case of the interval at ∼107 ka, the δ13C reduction is expressed as two short-lived intervals where the later at ∼106 ka might tentatively be linked to the intermediate δ18Oivf-sw anomaly at ∼107 ka in the Florida Straits. While this correlation is vague due to the chronostratigraphic uncertainty, the Ceara Rise δ13C reduction during MIS 5.2 clearly occurs contemporaneously to the high deep dweller δ18Oivf-sw values in SO164-17-2. The apparent link between intermediate salt accumulation in the northern subtropical gyre and reduced AMOC strength was also proposed by Wan et al.  based on coupled ocean-atmosphere model experiments. While the concept of salt and heat accumulation in the (sub)tropics during reduced overturning is not new [e.g., Carlson et al., 2008; Schmidt et al., 2004, 2006; Weldeab et al., 2006], Wan et al.  as well as other studies [Chang et al., 2008; Chiang et al., 2003; Wan et al., 2009; Zhang, 2007] point at the rather complex mechanisms that link oceanographic and climatic changes in low latitudes. In case of the western subtropical North Atlantic, lowered precipitation due to the southward shift of the ITCZ was invoked as dominant process [Chiang et al., 2008; Wan et al., 2010], while the reorganization of the oceanic circulation, in particular a reversal of the North Brazil Current, is proposed to be responsible for salt accumulation in the equatorial and southern (sub)tropical Atlantic [Chang et al., 2008; Wan et al., 2010]. Further, intensified wind stress resulting from a more pronounced meridional temperature gradient and stronger northeasterly trade winds might have caused enhanced Ekman downwelling within the subtropical gyre. Increased WNACW production, thus, appears to be a viable mechanism for the westward expansion of saline gyre waters into Florida Straits [Slowey and Curry, 1995].
3.4. A Role for AAIW in the Florida Straits During MIS 5?
 Considering the synchrony of the atmospheric and oceanographic changes observed during MIS 5.2 and 5.4 it remains speculative why the inferred vertical expansion of the gyre water in Florida Straits is much more pronounced during MIS 5.2 than in 5.4.
 In this context the waxing and waning influence of intermediate, southern source waters in the Gulf of Mexico, as for example AAIW, might be an important factor. Records covering the last deglaciation have shown that the formation and northward extent of AAIW was highly variable through time: Pahnke et al.  used ɛNd records from Tobago Basin and Brazil margin to infer that the northward penetration of AAIW was at maximum during Heinrich Event 1 and the Younger Dryas, concomitant to the inferred reduction of the AMOC, but at minimum during the Last Glacial Maximum [McManus et al., 2004]. Benthic Cd/Ca records from Florida Straits, in contrast, argue in favor of a reduced presence of southern source intermediate waters during the same periods of weak AMOC [Came et al., 2008]. The δ13C data from core SO164-17-2 obtained on the benthic foraminifer Cibicidoides wuellerstorfi might serve as an indicator for the relative contribution of southern versus northern source waters, with low δ13C values indicating a higher contribution of a nutrient-rich southern source. The total benthic δ13C amplitude of ∼0.7‰ during MIS 5 exceeds the 0.32‰ glacial/interglacial δ13C shift attributed to variations in the global carbon pool [Duplessy et al., 1988]. As profound variations in primary productivity were not recognized, water mass changes remain as a viable explanation for the observed δ13C variations. Our δ13C record (Figure 2h) shows the lowest values from Termination II to ∼107 ka, which is indicative of the presence of intermediate waters with a high contribution of AAIW-type water in Florida Straits during warmest parts of MIS 5. In turn, the higher δ13C values prevailing during later stages of MIS 5 would suggest a reduced influence of AAIW. Assuming that diluted AAIW, currently underlying the SUW, was absent after ∼107 ka, this would have enabled the SUW to expand to greater depths in Florida Straits (as in MIS 5.2; Figure 6c) during periods of enhanced gyre water production. Instead, at times of larger AAIW contribution, SUW expansion would have been impeded (Figures 6a and 6b). The rapid high-amplitude variations of the IWTMg/Ca record during MIS 5.2 may therefore be interpreted in terms of a period of hydrographic instability in Florida Straits (Figure 6b). Based on the available data, the reason for reduced AAIW contribution after ∼107 ka remains speculative. Possible mechanisms involve either a generally reduced AAIW production in the source area [Pahnke and Zahn, 2005] or a higher amount of subtropical gyre water entering the Caribbean and blocking AAIW intrusion.
 While the above discussions center around the impact of high-latitude climatic and oceanographic forcings on the low-latitude hydrography, the apparent expansion of saline water masses in the subtropics might have favored a positive feedback on the invigoration of the thermohaline circulation itself. Previous studies on rapid climatic rebounds, such as the Heinrich and Younger Dryas cooling events, already pinpointed the potential role of (sub)tropical heat and salt accumulation for the resumption of high-latitude deep water formation during periods of a reduced AMOC [Rühlemann et al., 1999; Schmidt et al., 2004, 2006]. Further modeling and proxy studies may show whether the accumulation of a vast amount of highly saline and slightly cooled water in the subsurface of the subtropics have played a similar role in limiting the magnitude of cold events during interglacial MIS 5 during phases of reduced northward heat advection.
 Combined foraminiferal Mg/Ca and δ18O data from Florida Straits core SO164-17-2 document abrupt increases in intermediate water salinity at times of transient cooling events during interglacial MIS 5, while surface salinities remained relatively constant. Based on further evidence from benthic δ13C measurements, we conclude that the intermediate depth salinity anomalies during MIS 5.2 and at ∼107 ka probably resulted from independent atmospheric and oceanic forcings, i.e., wind-driven intensification of the subtropical gyre circulation as well as a reduced northward expansion of AAIW. Variations in the δ18O gradients between deep and shallow dwelling foraminifera indicate that the density structure and in consequence the isopycnal tilt within Florida Straits might have changed during these intervals. The proposed subsurface salinity and temperature variations are broadly in line with paleoceanographic modeling studies and emphasize the sensitivity of the subtropical hydrography to relatively small variations in high- and low-latitude forcings.
 We kindly acknowledge the support and cooperation of captain and crew during R/V Sonne Cruise 164 (RASTA). Lulzim Haxhiaj (stable isotope laboratory, IFM-GEOMAR), Karin Kiessling (ICP-OES Laboratory, Kiel University), and Nadine Gehre provided valuable laboratory assistance. The contributions of Hanno Kinkel (Kiel), Jenny Lezius (Bremerhaven), and Jörg Steinlöchner (Stuttgart) to establish the stratigraphy of core SO164-17-2 are gratefully acknowledged. We thank Cyrus Karas, Carsten Eden, and Sascha Flögel for helpful discussions during the progress of this study as well as Editor Chris Charles and two anonymous reviewers for their constructive comments which significantly improved the manuscript. This work has been financed by the German Research Foundation (DFG) in the framework of the German Research Priority Program SPP1266 (INTERDYNAMIK, Project LOOP, Nu 60/17–1).