The various cleaning steps required for preparation of foraminiferal samples for Mg/Ca (and Sr/Ca) analysis are evaluated for their relative importance and effects on measured elemental ratios. It is shown that the removal of silicate contamination is the most important step for the measurement of Mg/Ca ratios. In an example, bulk sample Mg/Ca decreases from 10.5 to 2.5 mmol mol−1 during clay removal. Oxidation of organic material causes a lowering of sample Mg/Ca in the order of 10% or approximately 1°C when converted to temperature. Use of dilute acid leaching to remove adsorbed contaminants causes partial dissolution of the sample carbonate and a corresponding decrease in Mg/Ca. Reductive treatment also causes dissolution of the sample and a decrease in the Mg/Ca ratio of up to 10–15%. Sample preparation for Sr/Ca analysis does not require the same degree of rigor as is necessary for Mg/Ca work. The “within-run” reproducibility of the method described here for G. ruber in a core-top sample from the Arabian Sea was ±1.8% (mean sample ratio was 4.72 mmol mol−1). When converted to temperature, this becomes 28 ± 0.2°C. The equivalent result for Sr/Ca was ±0.5% (mean ratio = 1.44 mmol mol−1).
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Foraminiferal Mg/Ca ratios are now routinely analysed because of their value as indicators of past ocean temperature [e.g., Lea et al., 2000; Elderfield and Ganssen, 2000; Rosenthal et al., 2000; Lea et al., 2002]. Methods used to clean foraminiferal samples prior to analysis have mostly evolved from the Cd/Ca and Ba/Ca cleaning methods developed by Boyle , Boyle and Keigwin  and Lea and Boyle . Such methods involve a number of discrete sequential steps aimed at removing various contaminating phases: (1) removal of clay materials, (2) removal of organic matter, (3) removal of Mn-Fe-oxide coatings, and, if needed, (4) removal of barite, with (5) final “polishing” of the sample prior to analysis. However, there is some debate as to whether the removal of all the phases mentioned above is necessary when analysing for Mg/Ca ratios [e.g., Brown and Elderfield, 1996; Martin and Lea, 2002]. Consequently, the cleaning methods used differ between laboratories. The aim of this study was to investigate which cleaning steps are necessary and most important for the preparation of foraminiferal samples for Mg/Ca analysis.
 In this paper, we summarize results of a series of experiments that were made to provide a step by step evaluation of the various cleaning steps involved in sample preparation for Mg/Ca analysis. Because Sr/Ca ratios are obtained in our laboratory as a routine bi-product of Mg/Ca analysis, and because of their potential value as paleoceanographic indicators [e.g., Stoll et al., 1999], we have included Sr/Ca in our study. The behaviour of Sr/Ca during the cleaning methodology provides a useful contrast to that of Mg/Ca.
2. Approach and Methodology
 Our motivation for performing a detailed investigation into cleaning procedures was two fold. Firstly, we discovered that the Mg cleaning procedure routinely used in our laboratory yielded measured Mg/Ca ratios that were significantly higher than predicted in certain cases. It was noted that high values of Mg/Ca were associated with high levels of Fe and Al. This implied that silicate contamination was present in the samples even after cleaning [Barker and Elderfield, 2001]. Secondly, as will also be discussed in more detail in section 3, a preliminary test of two variants of the Mg/Ca cleaning procedure [Boyle and Keigwin, 1985; Elderfield and Ganssen, 2000] revealed a systematic offset in the measured Mg/Ca ratios. Samples cleaned by the Cd/Ca cleaning method of Boyle and Keigwin  gave almost consistently lower values of Mg/Ca than those cleaned by the Elderfield and Ganssen  method.
 Experiments were carried out on foraminiferal samples which were hand picked from pre-sieved and washed sediment samples. Before cleaning, foraminiferal tests were gently crushed between two glass plates. The aim here is to crack open the test chambers just enough to allow debris to escape while avoiding over-crushing, as this will lead to excessive sample loss during cleaning. The various cleaning steps are based on the method described in Boyle and Keigwin  with modifications which we discuss. First, we present results which we have selected to exemplify the two main issues we consider in this paper (the issue of silicate contamination and that of the Mn-Fe-oxide coating). These use what we term “an inefficient clay removal technique” which, nevertheless, is perfectly adequate for Mg/Ca work in the correct circumstances but is inadequate if a single methodology is used for all samples. Next, a number of individual monitoring experiments are described which use a refinement of the cleaning procedure (full details are given in Appendix 1) and demonstrate its effectiveness. Finally, we present an overview of the method and its reproducibility.
 Cleaning reagents were made from Aristar concentrates using Elga UHQ H2O for dilution (Appendix 2). Leaching and dissolution acids were made from Quartz Distilled (QD) concentrated HNO3 and QD H2O. Analyses were performed on a Varian Vista ICP-AES. Precision and Accuracy for Mg/Ca and Sr/Ca ratios, determined from replicate runs of a standard solution containing Mg/Ca = 5.13 mmol mol−1 and Sr/Ca = 2.09 mmol mol−1 at a Ca concentration of 60 ppm, are <0.5% [de Villiers et al., 2002].
3. Potential Problems and Preliminary Experiments
3.1. The Role of Silicate Minerals
 During his early work on the mineralogical and chemical composition of planktonic foraminiferal carbonate, Emiliani  demonstrated that measured foraminiferal Mg content varied linearly with Al (Figure 1a) and suggested that Mg was probably not, therefore, associated with the calcite lattice but must instead be related to sedimentary (i.e., silicate) contamination. Further, due to the repeated occurrence of Al and Si during the analysis of foraminiferal carbonate, he reported that “it is impossible to wash specimens absolutely clean in preparation for analysis”. The presence of Mg within the lattice of foraminiferal calcite is now undisputed. The importance of silicate removal was recognized by Boyle  for the analysis of Cd/Ca ratios. However, little investigation has been made into its significance for Mg/Ca or Sr/Ca measurements.
 Clay minerals represent a significant proportion of marine sediment, the four most abundant species being illite, kaolinite, montmorillonite and chlorite [Riley and Chester, 1971]. On average, these clays contain between about 1–10% Mg by weight (though insignificant amounts of Sr) [Deer et al., 1992]. An average planktonic foraminifer may weigh 15 μg and have a Mg/Ca ratio of 3 mmol mol−1 (i.e., it will contain 1 × 10−8 g of Mg). An increase of 10% in the measured ratio would then require 1 × 10−9 g of Mg. If a clay comprised just 1% Mg by weight, then 0.1 μg would be sufficient to cause significant uncertainty in the measured foraminiferal Mg/Ca ratio.
 The effect of silicate contamination on Mg/Ca and Sr/Ca ratios will vary significantly between samples and locations. For example foraminiferal samples from cold regions and correspondingly low (<2 mmol mol−1) Mg/Ca ratios will be more prone to significant contamination than those from warmer locations and with higher Mg contents. A sample taken from sediment with a particularly high clay content will require more efficient clay removal than one from a clay poor sediment. Similarly, samples containing ice rafted debris (IRD) or similar will require special attention to remove these particles which may cause significant contamination to measured Mg/Ca ratios (section 4).
 The potential for silicate contamination is also sensitive to the gross morphological structure of the foraminiferal species used. For example, the thin walled, open chambered tests of Globigerina bulloides are more prone to contamination than the thicker shelled benthic species, Uvigerina perigrina (Figure 1b). Plotting Mg/Ca against Al/Ca (or Fe/Ca) allows approximate determination of the contaminant composition. Offsets in Mg/Ca between species reflect differences in habitat temperature during calcification.
 Contamination of measured Mg/Ca ratios by silicate material will also be highlighted by a covariance between Mg/Ca and Fe/Ca or Al/Ca data in downcore records (Figure 2a). An estimate of the error associated with the silicate contamination is provided by the ratio of Fe/Mg or Al/Mg of the contaminant phase and the measured sample Fe/Mg or Al/Mg ratio. For example if the Fe/Mg ratio of silicate contamination is 1 mol mol−1, a measured sample Fe/Mg ratio of 0.01 mol mol−1 would show that the corresponding foraminiferal Mg/Ca ratio is 1% too high. The absolute error in the measured Mg/Ca ratio of a sample is given by:
where Fe may be substituted for by Al. In principle, this method could be used to provide a “silicate-corrected” Mg/Ca ratio by subtracting the calculated uncertainty from the measured ratio (Figure 2b). However, due to the uncertainty in estimating the contaminant composition, and the need to analyse additional elements, it is clear that data corrected in this way cannot be quoted with significant confidence. The example shown in Figure 1b and 2a serves to emphasise the consequence of inadequate clay removal. Using the cleaning procedure outlined in Appendix 1, typical values of Fe/Ca and Fe/Mg are <0.1 mmol mol−1 and <0.03 mol mol−1 respectively. The presence of Al is generally below detection for determination by ICP-AES. We reject measured sample Fe/Mg ratios of >0.1 mol mol−1 as potentially significantly contaminated by silicate phases.
3.2. The Role of Mn-Fe-Oxide Coatings
 Mn-oxide coatings (the term “Mn-oxide” is used interchangeably with “Mn-Fe-oxide” due to generally high Mn/Fe ratios observed in foraminiferal samples) are removed by reductive treatment using a buffered solution of hydrous hydrazine [Boyle and Keigwin, 1985]. It is clear from the work of Boyle , Boyle and Keigwin  and Lea and Boyle  that removal of such authigenic coatings is a vital part of the cleaning procedure for Cd/Ca and Ba/Ca determination. Foraminiferal shells from sediment cores commonly possess a Mn-rich oxide coating formed if Mn2+ has been mobilized during anoxic breakdown of organic matter deeper in the sediment column. A further coating of Mn-rich carbonate, formed below the manganese redox boundary, may also be present [Boyle, 1983].
 Analyses of manganese nodules and micronodules (common forms of Mn-Fe-oxide) consistently give Mg/Mn ratios of about 0.1 mol mol−1 [de Lange et al., 1992; Pattan, 1993]. A study of Mn-carbonate in marine sediments of the Panama Basin by Pederson and Price  also revealed Mg/Mn ratios of about 0.1 mol mol−1. Mn/Ca ratios for cleaned foraminiferal samples (without any attempt at removal of Mn phases) are typically around 102 μmol mol−1. This would suggest that the contribution of Mg from coatings to the corresponding Mg/Ca ratio of a typical sample is about 1%. Nevertheless, analysis of replicate samples through two variants of the Mg/Ca cleaning procedure: the method of Boyle and Keigwin , which includes a reductive cleaning stage, and the method of Elderfield and Ganssen , which does not, revealed a much larger systematic offset in the measured Mg/Ca ratios (Figure 3). Samples cleaned by the Cd/Ca cleaning method of Boyle and Keigwin  gave almost consistently lower values of Mg/Ca (by an average ∼15%) than those cleaned by the Elderfield and Ganssen  method.
 The primary difference between these two methods is that ‘Cd cleaning’ involves a reductive step aimed at removing the Mn-Fe-oxide coating. The reducing reagent (hydrous hydrazine in a citric acid/ammonia buffer) is corrosive to carbonate and causes partial dissolution of the sample; this is further exacerbated if the modification used for Ba/Ca is employed [Lea and Boyle . The effect of partial dissolution on foraminiferal calcite is known to lower Mg/Ca ratios [Brown and Elderfield, 1996; Rosenthal et al., 2000]. The question then arises; are the higher Mg/Ca ratios observed for samples cleaned by the Elderfield and Ganssen  method due to the presence of Mg in the Mn-oxide coating, or are the lower Mg/Ca ratios observed for reductively cleaned samples due to partial dissolution during treatment?
4. Experiments to Monitor Clay Removal
 The effect of clay removal on sample Mg/Ca and Sr/Ca ratios may be demonstrated by analysing the solutions removed during the clay removal stage of the cleaning process (Figure 4). In this experiment, a sample comprising 20 individuals of G. bulloides from the 122 cm interval in core NEAP-8K from the North Atlantic (59°48′N 23°54′W, water depth 2360 m [McCave, 1984]) was used. This sample has been shown to be particularly prone to silicate contamination since it contains significant amounts of IRD [Barker and Elderfield, 2001]. In this and in subsequent experiments the method described in Appendix 1 was used. An important feature of the clay removal protocol employed here is a “minimal settling” technique whereby the supernatant containing suspended clays and fine silicates is removed almost immediately from the sample after ultrasonification and agitation (Appendix 1- clay removal step 6). To demonstrate the benefit of minimal settling, the supernatants were removed and allowed to settle again before being separated into a further supernatant phase (termed “fines”) and a residual, or “coarse” fraction. Both portions were then acidified and analysed. The high Mg/Ca and Fe/Ca ratios of the coarse fraction solutions suggests that a considerable amount of silicate material settles rapidly (Figures 4b and 4c).
 The cumulative effect of clay removal on sample Mg/Ca is considerable; a drop from about 10.5 to 2.5 mmol mol−1 (Figure 4e). The progress of clay removal can be monitored from the change in slope of evolving Fe/Mg (Figure 4e). As the gradient approaches zero, this particular stage of cleaning has reached its full potential. Sample Fe/Mg ratios greater than about 0.01 mol mol−1 suggest that contamination may be a problem because contaminant Fe/Mg ratios are typically in the region of 1 mol mol−1. The Fe/Mg ratio in this example is about 0.2 mol mol−1 even after complete clay removal. This is because Fe may also be present in larger silicate grains which are not removed during clay removal (see below). Additionally, Fe may be present in an oxide coating and thus Fe/Mg(sample) > 0.01 mol mol−1 does not necessarily mean incomplete clay removal; it is primarily an indicator. In this respect Al/Mg would be a better indicator of clay but Fe/Mg is used in this laboratory because of its higher sensitivity when determined by ICP-AES.
 Sr/Ca ratios of the initial fine fractions are fairly high and may indicate that Sr is present in the same phases of contamination as those containing Fe and Mg (Figure 4d). Alternatively, the high Sr/Ca values could be associated with the presence of coccoliths within the shell fill. The ratio of Fe/Sr in the contaminating silicate phase may be calculated by mass balance and yields a value of ∼102 mol mol−1. Therefore if sample Fe/Sr ratios fall below about 1 mol mol−1 the risk of Sr contamination from silicate sources will be minimal. Hence the degree of silicate removal required for measurement of Sr/Ca ratios is much less than for Mg/Ca (Figure 4f).
 Although the focus of the cleaning procedure is to remove clay contamination, foraminiferal tests will occasionally contain silicate grains larger than a few μm (e.g., if there is IRD present in the sample). Theseso-called “particulates” prove hard to remove by traditional methods of clay removal because they tend to settle rapidly. Yet they have the potential to cause contamination of Mg/Ca ratios either through leaching in the dissolution acid or by entering the plasma during analysis. Removal of silicate particulates is best accomplished by hand under a microscope using a fine hair brush (Appendix 1). The sample described above contained large amounts of IRD and foraminiferal test chambers contained appreciable numbers of larger particulate grains. Particulates were removed before final dissolution of the sample and subsequently acidified and analysed. Had these particulates not been removed prior to sample dissolution, the measured Mg/Ca ratio would have been higher by >0.15 mmol mol−1 (giving a calculated temperature of approximately 1°C higher).
5. Experiments to Monitor Removal of Organic Matter
 The presence of organic matter in foraminiferal samples may produce elevated values of Mg/Ca [Hastings et al., 1996; Rathburn and De Deckker, 1997]. Mg may be contained within the organic material itself or released from clay minerals trapped within the organic matrix. Organic matter is removed by oxidation in a hot, alkali-buffered solution of H2O2. The product CO2 is allowed to escape and any associated impurities are brought into suspension. Multiple water washes are then employed to remove suspended impurities and any remaining reagent. The following experiment was designed to determine how much oxidative treatment is required when preparing sediment core samples. It should be noted that preparation of sediment trap or plankton tow samples may require further treatment as such samples typically contain very high levels of organic material [Anand et al., 2003].
 A core-top sample from the oxygen minimum zone of the Arabian Sea was selected (core 243K, 14.7°N, 51.6°E, water depth 1041 m) with high bulk sediment organic carbon content (2.58%) and a CaCO3 content of 55.5% [Sirocko, 1994]. 100 individuals of Globigerinoides ruber were picked from the 300–355 μm fraction. These were crushed together and homogenized before being split 5 ways. Each sample then underwent identical clay removal treatment (Appendix 1) before oxidation tests were performed. The 5 samples were subjected to 0, 2, 4, 6, or 10 oxidation steps (a single step involved adding 250 μl of 1% oxidising agent and heating in a boiling water bath for 5 minutes, ultrasonicating briefly at 2.5 minutes to allow any gaseous build-up to escape and maintain contact between sample and reagent). After oxidation and particulate removal, samples were leached 4 times for 30 seconds in 0.001M HNO3 before dissolution.
 Samples subjected to less than 20 minutes of oxidative treatment display elevated Mg/Ca ratios before leaching (Figure 5b). Leaching in dilute acid acts to remove some of the remaining organic matter but also causes partial dissolution of the sample and an associated decrease in foraminiferal Mg/Ca (section 7). The sample with no oxidative treatment had a higher Mg/Ca ratio than all of the other samples even after 4 leaches. In this example, the removal of organic matter was associated with a decrease in measured Mg/Ca of approximately 0.4 mmol mol−1 (or about 1°C). Removal of organic matter does not seem to affect Sr/Ca ratios to a significant extent.
6. Experiments to Monitor Removal of Mn-Oxides
 To demonstrate the effect of this treatment, samples of two foraminiferal species were picked from core-top sediments from the North Atlantic [Elderfield and Ganssen, 2000]. From each sample, 40 individual tests were picked, crushed and homogenized before being split into two subsamples. All samples underwent the refined silicate removal and oxidation stages. The first set of subsamples were then leached once and dissolved while the second were subjected to reductive cleaning before leaching and dissolution. This experiment differs from the experiment described in section 3 in two respects: (1) the refined cleaning method (Appendix 1) was used here and (2) sample pairs in the preliminary study were not homogenized before cleaning.
 Mg/Ca ratios of the reductively cleaned samples are systematically lower than non-treated samples by about 10–15% (Figure 6). This compares with a difference of 15–20% in the experimental results illustrated in Figure 3. Of course, Mn/Ca ratios also decrease as a result of the reductive treatment but Sr/Ca ratios show no significant change. The initial Mn/Ca ratios of the samples are low (<15 μmol mol−1) and typical for core-top samples [Boyle, 1983] (Figure 7a). Assuming a Mg/Mn ratio of ∼0.1 mol mol−1 for any Mn-rich coating present (section 3), the decrease in Mn/Ca through reductive cleaning (generally <5 μmol mol−1) would be expected to correspond to a decrease in Mg/Ca of about 0.5 μmol mol−1 or about 0.03%. This is clearly much lower than the observed decrease (even with an initially high Mn/Ca ratio of say 300 μmol mol−1, we would expect a decrease in Mg/Ca of only ∼0.03 mmol mol−1 if all the coating were removed by reductive cleaning). Similarly, if the observed average decrease in Fe/Ca (∼0.03mmol mol−1) through the reductive treatment were associated with removal of silicate contamination, we may expect a similar decrease in Mg/Ca (of about 1%). Monitoring of the proportional dissolution of the samples throughout cleaning shows that the reductive step causes significant dissolution of the sample calcite (Figure 7b). Therefore it appears, contrary to the findings of Martin and Lea , that reductive cleaning caused a lowering of sample Mg/Ca by partial dissolution of the sample rather than by removal of a Mn-oxide coating.
 Foraminiferal samples used for paleoceanographic reconstructions often come from sediments where redox conditions enhance the formation of Mn-rich oxide coatings. In order to investigate the effect of reductive treatment on the Mg/Ca ratio of such a sample, we again used tests of G. bulloides from the 122cm interval in NEAP 8K. Mn/Ca ratios measured on down-core samples of G. bulloides from NEAP 8K (without reductive cleaning) show variability which probably reflects changing redox conditions within the sediment column at that site (Figure 7a). A sample of G. bulloides from 122 cm, cleaned without reductive treatment, gives a measured Mn/Ca ratio of approximately 300 μmol mol−1; indicating the presence of a Mn-rich coating. Because reductive cleaning induces partial dissolution of sample calcite, which can cause a decrease in foraminiferal Mg/Ca ratios (see section 7) we treated the sample to several (20x) dilute acid leaches after clay removal and oxidation but before reductive cleaning.
 Dilute acid leaching causes partial dissolution (∼60%) of the sample (Figure 7b). Reductive treatment causes further dissolution (∼50% of the remaining sample). The sample treated with reductive cleaning shows a significantly lower Mn/Ca ratio than samples without this treatment and reflects the removal of a Mn-rich oxide phase (green circle, Figure 7c). It also gives a Mg/Ca ratio about 0.15 mmol mol−1 lower than a sample with a single leach and no reductive cleaning (red square, Figure 7c). However, the multiple acid leaching performed prior to reductive treatment has the effect of lowering sample Mg/Ca by approximately the same extent (blue triangle, Figure 7c). In this case we suggest that removal of the Mn-oxide coating per se does not cause a significant lowering of foraminiferal Mg/Ca.
 Occasionally, foraminiferal samples will have high (>102 μmol mol−1) Mn/Ca ratios even after reductive cleaning (the sample described above is an example). This is thought to be due to the presence of a Mn-rich carbonate coating [Boyle, 1983]. Removal of this coating is difficult but the use of acid-leaching has been suggested [Boyle and Keigwin, 1985]. This treatment may also cause dissolution of the foraminiferal sample and lowering of Mg/Ca.
7. Experiments to Monitor the Final Leaching Stage
 Published methods for cleaning foraminiferal carbonate for Cd/Ca, Ba/Ca, Mg/Ca and Sr/Ca ratios all include a dilute acid (∼0.001M HNO3) leaching step prior to final dissolution and analysis. The purpose of this step is to remove any contaminants that may have adsorbed to the shell surfaces during cleaning [Boyle and Keigwin, 1985].
 To evaluate the effect on Mg/Ca and Sr/Ca, 20 tests of G. bulloides from 122 cm in core NEAP-8K were taken through the full cleaning procedure (without the reductive step) and then subjected to 20 dilute acid leach steps each involving the addition of 250 μl of 0.001M HNO3 and ultrasonication for 30 seconds. Leaching caused significant (∼60%) dissolution of the sample and a decrease in Mg/Ca of about 0.2 mmol mol−1 or 10% (Figure 8). This decrease corresponds to a calculated temperature drop of almost 1°C. Sr/Ca was not significantly affected by the partial dissolution procedure (Figure 8d).
 Core-top observations provide strong evidence that partial dissolution at the seafloor causes a decrease in foraminiferal Mg/Ca ratios [e.g., Russell et al., 1994; Brown and Elderfield, 1996; Rosenthal et al., 2000]. It is not clear whether the effect of partial dissolution as recreated in the laboratory is similar to that occurring in nature. Nevertheless, it is clear that the two processes have similar results, i.e., a decrease in bulk sample Mg/Ca. Possible explanations for the observed decrease in Mg/Ca include: (1) shells within the sample population with higher Mg/Ca dissolve preferentially (2) preferential removal of Mg-rich portions of shells (3) “artificial” leaching of Mg from the calcite lattice (4) selective dissolution of contaminant phases with high Mg contents. Brown and Elderfield  demonstrated that Mg/Ca ratios of the planktonic foraminifera, G. tumida, could also be lowered by progressive dissolution in the laboratory. They suggested that dissolution proceeded by the preferential removal of regions of the test formed in warmer waters, i.e., those with higher Mg/Ca ratios. Benway et al.  use a flow-through Mg/Ca dissolution method [Haley and Klinkhammer, 2002] to demonstrate the same process as illustrated here. They propose that early dissolution involves the removal of test portions formed in warmer waters (analogous to option 2 above). It is clear that partial dissolution induced in the laboratory may cause a significant lowering of measured Mg/Ca ratios. Therefore we suggest limiting the use of weak acid leaching to a single leach before dissolution and analysis.
8. Overview of the Procedure
 In order to demonstrate the relative importance of the various steps outlined above (excluding the reductive and chelation procedures) two samples of G. bulloides were taken through the complete procedure (Appendix 1) plus multiple leaching steps and all removed portions were analysed (Figure 9). Depending on the degree of leaching prior to dissolution, loss of sample carbonate during cleaning may be 30% or more. It is apparent that silicate removal has by far the greatest influence on sample Mg/Ca values (Figure 9b). Compared with this, the oxidation and leaching stages have minor effects. Sr/Ca ratios are sensitive to the first two or three clay removal steps but do not vary much later in the procedure (Figure 9f). This confirms that measurement of foraminiferal Sr/Ca ratios does not require a very rigorous cleaning protocol.
 Because all foraminiferal cleaning methods cause some partial dissolution, Mg/Ca data reported by this, and other methods do not necessarily reflect the original Mg/Ca ratios of the pure foraminiferal carbonate before cleaning. Some sample loss by fragmentation or dissolution during cleaning is inevitable. Because Mg is heterogeneous within tests and varies from test to test within a sample, some change in Mg/Ca is also inevitable. The method described here seeks to minimize sample loss but not at the expense of thorough cleaning. The key issues then are (1) is the cleaning method reproducible and (2) are there systematic differences between results of methods that use different cleaning procedures.
8.1. Differences Between Methods
 The results summarized in section 6 demonstrate that use of reductive cleaning to remove Mn-rich coatings causes sample dissolution, leading to a decrease in the Mg/Ca ratio (Figure 6a). It is likely that the magnitude of this effect may vary somewhat from sample to sample (for example, foraminifera from dissolved sediment sections may perhaps be less affected) but we have not investigated this in detail. The comparison shown in Figure 6 indicates that samples cleaned using the reductive cleaning stage give ∼10–15% lower Mg/Ca ratios than those cleaned without this step. This corresponds to a systematic temperature difference of about 1–1.5°C between the two methods. This is somewhat larger than the error generally quoted for Mg/Ca-thermometry (±0.5–1°C).
 An alternative evaluation of the difference between cleaning methods is to compare published calibrations from different laboratories where contrasting approaches have been employed. To do this, it is important to compare data where temperature is estimated by comparable methods and using the same size fraction of a species. A potentially suitable comparison is between two calibrations, both using G. ruber (white) 250–350 μm, one based upon core top samples from the equatorial Pacific obtained using reductive cleaning [Lea et al., 2000] and one from this laboratory using the method outlined in Appendix 1 (except that a somewhat more rigorous oxidation stage was employed as the samples were obtained from N. Atlantic sediment traps and therefore contained more organic matter than a typical core-top sample) [Anand et al., 2003]. Although the core top samples were calibrated against sea surface temperature and the sediment trap samples against calcification temperature, it may be assumed that these estimates are comparable for this species.
 The calibration for the sediment trap samples using the method described here [Anand et al., 2003] gave:
and that for the core top samples using reductive cleaning [Lea et al., 2000] gave:
However, it is unfortunate for this comparison that the core top samples are affected slightly by dissolution and the calibration has been subsequently corrected [Dekens et al., 2002] by the equivalent of about 1.5°C to:
If the difference between cleaning methods seen in our experiments is reflected in the calibrations it will be shown by differences in the pre-exponential constant. It can be seen that all three calibrations are identical within error and values of the constant obtained using reductive cleaning (0.30 corrected upward to 0.38) span the value obtained without reductive cleaning (0.34). This serves to illustrate that the differences discussed between the two cleaning methods are small relative to the uncertainty in calibrations. The difference discussed (equivalent to ∼1°C) compares with standard deviations in temperature estimates based on the calibrations of ±1.13°C [Anand et al., 2003] and ±1.2 to ±1.4°C [Dekens et al., 2002]. Nevertheless, this is a systematic error and it will be important in future work to firmly establish the comparability of data using the two approaches. It also serves to demonstrate that the principal issue in preparation of samples for Mg/Ca thermometry is the efficient removal of silicate material rather than the reductive cleaning issue.
8.2. Reproducibility of the Method
 Any method of sample preparation needs to give reproducible results. To investigate the reproducibility of the Mg/Ca cleaning method used in this study, a total of 400 individuals of G. ruber (300–355 μm) were picked from a single core-top sample from the Arabian Sea (core 36KL, 17.1°N, 69.0°E, water depth 2055 m [Sirocko, 1994]). Ten sub-samples, each containing 20 shells, were crushed individually and cleaned using the procedure given in Appendix 1. Results for these samples help define the natural variability in Mg/Ca associated with core-top samples (Figure 10). The remaining 200 shells were crushed together and mixed with the aim of homogenizing them as far as possible (it is unlikely that complete homogenization is attainable since tests are only crushed sufficiently to open all chambers; not to produce a powder). These were then split into 10 further sub-samples and cleaned separately. All measurements were made within a single analytical run.
 The natural variability attained for the 10 unmixed sub-samples was ±3.7% (1σ) for Mg/Ca and ±1.2% (1σ) for Sr/Ca. The deviation for Mg/Ca is better considered with respect to calcification temperature. Using the generic Mg-calibration equation of Anand et al. :
the calculated temperature for separate samples of 20 individuals was 28.3 ± 0.4°C. This variability is within the error generally quoted for Mg/Ca-thermometry (±0.5–1°C).
 Variability associated with the homogenized samples was less than half that of the non-mixed samples for both Mg/Ca (1.8% ≡ 0.2°C) and Sr/Ca (0.5%). Therefore it can be stated that the reproducibility attainable from use of the Mg-cleaning method described here is at least twice as good as may be expected from natural variation between populations of 20 individuals. If all the variability observed in these data was a reflection of the cleaning procedure, we would not expect to see any difference between the mixed and un-mixed samples.
 The various steps required for the preparation of foraminiferal samples for Mg/Ca and Sr/Ca analysis have been evaluated for their relative importance and effects on measured elemental ratios. It has been shown that the removal of silicate contamination is by far the most important step for the measurement of Mg/Ca ratios. In comparison, the oxidation and leaching stages have relatively minor effects. Reductive treatment causes partial dissolution of sample carbonate and may lead to a corresponding decrease in foraminiferal Mg/Ca ratios of up to 10–15%. This effect is similar in magnitude to the uncertainties associated with published Mg/Ca thermometry calibrations. Because foraminiferal Mg/Ca thermometry involves comparison of sample data against empirical calibration curves, the most important considerations are that analyses are reproducible and that temperature estimates are obtained with a cleaning procedure comparable to that used in the calibration. The reproducibility of the method described here obtained for G. ruber in a core-top sample from the Arabian Sea was ±1.8% (or 0.2°C) for Mg/Ca and ±0.5% for Sr/Ca. Sample preparation for Sr/Ca analysis does not require the same degree of rigor as is necessary for Mg/Ca work.
 Clearly it would be desirable for workers involved in applying foraminiferal Mg/Ca thermometry to develop a consensus on the cleaning protocol employed during sample preparation. The issues outlined here highlight the importance of the various stages in cleaning foraminiferal calcite but also suggest that samples from differing locations/time periods may require varying degrees of rigor and potentially different treatments. It is clear that the removal of silicate minerals is of primary importance for the measurement of Mg/Ca ratios. In regions where clay content may be high, use of a rigorous clay removal technique is necessary. On the other, hand, if clay contamination is judged to be of less significance, fewer steps may be sufficient. Similarly, where larger silicate grains are present, such as IRD in the Glacial North Atlantic, these should be removed. By monitoring for silicate contamination it is possible to assess cleaning efficacy at all times.
 The effects of Mn-Fe-oxide removal on foraminiferal Mg/Ca have been discussed. While important for measurement of Cd/Ca ratios, removal of oxide coatings does not appear necessary for Mg/Ca analysis per se since the Mg content of such phases is probably low. Boyle  illustrated that Al/Ca ratios decreased through the use of reductive cleaning, i.e., that silicate contamination could be lessened by reductive treatment. This could imply that the more rigorous treatment, including additional ultrasonic steps, involved during reductive cleaning acts to remove more silicate minerals than by straight forward clay removal. Alternatively, silicate minerals may be trapped by the oxide coating and therefore only released by its removal. In this case, the presence of silicate contamination (and potential Mg contamination) can again be monitored during analysis. If silicate contamination persists after even rigorous clay and particulate removal, the use of reductive cleaning may be considered, bearing in mind its corrosive effects. Furthermore, just how constant the offset between samples treated with and without reductive cleaning may be has not yet been investigated. For example, it is not clear whether samples with initially different Mg/Ca ratios or those from different preservational settings would respond in the same way to reductive cleaning. Future work may involve a downcore study to investigate the comparative effects of the various treatments.
 Two other issues require some comment. One concerns the initial treatment of sediment samples to remove the fine (<63 μm) fraction prior to picking of foraminifera. Treatment of samples with bleach or by “dry-ashing” to remove organics is not recommended. The objective of the washing process must be to eliminate the maximum amount of silicates at this stage but not at the expense of fragmentation and dissolution of the sample. Un-buffered dispersion agents should not be used. Employment of ultrasonic dispersion requires careful monitoring.
 The second issue is speed of analysis relative to rigor. It is desirable that any procedure used in paleoceanographic research should be as straightforward and time effective as possible. The addition of a reductive cleaning procedure makes for a more lengthy procedure. The method described here for removal of silicate material is more lengthy than a batch method of treating samples. However, inclusion of steps to account for all contingencies may not be necessary. Overall, it is advisable that any sample set should receive initial screening for the degree of cleaning necessary before significant numbers are analysed. Continued monitoring is advisable throughout the entire sample set.
Appendix 1.: Stepwise Cleaning Procedure for the Preparation of Foraminiferal Calcite for Elemental Analysis
 Foraminiferal tests are crushed using two clean glass plates. The aim here is to allow any chamber fill to escape during subsequent cleaning stages.
 1. Place the tests in a single layer on the lower glass slide. Keep the sample moist with excess water.
 2. In a controlled manner, lower the second plate onto the sample and apply gentle pressure in order to open every shell chamber. Take care not to over-crush the sample; this will lead to excessive loss of sample during cleaning.
 3. Remove the upper glass plate and transfer all particles to the lower plate.
 At this time, a piece of light coloured paper should be positioned beneath the sample while under a microscope. This will reveal the presence of any larger silicate grains that may not be removed during the following clay removal steps. It is not necessary to remove such grains at this stage but only to note their presence so that action may be taken later on.
 Steps (4) to (6) should be followed if the sample is intended for paired analyses of trace metals and stable isotopes or similar.
 4. Add water to the sample in order to bring the shell particles into suspension.
 5. Mix particles thoroughly with a brush with the aim of homogenizing the sample as far as possible.
 6. Remove any excess water and divide into the desired proportions.
 7. Using a moistened brush, transfer the crushed sample to an acid cleaned 500 μl micro-centrifuge tube or relevant sample tube.
A1.2. Clay Removal
 During this stage, all samples should be treated individually in order to maximize cleaning effectiveness (batch treatment for ultrasonication is appropriate). It is important to use separate pipette tips for adding and removing reagents.
 Having opened the test chambers during crushing, much of the test fill will be loosened and easily brought into suspension.
 1. Squirt 500 μl of UHQ H2O onto the crushed sample (trapped air bubbles may be freed by flicking the tube end with a fingernail).
 2. Allow the sample to settle for 30 seconds or so.
 3. Remove the overlying solution (supernatant) with a separate pipette. The size of a 500 μl pipette tip is suitable for removing most of the overlying liquid from the tube without risk of sample loss.
 At this stage, all tubes should still contain about 10–20 μl of H2O.
 4. Place the sample rack in an ultrasonic bath for 1–2 minutes. This will encourage separation of more tightly bound clays from the test surfaces.
 Suspended clays will appear as a milky residue in the liquid just above the sample.
 5. Squirt 500 μl of UHQ H2O onto each sample. This will agitate the sample and bring loose clays into suspension.
 6. Briefly allow the sample to settle (minimal settling technique). Sufficient settling will only take a number of seconds (long enough for the distinct carbonate grains to reach the bottom). After this period the remaining settling material will mainly comprise unwanted silicate particles.
 7. Remove the overlying solution.
 8. Repeat steps (4) to (7) a further 4 times. More repetitions may be necessary for as long as clays are being visibly brought into suspension by ultrasonication.
 After the water cleaning steps, methanol is used for further clay removal. The lower viscosity of this reagent should dislodge material still attached to the carbonate tests.
 9. Squirt 250 μl of Aristar methanol into each tube.
 10. Ultrasonicate the tubes for 1–2 minutes.
 11. Treating each tube individually, lift the methanol off the sample with a pipette and quirt straight back in to bring clays into suspension.
 12. Allow sample to settle for a few seconds and remove the methanol.
 13. Repeat steps (9) to (12).
 14. Repeat steps (5) to (7) in order to remove any remaining methanol (further ultrasonication may be applied if desired).
A1.3. Removal of Organic Matter
 1. Add 250 μl of alkali buffered 1% H2O2 solution to each tube and secure the rack with a lid to prevent tubes popping open while under pressure.
 2. Place the sample rack in a boiling water bath for 10 minutes. At 2.5 and 7.5 minutes remove the rack momentarily and rap on the bench top to release any gaseous build-up. At 5 minutes place the rack in an ultrasonic bath for a few seconds and return to the water bath after rapping on the bench. The aim of these interim steps is to maintain contact between reagent and sample.
 3. Remove the oxidizing reagent using a pipette.
 4. Repeat steps (1) to (3).
 5. Remove any remaining oxidizing reagent by filling the tube with UHQ H2O and removing after settling. This step should be repeated 1–2 times.
A1.4. Removal of Coarse-Grained Silicates
 This step is necessary if silicates were observed in the sample after crushing. It is a good idea to follow this step even if no larger particles were seen; any foreign body that has entered the sample during the proceeding steps might bias the desired measurement.
 Particulate removal at this stage is less time consuming than straight after crushing as the clay treatment and oxidation steps will have broken down and removed some particles already.
 A 100 μl pipette is used to transfer the sample into a 1ml glass micro-beaker.
 1. Squirt 100 μl of UHQ H2O into the sample and immediately transfer to a micro-beaker avoiding settling.
 2. Repeat step (1) 3–4 times ensuring that the entire sample is transferred.
 3. Remove excess H2O leaving approximately 100μl overlying the sample.
 The sample is viewed under a microscope using a dark and light background in turn.
 4. Remove any particles that are not apparently carbonate using a fine brush. Strongly discoloured carbonate should also be removed.
 5. Transfer the sample into a clean micro-centrifuge tube using the technique in step (1) and remove excess H2O.
 Note, if particles are not removed using the technique outlined above, it is nevertheless, good practice to transfer all samples to clean tubes before continuing. If a transfer is not performed, thorough cleaning of the sample tube (including cap) with UHQ H2O should be carried out to ensure that all oxidizing reagent is removed.
A1.5. Weak Acid Leach
 A weak acid is used to remove any adsorbed contaminants from the test fragments.
 1. Add 250 μl of 0.001M HNO3 to each sample.
 2. Ultrasonicate all samples for 30 seconds.
 3. Remove acid from each sample.
 4. Squirt UHQ H2O into each tube.
 It is important to replace the leach acid with H2O as soon as possible for all samples in order to prevent excess dissolution.
 5. Remove the overlying H2O.
 6. Repeat steps (4) and (5).
 7. Using a 10 μl pipette, carefully remove any remaining solution from each sample.
 Dissolution should be performed with consideration given to any non-carbonate particles that may still be present in the sample. The following steps should act to reduce the risk of contamination from such phases.
 1. Add 500 μl of 0.075M HNO3 to each sample (for small samples 300 μl is adequate).
 2. Place the sample rack in an ultrasonic bath to promote reaction.
 3. Momentarily remove each tube in turn and flick with a fingernail to allow any build-up of CO2 to escape and the reaction to continue. As soon as production of CO2 ceases in any sample tube, remove that tube from the ultrasonic bath and leave to settle.
 4. Once all samples are dissolved, they should be transferred to clean sample tubes.
Appendix 2.: Preparation of Reagents
A2.1. Oxidising Reagent
 Alkali buffered 1% H2O2 solution.
250 μl oxidising solution used per sample.
Prepare fresh mixture for each batch of samples.
30% w/v (Aristar grade).
0.1 M sodium hydroxide (Aristar grade)
Add 100 μl H2O2 to ∼10 ml 0.1 M NaOH, sufficient for ∼20 samples.
A2.2. Dilute Acid Leach
 QD 0.001 M HNO3
250 μl used per sample.
75 fold dilution of 0.075 M HNO3
6 ml 0.075M HNO3 in 450ml UHQ H2O
A2.3. Dissolution Acid
 QD 0.075 M HNO3
500 μl used per sample.
200 fold dilution of QD conc. HNO3 (15M)
5 ml conc. HNO3 in 1 litre UHQ H2O
 We thank David Lea, Thorsten Kiefer, Ed Boyle, Michael Wara and Peggy Delaney for useful discussions and reviews. The Natural Environment Research Council (GT04/98/44/ES, GR3/1310 and GR3/JIF/05a), The European Commission (EVR1-CT-40018: CESOP) and the Comer Foundation provided financial aid.