The influence of temperature and seawater carbonate saturation state on 13C–18O bond ordering in bivalve mollusks

The shells of marine mollusks are widely used archives of past climate and ocean chemistry. Whilst the measurement of mollusk δ18O to develop records of past climate change is a commonly used approach, it has proven challenging to develop reliable independent paleothermometers that can be used to deconvolve the contributions of temperature and fluid composition on molluscan oxygen isotope compositions. Here we investigate the temperature dependence of 13C–18O bond abundance, denoted by the measured parameter Δ47, in shell carbonates of bivalve mollusks and assess its potential to be a useful paleothermometer. We report measurements on cultured specimens spanning a range in water temperatures of 5 to 25 °C, and field collected specimens spanning a range of −1 to 29 °C. In addition we investigate the potential influence of carbonate saturation state on bivalve stable isotope compositions by making measurements on both calcitic and aragonitic specimens that have been cultured in seawater that is either supersaturated or undersaturated with respect to aragonite. We find a robust relationship between Δ47 and growth temperature. We also find that the slope of a linear regression through all the Δ47 data for bivalves plotted against seawater temperature is significantly shallower than previously published inorganic and biogenic carbonate calibration studies produced in our laboratory and go on to discuss the possible sources of this difference. We find that changing seawater saturation state does not have significant effect on the Δ47 of bivalve shell carbonate in two taxa that we examined, and we do not observe significant differences between Δ47-temperature relationships between calcitic and aragonitic taxa.


Introduction
Molluscan carbonate was amongst the first biologically precipitated materials investigated during the development of the oxygen isotope paleotemperature scale (Epstein et al., 1953). Subsequently fossil mollusks have been widely used as an archive of past environmental change and seawater chemistry (e.g., Keith et al., 1964;Killingley and Berger, 1979;Grossman and Ku, 1986;Taviani and Zahn, 1998;Veizer et al., 1999;Tripati et al., 2001;Tripati and Zachos, 2002;Ivany et al., 2008;Wanamaker et al., 2011). However it has proven challenging to develop robust independent paleothermometers in mollusk carbonate; for example, approaches using trace element partitioning (Mg / Ca, Sr / Ca) into mollusk shell carbonate are often hampered by strong biological controls and high inter-and intra-specimen variability (e.g., Dodd, 1965;Lorens and Bender, 1980;Klein et al., 1996;Gillikin et al., 2005;Freitas et al., 2006Freitas et al., , 2008Freitas et al., , 2009Heinemann et al., 2011;Wanamaker et al., 2008). Therefore it has not yet been possible to reliably partition the contributions of temperature and seawater δ 18 O to bivalve mollusk carbonate δ 18 O with a high level of confidence in environments where both parameters could be expected to vary.
"Clumped" isotope paleothermometry is an emerging approach for reconstructing the temperatures of carbonate mineral precipitation (Eiler, 2011). The technique is founded on the principle that rare isotopes of carbon and oxygen have a thermodynamically driven tendency to bond with each other, or "clump", and that this effect increases as temperature decreases (Wang et al., 2004;Schauble et al., 2006). In practice the abundance of 13 C-18 O bonds in carbonate minerals is measured from the abundance of mass-47 CO 2 (predominantly 13 C 18 O 16 O) liberated on phosphoric acid digestion of carbonate minerals . Measured values are compared to a reference frame where isotope abundances from sample gases are compared to reference gases that have been heated to 1000 • C, producing a nearly random distribution of isotopes among all isotopologues Affek and Eiler, 2006;Huntington et al., 2009;Passey et al., 2010). More recently, standardization to CO 2 equilibrated with water at two or more controlled temperatures has been proposed as an "absolute reference frame" in an effort to reduce interlaboratory differences due to mass spectrometric effects such as bond breaking and reordering during sample gas ionization (Dennis et al., 2011). Here we refer to data presented relative to heated gases only as "relative to the stochastic distribution" although this reference frame has also been described as the "Caltech intralab reference frame" as strictly speaking our heated gases do not reach the full stochastic distribution Huntington et al., 2009). Data presented relative to the newly proposed reference frame is referred to as being on the "absolute reference frame" (Dennis et al., 2011). In both cases, we report data using the 47 parameter, which expresses the abundance of 13 C-18 O bonds found in a sample as an enrichment, in per mil, above that expected if isotopes were distributed randomly Huntington et al., 2009).
Following the calibration of the clumped isotope thermometer in inorganically precipitated calcite  detailed calibration studies of foraminifera, coccoliths, tooth bioapatite, and corals from our laboratory have shown that these biologically precipitated materials appear to yield a relationship between 47 and temperature ( Fig. 1) that is very similar to inorganic calcite (Tripati et al., 2010;Thiagarajan et al., 2011). The close relationship between the inorganic calcite calibration and 47 data from foraminifera, coccoliths, and corals -even in taxa that show deviations of up to ∼ 4 ‰ from the δ 18 O values expected given the temperature and δ 18 O of the fluid from which they precipitate -suggests either that inorganic calcite and biogenic carbonates are close to equilibrium or that all exhibit non-equilibrium effects of similar magnitude. In contrast, a study on otoliths and data from a single Porites coral specimen exhibit deviations from the inorganic calibration line (Ghosh et al., , 2007. In the case of otoliths this could be explained by uncertainties on the precise formation temperature of the samples, as appears to be also a factor in measurements on thermocline dwelling foraminifera (Tripati et al., 2010), or due to small systematic analytical errors that were likely more common early in the history of 47 measurements. The difference between Porites coral and the inorganic calibration in Ghosh et al. (2006) is relatively large and could be the result of a growth rate related "vital effect" (Saenger et al., 2012).
It is unclear why some biogenic carbonates exhibit relationships between temperature and 47 that resemble the inorganic calibration of Ghosh et al. (2006) whereas other biogenic materials do not. It is possible that this difference in behavior will shed new light on the long-standing problem concerning the origin of stable isotope "vital effects" (Weiner and Dove, 2003); namely, differences in isotopic composition between biogenic materials and compositions expected for thermodynamic equilibrium with their environment. Various explanations have been advanced for vital effects on the δ 18 O of biogenic carbonates, one invoking kinetic isotope effects associated with processes such as the hydration and hydroxylation of CO 2 in solution or crystal growth rate (e.g., McConnaughey, 1989); a second set of explanations invoke an equilibrium isotope fractionation associated with the fractionation of isotopes between species of dissolved inorganic carbon present in an organisms calcifying fluids (i.e., isotope fractionation between CO 2− 3 and HCO − 3 ), which then gets preserved in the solid phase (e.g., Spero et al., 1997;Zeebe, 1999;Adkins et al., 2003;and Tripati et al., 2010). Other models have invoked kinetic effects associated with element partitioning or isotope effects at the surface of a growing crystal, which is influenced by both crystal growth rate and dissolved inorganic carbon (DIC) speciation (Watson, 2004;Tripati et al., 2010). Preliminary predictions suggested a difference in 13 (Ghosh et al., , 2007Came et al., 2007;Tripati et al., 2010); and now has the data from Thiagarajan et al., included (Thiagarajan et al., 2011). Data from Zaarur et al., was not included due to uncertainties over exactly what environmental conditions the materials analyzed should reflect (Zaarur et al., 2011), and the Porites coral analyzed by Ghosh et al., was also excluded due to apparent kinetic isotope effects on 47 values . Also shown is a regression through the same compilation of published materials now converted into the absolute reference frame (Table S1) via the secondary transfer function method (Dennis et al., 2011). Note that the 10 6 /T 2 scale with T in degrees Kelvin is the primary temperature scale used for data plots, with a secondary x-axis in degrees Celsius presented as a guide only. All regression lines were recalculated from original data (see methods for details).
is small and would not necessarily be measurable were it to be preserved in the solid phase (Guo et al., 2008), whereas more recent solution phase ab initio calculations predict a slightly larger effect which may potentially be measurable in carbonates precipitating from a large pH range but is still probably too small to be measured across the typical range of pH seen in the modern ocean (Hill et al., 2013).
The similarity in 47 between inorganic calcite and some biogenic carbonates (foraminifera, coccoliths and some corals) is consistent with pH effects on carbonate isotopic composition, though the effects are not necessarily required (Tripati et al., 2010;Thiagarajan et al., 2011), and suggest that any kinetic isotope effects must have negligible influence on 47 values. Conversely the discrepant 47 values of a Porites coral  are more consistent with a larger kinetic isotope effect and not a pH effect (Saenger et al., 2012). Here, we investigate the controls on 13 C-18 O bond abundance in the shells of bivalve mollusks, with the dual aim of providing an empirical proxy calibration for paleoclimate studies as well as giving some new perspectives on the fractionation of isotopes during carbonate biomineralization.

Mollusk culturing
We analyzed cultured bivalve specimens from several different laboratories. We briefly summarize the methods and materials of these culturing experiments here and refer to previous publications for more detailed descriptions of culturing conditions where appropriate.
Specimens of Arctica islandica were cultured at 10.3 and 15 • C at the Darling Marine Center in Walpole, Maine. Approximately 30 juvenile (∼ 3 yr; shell height = ∼ 40 mm) specimens of A. islandica were grown in muddy sediment in a temperature-controlled environment for 15 weeks. Ambient seawater (salinity = 30.4 to 30.7; Hydrolab ® Min-iSonde ± 0.2) from 10 m water depth was pumped into the flowing seawater labs, where the water flow was reduced (∼ 6 L min −1 ) and the water was heated or cooled to maintain the desired temperature in the 1500 L holding tank. Prior to the start of the growth experiment, individuals (ind.) were immersed and marked with a biomarker stain, calcein, according to methods outlined previously (Beirne et al., 2012). The clams were exposed to 10 • C seawater for five weeks (8 April to 12 May, 2011), then briefly removed from the growth experiment and re-marked with calcein stain. The animals were then reintroduced to the growth experiment and exposed to 15 • C seawater for 10 weeks (14 May to 21 July, 2011). The clams were only exposed to ambient food. On 21 July 2011, all animals were harvested. The soft tissues were removed and the intact valves were rinsed and air-dried. Samples were then shipped to Iowa State University. Prior to sampling the aragonitic shell material, the periostracum was physically removed with a Dremel ® hand drill. Although growth marks were visible on the shell surface for each temperature treatment, sampling was further guided by the calcein stains (Beirne et al., 2012). Approximately 50 mg of CaCO 3 was removed from the outer shell layer of the left www.biogeosciences.net/10/4591/2013/ Biogeosciences, 10, 4591-4606, 2013 valve of one shell with a Dremel ® hand drill equipped with a diamond tipped bit on low speed. 5 • C cultures of A. islandica and Mytilus edulis were conducted at the Helmholtz Centre for Ocean Research Kiel (GEOMAR), Germany. Young M. edulis specimens were collected in Kiel Fjord (southwestern Kiel Bight) where salinity is on average 16.3 (±2.4 SD) and surface water temperatures range from 0.15 • C in winter to 23.4 • C (mean 10.48 ± 6.13 SD) in summer. A. islandica specimens were collected at 24 m depth at the station Süderfahrt (54 • 32.6 N, 10 • 42.1 E) in the central area of Kiel Bight where salinity is on average 21.8 (±2.4 SD) and temperatures vary between 0.6 and 17.5 • C (mean: 9.03 ± 4.23 SD; Bivalves were kept in temperature-insulated 4 L containers (with 10 ind. of M. edulis, and 7 ind. of A. islandica in each container) and were fed 0.5 mL ind. −1 d −1 of a concentrated livingphytoplankton suspension 5 times a week (DT's Premium Blend; DT's Plankton Farm). Bivalve individuals were allowed to slowly acclimatize to the respective treatments. Temperature and salinity were kept constant for the experimental duration of 15 weeks. Salinity levels were set by admixing freshly collected Baltic Sea water with either ionexchanged water or artificial marine salt (SEEQUASAL). The sample culturing setup is described in detail elsewhere (Hiebenthal et al., 2012). The shell material here used was grown at 5 • C and a salinity of 35. Shell sizes were measured at the beginning of the culturing phase and again prior to sampling using a caliper so that new growth could be identified. After 15 weeks of culturing, the whole soft tissue of the bivalves was removed from the shells and the shells were air-dried (7 d at 20 • C). Care was taken to remove with a Dremel ® hand drill approximately 10 mg from the very outer shell layer, representing new shell growth.
M. edulis and Pecten maximus cultures between 10 and 20 • C were carried out at the School of Ocean Sciences, Bangor University, UK. All animals were acclimated to the laboratory environment at a temperature of ∼ 13 • C for more than two months. Animals of similar size (< 1 yr) were then moved into separate aquaria and slowly acclimated to different but constant temperatures (maximum resolution of 1 • C), constant dimmed-light conditions and controlled food conditions; the aquaria were routinely cleaned of all detritus. Animals were fed a mixed algae solution from containers with a drip tap. For the duration of the experiments, animals were kept in individual plastic mesh cages within each aquarium. Natural seawater pumped from the Menai Strait was conditioned for a few days in settling tanks, and then pumped into holding tanks and introduced as a common supply into the laboratory aquaria. Due to variable growth rates, the duration of the experiments varied with species and aquarium temperature. Because of the limited number of aquaria available, separate temperature-controlled experiments were completed. Animals from the two species can be divided into three groups: one experiment with M. edulis at 12, 15 and 18 • C; a second experiment with M. edulis and P. maximus at 10, 15 and 20 • C; and a third with P. maximus and some M. edulis specimens at 18 • C. Seawater temperature was monitored every 15 min in each aquarium using submerged temperature loggers. Samples for pH measurements were obtained manually every other day by immersing 20 mL plastic syringes below the surface of the seawater in all the aquaria. The samples were subsequently allowed to warm up to room temperature (20 ± 2 • C) in the dark before measurement with a commercial glass electrode (Mettler Toledo Inlab 412). The electrode was calibrated using NBS pH 6.881 and 9.225 buffers (20 • C) and was then allowed to stand until a stable reading was obtained (∼ 1 min). Shell calcite from each specimen was sampled across each growth interval along the main axis of growth, as described previously (Freitas et al., 2008).
Bivalve specimens cultured at 25 • C and at different aragonite saturation states are described in Ries et al. (2009). Specimens of Mytilus edulis, Mercenaria mercenaria, Argopecten irradians, Crassostrea virginica, and Mya arenaria were collected from Nantucket Sound and then transferred into aquaria at the Woods Hole Oceanographic Institution. Briefly, seawater tanks were maintained at 25 ± 1 • C and were illuminated for 10 h per day with 213 W m −2 illuminance. Approximately every 24 days 75 % of the seawater was changed. Air-CO 2 mixtures of 409 and 2856 ppm pCO 2 were introduced into the aquaria with 6-inch micro-porous air stones. Salinity, temperature, and pH of aquarium seawater were measured weekly, and alkalinity biweekly using methods described previously (Ries et al., 2009). Aragonite saturation state, DIC, and pCO 2 were calculated from these parameters. Bivalve shells were sampled from their outermost growth line along their main axes of growth.

Field collected samples
Specimens were collected at the locations given in Table 2. The length of bivalve mollusk growing season will vary somewhat between taxa and this presents an additional source of uncertainty in the calibration. However, in the results section below we show that the slope of our calibration line is not significantly impacted by assumptions over the predominant season of field collected bivalve growth. In the figures and tables presented here we have assumed that there is a bias in the predominant season of shell growth to the three warmest months of the year. In order to obtain seawater temperatures at the sites where specimens were collected from we used the Levitus database (Levitus and Boyer, 1994), and in the case of the specimen from San Diego data from the Scripps Pier coastal water monitoring project (http://www.nodc.noaa.gov/dsdt/cwtg/spac.html).

Cleaning protocols
To evaluate the necessity of sample cleaning, 30-50 mg of each specimen were lightly crushed and treated for 60 min at Biogeosciences, 10, 4591-4606, 2013 www.biogeosciences.net/10/4591/2013/ room temperature in a 3 % H 2 O 2 solution. Samples were then washed three times in excess deionized water and dried in a 50 • C oven overnight. The majority of samples in this study were not cleaned as this cleaning was not found to impact 47 values, as described below.

Stable isotope measurements
Data were collected on two Thermo Finnigan MAT 253 gas source mass spectrometers. Carbonate samples and standards were reacted on the online common acid bath system with automated sample gas purification described previously (Passey et al., 2010). Acid digestion of carbonate minerals was carried out at 90 • C. For full details of analytical methods see previous publications (Huntington et al., 2009;Passey et al., 2010). In brief, 8-10 mg of calcium carbonate samples were crushed and reacted in phosphoric acid on an automated online acid reaction system (Passey et al., 2010) where evolving CO 2 gas is immediately frozen in a liquid nitrogen trap. Sample gases are passed through a Porapak Q 120/80 mesh GC (gas chromatograph) column held at −20 • C to remove potential organic contaminants. Gases are also passed through silver wool to remove sulfur compounds. 48 values were measured and were used as empirical indicators of potential organic contamination (not shown) as has been described previously (Huntington et al., 2009).

Data processing
47 values are defined as where R i represents mass i/mass 44 and R * represents isotopologues in the random (stochastic) distribution . As measurements were made on CO 2 liberated from carbonates by digestion with phosphoric acid heated to 90 • C they are significantly offset from previous published data on carbonates reacted at 25 • C. Passey et al. (2010) empirically determined a value of 0.08 ‰ for this offset based on measurement of carbonate standards, and previous studies have assumed this offset to be constant (Passey et al., 2010;Csank et al., 2011;Finnegan et al., 2011;Suarez et al., 2011;Eagle et al., 2011). Therefore, in order to compare mollusk data to previously published data reacted at 25 • C and reported relative to the stochastic distribution a correction of 0.08 ‰ was made.
We report data using both the stochastic reference frame for 47 values (as reported in previous studies such as Ghosh et al., 2006) and the "absolute reference frame" of Dennis et al. (2011), which assumes a certain value for the difference between heated gases and CO 2 gas standards equilibrated at other temperatures. As the majority of data here was collected before the proposition of the absolute reference frame, we convert 47 values to this reference frame using carbonate standards that were analyzed over the analytical time period. Accepted 47 values for Carrara marble and 102-GC-AZ01 on the absolute reference frame determined in our laboratory are 0.392 ‰ and 0.724 ‰ respectively (Dennis et al., 2011) and these were used to construct an empirical transfer function to generate 47 values on the absolute reference frame, as described previously (Dennis et al., 2011). For the conversion of the compiled published biogenic data (Tripati et al., 2010;Thiagarajan et al., 2011) and inorganic data to the absolute reference frame we also used the secondary transfer function approach, using standard values given in each publication, or where no standard data was given a Carrara marble or NBS-19 value of 0.392 ‰ was used (Dennis et al., 2011). All published data (Ghosh et al., , 2007Came et al., 2007;Tripati et al., 2010;Thiagarajan et al., 2011) and new bivalve data converted to the absolute reference frame is given in Tables 3 and S1, which include the standard values and the slope and intercepts that were used in the transfer function used to convert from the "stochastic reference frame" to the absolute reference frame.
For aragonite δ 18 O calculations an acid digestion fractionation factor of 1.00854126 was used, calculated by extrapolation from a published calibration Kim et al., 2007). For calcite a value of 1.00821000 was used (Swart et al., 1991).

The effect of sample cleaning on stable isotope measurements from bivalve shell carbonate
Bivalves calcify onto a protein matrix (Addadi et al., 2006), which results in the interlocking of organic material and carbonate shell. Organic contamination has the potential to provide isobaric interferences with mass-47 CO 2 measurements, and so we investigated the effect of oxidative sample cleaning on measured 47 values using a treatment of 30 min in 3 % H 2 O 2 . We found that cleaning did not impact measured produced from reaction of bivalve shell carbonate in phosphoric acid (Passey et al., 2010). It is also possible that the majority of the organic matter present in mollusk shell is refractory. This is a different result than seen in biogenic phosphate minerals where sample cleaning does seem to be necessary for accurate measurements . This indicates either that phosphates tend to have higher levels of contaminants that provide isobars for 47 measurements or that the larger sample size reacted to produce CO 2 from phosphate minerals tends to lead to higher levels of contaminants or incomplete reactions of uncleaned samples. Therefore in the remaining analysis presented here we did not perform any sample cleaning.

The relationship between temperature and 47 values in bivalve mollusks
An initial study of the temperature effects on 47 values in modern bivalve mollusks examined three samples (Came et al., 2007). Here we greatly expand the number of specimens measured as well as the range of temperatures encompassed by the calibration. We present data both relative to the stochastic reference frame (to aid comparison with previously published data) and in the recently proposed absolute reference frame (Tables 1-6 and S1). The most direct analysis of our data (i.e. involving a minimum of calculations) is the empirical correlation between known growth temperature and 47 value of bivalve carbonate relative to the stochastic reference frame, using a 90 • C phosphoric acid digestion reaction ( Fig. 2; Table 3). This is the temperature that is now standardly used on our automated online sample reaction and gas purification systems (Passey et al., 2010). We then applied the empirically determined acid digestion correction of 0.08 ‰ to derive data relative to the stochastic distribution that could be compared to previously published data collected on CO 2 produced by digesting carbonates in phosphoric acid at 25 • C (Fig. 2). Linear regressions through each dataset are presented in Fig. 2, and are tabulated with calculated uncertainties and alongside previously published regressions in Table 4.
Individual bivalve samples generally conform reasonably well to the temperature relationship defined by the total population of bivalve data. However a small number of samples, for example the specimen of Zygochlamys patagonica, show a significant departure from this relationship (i.e. fall outside the 95 % confidence intervals of the linear regression; Fig. 2). This appears to represent a unique property of the sample (possibly a "vital effect") on 47 rather than an imprecise measurement as the result is confirmed by analysis of CO 2 extracted from this specimen 6 times ( Table 2). The Levitus atlas of ocean temperatures also calls for a minor difference in mean annual temperature (∼ 8 • C) versus warm summer month (∼ 9 • C) temperature at the location and water depth on the Patagonian shelf where this sample was recovered from. Therefore if the database is correct, then 95 % confidence intervals through 47 measurements made on both cultured (circles) and field collected (triangles) mollusks grown at different temperatures. Shells were reacted with phosphoric acid heated to 90 • C to produce analyte CO 2 . These data are relative to the stochastic distribution as described previously (Huntington et al., 2009) and do not have the empirically derived acid digestion correction of 0.08 ‰ added (Passey et al., 2010), which is used to compare data to that derived from a 25 • C acid digestion reaction. The middle panel is the data with this correction. The bottom panel is bivalve calibration data with the acid digestion correction, then converted into the absolute reference frame (Dennis et al., 2011) using a secondary transfer function. Equations for the relationship between measured 47 and bivalve growth temperature are given in each case. , 10, 4591-4606, 2013 www.biogeosciences.net/10/4591/2013/ incorrect attribution of the season of growth to the summer months in Fig. 2 does not seem a likely explanation (Levitus and Boyer, 1994). Additional work on specific taxa will be needed to confirm this observation. Amongst the most significant departures from previous calibration lines are those from both calcitic and aragonitic specimens forming in the coldest environments, near-freezing shallow marine waters of the Ross Sea off Antarctica that do not reach temperatures significantly above 0 • C all year. The R 2 value of our bivalve mollusk calibration line is 0.7258 (Table 4) using data on the absolute reference frame, and the standard deviation of the residuals (SDR) is 0.017 ‰. This suggests that there is somewhat larger variability in bivalve 47 data compared to other biogenic calibration datasets. For example the linear regression through the foraminifera calibration of Tripati et al. (2010) has an R 2 value of 0.8998 and a SDR of 0.014 ‰, and for the study of corals by Thiagarajan et al. (2011) the R 2 value is 0.8703 with a SDR of 0.015 ‰ (Tripati et al., 2010;Thiagarajan et al., 2011). It is possible that this reflects very subtle biological or mineralogical effects on bivalve 47 data, although, as we describe below, we cannot resolve these effects in our dataset.

Biogeosciences
In the case of field collected bivalves in the figures and regression analysis presented we assumed that preferential growth occurred in the three warmest summer months. However we accept that many taxa do also grow at other times of the year and so in order to assess the impact of our assumption on the resulting regression lines through 47 versus temperature data we also created a regression line using mean annual water temperatures (data not shown) for field collected specimens. The slope of a linear regression line through all bivalve data including field collected specimens assumed to reflect mean annual temperature (rather than warm month average temperatures as in figures and tables) is 0.0350 on the absolute reference frame. This compares to a slope of 0.0362 assuming the warm month average temperature is the predominant growing season for field collected bivalve shells (Table 4). These slopes are not significantly different in an analysis of covariance (ANCOVA) test (p = 0.68). Therefore we conclude that our assumptions over the predominant growing season for bivalve mollusks do not significantly impact the slope of the linear regression lines presented here.

Comparison of bivalve 47 calibration with other theoretical and empirical calibrations
A linear regression through the plot of 1/T versus 47 values for our measurements from bivalves produces a significantly shallower slope than a regression through previously published calibration materials analyzed in our laboratory (Fig. 3). Previous publications did not use the same software or approaches for calculating linear regressions (e.g., Ghosh et al., 2006;and Huntington et al., 2009). Therefore in order to compare regressions precisely, as in Figs. 3 and 4, we recalculate all linear regressions using GraphPad Prism software (Zar, 1984) and it is these values that are presented in Table 4. In practice however these different methods do not yield slopes and intercepts that are markedly different; for example the linear regression presented by Ghosh et al. (2006) yielded a slope of 0.592, whereas using the software utilized here we yield a slope of 0.598. Linear regressions presented here do not take into account errors in carbonate formation temperatures or isotope measurements; in this dataset these tend to be quite similar on average and do not significantly impact the slope of the regression (data not shown). The slopes of the bivalve calibration regression and the Ghosh et al. (2006) inorganic calcite regression are significantly different (Table 5). Additionally, the bivalve mollusk calibration is shown to be significantly different than a compilation of published biogenic data from our laboratory (Table 5). The slopes of the bivalve calibration regression and the inorganic calibration regression of Dennis and Schrag (2010) are not significantly different (Table 5). However, the intercepts of the Dennis and Schrag regression and our bivalve data are significantly different (p = 0.0012). Thus, even though the slopes of these calibrations are statistically www.biogeosciences.net/10/4591/2013/ Biogeosciences, 10, 4591-4606, 2013 1 Cultured specimen growth temperature is accurate to within 0.5 • C on average (see methods). For field collected specimens temperatures correspond to average temperatures for the three warmest months (assumed to be the predominant growing season), it is assumed that there is a 1 • C error in growth temperatures on average. Ocean temperatures determined from the Levitus database. All temperatures are rounded to the nearest integer. 2 C, calcite; A, aragonite. refers to a mixed mineralogy with one mineral predominating. For the purpose of isotope calculations the dominant mineralogy is used. 3 Represents the number of distinct extractions of CO 2 from all samples, which are then purified and analyzed. 4 Relative to the stochastic distribution. Also referred to as data in the Caltech intralaboratory reference frame. Includes the acid digestion correction of 0.08. ± . Values are 1 s.e. 5 Values given on the absolute reference frame.
indistinguishable, there could be an offset in the absolute values of the two. We also note that the apparently higher variability in the bivalve mollusk dataset compared to other biogenic calibration datasets is taken into account by the statistical analysis of slopes presented in Table 5 and so this variability itself cannot explain the statistically significant differences in slopes we observe.
In order to consider whether the slope of the bivalve linear regression could be significantly effected by a few anomalous datapoints we tested the effect of excluding the five specimens recovered from the coldest temperatures from Antarctica (Laternula elliptica and Adamussium colbecki) that are also amongst the most different from the calibration line of Ghosh et al. (2006), yielding 47 values of 0.72-0.74 ‰ rela-tive to the stochastic distribution (Table 3) compared value of 0.80 ‰, which is predicted for carbonates growing at −1 • C if they conformed to the calibration of Ghosh et al. (2006). One possibility is that cold environments favor the expression of kinetic isotope effects on the 13 C-18 O bond abundance in carbonates, as, for example, the rate of reaction for the hydration of CO 2 in solution decreases significantly between 25 and 0 • C and is a potential source of disequilibrium isotope effects in the dissolved inorganic carbon pool from which carbonate forms (Johnson, 1982;Zeebe, 2009). In order to assess potential bias from these datapoints on regression lines we recalculated the linear regression through our dataset excluding taxa that grow in coldest environments and give potentially anomalous 47 values. Exclusion of the   Fig. 3. Comparison of bivalve 47 measurements to previously published calibration data. Here we compare the linear regressions through our mollusk data shown in Fig. 2 to published calibration lines, relative to both the stochastic distribution (left panels) and the absolute reference frame (right panels). In all cases a correction of 0.08 ‰ was made to compare mollusk data to older data collected in our laboratory using 25 • C acid digestion reactions. Mollusk calibration lines have a clearly shallower slope than the inorganic calcite calibration line of Ghosh et al. (2006) and have a similar slope to the calibration of Dennis and Schrag, but with a slight offset to that calibration (Dennis and Schrag, 2010). The mollusk calibration line is also significantly shallower than the linear regression through the compilation of other published materials from our laboratory (bottom panels), with previously published data plotted in this graph given in Table S1. specimens from Antarctica from the mollusk dataset does yield a steeper slope (Table 4) of 0.0402 ± 0.0050 (1 s.e.) on the absolute reference frame, however it does not change the results of our statistical analysis (Table 5) showing that the bivalve mollusk calibration dataset is has a significantly different slope to the previously published biogenic compilation produced in our laboratory and the inorganic calcite calibration of Ghosh et al. (2006). Whilst it is useful to examine the effect of excluding these samples on the regression line, it is also important to note that at present we do not have a good reason to exclude these Antarctic specimens from the regression analysis in this way. There is some rationale for supposing that carbonates that form at low temperatures could be more prone to record kinetic isotope effects, as described above, however previously published studies on L. elliptica and A. colbecki from Antarctica report that their measured δ 18 O are close to their expected equilibrium values (Barrera et al., 1994(Barrera et al., , 1990. Whilst we cannot rule out disequilibrium effects in regard the regression line through all our mollusk data as the most robust calibration.

Calcite versus aragonite
Theoretical calculations predict that there would be an offset between 47 values derived from calcite compared to aragonite Guo et al., 2009). However measurements from foraminifera and corals have not resolved any mineralogical effect (Tripati et al., 2010;Thiagarajan et al., 2011). In our mollusk dataset there is a slight offset between the slopes of regression lines between calcitic and aragonitic mollusks (Fig. 4), however the offset is in the opposite direction to that predicted from theory Guo et al., 2009). The slopes of linear regressions through the temperature-47 data for calcitic and aragonitic taxa (Fig. 4) were not significantly different (p = 0.520). If a difference between calcitic and aragonitic mollusks exists, then it is not easily resolvable. In some cases bivalves that precipitate shells with mixed mineralogy were selectively sampled to only acquire the calcite phase, such as the M. edulis specimens grown at Bangor University (Freitas et al., 2008). However, in other cases this distinction was not made and both mineralogies were sampled, as detailed in Table 4. For the calcite versus aragonite comparison samples with mixed mineralogy were excluded. When comparing the regression lines through the aragonite data to other calibrations (Table 5) it is worth noting that there does not appear to be enough data to statistically determine which of the two different inorganic calcite calibration lines Dennis and Schrag, 2010) the aragonitic mollusk data fits best with. Therefore it remains possible that the lack of a mineralogical difference in our study could be further resolved in the future with larger datasets.

The influence of seawater carbonate saturation state on bivalve stable isotopes
In a number of biogenic carbonates it has been suggested that changes in solution pH can influence carbonate δ 18 O (Spero et al., 1997;Rollion-Bard et al., 2003;Adkins et al., 2003). The effect of changing solution pH and carbonate chemistry on 13 C-18 O bond abundance in carbonate minerals has not been explicitly investigated. Here we analyzed specimens of Mya arenaria, and Argopecten irradians that were cultured at 25 • C and with CO 2 bubbled into the aquarium at either 409 ppm or 2856 ppm producing seawater that was either supersaturated or undersaturated with respect to aragonite (Ries et al., 2009). M. arenaria predominantly precipitates aragonite, whilst A. irradians precipitates low-Mg calcite. Both species showed a reduction in calcification in undersaturated seawater, but care was taken to only sample new growth in each case (Ries et al., 2009). In both cases no significant effects on δ 18 O and 47 values were observed in carbonate There is an offset between the regressions between calcite and aragonite, but it is not statistically significant.
that was formed by specimens cultured in seawater undersaturated with respect to aragonite (Table 6).

Discussion
The data presented here reaffirms the potential of 47 measurements to provide independent constraints on mineral formation temperatures and provides an empirical calibration that can be applied to paleoclimate studies using bivalve www.biogeosciences.net/10/4591/2013/ Biogeosciences, 10, 4591-4606, 2013  Table S1 for the data used for these regression line calculations. b Includes coral data from Ghosh et al., 2006 (but excludes Red Sea Porites), and data from Ghosh et al. (2007), Came et al. (2007), Tripati et al. (2010), Thiagarajan et al. (2011). See Table S1 for values for these data. c Excluding data from the five individuals of Laternula elliptica and Adamussium colbecki (which are Antarctic specimens from the coldest environments sampled in this study) as a means for determining whether the calibration slope could be significantly influenced by these samples alone. d Excluding specimens with mixed mineralogy. e Linear regressions through previously published data are all recalculated here using GraphPad Prism software (Zar, 1984) so that they are directly comparable to the new mollusk data presented here, and as a result may have slight differences from the slopes and intercepts given in original publications at the third or fourth decimal place. All regressions are on data that include an acid digestion temperature correction where appropriate (Passey et al., 2010). Errors are given as 1 s.e.  (Zar, 1984) to linear regressions through other previously published calibration studies datasets. Calculations are done with values on the absolute reference frame (ARF). The table displays the ANCOVA p value and whether the two slopes being compared are statistically different; (Y) = Yes, (N) = No. In this case we consider a p value < 0.05 as indicating statistically significant differences between the two slopes. b Excluding the five specimens of Laternula elliptica and Adamussium colbecki (which are specimens from the coldest Antarctic environments) as a means for determining whether the calibration slope could be significantly influenced by these samples alone. c Excluding specimens with mixed mineralogy. d Includes coral data from Ghosh et al. (2006) (but excludes Red Sea Porites), and data from Ghosh et al. (2007), Came et al. (2007), Tripati et al. (2010), Thiagarajan et al. (2011). See Table S1 for values for these data. mollusks. We also show that changing solution pH and carbonate chemistry should not be a confounding factor in the interpretation of bivalve based 47 or δ 18 O measurements, at least in the taxa studied, and that there is no significant mineralogical difference between calcite and aragonite. The errors in slope and intercepts for linear regression lines given in Table 4 highlight that successful calibration of the carbonate "clumped isotope" thermometer is dependent on having large datasets. For example, a linear regression through the initial inorganic calcite calibration dataset  has much larger uncertainties than a calibration line based on all the published biogenic calibration data from our Biogeosciences, 10, 4591-4606, 2013 www.biogeosciences.net/10/4591/2013/ laboratory due to having fewer datapoints. However we have shown statistically that the uncertainties in these calibration lines cannot alone explain the difference between our bivalve mollusk calibration line and other data produced in our laboratory, which (i) highlights that empirical calibrations of the carbonate clumped isotope paleothermometer are vital for each type of material and experimental setup, and (ii) suggests that initial papers showing close similarity of some biogenic materials to the inorganic calcite calibration of Ghosh et al. (2006;Tripati et al., 2010;Thiagarajan et al., 2011) should not be assumed to hold in all cases. We also note that after this manuscript was published as a discussion paper, another study of brachiopods and mollusks in a different laboratory also reported a similarly shallow slope (Henkes et al., 2013), although as these measurements were conducted using a very similar methodology to that used in our study the similarity between our calibration slopes does not entirely resolve the possible methodological differences between calibration studies described below. There are two possible explanations that are immediately apparent for the differences between calibration lines generated from different materials in our laboratory. First, the bivalve mollusk data presented here was obtained using the automated online sample reaction system described in Passey et al., 2010, whereas the in-depth calibration studies of corals, foraminifera and coccoliths were conducted using offline reactions with cryogenic and gas chromatography cleanup steps performed manually (Passey et al., 2010;Tripati et al., 2010;Thiagarajan et al., 2011). The calibration study on bioapatite  was conducted on the automated system, but it did not examine specimens grown at temperatures lower than ∼ 24 • C and so would not necessarily have resolved a difference in slope that would be most apparent at low temperatures. Therefore we must consider the possibility that an experimental effect, such as fractiona-tion of gases in either offline or online systems, or an effect due to the differences in acid digestion temperature between the two systems (25 • C for the offline reactions, 90 • C for the automated systems, which is presently addressed using a correction of 0.08 ‰ on the Caltech intralab reference frame) is not being correctly accounted for. Evidence against an experimental artifact from these two sources comes from the broadly comparable results that have been generated in different labs that use different systems for purifying CO 2 gas and different acid digestion temperatures as part of an interlaboratory comparison, which included measurements on a cold-water coral standard in four laboratories that consistently yielded a 47 value in the range of 0.78-0.80 ‰ on the absolute reference frame (Dennis et al., 2011). Additionally a number of applied studies using the automated sample preparation system have found that the calibration of Ghosh et al. (2006) generally yields plausible results including on modern specimens where we have good controls over growth temperature (e.g., Passey et al., 2010;Eagle et al., , 2011Finnegan et al., 2011;Csank et al., 2011;and Suarez et al., 2011). Nevertheless most applied studies have focused on samples formed at temperatures of 20 • C or more, and so there is a possibility that experimental differences such as small amounts of gas fractionation or equilibration during sample gas purification could preferentially affect samples with heavier 47 values (> 0.75 ‰). This is an area that should be explored in the future. Another possibility is that there are variations in acid digestion fractionation factors for samples of different isotopic composition or of different mineralogy, and whilst the aragonitic cold-water coral did not show this effect (Dennis et al., 2011) it would be useful to check if this is the case in other materials.
A second possible explanation for the differences in calibration lines revolves around fundamental differences in shell calcification in bivalve mollusks compared to other www.biogeosciences.net/10/4591/2013/ Biogeosciences, 10, 4591-4606, 2013 biogenic carbonates that could result in "vital effects" on 47 . In this scenario the closer match of deep sea corals to the calibration of Ghosh et al. (2006) at cold temperatures actually reflects the expression of a small kinetic isotope effect in all of these materials, one that is not found in mollusks. The data from foraminifera at cold temperatures is relatively sparse, with some samples from the Arctic Ocean showing deviations from the Caltech inorganic calcite calibration and so are analogous to the mollusk data presented here, but other datapoints from specimens from slightly warmer environments fall closer to the calibration of Ghosh et al. (2006), and Tripati et al. (2010). This highlights the relative paucity of data from carbonates forming at low temperatures and this is an obvious area to focus on in future calibration studies.
Bivalve mollusks frequently precipitate their shells close to equilibrium with respect to oxygen isotopes, with maximum deviations typically in the range of 0.5 ‰ (e.g., Horibe and Oba, 1972;Romanek and Grossman, 1989;Grossman and Ku, 1986;Barrera et al., 1994;and Wanamaker et al., 2006). This is in contrast to deep-sea corals, which exhibit nonequilibrium δ 18 O values of up to 4-5 ‰ in some cases (e.g., Adkins et al. 2003). Therefore we might expect that bivalve mollusk derived 47 values may also record close to equilibrium values, unless there is a source of biological fractionation of 47 in bivalves that has not yet been identified but hypothetically could be linked to mollusk specific mechanisms of biomineralization such as the use of organic templates for carbonate precipitation (Weiner and Dove, 2003;Addadi et al., 2006). If it was the case that mollusks are recording close to equilibrium values, the calibration of Ghosh et al. (2006) would have to include a kinetic isotope effect that fortuitously matches "vital effects" in previously published biogenic data from a temperature range of 0-10 • C that falls close to the inorganic calcite values. Finally, we note that even though a mineralogical difference between calcite and aragonite could not be resolved in our dataset it is still possible that very subtle mineralogical effects do exist and these effects contribute to the variability in measured 47 values. Larger datasets may be required to constrain this possibility with more certainty.
In conclusion, if the experimental effects described above can be either ruled out or better constrained, we will be able to say more about whether there may be small biological fractionations in 47 that differ between corals, foraminifera, and bivalves, and why these fractionations are most apparent at cold temperatures.