Clumped isotope composition of cold-water corals: A role vital

The carbonate clumped isotope thermometer is a promising tool for determining past ocean temperatures. It is based on the temperature dependence of rare isotopes ‘clumping’ into the same carbonate ion group in the carbonate mineral lattice. The extent of this clumping eﬀect is independent of the isotope composition of the water from which carbonate precipitates, providing unique advantages over many other paleotemperature proxies. Existing calibrations of this thermometer in cold-water and warm-water corals suggest clumped isotope ‘vital eﬀects’ are negligible in cold-water corals but may be signiﬁcant in warm-water corals. Here, we test the calibration of the carbonate clumped isotope thermometer in cold-water corals with a recently collected and well characterised sample set spanning a range of coral genera ( Balanophyllia , Caryophyllia , Dasmosmilia , Desmophyllum , Enallopsammia and Javania ). The clumped isotope compositions ( D 47 ) of these corals exhibit systematic dependences on their growth temperatures, conﬁrming the basis of the carbonate clumped isotope thermometer. However, some cold-water coral genera show D 47 values that are higher than the expected equilibrium values by up to 0.05 ‰ (equivalent to underestimating temperature by (cid:1) 9 (cid:1) C) similar to previous ﬁndings for some warm-water corals. This ﬁnding suggests that the vital eﬀects aﬀecting corals D 47 are common to both warm- and cold-water corals. By comparison with models of the coral calciﬁcation process we suggest that the clumped isotope oﬀsets in these genera are related to the kinetic isotope eﬀects associated with CO 2 hydration/hydroxylation reactions in the corals’ calcifying ﬂuid. Our ﬁndings complicate the use of the carbonate clumped isotope thermometer in corals, but suggest that species- or genus-speciﬁc calibrations could be useful for the future application of this paleotemperature proxy.


INTRODUCTION
Scleractinia or 'stony corals' precipitate aragonite skeletons and range from small, solitary corals to hermatypic or 'reef-building' colonial corals that can support extremely biodiverse ecosystems. Many species living in shallow seas host photosynthesising algae, which provide the corals with carbon for metabolisation Porter et al., 1984). Such corals are known as zooxanthellate, whereas corals that lack algal symbionts are known as azooxanthellate. Around 90% of azooxanthellate scleractinian corals live in cold/dark water and as such are often referred to as cold-water or deep-water corals (Roberts et al., 2009). The term 'cold-water coral' is also used to refer to animals from two further orders (Gorgonacea and Anthoathecata) that can have differing skeletal mineralogy. In this study we use the term 'cold-water coral' to refer specifically to azooxanthellate scleractinian corals (Roberts et al., 2006;Roberts and Cairns, 2014).
Cold-water corals are found in all major ocean basins at depths from the surface to deeper than 4000 m (Stanley and Cairns, 1988;Roberts et al., 2006;Cairns, 2007). Their skeletons can be collected from the seafloor and precisely dated using uranium-series techniques (Cheng et al., 2000). As such, they are a promising target as an archive of past ocean dynamics and properties, particularly for regions lacking sediment-core climate records .
Given the important role of temperature in governing seawater density, circulation and heat transport, numerous geochemical proxy-methods have been developed and applied to reconstruct past ocean temperature (eg. McCrea, 1950;Alibert and McCulloch, 1997;Barker et al., 2005;Rueggeberg et al., 2008;Thiagarajan et al., 2014). However, many paleotemperature proxies used successfully in other archives (eg. d 18 O and Mg/Ca in foraminiferal tests) cannot be straightforwardly applied in cold-water corals (eg. Smith et al., 2000;Adkins et al., 2003;Gagnon et al., 2007;Case et al., 2010). This complication is due to the presence of 'vital effects', i.e. biological processes that cause deviations of coral geochemical compositions from thermodynamic equilibrium values. For example, the stable isotope compositions (d 18 O and d 13 C) of cold-water coral skeletons are known to be depleted relative to the expected carbonate-seawater equilibrium values by up to 6-12‰ (Adkins et al., 2003;Lopez Correa et al., 2010;Lutringer et al., 2005;Mortensen and Rapp, 1998;Smith et al., 2000;Spiro et al., 2000;Rollion-Bard et al., 2003Marali et al., 2013). The magnitude of these stable isotope vital effects varies even within a coral skeleton and is thought to be related to kinetic and/or pH-driven processes during coral calcification (eg. McConnaughey, 1989;Adkins et al., 2003). A proxy that is neither affected by biological mediation nor requires knowledge of the seawater isotopic composition would be a valuable tool for the paleoceanographic community.
A promising paleotemperature proxy in this regard is the clumped isotope composition of coral skeletal carbonate (Ghosh et al., 2006;Thiagarajan et al., 2011;Saenger et al., 2012). It is based on the homogeneous isotope equilibrium within the carbonate lattice, and thus can constrain carbonate formation temperature without knowledge of the isotopic composition of the water from which the carbonate grew. An example of such an isotope equilibrium is: Under thermodynamic equilibrium, 13 C and 18 O in the carbonate lattice have a tendency to clump into the same carbonate ion group, leading to a relative enrichment in the abundance of the 13 C 18 O 16 O 2 2À isotopologue over that which would be expected if all isotopes were randomly distributed. The magnitude of this enrichment varies as a function of the carbonate formation temperature and can thus be used as a geothermometer.
In practice this enrichment is quantified by measuring the abundance anomalies of the mass 47 isotopologues (mostly 13 C 18 O 16 O) in CO 2 derived from phosphoric acid digestion of carbonates, which is proportional to the enrichment of the 13 C 18 O 16 O 2 2À isotopologue in the carbonate mineral (Ghosh et al., 2006;Guo et al., 2009b): where R 47, R 46 , and R 45 are the measured abundance ratios of masses 47, 46 and 45 relative to mass 44, and the denominator terms R * are the expected ratios if all isotopes were randomly distributed (i.e. if the sample had the stochastic distribution of isotopologues based on its measured d 18 O and d 13 C values) (Eiler and Schauble, 2004;Huntington et al., 2009). Ghosh et al. (2006) first calibrated the carbonate clumped isotope thermometer by precipitating synthetic calcites in the lab, yielding a temperature relationship of D 47 = 0.0592 Â 10 6 /T 2 À 0.02 (T in kelvin). Subsequently, Dennis and Schrag (2010) precipitated calcites and found a D 47 temperature dependence of D 47 = 0.0337 Â 10 6 / T 2 + 0.247, significantly different from that of Ghosh et al. (2006). We refer to these contrasting calibrations as the Ghosh calibration line and the Dennis calibration line throughout this study. The discrepancies between these calibration lines are not resolved by converting the measured D 47 values to the absolute reference frame (ARF) (Dennis et al., 2011). Recently Zaarur et al. (2013) repeated similar precipitation experiments to Ghosh et al. (2006) and found a temperature dependence close to the Ghosh calibration line for calcite and aragonite. In contrast, weaker temperature dependences of the clumped isotope thermometer in aragonite similar to the Dennis calibration line have also since been reported (Kim et al., 2010;Defliese et al., 2015). The causes of the different calibration slopes -which also exist in the calibration of some biogenic carbonates, e.g. brachiopods (Henkes et al., 2013;Came et al., 2014;Petrizzo et al., 2014) -are a matter of current debate within the clumped isotope community. It has been postulated that these inter-laboratory differences are related to differences in the analytical protocols in different labs, e.g. the extent of equilibration between the produced CO 2 and the water associated with the digestion acid (Came et al., 2014;Defliese et al., 2015).
Despite the complications described above, calibrations of the clumped isotope thermometer in many (although not all) biogenic carbonates appear to agree closely with the Ghosh calibration line (Came et al., 2007;Tripati et al., 2010;Thiagarajan et al., 2011;Zaarur et al., 2011;Eagle et al., 2013). However, with the exception of Eagle et al. (2013), these studies used the same analytical methods as Ghosh et al. (2006). Confidence in the application of the carbonate clumped isotope thermometer in cold-water corals increased with the finding that the relationship between their clumped isotope compositions and growth temperature was indistinguishable from the Ghosh calibration line (Thiagarajan et al., 2011). This result highlighted the possibility that the processes generating large vital effects in coral d 18 O and d 13 C did not affect clumped isotope fractionations in cold-water corals. The carbonate clumped isotope thermometer has since then been applied to the cold-water coral Desmophyllum dianthus, to explore past ocean temperatures (Thiagarajan et al., 2014). In contrast, the clumped isotope compositions of some warm-water scleractinian corals were shown to be systematically offset from the Ghosh calibration line by $0.04‰ (Saenger et al., 2012). Such clumped isotope vital effects in warm-water corals did not seem to be related to the presence or absence of algal symbionts (Saenger et al., 2012). In this study we increase the sample density of the calibration of the clumped isotope thermometer in cold-water corals with a new, recently-collected and well-characterised sample set spanning a range of coral genera. We test and correct for inter-laboratory differences in clumped isotope measurements, allowing direct comparison of our results with results from previous studies, and we further evaluate the possible causes of clumped isotope vital effects in scleractinian corals.

Coral samples
Corals analysed in this study were either collected during recent cruises to the Atlantic and Southern Oceans (Table S1), or have been measured before (Thiagarajan et al., 2011). Samples from the tropical Atlantic Ocean were collected during the JC094 cruise (Oct-Nov 2013, Robinson, 2014 by Remote Operated Vehicle (ROV) from the flanks of seamounts or fracture zone escarpments, with the depth of collection ranging from 200 m to 2500 m. These samples were all living immediately prior to collection. Coral samples from the Southern Ocean were collected during three cruises to the Drake Passage (NBP-0805, NBP-1103 and LMG-0605) (Burke et al., 2010;Margolin et al., 2014). These samples were collected using dredge or trawl deployments, on the flanks of seamounts, fracture zones and the continental margin between 330 m and 1879 m. The corals were not alive when collected, but were determined to be less than 1000 years old using uranium-series and/or reconnaissance radiocarbon dating (Burke and Robinson, 2012;Margolin et al., 2014). If necessary, coral samples were bleached and then rinsed with fresh water while on board ship to remove organic tissue and were stored at room temperature.

Temperature
Temperature estimates at each sample site were derived from at least two (up to three) different sources: (1) annual average temperature data from the Levitus94 (Levitus and Boyer, 1994) database from within 0.5°latitude and longitude, and extrapolated to the depth of sample collection; (2) ROV CTD data at the sample collection site (JC094 cruise only, including the maximum and minimum ROV CTD temperature recorded within ±3 m depth of the collection site) and (3) rosette CTD data taken near the sample site and extrapolated to the depth of sample collection (in some cases two rosette casts were available from different cruises). For most cases, data from the available sources agreed to within ±0.4°C and we take this to be the uncertainty on our estimates of coral growth temperature (Table S1). The exceptions were for samples JC094-B0457-Daslm-001 and JC094-B0468-Daslm-001 collected from within the Atlantic thermocline. The rosette CTD measured temperatures were 1°C and 2°C warmer than those measured using the ROV for the two samples respectively, while the Levitus94 temperatures agreed with the ROV data to within 0.2°C. The differing temperature estimates here are likely due to changes in the thermocline depth with proximity to the seamount and/or diurnal changes. It is also possible that seasonal variation in the thermocline is important for this site. This uncertainty does not alter our conclusions. In all figures and tables we report the temperatures measured by the CTD mounted on the ROV or rosettes.

Carbon and oxygen isotopic compositions of seawater
The isotopic compositions of seawater were determined based on waters collected in Niskin bottles during all cruises, which were attached to either the rosette system (12 collections per deployment) or the ROV (JC094 cruise only, up to 5 collections per deployment).
For d 13 C measurements, unfiltered seawater samples were collected in 250 ml acid-cleaned and ashed glass bottles from the Niskin bottles using acid-cleaned silicone tubing, first rinsing and then overfilling the bottle by at least 50%. The sample, leaving head-space, was poisoned with 50 ll of saturated mercuric chloride solution, and the bottle was then sealed with a plastic screwcap lid and o-ring. The d 13 C of dissolved inorganic carbon (DIC, 1 standard deviation S.D. = 0.1‰) was analysed at the University of California (Irvine) for tropical Atlantic samples (Gao et al., 2014), and at the National Ocean Sciences Accelerator Mass Spectrometry facility (NOSAMS) of Woods Hole Oceanographic Institution for Southern Ocean samples. The speciation of DIC in the seawater was calculated based on the seawater alkalinity and total DIC concentration estimated from the GLODAP database and measured temperatures, using the CO2SYS program (van Heuven et al., 2011). The d 13 C of the dissolved HCO 3 À was then calculated based on the measured d 13 C DIC and published carbon isotope fractionation factors (Mook et al., 1974). The calculated d 13 C HCO À For d 18 O measurements, unfiltered 60 ml subsamples were sealed using rubber stoppers and aluminium crimp seals in glass bottles and stored in cool ($4°C), dark conditions for transportation. For the tropical Atlantic samples, the d 18 O measurements were conducted using the IsoPrime100 with Aquaprep system at the NERC Isotope Geosciences Laboratory, with typical precisions of ±0.02‰ (1 S.D.). For Southern Ocean samples, the d 18 O measurements were conducted using the Isoprime Multiprep system at Tulane University, with precisions of ±0.24‰ (1 S.D.). In the tropical Atlantic, measurements of d 18 O had poorer spatial coverage than measurements of d 13 C. We therefore elected to calculate the ocean atlasderived seawater d 18 O for sample sites from this cruise, following the methods in Thiagarajan et al. (2011). A regional relationship between d 18 O and salinity was established using the NASA Global Seawater Oxygen-18 Database -v1.21 (Schmidt et al., 1999). The more finely gridded Levi-tus94 salinity database was then used to estimate d 18 O near the sample site. Where possible, the calculated d 18 O values were compared with the values derived from measurements of water samples at the respective sites. The maximum offset between the ocean atlas estimate and in-situ measured d 18 O was 0.1‰, about equal to the error on the ocean atlas estimate (Thiagarajan et al., 2011) and small compared to the range of d 18 O variability we observed amongst different corals ($4‰). Such small uncertainties in seawater d 18 O shall not affect our conclusions. The d 18 O of the seawater samples are reported relative to Vienna Standard Mean Ocean Water (VSMOW).

Stable isotope measurements of coral skeletons
The surfaces of each coral sample were scraped with a dremel tool to remove residual organic material (Thiagarajan et al., 2011). Aliquots of approximately 30 mg were cut from each specimen with a diamond blade, rinsed with 18 MX cm water and ground to powders/fine grains with a pestle and mortar. For seven solitary corals, subsamples were taken from both the theca and septa of the skeleton to assess any systematic differences in the clumped isotope composition between these two important skeletal architectures (Table 2). We also carried out an experiment on NBS-19 and two coral samples to evaluate the impact of oxidative cleaning on the measured D 47 (Table 3). For this experiment, crushed samples were subjected to water rinses, a 15 min wash in 30% H 2 O 2 + 1 M NaOH (mixed in the ratio 1:1) at 40°C, and a rinse lasting 1-2 min in (0.1 M) perchloric acid, similar to methods designed to vigorously remove organic material in preparation for uranium-series dating of cold-water coral skeletons (Cheng et al., 2000). Both cleaned and uncleaned powders of these samples were analysed. Except for samples employed in this experiment, all other samples in this study were not chemically cleaned prior to analysis, consistent with the procedure of previous clumped isotope studies of corals (Thiagarajan et al., 2011;Saenger et al., 2012).
Multiple aliquots ($4 mg each) of each coral sample were weighed into silver capsules and added to an autosampler in a random order. This randomising procedure was used in order to reduce the possibility of systematic instrumental effects affecting repeat measurements of a single sample. Each aliquot of a single powdered sample was generally analysed within a single measurement session, over the course of 3-4 days. An aliquot of one of two in-house carbonate standards (NBS-19 or 102-GC-AZ01) was analysed after every $7 coral aliquots. Between 3 and 5 aliquots were analysed for the majority of coral subsamples, resulting in 8 or more replicate measurements of many coral specimens (where the theca and septa of the same specimen were sampled).
Clumped isotope analyses were performed at Woods Hole Oceanographic Institution (WHOI), within two measurement sessions (May-July 2014 and September 2014), on a Thermo Scientific MAT-253 mass spectrometer coupled to a custom-built automated acid reaction and gas purification line. The purification line is similar to the one described by Henkes et al. (2013), except both heated gases and equilibrated gases were introduced by first freezing them in liquid nitrogen (LN 2 ) as opposed to directly injecting them into a helium stream.
Aliquots of carbonate samples and standards were digested in 103% H 3 PO 4 (q = 1.92 g/cm 3 ) at 90°C and evolved CO 2 was purified by passing through several cryogenic traps (À78°C dry ice/ethanol slush, and liquid nitrogen trap) and a custom-built packed 60 cm-long gas chromatograph column (Porapak Q, 50-80 mesh) held at À20°C. Purified CO 2 gases were then expanded into the sample bellow of the mass spectrometer and analysed at a bellow gas pressure corresponding to a signal of 12 V on the Faraday cup measuring mass 44 of CO 2 . A bottle of Oztech CO 2 (d 13 C VPDB = À3.63‰, d 18 O VSMOW = +25.04‰) was used as the working reference gas during isotope measurements. Each measurement sequence consisted of six acquisitions, and each acquisition consisted of 9 cycles of sample-reference comparison with 20 or 26 s of integration time.
The d 18 O and d 13 C values of carbonates were calculated using the working reference gas and then normalised by reference to the NBS-19 standard analysed in the same measurement session. CO 2 gases equilibrated at 1000°C and 25°C (i.e. 'heated gases' and 'equilibrated gases') were processed using the same purification line and analysed regularly to convert all clumped isotope values to the absolute reference frame (Dennis et al., 2011). An acid digestion fractionation of 0.092‰ was applied to normalise the clumped isotope composition of all carbonate standards and samples to acid extractions at 25°C (Henkes et al., 2013). The analytical precision of each measurement session was ±0.016‰, ±0.13‰ and ±0.06‰ (1 S.D.) for D 47, d 18 O and d 13 C respectively, based on repeated measurements of the two in-house carbonate standards. Where n subsamples of a coral were measured, standard errors (1 S.E.) on isotope ratios were calculated by dividing the 1 S.D. of those measurements by p n. In addition to isotope measurements made at WHOI, homogenised powders of five coral specimens were shared between WHOI and the California Institute of Technology (Caltech) to evaluate potential inter-laboratory differences in clumped isotope measurements. These samples were from four different coral genera (Table 4).

Processing of coral clumped isotope data from previous studies
Clumped isotope compositions of coral skeletons have been reported in previous studies, e.g. Thiagarajan et al. (2011) and Saenger et al. (2012). Saenger et al. (2012) presented their data in the ARF based on a secondary transfer function constructed from three in-house standards (a Carrara marble, corn CO 2 , and CO 2 equilibrated with water at 25°C). The data from Thiagarajan et al. (2011) were not originally reported in the ARF. Previous attempts to convert the Thiagarajan et al. (2011) data were based on measurements of heated gases and NBS-19 made around the time of the study, and used the 'tertiary reference frame' approach in Dennis et al. (2011), also often used to convert the original Ghosh et al. (2006) data (Dennis et al., 2011;Saenger et al., 2012;Eagle et al., 2013).
Recently, four of the corals measured in Thiagarajan et al. (2011) have been re-sampled and analysed at Caltech in the ARF. Together with another cold-water coral that has previously been measured in the ARF (Dennis et al., 2011), these data enable us to provide an updated conversion of the Thiagarajan et al. (2011) data to the ARF. Our approach is the same as the 'tertiary reference frame' approach above, with the exception that we use the re-analysed corals as additional carbonate standards to better-constrain the conversion. In addition, rather than a single conversion for data generated during multiple sessions, we use a session-by-session conversion for each of the seven sessions in 2008 in which cold-water corals were measured for Thiagarajan et al. (2011). For each session, we constructed an empirical transfer function based on the known ARF D 47 values for heated gases, NBS-19 and/or re-analysed corals run during that session. If a coral was not re-analysed in the ARF, this function was used to convert the clumped isotope data reported in Thiagarajan et al. (2011) to the ARF. For corals that were re-analysed in the ARF, we use the measured ARF value. The updated coral data, including the standards/corals used for the conversions, are presented in Table S4.
For each coral analysed, we estimated the expected d 18 O and d 13 C values if they had formed in isotope equilibrium with seawater, based on seawater data for the sample site (T, d 18 O sw , d 13 C HCO À 3 ) and published equilibrium isotope fractionation factors (Grossman and Ku, 1986;Romanek et al., 1992). The d 18 O and d 13 C of the majority of samples were depleted with respect to the expected equilibrium values, consistent with findings from previous studies (Fig. 1, Table 1, Table S1) (Adkins et al., 2003;Lutringer et al., 2005;Mortensen and Rapp, 1998;Smith et al., 2000;Rollion-Bard et al., 2003. Oxygen and carbon isotope depletions (i.e. Dd 18 O and Dd 13 C) are known to be linearly correlated across most parts of a cold-water coral skeleton, with Dd 13 C/Dd 18 O slopes typically around 2.3 and intercepts ranging from À5‰ to À1‰ (Adkins et al., 2003;Lutringer et al., 2005;Mortensen and Rapp, 1998;Smith et al., 2000;Rollion-Bard et al., 2003. The isotopic compositions of corals measured in this study fall within the region covered by these previously-determined slopes and intercepts, with the magnitude of the disequilibrium effects varying amongst different coral genera (Fig. 1). For example, the majority of data for the genera Caryophyllia and Javania fall within 1‰ of the expected equilibrium d 18 O values, whereas the majority of data for Balanophyllia, Dasmosmilia and Enallopsammia are more than 1‰ from the expected equilibrium values, with the species D. dianthus straddling these two extremes. However, previous micro-sampling measurements of Caryophyllia and Desmophyllum corals show that both genera can contain the entire range of isotopic depletion observed for cold-water corals (Smith et al., 2000;Adkins et al., 2003). It therefore seems probable that the differences we observe between genera are due to sampling biases, rather than the true extent of d 18 O and d 13 C disequilibrium.

Clumped isotope compositions
3.2.1. Effect of oxidative cleaning We compared measurements of both cleaned and uncleaned powders from one carbonate standard  and two coral samples (Table 3). Each powder was measured 3 times, with the exception of the uncleaned NBS-19, which was measured 12 times. The maximum difference in D 47 between cleaned and uncleaned powders was 0.016‰ for the coral JC094-B0561-JaClm-001 (Javania) and the minimum was 0.002‰ for the coral JC094-B0561-Carlm-001 (Caryophyllia). T-tests indicated that there was no significant difference between the D 47 of cleaned and uncleaned powders for any of the samples (p = 0.10 for NBS-19, p = 0.20 for JC094-B0561-JaClm-001 and p = 0.63 for JC094-B0561-Carlm-001). Two previous studies of biogenic aragonites, employing only H 2 O 2 treatment, also indicated that oxidative cleaning does not have a statistically significant effect on measured D 47 values (Table 3, Saenger et al., 2012;Eagle et al., 2013). Therefore coral samples in this study were not cleaned prior to isotope analysis with the exception of shipboard bleaching, water rinses and physical scraping, consistent with the procedure of Thiagarajan et al. (2011) andSaenger et al. (2012).
The magnitudes of our observed offsets (0.027-0.046‰) relative to the mean values reported in previous studies are close to the range of inter-laboratory difference observed in previous studies, which are up to 0.031‰ for NBS19 (0.057‰ for Carrara marble) and 0.023‰ for 102-GC-AZ01 (Dennis et al., 2011;Rosenheim et al., 2013;Zaarur et al., 2013;Tang et al., 2014). The exact cause of these offsets could be related to the differences in our analytical procedure and/or the phosphoric acid digestion fractionation factor (0.092‰) we adopted, and warrants further investigation. In this study, we corrected our clumped isotope data based on the linear function required to bring the values of these in-house standards measured at WHOI into agreement with their previously reported values (see details below). This linear function is a D 47 -dependent empirical transfer function, equivalent to that described in Dennis et al. (2011), but applied after the 'heated gas', 'equilibrated gas' and acid digestion corrections that transfer data into the ARF. Therefore, this additional correction means that our reported clumped isotope data are no longer strictly in the ARF.
Note that the clumped isotope value of the 102-GC-AZ01 standard reported by the Caltech laboratory in Dennis et al. (2011) was greater than the inter-lab average value (0.724‰ vs. 0.713‰). We therefore corrected our inter-laboratory comparison dataset in two ways, by assuming two different values for the 102-GC-AZ01 standard, i.e. 0.713‰ and 0.724‰. We evaluated the robustness of these corrections based on comparisons with measurements made at both WHOI and Caltech on five homogenised coral powders (Section 2.3.). Excellent agreement was observed between the WHOI and Caltech measurements of these five coral powders, when we corrected our data using the 0.724‰ value for 102-GC-AZ01, i.e. the value specifically from the Caltech laboratory (Dennis et al., 2011) (Fig. 2, Table 4). Such agreement highlights the usefulness of using carbonate standards to remove inter-lab bias, which can exist despite converting all measurements into the ARF. Given these results, and our aim of comparing our values with previous measurements made at Caltech (Thiagarajan et al., 2011), all data measured at WHOI presented in this study were corrected using the 0.724‰ value for 102-GC-AZ01 (Table S2). Data corrected using the inter-lab average value (0.713‰) are presented in the Supplementary data (Table S3).
To compare our results with previous results of warmwater corals reported by Saenger et al. (2012) from the Yale laboratory, we used their results reported in the ARF (Saenger et al., 2012). These authors also shared samples with Caltech, achieving consistent results between the two labs. Therefore, the warm-water coral results in that study should also be comparable to our corrected results.

Clumped isotope compositions of cold-water corals
Our measured D 47 values ranged from 0.797‰ to 0.875‰ for corals covering a temperature range of 2.2-17.4°C (Tables 1 and 2, Fig. 3). We include detailed sample provenance and the temperature and seawater chemistry estimates from each different source in the Supplementary material (Table S1). Raw data of clumped isotope analyses, including all carbonate and gas standards' values, are also presented in the Supplementary material (Tables S2 and  S3).
We find that, while an overall temperature dependence is observed, variations in D 47 at any single temperature are greater than found by Thiagarajan et al. (2011), probably due to the larger number of individual specimens and genera analysed here. For example, the D 47 of 15 corals collected from temperatures between 3 and 5°C vary by 0.06‰, equivalent to a 12°C range in estimated temperature using the calibration line of Ghosh et al. (2006). If data from Saenger et al. (2012) are included, this range increases to 0.08‰. This range of D 47 is significantly greater than our measurement uncertainties ($0.006-0.008‰ 1 S.E. for samples with multiple replicates, Section 2.3.). In contrast, we found no significant difference in D 47 (<0.02‰) between the septa and theca for any of the 7 corals studied ( Table 2).   Romanek et al. (1992) (for d 13 C) and Grossman and Ku (1986) (for d 18 O). Shown as symbols are the bulk coral isotope measurements from this study, Thiagarajan et al. (2011) and Saenger et al. (2012). Also shown as lines are selections of data from Adkins et al. (2003) and Smith et al. (2000). Each line represents the best fit through multiple measurements on a single coral specimen, and is not extrapolated past the maximum and minimum d 18 O measured in the specimen. Zero offsets from the expected equilibrium for both d 13 C and d 18 O are shown as dashed black lines. For simplicity, errors bars for the stable isotope measurements are not shown in the figure, but are around ±0.6‰ 2 S.E. for d 13 C (based on the error in Romanek et al., 1992) and ±0.2‰ 2 S.E. for d 18 O for the data symbols.

Inter-genus variations in cold-water coral D 47
The combined coral D 47 data from WHOI, Caltech and Yale highlights the previously observed first-order dependence of corals' clumped isotope compositions on their growth temperatures (Fig. 3). In accordance with previous clumped isotope studies of coral skeletal carbonates, the absolute values of our data lie more closely to the Ghosh calibration line than calibrations with shallower slopes, such as the Dennis calibration line (Fig. 3). We therefore discuss our results by reference to the Ghosh calibration line in the following sections, assuming it represents the carbonate clumped isotope equilibrium, consistent with previous clumped isotope studies of coral skeletal carbonates (Thiagarajan et al., 2011;Saenger et al., 2012).
Note: two versions of the Ghosh line have been derived in the ARF (Fig. 3. Dennis et al., 2011;Eagle et al., 2013). Although both of these were based solely on the original data reported by Ghosh et al. (2006), they are offset by $0.01‰ due to differences in the ways in which clumped isotope measurement uncertainties were incorporated into the regression analyses. Specifically, Dennis et al. (2011) weighted their regression model based on the analytical uncertainties of each measurement, while Eagle et al. (2013) did not. In addition, Zaarur et al. (2013) recently performed additional carbonate precipitation experiments and derived a new calibration line by combining their new data with the original Ghosh et al. (2006) data. Zaarur et al. (2013) used residual-based iterative weighting to include the analytical uncertainties in their regression, resulting in a calibration line similar to that reported by Eagle et al. (2013). Although these different calibration lines lie within each other's confidence envelopes, the exact version of the calibration line we adopt may slightly alter our conclusions. Here, in the absence of a clear community agreed 'best practice', we adopt the Dennis et al. (2011) version of the Ghosh line in our following discussion, since it lies closest to the coral data.
In the absence of clumped isotope vital effects, we would expect our measured coral clumped isotope compositions to cluster evenly around the Ghosh calibration line, the assumed clumped isotope equilibrium line. In this ideal case, the offsets of measured coral D 47 from the Ghosh calibration line would follow a normal distribution of mean zero, with a standard deviation equal to the long-term standard deviation for our carbonate standards representing the analytical uncertainty (Fig. 4, solid red curve). In the following analyses comparing our results to this ideal case we use only data from this study, for which all values of individual replicate measurements were available.
The distributions of offsets within each genus were indeed consistent with normal distributions, but did not have mean values of zero. Instead, with the exception of Caryophyllia, they tended to be offset towards higher values, with mean offsets ranging from $0‰ to 0.04‰ (Table 5). The genera Balanophyllia, Enallopsammia and Dasmosmilia had higher offsets (0.03-0.04‰) than the genera Desmophyllum, Caryophyllia and Javania (À0.005-0.005‰).
With the exception of Caryophyllia, chi-squared (v 2 ) tests for variance suggested that the variances of the offset distributions for each genus were not distinguishable from that of multiple measurements on the carbonate standard 102-GC-AZ01 (Kanji, 2006) (Table 5). This result suggests that the spread in results for most genera can be attributed to analytical uncertainties. Samples from the genus Caryophyllia had a significantly greater variance than would be expected from analytical uncertainties (Table 5), which we suspect arises from the greater variety of species and morphologies sampled for this genus. Note also that, considering data from all genera, the measurement data have a broader, flatter distribution than the idealised case.
We used Kolmogorov-Smirnov (K-S) tests to evaluate the significance of the difference between the observed offset distributions for each genus and the idealised case -in this case the expected offset distribution around the Ghosh line given our measurement uncertainty (Kanji, 2006). Samples with greater numbers of replicates are likely to better reflect their true D 47 . We therefore weighted the statistical analysis according to the total number of measurements on each genus, i.e. by considering each replicate measurement as an individual sample. Such a procedure means that samples with more replicates are weighted more highly. The number of measurements on any one genus ranged from 12 (for Balanophyllia) to 65 (for Caryophyllia). For the genera Balanophyllia, Enallopsammia and Dasmosmilia, the measured coral clumped isotope compositions did not agree with the distribution of the idealised case, instead forming distributions that were significantly offset above it (>99% likelihood of a difference, Table 5). The same results were obtained using T-tests, confirming that the mean offsets for each of these three genera are significantly greater than zero. It should be noted that if other versions of the  Table 4). The dashed line denotes a 1:1 relationship. Errors bars are 1 S.E.
inorganic Ghosh calibration are used (eg. Eagle et al., 2013;Zaarur et al., 2013), all the genera studied here have mean positive offsets that are significant according to the K-S and T-tests, but that the genera Desmophyllum, Caryophyllia and Javania remain close to the lines. Further Kolmogorov-Smirnov tests were carried out to evaluate differences between different coral genera. These tests showed that there was no significant difference in the distributions of offsets between the genera Desmophyllum and Javania, and between Dasmosmilia and Enallopsammia. However, with the exception of these two pairs of genera, every coral genus had distributions of offsets that were significantly different to those of every other genus (Fig. 4b).
As described above, uncertainties in the Ghosh calibration line make it difficult to establish how close any particular genus lies to the inorganic equilibrium, or how large the mean offsets are for each genus. The existence of other inorganic carbonate calibration lines in the literature further complicates the interpretation of our results (Dennis and Schrag, 2010;Kim et al., 2010;Zaarur et al., 2013;Tang et al., 2014;Defliese et al., 2015;Kluge et al., 2015). For example, had we assumed that the Dennis calibration line represents the clumped isotope equilibrium, the calculated clumped isotope offsets would have been greater for all genera (e.g. $0.1‰ for Balanophyllia). However, the finding that different genera have different offsets from the equilibrium values is not dependent on the assumed clumped isotope equilibrium line (Fig. S1).
Our results thus provide robust evidence of inter-genus variations in clumped isotope vital effects in cold-water corals. Inter-genus differences in clumped isotope compositions similar to those that we report here for cold-water corals have also been observed in warm-water corals. For example, the genus Porites was found to have D 47 values systematically offset above the Ghosh calibration line, whereas the genus Astrangia was not (Saenger et al., 2012).
The presence of clumped isotope vital effects in certain cold-water coral genera provides further evidence that the presence or absence of symbionts is not the driver of such effects. In addition, the suggestion of Saenger et al. (2012) that corals exhibiting clumped isotope vital effects could be identified by the relationship between d 18 O and d 13 C in the skeleton does not appear to be true for cold-water corals, because all cold-water corals exhibit very similar slopes in their depletions of d 18 O and d 13 C (Fig. 1). Given our new data, it appears that the mechanism causing clumped isotope vital effects may be common to all scleractinian corals -regardless of whether they grow in cold-water or warmwater -and may act to different degrees within different genera.
Note that despite certain genera (Balanophyllia, Dasmosmilia, Enallopsammia and Porites) showing significant clumped isotope vital effects, a relationship between D 47 and temperature can still be observed within each of these genera. We suggest that with more work, species-or genus-specific temperature calibrations could be a useful approach for application of the carbonate clumped isotope thermometer in coral skeletons. However, before attempting to use such an approach, it is important to determine the causes of the observed clumped isotope vital effects. We examine possible mechanisms below, beginning with an outline of the coral calcification process that will inform further evaluations of the existing d 18 O, d 13 C and D 47 data.

Mechanisms of coral calcification and stable isotope vital effects
Corals are thought to precipitate carbonate from an extracellular calcifying fluid (ECF), separated from the surrounding seawater by the calicoblastic cell membrane (Fig. 5, Saenger et al., 2012). An enzymatic alkalinity pump, e.g. CaATPase, is thought to operate across the cell membrane. It removes two protons from the ECF for every  Table 5 Statistical analysis of the differences in the distributions of measured coral D 47 values for each coral genus, relative to an idealised distribution around the inorganic calibration line determined by Ghosh et al. (2006) À0.0051 ± 0.022 1.2 Â 10 À2 6.7 Â 10 À2 6.5 Â 10 À5 Ca 2+ pumped into it (Ip et al., 1991). This ionic exchange increases the pH and alkalinity of the ECF, causing DIC to speciate towards CO 3 2À . Accordingly the aragonite saturation state in the ECF increases, leading to enhanced aragonite precipitation. Two sources of carbon contribute to the calcification within the ECF; the CO 2(aq) diffused across the cell membrane, and seawater DIC from seawater leakage (Fig. 5, eg. Erez, 1978;Furla et al., 2000).
Within this framework of coral calcification, several mechanisms have been proposed to explain the stable isotope (d 18 O and d 13 C) vital effects observed in corals. For example, McConnaughey (1989) attributed the co-depletion of d 18 O and d 13 C in coral skeletons (Fig. 2) to kinetic isotope effects associated with CO 2 hydration/ hydroxylation reactions. The kinetic isotope effects associated with these reactions can be transferred from the resulting HCO 3 À to the carbonate mineral because the rate of coral calcification (i.e. carbonate precipitation) is usually fast.
In comparison, Adkins et al. (2003) showed that the depletions of d 18 O and d 13 C in coral skeletons are actually decoupled in the dense and rapidly precipitated centres of calcification (COCs), and therefore argued that both these decoupled depletions, and the coupled depletions in the rest of the skeleton could rather be explained by changes in the pH of the ECF and the extent of CO 2 diffusion across the cell membrane. Speciation of DIC in the coral ECF varies as a function of its pH, with higher pH leading to larger fractions of CO 3 2À relative to HCO 3 À . Under isotope equilibrium, CO 3 2À has a d 18 O that is depleted relative to HCO 3 À (McCrea, 1950;Usdowski et al., 1991;Usdowski and Hoefs, 1993). Therefore, if coral calcification makes use of both CO 3 2À and HCO 3 À to form CaCO 3 , as suggested for other carbonates (Zeebe, 1999), the higher proportion of CO 3 2À at high pH should lead to more depleted skeletal d 18 O than at lower pH (Adkins et al., 2003). Higher pH of the ECF would also promote CO 2 diffusion through the cell membrane, the depleted d 13 C of which would cause depleted skeletal d 13 C. These effects could thus result in the co-variance in d 18 O and d 13 C observed in the majority of the skeleton, while allowing for the observed limit to d 13 C depletion in the rapidly precipitated COCs caused by a limit in the capacity for CO 2 diffusion. However, this pH-based vital effect model requires the COCs to be precipitated at the highest pH of any part of the coral, which appears to be at odds with boron isotope data (Blamart et al., 2007;Rollion-Bard et al., 2011).

Possible causes of observed intra-and inter-genus variations
Recent studies of coral clumped isotope compositions have examined the potential correlations between the clumped isotope and oxygen isotope variations, focusing on several key processes involved in coral calcification, e.g. diffusion, CO 2 hydration/hydroxylation and equilibrium isotope fractionation amongst different DIC species (Guo et al., 2009a;Thiagarajan et al., 2011;Saenger et al., 2012;Tripati et al., 2015). Guo et al. (2009a) made first-order estimates about the effects of CO 2 hydration/ hydroxylation reactions on the carbonate clumped isotope composition and suggested that they could result in negative slopes of D(D 47 )/D(d 18 O) on the order of À0.05 to À0.01 respectively, where D(D 47 ) = D 47 -D 47(eq) and D(d 18 O) = d 18 O-d 18 O (eq) are defined as the deviations of the coral clumped isotope and oxygen isotope compositions from their expected equilibrium values. An estimate for the effect of CO 2 diffusion through the lipid-bilayer (based on Knudson diffusion) predicted a similar slope of À0.023 (Thiagarajan et al., 2011), although this diffusion model predicts smaller than observed depletions in d 13 C for a unit depletion in d 18 O (Thiagarajan et al., 2011;Saenger et al., 2012). Importantly, both of the above processes apply only to DIC derived from the CO 2 diffused across the cell membrane and not to DIC from the leaked seawater (ambient DIC). In the absence of other processes, pH-driven changes in the ratio of CO 3 2À to HCO 3 À in the ECF, and thus that incorporated into the skeleton, are not thought to have a large effect on the resulting D 47 , because the difference in the 13 C-18 O clumping effects between these species is quite small; (D 63 HCO À 3 )-(D 63 CO À 32 ) = 0.018‰ (Guo et al., 2008) or 0.033‰ (Hill et al., 2014) (Guo et al., 2012;Tripati et al., 2015). We refer to the pH effect modelled in Hill et al. (2014) for the remainder of the study, noting that it represents an intermediate estimate of the impact of pH change on carbonate D 47 .
We compare our coral isotope data with these preliminary theoretically predicted D(D 47 )/D(d 18 O) slopes to determine the likely processes governing the clumped isotope vital effects in scleractinian corals. Again we assume that the Ghosh calibration line, as recalculated in Dennis et al. (2011), represents the carbonate clumped isotope equilibrium in the following discussion. Our broad conclusions are not affected by this choice of calibration line, but we acknowledge that our detailed interpretations may change should a different calibration line prove more realistic (e.g. Figs. S2, S3).
As described above (Section 3.2.), the intra-coral variation of d 18 O and D 47 , e.g. septa vs. theca, are small in all the coral specimens examined (<1.5‰ for d 18 O, <0.02‰ for D 47 respectively). This result suggests that the D 47 ranges observed amongst our corals are not simply related to differences between these two macroscopic skeletal architectures. The coral JC094-B0468-Daslm-001 (Dasmosmilia sp.) showed the maximum difference in d 18 O between the septa and theca, and was associated with a positive D (D 47 )/D(d 18 O) slope of 0.006 ± 0.016. This slope is very similar to the slope predicted for pH-driven isotope fractionation (0.002-0.01), although both septal and thecal D 47 were offset significantly above the expected equilibrium values (by $0.016-0.026‰), lying at least partially within the region predicted for kinetic effects associated with CO 2 hydration and hydroxylation reactions (Fig. 6a). Given estimated slopes of À0.05, À0.01 and À0.02 for the CO 2 hydration and CO 2 hydroxylation reactions and diffusive processes respectively, and the uncertainties on the observed theca-septa D(D 47 )/D(d 18 O) slope, this sample suggests that the more extreme slopes predicted for kinetic effects may not explain the intra-coral variation here. However, the large uncertainties on the slope do not preclude less extreme kinetic isotope effects (i.e. CO 2 hydroxylation) acting within this sample. The sample JC094-B0561-Jav-001 (Javania) also had a relatively large difference in d 18 O between the septa and theca (0.9‰). In this case the slope between the two data points was negative (À0.020 ± 0.024), very similar to the slope for diffusional effects and lying within the region bound by the slopes of the CO 2 hydration and hydroxylation reactions (Fig. 6b). However, the large uncertainties do not rule out a slope equal to that of a pH-driven vital effect. Therefore, our isotope measurements within single specimens cannot isolate one mechanism causing intra-coral isotope vital effects at the macro-scale. Further detailed work could aim to investigate intra-coral variation further, including the impacts of micro-scale features, such as centres of calcification.
The greater range in d 18 O within and between different coral genera allows us to evaluate the mechanisms of observed vital effects further. We plot isotope data for those   Fig. 7a, and plot data for the other genera separately in Fig. 7b. In general, warm-water corals appear to exhibit a larger depletion in d 18 O than cold-water corals despite similar ranges in D(D 47 ). All genera appear to be consistent with near-zero slopes of D(D 47 )/D(d 18 O), but as previously highlighted, different genera have different offsets from the expected D 47 equilibrium, suggesting differences in the vital effect mechanisms between them. For example, a line of best fit through the Desmophyllum data (WHOI and Caltech) has a slope of À0.0013 ± 0.0024 and an intercept of 0.0066 ± 0.0045, suggesting variations in the pH of the ECF are a primary control on the variability of  Dennis et al. (2011) -andKu (1986) respectively. Dashed lines show the modelled trajectories of hydration/hydroxylation kinetic effects, diffusive effects and equilibrium precipitation at varying pH (Section 4.2.2). Error bars are 1 S.E. D 47 amongst the different individuals of this genus. The warm-water genus Astrangia could also be consistent with this pattern. However, given the uncertainties associated with the Ghosh line described previously, it is possible that the D 47 of the genus Desmophyllum may also be systematically above the clumped isotope equilibrium (Figs. S2, S3). Better constraints on the inorganic equilibrium line will be needed to fully address this question.
The cold-water genera Balanophyllia, Dasmosmilia and Enallopsammia seem to be inconsistent with pH being the sole driver of isotope variations, lying in the region broadly covered by the first-order models of the effects of CO 2 hydration/hydroxylation and/or diffusion (Fig. 7a). This observation is also true for the two data points for the genera Caryophyllia and Javania that have d 18 O depletions of >1‰ (Fig. 7b). The warm-water genus Porites exhibits similar D 47 offsets to the cold-water corals in Fig. 7a, ranging from $0-0.04‰, suggesting that a similar range of D 47 vital effects could be in operation for both groups of corals. As noted above, the diffusion model predicts smaller than observed depletions in d 13 C per unit depletion in d 18 O (Thiagarajan et al., 2011;Saenger et al., 2012). Therefore, kinetic effects associated with CO 2 hydration/hydroxylation reactions would appear to be the more likely candidate for causing the clumped isotope vital effects observed for these samples.
Of course, it is likely that multiple processes act in concert during coral calcification and simultaneously affect the isotope composition of coral skeleton, producing the whole range of isotope signatures observed, including the genusspecific differences observed here. If true, this may complicate the comparison between the experimental data and the model predicted D(D 47 )/D(d 18 O) slopes for each individual process acting independently. Future modelling work should aim to systematically take into account the effects of multiple processes in order to more accurately constrain the role of each process in determining the isotope composition of coral skeletons, and to test whether species-or genus-specific calibration of clumped isotope thermometers would be appropriate.

CONCLUSIONS
We have tested the calibration of the carbonate clumped isotope thermometer in cold-water scleractinian corals, using a new set of samples that includes six of coral genera. We validated our clumped isotope data by directly comparing measurements made in two different laboratories (WHOI and Caltech) on the same sample powders. We found excellent agreement between laboratories once the data were normalised with reference to the in-house carbonate standards, suggesting that this normalisation may be a useful way to remove inter-laboratory biases which were not removed by converting clumped isotope data into the absolute reference frame.
We show that some cold-water coral genera exhibit clumped isotope vital effects, with their D 47 values consistently higher than the expected equilibrium values, similar to findings for some warm-water corals (Saenger et al., 2012). The similarity between cold-water and warm-water corals with respect to the ranges of clumped isotope vital effects observed supports previous suggestions that the presence or absence of symbiotic algae is not a governing factor for clumped isotope vital effects in scleractinian corals. Instead, we suggest that these vital effects are most likely related to kinetic effects associated with CO 2 hydration/hydroxylation reactions within the extracellular calcifying fluid of the corals and are common to both warm-and cold-water corals. We found no significant difference in D 47 between the theca and septa of individual specimens, suggesting that macro-scale skeletal features are not related to variations in D 47 . Current uncertainty in the appropriate clumped isotope equilibrium line complicates the detailed interpretations of our results. However, the fact that the magnitudes of the clumped isotope offsets from expected equilibrium vary between different coral genera is not dependent on the choice of clumped isotope equilibrium line, and suggests that clumped isotope vital effects are present and expressed more in some genera than others. Further research into the differences between different inorganic calibration lines will be important in determining the presence and causes of clumped isotope vital effects in any given coral genus.
Our data complicate the use of the carbonate clumped isotope thermometer in corals, suggesting that it cannot be straightforwardly applied across all species of coldwater coral without introducing significant uncertainty into reconstructed temperatures. However, given the apparent temperature dependence of D 47 , even in genera exhibiting significant clumped isotope vital effects, we suggest that the use of species-specific calibrations could still be a useful approach. In order to determine the reliability of this thermometer back through time, further work should also test it in fossil samples by comparison with other, independent temperature proxies, such as the Li/Mg ratio in corals (Case et al., 2010;Hathorne et al., 2013;Raddatz et al., 2013;Montagna et al., 2014).