NIST RM 8301 Boron Isotopes in Marine Carbonate (Simulated Coral and Foraminifera Solutions): Inter‐laboratory δ11B and Trace Element Ratio Value Assignment

The boron isotopic ratio of 11B/10B (δ11BSRM951) and trace element composition of marine carbonates are key proxies for understanding carbon cycling (pH) and palaeoceanographic change. However, method validation and comparability of results between laboratories requires carbonate reference materials. Here, we report results of an inter‐laboratory comparison study to both assign δ11BSRM951 and trace element compositions to new synthetic marine carbonate reference materials (RMs), NIST RM 8301 (Coral) and NIST RM 8301 (Foram) and to assess the variance of data among laboratories. Non‐certified reference values and expanded 95% uncertainties for δ11BSRM951 in NIST RM 8301 (Coral) (+24.17‰ ± 0.18‰) and NIST RM 8301 (Foram) (+14.51‰ ± 0.17‰) solutions were assigned by consensus approach using inter‐laboratory data. Differences reported among laboratories were considerably smaller than some previous inter‐laboratory comparisons, yet discrepancies could still lead to large differences in calculated seawater pH. Similarly, variability in reported trace element information among laboratories (e.g., Mg/Ca ± 5% RSD) was often greater than within a single laboratory (e.g., Mg/Ca < 2%). Such differences potentially alter proxy‐reconstructed seawater temperature by more than 2 °C. These now well‐characterised solutions are useful reference materials to help the palaeoceanographic community build a comprehensive view of past ocean changes.

levels of atmospheric carbon dioxide (CO 2 ) during intervals of climate change in the geological past, to better understand and anticipate potential future changes to the ocean/ atmosphere system and the impacts on marine bio-carbonate organisms. To this end, attention has been focused on the development of ocean pH proxies, with the differences in boron isotopic composition (expressed as δ 11 B SRM951 , relative to National Institute of Standards and Technology (NIST) Standard Reference Material (SRM) 951 Boric Acid, in parts per thousand; δ 11 B = (R( 11 B/ 10 B) Sample /R ( 11 B/ 10 B) SRM951 -1)) of modern and fossilised marine calcifying organisms (e.g., coral and foraminifera) showing particular promise (Hemming and Hanson 1992).
In addition to boron, the trace element composition of marine carbonates is vital to understanding past ocean characteristics and composition (e.g., Algeo and Rowe 2012). For this reason, palaeoceanographers routinely measure carbonate molar ratios of Mg/Ca (sometimes also paired with Li/Ca) and Sr/Ca as proxies for temperature (Beck et al. 1992, Anand and Elderfield 2003, Case et al. 2010; Al/Ca, Mn/Ca, Fe/Ca, and Ba/Ca to assess seawater metal content Boyle 1989, Guzmán andJiménez 1992) and sample diagenesis/contamination (Barker and Greaves 2003); Cd/Ca to estimate nutrient content (Rickaby and Elderfield 1999); and U/Ca to assess carbonate ion saturation states and calcification rates (Russell et al. 2004, DeCarlo et al. 2015. In this way, trace element proxy data can provide a holistic view of past and present ocean-climate interactions. Before palaeoceanographic interpretation can be made from any δ 11 B SRM951 or trace element dataset, rigorous assessment of uncertainty is required. Initial interlaboratory comparison exercises measuring boron isostopes in natural materials revealed large discrepancies in results across laboratories (>> AE1‰; Gonfiantini et al. 2003, Aggarwal et al. 2009). Despite much analytical improvement since then, recent inter-laboratory studies still report significant inter-laboratory disagreement for both boron isotope ) and trace element (Hathorne et al. 2013) measurements. Hence, well-characterised boron isotopic reference materials in a carbonate matrix are urgently needed to assess the accuracy and precision of carbonate δ 11 B SRM951 measurements through the entire procedural treatment: from dissolution of carbonate, ionic separation of boron from the carbonate matrix, to the final δ 11 B SRM951 measurement. To date, only two authentic carbonate boron isotope reference materials exist that have been value-assigned by the palaeoceanographic community: JCp-1 (Porites coral) and JCt-1 (Giant Clam) (Okai et al. 2002, Inoue et al. 2004, Hathorne et al. 2013) (see companion inter-laboratory study by Gutjahr et al. (2020 in press). While many carbonate geochemistry laboratories routinely use these materials in-house, recent changes to regulations by Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES) now restrict the distribution of both biogenic carbonates due to their animal origins. NIST has therefore supplemented these with NIST RM 8301 Boron Isotopes in Marine Carbonate (Simulated Coral and Foraminifera Solutions), hereafter abbreviated as NIST RM 8301, providing new solutionbased inorganic carbonate boron reference materials synthetically produced to imitate typical coral (NIST RM 8301 (Coral)) and foraminiferal (NIST RM 8301 (Foram)) δ 11 B SRM951 and trace element contents (Li, B, Na, Mg, Al, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Rb, Sr, Mo, Ag, Cd, Sn, Sb, Cs, Ba, Nd, W, Pb, U). The solutions comprising NIST RM 8301 will not only ensure quality control of procedural chemistry postdissolution across laboratories, but have the added benefits of having a high mass fraction of B to ensure stability during storage (NIST RM 8301 (Coral) ≈ 7.1 μg g -1 and RM 8301 (Foram) ≈ 1.9 μg g -1 ), free from any shipping restrictions associated with shipment of protected species, and are in abundant supply (5000 bottles each) to ensure long-term continuity of measurements into the future. Here, we present δ 11 B SRM951 and trace element data from NIST and other leading boron isotope laboratories to assign values to NIST RM 8301 and evaluate analytical performance between these laboratories.

Methodology NIST RM 8301 reference material production
The production of NIST RM 8301 reference materials is summarised in Figure 1. Six kilograms of high-purity powdered calcium carbonate (ACS 337 calcium carbonate, chelometric material; GFS Chemicals, OH, USA; Lot # C474630) was dissolved in TraceMetal ™ Grade concentrated nitric acid (Fisher Scientific, MA, USA) while cooled in a water bath. The dissolved solution was passed through Whatman ® GF/F 0.7 μm microfibre filters (142 mm diameter; 0.42 mm thickness; Cat # 1825-142) to remove small quantities (< 3 g total mass) of residual solids (trace oxides, organics, and/or silicates) that formed during the dissolution process. After filtration, the solution was transferred into two acid-cleaned 20 L HDPE carboys: one for NIST RM 8301 (Coral), and one for NIST RM 8301 (Foram). The matching carbonate matrix of these two solutions means that they share a common RM identifier (8301). However, with their contrasting boron isotope and trace element composition (see below) we distinguish between the two levels of this RM as 'Coral' and 'Foram'. The end user is not necessarily 7 8 expected to use both solutions in tandem as part of their quality control procedure; rather, they should choose the solution most appropriate to their sample type of interest.
Single-element trace element solutions were added in suitable quantities to each solution of the RM to reproduce the trace element to calcium ratios typical of a dissolved coral aragonite and foraminiferal calcite. The stock solution used for each element and the masses added to each RM solution are given in Table 1. Note that these values are given for reference to show the approximate element/Ca ratios in the solutions. Gravimetric concentrations are provided for documenting the production procedure, and the masses in Table 1 do not account for concomitant trace elements contained within the starting powdered carbonate and single-element standards, or for ions leached during dissolution and filtration. Inter-laboratory consensus information values for the trace element composition of the RMs are reported later in the manuscript.
The boron isotope spike customised and added to each trace element-doped dissolved carbonate RM solution was a mixture of NIST SRM 951a boric acid (δ 11 B = 0‰; 11 B/ 10 B = 4.0437) and a > 99% enriched 11 B spike (Trace Sciences) to give B/Ca and 11 B/ 10 B ratios typical of coral (B/Ca ≈ 550 μmol mol -1 ; δ 11 B SRM951 ≈ 25‰) and foraminifera (B/Ca ≈ 150 μmol mol -1 ; δ 11 B SRM951 ≈ 15‰) ( Table 1). NIST SRM 951a was dissolved in boron-free highpurity water while the more recalcitrant 11 B-enriched metal was microwave-digested in Optima ™ concentrated nitric acid and hydrogen peroxide (Fisher Scientific) using quartz vessels in an Anton Parr Multiwave 3000 Microwave Reaction System. Following addition of this final boron isotope spike, the resultant solutions were diluted with boron-free water (resistivity: 18.2 MΩ cm) to yield a total volume of 20.2 l in each carboy and final calcium concentrations of approximately 50 mg ml -1 and nitric acid content of approximately 3 mol l -1 . Gravimetric calibration and measurement by inductively coupled plasma-mass spectrometry (ICP-MS) revealed final calcium mass fractions of 49.7 and 51.3 mg g -1 for NIST RM 8301 (Coral) and NIST RM 8301 (Foram), respectively (see Analytical techniques section). The boron isotope spike weights given in Table 1 are for reference to show the target boron mass fraction and 11 B/ 10 B ratio of the solutions. As above, gravimetric preparation values are expected to differ from actual B/Ca and 11 B/ 10 B ratios in the RM solutions as small amounts of boron are added from the original carbonate and/or leached during dissolution. Each NIST RM 8301 solution was well mixed and aliquoted sequentially through preconditioned peristaltic pump tubing into 5000 acid cleaned 4 ml HDPE screw top vials for production. All vial cleaning and RM

9
dispensing were carried out in the NIST Biorepository ISO Class 5 clean room at the NIST Charleston Laboratory.

Inter-laboratory comparison outline
For the NIST contribution to the inter-laboratory study, nine vials of each RM were selected for measurement of δ 11 B SRM951 and trace element composition. Additional participant laboratories include LSCE, GEOMAR, Yale and Universities of Southampton, St Andrews, and National Cheng Kung that were selected based on their active research on boron isotopes in marine carbonates. Each laboratory was assigned a laboratory number at random to protect data anonymity. Participants were sent three vials (selected by random number generator) of each simulant of NIST RM 8301 and asked to make at least three separate boron isotopic measurements of each vial, providing a relevant citation detailing their matrix separation and analytical protocol of choice. The high nitric acid content in NIST RM 8301 (approximately 3 mol l -1 ) makes these solutions unsuitable for negative thermal ionisation mass spectrometry (e.g., Hönisch and Hemming 2004, B. Hönisch pers. comm.). All laboratories in this study therefore opted to employ multi-collector (MC)-ICP-MS analytical techniques.
Participants were asked to provide supplementary data for each vial for their typical carbonate trace element suite (element ratioed to calcium); however, these analyses were optional. In order for us to present a wide array of useful trace element ratios in these solutions, we did not ask laboratories to produce a detailed uncertainty budget beyond a combined estimation of analytical reproducibility and variability between the vials measured. For this reason, an expanded uncertainty is not provided for trace element values presented here. Consequently, they cannot be used to establish metrological traceability to the mol or kg, but are nonetheless useful guide values determined by inter-laboratory consensus. Trace element data were provided by NIST, LSCE, GEOMAR, Southampton and St. Andrews. Additional trace element data were provided by the Universities of Bristol and Oxford; however, these two laboratories measured fewer individual vials. Laboratories providing trace element data were again assigned a random number to protect data anonymity. All laboratories used either quadrupole or sector-field ICP-MS instrumentation to analyse the solutions, matrix-matching samples and calibration solutions with calcium mass fractions of between 25 and 100 µg g -1 . Note that the laboratory numbers used for boron isotopic comparison in Figure 2 do not correspond to the same laboratory number reporting elemental ratios in Figures 4 and 5.

Analytical techniques
Analytical techniques used by the participant laboratories are summarised in Tables 2 and 3. Details of the analytical approach used by NIST are given below.
Boron isotope measurements at NIST: Analyses at NIST followed protocols previously described in Foster (2008), Rae et al. (2011) and Foster et al. (2013). The exact volume of RM used varied between column batches, yet in all cases more than 100 ng of B was targeted for each analysis (typically 15 and 50 μl of NIST RM 8301 8 0 (Coral) and NIST RM 8301 (Foram), respectively). Each aliquot from the individual RM 8301 vials was buffered in 2 mol kg -1 sodium acetate to 0.5 mol kg -1 acetic acid (20:1 buffer to sample ratio) before boron was separated from the carbonate matrix using 20 μl micro-columns containing IRA 743 boron-specific anionic exchange resin (Kiss 1988). All boron must be recovered from columns to avoid isotopic fractionation; therefore, following elution of the boron fraction, an additional elution was checked to ensure > 99% of sample boron was recovered.
At NIST, the purified boron samples were diluted to a boron mass fraction of approximately 100 ng g -1 for analysis on a Nu Plasma II MC-ICP-MS concurrently with NIST SRM 951a Boric Acid Isotopic Standard at matrixmatched mass fractions of B and acid concentration (0.5 mol l -1 HNO 3 ). An on-peak zero was acquired as a 60 s acid blank measurement before each sample.
Immediately after sample wash in, a peak centre was performed using the 11 B mass before both boron isotopes were collected on H8 and L4 Faraday cups. Sample data were acquired as one block of sixty cycles each with an integration time of 2 s. Data acquisition was followed by a 2 min wash between samples to ensure minimal sample carry over (i.e., blank intensity < 1% of sample). Note that all laboratories in this study used either SRM 951 or SRM 951a to correct for mass bias, using similar sample-standard bracketing techniques.
The precision of δ 11 B SRM951 results in the NIST laboratory was assessed by repeat measurements of boric acid standard BAM ERM-AE121 (certified value, 19.9‰ AE 0.6‰; Vogl and Rosner 2012) during analytical runs. In addition, assessment of the full powdered sample processing methodology was performed using the carbonate reference material JCp-1 (robust mean including robust Six total procedural blank measurements were made at NIST alongside samples in this study (mean absolute blank of 116 pg of boron). These blanks are small relative to the sample size (< 0.09% of sample boron) resulting in minimal impact on δ 11 B SRM951 results (i.e., < 0.1‰); hence, a total procedural blank correction was not applied.
Elemental determination at NIST: Determinations were carried out on an Element XR sector-field ICP-MS and broadly followed the protocol of Marchitto (2006) to yield Li/Ca, B/Ca, Na/Ca, Mg/Ca, Al/Ca, Mn/Ca, Fe/Ca, Sr/ Ca, Cd/Ca, Ba/Ca, Nd/Ca and U/Ca ratios. Sample aliquots were diluted to an equal mass fraction of Ca (80 µg g -1 ), and every three samples were bracketed by a matrixmatched, gravimetrically prepared, primary calibrant that was traceable to NIST 3100 series single-element solutions. Measured intensities (counts per second) of each individual  sample were blank corrected using blank acids measurements before and after each sample. Typical precision (1s) for these element/calcium ratios is < 4% based on repeat measurements (n = 38) of an in-house matrix-matched gravimetric solution.
Reference value assignment for δ 11 B SRM951 and supplemental data Reference values and expanded uncertainties for δ 11 B SRM951 in NIST RM 8301 (Coral) and NIST RM 8301 (Foram) published in the final NIST Report of Investigation for NIST RM 8301 were determined from all inter-laboratory study data using the DerSimonian-Laird analysis method within the NIST Consensus Builder (Koepke et al. 2017). Value assignment such as this, by consensus approach involving multiple participants, is not conducive to obtaining a fully comprehensive uncertainty budget (e.g., Rosner 2012, Geilert et al. 2019). The DerSimonian-Laird method was therefore chosen to account for 'dark uncertainty' (unaccounted sources of uncertainty among laboratories) as the reported data only included uncertainties related to replication. Because no comprehensive uncertainty budgets were reported, traceability to the SI cannot be established. For this reason, we do not advocate the use of NIST RM 8301 for calibration purposes (e.g., using as a bracketing calibrator) or establishing metrological traceability. Consensus element/calcium ratios published in the final NIST Report of Investigation for NIST RM 8301 as non-certified information values, are based on the overall means calculated from the mean values reported for each material from each participant laboratory.

Results and discussion
Boron All boron isotope data for NIST RM 8301 collected by the seven contributing laboratories are shown in Table 4 and summarised in Table 5 and Figure 2. These data show Expanded uncertainties estimated using this approach are small (< AE0.2‰) in comparison with the laboratory mean values (> AE0.3‰; 2s). DerSimonian-Laird consensus values are weighted with uncertainty surrounding the consensus value decreasing roughly in proportion to the square root of the number of results being combined. Therefore, this approach can underestimate dark uncertainty in cases such as this where the number of laboratories is small. However, we note the alternative hierarchical Bayesian approach available in the NIST Consensus Builder offers similar results (AE 0.20‰ expanded uncertainty at 95% confidence), suggesting that DerSimonian-Laird analysis provides a reasonable estimate of the consensus value and its uncertainty.
Non-certified values such as this are a best estimate of the true value; however, they may reflect only the measurement repeatability and may not include all sources of uncertainty (May et al. 2000). An example of unaccounted uncertainty could be from the boron isotope ratio of NIST SRM 951 bracketing standard itself that was used by all laboratories in this study (NIST SRM 951 10 B/ 11 B absolute abundance ratio of 0.2473 AE 0.0002). While the δ 11 B value of NIST SRM 951 of 0‰, by definition, carries no uncertainty, heterogeneities between different batches of NIST SRM 951 used in each laboratory could potentially have an impact on absolute reported δ 11 B values. We consider this source of uncertainty and its impact on interlaboratory results small; however, because of the close agreement of boric acid standard BAM ERM-AE121 (no matrix removal step required) measurements provided by five out of seven laboratories in this study.
Inter-laboratory measurement discrepancy and calculated pH: Boron isotope measurements in marine carbonates are commonly used to calculate seawater pH values using the simplified relationship described in Equation (1) (Zeebe and Wolf-Gladrow 2001); see also full expression in Rae (2018): where α B is the fractionation factor between the two major species of boron in seawater (boric acid and borate; 1.0272; Klochko et al. 2006), pK B * is the dissociation constant for boric acid in seawater, and δ 11 B borate and δ 11 B sw are the respective boron isotopic ratios of the borate ion (thought to be incorporated into marine carbonates) and total boron in seawater (39.61‰; Foster and Pogge von Strandmann 2010). Although the inter-laboratory range in mean δ 11 B SRM951 values reported from each laboratory was similar (approximately 0.5‰) for each RM, if treated as a true carbonate sample (assuming a sensitivity of δ 11 B SRM951 to pH equal to borate ion; Equation 1), seawater pH values calculated using these δ 11 B SRM951 values result in a range in pH of 0.02 pH units for NIST RM 8301 (Coral) and a larger 0.10 pH unit range for NIST RM 8301 (Foram). This difference largely reflects the non-linear relationship between carbonate δ 11 B and pH (Equation 1) but also highlights the significant potential differences in calculated pH from boron isotope data produced in different laboratories, hence the need for reference materials like NIST RM 8301 to help tighten pH reconstructions using this proxy.
Boron isotope values and NIST RM 8301 homogeneity testing: Although the solutions comprising NIST RM 8301 were dispensed from carboys into the 4 ml distribution vials and capped as soon as possible, the sequential nature of the dispensing through single lengths of tubing has the potential to introduce heterogeneities across a reference material batch. Despite such potential bias, δ 11 B SRM951 values for each vial from all participant laboratories agree well across the entire seventeen racks (each rack contained 289 vials) of each reference material (Figure 3). Vial δ 11 B SRM951 values from the start (first four racks; first 24% of vials; n = 9) and end (last four racks; last 24% of vials; n = 8) of the dispensing sequence vary by < 0.03‰ in both RM 8301 (Coral) solution (with a typical processed aliquot size of ≈ 10 µl), and NIST RM 8301 (Foram) solution (with a typical processed aliquot size of ≈ 50 μl). The individual vial differences are indistinguishable at the quoted precision. Sampling of the population was limited by the 8 5 labour-intensive nature of boron isotope data collection. Shapiro-Wilk and F-tests show data subsets are, respectively, normally distributed and of similar variance (p > 0.05). A two-sample t-test comparing available data could therefore be performed that showed no statistically significant difference (p > 0.05) could be resolved between mean δ 11 B SRM951 values for vials at the start and end of the batch and speak for homogeneity of the reference materials.
Stability of reference materials is also of great importance; however, as yet insufficient time has elapsed for a rigorous investigation into NIST RM 8301 stability. Release of these solutions for use by the community, without lengthy delay, was considered a priority. We note that many interlaboratory participants have continued to use these solutions for more than 1 year and have obtained similar results to those that they report here. Full stability testing will be performed by NIST at a later date, with results made available on the NIST website (https://www.nist.gov/srm) along with recommendations of storage conditions. NIST RM 8301 (Foram) reproducibility issues: Multiple laboratories using the well-established ion exchange column matrix separation technique reported poorer reproducibility when using the lower B mass fraction NIST RM 8301 (Foram) solution. As an example, NIST results for this solution varied by AE 0.55‰ (2s) across all replicates in contrast with AE 0.31‰ variation across replicates of the higher boron mass fraction NIST RM 8301 (Coral) solution, despite a similar mass of boron being loaded. Even the lower boron concentration solution NIST RM 8301 (Foram) has a relatively large boron mass fraction at 2 μg g -1 . Therefore, we consider true heterogeneities within a single vial solution to be unlikely. Furthermore, the impact of the respectively, normally distributed and of similar variance (p > 0.05). A parametric two-sample t-test was therefore applied, which revealed no statistically significant (p > 0.05) difference between mean δ 11 B SRM951 across each batch of RM.

6
total procedural blank is also considered to be negligible (pg level) compared with the high mass of boron used per analysis (> 100 ng; cf. foraminifera samples often < 10 ng of B; Foster 2008). However, 11 B/ 10 B ratios are known to fractionate strongly on ion exchange columns unless near complete recovery of boron (> 99%) is achieved (Lemarchand et al. 2002). Boron can be lost at two stages during matrix separation (i) during sample loading and (ii) during sample elution steps; both being potentially detrimental to the isotopic integrity of the sample. While elution tails at NIST were all found to be low (less than 0.11 ng of B) and boron recovery from the columns was considered complete (> 100 ng loaded; tail < 0.1% of sample), boron loss during loading was not accurately quantified; thus, small amounts of B loss cannot be discounted.
Significant loss of boron during loading of foraminiferal calcite samples using these techniques has not been documented previously; however, the acid concentration in these RMs (21% by volume; 3 mol l -1 HNO 3 ) is considerably higher than that commonly used to dissolve marine carbonates (e.g., 0.5 mol l -1 HNO 3 ). A greater volume of buffer is therefore required to raise the pH of the solution before loading onto columns (20:1; cf. 2:1 Foster 2008). Rapid loading of high volume (> 1 ml) samples will cause solutions to pass quickly through the resin, reducing sample-resin interaction times and NIST reference values and 95% expanded uncertainties (reported at the bottom of the table in bold) were determined from reported laboratory data using the DerSimonian-Laird analysis method within the NIST Consensus Builder (Koepke et al. 2017). A reference value is a non-certified value that is the best estimate of the true value; however, the value may reflect only the measurement repeatability and may not include all sources of uncertainty (May et al. 2000).
8 7  Mean values and percentage relative standard deviation based on the replicate analysis of n vials are shown. Where only one vial was measured within laboratory, precision is based on replicate analysis of the same vial.

9
potentially resulting in incomplete and variable boron adherence on columns. Other participating laboratories loading only 300 μl maximum per column and step did not encounter these problems.
Occurrences of some boron loss during loading may account for the slightly lower mean δ 11 B SRM951 value for NIST RM 8301 (Foram) measured at NIST (14.24‰) with the lowest replicate measurement more than 0.8‰ lower than the inter-laboratory consensus reference value of 14.51‰. This effect was less pronounced or absent in analyses of the higher B mass fraction in NIST RM 8301 (Coral) (NIST mean only 0.03‰ lower than consensus reference value) where smaller volumes of RM were needed to achieve ≈ 100 ng of B, keeping total loading volumes low (≈ 300 μl). We note, however, that removal of the NIST RM 8301 (Foram) vials for which NIST δ 11 B SRM951 values were most variable (> AE 0.5‰, 2s; 3A2, 4K11, 17A2, 17G15) increases the NIST mean δ 11 B SRM951 value of NIST RM 8301 (Foram) by only 0.06‰ and thus has little impact on the overall consensus values.
Although the influence on final consensus values is considered small, compiled data here suggest the potential impact of incomplete boron loading onto columns on individual replicates of NIST RM 8301 (Foram) is potentially large, so RM 8301 users should evaluate their routine B separation methods carefully before processing the material. This highlights the importance of adequate buffering of samples prior to column loading and suggests that loading of samples low in B (e.g., dissolved foraminifera) should be performed slowly, in sequential small volumes (e.g., 100 μl), to maximise initial boron adhesion to the resin.

Trace elements
Trace element determinations were contributed by seven participant laboratories (Table 6). As laboratories were only asked to run their typical method for carbonate samples, not all laboratories were able to provide data for all trace elements that were added during the preparation of these RMs. We therefore only present means of trace element values provided by four or more laboratories as summarised in Figures 4 and 5.
For all analytes investigated here, the variance of measurements across vials within each laboratory (relative standard deviation up to AE 9% for Nd/Ca and AE 16% for Fe/Ca, but typically AE 2%; 1s) was better than the overall variance of elemental ratios in these RMs reported among laboratories (up to AE 21% for Al/Ca, but typically AE 6%; 1s). Similarly, poor reproducibility of values across laboratories was observed in the inter-laboratory study for carbonate powders JCp-1 and JCt-1 (Li/Ca, B/Ca, Ba/Ca, U/Ca, > AE 10% (1s); Hathorne et al. 2013). This struggle for accuracy, particularly for challenging to measure elements like Al, further highlights the need for reference materials for this type of analyses. While analytical offsets across laboratories for any trace element have the potential to bias interpretation of palaeoceanographic results, here we consider the impact on commonly used temperature proxies in corals and foraminifera Mg/Ca, Li/Mg (i.e., (Li/Ca)/ (Mg/Ca)), and Sr/Ca that typically reproduced well within each laboratory.
The ambient seawater temperature in which biogenic carbonates were formed is routinely estimated using these trace element ratios (given in mmol mol -1 ) and calibration equations such as:  2). Thus, we find inter-laboratory measurement discrepancy yields potentially more than 4°C inaccuracy in reconstructed temperature, even before calibration uncertainty is incorporated. Inter-laboratory 9 1 discrepancies therefore far outweigh the approximate AE 1.5°C calibration uncertainty typically quoted for this proxy (Case et al. 2010, Montagna et al. 2014, Fowell et al. 2016, Cuny-Guirriec et al. 2019. Laboratory mean Sr/Ca ratios for NIST RM 8301 (Coral) vary by only AE 1.7% (1s); however, using Equation (3), such discrepancy results in more than 3°C difference in calculated seawater temperatures. We note however that Sr/Ca values in NIST RM 8301 (Coral) (≈ 8.1 mmol mol -1 ) are slightly lower than typical coral values (≈ 9 mmol mol -1 ); thus, reconstructed temperatures based on this calibration (≈ 35°C) are higher than those typically found in the surface ocean. Similarly, for NIST RM 8301 (Foram), laboratory mean Mg/Ca values vary by AE 5.4% (1s); therefore, using Equation (4), implied seawater temperatures would vary by more than 3°C (maximum 17.3°C; minimum 14.2°C). This again exceeds typically quoted calibration uncertainty for temperature estimated from foraminiferal Mg/Ca (AE 1.2°C; Gray and Evans 2019).
This exercise highlights the large uncertainties that are potentially introduced to palaeoceanographic proxy reconstructions by discrepancies between laboratories and underestimated measurement uncertainties. With good within laboratory precision, but little accuracy with respect to known reference materials, data produced by a single methodology will yield results that elucidate robust relative palaeoceanographic changes (e.g., temperature rise or fall); however, absolute target values may be inaccurate. Such inaccuracies become especially significant in cases where paired carbonate trace element values are used to assess temperature-induced temporal changes in pK B * for δ 11 B-based pH or pCO 2 records (e.g., Martınez-Botı et al. 2015). Commutability of data is particularly important in calibration studies that set the ground work for proxy application and where data are often compiled from multiple laboratories (e.g., Montagna et al. 2014, Fowell et al. 2016, Marchitto et al. 2018, Cuny-Guirriec et al. 2019).

Summary
We used boron isotope data compiled from seven leading research laboratories to assign respective NIST reference values and 95% expanded uncertainties for δ 11 B SRM951 of 24.17‰ AE 0.18‰ and 14.51‰ AE 0.17‰ to the new marine carbonate reference materials NIST RM 8301 (Coral) and NIST RM 8301 (Foram). These reference values were assigned by consensus approach and by their nature do not necessarily account for all sources of uncertainty. However, these solutions were found to be homogeneous across the batches of vials and had characteristics suitable for use as analytical quality controls.
Trace element data were provided by seven participants, and inter-laboratory consensus information values for key trace elements in marine carbonates that include Li/Ca, B/Ca, Na/Ca, Mg/Ca, Al/Ca, Mn/Ca, Fe/Ca, Sr/Ca, Cd/Ca, Ba/Ca, Nd/Ca and U/Ca are given here. For all analytes in question, reported trace element values for RMs were considerably more variable across laboratories than reported precision based on a single methodology. The simulated marine carbonate solutions comprising NIST RM 8301 will not be subject to CITES limitations restricting distribution of authentic biogenic material and will help minimise analytical artefacts caused by sample pre-treatment in respective laboratories (e.g., oxidative cleaning and/or dissolution; Gutjahr et al. 2020). NIST RM 8301 (Foram) and NIST RM 8301 (Coral) are therefore valuable tools for evaluating the quality of marine carbonate geochemical analyses, thus improving confidence in palaeoceanographic interpretation.
Marine Carbonate (Simulated Coral and Foraminifera Solutions), available on the NIST SRM website, https:// www.nist.gov/srm. Any mention of commercial products is to specify adequately the analytical procedures used. It does not imply recommendation or endorsement by NIST or that the products mentioned are necessarily the best available for the intended purpose. Handling of NIST RM 8301 solutions will result in possible exposure to nitric acid, and appropriate personal protective equipment should be used.