Refining Boron Isotopic Measurements of Silicate Samples by Multi‐Collector‐Inductively Coupled Plasma‐Mass Spectrometry (MC‐ICP‐MS)

Solution MC‐ICP‐MS is an established technique for high precision boron isotope measurement results (δ11BSRM 951) in carbonates, yet its application to silicate rocks has been limited. Impediments include volatilisation during silicate dissolution and contamination during chemical purification. To address this, we present a low‐blank sample preparation procedure that couples hydrofluoric acid‐digestion and low‐temperature evaporation (mannitol‐free), to an established MC‐ICP‐MS measurement procedure following chemical purification using B‐specific Amberlite IRA 743 resin. We obtain accurate δ11BSRM 951 values (intermediate precision ±0.2‰) for boric acid (BAM ERM‐AE121 19.65 ± 0.14‰) and carbonate (NIST RM 8301 (Coral) 24.24 ± 0.11‰) reference materials. For silicate reference materials covering mafic to felsic compositions we obtain δ11BSRM 951 with intermediate precision < ±0.6‰ (2s), namely JB‐2 6.9 ± 0.4‰; IAEA‐B‐5 ‐6.0 ± 0.6‰; IAEA‐B‐6 ‐3.9 ± 0.5‰ (2s). Furthermore, splits of these same reference materials were processed by an alternative fusion and purification procedure. We find excellent agreement between δ11BSRM 951 measurement results by MC‐ICP‐MS of the reference materials using both sample processing techniques. These measurement results show that our sample processing and MC‐ICP‐MS methods provide consistent δ11BSRM 951 values for low B‐mass fraction samples. We present new data from Mid Ocean Ridge Basalt (MORB) glass, documenting a range in δ11BSRM 951 from ‐5.6 ± 0.3‰ to ‐8.8 ± 0.5‰ (2s), implying some upper mantle δ11BSRM 951 heterogeneity.

Isotopic fractionation of boron during surface (e.g., Spivack et al. 1987) and subduction processes (Ishikawa andNakamura 1993, De Hoog andSavov 2018) followed by its deep recycling can strongly influence the present-day boron isotopic composition of the mantle (see review in Marschall 2018).The 11 B/ 10 B ratio (expressed as δ 11 B SRM 951 , relative to National Institute of Standards and Technology (NIST) Standard Reference Material (SRM) 951 Boric Acid Isotopic Standard, in parts per thousand: varies by $ 50‰ between chondrites and modern seawater (Foster et al. 2018), making it a powerful means to differentiate between reservoirs within the mantle with different contributions of recycled components, providing sufficient precision can be attained.
Characterising the boron isotopic composition of the upper mantle, as sampled by MORB is important, but challenging, as these depleted samples often contain < 1 μg g -1 boron mass fraction (Ryan and Langmuir 1993).A recent, comprehensive set of MORB measurement results by a refined secondary ionisation mass spectrometry (SIMS) approach reported an isotopic range of $ 4‰ (δ 11 B SRM 951 of -4.0 to -7.8‰ (Marschall et al. 2017)).However, large measurement uncertainties (typically AE2 to AE4‰; 2s), make it difficult to resolve possible source variability from measurement imprecision (Marschall et al. 2017).Advances in the use of boron isotope ratio measurements to investigate the composition of the Earth's upper mantle therefore requires improvement in measurement uncertainties (Foster et al. 2018).The determination of boron isotopes by solution MC-ICP-MS offers a means to obtain high signal to noise ratios given sufficient sample (10 to 50 ng of B, this study) (e.g., Wei et al. 2013, Zhu et al. 2021, Cai et al. 2021, Li et al. 2019).If this can be combined with control of instrumental mass bias via regular sample-standardbracketing, and chemical purification of boron with minimal isotopic fractionation and contamination, then higher precision measurement results are viable (Li et al. 2019).Boron isotope ratio measurements by MC-ICP-MS samplestandard-bracketing techniques do not require close matching of boron concentration in samples and isotope standards (i.e., samples in the range of 5 to 50 ng g -1 boron mass fraction can be bracketed by a 50 ng g -1 boron isotope standard (Foster 2008, Rae et al. 2011, Devulder et al. 2013)).Matrix impurity has not shown to result in significant deviation of δ 11 B SRM 951 results in experimental runs (e.g., Devulder et al. 2013).Yet, we find that monitoring buffer reagent signal intensity correlates to some extent with larger measurement uncertainty in our early experimental runs.
The extraction and separation of boron from silicate matrices is challenging compared with carbonate samples, and numerous techniques have been trialled, including pyrohydrolysis, acid digestion, and flux fusion in preparation for analysis (Nakamura et al. 1992, Spivack et al. 1987, Tonarini et al. 1997).The standard-sample bracketing approach used in MC-ICP-MS critically requires matrix removal to achieve similar measurement conditions (matrix effects) between samples and isotope standards.However, certain boron species are volatile, particularly when drying samples in aqueous/acidic solution (e.g., Xiao et al. 1997), and procedural blank contributions are also a significant concern during sample processing.Achieving sufficient purification of boron from silicate samples for bias free measurement results without isotopic fractionation during dissolution and chromatographic separation remains problematic.In previous studies, significant emphasis has been placed on the matrix removal through multiple ionexchange columns (e.g., Tonarini et al. 1997), however recent studies have sought to reduce the column procedure complexity (e.g., Li et al. 2019).Some studies have trialled the use of acids (e.g., HF) to dissolve silicate samples as an alternative to flux fusion techniques, with the aim of eliminating the need for expensive platinum crucibles and reducing the procedural blank.Reasonable agreement has been achieved (δ 11 B SRM 951 values within 0.7‰) between these flux fusion and hydrofluoric acid (HF) digestion experiments using thermal ionisation mass spectrometry (TIMS) (e.g., Rosner et al. 2003).These difficulties have at least in part delayed the application of the advantages of the MC-ICP-MS methodology, routinely applied to carbonates (Foster et al. 2008, Stewart et al. 2020, Gutjahr et al. 2020), to boron isotope measurement of silicate samples.
Here we present a hydrofluoric acid (HF) dissolution and dry down technique followed by a three-step chemical separation, involving no further sample dry down between columns.This protocol draws on previously documented insights on boron volatilisation and processing, which have been performed over the last three decades (Nakamura and Ishikawa et al. 1992, Xiao et al. 1997, Gaillardet et al. 2001).This chemical purification approach is coupled with a precise MC-ICP-MS technique (Foster 2008, Rae et al. 2011, Stewart et al. 2020), that has been calibrated against widely used carbonate reference materials (Foster et al. 2013, Gutjahr et al. 2020, Stewart et al. 2020).In light of the previously discussed poor to moderate reproducibility of boron isotopic measurements on silicate samples by different measurement techniques and independent sample preparation procedures (e.g., Gonfiantini et al. 2003), we compare MC-ICP-MS measurement results of sub-samples from the same batches of reference materials purified by our new HF digestion method and a flux fusion technique (Tonarini et al. 1997), traditionally used in conjunction with TIMS.Given notable differences in these sample preparation techniques, this test of consistency is a valuable assessment of the robustness of our approach.
We monitored intermediate precision, independent of column chemistry, using the BIG D boric acid solution (Foster et al. 2013) and BAM ERM-AE121 (Boric Acid in Water certificate value of 19.9 AE 0.6‰; (Vogl and Rosner 2012)).To test for possible fractionation during chemistry and sample preparation, we used the NIST RM 8301 (Coral).This solution, with a composition mimicking a dissolved coral, has well-characterised δ 11 B SRM 951 from a recent interlaboratory comparison exercise (certificate value of 24.17 AE 0.18‰, Stewart et al. (2020)).
For natural rock reference materials, we have focussed on three volcanic samples widely analysed in previous interlaboratory comparisons (Table 1).These include two basaltic samples: the Japanese Geological Survey JB-2 (which is a subduction zone basalt from the Izu volcanic arc (Japan)) and IAEA-B-5 (which is a basalt from Mount Etna (Italy)), distributed by the International Atomic Energy Agency for interlaboratory calibration of boron isotope measurements (Tonarini et al. 2003).Although basalts are our prime interest, we additionally analysed the International Atomic Energy Agency standard, IAEA-B-6, a natural rhyolite glass (obsidian) from Lipari Island (Italy).
To illustrate the utility of our approach we also present a small set of δ 11 B SRM 951 measurement results of MORB glasses, from three major ocean basins comprising three normal and one enriched MORB (Table 2) (Niu et al. 1999, Regelous et al. 1999, Robinson et al. 1996, Niu and Batiza 1994).These samples were previously analysed by SIMS (Marschall et al. 2017) and thus provide intercomparison of SIMS and MC-ICP-MS techniques.The sample set includes two samples with a Cl/K ratio of < 0.08, which is considered to be the threshold for MORB glass uncontaminated by seawater (Marschall et al. 2017).
Our Pacific E-MORB sample has higher Cl/K and so we compare this with a Pacific N-MORB sample, which also has elevated Cl/K (Table 2).As another more B-enriched basaltic sample, we also analysed one glassy ocean island basalt from Pitcairn Island (Table 2), which has been analysed previously by SIMS (Walowski et al. 2021).

Experimental
Measurement procedure for volcanic glass Shards of volcanic glass, free of visible inclusions, were hand picked under a binocular microscope, avoiding contaminant mineral phases or whole rock particles and clasts.Typically, 30 to 60 mg were separated for analysis.
The hand picked glass shards were then gently washed in de-ionised boron free water and treated ultrasonically for 5 min.The samples were then partly covered using para-film and left to dry in a low airflow box fitted with B-free HEPA filters.

Hydrofluoric acid digestion of silicate materials
Samples were prepared in a clean laboratory fitted with a B-free HEPA air filtration system (University of Bristol).Sufficient sample was weighed out to yield approximately 10 to 100 ng of boron and placed into a 7 ml PFA screwtop beaker before the following dissolution and drying steps were carried out in a class 100 fume hood.Concentrated HF (Romil UltraPure grade) was added to samples in a proportion 10:1 by mass for dissolution.The beaker was sealed and placed on a hot plate at 60 °C for 24 to 48 h (class 100 fume hood).During dissolution, condensate on the sides and top were carefully tapped back to the bottom of the beaker.This experimental set-up was tested on low-SiO 2 bulk rock materials and volcanic glasses only therefore its application to higher-SiO 2 bulk rock material was not tested in this study.Beaker lids were removed for dry down, and in their place, a second 7 ml uncapped beaker was placed upside down on top of the sample beaker with a small gap ($ 1 mm) left between upright and inverted beakers (Figure 1).This arrangement, which is a modification after Gaillardet et al. (2001), allows for controlled evaporation, in an equilibrated environment and protects the sample from particles falling in during the protracted, low temperature dry down (Gaillardet et al. 2001).The time elapsed until dryness was a function of total acid used for digestion and varied between 6 and 12 h, at 60 °C on the hot plate.Earlier studies used mannitol (C 6 H 14 O 6 ) to minimise boron volatilisation during dry down (Nakamura et al. 1992), but others have questioned the value of its addition (Gaillardet et al. 2001).Although practitioners who use mannitol have appropriately minimised its impact on the boron blank, we have adopted an approach that does not include mannitol, thus removing this potential source of boron contamination altogether.
The dried samples were re-dissolved in 200 μl of 0.5 mol l -1 distilled nitric acid.This re-dissolution was often incomplete, likely due to insoluble fluorides, but these residues have been found not to fractionate boron from the sample (Rosner et al. 2003, Pi et al. 2014, Liu et al. 2018).The amount of insoluble fine particles typically correlates with digested sample mass.A minimum of 150 μl (depending on the volume of largely precipitate free solution) was transferred into a plastic micro centrifuge tube and centrifuged for 5 min (at 13,000 rpm) to separate remaining fine particles that did not re-dissolve.After centrifuging, the supernatant from the step above was carefully pipetted off and loaded onto the first chromatography column.

Ion exchange column chromatography procedure for boron extraction and purification
Ion exchange procedures, summarised in Figure 2, are modified from well-tested boron separation techniques from carbonate matrices established in the Bristol laboratory (Foster 2008, Rae et al. 2011), with the addition of cationic columns either side of the boron specific separation, to remove the greater matrix load from typical silicate samples (Tonarini et al. 1997).These extra steps were implemented in a protocol that does not require intermediate sample dry down between column separations and the associated risk of sample loss or fractionation.Thus, in our preparation procedure, samples are only evaporated to dryness after initial dissolution (see previous section).
Samples were first passed through a micro-cation exchange column (resin: Dowex AG50W-X8 100-200 mesh, 200 μl volume, see Figure 2), to remove the majority of the major element fraction.Centrifuged samples were loaded and then eluted with an additional 50 μl of 0.5 mol l -1 nitric acid.All eluents from this initial column were collected.Experiments with reference solutions (NIST SRM 951 and NIST RM 8301 (Coral)) did not reveal significant boron isotopic fractionation from this cation exchange resin (a known problem when using the boron specific anion exchange resin Amberlite IRA 743 if < 99% of boron is recovered (Lemarchand et al. 2002, Nakamura et al. 1992)).
Solutions were then buffered to a pH of 5 using an equal volume of 2 mol l -1 Na acetate and 0.5 mol l -1 acetic acid buffer.The buffer itself was first purified by passing through a $ 500 μl volume Amberlite IRA 743 column, twice.The buffered sample was loaded stepwise (100 μl at a time) onto a 20 μl column (see Figure 2) containing Amberlite IRA 743 anion-exchange resin.Samples were purified by elution of matrix components with B-free high purity water (HPW Milli-Q company, resistivity: 18.2 MΩ cm), before the sample fraction containing boron was eluted with 0.5 mol l -1 nitric acid (5 × 110 μl).An additional elution step (the "tail") was performed after sample collection and analysed for boron concentration to ensure recovery from columns was complete (> 99%).These resin volumes are deliberately small (20 μl) so that matrix free samples can be obtained in low volumes of eluted acid (i.e., only 550 μl) that require no further dry-down prior to analysis.Resin within the columns should be routinely refloated prior to use to ensure no air bubbles or resin displacement.
Large volume silicate samples at this point may still have retained some matrix elements.Therefore, for further purification we pass samples through another cation exchange procedure, identical to the first column.This final column should also remove occasional contamination of sample with buffer that can occur from breakthrough in the Amberlite IRA 743 column.

Instrumentationsolution mode MC-ICP-MS
All isotope ratio measurements were performed at the University of Bristol using a Thermo-Finnigan Neptune MC-ICP-MS fitted with an ESI sapphire injector, a PFA Scott barrel spray chamber and a 50 μl min -1 PFA nebuliser for sample introduction.Nickel skimmer (X-geometry) and standard sample cones were used.The Neptune is equipped with two amplifier boards containing 10 12 Ω resistors that are used for measurement of 10 B + and 11 B + (at mass to charge ratio (m/z) = 10 and 11 respectively).
The NIST SRM 951 bracketing isotope reference material was run at a boron concentration of 50 μg l -1 .Prior to analysis, a 10 μl aliquot of each 550 μl sample solution was diluted and pre-screened for boron concentration and matrix (sodium concentration) content so that remaining sample solutions could also be diluted to 50 μg l -1 boron concentration.Some samples did not have enough boron to yield a 50 μg l -1 solution in the 550 μl eluted from the final columnthese samples were analysed undiluted at lower concentrations (note none were below 19 μg l -1 boron concentration).Previous work has shown that samples with boron concentration as low as 10 μg l -1 can be bracketed with 50 μg l -1 NIST SRM 951 solutions while still obtaining accurate results (Foster 2008, Rae et al. 2011).Concentrated HF (6 μl) is added to all sample solutions immediately prior to analysis to yield a solution of 0.3 mol l -1 HF and 0.5 mol l -1 HNO 3 .Similarly, all wash, blanks and bracketing standards are run as 0.3 mol l -1 HF and 0.5 mol l -1 HNO 3 solution, ensuring all boron is present as soluble BF 4 -and not volatile B(OH) 3 (Zeebe and Rae 2020).
Using this HF addition approach, instrumental blank is reduced to < 1 % of sample after 2 min wash.This method avoids poor instrumental washout that can occur due to boron volatilisation from other acid media (Misra et al. 2014) and is more effective than the previous use in the Bristol laboratory of ammonia gas addition during sample introduction (Al-Ammar et al. 2000, Foster 2008).
Acquisition time per measurement was 120 s (integration time of 4 s, thirty ratios, one block).Together with up-take time, this means that each purified sample can be analysed up to three times using a 50 μl min -1 nebuliser.Typically, samples are run twice, in different parts of the sequence, to reflect representative instrumental variability.The boron isotope ratio reported for a sample is the mean of the measurement results.As there is enough material for triplicate runs, if a particular part of a run is noisy, or uptake imperfect, a third measurement is possible.When made, a pooled mean and 2s of the three analyses are reported.Each individual analysis was blank corrected by subtracting from the sample intensities at m/z 10 and 11 interpolated between measurement results from bracketing 'blank pots', which are beakers containing equal volumes of blank acid, run in the same fashion as the samples to monitor the small cumulative addition of airborne boron over the analysis time of a given sequence (Rae et al. 2011).Intercomparison method -K 2 CO 3 flux fusion (IGG-Pisa) sample preparation procedure The three selected silicate reference materials were also prepared at IGG-CNR in Pisa (Italy), using the B-separation method described by Agostini et al. (2021).Approximately 0.2 to 0.4 g of powdered rock samples were mixed with purified, powdered K 2 CO 3 in Pt-Ir (95%-5%) crucibles, at a K 2 CO 3 /rock ratio > 4. K 2 CO 3 is used since its high solubility facilitates rapid aqueous leaching of the resulting fusion cake.The sample+flux mixtures are fused at a temperature of $ 1000 °C.After cooling, boron was

Anion exchange column
10 × H 2 O (160 µl dropwise) brought into aqueous solution over-night by addition of Bfree water.The solution, along with the insoluble phases, was transferred into polypropylene tubes, prior to centrifugation.Boron was then extracted and purified from the solution with an optimised three-step chemical separation procedure, as described in Agostini et al. (2021).
Initial sample purification was achieved using boronspecific Amberlite IRA 743 20-50 mesh resin, in $ 2.35 ml volume PFA columns (resin radius = 5 mm, resin height = 30 mm, frit thickness = 2 mm).The Amberlite IRA 743 was used once per chemistry and discarded afterwards.The resin was cleaned using 1.5 mol l -1 HCl (2 × reservoir volume) and pre-conditioned using an aqueous NH 3 (pH $ 10).The sample was loaded at pH > 10 and rinsed with an aqueous NH 3 (pH $10), followed by boron elution using 1.5 mol l -1 HCl, collected in concave bottom PFA beakers (Savillex) and dried overnight.To avoid boronloss and fractionation during this step, mannitol was added to the eluted sample, and the hot plate temperature maintained < 70 °C.
The day after, samples were re-dissolved in 0.015 mol l -1 HCl and then passed through AG50W-X8x (200 to 400 mesh) cation-exchange resin (2.35 ml volume, radius = 5 mm, height = 30 mm, Tonarini et al. 1997).Boron was immediately collected with 0.015 mol l -1 HCl.The sample solution was adjusted to pH > 10 adding 0.8 ml of 1.5 mol l -1 NH 3 , following a third purification step.The sample was passed through 20 to 50 mesh Amberlite IRA 743 resin, following the procedure outlined above, modified by collecting boron from Amberlite IRA 743 in w = 2% HNO 3 , ready to be measured via MC-ICP-MS.
All the chemical purification steps were performed in a class 1000 clean room, using B-free ultrapure reagents.High purity water (HPW) was obtained by sub-boiling distillation using a Savillex DST-1000; HCl was obtained starting from azeotropic solution of Supelco (Sigma-Aldrich) HCl and two subsequent steps of sub-boiled distillation using a Savillex DST-1000; HNO 3 was obtained starting from Suprapur (Merck) HNO 3 and two subsequent steps of sub-boiling distillation using a Savillex DST-1000; B-free NH 3 was obtained with sub-boiling distillation of ultrapure NH 3 , adding mannitol to the starting solution to prevent boron volatilisation.
Again, prior to analysis by MC-ICP-MS (Bristol), a 10 μl aliquot of each 550 μl sample solution was diluted and prescreened for boron and matrix (Na + ) content so that remaining sample solutions could also be diluted to 50 μg l -1 boron concentration.
The synthetic coral solution NIST RM 8301 (Stewart et al. 2020) was used to monitor potential influences of the column chemistry procedure both with and without HF dry down (Table 3).The long-term mean δ 11 B SRM 951 of artificial coral solution NIST RM 8301 (Stewart et al. 2020) purified following our new three-step column procedure is 24.24 AE 0.11‰ (n = 23), which is within uncertainty of interlaboratory consensus values (24.17 AE 0.18‰ (Stewart et al. 2020)).This provides a useful assessment of intermediate precision for 10 to 100 ng boron samples processed through our full boron purification procedure.

Blanks
Particularly for the small amounts of boron possible to analyse by MC-ICP-MS, laboratory and procedural blanks can be a significant concern at low sample/blank ratios (e.g., when the sample has less than 5 ng of boron).In the previous section we reported how instrumental memory is reduced by aspirating 0.3 mol l -1 HF solutions and cumulative 'fall-in' during measurement is corrected using 'blank pots' of equal volume (see section Instrumentationsolution mode MC-ICP-MS) which are used to subtract the boron 'fall-in' from the air during the period of sample measurements, which is typically 12 h.These contributions are subtracted directly from beam intensities during the analysis routine.
We have also monitored total procedural blanks for our full chemical separation procedure.Total procedural blanks for the method described here were established by running blank samples through the complete procedure (using all reagents as used for samples and including dry down).These blanks are measured by the same sample-standardbracketing procedure as samples.Measured boron signals for these total procedural blanks barely exceeded the background on the MC-ICP-MS.For this reason, it was only possible to obtain reliable mass estimates (in pg) for these blank solutions and their δ 11 B SRM 951 ratios should be considered as a guide only.Four total procedural blanks measured alongside the measurement results reported here vary from 28 to 58 pg (Appendix S1), which are in keeping with a mean of 38 AE 36 pg for larger number of total procedural blanks (n = 41) for Amberlite column chemical separation alone over a longer, three-year period (Appendix S1).Even for our smallest sample sizes of 10 ng, the contribution of this total procedural blank is negligible, and we make no correction for it (the lowest sample boron mass fraction (10268 pg) was 270 times greater than the mean total procedural blank of 38 pg).Typical ranges for blank δ 11 B SRM 951 are approximately -10 to 10‰ (Tonarini et al. 1997), in agreement with our mean estimate of -10‰ for the silicate processing (Appendix S1).
In our worst case (sample 10 ng and blank 58 pg), the blank contributes $ 0.58% towards the δ 11 B SRM 951 .A good overview of blank correction effects is given in Tonarini et al. (1997), and there a 2% blank contribution is needed to result in an 0.2‰ shift of the unknown.This is well within the limit of our here suggested intermediate precision of 0.6‰ for silicates.Additionally, we achieve a better sample/ blank than in Tonarini et al. (1997), which further diminishes the effect of blank composition on our unknowns.
Total procedural blanks for IGG-Pisa column chemistry are in the range of 10 to 40 pg and are similarly insignificant for the sample sizes studied here (10 to 50 ng).
δ 11 B SRM 951 of silicate rock reference materials Our results for silicate reference materials and a comparison with literature values is given in Table 1, Figure 3 and briefly summarised below.The natural silicate rock reference material best characterised for its δ 11 B SRM 951 is the JB-2 basalt, with a relatively well-defined consensus δ 11 B SRM 951 value between 6.7 and 7.7‰ based on results from several laboratories and their digestion methods (see Table 1).The mean δ 11 B SRM 951 value for JB-2 obtained here using our HF digestion method is 6.9 AE 0.4‰ (n = 12, 2s).Our measured value is identical to that we obtain from the same powder lot processed by the flux fusion and chemical separation procedure of the Pisa CNR-IGG laboratory (6.9 AE 0.1‰; n = 1, 2SE).
We obtained mean δ 11 B SRM 951 values of -6.0 AE 0.6‰ (n = 6) and -6.1‰ (n = 1) for the same material lot of International Atomic Energy Agency reference material IAEA-B-5 (Mt.Etna volcano basalt) using our HF digestion and the flux fusion protocol respectively.The concordant results of this study are significantly lower than the literature value (δ 11 B SRM 951 $ -4.0‰, Gonfiantini et al. 2003).
Agreement between samples processed by HF dissolution and flux fusion are poorer for the obsidian glass (IAEA-B-6).This yields a mean δ 11 B SRM 951 of -3.9 AE0.5‰ using our HF digestion protocol, but δ 11 B SRM 951 of -2.9 AE 0.1‰ for the aliquot processed by flux fusion.Yet both values lie within the wide bounds of literature data (δ 11 B SRM 951 = -0.45 to -4.46‰).
δ 11 B SRM 951 of oceanic basalts Our four MORB samples show a range of δ 11 B SRM 951 from -5.6 AE 0.3‰ to -8.8 AE 0.5‰ (Table 2).The difference between the mean δ 11 B SRM 951 of MC-ICP-MS and SIMS measurement results is statistically insignificant, but in contrast to the SIMS data, MC-ICP-MS measurement results show resolvable differences (at measurement precision level) in δ 11 B SRM 951 between samples (Figure 4).
The Pitcairn Island glass provides a clearer comparison between our bulk measurement results and SIMS, given its higher boron mass fraction the latter measurement result is more precise than for the depleted MORB samples.The agreement between the two approaches is good (Table 2),  Mean: 6.9 ± 0.4‰ Mean: -6.0 ± 0.6‰ Mean: -3.9 ± 0.5‰ with a mean of -5.3 AE 0.9‰ (n = 6) from repeat SIMS measurement results versus -5.1 AE 0.1‰ (n = 4) from a mean of measurement results of two different glass shards by MC-ICP-MS (Table 2).

Incorporation of previous study observations on boron volatilisation
Volatilisation of boron during sample processing of silicate samples is a well-known concern (Nakamura et al. 1992).Some studies exploited this boron volatilisation, by achieving total evaporation and collecting and condensing the vapor fraction, which serves as a first matrix removal step by leaving behind non-volatile elements and compounds (Xiao et al. 1997, Gaillardet et al. 2001, Liu et al. 2013, Van Hoecke et al. 2014).The use of HF in silicate dissolution raises the spectre of possible boron loss as BF 3 and has commonly led to the addition of mannitol during dissolution to guard against this (Nakamura et al. 1992, Tonarini et al. 1997).Not only has the effectiveness of mannitol in binding to boron through complexation in typical dissolution scenarios been questioned (Xiao et al. 1997, Gaillardet et al. 2001), but recent calculations have further shown that the highly soluble BF 4 -is the dominant species in strong solutions of HF (Zeebe and Rae 2020).Studies by Xiao et al. (1997) and Gaillardet et al. (2001) concluded that boron isotopic fractionation is not necessarily prevented by the addition of mannitol, but instead that samples slowly dried at temperatures < 65 °C, from acids such as HCl and HF, retain their true boron isotopic composition, if recovered (i) immediately or (ii) just before dryness is achieved.
We have conducted a series of evaporation experiments of our own, in an attempt to reproduce the observations of Gaillardet et al. (2001), in the context of silicate rock samples.We changed multiple parameters to test their impact on δ 11 B SRM 951 results: (i) acid matrix was either HF conc or HNO 3 conc , (ii) the dry down temperature set at 60 °C and 120 °C and (iii) the vapour was allowed to escape rapidly or in a semi-enclosed set-up that allows only slow vapour loss.Our experimental results support the finding of Gaillardet et al. (2001), in that open evaporation in a nitric acid medium causes obvious isotopic fractionation (Figure 5, Table 4).In one experimental run, samples were dried excessively, which led to an approximately 3 to 4‰ isotopic shift towards lower δ 11 B SRM 951 in the evaporated residue (Figure 3b).Subsequently, we adopted drying at 60 °C for our final procedure of silicate sample processing, removing samples from the hot plate just before or when dryness is achieved.When collecting before dryness was achieved, residual HF was neutralised by addition of Na-acetate, forming NaF, visible as precipitate occasionally.Testing of pH is necessary for samples collected prior to dryness, to ensure effective ion exchange resin performance.

Residual matrix contamination
Residual matrix components (major ions: Si, Ca, Na, Mg, etc.) in samples can cause differences in mass bias and shifts in the measured 11 B/ 10 B ratio of the following bracketing isotope standard (Xiao et al. 1997).While a good separation of boron from other elements is achieved by our chromatographic procedures, in a very few instances, some sample matrix and/or Na + still leaked through into the sample aliquot during the testing and adjustment phase of the method development.Before analysis we therefore measured the m/z = 23 (Na + ) beam intensity during the boron concentration screening measurement, to ensure all matrix cations had been separated from the boron (Rae et al. 2011).During the evolution of our procedure, we found using a second cationic column, guarded against this eventuality and further purified the boron from silicate matrix components.The intermediate precision of JB-2 basalt improved as we developed our modified measurement procedure by adding the final cation exchange column step.
All data we report here were processed using this second cation exchange column.our three-step purification chemistry at Bristol and sample screening for Na, we infer from the poorer intermediate precision of the measurement result of silicate reference materials ($ AE0.6‰; 2s) compared with the simpler matrix boric acid and calcium carbonate reference materials (AE < 0.2‰; 2s; Table 3) that matrix separation may still be imperfect for silicate samples (presumably matrix ions other than Na + ).We employed a cut-off for sample rejection and/or re-processing through the cationic column when the raw m/z = 23 (Na + ) beam intensity of the prescreening test (fifteen-fold dilution of 10 μl sample aliquot; see section Instrumentation -solution mode MC-ICP-MS)) exceeded 0.5 V prior to sample analysis.Future testing of matrix sensitivity of silicate bulk rock samples following the procedures outlined in Chen et al. (2016) and Devulder et al. (2013) may provide an angle for further improvement of the method described here.Previous work by Devulder et al. (2013) showcased promising results using lithium as an isotope dopant, which encourages further research.Based on the intermediate precision of boron in silicate materials of 0.6‰ (discussed here) vs. the typical intermediate precision of 0.2‰ for carbonate materials, we infer that up to 0.4‰ of the total measurement precision derives from matrix effects.This, however, conflicts with observations by Devulder et al. (2013), Foster (2008) and Rae et al. (2011), who find no significant effect of matrix elements studied by them.An extensive study of elements interfering with B in MC-ICP-MS would thus be required to solve this discrepancy.
δ 11 B SRM 951 comparabilitydiminishing interlaboratory differences for basalt matrixes A recurring problem in studies of silicate rocks is the large variability of reported δ 11 B SRM 951 for silicate reference materials (Gonfiantini et al. 2003), particularly when comparing different sample preparation and measurement procedures (Gonfiantini et al. 2003).Several permil differences between analytical set-ups are commonly observed, but large discrepancies can still occur within datasets obtained by similar measurement procedures (Gonfiantini et al. 2003) (Table 1).In this light, we compare our new δ 11 B SRM 951 results with literature data.
We stress the excellent agreement between the δ 11 B SRM 951 we report for basaltic reference materials JB-2 and IAEA-B-5 using our two preparation techniques.Given our MC-ICP-MS applied measurement procedure approach has been benchmarked against a wide number of pure isotope reference materials, we expect that silicate sample processing (dissolution and matrix separation) should be the most critical aspect of sample accuracy.Therefore, the concordant results for basaltic reference materials obtained by (i) HF dissolution followed by our three-column sample preparation procedure at Bristol and (ii) alkali flux fusion followed by three steps of purification in Pisa, is strongly supportive of the reliability of these values.
In a previous section we noted the sensitivity of sample measurement precision of the measurement results to the inferred difference in residual matrix between purified silicate samples and boric acid solution reference materials.Matrix sensitivity has also been emphasised by Chen et al. (2016).Contrasting residual matrix compositions may therefore provide a possible explanation of differences between recent MC-ICP-MS measurement results of the same reference materials (Table 1).For example, Li et al. (2019) and Wei et al. (2013) who use a single column separation procedure, obtain δ 11 B SRM 951 for B5 > 1.5‰ higher than our measurement results at measurement precision.Such discrepancies require further investigation and quantification of abundances of residual species after boron purification seems worthwhile.However, we also note that unlike the δ 11 B SRM 951 we obtained for BAM ERM-AE121 (19.65 AE 0.14‰), the value reported by Li et al. (2019) is (1.5‰ lower) outside of uncertainty of both the certified value of 19.9 AE 0.6‰ (Vogl and Rosner 2012) and the recent interlaboratory consensus value (19.64 AE 0.17‰; 2s of measurement results from five boron isotope laboratories (Stewart et al. 2020)) for this matrix-free reference material.Further inter-lab calibration work and measurement uncertainty budget evaluation is therefore required, for instance, testing of common analyte internal standardisation (e.g., addition of a lithium isotope spike to samples) that may help to improve sample-standard-bracketing techniques (e.g., Devulder et al. (2013)).
Agreement between δ 11 B SRM 951 values for the obsidian glass reference material IAEA-B-6 between our HF acid dissolution and flux fusion approaches is imperfect.Yet Tonarini et al. (1997) report scepticism about their IAEA-B-6 results, suggesting the flux fusion to have formed silicate structures that retained boron in the fusion cake; this may explain the differences we obtained.Indeed, high silica samples such as IAEA-B-6 obsidian potentially fractionate boron during fusion (Tonarini et al. 2003).Our HF dissolution measurement results are again isotopically lighter than recent MC-ICP-MS measurement results of Li et al. (2019) and Wei et al. (2013), but are in agreement with two recent analyses of another IAEA-B-6 batch carried out at IGG-Pisa after flux fusion.Those measurement results are -3.60 AE 0.07‰ (2SE) and -3.48 AE 0.04‰ (2SE), within uncertainty of our value (-3.9 AE 0.5‰).In the latter, special care was taken to avoid incomplete fusion and residue in the fusion cake.To this end, the amount of sample powder was reduced to 100 mg, and the amount of K 2 CO 3 flux was increased to 1 g, raising the flux/sample ratio > 10.This reduction of sample volume may (i) increase the role of potential B inhomogeneity, negatively impacting uncertainty of measurement results and (ii) increase blank contribution to the measurement uncertainty budget.
However, petrographic observations of IAEA-B-6 have shown that the parental material for this reference material is heterogeneous (Cooper et al. 2019).Li et al. (2019) further reported lighter δ 11 B SRM 951 isotopic composition (by up to 1.5‰) for IAEA-B-6 after leaching, indicative of alteration phases in the powder.It is possible that different batches of IAEA-B-6 are heterogeneous and so it is not the best reference material with which to compare data reproducibility between laboratories.
Generally, the HF dissolution and flux fusion procedure discussed and tested here returned comparable results.We would therefore recommend the use of HF dissolution for labs without flux fusion capabilities and particularly advocate its use for small sample volumes (< 600 μl).

Applications of boron isotopes in source reservoir identification
A growing body of data show a mean MORB δ 11 B SRM 951 of $ -7.5‰ (Dixon et al. 2017, Marschall et al. 2017), but discerning significant δ 11 B SRM 951 variability within these relatively imprecise (typically AE2‰) in situ measurement results is difficult.While several MORB have previously been identified with δ 11 B SRM 951 distinctly higher than this mean, they are also associated with anomalously high Cl/K (Marschall et al. 2017, Li et al. 2019) or water contents (Dixon et al. 2017) indicative of the influence of crustal contamination.By contrast, South Atlantic MORB with δ 11 B SRM 951 as low as -11.7 AE 2‰ appear to resolve a component with isotopically lighter boron (Dixon et al. 2017).
In general, our new MORB data have δ 11 B SRM 951 values in keeping with the literature (Figures 4 and 6), including the recent MC-ICP-MS study of Li et al. (2019).Yet it is notable that for samples that show no evidence of seawater contamination, we demonstrate variability in δ 11 B SRM 951 beyond our intermediate precision (AE0.6‰).Namely, two samples with low Cl/K (< 0. δ 11 B SRM 951 that ranges from -6.2 to -8.8‰, implying significant variability in the upper mantle source (Table 2).
We note that the sample volume of the SIMS measurement results is significantly lower than the bulk glass (25.9 to 63.2 g) measurement with which we compare it here.The low abundance of boron in mafic volcanic glass means it is not easy to decipher heterogeneity at the micrometre scale that would be homogenised when using several grams of glass shards.This requires either further reduction of sample volumes analysed by solution MC-ICP-MS or higher precision for in situ techniques.Future work could include further sequential leaching experiments of reference materials (e.g., Li et al. (2019)).This, however, first requires a better consensus on natural reference material δ 11 B SRM 951 values.
As in other studies, there is no clear trend of δ 11 B SRM 951 with upper mantle enrichment (Figure 4).Our E-MORB from the East Pacific Rise and enriched oceanic basalt from Pitcairn Island have δ 11 B SRM 951 indistinguishable from two of our three N-MORB samples (Figure 2, Table 2).This shows that substantial mantle enrichment can occur without markedly influencing δ 11 B SRM 951 .The preliminary observation is that there is greater variability in boron isotope ratios in the more depleted samples.

Conclusions
The dissolution of silicate materials using hydrofluoric acid prior to column chemistry and measurement by MC-ICP-MS presented here permit the investigation of samples limited by mass and low-B mass fraction samples, such as MORB glass.Key findings include: (1) Boron can be liberated from silicate rock materials without isotopic fractionation by use of HF acid digestion in the absence of volatilisation suppression agents (e.g., mannitol).This requires careful, low temperature (< 60 °C) dry down following dissolution.Evaporating samples from a beaker largely covered by an upturned beaker helps control evaporation, prevents particle fall in and appears to guard against sample fractionation during higher temperature evaporation, for a short duration of time.
-  (2) Our three-column (cation-anion-cation) method of removing matrix elements and measurement of low-B samples by solution MC-ICP-MS yields marked improvement in measurement precision over results from previous thermal ionisation (TIMS) and in situ (SIMS, LA-MC-ICP-MS) techniques, with measurement precision now better than AE0.6‰ (2s).We obtain consistent δ 11 B SRM 951 for basaltic reference materials with boron separated by our acid digestion approach and an independent, flux fusion technique.(3) A comparison of SIMS and newly acquired MC-ICP-MS data of volcanic glasses from MORB settings shows a general overlap of SIMS and MC-ICP-MS results, but considerably better measurement precision is offered by MC-ICP-MS for samples with B mass fractions < 2 μg g -1 .This allows us to resolve boron isotopic heterogeneity in the upper mantle.

Figure 1 .
Figure 1.Schematic drawing of the beaker arrangement for the sample dry-down after HF digestion.

Figure 2 .
Figure 2. Procedure of the ion exchange chromatography in table format with schematic drawings of the column design and dimensions used in this study, for the respective steps.

Figure 3 .
Figure 3. δ 11 B SRM 951 of silicate reference materials using the HF digestion and three-column (cationanion-cation) ion exchange procedure presented in this study.The range bars are 2s of replicate measurement results.Thick black lines represent mean values measured here and the dashed black line the standard deviation.The silicate reference material JB-2 (a) yielded the greatest reproducibility.Note that two samples (orange points) in (b) were left intentionally on the hot plate after dryness to investigate the effect of isotope fractionation from a dry mediumthese are not included in our mean value.The approximately 2 to 4‰ lighter isotopic composition measured is consistent with other excessive dry down experiments (Nakamura et al. 1992, Gaillardet et al. 2001).In (c), one IAEA-B-6 triplicate analysis (δ 11 B SRM 951 -3.44 AE 1.03‰) is constituted of the values: -3.61 AE 0.26‰, -3.95 AE 0.23‰ and -2.94 AE 0.23‰ (all 2s).In the absence of indicators that would suggest anomalous run conditions (Rae et al. 2011), the value of -2.94 AE 0.23 is retained in the triple replicate calculation.Black squares denotes flux fusion with measurements in Bristol, whereas whites square with black outline denotes flux fusion and measurement at Pisa CNR (MC-ICP-MS). 

Figure 5 .
Figure 5. δ 11 B SRM 951 of synthetic coral isotope reference material (NIST RM 8301 (Coral)), that was dried to convert from strong HF or HNO 3 acid to 0.5 mol l -1 HNO 3 , simulating the dissolution procedure.The dashed yellow lines represent the 2s variability of the interlaboratory consensus value, Stewart et al. (2020).Open circles denote samples where vapour was allowed to escape rapidly during the drying process (uncovered beakers), while filled circles represent samples where evaporation was carried out in a semienclosed environment (see Figure1).Red symbols indicate HNO 3 acid matrix, black symbols indicate HF acid matrix.
Geostandards and Geoanalytical Research published by John Wiley & Sons Ltd on behalf of International Association of Geoanalysts.
of volcanic glass obtained by SIMS (open circles) with uncertainty indicated at the 2s level.The δ 11 B SRM 951 of volcanic glass obtained by LA-ICP-MS (open squares) with uncertainty indicated at the 2s level.Sources of data (δ 11 B SRM 951 and K/Ti) displayed: Li et al. (2019), Marschall et al. (2017), Dixon et al. (2017), le Roux et al. (2004).We analysed the same material as Marschall et al. (2017) and utilise the K/Ti values from Marschall et al. (2017) together with our new, solution MC-ICP-MS δ 11 B SRM 951 values.

Table 1 .
Measured δ 11 B SRM951 values of silicate reference materials and previously published values

Table 3 .
Individual run results of reference materials measured in this study.Within-run measurement precision represents the mean of two to three repeat measurement results of the same aliquot

Table 4 .
Data presented in Figure4 (Dixon et al. 2017, le Roux et al. 2004)17)LA-ICP-MS(Dixon et al. 2017, le Roux et al. 2004)Figure6.Solution MC-ICP-MS measurement results (red diamonds) for glasses previously measured by SIMS.MC-ICP-MS data plotted with uncertainty of AE0.6‰, reflecting the intermediate precision of basalt reference materials.Detailed information on individual measurement precision is given inTable 1.The δ 11 B SRM 951