Inter‐laboratory Characterisation of Apatite Reference Materials for Chlorine Isotope Analysis

Here we report on a set of six apatite reference materials (chlorapatites MGMH#133648, TUBAF#38 and fluorapatites MGMH#128441A, TUBAF#37, 40, 50) which we have characterised for their chlorine isotope ratios; these RMs span a range of Cl mass fractions within the apatite Ca10(PO4)6(F,Cl,OH)2 solid solution series. Numerous apatite specimens, obtained from mineralogical collections, were initially screened for 37Cl/35Cl homogeneity using SIMS followed by δ37Cl characterisation by gas source mass spectrometry using both dual‐inlet and continuous‐flow modes. We also report major and key trace element compositions as determined by EPMA. The repeatability of our SIMS results was better than ± 0.10‰ (1s) for the five samples with > 0.5 % m/m Cl and ± 0.19‰ (1s) for the low Cl abundance material (0.27% m/m). We also observed a small, but significant crystal orientation effect of 0.38‰ between the mean 37Cl/35Cl ratios measured on three oriented apatite fragments. Furthermore, the results of GS‐IRMS analyses show small but systematic offset of δ37ClSMOC values between the three laboratories. Nonetheless, all studied samples have comparable chlorine isotope compositions, with mean 103δ37ClSMOC values between +0.09 and +0.42 and in all cases with 1s ≤ ± 0.25.

Moreover, it is recommended to report the relative difference of isotope ratios as 10 3 δ 37 Cl rather than in commonly used per mil (‰) notation (Meisel 2019). In this manuscript, however, we used ‰ notation both to express measurement repeatability, which was calculated as relative standard deviation, and to report the difference in instrumental mass fractionation (IMF) in SIMS analyses. Furthermore, when quantifying the actual divergence of measured isotope ratios from the true values defined by gas source isotope ratio mass spectrometry (GS-IRMS), we used the widely employed notation:  (2) which is a unitless value directly related to the IMF induced bias according to: The anchor point for the 10 3 δ 37 Cl SMOC scale is +0.43 (Brand et al. 2014), which corresponds to the sodium chloride NIST SRM 975 (Xiao et al. 2002). However, due to the fact that the absolute ratio for this material ( 35 Cl/ 37 Cl = 3.1272) is associated with a significant uncertainty of AE 0.0082 (2s, Shields et al. 1963), throughout this paper we recalculated δ 37 Cl SMOC values obtained by GS-IRMS to 37 Cl/ 35 Cl ratios using the SMOC ratio of 0.319533 (zero point on the δ 37 Cl SMOC scale). This value was calculated based on the data reported for the sodium chloride secondary reference material (RM) ISL-354: 10 3 δ 37 Cl SMOC value of +0.05 AE 0.02 (1σ; Xiao et al. 2002) and 37 Cl/ 35 Cl ratio of 0.319549 AE 0.000044 (2s; Wei et al. 2012). The recalculation of δ 37 Cl values obtained by conventional methods to 37 Cl/ 35 Cl ratios for our samples was necessary for quantifying the IMF of our SIMS analyses.
Most of the existing research on the chlorine isotope composition of geological materials has been based on bulk analyses using the GS-IRMS (e.g., Eggenkamp 1994Eggenkamp , 2014. However, in those cases where limited amounts of material are available in situ analysis using SIMS is the method of choice. NanoSIMS instruments have proven useful for the analysis of lunar samples, where repeatability of measurements on the reference materials of~1‰ is acceptable in view of the large natural variations of 10 3 δ 37 Cl SMOC in lunar materials (from -1 up to +81, e.g., Sharp et al. 2010, Potts et al. 2018, Stephant et al. 2019, Wang et al. 2019. In contrast, the much narrower range of δ 37 Cl values found in most terrestrial geological samples makes chlorine isotope ratio determinations challenging, both in terms of data acquisition and interpretation. Two recent studies of apatite crystals from Proterozoic and Archaean rocks have shown that chlorine isotope signatures of apatite can help in tracing metasomatic processes overprinting terrestrial rocks; such data can provide valuable information, despite displaying much narrower ranges of 10 3 δ 37 Cl SMOC values (from -1.2 to +3.8, Kusebauch et al. 2015a or from -0.8 to +1.6, Wudarska et al. 2020) compared with those recorded in the extra-terrestrial materials.
Quantitative isotope measurements by SIMS require homogeneous RMs of known isotope composition. Recent SIMS studies of δ 37 Cl in apatite (Kusebauch et al. 2015a, Bellucci et al. 2017, Sarafian et al. 2017 have been based on two endmembers of the solid solution series represented by chemical formula Ca 10 (PO 4 ) 6 (F,Cl,OH) 2 : natural fluorapatite from Durango (Mexico) and synthetic chlorapatite. A linear correction between IMF and Cl mass fraction was applied based on the analyses of these two materials (Kusebauch et al. 2015a). Unfortunately, these RMs are available only in small amounts. Taking into consideration the narrow range of chlorine isotope ratios in terrestrial materials, SIMS measurements require calibration using RMs closely matched to the chemical composition of the unknowns. Therefore, the goal of our current research was to develop a suite of well-characterised, homogeneous RMs covering a wide range of chlorine mass fractions in apatite making it possible to precisely define the matrix-dependent IMF of SIMS measurements. This issue has also been highlighted in the recent paper by Li et al. (2020) who concluded that additional RMs with a range of Cl mass fractions are required for routine δ 37 Cl measurements, which would be in addition to a newly developed apatite RM (Eppawala-AP, Sri Lanka) described therein.
Previous SIMS isotopic works on magnetite, haematite and sphalerite showed that there may be a significant crystal orientation effect on measured oxygen and sulfur isotope ratios (Huberty et al. 2010, Kita et al. 2011. In view of the anisotropic behaviour of apatite during EPMA measurements (Stormer et al. 1993, Goldoff et al. 2012, Stock et al. 2015, which results from the incorporation of F -, Cland OHin the anion column parallel to the c-axis of the hexagonal structure Rakovan 2002, Hughes et al. 2016), we also conducted a SIMS investigation of apatite fragments that were intentionally embedded in 1 2 2 epoxy mounts so as to provide differing crystallographic orientations.

Sample selection and description
A total of fifty-six natural apatite samples provided by mineralogical museums were tested for their chlorine isotope compositions ( 37 Cl/ 35 Cl) using SIMS. A suite of six specimens (Table 1, Figure 1, Appendix S1) that proved to be suitably homogeneous in 37 Cl/ 35 Cl (repeatability of measurements within AE 0.2%, 1s) and representing a broad range of Cl mass fractions were further characterised using scanning electron microscopy-energy dispersive spectroscopy (SEM-EDS), EPMA, SIMS and GS-IRMS. Durango apatite, which is often used as a reference material in various analytical methods (e.g., McDowell et al. 2005, Teiber et al. 2015, Kusebauch et al. 2015a, has served in this study both as a test sample and as a quality control material. The six selected materials, previously cleaned in acetone and distilled water, were crushed and sieved to a grain-size fraction between 63 and 1400 µm (or in case of Durango between 63 and 1000 µm). These subsequently underwent mineral separation to remove any foreign phases and were finally split using a riffle splitter in order to generate multiple, metrologically identical units (Table 1). The purity of all split materials is estimated to be ≥ 98%.
MGMH#133648 consists of seven yellow-white crystals of chlorapatite from Norway, of which only the one fragment (~11.4 g) shown in Figure 1 was selected for this study, while the remaining material (~13.2 g) was returned to the Mineralogical and Geological Museum at Harvard University. A white crust of hydroxylapatite was removed from the surface of the chlorapatite crystal using a circular saw and a micro-drill. Crushed material underwent magnetic (Frantz) and heavy liquid (bromoform) separation. MGMH#133648 contains minor amounts of inclusions of hydroxylapatite, Ca-Mg-Fe-silicates and aluminosilicates, and calcite.
TUBAF#38 is a yellow crystal of chlorapatite from Bamble, Norway. A whitish grey crust of hydroxylapatite was removed from the surface of the chlorapatite crystal using a circular saw. Crushed material underwent magnetic (Frantz) and heavy liquid (bromoform) separation. TUBAF#38 contains minor amounts of inclusions of hydroxylapatite, Mg-silicates, Mg-Fe-Ca-and K-bearing aluminosilicates, and monazite.
MGMH#128441A is a yellow crystal of fluorapatite, which was selected for this study from eight specimens from Colorado, US; the other seven crystals, having similar Cl isotope compositions, were returned to the Mineralogical and Geological Museum at Harvard University. Dark-green mineral (possibly amphibole) was removed from the surface and from a cavity inside of the selected crystal using a microdrill. Crushed material underwent magnetic (Frantz) separation. MGMH#128441A contains minor amounts of inclusions of Fe-oxides, Ca-Mg-Fe-Mn-silicates and aluminosilicates (some of which probably contain minor amounts of Cl), and monazite.
TUBAF#50 is a yellow, multi-crystal fluorapatite from Spain. Prior to crushing, the sample was hand-picked to remove non-apatite material (e.g., dust particles and contaminating fibres). TUBAF#50 is almost free from any inclusions; only a single inclusion of Fe-silicate was identified using SEM-EDS.
Dur is a yellow fluorapatite from Durango, Cerro de Mercado, Mexico. A gem-quality single crystal contained only minor amounts of Fe-oxides, which could mostly be eliminated by careful hand-picking.

Sample preparation
For further EPMA and SIMS characterisation, apatite crystals were mounted in epoxy resin in the centre of 25.4 mm diameter acrylic discs in which holes had been drilled. The mounts were subsequently polished to a < 5 µm surface quality as evaluated by white light interferometry. A total of five mounts were produced; they contained the following materials obtained from the splitting process described above: 1 2 3 • Mount #1: one unit of each sample (random crystals) and one fragment of both: the Durango apatite and the NIST SRM 610 silicate glass for tuning purposes (Figure 2).
• Mount #2: a few hand-picked crystals of TUBAF#37, which showed different hues of yellow colour under the binocular microscope, and a few crystals of TUBAF#38 having white, usually opaque, OH-enriched fragments (Appendix S2, Figure S2.1).
• Mount #3: a part of the white crust of hydroxylapatite mechanically removed from the surface of the MGMH#133648 sample (see section 'Sample selection and description'), which was embedded in epoxy to determine its isotopic composition, for addressing the minor amounts that we were unable to exclude from the sample splits (Appendix S2, Figure S2.2).
• Mount #4: three fragments of the Durango single crystal, which represent different orientations with respect to c-axis, along with a piece of NIST 610 glass for tuning purposes (Appendix S2, Figure S2.3).  Conventional δ 37 Cl determinations in three independent GS-IRMS laboratories were performed on identical mineralogical splits that were hand-picked in Potsdam under a binocular microscope beforehand in order to remove most of the non-apatite material (minor amounts of inclusions were present in some analysed samples, but special attention was paid to remove any Cl-rich phases such as amphibole).

SEM-EDS and FE-SEM-EBSD analyses
Polished crystals embedded in epoxy as well as loose grains mounted on conductive carbon tabs were investigated at the Helmholtz Centre Potsdam (GFZ) using an Ultra Plus (Carl Zeiss, Oberkochen) field emission scanning electron microscope equipped with an EDS detector (Thermo Fisher Scientific) operated at 20 kV accelerating voltage in order to identify the chemical compositions of those phases associated with apatite samples.
The crystal orientations of the Durango apatite fragments selected for the assessment of crystallographic orientation effects on measured 37 Cl/ 35 Cl ratios were determined using the electron backscatter diffraction technique (EBSD). Sample mount #4 (see above), which had been polished with silica colloid, was placed on a SEM stub using a carbon tape. EBSD point analyses and mapping (at 0.2 µm step size) were performed using a FEI Quanta 3D field emission scanning electron microscope equipped with an EBSD detector (AMETEK-EDAX Digiview 5 Camera) in low vacuum mode (~100 Pa) operated at 20 kV acceleration voltage and 23 nA beam current. The raw data were further processed using EDAX OIM software. The EBSD mapping results were filtered based on the Confidence Index (CI) and only the points having CI > 0.15 were taken into consideration for further interpretation.

Chemical analysis by EPMA
Quantitative chemical analyses of apatite crystals were performed using the JEOL Hyperprobe JXA-8500F microprobe at the GFZ. The samples were coated with a 20 nm thick carbon film and analysed for major and trace elements using an acceleration voltage of 15 kV, a spot size of 10 µm and two different beam current settings of 5 and 40 nA. Very low electron beam current of 5 nA and relatively large beam diameter were applied in order to minimise volatile (F, Cl) diffusion (Stormer et al. 1993, Goldoff et al. 2012, whereas a high electron beam current of 40 nA was used to detect REE and other trace elements. The details on the spectral lines, corresponding peak counting times and the calibration materials are given in Appendix S3. Calculations of the apatite formulae based on the EPMA data were performed using the method of Ketcham (2015) with normalisation done based on 25 oxygen equivalents. The OH content was calculated with the assumption that the Xsite is fully occupied by F, Cl and OH (F + Cl + OH = 2).

Chlorine isotope homogeneity test by SIMS
In situ analyses of 37 Cl/ 35 Cl ratios were conducted using the Cameca 1280-HR large-geometry SIMS instrument at the GFZ. Prior to analyses, the sample mounts were cleaned in high-purity ethanol and were then argon sputter coated with a 35 nm thick high-purity gold film to assure electrical conductivity. A~2 nA, nominally 10 keV 133 Cs + primary ion beam focused to a~5 µm diameter spot was used to produce 35 Cland 37 Clsecondary ions following a modified protocol based on that of Kusebauch et al. (2015a). Using a 25 × 25 μm raster, the sample surface was sputtered for 60 s to remove the gold coat and to suppress surface contaminants. The raster was then reduced to a 15 × 15 μm, from which 35 Cland 37 Clsecondary ions were collected simultaneously on Faraday cups, over twenty cycles with each According to earlier research on aluminosilicate glasses by Layne et al. (2004), a MRP of 5142 is necessary to separate adjacent peaks corresponding to 35 Cland 34 S 1 H -. Sulfur is commonly incorporated into the apatite structure (Pan and Fleet 2002); therefore, we have carefully developed our SIMS measurement protocol taking into consideration the risk of this potential interference. In the case of TUBAF#37, which has the lowest Cl/S ratio among all tested samples with 0.27 AE 0.04% m/m of Cl and 0.25 AE 0.09% m/m of SO 3 , the 34 S 1 Hpeak is nearly negligible (ca. 65 counts per second) in comparison with the very intense 35 Clpeak (ca. 3.15 × 10 5 counts per second, Figure 3). Hence, this added precaution seems worth considering only at very low Cl/S ratios during routine analyses.
The peak corresponding to 34 S 1 Hhas been noted only using electron multiplier (EM) detection, while it could not be distinguished from the background of the Faraday cup (FC). Taking into consideration a very low intensity of this peak in EM detection, we have concluded that the isobaric interference of 34 S 1 Hwith 35 Clis irrelevant for our apatite homogeneity test, and therefore, we have conducted the measurements at a mass resolving power of~1800 (Figure 4), which greatly improves ion counting rates leading to overall better data quality.
Multiple SIMS analysis points on multiple fragments were selected on all six apatite samples; these locations were broadly spaced and selected in a quasi-random fashion. Our SIMS data were collected in a single automatic, chainanalysis sequence that lasted~17.5 h. Instrumental drift was monitored via closely spaced analyses both on the fragment of the Durango apatite and on one of the MGMH#133648 crystals that were regularly interspersed throughout the run. Additional tests of selected fragments of TUBAF#37, TUBAF#38 and MGMH#133648 embedded in two other mounts were conducted during the same session under identical analytical conditions. Three oriented fragments of the Durango apatite had been analysed a couple of months earlier following similar analytical protocol.
A typical count rate for 37 Clions was in the range from 3.8 × 10 6 (for fluorapatite TUBAF#37) to~9.4 × 10 7 (for chlorapatite MGMH#133648). The depth of sputtered sample pits was~0.9 µm, and the amount of consumed material during each analysis was~250 pg as based on white light interferometry (for a 3D image of a typical SIMS crater, see Appendix S4).

Conventional δ 37 Cl value determination by GS-IRMS
Isotopic studies can be affected by inter-laboratory bias, which results from differences in analytical protocols (e.g., Gonfiantini et al. 2003, John et al., 2010, Manzini et al. 2017. In order to establish the extent of inter-laboratory bias for chlorine isotope measurements of apatite, conventional δ 37 Cl values were determined independently in three gas source isotope ratio mass spectrometry laboratories. A traceability material of sodium chloride ISL-354 of known chlorine isotope composition ( 37 Cl/ 35 Cl = 0.319549, Wei et al. 2012; 10 3 δ 37 Cl SMOC = +0.05, Xiao et al. 2002) was analysed along with the apatite samples.
Dual-inlet isotope ratio mass spectrometry in laboratory 1: Chlorine isotope analyses at the Institut de Physique du Globe de Paris (France) were conducted using dual-inlet IRMS following the methods described in Eggenkamp (1994), improved by Godon et al. (2004), and used at IPGP ever since (e.g., Li et al. 2015, Eggenkamp et al. 2016, 2019. Apatite samples were fully dissolved in 15.5 ml of 4.4 mol l -1 HNO 3 and placed in an oven at 75°C for~7 h. All samples, except one (MGMH#133648), were independently dissolved twice (i.e., eleven separate dissolutions) before independent replicate preparation for δ 37 Cl measurements (n = 18 in total). The amount of the processed material depended on the Cl concentration: from~24 mg for chlorapatite MGMH#133648 up to~360 mg for fluorapatite TUBAF#37. Chlorine was completely precipitated as AgCl after the addition of 5 ml of 1 mol l -1 AgNO 3 . In contrast, following the recommendation of Eggenkamp (1994), for the precipitation of the internal seawater reference Atlantique 2 (n = 10) and of the international reference material ISL-354 (previously dissolved in ultrapure water; n = 9), we consecutively added: 4 ml of 1 mol l -1 KNO 3 (to increase the ionic strength of the solution), 2 ml of McIlvaine solution (to buffer solution at pH 2.2) and 1 ml of 0.2 mol l -1 AgNO 3 .
After precipitation, AgCl was filtrated and protected from light on a 0.7 µm Whatman GF/F glass fibre filter. The filter was dried overnight in an oven at 80°C and subsequently loaded into a Pyrex™ tube with excess of CH 3 I, sealed under vacuum and reacted at 80°C for 48 h. The tube was then cracked under vacuum, and the CH 3 Cl was separated from the excess of CH 3 I by gas chromatography (two columns filled with Porapak Q). The final amount of pure CH 3 Cl generated was quantified by a capacitive gauge.
This provided a control of the quality of the whole procedure: apatite dissolution, AgCl precipitation, CH 3 Cl fabrication and CH 3 Cl purification. Yields for the whole procedure were calculated for each prepared aliquot of CH 3 Cl, based on the mass of apatite dissolved and the results of EPMA measurements. Finally, CH 3 Cl was collected in a sample 0.0E+00 5.0E+04  Figure 3. Secondary ion spectrum (registered using EM) at nominal mass of 35 Clfor the apatite sample TUBAF#37 illustrating isobaric interference with 34 S 1 Hat M/ΔM ≈ 5400 (at 10% peak height). Lower part of the figure shows a magnified spectrum from the upper part of this figure.

2 7
tube and transferred to a Thermo Scientific Delta V mass spectrometer.
All δ 37 Cl measurements were normalised against at least two measurements of Atlantique 2 seawater [the internal seawater reference used by Godon et al. (2004) and since at IPGP] run each day under the same conditions of analyses as the apatite samples. All data were acquired on three consecutive days at IPGP. The uncertainty on each 10 3 δ 37 Cl data is AE 0.04 (1s) based on long-term δ 37 Cl measurements of Atlantique 2 (n = 150, and also checked on ten seawater analyses run over the course of this study). As a quality check for this normalisation, the 10 3 δ 37 Cl values on nine aliquots of the reference material ISL-354 averaged -0.01 AE 0.10 (Table 5), which is consistent with the recommended value (+0.05, Xiao et al. 2002). To further check for the quality of the whole procedure for the δ 37 Cl determination, we also ran five aliquots of seawater prepared with addition of Na 2 HPO 4 Á 2H 2 O salt and concentrated HNO 3 , and subsequently precipitated following the protocol used for the dissolved apatite samples (i.e., with the addition of AgNO 3 reactant only). The results of these tests are presented in Table S5.1 (Appendix S5). Total Cl amounts in all aliquots produced from the processed materials (the apatite samples and ISL-354) as well as in the reference seawater samples were within the range of 10-20 micromoles.
Finally, it is noteworthy that the calculated yields for the whole procedure (i.e., all steps of apatite dissolution, AgCl precipitation, CH 3 Cl fabrication, CH 3 Cl purification) averged 99 AE 9% (Table S5.2 in Appendix S5), reflecting an excellent match between the Cl content estimated by EPMA with those measured based on apatite dissolution at IPGP.
Dual-inlet isotope ratio mass spectrometry in laboratory 2: Chlorine isotope analyses at the University of Delaware (USA) were conducted using dual-inlet IRMS. Each apatite sample was dissolved entirely (50-580 mg) in 0.5 mol l -1 HNO 3 at room temperature. Dissolutions were performed for two separate units of TUBAF#50, and a single dissolution was performed for one unit of each of the other samples. The reference materials ISL-354 and NIST SRM 975 were dissolved in ultrapure de-ionised water. Two or three replicate measurements of each sample and reference material were performed on separate aliquots of solution. The measurements were performed in two measurement sessions a year apart using two different mass spectrometers and applying additional steps in the sample preparation procedure in the second session.
Enough sample solution to yield between 10 and 30 micromoles Cl was aliquoted into a 50-cm 3 centrifuge tube.
The aliquot volume was then adjusted to 10 ml by addition of ultrapure de-ionised water or by evaporation at 90°C. Each adjusted sample aliquot was then further acidified with 200 µl of concentrated HNO 3 and then 3 ml of 1 mol l -1 AgNO 3 was added. Capped centrifuge tubes were warmed in a 90°C oven for 30 min and then moved into a dark cabinet to cool and ripen overnight. AgCl solids were centrifuged at 2100 g for 3.5 min, and the supernatant solution was removed. The solids were rinsed three additional times with dilute HNO 3 followed by centrifugation. The solid AgCl was transferred into a pre-cleaned 9 mm diameter Pyrex combustion tube using a small section of a quartz fibre filter swab and small ultrapure water rinses. The AgCl was then evaporated to dryness at 90°C in a dark oven.
The apatite sample TUBAF#37 produced unusual, superfine precipitates that did not separate from solution even after 20-min centrifugation. This sample was left in a dark cabinet for 8 months. The finally clear supernatant was removed and analysis by ion chromatography (IC) showed negligible chloride in solution. The solids were re-dissolved using two drops of concentrated NH 4 OH and 10 ml deionised H 2 O. The chloride concentration of this re-dissolved solution was comparable to the original raw solution. The redissolved TUBAF#37 solution was then re-acidified, and 3 ml of 1 mol l -1 AgNO 3 was added. The resulting precipitates were the expected AgCl solids. Aliquots of four other samples measured in the second session (MGMH#133648, MGMH#128441A, TUBAF#38 and TUBAF#40) and ISL-354 were also treated in the same manner.
Under vacuum, an excess of CH 3 I (~200 micromoles) was cryogenically loaded into each combustion tube and sealed under vacuum with a methane-oxygen torch. Samples were then reacted at 300°C for 2 h (Holt et al. 1997). The mixture of CH 3 Cl and CH 3 I was chromatographically separated with a 6 m ¼" PoraPlot packed column with helium carrier (70 ml min -1 , 65°C). The CH 3 Cl peak was monitored by a thermal conductivity detector and trapped using liquid nitrogen, while the CH 3 I was back-flushed to waste and the helium carrier gas was pumped away.
In the first measurement session, the cryofocused CH 3 Cl was directly expanded into the dual-inlet bellows of a Thermo Finnigan Delta Plus XL IRMS via a series of switching valves. In the second session, the CH 3 Cl was cryogenically transferred into a 5-ml glass bulb with a PTFE HiVac valve. These samples were then manually introduced to the dualinlet bellows of a Thermo Scientific Delta V-Plus IRMS the following day. The same reference CH 3 Cl monitoring gas was used for both mass spectrometers. The m/z 52/50 ratios of the sample gas were determined relative to this reference gas with its δ 52 value set to 0.0. The ISL-354 reference material prepared along with the samples was used to calibrate results to the SMOC scale. The NIST SRM 975 reference sample was used as an additional quality control material (Table S5.3 in Appendix S5).
Continuous-flow isotope ratio mass spectrometry in laboratory 3: Chlorine isotope measurements at the University of Texas (Austin, USA) were conducted using continuous-flow isotope ratio mass spectrometry (CF-IRMS) following the methods outlined in Eggenkamp (1994) and modified by Sharp et al. (2007). Clwas extracted from approximately 15-100 mg of pre-weighed apatite via pyrohydrolysis, in which the sample was melted in the presence of an ultrapure de-ionised water vapour; Clwas then collected in 10 ml of 20 mmol l -1 NaOH (Schnetger andMuramatsu 1996, Shimizu et al. 2015). Two separate pyrohydrolysis extractions were prepared for each sample. Reference material ISL-354 was dissolved in ultrapure deionised water, and separate aliquots were analysed from the same solution. Once the Clwas in solution, samples were prepared following the methods outlined in Eggenkamp (1994) and as described in detail above. The AgCl was reacted with excess CH 3 I to produce CH 3 Cl, which was purified on a column filled with Porapak Q and introduced into a Thermo Electron MAT 253. Measurements were calibrated against three seawater reference samples. Yields, based on sample peak areas compared with reference peak areas of known Cl contents, were 104 AE 25%. Cl amounts in all aliquots produced from the processed materials were within the range of~0.8 to 2.4 micromoles. Samples were analysed during three separate measurement sessions.
This method's 10 3 δ 37 Cl uncertainties during the measurement sessions were AE 0.13 (1s), which is slightly better than the laboratory's long-term uncertainty of AE 0.20 (1s). The solution leftovers from the halogen extraction via pyrohydrolysis were also analysed by IC at the GFZ in order to determine Cl mass fraction and to assess yields of the Cl extraction process. Method description and the results of the IC measurements are included in Appendix S5.

Chemical composition
All apatite samples were investigated for their chemical composition using EPMA with two different electron beam current settings -5 nA (Table 2) and 40 nA ( Table 3) that enabled us to obtain high-quality information on the mass fractions of both volatile (F, Cl) and trace elements, respectively. No zoning or margin-to-core chemical composition variations were detected on the random fragments of crushed material that were investigated. For full data sets of the individual EPMA measurements, see Appendix S6.
Selected apatite samples cover almost the full compositional space of the F-Cl-OH system ( Figure 5) with the lowest Cl mass fraction determined for fluorapatite TUBAF#37 (0.27 AE 0.04% m/m, 1s) and the highest content in chlorapatite MGMH#133648 (6.34 AE 0.16% m/m, 1s). Due to volatile migration under electron beam influence (e.g., Goldoff et al. 2012), fluorine content was most probably overestimated in the data set obtained using 40 nA current, and therefore, further calculations of H 2 O content by difference and the structural formulae were carried out using exclusively our 5 nA data set ( Table 2). The chlorine mass fraction of MGMH#133648 varies notably depending on the electron beam current (median values are 6.34 AE 0.16% m/m (1s) and 6.09 AE 0.71% m/m (1s) for 5 and 40 nA, respectively). We observed grain-to-grain variations for MGMH#133648 (Appendix S6), which may suggest the crystal orientation effect previously noted in EPMA studies (e.g., Goldoff et al. 2012). Grain-to-grain variations were also observed for TUBAF#38, but Cl contents are similar with median values at 4.28 AE 0.12% m/m (1s) and 4.29 AE 0.19% m/m (1s) for 5 and 40 nA measurements, respectively. Both MGMH#133648 and TUBAF#38 contain OH-enriched areas identified in our SEM-based investigation of the inclusions, which could be the reason for grain-to-grain variations. It should be noted that these variations have not been detected in the 5-nA measurements, which were carried out at random locations on the very same crystal fragments as the 40 nA analyses. We note, however, that the sample mount was re-polished between the two sessions devoted to 5 and 40 nA measurements. Therefore, we advise using MGMH#133648 and TUBAF#38 with caution for chlorine abundance determinations due to a possible heterogeneous Cl/OH ratio in these two particular chlorapatite specimens.
The minerals of the apatite group can commonly incorporate different minor and trace elements into both cationic (e.g., La 3+ , Ce 3+ , Pr 3+ , Nd 3+ , Sr 2+ , Mg 2+ , Na + ) and anionic positions (e.g., SiO 4 4-, SO 4 2-, Pan and Fleet 2002). Accordingly, in the studied apatite samples various types and amounts of minor and trace elements were investigated. TUBAF#50 and MGMH#128441A are very rich in REEs, while MGMH#133648 and TUBAF#40 contain fewer such trace elements. Significant amounts of other components such as Sr, Mg and Na were detected in TUBAF#50. The 1 2 9 chemical composition of Dur (Durango, Mexico) closely matches the composition of other crystals from the same locality (e.g., Förster 2003, Marks et al. 2012), which confirms the validity of the EPMA protocols applied in this study.

SIMS homogeneity test of 37 Cl/ 35 Cl ratios
The homogeneity test conducted on random apatite fragments from one unit of each sample (Figure 2) confirmed that they are sufficiently homogeneous at the sub-nanogram sampling scale in terms of their chlorine isotope composition. The repeatability precision of ≥ 30 measurements performed on a minimum of fifteen fragments was better than AE 0.2‰ (1s, i.e., 1 standard deviation) for all samples (Table 4, Figure 6, Appendix S7); in fact, it was better than AE 0.1‰ for all five samples having > 0.5% m/m Cl. All specimens were tested during the same measurement session under the identical conditions, and therefore, the measurements yielded better counting statistics and repeatability for Cl-rich samples. A mean uncertainty of single analysis determined for each sample was between AE 0.01‰ (1s) for chlorapatite MGMH#133648 and AE 0.11‰ (1s) for fluorapatite TUBAF#37 (Figure 6). Our instrumental drift monitor analyses interspersed periodically with analyses of the studied apatite samples showed that the data were collected under stable conditions, and therefore, no offline drift corrections were necessary.
Additional tests were conducted on hand-picked grains of TUBAF#37 and TUBAF#38, which looked visually different under the binocular microscope. A few grains representing yellow and colourless crystals of TUBAF#37 as well as milky and transparent crystals of TUBAF#38 (Appendix S2) were investigated using SEM-EDS and further analysed by SIMS for their chlorine isotope compositions (Table 4, Figure 7, Appendix S7). BSE images collected for TUBAF#37 reveal that the yellow colouring results from a thin calcite layer, which covers some of the crystals, and which accordingly would have no impact on the 37 Cl/ 35 Cl ratios. Therefore, the chlorine isotope composition of yellow crystals in TUBAF#37 is consistent with the composition of colourless grains. Furthermore, SEM investigation showed that the white, usually opaque fragments of TUBAF#38 preserved traces of OH-replacement processes in this chlorine-rich sample. The 37 Cl/ 35 Cl ratios of OH-enriched areas determined by SIMS were measured and compared with the values collected for transparent, light-yellow crystals of TUBAF#38. This test showed only very small differences between the two crystal types (0.25‰ between two most extreme values; n = 64). In contrast, a similar test conducted on a piece of white crust removed from the surface of MGMH#133648 showed significant differences (up to 1.75‰, n = 60; Table 4, Figure 7, Appendix S7) between the most extreme value determined in OH-enriched area and the mean 37 Cl/ 35 Cl ratio for the volumetrically dominant, transparent, yellow crystals of MGMH#133648. The white crust was cut from the MGMH#133648 prior to crushing, but some minor amounts are still present in the split material. Therefore, we recommend that careful BSE and reflected-light optical observations of mounted polished crystals should be made prior to isotopic measurements so as to exclude any inclusions in all reference materials or OH-enriched areas in TUBAF#38 and MGMH#133648.

Crystallographic orientation effect on SIMS measured 37 Cl/ 35 Cl ratios
Minerals of the apatite group usually crystallise in the hexagonal crystal system (with space group P6 3 /m), but 1 3 0 lowering of the symmetry is possible due to extensive cationic and anionic substitutions in the lattice (Hughes and Rakovan 2002). The apatite crystal structure parameters readily respond to the various content of the anion position, which is parallel to the c-axis of the hexagonal structure (e.g., Hughes et al. 2016). Accommodation of monovalent anions such as Cland Fin the anion column entails migration and exchange of these elements not only during the postcrystallisation processes, but also under the influence of the electron bombardment during the quantitative measurements by EPMA (Stormer et al. 1993, Goldoff et al. 2012, Stock et al. 2015. It has been shown that the crystallographic orientation of apatite has a significant effect on the time-dependent X-ray intensities of fluorine and chlorine (e.g., Goldoff et al. 2012). Furthermore, previous SIMS works devoted to, for example, magnetite, haematite and sphalerite have also shown a significant crystal orientation effect on measured oxygen and sulfur isotope ratios (Huberty et al. 2010, Kita et al. 2011. Therefore, in this portion of our study three apatite fragments from a single crystal, but having differing orientations to the c-axis, were investigated by SIMS. A portion of an euhedral crystal of the Durango fluorapatite was cut approximately perpendicular, parallel and slantwise (bevel cut) to the c-axis anion channel of the apatite structure. The orientation of these crystal fragments was confirmed by EBSD mapping (Figure 8), which showed that Dur-A is perpendicular to the c-axis and is close to the (0001) plane, Dur-B is parallel to the c-axis (close to the ð1120Þ plane), while Dur-C has an intermediate orientation. 1 3 1 EBSD mapping of the Durango fragment cut slantwise to the c-axis channel (Dur-C) showed some variability in crystal orientation (visible as yellowish areas in the violet background in Figure 8). This is possibly an instrumental artefact related to the quality of the surface, but this interpretation needs further confirmation. Moreover, it was not possible to establish the Miller-Bravais indices for Dur-C due to the inconclusive results from EBSD point analyses; therefore, it is unclear which crystallographic plane is represented by the Dur-C fragment.
Repeatability of widely distributed SIMS analyses conducted on the individual oriented fragments was in all three cases better than AE 0.09‰ (1s, n = 10-18; Table 4, Appendix S7), proving their internal homogeneity, while the repeatability of the measurements of all three fragments was AE 0.16‰ (1s, n = 38). For the Cameca 1280-HR SIMS instrument operated at primary and secondary accelerating voltages of +10 and -10 kV, respectively, the incident Cs + beam angle is 21°from the normal to the sample surface (Huberty et al. 2010). Consequently, during the measurements of Dur-A and Dur-B the incident primary beam angle was about 20°and 70°from the c-axis channel, respectively. The SIMS analyses yielded higher (by 0.16%) 37 Cl/ 35 Cl measured ratios for Dur-A (0.32072 AE 0.00002, 1s) compared with the ratios measured in Dur-B

Conventional δ 37 Cl values
Conventional analyses (GS-IRMS) of stable chlorine isotopes have been successfully applied to various geological materials, mostly fluids, silicates and salts (e.g., Godon et al. 2004, Sharp et al. 2007, Bonifacie et al. 2007, Li et al. 2015, Eggenkamp et al. 2019. GS-IRMS data for the minerals of the apatite group are scarce (e.g., Kusebauch et al. 2015c), and the impact of choices made when designing an analytical protocol has not previously been established.
All apatite samples investigated in this study have comparable chlorine isotopic compositions to the mean For full data sets of all individual SIMS analyses, see Appendix S7. a Not corrected for IMF. b The data collected in February 2018: 10 3 δ 37 Cl SMOC values were calculated based on the correlation of α inst with Cl concentration (for further details, see Figure 10 and subsection 'Matrix effect and IMF'). The data collected on the oriented crystals in June 2017: 10 3 δ 37 Cl SMOC values were calculated based on the α inst determined using the mean 37 Cl/ 35 Cl ratio for Dur-A, Dur-B and Dur-C. The data reduction for all measurements was conducted using the absolute SMOC 37 Cl/ 35 Cl ratio of 0.319533 (zero point on the δ 37 Cl SMOC scale; see the third paragraph of Introduction for further details). c Instrumental drift monitor analyses regularly interspersed with analyses of the studied apatite samples. Dur-1, Dur-2 and Dur-3 indicate three randomly oriented fragments of the Durango apatite, which were used for monitoring the instrumental drift.
1 3 3 10 3 δ 37 Cl SMOC values clustering between +0.09 and +0.42, which is in contrast to a wider range of values represented by the two reference samples -Durango (+0.5) and synthetic chlorapatite (+2.0)characterised by Kusebauch et al. (2015a) or the Norwegian samples studied therein which span a range from -1.2 up to +3.7. Determined 10 3 δ 37 Cl SMOC values do not correlate with Cl mass fractions (Table 5, Figure 9). There are significant differences between individual data sets generated in three independent GS-IRMS laboratories. The offset of 10 3 δ 37 Cl SMOC values is not constant for all studied specimens, and it varies from 0.05 to 0.67. However, the values determined at the UT Austin are generally the highest, whereas those measured in IPGP are usually the lowest and the values reported by the Delaware laboratory in most cases fall between the data provided by the other two 1 3 4 laboratories. In contrast, for all three GS-IRMS laboratories the 10 3 δ 37 Cl SMOC values determined for sodium chloride ISL-354 are in close agreement with the published value of +0.05 (Xiao et al. 2002), which confirms that the analytical protocols for silver chloride precipitation and measurement were developed and applied properly in all three laboratories.
In general, continuous-flow IRMS determinations combined with pyrohydrolysis are characterised by higher uncertainty (AE 0.20 for UT Austin) than those conducted in dual-inlet IRMS mode combined with dissolution in HNO 3 (AE 0.04 for both IPGP and U. Delaware). This results not only from the principles of these two methods, but also from the fact that Cl concentrations in all aliquots produced from the processed materials via pyrohydrolysis (from~0.8 to 2.4 micromoles) were significantly lower than those obtained via apatite dissolution in HNO 3 (10-30 micromoles). The offset between laboratories is not observed for the sodium chloride ISL-354, but it should be noted that this reference material did not undergo pyrohydrolysis and was dissolved in deionised water in all three laboratories. This could further suggest that the process of pyrohydrolysis may have had a significant role regarding the variations in the determined δ 37 Cl SMOC values. However, the ion chromatography measurements conducted on the solution leftovers from the halogen extraction via pyrohydrolysis showed that Cl concentrations in the solutions produced from the apatite samples match the compositions determined by EPMA (Appendix S5, Table S5.4). In the case of the Cl-poor fluorapatite samples, the estimated yields are lower (e.g., 83-86% for TUBAF#37) compared with the yields for chlorapatite samples (e.g., 94-96% for MGMH#133648), but these differences do not correlate with the offsets between laboratories.
Furthermore, a set of seawater aliquots were prepared at IPGP with the addition of Na 2 HPO 4 Á 2H 2 O and HNO 3 ; the IRMS measurements yielded a mean 10 3 δ 37 Cl SMOC value of -0.05 AE 0.04 (1s; Appendix S5, Table S5.1), which is in very good agreement with the value for seawater (SMOC, 10 3 δ 37 Cl = 0; Brand et al. 2014). This suggests no influence of the phosphate matrix on the precipitation of AgCl and measured chlorine isotope compositions.

3 5
The δ 37 Cl measurements at the University of Delaware were performed in two measurement sessions a year apart using two different mass spectrometers and applying additional steps in the sample preparation procedure, which included an additional re-dissolution step of previously precipitated AgCl solids. The 10 3 δ 37 Cl SMOC values reported for these two sessions differ by only~0.1, which confirm the high reproducibility of these measurements, suggesting the mass spectrometer has little or no effect on the data. Additionally, the sodium chloride isotopic reference material NIST SRM 975 was measured at the University of Delaware as a secondary quality control material (Appendix S5, Table S5.3). The mean 10 3 δ 37 Cl SMOC value for two measurement sessions (+0.42 AE 0.04, 1s) agrees very well with the published 10 3 δ 37 Cl SMOC value of this reference sample (+0.43; Xiao et al. 2002).
Additional tests of the analytical procedures described above confirm the validity of our bulk methods and suggest that the variable offset between GS-IRMS laboratories results from apatite behaviour during the sample preparation. In general, an advantage of the used procedures is the fact that the chloride in all samples is in a pure silver chloride matrix which should (nearly) eliminate matrix effect issues during GS-IRMS measurements (Eggenkamp 1994). However, the samples that contain high sulfur mass fractions may precipitate Ag 2 S instead of AgCl when AgNO 3 is added to the solution. As a result, the AgCl yield could be lower but also (CH 3 ) 2 S could be produced during the reaction with CH 3 I, and this would impact the isotope ratio determination (Eggenkamp 2014). Two apatite samples TUBAF#37 and TUBAF#40 contain pyrite inclusions, which could not be fully removed from the split material (by mineral separation or hand-picking), and therefore, δ 37 Cl determinations for these two specimens may have been impacted by Ag 2 S precipitation. However, this could not be the case for other, sulfidefree samples (especially the inclusion-free TUBAF#50 specimen), which also show significant offsets between δ 37 Cl values reported by different laboratories. Also, we could not correlate the variations of minor and trace elements (including SO 4 2in the apatite structure) detected by EPMA with variable offsets between GS-IRMS data sets. This further  (Figure 10).

3 7
suggests more complex influence of chlorine extraction from apatite on the measured ratios and, more importantly, highlights the importance of the characterisation of reference materials for microanalysis with the support of multiple laboratories applying independent analytical protocols.

SIMS matrix effect and IMF
Matrix-dependent isotope mass fractionation is a common feature of isotopic studies of minerals crystallising as solid solutions (e.g., Śliwiński et al. 2016Śliwiński et al. , Siron et al. 2017, and in most cases two or more reference materials should be used for defining the IMF correction factor in order to assess the presence of any such effect. In the study by Kusebauch et al. (2015a), it was noted that a matrix correction should be applied to SIMS measurements conducted on calcium phosphates of the apatite group, in which F can be extensively substituted by Cl and OH. The apatite samples characterised in our study cover nearly the entire spectrum of F-Cl-OH compositions ( Figure 5). Based on our working values (Table 5) derived from GS-IRMS, the mean IMF value determined for the two endmembers TUBAF#37 and MGMH#133648 differs by~0.24‰ (Figure 10a), which is significant when compared with the small range of chlorine isotope ratios found in terrestrial materials, but it is rather negligible in view of larger variations of δ 37 Cl values determined by different GS-IRMS laboratories and possible crystallographic orientation effect in SIMS analyses.
SIMS data collected on our six reference materials and corrected for IMF assuming a best-fit linear mass fraction dependency of IMF correspond well with 10 3 δ 37 Cl values determined by GS-IRMS (Figure 10b). In contrast, the same SIMS data set corrected for IMF based on the analyses of the single fragment of Dur fluorapatite shows systematic offsets from the data reported by GS-IRMS (Figure 10c). The highest offsets were recorded for the two chlorapatite samples MGMH#133648 and TUBAF#38, for which 10 3 δ 37 Cl values obtained using SIMS were +0.10 (based on a linear correction of IMF) vs. -0.34 (when IMF correction was based on Dur only) for MGMH#133648, and +0.11 (based on a linear correction) vs. -0.22 (Dur correction) for TUBAF#38. GS-IRMS analyses yielded recommended 10 3 δ 37 Cl values of +0.09 for both chlorapatite specimens.
The SIMS IMF value of the Dur fluorapatite is slightly higher (by~0.2%) than IMF factors of other Cl-poor samples, which may contain a contribution from the observed crystallographic orientation effect; this IMF value was determined based on the measurements conducted on a single fragment of the Dur crystal (Dur-1 in mount #1), which was used for monitoring of the instrumental drift during the SIMS homogeneity test. Alternatively, a slightly different IMF may be a characteristic feature of Dur apatite. A higher IMF value for Dur was also noted in the additional test of selected crystals of TUBAF#37 and TUBAF#38 embedded in mount #2. In that case, the IMF value for a single crystal of Dur (Dur-2) was higher by 0.16‰ than a mean value for multiple crystals of TUBAF#37-B. The reason for IMF variations between RMs having low Cl mass fractions (≤ 1.4% m/m) is uncertain at the moment. Crystallographic orientation effect can be one of the reasons, but a variable offset between individual data sets generated in three independent GS-IRMS laboratories may also play a role here. Therefore, we recommend using multiple, randomly oriented fragments of at least two RMs to account for crystallographic orientation effect on collected 37 Cl/ 35 Cl ratios, thereby reducing the risk of potential bias associated with conventional δ 37 Cl determinations. Application of such an approach improved quality and validated the SIMS data in the previous study by Wudarska et al. (2020), in which the data reduction was conducted using four RMs. Furthermore, the IMF calibration based on apatite reference materials closely matching the composition of the 1 3 8 unknown samples would not only minimise the matrix effect but also enable the use of the analytical protocols tailored to the materials being studied. In the case where Cl-poor samples are analysed along with a RM having much higher Cl content, the current of primary ion beam should be limited to avoid too high ion counting rates for Faraday cups during RM measurements. Consequentially, measurement precision for the unknowns must be worse than that in the case where the applied primary ion beam current is optimised to the unknowns. In this study, all specimens were tested under the identical conditions, and therefore, the measurements yielded better counting statistics and repeatability for Cl-rich samples ( Figure 6).

Conclusions
Metrological splits of the apatite samples described here can be obtained through the IAGeo Limited website (www.ia geo.com). The 10 3 δ 37 Cl working values are as follows: MGMH#133648 = +0.09 AE 0.16 (1s), TUBAF#38 = + 0.09 AE 0.19, TUBAF#40 = +0.20 AE 0.24, MGMH# 128441A = +0.42 AE 0.20, TUBAF#50 = +0.32 AE 0.25, TUBAF#37 = +0.20 AE 0.13. All six reference materials are suitable for calibration of the in situ δ 37 Cl measurements. We recommend collecting BSE and reflected-light optical images of mounted polished crystals prior to isotopic analyses in order to avoid any possible inclusions and any crust that might be present. We also advise using MGMH#133648 and TUBAF#38 with caution for chlorine abundance determinations due to possible variations of Cl/ OH ratio distribution in these two specific chlorapatite specimens. Taking into consideration a potential crystal orientation effect on measured 37 Cl/ 35 Cl ratios and matrixdependent IMF, our recommendation for future SIMS-based δ 37 Cl SMOC determinations on minerals from the apatite group is to use multiple fragments of (at least two, but preferably three) reference materials matching as closely as possible the Cl content of the unknown samples. Due to the 1 3 9 offset between 10 3 δ 37 Cl values determined in three GS-IRMS laboratories, when interpreting SIMS data we recommend taking into account that the total uncertainty of SIMS results based on our reference samples would be at the level of AE~0.3, despite better capabilities offered by largegeometry SIMS instruments.