Supplementary Material

S1. Methods and analytical conditions for analysis of Si isotopes

S1.1 Samples and reagents

Samples and standards were processed using identical techniques throughout this research. The widely-accepted standard NBS28 (now called NIST RM8546; historically referred to as NBS28), a sample of quartz sand distributed by the National Institute of Standards and Technology (USA) was used as a bracketing standard. The external standards BHVO-2 (basalt, USGS) and Diatomite (University of California Santa, Barbara) were used to monitor data accuracy and check for instrumental mass-bias.

All acids used in this research were of analytical grade, purified at least once by sub-boiling in PTFE elbows. The water used in this study was deionised using a MilliQ-element (MQ-e) water polisher, providing high purity water with a resistivity of 18.2 MΩ cm -1. Samples were stored in pre-cleaned polypropylene bottles. Fusions were done in silver crucibles, made of 99.99% pure Ag sheet. The flux used was semiconductor grade 99.99% NaOH (Sigma Aldritch, in pellet form), which has a Si blank of around 5ppm. To clean the crucibles, they were first leached in weak, distilled HCl overnight. Following this, a blank fusion using an excess of NaOH was made in each crucible, after which the crucibles were rinsed in MQ-e and left overnight in another HCl leach.

S1.2 Alkali fusion

The alkali fusion method used in this research is described in detail by Georg et al. (2006). Around 10 mg of sample was weighed into a silver crucible. This amount was adjusted roughly, taking into account sample composition and dilution volume to yield a final solution concentration of between 10 and 20 ppm Si. Around 200 mg of NaOH flux was then added to the crucible; one pellet weighs roughly 100mg. NaOH flux is used as it recovers >95% of the sample during fusion and the chemical composition of NaOH is comparatively simple. Also, upon dissolution with HNO3, this flux adds only Na cations to the sample; hence, the amount of cations artificially added to the sample is kept at a minimum, which makes chemical separation via ion-chromatography relatively simple.

The crucible containing the sample and flux was placed in a muffle furnace (preheated to 720°C) for 12 minutes, which allows enough time for complete fusion to take place. After removal from the furnace, the crucible was left to cool slightly for 30 seconds and then transferred into 20 ml MQ-e water in a 30 ml PTFE vial, capped and left to equilibrate in the dark for approximately 24 hours.

Before transferring the sample into solution, the capped PTFE vial was placed in an ultrasonic bath and heated to ~ 60°C (as suggested by Zambardi and Poitrasson, 2011) to agitate the fusion cake. The sample was transferred into solution by 10 ml pipette into pre-cleaned PP bottles – the crucible was rinsed a number of times in MQ-e to ensure all sample was transferred. The solution was then acidified to 1% HNO3 v/v and left overnight to stabilise. HCl can also be used to acidify the sample solution and some research groups prefer this, due to the possibility that it may reduce the 14N16O+ interference on the 30Si+ beam (although most of the N and O is introduced into the instrument from the ambient environment). However, use of HCl could cause a drop in machine sensitivity, due to chlorine's high 1st ionisation potential (12.97 eV). In either case, work by

Armytage et al. (2011) shows that there is no isotopic bias introduced by storing samples in either HNO3 or HCl; our choice of HNO3 is merely historical.

The Si concentration in each solution was checked on a Thermo Scientific Evolution 60 desktop photospectrometer using the “Heteropoly-Blue” method with the Hach-Lange Ultra low range Si reagent kit (http://www.hach.com). For samples with pre-existing SiO2 (wt.%) measurements, typical Si fusion yields were between 95 and 99%.

S1.3 Ion exchange chromatography

To purify samples for Si isotope analysis, we employed a one-step column procedure using BioRad AG50 X-12 (200-400 mesh) cation exchange resin. The resin was precleaned in a PTFE container before it is loaded, using the method given in Table S1. Standard BioRad Poly-Prep columns with a 2 ml bed volume and 10 ml reservoir volume were used, which were cleaned in 1% HNO3 v/v. A resin bed volume of 1.8 ml was packed into each column and the resin was further cleaned using the method described in Table S1. Before any sample was loaded, the resin was cleaned twice and between samples it was cleaned once. The resin is robust enough to continue performing after 4–5 sets of samples and cleaning acids have passed through the column.

This simple purification method is possible because, between pH2 - pH8, Si is in the form Si(OH)4 in equilibrium with the anionic species H3SiO4

- (the prevailing anionic species in dissolved silicate samples). Most matrix elements are cationic and are quantitatively retained by the resin as the sample passes through the column. Column calibration curves show that only anionic species such as SO4

2- and NO3- are present in the elute (Georg et al., 2006; even the large amount of Na,

introduced by the fusion procedure, was quantitatively stripped from solution).

Sample solutions were loaded directly on to the column and are collected immediately in 15 ml PTFE vials. The amount of sample solution was calculated such that 30 μg of Si is loaded on to each column, which will result in final Si running concentrations of 3.0 ppm when diluted to 10ml. The Si was then eluted in MQ-e, the volume of which is at least twice the resin volume, i.e. 5 ml. An example column recipe is given in Table S2.

Total procedural blanks, measured on the Neptune, gave a total maximum blank of 0.1 μg Si, which, at 0.35% of the total signal, is negligible. The isotopic composition of the blank was impossible to measure effectively, due to the extremely low concentrations.

S1.4 Standard addition

To test for the presence of matrix effects, we performed a standard addition test on the EH chondrite Qingzhen. The theory and rationale behind this technique is described in Tipper et al. (2008). Four mixtures were prepared, with varying fractions of Si from the standard, ranging from 0.14 to 0.79 (Table S3). The Si isotope data plot on a linear array with intercept of -0.05 ± 0.07 ‰ (NBS28 = 0 ‰) and the regression gives a calculated Qingzhen composition of δ30Si = -0.85± 0.14 ‰. This is almost identical to the measured composition of Qingzhen (Table 1). The linearity of the data, zero intercept and comparable calculated composition of the standard indicate that the sample has not been compromised by matrix effects. See also the discussion in Section 4.1 of the main text.

S1.5 Mass spectrometry

Silicon isotope measurements were made on a Neptune Plus (Thermo Fischer Scientific, Bremen, Germany) Multi-Collector Inductively-Coupled-Plasma Mass-Spectrometer (MC-ICP-MS) at Washington University in St. Louis. The instrument was operated at medium resolution (M/ΔM ~ 7500 where ΔM is measured at 5% and 95% of peak height) to avoid polyatomic interferences (e.g. 12C16O+, 28SiH+, 14N16O+) – a typical peak shape, showing the Si beams, approximate position of measurement, interferences and mass resolution, is given in Figure S2. The isotope beams 28Si, 29Si and 30Si were measured using Faraday cups L3, C and H3 respectively. Samples were introduced into the plasma via a 100 μl min-1 PFA ESI (Elemental Scientific, Omaha, USA) microflow nebuliser and a double-pass glass cyclonic spray chamber. Depending on instrumental conditions, a total analyte signal of 7.0 V was common, giving beam strengths of 29Si ~ 350 mV and 30Si ~ 250 mV. Instrument background signal (typically ~10 mV total Si) was measured at the beginning of each analysis session and the subsequent sample measurements were corrected using this data. Analytical settings are listed in Table S4.

Isotope ratios were measured in static mode, with each measurement consisting of 25 cycles of 8.4 second integrations, with a 3 second idle time. Ratios were calculated in the Thermo Neptune Data Evaluation software, which discarded any outliers at the 2 sigma confidence level. Absolute errors on the 29Si/28Si and 30Si/28Si ratios were ~7.5 × 10-7 and ~7.0× 10-7 respectively, which equates to 0.015‰ and 0.018‰.

To correct for instrumental mass bias, isotope measurements were calculated using the standard-sample bracketing protocol relative to the NBS28 standard. Variations in Si isotopes are defined using the delta notation (δ30Si or δ29Si) as the per mil (‰) deviation from the standard, as such:

δxSi = [(xSi/28Sisample / xSi/28SiNBS28) – 1] × 1000;

where x = 29 or 30. We discuss the Si data in terms of δ 30Si values which are approximately twice that of δ29Si (no mass-independent Si isotope variations have so far been measured in bulk meteorites). Figure S3 shows a 3-isotope plot of all Si isotope data from this study (excluding the highly-fractionated Diatomite) and shows that all data is mass-dependent within the 95%s.e. error of each sample – the average deviation from mass dependence is 0.00 ± 0.02 ‰ amu-1, with a maximum deviation of 0.03 ‰ amu-1.

Each measurement session consisted of 10 samples, two of which were always the external standards BHVO-2 and Diatomite. All samples (including the standards) were analysed once in a sequence, bracketed by NBS28, which was then repeated 3 to 6 times (e.g. NBS-std1-NBS-std2-NBS-sample1-NBS-sample2-NBS…) such that a typical measurement session lasted between 8 and 16 hours. Instrumental drift over this time period can be significant, as shown Figure S4, but the drift is typically linear such that the standard-sample bracketing technique is sufficient to correct for this. Calculating the δ30Si value of NBS28 using adjacent bracketing standard measurements for the two analysis sessions shown in Figure S4 gives δ30Si= 0.00 ± 0.09 ‰ (2s.d., n=29) for April 5th 2012, and δ30Si= 0.00 ± 0.10 ‰ (2s.d., n=60) for April 20th 2012.

To assess method accuracy, and to ensure inter-laboratory comparison, we routinely analysed the well-characterised geo-standards Diatomite (Reynolds et al., 2007) and BHVO-2 (Abraham et al., 2008) alongside samples. Our data for these standards (δ 30SiDiatomite = 1.22 ± 0.10 ‰; δ30SiBHVO-2 = -0.28 ± 0.09 ‰; 2 s.d.; Table 1) are in very good agreement with the literature values, indicating that our methods are robust.

S2. Kamacite δ30Si calculations using Monte Carlo simulations

To calculate a robust set of estimates and associated errors for the Si isotope composition of kamacite, we employed a series of Monte Carlo simulations. In particular, we employed the ‘ randn’ function in MatLab to generate an array of pseudorandom values, following a normal distribution, and used these to generate an array of values centred about each variable in the calculations.

The kamacite Si isotope composition was calculated using the mass balance equation:

δ 30 Sikamacite=δ30Sibulk−( f Sisilicate×δ

30 Sisilicate)1−f Si silicate

where fSisilicate is the fraction of Si in the meteorite in the silicate phase, as calculated by:

f Sisilicate=χ silicate× [Si ]silicate

χsilicate× [Si ]silicate+ χkamacite× [Si ]kamacite

where [Si]x is the concentration of Si in phase x, and χx is the modal abundance of phase x. In the first set of values in Table 2, χkamacite = abundance of metal, whereas in the second set χ kamacite = abundance of metal + troilite – this is assuming that the troilite formed by later sulphidisation of the metal grains (i.e. Rubin, 1984).

The Monte Carlo simulations consisted of 10000 iterations of the formulae above, using a series of variables calculated as a randomised normal distribution about the mean values. The standard deviation from the mean of each variable was assigned as follows:

δ30Si values: measured standard deviation of each measurement, as given in Table 1; χx values: assuming a 2sd of ± 0.05 modal abundance; [Si]x values: assuming a ±10% error on the Si measurements.

A histogram of the data output from a simulation for the EH4 chondrite Indarch, is given in Figure S5.

References

Abraham, K., Opfergelt, S., Fripiat, F., Cavagna, A. -J., de Jong, J. T. M., Foley, S. F., André, L., Cardinal, D., 2008. δ30Si and δ29Si Determinations on USGS BHVO-1 and BHVO-2 Reference Materials with a New Configuration on a Nu Plasma Multi-Collector ICP-MS. Geostand. Geoanal. Res. 32, 193-202.

Armytage, R. M. G., Georg, R. B., Savage, P. S., Williams, H. M., Halliday, A. N., 2011. Silicon isotopes in meteorites and planetary core formation. Geochim. Cosmochim. Acta. 75, 3662-3676.

Georg, R. B., Reynolds, B. C., Frank, M., Halliday, A. N., 2006. New sample preparation techniques for the determination of Si isotopic compositions using MC-ICPMS. Chem. Geol. 235, 95-104.

Reynolds, B.C., Aggarwal, J., André, L., Baxter, D., Beucher, C., Brzezinski, M.A., Engström, E., Georg, R.B., Land, M., Leng, M.J., Opfergelt, S., Rodushkin, I., Sloane, H.J., Van Den Boorn, S.H.J.M., Vroon, P.Z., Cardinal, D., 2007. An inter-laboratory comparison of Si isotope reference materials. J. Anal. At. Spectrom. 22, 561–568.

Rubin, A.E., 1984. The Blithfield meteorite and the origin of sulfide-rich, metal-poor clasts and inclusions in brecciated enstatite chondrites. Earth Planet. Sci. Lett. 67, 273–283.

Tipper, E. T. and Louvat, P. and Capmas, F. and Galy, A. and Gaillardet, J. (2008) Accuracy of stable Mg and Ca isotope data obtained by MC-ICP-MS using the standard addition method. Chemical Geology, 257 (1-2). pp. 65-75.

Zambardi, T. and Poitrasson, F., 2011. Precise Determination of Silicon Isotopes in Silicate Rock Reference Materials by MC-ICP-MS. Geostand. Geoanal. Res. 35, 89-99.

Table S1 – Ion chromatography resin pre-cleaning and cleaning procedures.

Resin pre-clean procedure Resin cleaning procedureLiquid Notes Liquid

2x MQ-e 10ml 3N HNO3

1x 6N HNO3 6ml 6N HNO3

1x MQ-e 5ml 6N HCl1x 6N HNO3 3ml 6N HNO3

1x MQ-e 3ml 3N HNO3

1x 1:1 v/v HNO3 leave for 30 mins 10ml MQ-e2x MQ-e 3ml MQ-e1x 1:1 v/v HNO3 leave for 30 mins 3ml MQ-e2x MQ-e1x 6N HNO3 leave overnight1x MQ-e Store

Table S2 – Example column recipe.

Example conditionsConcentration required 3.0 ppmSolution concentration 20.0 ppmFinal sample volume 10.000 ml

Column recipeSample solution 1.500 mlMQ-e 5.000 ml (added in two steps; 2+3ml)

MQ-e added to elute 3.357 mlAcid added to elute (70% HNO3) 0.143 ml (1% v/v HNO3)

Table S3 – Standard addition experiment data.

NBS28 18.34 ppm StandardQingzhen 12.18ppm Sample

% of solutionMixture NBS28 Qingzhen ppm f(sample) δ30Si

(‰)2s.d. δ29Si

(‰)2s.d. n

1 0.15 0.85 13.104 0.79 -0.69 0.06 -0.38 0.08 32 0.30 0.70 14.028 0.61 -0.52 0.08 -0.28 0.06 33 0.50 0.50 15.260 0.40 -0.38 0.11 -0.18 0.13 34 0.80 0.20 17.108 0.14 -0.16 0.05 -0.10 0.08 3

Regression

δ30Si vs. f 2sd δ29Si vs. f 2sd

Slope -0.81 0.120 -0.38 0.170Intercept -0.05 0.064 -0.06 0.095

Qingzhen compositionδ30Si 2sd δ29Si 2sd

Calculated -0.85 0.14 -0.44 0.19Measured -0.82 0.11 -0.41 0.06

Table S4 – Neptune Plus MC-ICP-MS typical analysis settings.

RF forward power 1100 WExtraction -2000 VFocus -664 VSkimmer cone Standard “A”-type, NickelSlit resolution MediumResolving power ~7500 (M/ΔM)Faraday cup setup 28Si (L3); 29Si (C); 30Si (H3)Measurement mode StaticSample gas 1.1 l min-1 ArCoolant gas 15.0 l min-1 ArAuxiliary gas 0.70 l min-1 ArRinse time 150 sUptake time 50 sAcquisition time 285 sRunning concentrations 3.0 ppmNebuliser flux ~ 100 μl min-1

Figure S1 – Results of the standard addition experiment.

Silicon isotope measurements of the standard:sample mixtures prepared for the standard addition experiments – data from Table S3. The regression statistics were calculated using IsoPlot; errors were 2sd of the isotope measurements, and a fixed 2sd of 0.025 for the calculated fraction of sample, based on errors of the Si concentration and introduced during mixing. The orange diamond is the measured composition of Qingzhen, as given in Table 1.

Figure S2 – Mass scans of Si on the Neptune Plus at Washington University.

Figure S3 – 3-isotope plot of all Si isotope data from this study.

Three isotope plot of all data from this study, minus the highly fractionated Diatomite geostandard. Error bars are 95%se. The lines are the calculated mass dependent fractionation lines for kinetic and equilibrium fractionation (b = slope). No offset within error from either fractionation line is apparent for any of the data.

Figure S4 – Mass bias drift over the period of a two typical Si isotope measurement sessions.

Figure S5 – Example of the data output from the Monte Carlo simulation for the Si isotope composition of kamacite in EH4 Indarch.