Chemical Geology 345 (2013) 50–61

Contents lists available at SciVerse ScienceDirect

Chemical Geology

j ourna l homepage: www.e lsev ie r .com/ locate /chemgeo

MC-ICP-MS measurement of δ34S and Δ33S in small amounts ofdissolved sulfate

Guillaume Paris ⁎, Alex L. Sessions, Adam V. Subhas, Jess F. AdkinsDivision of Geological and Planetary Sciences, California Institute of Technology, 1200 E California Blvd, Pasadena, CA 91125, USA

⁎ Corresponding author. Tel.: +1 626 395 3765.E-mail address: gparis@caltech.edu (G. Paris).

0009-2541/$ – see front matter © 2013 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.chemgeo.2013.02.022

a b s t r a c t

a r t i c l e i n f o

Article history:Received 16 November 2012Received in revised form 16 February 2013Accepted 19 February 2013Available online 5 March 2013

Editor: U. Brand

Keywords:Sulfur isotopesMC-ICPMSTrace sulfateMass independent fractionation

Over the last decade, the use ofmulti-collector inductively coupled plasmamass spectrometry (MC-ICP-MS) hassignificantly lowered the detection limit of sulfur isotope analyses, albeit typically with decreased precision.Moreover, the presence of isobaric interferences for sulfur prevented accurate analysis of the minor isotopes33S and 36S. In the present study, we report improved techniques for measuring sulfur isotopes on theMC-ICP-MS Neptune Plus (Thermo Fischer Scientific) using a heated spray chamber coupled to a desolvatingmembrane (Aridus, Cetac). Working at high mass resolution, we measured δ34S values of natural samples witha typical reproducibility of 0.08–0.15‰ (2sd) on 5 to 40 nmol sulfur introduced into the instrument. We appliedthis method to two seawater profiles, using 25 μl of sample (700 nmol of sulfate). The average δ34SVCDT value is20.97 ± 0.10‰ (2sd, n = 25). We show that the amount of sulfate required for an analysis can be decreased to5 nmol. Because the plasma is sustained by Ar, measurement of 36S is impossible at the current mass resolutiondue to the presence of 36Ar+, but a reproducibility of 0.1–0.3‰ (2sd) is achieved on the measurement of massindependent fractionations (Δ33S). This is the first time such precision has been achieved on samples of this size.

© 2013 Elsevier B.V. All rights reserved.

1. Introduction

Sulfur isotopes (32S, 33S, 34S, 36S) are valuable tracers in many(paleo)environmental studies, but several analytical limitationsremain that limit their utility. Sulfur isotopic compositions were origi-nally measured on gas source isotope-ratio mass spectrometers(GS-IRMS) as SO2 (Wanless and Thode, 1953; Thode et al., 1961) orSF6 (Hulston and Thode, 1965). These methods require extraction of afew μmol of sulfate as a solid (BaSO4), which is then combusted toSO2, to make a measurement with a precision down to 0.05‰. Whilethis amount of sulfur is readily achieved in many marine samples, itmay become limiting in samples with low sulfate (or sulfur) content,such as river waters, aerosols, ice cores, Precambrian carbonates, andforaminiferae. Moreover, the need to correct for 17O and 18O isobarsin themass spectrum of SO2 further limits accuracy and precision, espe-cially for 33S and 36S abundances. The use of SF6 as an analyte gas hasthree main advantages over SO2. Unlike oxygen, fluorine is monoisoto-pic so there are no isobaric interferences and the entire S isotopicsystem can be measured at high precision with little to no memoryeffect (Hulston and Thode, 1965). However, the SF6 method requiresthe use of a fluorination vacuum line, which is not widely available,restricting this method to specialized laboratories. The introduction ofisotope-ratio-monitoring (‘continuous flow’) techniques (IRM) hasallowed researchers to measure samples as small as 200 nmol S byin-situ fluorination (Ono et al., 2006a, 2006b). Secondary ion mass

rights reserved.

spectrometry (SIMS) is also suitable for in-situ analysis of sulfide min-erals (Deloule et al., 1986; Chaussidon et al., 1989) and kerogen sulfur(Bontognali et al., 2012).

Because of major interferences of O2 and molecular hydrides ateach mass of its isotopes, sulfur was not analyzed by inductivelycoupled plasma mass spectrometry (ICP-MS) before the late 1990s,though this widespread technique had the potential to surmountmost of the previously described difficulties and appears especiallysound for sulfate in solution or trace sulfate in carbonates. To avoidthe cardinal mass interferences, the isotopic composition of sulfurwas first measured on a quadrupole ICP-MS using masses 48 (32SO+)and 50 (34SO+) to get 34S/32S ratios. Precision on the ratios was betterthan 10‰ (1sd). Elemental S+ ions were first analyzed directly usingan RF-only hexapole ICP-MS where the collision cell was filled with amixture of Xe and H2 to decrease the S+/O2

+ ratio. This method resultedin precisions of better than 3‰ (1sd) for solutions containing between300 and 1500 μmol S/l (Prohaska et al., 1999). Combining a double-focusing sector field mass spectrometer and a desolvating membraneto decrease (16O)2 interference on mass 32, Prohaska et al. obtained34S/32S ratios with precision of 0.4‰ (1sd) using less than400 nmol S, and precision of 1‰ using 4 nmol S for one measurement(Prohaska et al., 1999). Multicollector ICP-MS (MC-ICP-MS) allows forsimultaneous collection of 32S+, 33S+, 34S+ and 36S+, decreasing theeffect of fluctuations in the plasma on the measured ratios (Cloughet al., 2006). Isotopic compositions are usually reported using the δ no-tation (‰): δ'3xS=1000×[(3xS/32S)sample/(3xS/32S)standard–1]. Morerecently, Craddock et al. (2008) developed bulk and in-situ (by laser ab-lation) measurements of δ34S on the MC-ICP-MS Neptune (Thermo

Filter Cleaning

Add HNO

3+HCl

Dissolution in HCl

EvaporateCentrifugation,

supernatant recovery

Dilute in HCl 0.25%

Aliquot Neptune/IC

Run on the Neptune

Dilution in H

2O

Run on the IC Drying

Dilution in HNO

3

Add NaCl

Waters Carbonates

DX500 Neptune

Purification on Bio-Rad AG50X8

Fig. 1. Chart flow of the protocol described in the paper.

51G. Paris et al. / Chemical Geology 345 (2013) 50–61

Fischer Scientific). Because of the 32S1H+ interference at mass 33, theycould not measure δ33S values with high precision however, and there-fore did not report data for the mass-independent fractionation tracerΔ33S (Craddock et al., 2008). Δ33S is defined as δ'33S–0.515×δ'34S,where δ'3xS=1000×ln[(3xS/32S)sample/(3xS/32S)standard] (Hulston andThode, 1965). The use of MC-ICP-MS for S isotopes is becoming morewidespread. Due to high useful ion yields (which combines ionizationefficiency and sample usage) and low detection limits, sample sizesare order(s) of magnitude smaller than for SO2 and SF6 methods,down to 60 nmol with reproducibility of ~0.3‰ for δ34S (Das et al.,2012). In many cases, the use of MC-ICP-MS allows higher throughputand requires less workup (e.g., no need for precipitation of dissolvedsulfates or sulfides, no combustion to SO2 or fluorination to SF6). Work-ing at higher mass resolution can further reduce these interferences. Incombination, these aspects ofMC-ICP-MSprovidemuch better accuracythan SO2 measurements for both 33S/32S and 34S/32S ratios. 36S is notmeasureable by ICP-MS due to a severe interference by 36Ar.

Herewe further push the limits of precision and sensitivity for sulfurisotope analysis on the MC-ICP-MS Neptune Plus (Thermo FischerScientific), and analyze the current limitations to our precision. Follow-ing the advent of GC–ICP-MS measurements of single organiccompounds for their δ34S values at Caltech (Amrani et al., 2009), wepresent the development and application of trace sulfate analyses onthe Neptune Plus and apply it to seawater profiles. Three points areassessed in this paper: purification of the samples, a decrease of isobaricinterferences on the Neptune and measurement of 33S, and an increaseof sensitivity and precision compared to previous MC-ICPMS studies.Because analysis of 33S on the Neptune is a new aspect of this work,we explore the limitations encountered by measuring this isotopeabundance. Themethod developed in this paper is a novel and straight-forward way to analyze δ33S, δ34S and Δ33S from carbonate-associatedsulfate (CAS) or trace dissolved sulfate in waters down to ~5 nmol S.

2. Description of the method

2.1. Wet chemistry

Three types of sample have been analyzed; solid BaSO4, dissolvedsulfate in natural waters, and CAS. Each kind of sample is processeddifferently to the point of yielding dissolved aqueous SO4

2−; from thatpoint on, all are treated identically. Samples are typically split for mea-surement of both sulfate concentration by ion chromatography (IC) andsulfur isotopic composition by MC-ICP-MS (Fig. 1). Except if otherwiseindicated, all acids are ultrapure ‘Baseline’ grade (Seastar ChemicalsInc.).

2.1.1. BariteBarites are dissolved by sitting at room temperature for three days in

a solution of 50 mmol/l of diethylene triamine pentaacetic acid (DTPA)and NaOH and manually shaking the vials regularly. Samples are left inthe solution for three days and then purified according to the protocoldescribed in Section 2.2. The final volume should not be greater thanthe volume of resin used for purification of the sample.

2.1.2. Aqueous sulfateWater samples are aliquotted for measuring concentration and iso-

topic ratio. Ideally, 200 to 500 μl of sample with a sulfate concentrationbetween5 and 25 μmol/l are introduced on the IC. Sulfate extraction forisotopic measurement requires at least 5 nmol S. Samples are acidifiedwith one drop of 10% HCl and one drop of 5% HNO3, evaporated todryness at 100 °C, and then rediluted in 0.25% HCl. Samples are thenaliquotted and purified.

2.1.3. Carbonate samplesCarbonate samples are weighed, cleaned and ground to a fine

powder. The deep-sea coral analyzed for this work is both physically

and chemically cleaned before dissolution. After removing the exter-nal part of the coral with a Dremel tool, it is ground and rinsed threetimes in MilliQ water. Corals are then left for 12 h in a 1:1 mixture of30% H2O2 and 0.1 M NaOH solution, after which they were rinsed 3times in MilliQ water.

After cleaning, the sample is dissolved in 5% HCl. An aliquot issampled for concentration and another one for isotopic composition.They are independently dried down and the IC aliquot is rediluted inwater while the isotopic composition aliquot is rediluted in 0.25% HCl.

2.2. Purification

Our first attempt to purify the samples consisted of running themthrough a Dionex Anion Self-Regenerating Suppressor membrane, inorder to extract cations from the solutions. Quantitative sulfate recov-ery (>95%) and complete purification of the sample have never beenachieved. A second attempt using the strong anionic resin AG1X8(Bio-Rad) was also performed, however, its affinity for sulfate madeits recovery difficult for low-concentration samples. Instead, samplesare thus purified using the strong cationic resin AG50X8 (Bio-Rad).Active exchange sites are sulfonyl groups, potentially contributingto the sulfur blank, though this resin has been previously used for pu-rification of sulfate minerals (Craddock et al., 2008). The resin is batchcleaned by rinsing it first in MilliQ H2O three times to remove smallerparticles. It is then rinsed in 8 N reagent grade HNO3 and then alter-natively rinsed in 10% HCl and MilliQ water, where each rinse is leftfor at least 3 h on a shaking table.

As the capacity of the resin is 1.7 meq/ml, varying cation/SO42−

ratios in the samples require variable amounts of resin. The requiredamount of resin is introduced on columns made of shrink-wrap PTFEtubing (Small Parts, Inc.). The column is rinsed with MilliQ H2O,acetone, MilliQ H2O again and 10% HCl before loading the resin. Once

Table 1Neptune parameters.

Neptune parametersRf power 1200 WCool gas 15 l/minAux gas ~0.8 l/minSample gas ~0.8 l/min

AridusSpray chamber temp. 110 °CDesolvating membrane temp. 160 °CAr sweep gas 4 l/minN2 flow 0.02 l/minNebulizer PFA-50 (ESI)Flow rate 60 μl/minCones Al X-cones

Measurement parametersCup configuration 32S (C), 33S (H1), 34S (H3)Resolution mode High (15 μm entrance slit)Acquisition 50 blocksIntegration time 4.194 sUptake time 60 sVolume of sample 270 μlWash + background measurement 13 min

52 G. Paris et al. / Chemical Geology 345 (2013) 50–61

loaded, the resin is rinsed twice with 20 column volumes (CV) of MilliQH2O and 20 CV of 10% HCl. The resin is conditioned with 3 × 5 CV of0.25% HCl. The sample is introduced and the resin is rinsed with3 × 1 CV of 0.25% HCl to ensure complete elution of sulfate. The finalvolume is then dried down and diluted in 5% HNO3 to reach a solutionof known sulfate concentration. NaCl is added to the final solution tomatch the concentration of sodium in the sample to the bracketingstandard. The sample is then ready to run on the Neptune. The yieldof this procedure has been checked by measuring concentrations ofseawater samples on the IC before and after purification. The measuredyield is 103 ± 5%. The total procedural blank depends on the size ofthe column, but for the columns used in this study (20 μl of resin) themeasured blank ranged from less than 0.1 to 0.3 nmol SO4. This rangeis partially explained by the small amounts of sulfate, which are closeto the detection limit of the IC.

Teflon vials used in this study are rinsed 5 times in MilliQ H2O,heated overnight in 8 N reagent grade HNO3, rinsed 5 times withMilliQ H2O, refluxed for 24 h in Seastar HNO3 and rinsed 5 timeswith MilliQ H2O. Vials have to be handled carefully because gloveshave been found to be a source of sulfur contamination. Neptuneautosampler vials are washed in 10% reagent grade HCl and rinsedfive times in MilliQ H2O. The resin is not recycled and columns arestored in 10% reagent grade HCl.

2.3. Description of the IC

Sulfate concentrations are measured using a DX500 Ion Chromato-graph (Dionex). The system is comprised of an AS40 autosampler, anEG40 eluent generator equippedwith a KOH eluent generator cartridgeEGC III KOH, an LC20 Analyzer equipped with an IonPac AS19 inorganicanion column and an ED40 electrochemical detector. Samples are intro-duced using 500 μl Dionex vials. They are injected via a 10 μl loop,mixed with in-situ generated KOH eluent and pushed through theguard column and then the analytical column that separates the anions.The sample flows through the Self Regenerating Suppressor to removethe eluent and enhance the detection of the analyte. Single injectionsare performed on each vial. Standards, and some samples, are analyzedin triplicate to evaluate reproducibility of the results.

A gradient of KOH eluent concentration is applied. The concentra-tion is 5 mmol/l for the first 5 min, then increases linearly from 5 to45 mmol/l between 5 and 25 min at a constant pressure (2200 psi).The area of each peak is directly proportional to the concentration ofthe anion of interest. Data are manually processed to ensure that thebaseline of each peak is accurate. Calibration standards are run duringeach session to ensure the validity of the calibration curve.

2.4. Description of the Neptune

Samples are analyzed for isotopic composition using a ThermoFischer Scientific MC-ICPMS Neptune Plus equipped with nine faradaycups. Typical settings are shown in Table 1. Samples are introduced asNa2SO4 in 5% HNO3 (Seastar) using a ESI PFA-50 nebulizer (with ameasured uptake rate of about 60 μl/min) and a Cetac Aridus desolvatorthat utilizes a PTFE membrane at 160 °C to remove solvent vapor. Atthe start of each run, the Neptune ion source is tune to: i) optimize sen-sitivity, ii) decrease isobaric interferences, and iii) increase the signalstability. Torch position is optimized first, followed by an iterative opti-mization of the various gas flows (sample gas and auxiliary gas on theNeptune, sweep gas and additional gas on the Aridus). The optimalposition and alignment of the peaks are determined first visually onthe interference-free shoulders and confirmed by collecting 34S/32Sand 33S/32S ratios to evaluate the Δ33S value. The position is chosen sothatΔ33S is 0‰within±0.1‰, ensuring that interferences do not affectthe signals measured on the collected masses. Isotope ratios are mea-sured in high resolution mode to allow optimal mass resolution, withM/ΔM typically between 8000 and 10,000 using the 5%–95% valley

definition (Weyer and Schwieters, 2003). The detection of 32S, 33S and34S is affected by isobaric interferences created by various isotopologuesof O2 and SH. Because of the major interference on 36S from 36Ar, thisisotope is not measured. Interferences are heavier than the S+ ions ofinterest, leaving an interference-free shoulder on the low-mass side ofthe peak (Fig. 2), similar to previous observations (Craddock et al.,2008). We measure 32S in the central faraday cup because it decreasesthe presence of a reflection on the low mass side of the peak. Theheavy isotopes 33S and 34S are measured in faraday cups H1 and H3,respectively. The signals from the three cups are registered throughfeedback electrometers with 1011 Ω resistors. Both 32S and 34S peaksare lined up on the left of the sulfur plateau and 33S is slightly offset.The relative position of the central cup is chosen such that the signalfor 33S is measured in the middle of the interference free plateau,which is 4 mDa wide.

Isotope ratios are collected in 50 cycles of 4.194 s integration timeeach. A sulfate concentration of 20 μmol/l yields a signal intensity of~10 V for 32S+, corresponding to 1.3·1011 collected ions. A run usesabout 0.3 ml of sample, which corresponds to ~3.06·1015 ions intro-duced. The useful ion yield is thus 0.036‰, somewhat lower than typ-ical for the ICP-MS, which is partially explained by the relatively highenergy of first ionization for sulfur and the use of high resolution.

Instrumental background is measured after each sample orbracketing standard and is subtracted from the average signal mea-sured on each cup. Each sample or bracketing standard is analyzedusing a Matlab script that removes outliers and/or problematic por-tions of the 50 cycles. Typically, sections of a run are removed whenthe signal is not stable at the start of the run, and/or the sampleruns out of solution before the 50 cycles are complete.

Mass drift is corrected using a bracketing standardmade of dissolvedreagent-grade Na2SO4. Both 34S/32S and the 33S/32S ratios are correctedindependently assuming a linear variation between two consecutivebracketing standards. The measured ratio for sample j (Rmj) is driftcorrected using Eq. (1):

Rcorr j ¼Rm j

αj−1−αjþ1

� �=2

h iþ 1

ð1Þ

with αi ¼Rmi−Rtrue

Rtrue.

Sign

al in

tens

ity (

V)

Mass in Central cup (amu)

Aridus - 20 μmol/l

Mass in Central cup (amu)

0

2

4

6

8

10

12

31.95 31.97 31.99 32.01 32.03 32.05 32.07 32.09

Aridus - 20 μmol/l

10-4

10-3

10-2

0.1

1

10

102

31.95 31.97 31.99 32.01 32.03 32.05 32.07 32.09

8

25.8

17.7

24.8

H1*121C

H3*21

16O-16O

16O-18O

32S-H

H-16O-16O

10-4

10-3

10-2

0.1

1

10

102

10-5

10-6

103

31.92 31.94 31.96 31.98 32 32.02 32.04 32.06

SIS - 20 μmol/l

0

2

4

6

8

10

12

31.92 31.94 31.96 31.98 32 32.02 32.04 32.06Mass in Central cup (amu) Mass in Central cup (amu)

SIS - 20 μmol/l

Sign

al in

tens

ity (

V)

a b

c d

Fig. 2. Peak shapes for 32S, 33S and 34S using the stable introduction system (a and b) and Aridus (c and d), in linear (a and c) and log scales (b and d). Interferences are indicated. Thegray bar represents the interference free plateau on which isotopic ratios are measured.

53G. Paris et al. / Chemical Geology 345 (2013) 50–61

3. Experimental results

3.1. Peak shape

Fig. 2 shows the high-resolution peak shapes for 20 μmol/l of Na2SO4

introduced on the Neptune using either the Aridus or a Stable Introduc-tion System (SIS). Measured ratios are affected by instrumental back-ground and isobaric interferences (Table 2). Craddock et al. (2008)showed that the main interferences are 16O16O on mass 32 and16O18O on mass 34. Isobars overlapping with 33S follow amore compli-cated pattern. The first interference on the right of the peak is 32S1H,characterized by a ΔM of about 8 mDa, which leaves a remaining3–4 mDa flat shoulder to measure the interference free 33S signal.The second main interference (ΔM = 25.2 mDa), corresponds to16O16O1H. The 16O17O interference is less than 4 mDa to the left of1H16O16O and is not well resolved in Fig. 2. To prevent the risk of inter-ferences due to Ni2+ on 32S and 34S we use Aluminum X-cones.

3.2. Matrix effects

All samples are run asNa2SO4 because ionization efficiency is relatedto the amount of Na+ in the solution (Fig. 3). Pure sulfuric acid solutions

Table 2Main interferences on the S peaks identified based on the difference between the atomicmass of the interference and the atomic mass of the concerned sulfur isotope (ΔM).

CalculatedΔM

MeasuredΔM

32S1H (33S) 8.44 816O16O1H (33S) 26.19 25.7516O16O (32S) 17.75 17.716O18O (34S) 26.2 24.78

have very low ion yields (Fig. 3b). We know of no simple explanationfor this dependence on the cation concentration. Such dependencehas not been previously reported and might be related to the use of adesolvating membrane, which is the main difference between thisstudy and Craddock et al. (2008). The sensitivity boost fromNa additionis also independent of whether it comes as NaCl or as NaOH. Variationsin the Na+/SO4

2− ratio also lead to variations in the measured isotopicratios (Fig. 3a) and this effect is stronger at higher sulfateconcentrations. Fig. 3 shows simultaneous intensity and ratio biases as afunction of both the Na+/SO4

2− ratio and the sulfate concentration. How-ever, for a same amount of Na+ (the one considered here is 40 μmol/l,similar to the bracketing standard, white dotted line in Fig. 3a) all of themeasured ratios are the same, suggesting that the matrix effect dependson the amount of sodium more than on the Na+/SO4

2− ratio, as longas the amount of Na+ is more than twice the amount of sulfate. Theaverage δ34S of the various concentrations of sulfate at 40 μmol/l ofNa+ is −0.06 ± 0.14‰ (2sd), not including the most concentratedsulfate solution (50 μmol/l). Our Neptune based method is aided byfirst establishing the sulfate concentration for each solution and thenma-trix and intensity matching to the bracketing standards.

3.3. Accuracy and sample size

The true δ34S value of the bracketing standard has been evaluated byrunning this solution against the IAEA S1 standard, which defines theVCDT scale. The value of the S1 standard is defined as being −0.3‰exactly (Coplen and Krouse, 1998). A known weight of standard(about 1.5 mg) has been dissolved in concentrated Seastar HNO3,dried down, dissolved again in 10% HCl + 5% HNO3, dried down againand then purified. This procedure has been repeated twice. Each sulfatesolution has been purified three independent times and the Na2SO4 hasbeen run against each of these solutions at least 3 times. The mean 34S/

5 10 15 20 25 30 35 40 45 500

2

4

6

8

10

−5

0

5

10

15

20

25

0‰

-0.2‰

0.2‰

δ34S[SO4

2-] μmol/l

Na+

/SO

42- m

ol/m

ol

0.0

0.5

1.0

1.5

2.0

2.5

0 2 4 6 8

rela

tive

sign

al

Na+/SO42- mol/mol

5 μmol/l

10 μmol/l

20 μmol/l

20 μmol/l - Na+ comes from NaOH

50 μmol/l

[SO4

2-]

a b

Fig. 3. Influence of sodium addition on measured 34S/32S ratios (a) and relative intensity (b) from a purified solution of bracketing Na2SO4 standard and post-purification sodiumaddition run against a 20 μmol/l solution of bracketing standard. Isotopic ratios are reported as a delta value calculated against the unpurified bracketing standard as a reference andrelative intensity is the ratio of the intensity of the sample against the intensity of the bracketing standard. All data are background and drift corrected. The star indicates thebracketing standard and the dotted white line represents a constant amount of Na+ (40 μmol/l), showing that the measured ratio is constant along that line as long as the amountof sulfate in solution is half the amount of sodium. Fig. 3b shows that Na+ is required for full ionization of the sulfate present in solution with a maximum intensity reached for aNa+/SO4

2− ratio of 2.

54 G. Paris et al. / Chemical Geology 345 (2013) 50–61

32S value measured for the bracketing standard is 0.0440939 ±6.9 × 10−6 (2sd, n = 35) and the 33S/32S value is 0.0078711 ±1.3 × 10−6 (2sd, n = 35). The Δ33S value of the Na2SO4 is 0.02 ±0.16‰ (2sd). To estimate the accuracy of the method, we evaluate theratios of three IAEA standard BaSO4 solutions: SO5, SO6, and NBS 127.Solutions of 5 mmol/l of sulfate were prepared by dissolving BaSO4

in DTPA. These solutions were purified on AG50X8 according to thepreviously described protocol and the ratios were compared againstpublished values. Table 3 shows the values of the measured 34S/32S

Table 3Reported δ34SVCDT and Δ33S for the three BaSO4 standards NBS 127, IAEA SO5 and SO6. Allindividual purification and for the entire population. Mean of the population is also reportedmeasured with lower concentrations of SO4

2− are also reported for each standard, with anderrors are the propagation of the internal standard error. Values from other studies are als13 values have been considered for the Δ33S of SO5, one of the values being an obvious ou

NBS 127 δ34SVCDT(Δ33S)

2sd S

Concentration20 μmol/l 1 21.14‰ 0.07

n = 31

2 21.23‰ 0.03n = 3

2

3 21.08‰ 0.12n = 3

3

4 21.07‰ 0.30n = 5

4

Mean of the population 21.12‰(0.05‰)

0.22(0.14)n = 14

5 μmol/l 4 21.09‰ 0.09 310 μmol/l 4 21.09‰ 0.07 3

Mean of the population 20.93‰ 0.245 μmol/l 4 20.48‰ 0.09 310 μmol/l 4 20.77‰ 0.07 3

Published valuesHopple et al. (2002)a 21.10‰ ?Halas and Szaran (2001)b 21.17‰ 0.12

a Values measured using the SF6 method.b Values measured using the SO2 method.

and Δ33S values for the three standards run, without and with back-ground subtraction. This table shows that if all of the data are back-ground subtracted, the measured values are accurate. This is true forconcentration as low as 5 μmol/l, with no significant loss of precisionon themeasured δ34S value, whichmeans that only 3 nmol of S is suffi-cient to accurately and precisely measure the isotopic composition ofsulfur. To explore the possibility of running small samples, we ran dilut-ed seawater on the columns so that only 5 and 10 nmol of sulfate werepurified and run on the Neptune. The measured δ34S values were

standards have been purified 4 times and mean values and 2sd are reported for eachfor the δ34SVCDT values with no instrumental background population (in italic). Valueswithout background subtraction (the latter in italic). For these lower concentrations,o presented. The nature of the reported errors is not specified by these authors. Onlytlier.

O5 δ34SVCDT(Δ33S)

2sd SO6 δ34SVCDT(Δ33S)

2sd

0.59‰ 0.07n = 3

1 −33.99‰ 0.07n = 3

0.57‰ 0.08n = 3

2 −34.04‰ 0.02n = 3

0.47‰ 0.03n = 5

3 −34.06‰ 0.12n = 5

0.43‰ 0.09n = 3

4 −34.12‰ 0.03n = 3

0.51‰(0.09‰)

0.15(0.34)n = 13

−34.05‰(0.09‰)

0.12(0.24)n = 14

0.75‰ 0.10 3 −34.04‰ 0.120.60‰ 0.08 3 −34.06‰ 0.080.48‰ 0.17 −33.74‰ 0.140.58‰ 0.1 3 −32.93‰ 0.110.55‰ 0.08 3 −33.45‰ 0.08

0.49‰ 0.11 −34.05‰ 0.080.15‰ 0.06 −34.12‰ 0.08

10-3

10-2

10-4

10-5

10-6

107 108 109 1010 1011

log(

RS

E)

log(Neff

)

33S/32S

34S/32S

Johnson noise

σcounting statistics

20 mol/l2.5 mol/lμ

μ

σ

Fig. 4. Errors (as relative internal standard error (RSE)) for 34S/32S and 33S/32S ratios of50 cycles with a 4.194 s integration time over a range of concentration from 0.25 to50 μmol/l. Signal intensity is given as the effective number of counts collected overthis period (Neff) (Eq. (6)). The predicted error due to counting statistics is based ona Poisson law (Eq. (7), slope = −1/2, solid line). The predicted error due to Johnsonnoise is calculated from independent measurements of Johnson noise (Eq. (9),slope = −1, dotted line). Two total sulfate concentrations are indicated as an examplefor both 33S/32S and 34S/32S ratios.

55G. Paris et al. / Chemical Geology 345 (2013) 50–61

20.7 ± 0.15‰ and 20.6 ± 0.15‰ respectively. The measured δ34Svalue for the blank being −2.2 ± 0.8‰, the measured seawaterisotopic compositions suggest a blank contamination of 0.05 to0.15 nmol S due to the purification process, in agreement with theamount of sulfate from the blank measured using the IC.

4. Discussion

4.1. Errors and statistics

Following the approach of John and Adkins (2010) on iron isotopes,we describe the processes affecting the precision of our measurementsby dividing them into three categories; internal, intermediate andexternal errors. Internal error is the standard error of the 50 cyclesmeasured for each sample. Intermediate error is the reproducibility ofdifferent measurements of the same solution post-purification andmatrix matching. External error is the reproducibility of separatealiquots of the original sample processed separately through all of thechemistry. The step from internal to intermediate errors is the largestof the progression from a single measurement to true external repro-ducibility so we discuss it last in this section.

4.1.1. Internal errorEstimation of the internal standard error has been described previ-

ously (Albarède and Beard, 2004; John and Adkins, 2010). Ion beamcollection is expected to follow a Poisson distribution (or ‘counting sta-tistics’, Eq. (2)) for its uncertainty:

σ ¼ ffiffiffin

p ð2Þ

where n is the number of ions collected at a given mass.For a measured ratio Rm of two ion beams

Rm ¼ XY: ð3Þ

The error on the ratio is:

σRm¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiσX

2

nX2 þ σY

2

nY2

sð4Þ

with ni = 4.194 × 50 × 62.5 × 106 × Vi (the measured voltage on1011 Ω resistors). Using Eq. (1), Eq. (4) can be rewritten as following:

σcounting statistics ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffinX þ nY

nXnY

r: ð5Þ

John andAdkins (2010) defined a quantityNeff (Eq. (6)) representingthe effective number of counts to be used in estimating counting statis-tics (Eq. (7)):

Neff ¼ nXnY= nX þ nYð Þ ð6Þ

σcounting statistics ¼ffiffiffiffiffiffiffiffi1

Neff

s: ð7Þ

This relationship is represented in a log–log plot of measured vari-ability versus signal strength by a linear trendwith a−0.5 slope (Fig. 4).

At low concentrations however, the measured isotope ratio vari-ance will be dominated by Johnson noise, the fluctuation of the back-ground signal due to thermal noise in feedback resistors of theamplifier circuit. This translates into (John and Adkins, 2010):

Vmeasured ¼ Vsignal þ VJohnson noise ð8Þ

where Vmeasured is the recorded voltage, Vsignal is the voltage due tothe ion beam captured in the Faraday cup, and VJohnson is the magni-tude of the baseline Johnson noise.

The error can then be calculated as

log σ Johnsonnoise

� �¼ − log Neffð Þ

þ logR

R þ 1ð Þ

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffinnoise a

R

� �2 þ nnoise b2

r !ð9Þ

with nnoise the number of ions corresponding to the Johnson noisemeanintensity and R themeasured isotopic ratio. This corresponds to a lineartrend with a −1 slope in the Neff plot. We evaluate Johnson noise bymeasuring the standard deviation of the intensity of a thousand cyclesof 4.194 s with the Neptune analyzer gate closed. The noise is 40 μVon the central cup and 17 μV on cups H1 and H3. The intercept inEq. (9) is 4.48 for 34S/32S and 4.49 for 33S/32S ratios, respectively. Asseen in Fig. 4, S isotopemeasurements follow the same trends, showinglittle or no deviation from theory in terms of internal error. It also showsthat the transition from Johnson noise to counting statistics is~20 μmol/l for 33S/32S and ~2.5 μmol/l for 34S/32S.

4.1.2. Background subtractionOne of the issues affecting sulfur isotope measurements on the

Neptune is the instrumental background signal. As seen in Table 3,background subtraction is necessary to get accurate measurements ofδ34S andΔ33S. This background usually represents 0.1 to 1% of themea-sured intensity on cups C, H1 and H3. The ratios measured in the blankfor H1/C and H3/C are close to 33S/32S and 34S/32S suggesting that thebackground signal is mostly due to sulfur. This constant backgroundmight be explained by the presence of sulfate in the solutions or amem-ory effect in the Aridus. For eachmeasured ratio, the intensitymeasuredin the blank solution before and after a given standard or sample isaveraged and subtracted from the respective signal. The backgroundvariance is larger than the internal errors and ultimately limits the

Table 4Average and standard deviation (sd) of the END populations of Runs 1 and 2 (see text)and for the seawater samples from this study. Time is the time between two bracketingstandards. Comparing Run 2 and seawater run indicate that the END standard devia-tion can significantly deviate from 1 even without analyzing actual samples. Therefore,the source of the instability lies within the mass spectrometry itself.

Run 1 Run 2 Seawater run

Time(min)

Average sd Time(min)

Average sd Time(min)

Average sd

34/32 6 0.01 1.36 6 0.13 1.8630 −0.12 1.94 30 −1.47 5.07 30 −0.89 2.3760 0.01 2.53 60 0.52 6.99

33/32 6 0.09 1.21 6 −0.09 1.2330 0.13 1.04 30 −0.48 1.49 30 −0.60 2.1360 0.02 1.3 60 0.1 2.11

56 G. Paris et al. / Chemical Geology 345 (2013) 50–61

precision of our measurements. In the following, we evaluate the influ-ence of the background correction on the error of the measured ratio.

Rm is the ratio measured on the Neptune and rblk is the ratio of theblank.

rblk ¼ xblkyblk

ð10Þ

To adjust the measured intensities of X and Y, a blank intensity issubtracted, respectively xblk and yblk. The blank corrected ratio is

Rblk corr ¼X−xblkY−yblk

: ð11Þ

Eq. (11) can be rewritten using ratios.

Rblk corr ¼RmY−rblkyblk

Y−yblkð12Þ

The error on blank corrected ratios can be calculated by propagat-ing errors through the different terms in Eq. (12), including their co-variances. However, the calculation can be somewhat simplified asthere is no covariance between Rm and yblk, Rm and rblk, rblk and Y,or Y and yblk., The propagated error on the blank corrected ratioscan then be written as Eq. (13):

σRblk corr

2 ¼ σY2 ∂Rblk corr

∂Y

� �2

þ σyblk2 ∂Rblk corr

∂yblk

� �2

þ σRm

2 ∂Rblk corr

∂Rm

� �2

þ σrblk2 ∂Rblk corr

∂rblk

� �2

þ 2σRmY∂Rblkcorr

∂Rm

� � ∂Rblk corr

∂Y

� �

þ 2σrblkyblk

∂Rblkcorr

∂yblk

� � ∂Rblk corr

∂rblk

� �ð13Þ

where the terms in the first row are the variances of the signals andthe blanks, and the two terms in the bottom row are the covariancesbetween the measured ratios and the intensities in both the signaland blank. The covariance σRmY is defined as:

σRmY ¼Xncycles

i¼1

Rmi−Rm� �

Yi−Y� �

ncycles−1� � ð14Þ

where ncycles is the number of 4.194 s integrations that are combinedinto a single ratio measurement (typically 25 or 50). However, thisformulation of the covariance does not let us do the calculation afterwe have processed the raw data into averaged ratios. To solve thisproblem, we follow (Rae, 2011) and start by reformulating Eq. (14):

σRmY ¼ 1

ncycles−1� � Xncycles

i¼1

Xi−RmYi−RmiY þ Rm Y� �

:

The next step is to use the linearity of the sum:

σRmY ¼ 1

ncycles−1� � Xncycles

i¼1

Xi−Xncycles

i¼1

RmYi−Xncycles

i¼1

RmiY þXncycles

i¼1

RmY

" #:

The covariance can now be written as:

σRmY ¼ 1

ncycles−1� � ncyclesX−ncyclesRmY−ncyclesRmY þ ncyclesRmY

h i;

because λ ¼ 1ncycles

Xncycles

i¼1

λi.

In the end, the covariance is:

σRmY ¼Xncycles

i¼1

ncyclesX−ncyclesRmY� �

ncycles−1� � : ð15Þ

Now all of the parameters required to calculate σRmY are availableafter averaging the 25 or 50 cycles, making it less cumbersome to esti-mate the covariance. The same calculation can be applied to σrblkyblk.It is thus possible to combine the internal error with the propagationof the background subtraction to get a final estimate of the instrumentblank corrected uncertainty on a measured ratio.

4.1.3. External errorTo evaluate external variability, a sample of surface seawater from

the Pacific Ocean (Catalina Island) is purified three times on separatecolumns, using 50 μl of seawater for each replicate. The averageδ34SVCDT value is 20.94 ± 0.08‰ (2sd of the three averagedmeasuredδ34SVCDT values) and the average Δ33S value is 0.04 ± 0.09‰ (2sd,n = 3). The same process is applied to aliquots of 5 mg of cleanedpowder of the deep-sea coral Desmophyllum diantus collected offTasmania during the Southern Surveyor cruise (SS0108 STA011 1/13/2008). The measured δ34SVCDT value is 22.06 ± 0.17‰ and Δ33S is0.12 ± 0.13‰ (2sd, n = 3). This variability is significantly higherthan the internal error. We thus want to understand if the increasedvariance is due to the purification or to the mass spectrometry, whichcan be done by investigating intermediate errors (where replicates areduplicate measurements of the same sample solution, not separatechemical processing of the sample CAS or water sample).

4.1.4. Intermediate errorIn the course of a Neptune run, the difference between two mea-

sured ratios of the same solution is larger than what would be expectedbased on the internal errors. John and Adkins (2010) use the END (ErrorNormalized Deviates, normalized to the expected variance from inter-nal statistics alone) statistic to quantify this extra variability and definedas:

END ¼ R2−R1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiσ2

1 þ σ22

� �q : ð16Þ

If the internal errors completely describe the variability betweenreplicates, then a large number of END values will describe a histo-gramwith a mean of 0 and a standard deviation of 1. For the seawatersamples that were run in this study, the standard deviation of theEND for the 34S/32S ratios is 2.34 and the mean is −0.89 (Table 4).While the internal errors follow the predictions of Johnson noise

57G. Paris et al. / Chemical Geology 345 (2013) 50–61

and Poisson statistics, replicates run over many hours show muchmore variability.

We designed two different experiments to explore the source ofthis variability. Run 1 is a continuous aspiration of a single samplewhere data are averaged every 25 cycles. The autosampler tube wassimply left in a large volume of solution (Fig. 5). Run 2 uses thesame solution but the autosampler tube moved up and down every25 cycles to inject air into the sample stream (Fig. 6). Means and stan-dard deviations of the different END populations are reported inTable 4, which shows that END standard deviations increase withtime between bracketing standards and with injecting air every25 cycles. Overall, the standard deviation of the END is larger forthe 34S/32S ratio than for the 33S/32S ratio. By comparing Run 2 withdata collected on seawater samples, we find that introducing thesame solution again and again induces the same intermediate repro-ducibility as a run with standard-sample-standard bracketing andwashes. As the highest variability of these runs overlaps with thetrue external sample replicates reported above, it is clear that the or-igin of our extra variance (after drift correction) is not due to prob-lems with sample matrix, machine memory, or poorly matchedstandards. Instead, the problem seems to be within the mass spec-trometry itself.

One possibility for the extra variability in intermediate replicatesis that we have not properly accounted for mass bias drift over thecourse of the run. Allan variance plots are a good test of this hypoth-esis. First introduced to estimate the stability of atomic clocks, thisstatistic provides an evaluation of the drift for any given system. Con-sidering three evenly spaced measured ratios during a run, xn, xn + 1,and xn + 2, spaced by a measurement interval τ, the difference be-tween ratios normalized for the sampling interval is given by yn =(xn + 1 − xn) / τ. The instability in the system may be representedby the change in this difference across the whole run (for a given

-0.30

-0.20

-0.10

0.00

0.10

0.20

41.90 42.00 42.10 42.20 42.30 42.40

40.25

40.35

40.45

40.55

40.65

40.75

78.55 78.65 78.75 78.85 78.95 79.05

δ'33

S (‰

)

δ'34S (‰)

δ33S (‰)

Δ33S

(‰)

y = 0.8726x-36.929R² = 0.827

MDF line

O 2 in

terfer

ence

a

c

Fig. 5. δ′34S vs. δ′33S and Δ33S vs. δ33S for raw data and drift corrected data during a run wheMass Dependent Fractionation line.

time interval): Δyn = yn + 1 − yn=. The Allan variance σ2y(τ) is

then:

σ2y τð Þ ¼ 1

2 N−1ð ÞXN−1

i¼1

yiþ1−yi� �2 ¼ 1

2 N−1ð ÞXN−1

i¼1

xiþ2−2xiþ1 þ xi� �2

:

where N is the total number of measured ratios used to calculate thesum. For a system with no drift in the ratios, the Allan variance willdrop as τ increases. We evaluate the Allan variance for two differentruns where the sample was continuously introduced into the instru-ment, Run 1 (same run as previously described) and Run 1b. The in-tegration time is 4.194 s in each case but Run 1 had an idle time of3 s between each time step while 1b did not, explaining the slightlydifferent starting points for the two runs in Fig. 7. The solid lines indi-cate the theoretical evolution of the variance in the case of no drift. Anunchanging Allan variance with τ indicates linear drift in the massbias of our measured ratios. Data are calculated from the integrationtime to 20,000 s (~5 h30), which is half the total duration of a run.

The influence of mass bias drift on 34S/32S and 33S/32S ratios is dif-ferent between the two runs, but no drift affects the Δ33S values, im-plicating true mass dependent mass bias drift as the source forvariability between measured ratios. Fig. 7a shows that differentruns have different timescales of variability. Run 1 has an oscillationof ~10–30 min but with no long term change in the mean value ofthe ratios, while Run 1b is characteristic of a linear drift in theuncorrected ratios. 33S/32S ratios appear to drift at a longer time-scales, but this is due to the smaller size of the 33S beam being inher-ently more variable (Fig. 7b). In a linearly drifting system with morenoise between the individual points, deviations from the ideal linefirst show up at longer periods. In each Run the drift is mass depen-dant because it does not affect the calculated Δ33S value (Fig. 7c).

-0.25

-0.15

-0.05

0.05

0.15

0.25

-1.10 -1.00 -0.90 -0.80 -0.70 -0.60

-1.10

-1.00

-0.90

-0.80

-0.70

-0.60

-1.85 -1.75 -1.65 -1.55 -1.45 -1.35

δ33S (‰)

δ'33

S (‰

)Δ33

S (‰

)

δ'34S (‰)

y = 0.9793x + 0.8177R² = 0.9464

32S-H

MDF lineb

d

re the autosampler probe remained in the same solution the entire time. MDF line is the

-1.1

-1.0

-0.9

-0.8

-0.7

-0.6

-0.5

-1.9 -1.7 -1.5 -1.3

-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

-1.1 -0.9 -0.7 -0.5-0.4

-0.3

-0.2

-0.1

0.0

0.1

0.2

41.4 41.6 41.8 42.0 42.2 42.4 42.6

δ'33

S (‰

)

δ'34S

δ33S (‰)δ33S (‰)

Δ33S

(‰)

δ'33

S (‰

)Δ33

S

41.6

41.7

41.8

41.9

42.0

42.1

42.2

81.0 81.2 81.4 81.6 81.8 82.0 82.2δ'34S (‰)

MDF line

32S-H

MDF line

O 2 in

terfer

ence

O 2 in

terf

eren

ce

a b

dc

Fig. 6. δ′34S vs. δ′33S and Δ33S vs. δ33S for raw data and drift corrected data during a run where the autosampler probe went out of the solution between two blocks of 25 cycles ofratio measurement.

58 G. Paris et al. / Chemical Geology 345 (2013) 50–61

This is confirmed when we make Allan variance plots with thecorrected ratios (not shown). Allan variance analysis demonstratesthat there is mass bias drift within the 50 cycles we use to averagefor a single ratio, and that we might improve the overall δ34S andδ33S ratios with fewer cycles per measurement. However, shorter ac-quisition would clearly adversely affect the uncertainty in the Δ33Svalue. This analysis also explains why points in Fig. 4 lie slightlyabove the counting statistics and Johnson noise lines.

4.2. Discussion on the source of variability

In this section we unpack the sources of variability that can giverise to an END standard deviation of 2–4 over and above that due tointernal errors. Fig. 5 shows the relationships between δ34S, δ33S,and Δ33S for the data in Run 1. Each point in the left hand column(a and c) is a non-drift corrected, and non-normalized, value of theNa2SO4 solution for a 25 cycle integration. The right column (b andd) are the same data but now using every other 25 cycle block todrift correct and normalize the data between blocks. A trend betweenthe scatter in the δ34S and δ33S data is presented in Fig. 5a, but Fig. 5bshows that drift correction shrinks the δ34S data much more than theδ33S data (the ranges on the x and y-axes are the same in a and b).Calculating the mass independent Δ33S variance for each sampleshows that this value is dominated by scatter in the δ33S data(Fig. 5c). The Δ33S and δ34S data do not correlate (not shown). Driftcorrection shrinks the δ33S vs. Δ33S trend to a relatively tight 1:1relationship. Fig. 5d is the signature of an isobaric interference on 33Sbut not 34S or 32S.

From the peak shape data in Fig. 2, the 32S1H hydride species is themost likely candidate for this interference on δ33S and Δ33S. The hy-dride of 16O dimer might also contribute, but it is almost an order ofmagnitude smaller and much easier to mass resolve. With only ab4 mDa interference free plateau, small changes in the absolute

mass stability could change the relative intensity of the hydride inter-ference. The combined high voltage and magnetic field stability of theNeptune is about 1 part in 105, which corresponds to 3 mDa over a300 Da full mass range. The peak shapes in Fig. 2 also show thatoxygen dimer interferences are the most likely cause of extra varianceat masses 32 and 34. However, these interferences are very hard todetect, because they affect the 34/32 ratio with a slope close to theregular mass dependent drift in the ratio (33S/32S = 0.007877, 34S/32S = 0.044163, 33O2/32O2 = 0.00038 and 34O2/32O2 = 0.00205).

When air is injected to the plasma in between measurements(Run 2), the overall range of variability nearly doubles (Fig. 6) andthe competing effects of oxygen interferences and mass bias driftcan be seen. In the non-drift corrected δ34S vs. δ33S data (Fig. 6a)the larger range compared to Run 1 comes from the increased massbias effect of changing the plasma conditions by injecting air ratherthan from increasing the O2 interferences, as the source of oxygenfor the interference is the water aspirated with the sample. Thisdrift can be normalized by sample-standard bracketing (Fig. 6b), butagain it does not change the δ33S data substantially. The predomi-nance of the 32S1H interference is still visible when plotting δ33S vs.Δ33S (Fig. 6c), but a larger scatter appears, due to mass dependentfractionation. When the data is mass bias corrected (Fig. 6d), thestrong 1:1 relationship returns.

Overall, it is clear that much, but not all, of the mass bias drift andO2 interferences on δ34S can be normalized with sample-standardbracketing. Unfortunately this technique has little effect on theshort term variability of 32S1H interference on δ33S. The END standarddeviation is smaller for the δ33S data because the internal errors arelarger, not because there is inherently less extra variability on thisratio. The absolute variance of δ34S can be reduced, but the internalerrors are so small (~20 ppm on normal sized samples) that theEND standard deviation is still large. A longer duration betweenbracketing standards leads to larger END standard deviations for

Alla

n va

rian

ceA

llan

vari

ance

Alla

n va

rian

ce

1 10 100 1000 10000

1.10-3

1.10-1

1.10-6

1.10-7

1.10-6

1.10-5

1

1

10 100 1000 10000

10 100 1000 10000

Time (s)

50 cyclestime between two

bracketing standards

b

c

34S/32S

33S/32S

Δ33S(‰)

a

Fig. 7. Square root of the Allan variance vs. time in a log–log space for Run 1 (open circles) and Run 1 bis (gray circles). The line represents the ideal time-dependant variationfollowing a Poisson law with time. Integration times are the same for both runs but each cycle of 4.194 s is separated by a 3 s idle time in Run 1 and not in Run 1 bis. The solidline (50 cycles) indicates the duration of one measurement (~210 s). The dashed line indicates the time span between two successive bracketing standards (~35 min). (For inter-pretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

59G. Paris et al. / Chemical Geology 345 (2013) 50–61

δ34S because non-linear changes in both mass biasing and the oxygeninterferences are more likely to occur.

To improve our reproducibility on the measured Δ33S, we need tobetter separate the 32S1H interference. We could achieve this goalusing an entrance slit thinner than 25 μm thus increasing the massresolution. END standard deviations for δ34S will have higher stan-dard deviations than internal errors as long as the sample-standardbracketing interval is longer than the variability in the water inducedO2 interference on masses 32 and 34.

5. Seawater profiles

The method described in this paper has been applied to a collectionof two seawater profiles from the San Pedro basin (Pacific Ocean, 10samples, (John et al., 2012)) and theAtlantic ocean (USGeotracesAtlan-tic, station 9, 24 samples). The first profile goes down to 895 m and thesecond one down to 3000 m water depth.

Since the development of the sulfur isotopicmeasurement, seawaterhas been stated to have a constant isotopic composition of ~20‰.The first published systematic study of seawater δ34S suggested a

value of 20.12 ± 0.74‰, n = 17 (Thode et al., 1961) followed by20.99 ± 0.17‰, n = 25 (Rees et al., 1978), 20.88 ± 0.95‰, n = 39(Kampschulte et al., 2001), and values ranging from 18.8 to 20.3‰(Krouse and Coplen, 1997; Coplen and Krouse, 1998; Ding et al.,2001). All of these numbers are an average over different locationsand depth. Rees et al. (1978) are the first ones who systematicallyexplored potential variations of seawater δ34S values with depth inthe open ocean, followed by a joint study of oxygen and sulfur isotopesof seawater (Longinelli, 1989) with values ranging from 18.8 to 20‰. Adepth transect of sulfur isotopes has also been established for the NorthSea (Böttcher et al., 2007) with very similar mean values for seawatersamples collected over springtime (δ34S = 20.9 ± 0.13‰; n = 131)and summertime (δ34S = 21.1 ± 0.13‰; n = 59) Unfortunately,studies prior to 1995 cannot be accurately compared with data morerecent than 1995. The δ34S scale was previously reported using the Can-yon Diablo Troilite standard (CDT). The supply of the standard waseventually depleted and a new scale was defined at the 1993 ViennaConference, the Vienna-CDT scale. Because the CDT standard turnedout to be slightly inhomogeneous, this new scale is based on the Inter-national Atomic Energy Agency (IAEA) S1 standard, defined as being

0

500

1000

1500

2000

2500

3000

3500

20.50 20.75 21.00 21.25 21.50 21.75 22.000

100

200

300

400

500

600

700

800

900

1000

20.50 20.75 21.00 21.25 21.50 21.75 22.00

Geotrace 9 San Pedro Basin

dept

h (m

)

δ34SVCDTδ34S

VCDT

Fig. 8. Isotopic ratios of two seawater profiles (San Pedro Basin and Geotrace transect 9) reported as δ34SVCDT values. Error bars are the internal error predicted by counting statisticsmultiplied by the standard deviation of the END (Eq. (16)).

60 G. Paris et al. / Chemical Geology 345 (2013) 50–61

−0.3‰ in the VCDT scale (Krouse and Coplen, 1997; Coplen andKrouse, 1998; Ding et al., 2001). Here we investigated two depth tran-sects, on a total of 34 samples that place seawater data on the VCDTscale. Results are presented in Fig. 8. They display a constant δ34Svalue with depth. The average on the two sites is 20.96. ± 0.12‰ and20.99 ± 0.10‰ respectively (2sd). These two values are indistinguish-able from one another. They appear however slightly lower than thevalue measured for the standard NBS 127, which is barite precipitatedfrom seawater, or compared to the value measured for modern barites(21.10 ± 0.20‰, 2sd) (Paytan et al., 1998). The standard deviation ofthe END for these runs is shown in Table 4. These profiles exhibitδ34SVCDT values that are constant with depthwithin error bars, in agree-mentwith the conclusion that the open ocean has an homogeneous iso-topic composition (Rees et al., 1978; Longinelli, 1989). The δ34S valuesare in agreement with recent studies showing values for standardIAPSO seawater by MC-ICPMS after column purification of 21.11 ±0.40‰ (Das et al., 2012) and 21.18 ± 0.27‰ (Craddock et al., 2008),or as SF6 by IRM after extraction from four natural samples as BaSO4

(21.34 ± 0.13‰, Ono et al., 2012). Average Δ33S for San PedroBasin 0.08 ± 0.10 (n = 10) and the Atlantic Ocean (0.07 ± 0.07‰;n = 24) agree well with the value of 0.050 ± 0.03‰ (2sd, n = 6samples) measured by gas source isotope-ratio-monitoring MS (Onoet al., 2012).

6. Conclusions

The method described in this paper allows for the first time a mea-surement of the triple isotopic composition of sulfur isotopes on sam-ples of a few nmol of sulfate. Although, significant progress hadalready been achieved in measurement of S isotopes on theMC-ICP-MS Neptune (Craddock et al., 2008; Das et al., 2012), we areable to work on samples one order of magnitude smaller using amethod than can be applied to a broad range of samples such aswaters,carbonates, sulfides or samples available in limited amount such asforaminiferae. As an example, 20–50 mg of carbonates with less a100 ppm of sulfate or a few ml of waters (such a rivers or rainwaters)

containing less than 100 μmol S/l carry enough sulfate to be measuredby this method. Applying this method to seawater yields a δ34SVCDTvalue of 20.97 ± 0.10‰ (2sd, n = 34).

Precision on the measured Δ33S could be improved by working ata higher resolution than achieved by the slits available for theNeptune. Such improvement will allow exploration of fine mass inde-pendent fractionation in modern samples with a simple method thandoes not require specific fluorination line and requires very smallamount of sulfate.

Acknowledgment

The authorswould like to acknowledgeNathanDalleska for precioushelp in using the IC, Eric Kort for helping with Allan variance analyses,Seth John for providing the seawater samples and Nivedita Thiagarajanfor the coral sample. We also thank two anonymous reviewers for theircomments and suggestions. This work was supported by the Camilleand Henry Dreyfus Postdoctoral Program in Environmental Chemistry.The Adkins and Sessions groups are also deeply thanked for feedbacksand support. Thanks also to Vicky Rennie for being a motivated earlyuser of this new method.

References

Albarède, F., Beard, B., 2004. Analytical methods for non-traditional isotopes. Reviewsin Mineralogy and Geochemistry 55 (1), 113–152.

Amrani, A., Sessions, A.L., Adkins, J.F., 2009. Compound-specific δ34S analysis of volatileorganics by coupled GC/multicollector-ICPMS. Analytical Chemistry 81 (21),9027–9034.

Bontognali, T.R., et al., 2012. Sulfur isotopes of organic matter preserved in 3.45-billion-year-old stromatolites reveal microbial metabolism. Proceedings of the NationalAcademy of Sciences 109, 15146–15151.

Böttcher, M.E., Brumsack, H.J., Dürselen, C.D., 2007. The isotopic composition of modernseawater sulfate: I. Coastal waters with special regard to the North Sea. Journal ofMarine Systems 67 (1–2), 73–82.

Chaussidon, M., Albarède, F., Sheppard, S.M.F., 1989. Sulphur isotope variations inthe mantle from ion microprobe analyses of micro-sulphide inclusions. Earth andPlanetary Science Letters 92 (2), 144–156.

61G. Paris et al. / Chemical Geology 345 (2013) 50–61

Clough, R., Evans, P., Catterick, T., Evans, E.H., 2006. δ34S measurements of sulfur bymulticollector inductively coupled plasma mass spectrometry. Analytical Chemistry78 (17), 6126–6132.

Coplen, T.B., Krouse, H.R., 1998. Sulphur isotope data consistency improved. Nature 392(6671), 32.

Craddock, P.R., Rouxel, O.J., Ball, L.A., Bach, W., 2008. Sulfur isotope measurement ofsulfate and sulfide by high-resolution MC-ICP-MS. Chemical Geology 253 (3–4),102–113.

Das, A., Chung, C.-H., You, C.-F., Shen, M.-L., 2012. Application of an improved ionexchange technique for the measurement of δ34S values from microgram quantitiesof sulfur by MC-ICPMS. Journal of Analytical Atomic Spectrometry 27 (12),2088–2093.

Deloule, E., Allegre, C.J., Doe, B.R., 1986. Lead and sulfur isotope microstratigraphy ingalena crystals from Mississippi valley-type deposits. Economic Geology 81 (6),1307–1321.

Ding, T., et al., 2001. Calibrated sulfur isotope abundance ratios of three IAEA sulfurisotope reference materials and V-CDT with a reassessment of the atomic weightof sulfur. Geochimica et Cosmochimica Acta 65 (15), 2433–2437.

Halas, S., Szaran, J., 2001. Improved thermal decomposition of sulfates to SO2 and massspectrometric determination of δ34S of IAEA SO-5, IAEA SO-6 and NBS-127 sulfatestandards. Rapid Communications in Mass Spectrometry 15 (17), 1618–1620.

Hopple, J.A., Böhlke, J.K., Krouse, H.R., Rosman, K.J.R., Ding, T., Vocke, R.D., Révész, K.M.,Lamberty, A., Taylor, P., De Bièvre, P., 2002. Compilation of minimum and maxi-mum isotope ratios of selected elements in naturally occurring terrestrial materialsand reagents. (USGS).

Hulston, J.R., Thode, H.G., 1965. Variations in the S33, S34, and S36 contents of meteoritesand their relation to chemical and nuclear effects. Journal of Geophysical Research70 (14), 3475–3484.

John, S.G., Adkins, J.F., 2010. Analysis of dissolved iron isotopes in seawater. MarineChemistry 119 (1–4), 65–76.

John, S.G., Mendez, J., Moffett, J., Adkins, J., 2012. The flux of iron and iron isotopes fromSan Pedro Basin sediments. Geochimica et Cosmochimica Acta 93, 14–29.

Kampschulte, A., Bruckschen, P., Strauss, H., 2001. The sulphur isotopic composition oftrace sulphates in Carboniferous brachiopods: implications for coeval seawater,correlation with other geochemical cycles and isotope stratigraphy. ChemicalGeology 175 (1–2), 149–173.

Krouse, H., Coplen, T., 1997. Reporting of relative sulfur isotope-ratio data. Pure andApplied Chemistry 69 (2), 293–295.

Longinelli, A., 1989. Oxygen-18 and sulphur-34 in dissolved oceanic sulphate andphosphate. In: Fritz, P., Fontes, J.C. (Eds.), The Marine Environment, A. Handbookand Environmental Isotope Geochemistry. Elsevier, Amsterdam, pp. 219–255.

Ono, S., Wing, B., Johnston, D., Farquhar, J., Rumble, D., 2006a. Mass-dependentfractionation of quadruple stable sulfur isotope system as a new tracer of sulfurbiogeochemical cycles. Geochimica et Cosmochimica Acta 70 (9), 2238–2252.

Ono, S., Wing, B., Rumble, D., Farquhar, J., 2006b. High precision analysis of all fourstable isotopes of sulfur (32S, 33S, 34S and 36S) at nanomole levels using a laserfluorination isotope-ratio-monitoring gas chromatography–mass spectrometry.Chemical Geology 225 (1–2), 30–39.

Ono, S., Keller, N.S., Rouxel, O., Alt, J.C., 2012. Sulfur-33 constraints on the origin of sec-ondary pyrite in altered oceanic basement. Geochimica et Cosmochimica Acta 87,323–340.

Paytan, A., Kastner, M., Campbell, D., Thiemens, M.H., 1998. Sulfur isotopic compositionof Cenozoic seawater sulfate. Science 282 (5393), 1459–1462.

Prohaska, T., Latkoczy, C., Stingeder, G., 1999. Precise sulfur isotope ratio measurementsin trace concentration of sulfur by inductively coupled plasma double focusingsector field mass spectrometry. Journal of Analytical Atomic Spectrometry 14 (9),1501–1504.

Rae, J., 2011. Boron Isotopes: Measurement, Calibration and Glacial CO2. University ofBristol, Bristol.

Rees, C.E., Jenkins, W.J., Monster, J., 1978. The sulphur isotopic composition of oceanwater sulphate. Geochimica et Cosmochimica Acta 42 (4), 377–381.

Thode, H.G., Monster, J., Dunford, H.B., 1961. Sulphur isotope geochemistry.Geochimica et Cosmochimica Acta 25 (3), 159–174.

Wanless, R.K., Thode, H.G., 1953. A mass spectrometer for high precision isotope ratiodeterminations. Journal of Scientific Instruments 30 (11), 395.

Weyer, S., Schwieters, J.B., 2003. High precision Fe isotope measurements with highmass resolution MC-ICPMS. International Journal of Mass Spectrometry 226 (3),355–368.