Chemical Geology 345 (2013) 50–61
Contents lists available at SciVerse ScienceDirect
Chemical Geology
journal homepage: www.elsevier.com/locate/chemgeo
MC-ICP-MS measurement of δ 34S and Δ 33S in small amounts of
dissolved sulfate
Guillaume Paris ⁎, Alex L. Sessions, Adam V. Subhas, Jess F. Adkins
Division of Geological and Planetary Sciences, California Institute of Technology, 1200 E California Blvd, Pasadena, CA 91125, USA
a r t i c l e
i n f o
Article history:
Received 16 November 2012
Received in revised form 16 February 2013
Accepted 19 February 2013
Available online 5 March 2013
Editor: U. Brand
Keywords:
Sulfur isotopes
MC-ICPMS
Trace sulfate
Mass independent fractionation
a b s t r a c t
Over the last decade, the use of multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS) has
significantly 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 isotopes
33
S and 36S. In the present study, we report improved techniques for measuring sulfur isotopes on the
MC-ICP-MS Neptune Plus (Thermo Fischer Scientific) using a heated spray chamber coupled to a desolvating
membrane (Aridus, Cetac). Working at high mass resolution, we measured δ34S values of natural samples with
a typical reproducibility of 0.08–0.15‰ (2sd) on 5 to 40 nmol sulfur introduced into the instrument. We applied
this method to two seawater profiles, using 25 μl of sample (700 nmol of sulfate). The average δ34SVCDT value is
20.97 ± 0.10‰ (2sd, n = 25). We show that the amount of sulfate required for an analysis can be decreased to
5 nmol. Because the plasma is sustained by Ar, measurement of 36S is impossible at the current mass resolution
due to the presence of 36Ar+, but a reproducibility of 0.1–0.3‰ (2sd) is achieved on the measurement of mass
independent 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 limitations
remain that limit their utility. Sulfur isotopic compositions were originally measured on gas source isotope-ratio mass spectrometers
(GS-IRMS) as SO2 (Wanless and Thode, 1953; Thode et al., 1961) or
SF6 (Hulston and Thode, 1965). These methods require extraction of a
few μmol of sulfate as a solid (BaSO4), which is then combusted to
SO2, to make a measurement with a precision down to 0.05‰. While
this amount of sulfur is readily achieved in many marine samples, it
may become limiting in samples with low sulfate (or sulfur) content,
such as river waters, aerosols, ice cores, Precambrian carbonates, and
foraminiferae. Moreover, the need to correct for 17O and 18O isobars
in the mass spectrum of SO2 further limits accuracy and precision, especially for 33S and 36S abundances. The use of SF6 as an analyte gas has
three main advantages over SO2. Unlike oxygen, fluorine is monoisotopic so there are no isobaric interferences and the entire S isotopic
system can be measured at high precision with little to no memory
effect (Hulston and Thode, 1965). However, the SF6 method requires
the use of a fluorination vacuum line, which is not widely available,
restricting this method to specialized laboratories. The introduction of
isotope-ratio-monitoring (‘continuous flow’) techniques (IRM) has
allowed researchers to measure samples as small as 200 nmol S by
in-situ fluorination (Ono et al., 2006a, 2006b). Secondary ion mass
⁎ Corresponding author. Tel.: +1 626 395 3765.
E-mail address: [email protected] (G. Paris).
0009-2541/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.chemgeo.2013.02.022
spectrometry (SIMS) is also suitable for in-situ analysis of sulfide minerals (Deloule et al., 1986; Chaussidon et al., 1989) and kerogen sulfur
(Bontognali et al., 2012).
Because of major interferences of O2 and molecular hydrides at
each mass of its isotopes, sulfur was not analyzed by inductively
coupled plasma mass spectrometry (ICP-MS) before the late 1990s,
though this widespread technique had the potential to surmount
most of the previously described difficulties and appears especially
sound for sulfate in solution or trace sulfate in carbonates. To avoid
the cardinal mass interferences, the isotopic composition of sulfur
was first measured on a quadrupole ICP-MS using masses 48 ( 32SO+)
and 50 (34SO+) to get 34S/32S ratios. Precision on the ratios was better
than 10‰ (1sd). Elemental S+ ions were first analyzed directly using
an RF-only hexapole ICP-MS where the collision cell was filled with a
mixture of Xe and H2 to decrease the S+/O2+ ratio. This method resulted
in precisions of better than 3‰ (1sd) for solutions containing between
300 and 1500 μmol S/l (Prohaska et al., 1999). Combining a doublefocusing sector field mass spectrometer and a desolvating membrane
to decrease ( 16O)2 interference on mass 32, Prohaska et al. obtained
34 32
S/ S ratios with precision of 0.4‰ (1sd) using less than
400 nmol S, and precision of 1‰ using 4 nmol S for one measurement
(Prohaska et al., 1999). Multicollector ICP-MS (MC-ICP-MS) allows for
simultaneous collection of 32S+, 33S+, 34S+ and 36S+, decreasing the
effect of fluctuations in the plasma on the measured ratios (Clough
et al., 2006). Isotopic compositions are usually reported using the δ notation (‰): δ'3xS = 1000× [(3xS/32S)sample/(3xS/32S)standard – 1]. More
recently, Craddock et al. (2008) developed bulk and in-situ (by laser ablation) measurements of δ34S on the MC-ICP-MS Neptune (Thermo
G. Paris et al. / Chemical Geology 345 (2013) 50–61
Fischer Scientific). Because of the 32S 1H+ interference at mass 33, they
could not measure δ 33S values with high precision however, and therefore did not report data for the mass-independent fractionation tracer
Δ 33S (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 and
Thode, 1965). The use of MC-ICP-MS for S isotopes is becoming more
widespread. Due to high useful ion yields (which combines ionization
efficiency and sample usage) and low detection limits, sample sizes
are 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 throughput
and requires less workup (e.g., no need for precipitation of dissolved
sulfates or sulfides, no combustion to SO2 or fluorination to SF6). Working at higher mass resolution can further reduce these interferences. In
combination, these aspects of MC-ICP-MS provide much better accuracy
than SO2 measurements for both 33S/32S and 34S/32S ratios. 36S is not
measureable by ICP-MS due to a severe interference by 36Ar.
Here we further push the limits of precision and sensitivity for sulfur
isotope analysis on the MC-ICP-MS Neptune Plus (Thermo Fischer
Scientific), and analyze the current limitations to our precision. Following the advent of GC–ICP-MS measurements of single organic
compounds for their δ34S values at Caltech (Amrani et al., 2009), we
present the development and application of trace sulfate analyses on
the Neptune Plus and apply it to seawater profiles. Three points are
assessed in this paper: purification of the samples, a decrease of isobaric
interferences on the Neptune and measurement of 33S, and an increase
of 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 isotope
abundance. The method developed in this paper is a novel and straightforward way to analyze δ33S, δ 34S and Δ 33S from carbonate-associated
sulfate (CAS) or trace dissolved sulfate in waters down to ~5 nmol S.
51
Waters
Carbonates
Filter
Cleaning
Add
HNO3+HCl
Dissolution in
HCl
Evaporate
Centrifugation,
supernatant
recovery
Dilute in HCl
0.25%
Aliquot Neptune/IC
DX500
Neptune
Dilution in
H2O
Purification on
Bio-Rad
AG50X8
Run on the IC
Drying
Dilution in
HNO3
Add NaCl
Run on the
Neptune
Fig. 1. Chart flow of the protocol described in the paper.
2. Description of the method
2.1. Wet chemistry
Three types of sample have been analyzed; solid BaSO4, dissolved
sulfate in natural waters, and CAS. Each kind of sample is processed
differently to the point of yielding dissolved aqueous SO42−; from that
point on, all are treated identically. Samples are typically split for measurement of both sulfate concentration by ion chromatography (IC) and
sulfur isotopic composition by MC-ICP-MS (Fig. 1). Except if otherwise
indicated, all acids are ultrapure ‘Baseline’ grade (Seastar Chemicals
Inc.).
2.1.1. Barite
Barites 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 in
the solution for three days and then purified according to the protocol
described in Section 2.2. The final volume should not be greater than
the volume of resin used for purification of the sample.
2.1.2. Aqueous sulfate
Water samples are aliquotted for measuring concentration and isotopic ratio. Ideally, 200 to 500 μl of sample with a sulfate concentration
between5 and 25 μmol/l are introduced on the IC. Sulfate extraction for
isotopic measurement requires at least 5 nmol S. Samples are acidified
with one drop of 10% HCl and one drop of 5% HNO3, evaporated to
dryness at 100 °C, and then rediluted in 0.25% HCl. Samples are then
aliquotted and purified.
2.1.3. Carbonate samples
Carbonate 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 external part of the coral with a Dremel tool, it is ground and rinsed three
times in MilliQ water. Corals are then left for 12 h in a 1:1 mixture of
30% H2O2 and 0.1 M NaOH solution, after which they were rinsed 3
times in MilliQ water.
After cleaning, the sample is dissolved in 5% HCl. An aliquot is
sampled for concentration and another one for isotopic composition.
They are independently dried down and the IC aliquot is rediluted in
water while the isotopic composition aliquot is rediluted in 0.25% HCl.
2.2. Purification
Our first attempt to purify the samples consisted of running them
through a Dionex Anion Self-Regenerating Suppressor membrane, in
order to extract cations from the solutions. Quantitative sulfate recovery (>95%) and complete purification of the sample have never been
achieved. A second attempt using the strong anionic resin AG1X8
(Bio-Rad) was also performed, however, its affinity for sulfate made
its recovery difficult for low-concentration samples. Instead, samples
are thus purified using the strong cationic resin AG50X8 (Bio-Rad).
Active exchange sites are sulfonyl groups, potentially contributing
to the sulfur blank, though this resin has been previously used for purification of sulfate minerals (Craddock et al., 2008). The resin is batch
cleaned by rinsing it first in MilliQ H2O three times to remove smaller
particles. It is then rinsed in 8 N reagent grade HNO3 and then alternatively rinsed in 10% HCl and MilliQ water, where each rinse is left
for 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 required
amount of resin is introduced on columns made of shrink-wrap PTFE
tubing (Small Parts, Inc.). The column is rinsed with MilliQ H2O,
acetone, MilliQ H2O again and 10% HCl before loading the resin. Once
52
G. Paris et al. / Chemical Geology 345 (2013) 50–61
loaded, the resin is rinsed twice with 20 column volumes (CV) of MilliQ
H2O and 20 CV of 10% HCl. The resin is conditioned with 3 × 5 CV of
0.25% HCl. The sample is introduced and the resin is rinsed with
3 × 1 CV of 0.25% HCl to ensure complete elution of sulfate. The final
volume is then dried down and diluted in 5% HNO3 to reach a solution
of known sulfate concentration. NaCl is added to the final solution to
match the concentration of sodium in the sample to the bracketing
standard. The sample is then ready to run on the Neptune. The yield
of this procedure has been checked by measuring concentrations of
seawater samples on the IC before and after purification. The measured
yield is 103 ± 5%. The total procedural blank depends on the size of
the column, but for the columns used in this study (20 μl of resin) the
measured blank ranged from less than 0.1 to 0.3 nmol SO4. This range
is partially explained by the small amounts of sulfate, which are close
to 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 with
MilliQ H2O, refluxed for 24 h in Seastar HNO3 and rinsed 5 times
with MilliQ H2O. Vials have to be handled carefully because gloves
have been found to be a source of sulfur contamination. Neptune
autosampler vials are washed in 10% reagent grade HCl and rinsed
five times in MilliQ H2O. The resin is not recycled and columns are
stored in 10% reagent grade HCl.
Table 1
Neptune parameters.
Neptune parameters
Rf power
Cool gas
Aux gas
Sample gas
1200 W
15 l/min
~0.8 l/min
~0.8 l/min
Aridus
Spray chamber temp.
Desolvating membrane temp.
Ar sweep gas
N2 flow
Nebulizer
Flow rate
Cones
110 °C
160 °C
4 l/min
0.02 l/min
PFA-50 (ESI)
60 μl/min
Al X-cones
Measurement parameters
Cup configuration
Resolution mode
Acquisition
Integration time
Uptake time
Volume of sample
Wash + background measurement
32
S (C), 33S (H1), 34S (H3)
High (15 μm entrance slit)
50 blocks
4.194 s
60 s
270 μl
13 min
2.3. Description of the IC
Sulfate concentrations are measured using a DX500 Ion Chromatograph (Dionex). The system is comprised of an AS40 autosampler, an
EG40 eluent generator equipped with a KOH eluent generator cartridge
EGC III KOH, an LC20 Analyzer equipped with an IonPac AS19 inorganic
anion column and an ED40 electrochemical detector. Samples are introduced using 500 μl Dionex vials. They are injected via a 10 μl loop,
mixed with in-situ generated KOH eluent and pushed through the
guard column and then the analytical column that separates the anions.
The sample flows through the Self Regenerating Suppressor to remove
the eluent and enhance the detection of the analyte. Single injections
are performed on each vial. Standards, and some samples, are analyzed
in triplicate to evaluate reproducibility of the results.
A gradient of KOH eluent concentration is applied. The concentration is 5 mmol/l for the first 5 min, then increases linearly from 5 to
45 mmol/l between 5 and 25 min at a constant pressure (2200 psi).
The area of each peak is directly proportional to the concentration of
the anion of interest. Data are manually processed to ensure that the
baseline of each peak is accurate. Calibration standards are run during
each session to ensure the validity of the calibration curve.
2.4. Description of the Neptune
Samples are analyzed for isotopic composition using a Thermo
Fischer Scientific MC-ICPMS Neptune Plus equipped with nine faraday
cups. Typical settings are shown in Table 1. Samples are introduced as
Na2SO4 in 5% HNO3 (Seastar) using a ESI PFA-50 nebulizer (with a
measured uptake rate of about 60 μl/min) and a Cetac Aridus desolvator
that utilizes a PTFE membrane at 160 °C to remove solvent vapor. At
the start of each run, the Neptune ion source is tune to: i) optimize sensitivity, ii) decrease isobaric interferences, and iii) increase the signal
stability. Torch position is optimized first, followed by an iterative optimization of the various gas flows (sample gas and auxiliary gas on the
Neptune, sweep gas and additional gas on the Aridus). The optimal
position and alignment of the peaks are determined first visually on
the interference-free shoulders and confirmed by collecting 34S/32S
and 33S/32S ratios to evaluate the Δ33S value. The position is chosen so
that Δ33S is 0‰ within ±0.1‰, ensuring that interferences do not affect
the signals measured on the collected masses. Isotope ratios are measured in high resolution mode to allow optimal mass resolution, with
M/ΔM typically between 8000 and 10,000 using the 5%–95% valley
definition (Weyer and Schwieters, 2003). The detection of 32S, 33S and
34
S is affected by isobaric interferences created by various isotopologues
of O2 and SH. Because of the major interference on 36S from 36Ar, this
isotope is not measured. Interferences are heavier than the S + ions of
interest, leaving an interference-free shoulder on the low-mass side of
the peak (Fig. 2), similar to previous observations (Craddock et al.,
2008). We measure 32S in the central faraday cup because it decreases
the presence of a reflection on the low mass side of the peak. The
heavy isotopes 33S and 34S are measured in faraday cups H1 and H3,
respectively. The signals from the three cups are registered through
feedback electrometers with 1011 Ω resistors. Both 32S and 34S peaks
are 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 signal
for 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 time
each. A sulfate concentration of 20 μmol/l yields a signal intensity of
~ 10 V for 32S +, corresponding to 1.3·10 11 collected ions. A run uses
about 0.3 ml of sample, which corresponds to ~ 3.06·10 15 ions introduced. The useful ion yield is thus 0.036‰, somewhat lower than typical for the ICP-MS, which is partially explained by the relatively high
energy of first ionization for sulfur and the use of high resolution.
Instrumental background is measured after each sample or
bracketing standard and is subtracted from the average signal measured on each cup. Each sample or bracketing standard is analyzed
using a Matlab script that removes outliers and/or problematic portions of the 50 cycles. Typically, sections of a run are removed when
the signal is not stable at the start of the run, and/or the sample
runs out of solution before the 50 cycles are complete.
Mass drift is corrected using a bracketing standard made of dissolved
reagent-grade Na2SO4. Both 34S/32S and the 33S/32S ratios are corrected
independently assuming a linear variation between two consecutive
bracketing standards. The measured ratio for sample j (Rmj) is drift
corrected using Eq. (1):
Rm j
i
Rcorr j ¼ h
α j−1 −α jþ1 =2 þ 1
with α i ¼
Rmi −Rtrue
.
Rtrue
ð1Þ
G. Paris et al. / Chemical Geology 345 (2013) 50–61
a
12
Aridus - 20 μmol/l
Signal intensity (V)
10
b
102
Aridus - 20 μmol/l
10
C
H1*121
H3*21
8
32
S-H
25.8
16
6
0.1
4
10-2
2
10-3
O-16O
H-16O-16O
17.7
16
O-18O
24.8
10-4
31.95 31.97 31.99 32.01 32.03 32.05 32.07 32.09
Mass in Central cup (amu)
Mass in Central cup (amu)
d
12
SIS - 20 μmol/l
10
3
102
10
Signal intensity (V)
8
1
0
31.95 31.97 31.99 32.01 32.03 32.05 32.07 32.09
c
53
SIS - 20 μmol/l
10
1
8
0.1
6
10-2
10-3
4
10-4
2
10-5
0
31.92
10-6
31.94
31.96
31.98
32
32.02
32.04
32.06
Mass in Central cup (amu)
31.92 31.94 31.96 31.98
32
32.02 32.04 32.06
Mass in Central cup (amu)
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. The
gray bar represents the interference free plateau on which isotopic ratios are measured.
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 Introduction System (SIS). Measured ratios are affected by instrumental background and isobaric interferences (Table 2). Craddock et al. (2008)
showed that the main interferences are 16O16O on mass 32 and
16 18
O O on mass 34. Isobars overlapping with 33S follow a more complicated pattern. The first interference on the right of the peak is 32S1H,
characterized by a ΔM of about 8 mDa, which leaves a remaining
3–4 mDa flat shoulder to measure the interference free 33S signal.
The second main interference (ΔM = 25.2 mDa), corresponds to
16 16 1
O O H. The 16O 17O interference is less than 4 mDa to the left of
1 16 16
H O O and is not well resolved in Fig. 2. To prevent the risk of interferences due to Ni 2+ on 32S and 34S we use Aluminum X-cones.
3.2. Matrix effects
All samples are run as Na2SO4 because ionization efficiency is related
to the amount of Na+ in the solution (Fig. 3). Pure sulfuric acid solutions
Table 2
Main interferences on the S peaks identified based on the difference between the atomic
mass of the interference and the atomic mass of the concerned sulfur isotope (ΔM).
32 1
S H (33S)
O16O1H (33S)
16 16
O O (32S)
16 18
O O (34S)
16
Calculated
ΔM
Measured
ΔM
8.44
26.19
17.75
26.2
8
25.75
17.7
24.78
have very low ion yields (Fig. 3b). We know of no simple explanation
for this dependence on the cation concentration. Such dependence
has not been previously reported and might be related to the use of a
desolvating membrane, which is the main difference between this
study and Craddock et al. (2008). The sensitivity boost from Na addition
is also independent of whether it comes as NaCl or as NaOH. Variations
in the Na+/SO42− ratio also lead to variations in the measured isotopic
ratios (Fig. 3a) and this effect is stronger at higher sulfate
concentrations. Fig. 3 shows simultaneous intensity and ratio biases as a
function of both the Na+/SO42− ratio and the sulfate concentration. However, 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 the
measured ratios are the same, suggesting that the matrix effect depends
on the amount of sodium more than on the Na+/SO42− ratio, as long
as the amount of Na+ is more than twice the amount of sulfate. The
average δ34S of the various concentrations of sulfate at 40 μmol/l of
Na+ is −0.06 ± 0.14‰ (2sd), not including the most concentrated
sulfate solution (50 μmol/l). Our Neptune based method is aided by
first establishing the sulfate concentration for each solution and then matrix and intensity matching to the bracketing standards.
3.3. Accuracy and sample size
The true δ 34S value of the bracketing standard has been evaluated by
running this solution against the IAEA S1 standard, which defines the
VCDT 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 again
and then purified. This procedure has been repeated twice. Each sulfate
solution has been purified three independent times and the Na2SO4 has
been run against each of these solutions at least 3 times. The mean 34S/
54
G. Paris et al. / Chemical Geology 345 (2013) 50–61
a
10
25
b
2.5
[SO42-]
20
50 μmol/l
2.0
6
10
4
5
relative signal
15
-0.2‰
Na+/SO42- mol/mol
8
1.5
20 μmol/l
1.0
0‰
20 μmol/l - Na+ comes from NaOH
0
2
0.5
0.2‰
−5
0
5
10
10 μmol/l
15
20
25
30
35
40
45
0.0
50
[SO42-] μmol/l
5 μmol/l
0
δ34S
2
4
6
8
Na+/SO42- mol/mol
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 sodium
addition 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 and
relative 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 the
bracketing 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 amount
of 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 a
Na+/SO42− ratio of 2.
32
and Δ33S values for the three standards run, without and with background subtraction. This table shows that if all of the data are background subtracted, the measured values are accurate. This is true for
concentration as low as 5 μmol/l, with no significant loss of precision
on the measured δ 34S value, which means that only 3 nmol of S is sufficient to accurately and precisely measure the isotopic composition of
sulfur. To explore the possibility of running small samples, we ran diluted seawater on the columns so that only 5 and 10 nmol of sulfate were
purified and run on the Neptune. The measured δ34S values were
S 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 Δ 33S value of the Na2SO4 is 0.02 ±
0.16‰ (2sd). To estimate the accuracy of the method, we evaluate the
ratios 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 the
previously described protocol and the ratios were compared against
published values. Table 3 shows the values of the measured 34S/32S
Table 3
Reported δ34SVCDT and Δ33S for the three BaSO4 standards NBS 127, IAEA SO5 and SO6. All standards have been purified 4 times and mean values and 2sd are reported for each
individual purification and for the entire population. Mean of the population is also reported for the δ34SVCDT values with no instrumental background population (in italic). Values
measured with lower concentrations of SO42− are also reported for each standard, with and without background subtraction (the latter in italic). For these lower concentrations,
errors are the propagation of the internal standard error. Values from other studies are also presented. The nature of the reported errors is not specified by these authors. Only
13 values have been considered for the Δ33S of SO5, one of the values being an obvious outlier.
Concentration
20 μmol/l
NBS 127
δ34SVCDT
(Δ33S)
2sd
SO5
δ34SVCDT
(Δ33S)
2sd
SO6
δ34SVCDT
(Δ33S)
2sd
1
21.14‰
1
0.59‰
−33.99‰
21.23‰
2
0.57‰
2
−34.04‰
3
21.08‰
3
0.47‰
3
−34.06‰
4
21.07‰
4
0.43‰
4
−34.12‰
21.09‰
21.09‰
20.93‰
20.48‰
20.77‰
0.75‰
0.60‰
0.48‰
0.58‰
0.55‰
0.07
n=3
0.08
n=3
0.03
n=5
0.09
n=3
0.15
(0.34)
n = 13
0.10
0.08
0.17
0.1
0.08
1
2
0.07
n=3
0.03
n=3
0.12
n=3
0.30
n=5
0.22
(0.14)
n = 14
0.09
0.07
0.24
0.09
0.07
−34.04‰
−34.06‰
−33.74‰
−32.93‰
−33.45‰
0.07
n=3
0.02
n=3
0.12
n=5
0.03
n=3
0.12
(0.24)
n = 14
0.12
0.08
0.14
0.11
0.08
21.10‰
21.17‰
?
0.12
0.49‰
0.15‰
0.11
0.06
−34.05‰
−34.12‰
0.08
0.08
Mean of the population
5 μmol/l
10 μmol/l
Mean of the population
5 μmol/l
10 μmol/l
Published values
Hopple et al. (2002)a
Halas and Szaran (2001)b
a
b
21.12‰
(0.05‰)
4
4
4
4
Values measured using the SF6 method.
Values measured using the SO2 method.
0.51‰
(0.09‰)
3
3
3
3
−34.05‰
(0.09‰)
3
3
3
3
G. Paris et al. / Chemical Geology 345 (2013) 50–61
20.7 ± 0.15‰ and 20.6 ± 0.15‰ respectively. The measured δ 34S
value for the blank being −2.2 ± 0.8‰, the measured seawater
isotopic compositions suggest a blank contamination of 0.05 to
0.15 nmol S due to the purification process, in agreement with the
amount of sulfate from the blank measured using the IC.
55
10-2
34
S/32S
S/32S
20 μ mol/l
2.5 μmol/l
33
on
ns
h
σ Jo
10-3
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 measurements
by dividing them into three categories; internal, intermediate and
external errors. Internal error is the standard error of the 50 cycles
measured for each sample. Intermediate error is the reproducibility of
different measurements of the same solution post-purification and
matrix matching. External error is the reproducibility of separate
aliquots of the original sample processed separately through all of the
chemistry. The step from internal to intermediate errors is the largest
of the progression from a single measurement to true external reproducibility so we discuss it last in this section.
4.1.1. Internal error
Estimation of the internal standard error has been described previously (Albarède and Beard, 2004; John and Adkins, 2010). Ion beam
collection is expected to follow a Poisson distribution (or ‘counting statistics’, Eq. (2)) for its uncertainty:
σ¼
pffiffiffi
n
ð2Þ
where n is the number of ions collected at a given mass.
For a measured ratio Rm of two ion beams
Rm ¼
X
:
Y
ð3Þ
σ Rm
ð4Þ
with ni = 4.194 × 50 × 62.5 × 10 6 × Vi (the measured voltage on
10 11 Ω resistors). Using Eq. (1), Eq. (4) can be rewritten as following:
σ counting statistics ¼
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
nX þ nY
:
nX nY
ð5Þ
John and Adkins (2010) defined a quantity Neff (Eq. (6)) representing
the effective number of counts to be used in estimating counting statistics (Eq. (7)):
Neff ¼ nX nY =ðnX þ nY Þ
σ counting statistics
sffiffiffiffiffiffiffiffi
1
:
¼
Neff
ð6Þ
ð7Þ
This relationship is represented in a log–log plot of measured variability versus signal strength by a linear trend with a −0.5 slope (Fig. 4).
At low concentrations however, the measured isotope ratio variance will be dominated by Johnson noise, the fluctuation of the background signal due to thermal noise in feedback resistors of the
amplifier circuit. This translates into (John and Adkins, 2010):
Vmeasured ¼ Vsignal þ VJohnson noise
10-4
10-5
σ
cou
ntin
g st
atis
tics
10-6
107
108
109
ð8Þ
1010
1011
log(Neff)
Fig. 4. Errors (as relative internal standard error (RSE)) for 34S/32S and 33S/32S ratios of
50 cycles with a 4.194 s integration time over a range of concentration from 0.25 to
50 μmol/l. Signal intensity is given as the effective number of counts collected over
this period (Neff) (Eq. (6)). The predicted error due to counting statistics is based on
a Poisson law (Eq. (7), slope = −1/2, solid line). The predicted error due to Johnson
noise is calculated from independent measurements of Johnson noise (Eq. (9),
slope = −1, dotted line). Two total sulfate concentrations are indicated as an example
for both 33S/32S and 34S/32S ratios.
where Vmeasured is the recorded voltage, Vsignal is the voltage due to
the ion beam captured in the Faraday cup, and VJohnson is the magnitude of the baseline Johnson noise.
The error can then be calculated as
log σ Johnson noise ¼ − logðNeff Þ
The error on the ratio is:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
σX2 σY2
¼
þ
nX 2 nY 2
log(RSE)
ise
no
4. Discussion
þ log
R
ðR þ 1Þ
r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi!
nnoise a 2
þ nnoise b 2
R
ð9Þ
with nnoise the number of ions corresponding to the Johnson noise mean
intensity and R the measured isotopic ratio. This corresponds to a linear
trend with a −1 slope in the Neff plot. We evaluate Johnson noise by
measuring the standard deviation of the intensity of a thousand cycles
of 4.194 s with the Neptune analyzer gate closed. The noise is 40 μV
on the central cup and 17 μV on cups H1 and H3. The intercept in
Eq. (9) is 4.48 for 34S/ 32S and 4.49 for 33S/32S ratios, respectively. As
seen in Fig. 4, S isotope measurements follow the same trends, showing
little or no deviation from theory in terms of internal error. It also shows
that 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 subtraction
One 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 of
δ34S and Δ 33S. This background usually represents 0.1 to 1% of the measured intensity on cups C, H1 and H3. The ratios measured in the blank
for H1/C and H3/C are close to 33S/32S and 34S/ 32S suggesting that the
background signal is mostly due to sulfur. This constant background
might be explained by the presence of sulfate in the solutions or a memory effect in the Aridus. For each measured ratio, the intensity measured
in the blank solution before and after a given standard or sample is
averaged and subtracted from the respective signal. The background
variance is larger than the internal errors and ultimately limits the
56
G. Paris et al. / Chemical Geology 345 (2013) 50–61
precision of our measurements. In the following, we evaluate the influence 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 the
blank.
rblk ¼
xblk
yblk
ð10Þ
To adjust the measured intensities of X and Y, a blank intensity is
subtracted, respectively xblk and yblk. The blank corrected ratio is
Rblk corr ¼
X−xblk
:
Y−yblk
ð11Þ
Eq. (11) can be rewritten using ratios.
Rblk corr ¼
Rm Y−rblk yblk
Y−yblk
ð12Þ
The error on blank corrected ratios can be calculated by propagating errors through the different terms in Eq. (12), including their covariances. However, the calculation can be somewhat simplified as
there is no covariance between Rm and yblk, Rm and rblk, rblk and Y,
or Y and yblk., The propagated error on the blank corrected ratios
can then be written as Eq. (13):
∂Rblk corr 2
∂Rblk corr 2
∂Rblk corr 2
þ σ yblk 2
þ σ Rm 2
∂Y
∂yblk
∂Rm
2
∂Rblkcorr
∂Rblk corr
2 ∂Rblk corr
ð13Þ
þ σ rblk
þ 2σ Rm Y
∂rblk
∂Rm
∂Y
∂Rblkcorr
∂Rblk corr
þ 2σ rblk yblk
∂yblk
∂rblk
σ Rblk corr 2 ¼ σ Y 2
where the terms in the first row are the variances of the signals and
the blanks, and the two terms in the bottom row are the covariances
between the measured ratios and the intensities in both the signal
and blank. The covariance σRmY is defined as:
X Rmi −Rm Yi −Y
¼
ncycles −1
i¼1
ncycles
σ Rm Y
ð14Þ
where ncycles is the number of 4.194 s integrations that are combined
into a single ratio measurement (typically 25 or 50). However, this
formulation of the covariance does not let us do the calculation after
we have processed the raw data into averaged ratios. To solve this
problem, we follow (Rae, 2011) and start by reformulating Eq. (14):
In the end, the covariance is:
X ncycles X−ncycles Rm Y
:
¼
ncycles −1
i¼1
ncycles
σ Rm Y
ð15Þ
Now all of the parameters required to calculate σRmY are available
after averaging the 25 or 50 cycles, making it less cumbersome to estimate the covariance. The same calculation can be applied to σrblkyblk.
It is thus possible to combine the internal error with the propagation
of the background subtraction to get a final estimate of the instrument
blank corrected uncertainty on a measured ratio.
4.1.3. External error
To evaluate external variability, a sample of surface seawater from
the Pacific Ocean (Catalina Island) is purified three times on separate
columns, using 50 μl of seawater for each replicate. The average
δ 34SVCDT value is 20.94 ± 0.08‰ (2sd of the three averaged measured
δ 34SVCDT values) and the average Δ 33S value is 0.04 ± 0.09‰ (2sd,
n = 3). The same process is applied to aliquots of 5 mg of cleaned
powder of the deep-sea coral Desmophyllum diantus collected off
Tasmania during the Southern Surveyor cruise (SS0108 STA011 1/13/
2008). The measured δ 34SVCDT value is 22.06 ± 0.17‰ and Δ33S is
0.12 ± 0.13‰ (2sd, n = 3). This variability is significantly higher
than the internal error. We thus want to understand if the increased
variance is due to the purification or to the mass spectrometry, which
can be done by investigating intermediate errors (where replicates are
duplicate measurements of the same sample solution, not separate
chemical processing of the sample CAS or water sample).
4.1.4. Intermediate error
In the course of a Neptune run, the difference between two measured ratios of the same solution is larger than what would be expected
based on the internal errors. John and Adkins (2010) use the END (Error
Normalized Deviates, normalized to the expected variance from internal statistics alone) statistic to quantify this extra variability and defined
as:
R2 −R1
END ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
:
σ 1 þ σ 22
ð16Þ
If the internal errors completely describe the variability between
replicates, then a large number of END values will describe a histogram with a mean of 0 and a standard deviation of 1. For the seawater
samples that were run in this study, the standard deviation of the
END 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
ncycles
X
1
Xi −Rm Yi −Rmi Y þ Rm Y :
σ Rm Y ¼ ncycles −1 i¼1
The next step is to use the linearity of the sum:
"n
#
ncycles
ncycles
ncycles
cycles
X
X
X
X
1
Xi −
Rm Yi −
Rmi Y þ
Rm Y :
σ Rm Y ¼ ncycles −1
i¼1
i¼1
i¼1
i¼1
Table 4
Average 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 bracketing
standards. Comparing Run 2 and seawater run indicate that the END standard deviation can significantly deviate from 1 even without analyzing actual samples. Therefore,
the source of the instability lies within the mass spectrometry itself.
Run 1
The covariance can now be written as:
h
i
1
ncycles X−ncycles Rm Y−ncycles Rm Y þ ncycles Rm Y ;
σ Rm Y ¼ ncycles −1
X
33/32
ncycles
because λ ¼ n
1
cycles
i¼1
34/32
λi .
Run 2
Seawater run
Time
(min)
Average
sd
Time
(min)
Average
sd
6
30
60
6
30
60
0.01
−0.12
0.01
0.09
0.13
0.02
1.36
1.94
2.53
1.21
1.04
1.3
6
30
60
6
30
60
0.13
−1.47
0.52
−0.09
−0.48
0.1
1.86
5.07
6.99
1.23
1.49
2.11
Time
(min)
Average
sd
30
−0.89
2.37
30
−0.60
2.13
G. Paris et al. / Chemical Geology 345 (2013) 50–61
and Poisson statistics, replicates run over many hours show much
more variability.
We designed two different experiments to explore the source of
this variability. Run 1 is a continuous aspiration of a single sample
where data are averaged every 25 cycles. The autosampler tube was
simply left in a large volume of solution (Fig. 5). Run 2 uses the
same solution but the autosampler tube moved up and down every
25 cycles to inject air into the sample stream (Fig. 6). Means and standard deviations of the different END populations are reported in
Table 4, which shows that END standard deviations increase with
time between bracketing standards and with injecting air every
25 cycles. Overall, the standard deviation of the END is larger for
the 34S/ 32S ratio than for the 33S/ 32S ratio. By comparing Run 2 with
data collected on seawater samples, we find that introducing the
same solution again and again induces the same intermediate reproducibility as a run with standard-sample-standard bracketing and
washes. As the highest variability of these runs overlaps with the
true external sample replicates reported above, it is clear that the origin of our extra variance (after drift correction) is not due to problems with sample matrix, machine memory, or poorly matched
standards. Instead, the problem seems to be within the mass spectrometry itself.
One possibility for the extra variability in intermediate replicates
is that we have not properly accounted for mass bias drift over the
course of the run. Allan variance plots are a good test of this hypothesis. First introduced to estimate the stability of atomic clocks, this
statistic provides an evaluation of the drift for any given system. Considering three evenly spaced measured ratios during a run, xn, xn + 1,
and xn + 2, spaced by a measurement interval τ, the difference between ratios normalized for the sampling interval is given by yn =
(xn + 1 − xn) / τ. The instability in the system may be represented
by the change in this difference across the whole run (for a given
57
time interval): Δyn = yn + 1 − yn=. The Allan variance σ 2y(τ) is
then:
2
σ y ðτÞ ¼
N
−1 N
−1 X
X
2
2
1
1
yiþ1 −yi ¼
x −2xiþ1 þ xi :
2ðN−1Þ i¼1
2ðN−1Þ i¼1 iþ2
where N is the total number of measured ratios used to calculate the
sum. For a system with no drift in the ratios, the Allan variance will
drop as τ increases. We evaluate the Allan variance for two different
runs where the sample was continuously introduced into the instrument, Run 1 (same run as previously described) and Run 1b. The integration time is 4.194 s in each case but Run 1 had an idle time of
3 s between each time step while 1b did not, explaining the slightly
different starting points for the two runs in Fig. 7. The solid lines indicate the theoretical evolution of the variance in the case of no drift. An
unchanging Allan variance with τ indicates linear drift in the mass
bias of our measured ratios. Data are calculated from the integration
time 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 different between the two runs, but no drift affects the Δ 33S values, implicating true mass dependent mass bias drift as the source for
variability between measured ratios. Fig. 7a shows that different
runs have different timescales of variability. Run 1 has an oscillation
of ~ 10–30 min but with no long term change in the mean value of
the ratios, while Run 1b is characteristic of a linear drift in the
uncorrected ratios. 33S/ 32S ratios appear to drift at a longer timescales, but this is due to the smaller size of the 33S beam being inherently more variable (Fig. 7b). In a linearly drifting system with more
noise between the individual points, deviations from the ideal line
first show up at longer periods. In each Run the drift is mass dependant because it does not affect the calculated Δ 33S value (Fig. 7c).
40.75
-0.60
32
n
re
rfe
40.65
ce
S-H
-0.70
e
nt
40.55
δ'33S (‰)
δ'33S (‰)
i
O2
40.45
40.35
-1.00
a
-1.10
0.20
0.10
Δ33S (‰)
-0.10
-0.20
MDF line
y = 0.9793x + 0.8177
R² = 0.9464
0.15
y = 0.8726x-36.929
R² = 0.827
0.00
b
-1.85 -1.75 -1.65 -1.55 -1.45 -1.35
δ'34S (‰)
0.25
δ'34S (‰)
Δ33S (‰)
-0.90
MDF line
40.25
78.55 78.65 78.75 78.85 78.95 79.05
-0.30
-0.80
0.05
-0.05
-0.15
c
-0.25
d
41.90 42.00 42.10 42.20 42.30 42.40
-1.10 -1.00 -0.90 -0.80 -0.70 -0.60
δ33S (‰)
δ33S (‰)
Fig. 5. δ′34S vs. δ′33S and Δ33S vs. δ33S for raw data and drift corrected data during a run where the autosampler probe remained in the same solution the entire time. MDF line is the
Mass Dependent Fractionation line.
58
G. Paris et al. / Chemical Geology 345 (2013) 50–61
42.2
32
S-H
-0.5
2
O
41.9
41.8
MDF line
41.7
-0.8
-0.9
-1.1
-1.9
δ'34S
d
0.2
0.0
0.1
-1.5
-1.3
0.0
-0.2
-0.1
-0.3
-0.2
-0.4
41.4 41.6 41.8 42.0 42.2 42.4 42.6
-0.3
-1.1
δ33S (‰)
-1.7
0.3
0.1
-0.1
MDF line
δ'34S (‰)
Δ33S
Δ33S (‰)
0.2
in
O2
-1.0
41.6
81.0 81.2 81.4 81.6 81.8 82.0 82.2
c
nc
re
rfe
te
-0.7
δ'33S (‰)
δ'33S (‰)
42.0
e
-0.6
in t
42.1
b
erf
ere
nce
a
-0.9
-0.7
δ33S (‰)
-0.5
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 of
ratio measurement.
This is confirmed when we make Allan variance plots with the
corrected ratios (not shown). Allan variance analysis demonstrates
that there is mass bias drift within the 50 cycles we use to average
for a single ratio, and that we might improve the overall δ 34S and
δ 33S ratios with fewer cycles per measurement. However, shorter acquisition would clearly adversely affect the uncertainty in the Δ 33S
value. This analysis also explains why points in Fig. 4 lie slightly
above 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 give
rise to an END standard deviation of 2–4 over and above that due to
internal errors. Fig. 5 shows the relationships between δ 34S, δ 33S,
and Δ 33S 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 the
Na2SO4 solution for a 25 cycle integration. The right column (b and
d) are the same data but now using every other 25 cycle block to
drift correct and normalize the data between blocks. A trend between
the scatter in the δ 34S and δ 33S data is presented in Fig. 5a, but Fig. 5b
shows that drift correction shrinks the δ 34S data much more than the
δ 33S data (the ranges on the x and y-axes are the same in a and b).
Calculating the mass independent Δ 33S variance for each sample
shows that this value is dominated by scatter in the δ 33S data
(Fig. 5c). The Δ 33S and δ 34S data do not correlate (not shown). Drift
correction shrinks the δ 33S vs. Δ 33S trend to a relatively tight 1:1
relationship. Fig. 5d is the signature of an isobaric interference on 33S
but not 34S or 32S.
From the peak shape data in Fig. 2, the 32S 1H hydride species is the
most likely candidate for this interference on δ 33S and Δ 33S. The hydride of 16O dimer might also contribute, but it is almost an order of
magnitude smaller and much easier to mass resolve. With only a
b4 mDa interference free plateau, small changes in the absolute
mass stability could change the relative intensity of the hydride interference. The combined high voltage and magnetic field stability of the
Neptune is about 1 part in 10 5, which corresponds to 3 mDa over a
300 Da full mass range. The peak shapes in Fig. 2 also show that
oxygen dimer interferences are the most likely cause of extra variance
at masses 32 and 34. However, these interferences are very hard to
detect, because they affect the 34/32 ratio with a slope close to the
regular mass dependent drift in the ratio ( 33S/ 32S = 0.007877, 34S/
32
S = 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) and
the competing effects of oxygen interferences and mass bias drift
can be seen. In the non-drift corrected δ 34S vs. δ 33S data (Fig. 6a)
the larger range compared to Run 1 comes from the increased mass
bias effect of changing the plasma conditions by injecting air rather
than from increasing the O2 interferences, as the source of oxygen
for the interference is the water aspirated with the sample. This
drift can be normalized by sample-standard bracketing (Fig. 6b), but
again it does not change the δ 33S data substantially. The predominance of the 32S 1H interference is still visible when plotting δ 33S vs.
Δ 33S (Fig. 6c), but a larger scatter appears, due to mass dependent
fractionation. When the data is mass bias corrected (Fig. 6d), the
strong 1:1 relationship returns.
Overall, it is clear that much, but not all, of the mass bias drift and
O2 interferences on δ 34S can be normalized with sample-standard
bracketing. Unfortunately this technique has little effect on the
short term variability of 32S 1H interference on δ 33S. The END standard
deviation is smaller for the δ 33S data because the internal errors are
larger, not because there is inherently less extra variability on this
ratio. The absolute variance of δ 34S can be reduced, but the internal
errors are so small (~ 20 ppm on normal sized samples) that the
END standard deviation is still large. A longer duration between
bracketing standards leads to larger END standard deviations for
G. Paris et al. / Chemical Geology 345 (2013) 50–61
50 cycles
59
time between two
bracketing standards
Allan variance
1.10-5
1.10-6
a
34
S/32S
Allan variance
1
100
1000
10000
10
100
1000
10000
100
1000
10000
1.10-6
1.10-7
b
1
Allan variance
10
33
S/32S
1.10-1
c
Δ33S(‰)
1.10-3
1
10
Time (s)
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 variation
following 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 solid
line (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 interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
δ 34S because non-linear changes in both mass biasing and the oxygen
interferences are more likely to occur.
To improve our reproducibility on the measured Δ 33S, we need to
better separate the 32S 1H interference. We could achieve this goal
using an entrance slit thinner than 25 μm thus increasing the mass
resolution. END standard deviations for δ 34S will have higher standard deviations than internal errors as long as the sample-standard
bracketing interval is longer than the variability in the water induced
O2 interference on masses 32 and 34.
5. Seawater profiles
The method described in this paper has been applied to a collection
of two seawater profiles from the San Pedro basin (Pacific Ocean, 10
samples, (John et al., 2012)) and the Atlantic ocean (US Geotraces Atlantic, station 9, 24 samples). The first profile goes down to 895 m and the
second one down to 3000 m water depth.
Since the development of the sulfur isotopic measurement, seawater
has been stated to have a constant isotopic composition of ~20‰.
The first published systematic study of seawater δ 34S suggested a
value of 20.12 ± 0.74‰, n = 17 (Thode et al., 1961) followed by
20.99 ± 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 locations
and depth. Rees et al. (1978) are the first ones who systematically
explored potential variations of seawater δ34S values with depth in
the open ocean, followed by a joint study of oxygen and sulfur isotopes
of seawater (Longinelli, 1989) with values ranging from 18.8 to 20‰. A
depth transect of sulfur isotopes has also been established for the North
Sea (Böttcher et al., 2007) with very similar mean values for seawater
samples collected over springtime (δ 34S = 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 more
recent than 1995. The δ34S scale was previously reported using the Canyon Diablo Troilite standard (CDT). The supply of the standard was
eventually depleted and a new scale was defined at the 1993 Vienna
Conference, the Vienna-CDT scale. Because the CDT standard turned
out to be slightly inhomogeneous, this new scale is based on the International Atomic Energy Agency (IAEA) S1 standard, defined as being
60
G. Paris et al. / Chemical Geology 345 (2013) 50–61
δ34SVCDT
δ34SVCDT
20.50
0
20.75
21.00
21.25
21.50
21.75
22.00
Geotrace 9
20.50
0
20.75
21.00
100
21.25
21.50
21.75 22.00
San Pedro Basin
200
500
300
400
500
1000
600
depth (m)
700
1500
800
900
1000
2000
2500
3000
3500
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 statistics
multiplied by the standard deviation of the END (Eq. (16)).
−0.3‰ in the VCDT scale (Krouse and Coplen, 1997; Coplen and
Krouse, 1998; Ding et al., 2001). Here we investigated two depth transects, on a total of 34 samples that place seawater data on the VCDT
scale. Results are presented in Fig. 8. They display a constant δ34S
value with depth. The average on the two sites is 20.96. ± 0.12‰ and
20.99 ± 0.10‰ respectively (2sd). These two values are indistinguishable from one another. They appear however slightly lower than the
value measured for the standard NBS 127, which is barite precipitated
from seawater, or compared to the value measured for modern barites
(21.10 ± 0.20‰, 2sd) (Paytan et al., 1998). The standard deviation of
the END for these runs is shown in Table 4. These profiles exhibit
δ34SVCDT values that are constant with depth within error bars, in agreement with the conclusion that the open ocean has an homogeneous isotopic composition (Rees et al., 1978; Longinelli, 1989). The δ34S values
are in agreement with recent studies showing values for standard
IAPSO 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 Pedro
Basin 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 = 6
samples) measured by gas source isotope-ratio-monitoring MS (Ono
et al., 2012).
6. Conclusions
The method described in this paper allows for the first time a measurement of the triple isotopic composition of sulfur isotopes on samples of a few nmol of sulfate. Although, significant progress had
already been achieved in measurement of S isotopes on the
MC-ICP-MS Neptune (Craddock et al., 2008; Das et al., 2012), we are
able to work on samples one order of magnitude smaller using a
method than can be applied to a broad range of samples such as waters,
carbonates, sulfides or samples available in limited amount such as
foraminiferae. As an example, 20–50 mg of carbonates with less a
100 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 measured
by this method. Applying this method to seawater yields a δ 34SVCDT
value of 20.97 ± 0.10‰ (2sd, n = 34).
Precision on the measured Δ 33S could be improved by working at
a higher resolution than achieved by the slits available for the
Neptune. Such improvement will allow exploration of fine mass independent fractionation in modern samples with a simple method than
does not require specific fluorination line and requires very small
amount of sulfate.
Acknowledgment
The authors would like to acknowledge Nathan Dalleska for precious
help in using the IC, Eric Kort for helping with Allan variance analyses,
Seth John for providing the seawater samples and Nivedita Thiagarajan
for the coral sample. We also thank two anonymous reviewers for their
comments and suggestions. This work was supported by the Camille
and Henry Dreyfus Postdoctoral Program in Environmental Chemistry.
The Adkins and Sessions groups are also deeply thanked for feedbacks
and support. Thanks also to Vicky Rennie for being a motivated early
user of this new method.
References
Albarède, F., Beard, B., 2004. Analytical methods for non-traditional isotopes. Reviews
in Mineralogy and Geochemistry 55 (1), 113–152.
Amrani, A., Sessions, A.L., Adkins, J.F., 2009. Compound-specific δ34S analysis of volatile
organics 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-billionyear-old stromatolites reveal microbial metabolism. Proceedings of the National
Academy of Sciences 109, 15146–15151.
Böttcher, M.E., Brumsack, H.J., Dürselen, C.D., 2007. The isotopic composition of modern
seawater sulfate: I. Coastal waters with special regard to the North Sea. Journal of
Marine Systems 67 (1–2), 73–82.
Chaussidon, M., Albarède, F., Sheppard, S.M.F., 1989. Sulphur isotope variations in
the mantle from ion microprobe analyses of micro-sulphide inclusions. Earth and
Planetary Science Letters 92 (2), 144–156.
G. Paris et al. / Chemical Geology 345 (2013) 50–61
Clough, R., Evans, P., Catterick, T., Evans, E.H., 2006. δ34S measurements of sulfur by
multicollector inductively coupled plasma mass spectrometry. Analytical Chemistry
78 (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 of
sulfate 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 ion
exchange technique for the measurement of δ34S values from microgram quantities
of 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 in
galena 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 sulfur
isotope reference materials and V-CDT with a reassessment of the atomic weight
of sulfur. Geochimica et Cosmochimica Acta 65 (15), 2433–2437.
Halas, S., Szaran, J., 2001. Improved thermal decomposition of sulfates to SO2 and mass
spectrometric determination of δ34S of IAEA SO-5, IAEA SO-6 and NBS-127 sulfate
standards. 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 maximum isotope ratios of selected elements in naturally occurring terrestrial materials
and reagents. (USGS).
Hulston, J.R., Thode, H.G., 1965. Variations in the S33, S34, and S36 contents of meteorites
and their relation to chemical and nuclear effects. Journal of Geophysical Research
70 (14), 3475–3484.
John, S.G., Adkins, J.F., 2010. Analysis of dissolved iron isotopes in seawater. Marine
Chemistry 119 (1–4), 65–76.
John, S.G., Mendez, J., Moffett, J., Adkins, J., 2012. The flux of iron and iron isotopes from
San Pedro Basin sediments. Geochimica et Cosmochimica Acta 93, 14–29.
Kampschulte, A., Bruckschen, P., Strauss, H., 2001. The sulphur isotopic composition of
trace sulphates in Carboniferous brachiopods: implications for coeval seawater,
correlation with other geochemical cycles and isotope stratigraphy. Chemical
Geology 175 (1–2), 149–173.
61
Krouse, H., Coplen, T., 1997. Reporting of relative sulfur isotope-ratio data. Pure and
Applied Chemistry 69 (2), 293–295.
Longinelli, A., 1989. Oxygen-18 and sulphur-34 in dissolved oceanic sulphate and
phosphate. In: Fritz, P., Fontes, J.C. (Eds.), The Marine Environment, A. Handbook
and Environmental Isotope Geochemistry. Elsevier, Amsterdam, pp. 219–255.
Ono, S., Wing, B., Johnston, D., Farquhar, J., Rumble, D., 2006a. Mass-dependent
fractionation of quadruple stable sulfur isotope system as a new tracer of sulfur
biogeochemical cycles. Geochimica et Cosmochimica Acta 70 (9), 2238–2252.
Ono, S., Wing, B., Rumble, D., Farquhar, J., 2006b. High precision analysis of all four
stable isotopes of sulfur (32S, 33S, 34S and 36S) at nanomole levels using a laser
fluorination 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 secondary pyrite in altered oceanic basement. Geochimica et Cosmochimica Acta 87,
323–340.
Paytan, A., Kastner, M., Campbell, D., Thiemens, M.H., 1998. Sulfur isotopic composition
of Cenozoic seawater sulfate. Science 282 (5393), 1459–1462.
Prohaska, T., Latkoczy, C., Stingeder, G., 1999. Precise sulfur isotope ratio measurements
in trace concentration of sulfur by inductively coupled plasma double focusing
sector field mass spectrometry. Journal of Analytical Atomic Spectrometry 14 (9),
1501–1504.
Rae, J., 2011. Boron Isotopes: Measurement, Calibration and Glacial CO2. University of
Bristol, Bristol.
Rees, C.E., Jenkins, W.J., Monster, J., 1978. The sulphur isotopic composition of ocean
water 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 ratio
determinations. Journal of Scientific Instruments 30 (11), 395.
Weyer, S., Schwieters, J.B., 2003. High precision Fe isotope measurements with high
mass resolution MC-ICPMS. International Journal of Mass Spectrometry 226 (3),
355–368.