Improvements in the determination of attogram-sized ²³¹Pa in dissolved and particulate fractions of seawater via multi-collector inductively coupled plasma mass spectrometry

Abstract

A technique is developed to quantify the ultra-trace ²³¹Pa (35–3904 ag) concentration in seawater using multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS). The method is a modification of the process developed by Shen et al. (Anal Chem 75(5):1075–1079, 2003. https://doi.org/10.1021/ac026247r) and extends it to the application of very low levels of actinides, and the 35 ag ²³¹Pa can be measured with a precision of 15%. The total process blank for the water column was 0.02 ag/g, while the values of the large and small particles were ~ 30 ag/g. The ionization efficiency (ions generated/atom loaded) varies from 0.7 to 2.4%. The measurement time is 2–5 min. The amount of ²³¹Pa needed to produce ²³¹Pa data with an uncertainty of ± 0.8–15% is 35–3904 ag (~ 0.9 × 10⁵ to 10 × 10⁶ atoms). Replicate measurements of known standards and seawater samples demonstrate that the analytical precision approximates that expected from counting statistics, and that based on detection limits of 52 ag, 55 ag, and 28 ag, protactinium can be detected in a minimum seawater sample size of ~ 2.6 L for small suspended particulate matter (> 0.8 μm and < 51 μm), ~ 3.0 L for large suspended particulate matter (> 51 μm), and ~ 56 mL for filtered (< 0.45 μm) seawater. The concentration of ²³¹Pa (several attograms per liter) can be determined with an uncertainty of ± 2–8% (2σ) for suspended particulate matter filtered from ~ 60 L of seawater. For the dissolved fraction, ~ 1 L of seawater yields ²³¹Pa measurements with a precision of 0.8–10%. The sample size requirements are several orders of magnitude less than traditional decay-counting techniques, and the precision is better than that previously reported for ICP-MS techniques. Our technique can also be applied to other environmental samples, including river, lake, and cave water samples.

1 Introduction

²³¹Pa is the only long-lived intermediate daughter product of ²³⁵U (t_1/2 = 32,760 years) (Robert et al. 1969) and has been broadly applied as a geochemical tracer in fields as diverse as oceanography, geochronology, and paleoclimatology (Anderson et al. 1983; Asmerom et al. 2000; Bradtmiller et al. 2014; Edmonds et al. 1998; Edwards et al. 1997; Hayes et al. 2014, 2015; Hoffmann et al. 2013; Moran et al. 1997). It is noteworthy that the concentrations of ²³¹Pa and ²³⁰Th in seawater and sediments, as well as their isotopic ratios, can elucidate the representative processes responsible for the removal of particle-reactive elements from the ocean (Anderson et al. 1983; Chase et al. 2003; Hayes et al. 2015; Kretschmer et al. 2011; Lao et al. 1993; Luo and Ku 1999, 2004; Siddall et al. 2005; Walter et al. 1997, 2001). The ²³¹Pa and ²³⁰Th in seawater not only contribute to the clarification of the mechanisms of global ocean circulation processes but also deepen our understanding of the paleoenvironmental record, including processes such as boundary scavenging (Anderson et al. 1983; Hayes et al. 2014, 2015; Kretschmer et al. 2011; Moran et al. 2001, 2002) and ventilation of the modern (Deng et al. 2014; Edmonds et al. 1998; Thomas et al. 2006) and last-glacial (Bradtmiller et al. 2014; Thomas et al. 2007) periods in the Atlantic, Pacific, Indian, and Arctic oceans.

²³¹Pa has an extremely low natural abundance in the marine environment, leading to uncertainty in its measurement. In surface seawater, the dissolved fraction of ²³¹Pa is ~ 500 ag/L, while the particulate fraction is only 10 ag/L (Deng et al. 2014; Edmonds et al. 1998; Hayes et al. 2014; Moran et al. 2001, 2002; Thomas et al. 2006). There are differences in the precisions and the required sample sizes of the different techniques. For example, the traditional a and ß counting techniques typically yield a precision of ± 9–28% (2σ) for the determination of ²³¹Pa (100–4000 fg) (Anderson and Fleer 1982). In addition, these methods require larger sample sizes (~ 75 large samples per measurement batch) and longer counting times (Anderson and Fleer 1982; Anderson et al. 1983; Nozaki and Nakanishi 1985). Thermal ionization mass spectrometry (TIMS) was first applied by Pickett et al. (1994) to perform Pa measurements with a precision of better than ± 1% for sample sizes of hundreds of femtograms of ²³¹Pa. Subsequently, Shen et al. (2003) used TIMS with a precision of ± 4–12% (2σ) for samples containing 100–1000 ag ²³¹Pa (~ 3 × 10⁵ to 3 × 10⁶ atoms). TIMS has been widely used for the determination of ²³¹Pa in different natural materials, such as carbonates, silicates, and seawater (Edmonds et al. 1998; Edwards et al. 1997; Pickett and Murrell 1997). In addition, inductively coupled plasma mass spectrometry (ICP-MS) has also been used for the measurement of ²³¹Pa in both dissolved and particulate forms throughout ocean water profiles. This method improves the detection limits and sample size requirements and has a precision of ~  ± 2% for 100–200 fg seawater samples (Choi et al. 2001). In contrast to the above techniques, a multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS) technique with lower detection limits and a higher precision is proposed in this study. This technique can be applied to various special environments, for example, the detection of small (0.8–51 μm) and large (> 51 μm) particles and surface dissolved samples with low ²³¹Pa concentrations in the Arctic with a better precision. The division of the particle size is of great oceanographic importance (Hayes et al. 2015). It has also been shown that scavenging intensity for the natural radionuclides ²³⁰Th and ²³¹Pa can vary with particle type in ocean. Changing proportions of particle types over geologic time may overprint other influences on the sediment record. While previous work has shown that there is still debate about what are the major carrier phases. The research result of Hayes et al. (2015) suggests that a negligible fraction of Pa or Th exists in the 0.45–0.8 μm size class. Most of Pa and Th exist in particle of size 0.8–51 μm size. The particle collected in ocean includes lithogenic material, particulate organic matter, calcium carbonate, biogenic opal and authigenic Fe and Mn oxides. And absorbed ²³⁰Th is mainly originating from decay of ²³⁴U in dissolved seawater namely < 0.45 μm. Absorbed ²³⁰Th originating from dissolution of U-bearing minerals usually exists small particulate about 0.8–51 μm. Particulate ²³⁰Th originating from decay of ²³⁴U within intact mineral lattices usually exists large particulate (> 51 μm). This detailed classification of particle size is fully linked to the differences in the chemical behavior of the ²³¹Pa and ²³⁰Th being measured.

(1) Samples passing through 0.8 μm filters contain colloidal U and dissolved U, and a negligible fraction of Pa or Th exists in the 0.45–0.8 μm size class. In the water column, Th has a tetravalent oxidation state with a high particle adsorption that is consistent with Pa, and the Pa in the water column is readily absorbed onto suspended particulate matter. (2) Samples between the 0.8 μm and 51 μm filters contain small particles, which usually occur in the convergence, adsorption, and reconvergence processes in the water column due to the slow settling velocity. (3) Samples passing through 51 μm filters usually exist as large particle and usually have a fast settling velocity, indicating that the detrital component in the water column is less affected before sinking to sediments and the sizes of the particles have with oceanographic significance. Among them, the Arctic surface dissolved fraction contains ~ 70 ag/L, and the ice hole contains ~ 30 ag/L. In addition, there are differences in the particulate fraction of ²³¹Pa at different depths in the same environment (Additional file 1: Figs. S1 and S2). The small particulate fraction of ²³¹Pa in the Pacific surface seawater is only 3 ag/L (10^–21–10^–20 molar or 10³–10⁴ atoms/L), while the large particulate fraction at a depth of ~ 1000 m is 15 ag/L. The particulates analyzed are mainly inorganic particles with a limited organic effect. The concentrations of the elements in the dissolved fraction cannot be measured via ICP-MS due to the high detection limits of several ppb and ppt. All of the samples measured in this study were at the ag and fg levels. The ²³⁰Th and ²³¹Pa concentrations of the particulates with sub-fg and sub-pg concentrations also cannot be measured via ICP-MS or CHNS analysis because of the high detection limits of several ppb and ppt. All measured samples here are all ag and fg levels. ²³⁰Th and ²³¹Pa of particulate with sub-fg and sub-pg also cannot be measured by ICP-MS or CHNS analysis because of high detection limits getting to ppb and ppt. In conclusion, the quantification of ultra-trace ²³¹Pa (35–3904 ag) concentrations in seawater via MC-ICP-MS is available for the measurement of samples of several orders of magnitude at the ag and fg levels, and this extension of the determination of very low ²³¹Pa concentrations is both novel and challenging.

The lowest value previously reported for the blanks was 16 ± 15 ag (Shen et al. 2003), and the detection limits were 38 ag for particulate matter and 49 ag for the dissolved fraction. In this study, we mainly modified the method proposed by Shen et al. (2003), and the main contribution of this study is the characterization and testing of this method for very small protactinium loads in the dissolved fractions of seawater for small and large particulates (Fig. 1). The signal/noise ratio was improved due to the high ionization efficiency and fast analysis of a large-quantity samples with a limited analysis time in a case producing more ²³³U decayed from ²³³Pa (decay constant λ_223Pa = 0.0256486 d⁻¹) for analysis with a slow time. However, the application of MC-ICP-MS to the analysis of the seawater particulate and dissolved fractions containing as little as tens of attograms of Pa still remains challenges. (1) The procedural blank should be kept at as a low level as possible. (2) A high ionization efficiency must be maintained (e.g., ~ 0.7–2.4%). (3) Measurements with small ion beams must be precise and accurate. Among the other characteristics, a low detector dark noise is required. With the introduction of ²³³Pa (t_1/2 = 27.0 days; Jones et al. 1986) as a spike in MC-ICP-MS analysis, the dark noise inevitably increases. Due to the decay of ²³³Pa to ²³³U in a short time range, the dark noise of the detector increases from 70 to 143 cpm in 10 h. When the samples have been measured, we increase the machine sensitivity (ionization efficiency get to from 0.7 to 2.4%) to obtain a high signal/noise ratio when we measure sample in a limited time from 2 to 5 min/per sample. TIMS analysis of ²³³Pa is slow, usually ~ 1 h/per sample, which is not good for a large quantity of samples per batch, because of the short decay time from ²³³Pa to ²³³U in the entire analysis period. 4) A high sample throughput is required to obtain large-scale ocean ²³¹Pa data. The analysis of ultra-trace ²³¹Pa in the dissolved seawater and particulate fractions requires the method with small sample sizes, short measurement times, a low procedural blank, a high instrumental sensitivity, and a high signal/noise ratio. In this study, we mainly utilized MC-ICP-MS to determine the concentration of ultra-trace ²³¹Pa (35–3904 ag) in seawater. This method is feasible and has a better sample size, precision, and accuracy than previously reported ICP-MS techniques.

Fig. 1

a Natural processes of ²³¹Pa in seawater and b procedures for the determination of ²³¹Pa concentration via the modified MC-ICP-MS technique developed in this study

2 Experimental methods

2.1 Reagents and materials

This study was performed in the clean room at the Minnesota Isotope Laboratory (MIL) for chemical procedure analysis. A Super-Q purification system was used to purify the deionized/distilled water, and the bottles and beakers were cleaned in an acid vat (Edwards 1988). The ultrapure reagents were obtained from Aristar Ultra Chemicals, including HCl, HF, HClO₄, and NH₄OH. The redistilled HNO₃ was obtained from GFS Chemicals Inc. First, 200 g of anion-exchange resin (Spectrum Spectra Gel, AG 1-X8, 100–200 mesh) was filled into a quartz column and was sequentially cleaned using the following cleaning steps: (1) 250 mL of H₂O, 250 mL of 14 M HNO₃; (2) 500 mL of H₂O; (3) 250 mL of H₂O, 250 mL of 14 M HCl; (4) 500 mL of H₂O; (5) 490 mL of H₂O, 10 mL of HCl; and (6) 500 mL of H₂O. The columns used for the ion separation were constructed of heat-shrink Teflon tubing (4:1, ½ in.) from Texloc Ltd. and were cleaned sequentially via boiling in aqua regia, AR HNO₃ (7 M), AR HCl (5 M), distilled HNO₃ (0.1 M), and Aristar Ultra HCl (0.1 M).

2.2 ²²³Pa spike solution preparation

Our modification of the ²³⁷Np milking method described by Shen et al. (2003) was to obtain the internal standard (²³³Pa), which is described below. The measurement conditions and improved separation methods for MC-ICP-MS are presented in Additional file 1 (Tables S1 and S2).

(1) The solution in the stored Np vial was dried, and five drops of HClO₄ were added to the ²³⁷Np solution. (2) The solution was dried, redissolved using 1 column volume (cv) of 9 M HCl, and passed through an 850 μL ion-exchange column, and the eluant was collected in the Np vial. (3) The Pa fraction was eluted twice via the addition of 2 CVs of 9 M HCl + 0.05 M HF to the Pa vial. The collected Pa solution was dried using three drops of HClO₄. (4) The Np fraction was eluted into the Np vial by adding 8 CVs of 1 M HCl + 0.5 M HF. (5) The collected Pa fraction was dissolved using 1 cv of 9 M HCl + 0.05 M HF and was loaded into column 2. (6) Steps (3), (4), and (5) were repeated 10 times in columns 2 and 3 to purify the Pa from the Np. (7) The collected Pa fraction was dissolved using 1 cv of 9 M HCl and was loaded into column 4. The Th fraction was eluted twice via the addition of 2 CVs of 9 M HCl to the Np vial. (8) The Pa fraction was eluted twice via the addition of 2 CVs of 9 M HCl + 0.05 M HF to the Pa vial. The collected Pa solution was dried using three drops of HClO₄. (9) Steps (7) and (8) were conducted one time in column 5 to further purify the Pa from Th interference. After collecting the last Pa fraction, the solution was dried and redissolved using five drops of HNO₃. In a clean room, the Pa solution was diluted to ~ 100 mL using 1 M HNO₃ + 0.05 M HF and at heated at 70 °C overnight to equilibrate the spike solution.

2.3 Standards and samples

The ²³¹Pa solutions (Pa-I, 3.04 × 10⁹ atoms/g; Pa-II, 4.09 × 10⁹ atoms/g) were separated from the uranium ore solutions. The ²³¹Pa was calibrated using a ²³³Pa tracer solution, which was calibrated with a ²³³Pa tracer solution, which was calibrated with a solution of a zircon from the Piper Gulch Granodiorite that records concordant U-Pa ages (Asmerom et al. 1990) and may be in secular equilibrium with ²³⁴U and ²³⁰Th (Cheng et al. 2000). Interlaboratory calibration between the Minnesota Isotope Lab (based on the same mixed ²³¹Pa–^230Th–²³²Th) and Lamont Laboratory demonstrated that the independent calibrations were reproducible to better that 0.2% (Edwards et al. 1997). A more dilute standard (Pa-III, 1.85 × 10⁹ atoms/g or 710 fg/g) was prepared via dilution of an aliquot of Pa-I.

The particulate fraction of seawater samples (~ 500 L) collected from three depths (25–5000 m) in the Pacific was used for the replicate analyses. An additional 25 seawater samples (including dissolved samples collected in the Arctic during Geotrace 2015 and small and large particulate fractions collected in the Pacific during Geotrace 2013) were filtered through an acid-cleaned 51 μm pore-size Sefar polyester mesh prefilter and a 0.8 μm pore-size Pall Supor 800 polyethersulfone filter, and approximately 500 L of seawater was filtered to collect the particulate fractions. The filtered seawater samples were stored and acidified (20 mL of 6 M HCl) in acid-cleaned polyethylene bottles.

2.4 Chemical analysis

The chemical procedures were optimized according to the methods used in previous studies, and Th samples were also collected. The measurement time has been 2–5 min according to the sample content based on the signal/noise ratio, usually the signal of ²³³Pa is 600 cps. When we measured a 1 L sample, the ²³¹Pa content of the sample varied from 35 to 4000 ag. The dissolved fractions of the seawater samples were acidified (pH of ~ 2) using 6 M HCl, 30–80 fg of ²³³Pa tracer was added, and 450 mg of Fe was added to 1 L of dissolved sample. To ensure sample-spike equilibrium, the solution was shaken by hands for 4 days, and then, NH₄OH was added to the solution to co-precipitate Pa with Fe (OH)₃ (pH of 8–9 for this solution). The concentration of ²³¹Pa (35–3904 ag) in seawater is an ultra-trace level, and the background value of ²³¹Pa is very low. A large number of samples (~ 56) were analyzed in the first batch and the other 28 samples were analyzed in the second batch to avoid contamination, and more magnetic stirrers were used at the same time. We preferred the hand-shaking method to the magnetic stirrer method. Hand shaking was conducted for 2 days during the experimental phase. Each sample was shaken hourly and eight times per day to ensure adequate mixing of the solution, diluent, and FeCl₂. The sample was then left to stand for 2 days in order to achieve effective flocculation and precipitation of Fe(OH)₃ using NH₃·H₂O. The overall chemical procedure yield was determined by measuring the ²³³Pa content of the diluent, and the yield consistently exceeded 90%. The replicate sample results of every batch were accurate within the error, and the results of the surface water (SW) standards STD 2010-1 and STD 2015-1 for every batch with the same column chemistry were also consistent within error. This precipitate was allowed to settle for 1–2 days before the overlaying seawater was siphoned off. Then, the Fe precipitate was transferred to a centrifuge tube for centrifugation and as rinsed with deionized H₂O (> 18 MΩ) to remove the major ions from the seawater. The precipitate was then dissolved in 14 M HNO₃ and transferred to a Teflon beaker. It was then dried and subjected to anion-exchange chromatography in 7 M HNO₃ using AG1-X8, 100–200 mesh-size resin, and a polyethylene frit. The initial separation was performed in Teflon columns with a 0.75 mL column volume. The sample was loaded in one CV of 7 M HNO₃, followed by the addition of 1.5 CVs of 7 M HNO₃, 3 CVs of 8 M HCl (to collect the Th fraction), and 3 CVs of 8 M HCl + 0.015 M HF (collect Pa fraction). The Pa and Th fractions were then dried down in the presence of two drops of HClO₄ and dissolved in 7 M HNO₃. They were each passed through the second and third columns (each with a 0.5 mL column volume) using similar elution schemes. The final Pa and Th fractions were then dried down in the presence of two drops of HClO₄ and were dissolved in 2% HNO₃ + 0.01% HF for analysis on the mass spectrometer.

The filtered particulate fraction was analyzed as follows: 1/8 of a filter was placed in a 30 mL Teflon beaker, and the aliquots of the artificial isotope yield monitors of ²²⁹Th and ²³³Pa were weighed. The filters were completely submerged in 7 M HNO₃ combined with 10 drops of HF, were tightly capped with a Teflon threaded cap, and were heated for 10 h at ~ 93.3 °C (200°F) to dissolve/leach the particulate samples under pressure. The leached solution was then transferred to a second acid-cleaned Teflon beaker and was separated from the remaining filter, and then, five drops of HClO₄ were added to the leached solution in the second beaker. The original beaker walls and caps were washed with small amounts of diluted HNO₃ for low concentration ~ 10% and the resulting solution was added to the second beaker. The solution was then dried down, dissolved in 2 M HCl, and transferred to a 15 mL centrifuge tube, along with the 2 M HCl rinse solution. One drop of dissolved Fe and six to nine drops of NH₄OH were added to raise the pH to 8–8.5, at which time the iron (oxy) hydroxide precipitated. This precipitate was then centrifuged, decanted, washed with deionized H₂O (> 18 MΩ), centrifuged, dissolved in 14 M HNO₃, and transferred to a Teflon beaker. It was then dried down and dissolved in 7 M HNO₃ for anion-exchange chromatography using AG1-X8, 100–200 mesh resin, and a polyethylene frit. The initial separation was performed using Teflon columns (internal diameter of ~ 0.35 cm) with a ~ 0.55 mL column volume. The sample was loaded in one CV of 7 M HNO₃, followed by 1.5 CVs of 7 M HNO₃, 3 CVs of 8 M HCl (collection of Th fraction), and 3 CVs of 8 M HCl + 0.015 M HF (collection of Pa fraction). The Pa and Th fractions were then dried down in the presence of two drops of HClO₄ and were dissolved in 7 M HNO₃. They were each passed through the second and third columns (each with a column volume of ~ 0.55 mL) using similar elution schemes. The final Pa and Th fractions were then dried down in the presence of two drops of HClO₄ and were dissolved in 2% HNO₃ + 0.01% HF for mass spectrometer analysis. The chemical recoveries were typically 90–95% for the dissolved and particulate fractions.

The concentrations of ²³²Th, ²³⁰Th, and ²³¹Pa were calculated via isotope dilution using nuclide ratios determined on a Thermo-Finnigan Neptune mass spectrometer. All of the measurements were conducted using a peak jumping routine in ion counting mode on the discrete dynode multiplier after the retarding potential quadrupole. A total of 50–250 sets of data were measured for each sample. Since the centering of peaks could not be easily verified with small ion beams for ²³¹Pa and ²³³Pa, an additional mixed ²³¹Pa–²³³Pa standard was used for this purpose. A solution of ²³³U–²³⁶U tracer was analyzed to determine the mass deviation correction (assuming that the mass fractionation for Th and Pa was the same as that for U). Each sample measurement was bracketed by measurements of an aliquot of the run solution in order to correct for the influence of the instrument background count rates on the masses measured. The analyses (MC-ICP-MS) were conducted within 24 h of the chemical processing, and the study revealed that the small amount of ²³³U produced via the decay of ²³³Pa during this interval evaporated prior to the start of the data acquisition to diminish this isobaric interference, i.e., to minimize the decay of ²³³Pa to ²³³U in the Pa cut (Thomas et al. 2006). Throughout the analysis, the multiplier dark noise gradually increased from an initial value of < 1.0 counts per second (cps) to 1.2–1.4 cps due to the introduction of the short-lived ²³³Pa radionuclide. This background dark noise was tracked between samples, and a correction was applied during the data processing.

An aliquot of two intercalibrated working standard solutions of ²³²Th, ²³⁰Th, and ²³¹Pa, SW STD 2010-1 and STD 2015-1, were added to a separate acid-cleaned Teflon beaker along with weighed aliquots of the ²²⁹Th spike and ²³³Pa spike. The spike and standard were equilibrated for 3 days. They were then dried down and dissolved in 7 M HNO₃ for anion-exchange chromatography using AG1-X8, 100–200 mesh-size resin, and a polyethylene frit, and they were also processed using the sample procedure used for the samples. The procedural blanks for the chemical and mass spectrometric analyses conducted in the Minnesota lab were approximately 28 ag for ²³⁰Th and 18 ag for ²³¹Pa.

An improved and more consistent ionization efficiency of 0.7–2.4% was used for the sample measurement, which was better than our previous work (ionization efficiency of 3–5‰) (Shen et al. 2003). Additional advantages include a reduced measurement time, from ~ 1 h to 2–5 min, and an increased ion beam intensity. The typical beam intensities for ²³¹Pa and ²³³Pa were 8–470 cps and 800–5000 cps, respectively, and thus, the signal/dark noise ratio was increased by a factor of 3–6.

2.5 Data processing

We completed the data processing offline, and the measurements were calibrated sequentially, including mass fractionation correction, instrumental tail correction, diluent factor adjustment, blank subtraction, and internal growth correction for ²³¹Pa and ²³⁰Th. The uncertainties were calculated to be at the 2σ level, representing the maximum of the counting statistics. The corrections were fully propagated via mass fraction correction, dark noise, abundant sensitivity, blanks, and spikes. The abundance sensitivity was 2.6 × 10^–7 at a one atomic mass unit (amu) difference and 5.2 × 10^–8 at a 2 amu difference after the RPQ filter. Radioactive decay corrections were performed for the ²³³Pa in the spike solution between calibration and analysis time.

3 Results and discussion

3.1 Blank and limit of detection

After 11 separation and purification step repetitions of the ²³³Pa spike from the ²³⁷Np solution, the ²³¹Pa was no longer measurable (²³¹Pa/²³³Pa < 7 ppm) and the amount of ²³¹Pa in each sample was negligible. The total ²³¹Pa blank from the chemicals (including the spikes, resins, iron, and acids) in the dissolved sample was 0.02 ± 0.006 ag/g (Table 1). The total procedural blanks, including filtration, were 34 ± 6 ag/g (1/8 filter) (0.8 μm Pall Supor800 polyethersulfone filters) and 29 ± 8 ag/g (1/8 filter) (51 μm Sefar polyester mesh prefilter), all of which are used for the seawater particulate sampling (Table 1).

Table 1 The total amount of ²³¹Pa in the blank derived from the chemicals, including the spike, resin, Fe, and acids, and the total procedural blank, including filtering for an acid-cleaned 0.8 μm Pall Supor800 polyethersulfone filter (for small particulates) and 51 μm Sefar polyester mesh prefilter (for large particulates)

The dark noise in our system was ~ 1.0 cps, and the noise/signal ratio was typically between 5 and 10% for samples with ²³¹Pa < 0.3 ag/g, and less than 3% for the samples had ²³¹Pa > 0.4 ag/g (Fig. 2). On the basis of the general definition of the detection limit, the blank + 3δ_B, where δ_B is the standard deviation of the blank, the minimum detectable amount of the dissolved sample ²³¹Pa was 38 ag, that for using one-eighth of a 0.8 μm Pall Supor800 polyethersulfone filter was 52 ag, and that for using one-eighth of a 51 μm Sefar polyester mesh prefilter was 53 ag.

Fig. 2

Relationship between the analytical uncertainty for the sample (triangle), chemical blank (diamond), and secondary electron multiplier (SEM) dark noise (squares) and the amount of ²³¹Pa measured in a the dissolved fraction of 25 seawater samples, b the small particulate fraction of 14 seawater samples, and c the large particulate fraction of six seawater samples. The seawater dissolved samples were collected from the Pacific Ocean and the particulate samples were collected from the Arctic Ocean

3.2 Precision and accuracy

The precision is eventually constrained by the counting statistics: 2σ = 2 (N^−1/2), where N is the total number of ions counted. Due to the high ionization efficiency (0.7–2.4%), the high ²³¹Pa ion beam to dark noise ratio (9–457), and the low procedural blank (0.02 ag/g), the ²³¹Pa measurements had a precision of 1–3.9 ag/g (0.8–2%), including the counting error for ²³³Pa. Figure 2 shows the relationship between the sources of error (including the sample size, chemical blank, and mass fractionation correction) and the total analytical uncertainty for the Pacific seawater particulate samples and Arctic seawater dissolved samples. For samples containing ²³¹Pa > 0.3 ag/g, the Pa counting error accounted for the majority of the total uncertainty, and the procedural blank accounted for less than 7%. For samples containing ²³¹Pa < 0.3 ag/g, the intensity of the ²³¹Pa ion beam was always less than 8 cps. When the amount of ²³¹Pa was less than 0.1 ag/g, the portion of the total uncertainty caused by the counting error was less than 10%, and the portion caused by the blanks was greater than 20%.

The results from replicate analyses of the ²³¹Pa-2010 and ²³¹Pa-2015 standards are presented in Table 2. The internal precision (intra-run) was 0.5% for a sample size of 18 fg. For multiple measurements of the sample sizes, the external uncertainties (between-runs) were comparable to the internal uncertainties for small particulates. These results indicate that the internal uncertainty was an accurate measure of the true uncertainty. The key sources of this uncertainty were the counting statistics (primarily for samples greater than 300 ag) and the blank correction uncertainties (primarily for samples less than 100 ag). The measured concentrations for all of the replicates were consistent with the gravimetric measurements and with measurements obtained in the Lamont–Doherty Earth Observatory (LDEO) Laboratory, Columbia University (Table 2), indicating that the MC-ICP-MS technique is accurate overall.

Table 2 Replicate analyses of the ²³¹Pa-2010 and ²³¹Pa-2015 standards

3.3 Seawater sample analysis

The ionization efficiency was 1.0% when ²³¹Pa was analyzed via MC-ICP-MS, and the precision was ± 0.8–6% when the sample size was 100–3900 ag ²³¹Pa. As a result, we would need 5 L of surface seawater (²³¹Pa of ~ 20 ag/L) or 2 L of deep seawater (²³¹Pa of ~ 50 ag/L) in order to obtain suspended particulate matter samples of this precision via filtration (Moran et al. 2001, 2002). In contrast, 0.2 L of seawater (²³¹Pa of 500–4000 ag/L) was required for the dissolved fraction samples (Moran et al. 2001, 2002). With respect to the detection limits, the minimum surface seawater sample size was 2.5 L for the particulate fraction and 0.2 L for the dissolved fraction. Therefore, for routine seawater ²³¹Pa analyses, we recommend sample sizes of 0.5–1 L for the dissolved fraction and 5–10 L for the suspended particulate fraction.

Figure 3 presents the results of the replicate analyses of seawater (Arctic and Pacific) and small/large particulate samples (Pacific). The dissolved and small particulate samples are consistent with each sampling depth within the margin of error. The similarity of the magnitudes of the external and internal uncertainties for each sample shows that the internal error can represent the true precision. Nevertheless, the external uncertainties for the Pacific large particle samples are larger than the individual internal uncertainties (Fig. 3c), indicating that the internal errors may not be representative of the true precision of the large particulate samples, but the external uncertainty was predominantly less than 12%. The suspended particulate matter samples obtained filtering ~ 500 L seawater at different depths through 1/8 of a filter contained 200–2000 ag of ²³¹Pa and had a measurement accuracy of ± 1.0–5.6%. The precision of the two replicates of one particle sample was ± 1.5–5.9% from two replicates of six small particulate samples obtained via 1/8 filtration of ~ 500 L of seawater containing 200, 280, 380, 1830, and 5780 ag. All of the samples were from the Geotrace program. The particulate matter is not only for Pa and Th but also for the analysis of other trace metal elements (Schlitzer et al. 2018), and all of the samples were filtered in the field when sampling seawater. The concentration detection limit was also useful for particles. Due to the heterogeneity of the particulate matter, in this study, 1/8 of a filter was used uniformly for the chemical analysis, and realistic sample sizes were used to obtain reproducible evaluation results. This method is useful for sub-femtogram size applications, with an error of ~ 1%, and it is useful for attogram size applications, with an error of ~ 15% (Fig. 3). From two replicates of five large particulate samples, two replicates of one large particulate sample were analyzed with an accuracy of ± 1–20% for 1/8 filtration of ~ 500 L of seawater from depths of 900–3900 m, with a ²³¹Pa content of 500–3800 ag. The results demonstrate that the ²³¹Pa chemically separated from the seawater had essentially the same analytical power as the ²³¹Pa in the standard solution.

Fig. 3

Replicate measurements of the dissolved seawater sample from the Arctic and the two particulate fractions filtered from the seawater collected from the Pacific: a concentration of ²³¹Pa per gram of dissolved seawater; b concentration of ²³¹Pa in the small particles (0.8–51 μm); c concentration of ²³¹Pa in the large particulate matter (> 51 μm). The blue color in the three panels indicates the actual measured concentration of ²³¹Pa in the sample, and the red color indicates the corresponding measurement error or uncertainty

4 Conclusions

In this study, we improved and characterized the MC-ICP-MS technique for measuring attogram-sized ²³¹Pa concentrations in filtered seawater and particulate matter samples. Compared to previous ICP-MS methods, the chemical blanks had low levels in our study. The total process blank for the water column was 0.02 ag/g, while the concentrations of the large and small particles were ~ 30 ag/g. Considering the higher ionization efficiency of the samples (0.7–2.4%), we achieved a relatively high ion beam signal to dark noise ratio, resulting in the results having reduced sample size capabilities (< 100 ag ²³¹Pa). In addition, a large volume of water sample is often required for routine seawater ²³¹Pa analyses, but in this study, we achieved a reduction in the sample volume. We recommend volumes of 0.5–1 L for the dissolved phase and 5–10 L for the suspended particulate matter. In addition, for the fluctuation of the dark noise during the sample analysis, we evaluated the stability of the dark noise via multiple corrections of the dark noise, multiple mass fractionation of ²³⁶U/²³³U (U-F), and cross-assessment of Pa standards STD 2010-1 and STD 2015-1. In summary, measuring ultra-trace ²³¹Pa (35–3904 ag) concentrations in seawater using MC-ICP-MS is applicable to measurement of samples with ag to fg concentrations. This method is not only applicable to surface water and deep water but also has broad applications to other earth science fields, including ²³⁵U–²³¹Pa geochronology, igneous geochemistry, and paleoclimatology.