Determination of toxic trace impurities in titanium dioxide by solid sampling-electrothermal vaporization-inductively coupled plasma mass spectrometry

Abstract

This paper reports on the performance of solid sampling-electrothermal vaporization-inductively coupled plasma mass spectrometry (SS-ETV-ICPMS) for the direct multi-element determination of six toxic trace impurities (As, Cd, Hg, Pb, Sb and Zn) in three different TiO₂ samples. Careful selection of the most suitable target nuclides for the different analytes and temperature program optimization permit the reliable monitoring of the analytes of interest, while avoiding the occurrence of potentially relevant spectral interferences. A pyrolysis-free program, in which simultaneous vaporization of all the analytes is carried out at a relatively low temperature (1700 °C), was used. In this way, sample matrix and analyte co-vaporization is avoided, considerably reducing matrix effects and allowing the determination of Zn, which would otherwise be hampered by a spectral overlap with TiO⁺ ions. Calibration against aqueous standards was found to be feasible. Addition of 50 ng Pd as carrier agent improved the linearity of the calibration curves. Two different internal standards (¹⁰⁵Pd⁺ and ¹²⁵Te⁺) were used to compensate for matrix effects. The method thus developed exhibits interesting features: low limits of detection (ng g⁻¹ range) for all of the elements, at least an order of magnitude lower than those for digestion-based procedures; high sample throughput (maximum 35 min per determination), contrasting with the 2 h required for sample digestion and subsequent analysis; low sample consumption (a few milligrams only); precision values usually in the 9–13% RSD range; and last but not least, the absence of any sample pre-treatment, with the subsequent lower risk of analyte losses, contamination or personal and instrumental harms derived from the necessity of using hazardous reagents for sample digestion. SS-ETV-ICPMS therefore seems to be a promising alternative for industrial control analysis of TiO₂ samples with analyte contents ranging from a few ng g⁻¹ to several hundreds of µg g⁻¹.

---

1. Introduction

Titanium dioxide (TiO₂) is a white crystalline substance, the chemical and physical properties of which rendered it into one of the compounds most used in various branches of modern industry and a variety of other applications. As a result of this extensive use, numerous international regulations limit the concentration levels of different elements in this material. The quality, purity and safety of titanium dioxide is thus controlled to make the product fit-for-purpose for every particular application.

In the field of advanced materials and technology, for instance, the main concern is assuring the purity of this material. The physicochemical properties of titanium dioxide can be significantly affected by some metal impurities present at trace and ultra-trace levels (including alkali, alkaline earth and a number of other elements such as Al, Co, Cr, Cu, Fe, Ni, Si, V or Zr) which may compromise its use in some applications. Keeping this in mind, it is not surprising to find several papers in the literature dealing with the determination of some of these elements by means of different solution-based and solid sampling analytical techniques, such as neutron activation analysis (NAA),^1,2X-Ray fluorescence spectroscopy (XRF),³ direct solid sampling, slurry sampling and solution-based ICP-atomic emission spectrometry (ICP-AES),^4–6 solution-based ICP-mass spectrometry (ICPMS)³ or solid sampling/slurry sampling graphite furnace atomic absorption spectrometry (GFAAS).^7,8

The largest quantity of TiO₂ however, is still used as a white pigment (European additive E171⁹), in the paint, food, pharmaceutical and cosmetic industries.¹⁰ In all these cases, surveillance of the product's properties is not as critical as human safety assurance, and this is why the contents of other target elements are normally controlled in TiO₂ intended for these kinds of applications. In particular, various national and international standards (European Commission Directive 95/45/EC, Food and Drug Administration (FDA) Regulation 21-CFR, European Pharmacopoeia, Pharmacopoeia of the USA, etc.) have set limiting values for the contents of six toxic impurities in TiO₂ samples intended for use in human consumption products. These limits are: As (3 µg g⁻¹), Cd (0.5 µg g⁻¹), Hg (0.5 µg g⁻¹), Pb (10 µg g⁻¹), Sb (50 µg g⁻¹) and Zn (50 µg g⁻¹). Thus, the development of suitable analytical methods, capable of providing reliable results for these six elements in titanium dioxide samples rapidly, is required. Surprisingly, only few works can be found in the literature focusing on those analytes,^11–13 and none of these reports the simultaneous determination of all of these elements in a fast and convenient way, meeting the requirements of industrial quality control analysis.

In principle, the extension of the existing methods for TiO₂ analysis cited in the second paragraph to the determination of these six analytes could be considered as the most reasonable solution for this situation. However, different problems arise when trying to do so in the context of routine analysis. First, the traditional approach, consisting of sample dissolution and subsequent analysis of the digest thus obtained by means of AAS, AES or ICPMS, cannot meet the requirements for industrial control analysis, as the protocols based on this scheme are usually labour-intensive, time-consuming and require the use of relatively large amounts of hazardous reagents, such as concentrated hydrofluoric acid and/or sulfuric acid.^3,5,6 Moreover, the dissolved matrix components and the reagents used for digestion may give origin to spectral and non-spectral interferences, so that matrix/analyte separation often needs to be carried out prior to analysis in order to achieve the limits of detection (LODs) required.^1,3,14,15 These sample pre-treatment procedures are accompanied by a higher risk of analyte loss or contamination, a critical issue in cases where highly volatile (e.g., Hg) or contamination-prone elements (e.g., Zn) are considered.¹

The use of direct solid sampling techniques can overcome these difficulties. However, most of the solid sampling methods proposed in the literature for TiO₂ analysis are not very well-suited for this particular situation either, as they require the use of complicated instrumentation, beyond the possibilities of routine analysis laboratories (neutron activation methods^1,2), lack sufficient sensitivity (XRF and ICP-AES^3–5), or provide mono-element determination only (GFAAS^7,8), increasing the analysis time considerably. As a consequence, there is still a need for the development of fast routine multi-element methods for this particular application.

In this context, the use of solid sampling-electrothermal vaporization-ICPMS (SS-ETV-ICPMS) could be very interesting, as it has proven itself to be: (i) a reliable technique, (ii) practically sample pretreatment-free, (iii) a multi-element technique and (iv) characterized by a high sample throughput. All these characteristics make the technique well-suited for industrial control analysis.¹⁶ Only one attempt at using this technique to titanium dioxide analysis can be found in the literature.¹³ In that paper, the use of slurry sampling ETV-ICPMS was evaluated, with satisfactory results in terms of accuracy. The main disadvantage of this approach was an additional sample pre-treatment step of 30 minutes duration that was required to achieve a suitable particle size. This step limits the sample throughput and increases the risk of contamination. Further, the absolute LODs reported were similar to those obtained after sample digestion, owing to the inherent dilution of the samples during the slurry sampling protocol used. Finally, one of the most volatile target analytes (Cd) was not investigated in that work. A truly direct solid sampling (SS)-ETV-ICPMS method for these samples would have significant benefits.

This article reports on a study wherein the potential of SS-ETV-ICPMS for the direct determination of As, Cd, Hg, Pb, Sb and Zn in TiO₂ samples was evaluated. In order to achieve this goal, three different samples intended for different uses and presenting very different analyte levels were selected for the study. Optimization of the instrumental parameters in order to permit the simultaneous determination of these elements, while at the same time avoiding the influence of spectral interferences, was carried out. Special attention was devoted to the development of an accurate, but straightforward calibration strategy using aqueous standards.

2. Experimental

2.1. Samples and standards

2.1.1. Standards and reagents. Ultra-pure water was purchased from ChemLab (Zedelgem, Belgium). 14 mol L⁻¹ HNO₃ (Panreac, Barcelona, Spain) and 18 mol L⁻¹ H₂SO₄ (Merck, Darmstadt, Germany) were further purified in-house by sub-boiling distillation in quartz equipment. Ammonium sulfate of analytical grade was purchased from Merck. Standard solutions were prepared by appropriate dilution of commercially available 1 g L⁻¹ single element standards (Sigma-Aldrich, Saint Louis, USA; Merck, Darmstadt, Germany and Alfa Aesar, Karlsruhe, Germany) with 0.14 mol L⁻¹ HNO₃.

2.1.2. Samples. The reference material Titanium Dioxide Standard Reference Material 154c was obtained from NIST (National Institute of Standards and Technology, Gaithersburg, USA). This material is presented as a fine powder (particle size < 45 µm) and has a certified purity of 99.591 ± 0.062% TiO₂. Information values for the concentration of various metallic and non-metallic impurities (including the analytes under study) are available from the certificate of analysis. For Cd and Hg, however, only the fact that their concentration is below 100 ng g⁻¹ is indicated.

Two other commercially available TiO₂ samples were also analyzed: TiO₂ > 99% from UCB (Leuven, Belgium) and White pigment—Blanc de Titane (99% TiO₂), purchased from Sennelier (Paris, France). All of these were available as powders.

2.2. Instrumentation

A Perkin Elmer (Boston, USA) HGA-600MS electrothermal vaporizer coupled to a Perkin Elmer Sciex Elan 5000 ICP-mass spectrometer was used for the solid sampling analysis. The characteristics of the ETV device and its connection to the ICPMS device were described in a previous work.¹⁷ Pyrolytic graphite-coated tubes and cups (cup-in-tube technique for solid sampling) were used in combination with this device. A microbalance (Sartorius M3P, Gottingen, Germany) with a readability of 1 µg was used for weighing the samples. The ICPMS OmniRange settings, an instrumental option which can reproducibly reduce the sensitivity of the mass spectrometer for every individual isotope monitored,¹⁸ was used in order to compensate for the marked difference in concentration levels found for the different analytes in the diverse samples analysed. OmniRange values were briefly optimized at the beginning of every working session in order to keep the maximum peak signal intensities for all of the elements in the range of 25000–200000 counts/s.

For the analysis of the digested samples, a Finnigan Element XR Sector Field-ICPMS instrument from Thermo Scientific (Bremen, Germany) was used. Sample introduction was accomplished by means of a micro-concentric nebulizer (200 µL min⁻¹) mounted onto a cyclonic spray chamber.

2.3. Procedure for the analysis of TiO₂ samples by means of SS-ETV-ICPMS

All samples were dried following the NIST recommendations (2 h in an oven at 110 °C) prior to analysis. After that, approximately 1 mg of sample was introduced into the graphite cup and directly weighed into it. The cup containing the sample was inserted into the furnace for subsequent analysis using an insertion tool available from Perkin Elmer. Finally, 10 µL of 5 mg L⁻¹Pd solution and 10 µL of 5 mg L⁻¹Te solution (50 ng of each element) were added into the graphite cup using the autosampler. The operating conditions are summarized in Table 1.

Table 1 Instrument operating conditions and data acquisition parameters for the ETV-ICPMS measurements

ICP-mass spectrometer PerkinElmer Sciex Elan 5000
a Monitored for diagnostic purposes. b ^105Pd^+ signal was also used as internal standard.
RF power	1200 W
Plasma flow rate	12 L min^−1
Auxiliary flow rate	1.2 L min^−1
Carrier flow rate	0.9 L min^−1
Sampling cone	Ni, 1.0 mm Ø
Skimmer	Ni, 0.75 mm Ø
Lens voltage	Optimized using pneumatic nebulization. No further optimization required when switching to ETV
Scanning mode	Peak hop transient
Dwell time per acquisition point	20 ms
Mass-to-charges (nuclides) monitored	^66Zn, ^75As, ^111Cd, ^121Sb, ^202Hg, ^208Pb, ^105Pd, ^125Te, ^49Ti^a, ^64Zn^a, ^80Ar2^+^a

---

HGA-600MS electrothermal vaporizer
Sample mass	1–3 mg
Chemical modifier added	50 ng Pd^b
Internal standard added	50 ng Te
Tube and cup lifetime	∼200 solid samples
Temperature program	Temperature/°C	Ramp/s	Hold time/s
Drying step	120	20	45
Vaporization step	1700	0	10
Cleaning step	2700	1	5

---

Five replicate measurements (every replicate representing the measurement of 1 solid sample aliquot) were carried out for each determination, and the median of the five was taken as the representative value in order to minimize the possible influence of outliers.¹⁹ Calibration was carried out against aqueous standard solutions, using the ¹⁰⁵Pd⁺ signal as internal standard for the determination of As and Sb, and the ¹²⁵Te⁺ signal as internal standard for the determination of Cd, Hg, Pb and Zn. Every working session (representing about 90 min of work) consisted of the construction of the calibration curve (based on the triplicate measurement of a blank and 3 standards, spanning a range of 2 orders of magnitude, in which the expected analyte concentration is situated) and the replicate measurements of all of the solid samples. 4 different determinations, representing the measurement of 20 solid sample aliquots, were carried out per session. 8 determinations were finally carried out per sample. Integrated peak area (counts) was selected as the measurement mode in all situations.

The NIST 154c sample was also analyzed by the standard addition method. For every determination, 3 replicates of a blank (10 µL of a 0.14 M HNO₃ solution), 5 replicates of the solid sample and 5 replicates of sample material to which 10 µL of an appropriate multi-element standard solution was added were measured. The same operating conditions as used for the external calibration methodology were used here, except for the addition of the chemical modifiers and internal standard. Also in this case, the median of five results (both in the spiked and non spiked solids) was taken as the representative value. 8 determinations in total were finally carried out.

2.4. Procedure for dissolution and analysis of the samples for comparison purposes

Analysis of the samples under investigation by means of solution-based sector field (SF)-ICPMS after sample digestion was also carried out in order to obtain reference values. This procedure was also applied to the NIST reference material, as only information values were available for the different analytes under consideration. The digestion method used was adapted from a method developed by Korn et al. for TiO₂ analysis by means of ICP-AES.⁶ Three samples of each material were weighed (∼0.1 g) and placed into 50 mL beakers. Subsequently, 1.5 g of (NH₄)₂SO₄ and 4 mL of 18 M H₂SO₄ were added to each of the beakers and they were heated smoothly on a hot plate (∼250 °C) for approximately 30 min, until complete dissolution of the sample. After cooling down for 15 minutes, the digests were transferred into 25 mL volumetric flasks and made up to volume with ultra-pure water. For analysis, the sample digests were diluted 10-fold and Co, In and Tl were added as internal standards. Single standard addition was used for calibration.

3. Results and discussion

3.1. Selection of nuclides to be monitored and optimization of temperature program

The selection of the nuclides most suited for monitoring and the optimization of the temperature program used for the vaporization of the analytes are very important issues that are closely interrelated. In fact, appropriate selection of the nuclides monitored and temperature programming can help in avoiding spectral overlap of analyte signals with those from ionic species originating from the matrix and/or the chemical modifiers added. These aspects were investigated first.

All the experiments carried out with solid samples for development and optimization of the SS-ETV-ICPMS method were performed with NIST 154c, for which reference values were already available for most of the analytes under consideration and better homogeneity could be expected. Because of their low abundance in this sample, Cd and Hg could not be determined by means of SS-ETV-ICPMS and additional amounts of these elements were added to the NIST material in order to use it for the entire optimization procedure. For this purpose, a spiking procedure adapted from that proposed by Tibi and Heumann was used.²⁰ A small amount of the sample (∼2 g) was taken, dried following the NIST recommendations indicated in section 2.3, weighed in a Teflon beaker and submersed in 2 mL of a solution containing Hg and Cd amounts finally resulting in a concentration in the solid sample aliquot of approximately 0.5 µg g⁻¹ for both analytes. The Teflon beaker was maintained at 60 °C in an oven till total dryness of the sample, which was finally homogenized by mixing with a Teflon pestle. This doped sample was then used for the optimization experiments.

A list of the different isotopes available for each analyte and potential interferences are gathered in Table 2. For most of the target elements, nuclide selection was not really problematic, and only the cases of As, Sb and Zn required special atention.

Table 2 Available isotopes for the target elements and list of the most relevant potential interferences considering the matrix composition and internal standards added. Isotopes finally selected for the determinations are indicated in bold

Isotopes (abundance)	Potential interferences
^75 As (100%)	^40Ar^35Cl^+
^106Cd (1.25%)	^106Pd^+, ^40Ar^66Zn^+
^108Cd (0.89%)	^108Pd^+, ^40Ar^68Zn^+
^110Cd (12.49%)	^110Pd^+, ^40Ar^70Zn^+
^111 Cd (12.80%)
^112Cd (24.13%)	^112Sn^+
^113Cd (12.22%)	^113In^+
^114Cd (28.73%)	^114Sn^+, ^102Pd^12C^+
^116Cd (7.49%)	^116Sn^+, ^104Pd^12C^+
^196Hg (0.14%)	^196Pt^+
^198Hg (10.02%)	^198Pt^+
^199Hg (16.84%)
^200Hg (23.13%)
^201Hg (13.22%)
^202 Hg (29.80%)
^204Hg (6.85%)	^204Pb^+
^204Pb (1.50%)	^204Hg^+
^206Pb (23.60%)
^207Pb (22.60%)
^208 Pb (52.30%)
^121 Sb (57.30%)	^105Pd^16O^+, ^108Pd^13C^+
^123Sb (42.70%)	^123Te^+
^64Zn (48.60%)	^64Ni^+, ^48Ti^16O^+, ^128Te^++, ^32S^32S^+, ^32S^16O^16O^+
^66 Zn (27.60%)	^50Ti^16O^+, ^40Ar^26Mg^+, ^32S^34S^+, ^34S^16O^16O^+
^67Zn (4.10%)	^49Ti^18O^+, ^33S^34S^+, ^32S^16O^18O^1H^+,
^68Zn (18.7%)	^50Ti^18O^+, ^40Ar^28Si^+, ^34S^34S^+, ^32S^36S ^+,^36Ar^32S^+
^70Zn (0.60%)	^70Ge^+, ^40Ar^30Si^+, ^34S^18O^18O^+, ^38Ar^32S^+

---

For As, the signal from ⁷⁵As⁺, the only isotope available, could be affected by the occurrence of the polyatomic ion ⁴⁰Ar³⁵Cl⁺. Although chlorine levels are not expected to be very high in titanium dioxide samples (11 mg/kg in the NIST 154c material), the potential for error was investigated in several experiments. Solutions containing NaCl in high concentration (10 µg of NaCl were vaporized during each firing) were analyzed using the temperature program finally selected for the analysis of the TiO₂ samples. A Background Equivalent Concentration (BEC) of 0.73 ng g⁻¹ was obtained under these worst-case-scenario conditions, indicating that the influence of ⁴⁰Ar³⁵Cl⁺ on the signal at a mass-to-charge ratio (m/z) of 75 is negligible with the temperature program used. This is in agreement with some results previously reported by other scientists.^21–23

As shown in Table 2, both isotopes of Sb (¹²¹Sb⁺ and ¹²³Sb⁺) are affected by spectral overlap from Pd- and Te-based interferences. These interferences could be relevant considering the addition of Pd and Te as chemical modifier and internal standard (see below and section 3.2 for further details). Finally, the most abundant isotope, ¹²¹Sb⁺, was selected, as the influence of ¹⁰⁵Pd¹⁶O⁺ and ¹⁰⁸Pd¹³C⁺ should be much less severe than the isobaric overlap of ¹²³Te⁺ and ¹²³Sb⁺. Further experiments showed that the blanks for ¹²¹Sb⁺ did not vary significantly with the amount of Pd used as modifier (50 ng)—a BEC of 0.53 ng g⁻¹ was obtained under these conditions—possibly owing to the low oxide formation taking place in a “dry” plasma.

The signals of all the major Zn isotopes (⁶⁴Zn⁺, ⁶⁶Zn⁺ and ⁶⁸Zn⁺) overlap with those of TiO⁺ ions (see Table 2), which is an important issue due to the nature of the sample (99% TiO₂). However, by using ETV for sample introduction, it is possible to render the signal of Zn ions practically free from spectral overlap by means of temperature programming, based on the different volatilities of Zn and Ti. This strategy, which has proved to be an excellent tool in ETV-ICPMS,^24,25 implies a careful optimization of the temperature program in order to find suitable vaporization conditions. This must be not only achieved for Zn, but for all of the analytes, while avoiding significant co-vaporization of the matrix.

Considering the high thermal stability of TiO₂ (TiO₂ m.p. 1840 °C, b.p. 2500–3000 °C²⁶) and the rather high volatility of the analytes of interest, the best way to tackle the analysis seemed to be the direct vaporization of the analytes at a relatively low temperature at which TiO₂ is still not vaporized without the need for any previous pyrolysis step. A cleaning step at high temperature could be used afterwards in order to remove the sample matrix from the graphite tube and prepare the latter for the next measurement.

Different temperatures were then tested for the vaporization of the analytes from the solid sample. Although all of the analytes start vaporizing at relatively low temperatures (1400 °C for Sb and As, 1200 °C for the rest), higher values (above 1600–1700 °C) are needed for obtaining well-defined peaks for all of them. In fact, it was established that the higher the temperature was, the better the signal shape. As a result, a careful selection of a compromise vaporization temperature leading to well-defined signals for all of the target elements, while still allowing the thermal separation of the Zn signal from that coming from the TiO⁺ interferences was crucial at this point.

For this purpose, the signal at m/z 49 (⁴⁹Ti⁺) was monitored together with those at m/z 64 and m/z 66 (⁶⁴Zn⁺ and ⁶⁶Zn⁺) during the vaporization of the solid sample at different temperatures starting from the minimum value for good signal definition, i.e. 1700 °C. It was assumed that substantial changes in the background value at m/z 49 (for which a high OmniRange setting had been selected in order to avoid excessively intense ion beams from reaching the detector) would represent a significant rate of matrix co-vaporization and thus, an influence of TiO⁺ species on the signals of the Zn isotopes could be expected. This effect is illustrated in Fig. 1, where the result obtained upon vaporization of an aliquot of NIST 154c, while ramping the termperature from 1700 °C to 2000 °C is presented. As seen in Fig. 1, no changes are observed for the signal at m/z 49 until a temperature of about 1900 °C is reached (at a time of around 4.5s). At this point, the sample matrix starts vaporizing, as indicated by the increasing signal intensity at m/z 49. Signals recorded at both m/z 64 and 66 increase dramatically after that point (especially in the case of m/z 64), most probably due to the contribution from the interfering TiO⁺ species at these m/z values. The integration of the peaks at m/z 64 and 66 from t = 0 to t = 4.5s (before the appearance of the peak at m/z 49) correspond to the Zn signals free from the TiO⁺ interferences. This assumption is further supported by two facts, (i) the ratio for those signals after mass discrimination correction corresponds well with the expected Zn isotope ratio (1.74 ± 0.06) and (ii) the recovery calculated for the integration of those initial peaks (once the working methodology was finally optimized) was close to 100% (98.3%). After some work with the NIST sample at different vaporization temperatures, and taking into account that variations in the condition of the graphite cup (upon subsequent firing) may affect the matrix vaporization, 1700 °C was finally selected as the safest compromise value still providing well-defined profiles for all of the elements, while assuring a minimal contribution of TiO⁺ even after a high number of firings.

Fig. 1 Thermal separation of the signals from Zn⁺ and TiO⁺. Signal obtained for the direct vaporization of NIST 154c TiO₂.

After this optimization of the temperature program, and owing to the “thermal resolution” offered by the ETV introduction system,²⁷ the TiO⁺ interferences should be finally overcome, thus allowing the selection of any of the Zn isotopes for the determinations. In this regard, selection of the most abundant isotope, ⁶⁴Zn⁺, for quantitative purposes was avoided due to the spectral overlap of this ion with the ⁶⁴Ni⁺ isobar (0.9%) and ¹²⁸Te²⁺ (the influence of which could be important considering the potential presence of Ni in the samples and the use of Te as internal standard) and the monitoring of ⁶⁶Zn⁺ was preferred for the determinations instead. However, in practice, ⁶⁴Zn⁺ was also monitored in all the measurements together with ⁶⁶Zn⁺ for diagnostic purposes. This strategy, (together with monitoring of the ⁴⁹Ti⁺ signal), was used as a means to detect (and eventually reject) the measurements affected by TiO⁺ formation, and proved to be especially helpful in the cases in which the high OmniRange setting for ⁴⁹Ti⁺ was actually hiding partial matrix co-vaporization. As can be seen from Fig. 1, the impact of the TiO⁺ interference on the signal at a mass-to-charge ratio of 64 is much higher than that observed at a mass-to-charge ratio of 66, as the abundance of the ion interfering at m/z 64 (⁴⁸Ti¹⁶O⁺) is much higher than that of the ion at m/z 66 (⁵⁰Ti¹⁸O⁺). As a consequence, a dramatic increase of the ⁶⁴Zn⁺/⁶⁶Zn⁺ ratio, by factors much higher than those expected taking into account the typical uncertainty values observed for isotope ratio monitoring with SS-ETV-ICPMS (typically below 3%²⁸), is seen when matrix co-vaporization occurs. This fact allowed to detect and discard some biased measurements (in practice, those values differing by more than 3% from the median of all the values measured during the session were discarded). The occurrence of this kind of outliers was observed to be very low in the beginning of the graphite cup and tube lifetime, typically below 5%. However, as a higher number of measurements were carried out (with the consequent changes in the condition of the graphite parts), the occurrence of those outliers was also observed to increase. Typically, up to 200 solid samples could be analyzed with the same graphite cup and tube with an outlier incidence lower than 5–10%. When frequencies higher than those values are observed for subsequent firings, replacing these graphite consumables is advised.

For the cleaning step, a temperature of 2700 °C was maintained during 5 s. With this approach, a thin “silvery layer”, previously identified by some other workers as Ti carbides,^8,12 was observed to appear in the graphite cup after some firings. This fact, which could lead to Ti accumulation in the system, has been previously associated with the appearance of memory effects in TiO₂ analysis by means of slurry sampling-ETV-ICPMS.¹³ In that work, the use of a cleaning step, consisting of the vaporization of a fluorinating agent (PTFE) at high temperature in order to promote Ti removal and reduce memory effects, was considered necessary. Thus, different experiments aiming at assessing the influence of those memory effects for the SS-ETV-ICPMS method under development were carried out in the context of the present work. The blank values obtained for the different analytes after TiO₂ vaporization with thermal cleaning at 2700 °C were compared with those obtained after using NH₄HF₂ as a fluorinating agent assisting the cleaning procedure. In the latter case, 10 µL of NH₄HF₂ 1% (w/w) were introduced in the graphite tube and, after drying at 120 °C, a temperature of 2700 °C was maintained during 5 s. The blank signals obtained after both cleaning strategies were very similar for all of the target elements, thus showing that memory effects are not an important issue in this particular case. As a consequence, the less cumbersome and time-consuming thermal cleaning without the use of any fluorinating agent was selected. A summary of the temperature program finally selected can be found in Table 1.

3.2. Calibration approaches

Calibration is still one of the weaker points for most solid sampling techniques, and the case of SS-ETV-ICPMS is not an exception. Among the different calibration approaches that have been succesfully used with this technique,²⁵ those based on the use of aqueous standards are normally preferred when the main goal is the development of a fast methodology for routine analysis, and this strategy was also tried here.

However, in spite of the careful temperature programming carried out in order to achieve a proper separation of analyte and matrix vaporization, the appearance of matrix effects could not be avoided in this particular case. As explained below in detail, different signal profiles and different peak areas were obtained for all of the analytes when vaporization from an aqueous standard and from a solid sample are compared. Therefore, additional measures had to be taken in order to efficiently correct for these differences and obtain accurate results.

For the vaporization of the aqueous standards on the one hand, the use of the relatively low vaporization temperature at 1700 °C resulted in poor analyte transport efficiencies for the standards with the lowest analyte concentration (corresponding to analyte amounts in the range 0.5–1 ng), especially for the more volatile elements. As a result, non linear-calibration curves were obtained for the analytes with that temperature program. This negative effect is well documented in the literature and, very often, the addition of a chemical modifier acting as a carrier agent has proven a suitable solution.^29,30 The search for a suitable carrier agent was then considered, and palladium was tested as a first candidate.

The first experiment carried out, consisted of the construction of different calibration curves (blank + 3 standards) for the different analytes with the addition of different amounts of Pd (as nitrate). The analytes can be divided into two different groups according to the behaviour observed. For the more volatile elements (Zn, Cd, Hg and Pb), increasing amounts of Pd resulted in better linearity for the calibration curves, as well as slightly improved sensitivity, as can be seen in Fig. 2A for Pb. For As and Sb on the other hand, the improvement in linearity observed for increasing amounts of Pd added is accompanied by a reduction in sensitivity after a certain point (50 ng Pd), as shown in Fig. 2B for As. This effect is probably due to thermal over-stabilization of these analytes in the graphite tube,¹⁶ as no suppression effects were observed for the argon dimer signal for the Pd quantities tested. In fact, the signals for As and Sb observed when using Pd amounts higher than 50 ng were much broader and, in some cases, not well-defined at all (multiple peaks and long tailings were observed in these cases). The use of higher vaporization temperatures would be required in order to improve the situation and obtain well-defined, unimodal peaks for these elements. However, as already discussed in section 3.1, this temperature increase would lead to some degree of matrix co-vaporization during analysis of the solid samples, an effect that obviously needs to be avoided. As a consequence, the addition of 50 ng Pd was finally selected as a compromise approach, in which the linearity problem is solved, and well-defined signals are still obtained for all of the analytes when using a vaporization temperature of 1700 °C. The typical signals obtained upon vaporization of aqueous standards under these conditions are shown in Fig. 3A.

Fig. 2 Influence of the amount of Pd added on the linearity and sensitivity of the calibration curves obtained for (A) Pb and (B) As, under the conditions used in this work (see Table 1).

Fig. 3 Comparison of the signal profiles obtained for: (A) an aqueous standard solution containing 0.5 ng As (peak area = 51200 counts), 0.5 ng Cd (peak area = 35300 counts), 0.5 ng Hg (peak area = 69840 counts), 1.0 ng Pb (peak area = 60770 counts), 1.5 ng Sb (peak area = 116250 counts) and 1.0 ng Zn (peak area = 114724 counts for ⁶⁶Zn⁺); and (B) 0.653 mg of the solid sample NIST 154c containing approximately 0.48 ng As (peak area = 127850 counts), 0.33 ng Cd (peak area = 29840 counts), 0.33 ng Hg (peak area = 59500 counts), 0.87 ng Pb (peak area = 69144 counts), 1.29 ng Sb (peak area = 249000 counts) and 0.98 ng Zn (peak area = 140760 counts for ⁶⁶Zn⁺).

In contrast to the behaviour observed for the aqueous standards, the addition of Pd did not result in any remarkable effect on the vaporization of the different analytes from the solid samples. This would indicate a strong interaction between the analytes and the sample matrix (probably they are adsorbed or included in the crystal structure of TiO₂) that is only broken at the moment in which the former are vaporized, and at that point, interaction with palladium does not seem to take place anymore. The typical signals obtained upon vaporization of the target analytes from a solid sample are shown in Fig. 3B.

Two different effects, contrasting with the situation found for aqueous standards, can be seen in Fig. 3B. First, the analytes appear clearly divided into two different groups according to their volatilities, with As and Sb appearing slightly delayed and closer to the Pd signal than the rest of the analytes. This situation is different from that observed for aqueous standards, since in that case the interaction with the modifier minimized the differences in volatility for the different analytes and all signals appeared at practically the same time. In addition to that, for the solid samples, a clear enhancement in signal intensity for all of the analytes was observed, as is clearly shown in Fig. 4. The same sensitivity cannot be obtained when vaporizing aqueous solutions, not even when Pd is added as a carrier.

Fig. 4 Comparison of the sensitivity obtained for the analytes and internal standards upon the vaporization of the aqueous standard and the solid sample presented in Fig. 3. The values obtained for Pd and Te were multiplied by a factor of 10 for representation in the figure, in order to improve readability of the graph. The enhancement factor (EF), defined as the ratio between the sensitivities (s_{solid sample}/s_{aqueousstandard}) for every element for which reference values for the solid sample were used in the calculation, is also displayed in the graph. The error bars are displayed as ± SD of the measurements.

This signal enhancement, which affects more markedly the elements in the second group, could be explained by the early vaporization of a small percentage of the sample matrix as TiC at moderate temperatures (1700 °C), as reported by some authors for TiO₂ analysis by means of GFAAS.⁸ The formation and vaporization of those carbides, which would improve the transport efficiency for the analytes vaporized from solid samples with respect to those vaporized from aqueous standards, should certainly follow an increasing trend along the vaporization step carried out at 1700 °C (i.e. the formation rate of TiC in the graphite tube should increase gradually as the furnace temperature stabilizes and is maintained for a period of time). Hence, this fact could adequately explain the increased matrix effect observed for As, Sb and Pd. On the other hand, the impact of this partial TiC vaporization on the formation of TiO⁺-based interferences should be minimal, considering the characteristics of a “dry” plasma and the high thermal stability of these carbides.

Taking all of this into account, it was not surprising that the results obtained for the determination of the different target elements by simply using the signal of the aqueous standards for calibration were overestimated (especially in the case of As and Sb). In the traditional way of analysis after sample digestion, the more obvious strategy to solve this problem and correct for the matrix effects would be the use of the single standard addition methodology (the use of isotope dilution was prevented by the monoisotopical nature of As). Applying this calibration methodology for solid sampling analysis, however, is not so straightforward, and the possibility of using a simpler method was considered here.

In this regard, the use of external calibration together with a suitable internal standard as a means to compensate for the difference in sensitivity caused by matrix effects had proved sucsessful in some previous works.^31,32 The advantages of using this calibration strategy over the standard addition method are clear in terms of sample throughput and ease of use, and this is why the former was preferred. In the case of standard addition, previous knowledge about the analyte concentration in the samples to be analyzed is crucial in order to prepare a suitable spike for each sample. Although some information in this direction is also needed in the case of external calibration with internal standardization (something which is true for any calibration methodology in any analytical technique), the requirements are much less stringent. By using the external calibration approach, a curve covering two orders of magnitude can be made in the beginning of the working session and can be afterwards applied to different samples with different concentrations in the same session. In the case of the standard addition method, every solid sample needs to be measured twice, (i.e. with and without the addition of the spike, representing 5 solid sample aliquots for each situation) which deteriorates considerably the sample throughput when more than one sample has to be analyzed. In any case, the standard addition methodology was also applied to the analysis of the NIST 154c sample for comparison reasons. Results are gathered in Table 3.

Table 3 Analysis of the TiO₂ samples by means of SS-ETV-ICPMS: quantitative results and limits of detection (3s definition). 8 determinations were performed for each sample (every one consisting of the measurement of 5 replicate solid samples, representing about 15 min of work). In the case of the reference values, three replicate analyses were carried out, as described in section 2.4. In all cases, uncertainties are provided as 95% confidence intervals

	As	Cd	Hg	Pb	Sb	Zn
a Theoretical value expected for Cd and Hg, after spiking the sample as described in section 3.1.
LODs ETV-ICPMS/ng g^−1	1.0	0.5	20	6.5	1.0	5.5
NIST 154 c/µg g^−1
SS-ETV-ICPMS—External calibration	0.68 ± 0.06	0.53 ± 0.05	0.48 ± 0.06	1.49 ± 0.19	1.89 ± 0.15	1.54 ± 0.15
SS-ETV-ICPMS—Standard addition	0.80 ± 0.07	0.55 ± 0.05	0.45 ± 0.05	1.39 ± 0.13	2.13 ± 0.16	1.47 ± 0.14
PN-SF-ICPMS	0.73 ± 0.08	0.50 ^a	0.50 ^a	1.33 ± 0.13	1.98 ± 0.17	< LOD
NIST informative values	0.7	—	—	1.1	3	1.5
Blanc de Titane Sennelier/µg g^−1
SS-ETV-ICPMS	1.84 ± 0.13	0.085 ± 0.007	0.42 ± 0.05	10.0 ± 0.7	0.56 ± 0.06	2.84 ± 0.19
PN-SF-ICPMS	2.06 ± 0.16	0.084 ± 0.006	0.19 ± 0.11	10.9 ± 0.8	0.52 ± 0.05	< LOD
TiO2 UCB/µg g^−1
SS-ETV-ICPMS	2.62 ± 0.22	0.69 ± 0.06	1.08 ± 0.11	49 ± 4	2.47 ± 0.21	85 ± 7
PN-SF-ICPMS	2.56 ± 0.23	0.79 ± 0.06	1.18 ± 0.15	45 ± 4	3.04 ± 0.27	88 ± 7

---

The next step in the method development was the search for a suitable internal standard. However, this is not always an easy task when working with SS-ETV-ICPMS. Most importantly, an appropriate internal standard should show an analogous thermochemical behaviour in the graphite furnace as that of the analytes, while being present at negligible levels in the samples under consideration.³³ The requirement of analogous thermochemical properties seems especially important in this case where, as discussed before, matrix effects affect the analytes differently according to their corresponding volatilities. Consequently, at least two different internal standards should be used, one for each of the groups of analytes identified, on the basis of the temporal separation observed during their direct vaporization from the solid sample.

At this point, the use of the ¹⁰⁵Pd⁺ signal as internal standard for the group consisting of As and Sb was considered, taking advantage of the similar furnace behaviour shown by these three elements. This strategy, already tested in previous works of our group,^34,35 seemed the most straightforward, as Pd had already been added as carrier agent and is not expected to be present at significant levels in this kind of samples. Experiments proved that the use of ¹⁰⁵Pd⁺ as internal standard can compensate for the matrix effects observed for As and Sb, ensuring results that do not deviate more than 15% from the expected value for every solid sample replicate.

However, as illustrated in Fig. 4, the use of the same internal standard for the rest of the analytes would lead to clearly underestimated results, as the matrix effects are not affecting them to the same extent. The search for an internal standard better suited for these elements was then considered. In this regard, the addition of Te seemed quite promising, as this element shows a similar volatility to that of the analytes and again its presence in TiO₂ samples is expected to be low. Moreover, Te is an element of medium mass and, thus, it seemed a good compromise situation for the analytes considered (covering a range from m/z66-Zn to m/z208-Pb). 10 µL of an aqueous solution containing 5 mg L⁻¹ of this element were introduced in the graphite furnace, just after addition of Pd, onto the solid sample. As seen in Fig. 3B, Te is co-vaporized with Cd, Hg, Zn and Pb and is affected by matrix-effects in the same way as these elements (see Fig. 4). The use of the ¹²⁵Te⁺ signal as internal standard for the determination of the elements in the first group improved the results obtained both in terms of accuracy and repeatability and, therefore, this approach was used for the final analysis of the solid samples.

3.3. Analysis of the samples

The samples investigated were finally analyzed by means of SS-ETV-ICPMS, as described in section 2.3. The results obtained for these samples are shown in Table 3.

The samples were also analyzed according to the procedure detailed in Section 2.4, using pneumatic nebulization-SF-ICPMS after sample dissolution in order to obtain reliable reference values for method validation. Some problems were observed for the acquisition of reliable reference values for some of the analytes, especially for Zn and Hg.

In the case of Zn, problems arose when dealing with the less concentrated samples, i.e. those from NIST and Sennelier. The occurrence of TiO⁺ interferences and of S-based ions coming from the sulfuric acid used for the sample digestion, together with the fact that Zn is a rather contamination-prone element, resulted in high LODs for this element for the solution-based ICPMS method. As a result, the content of Zn in these two samples could not be determined with this method. An alternative microwave digestion method avoiding the use of sulfuric acid and based on the use of mixtures of concentrated hydrofluoric and nitric acids was also tested,³⁶ but the blank values obtained with that approach were also inappropriate for the analysis of these two samples. Owing to its much higher value however, the Zn content in the UCB sample could be properly determined with the sulfuric acid-based method, and the reference value thus obtained is shown in Table 3, as well as the indicative value provided by NIST for SRM 154c.

In the case of Hg, the RSD obtained for the reference values was considerably higher than that for the other elements, especially in the pigment sample Blanc de Titane, with a value close to 25%. Moreover, in this specific case, the reference value obtained was considerably lower than the SS-ETV-ICPMS result, so that both results are significantly different at the 95% confidence level. The reason behind the bias between these results could be related with the digestion procedure used, which could lead to Hg losses. In fact, Hg is a complex analyte and its determination in solid samples by means of digestion approaches has been reported to be usually hampered by possible losses due to the high volatility of some Hg species, especially for samples that are difficult to dissolve.^37,38 According to these papers, Hg losses can already be expected at relatively low temperatures used for sample digestion, which would explain the underestimated results and the high RSD values obtained for this sample.

Except for this discrepancy, a closer look at the results gathered in Table 3 shows that there is a good agreement (differences within experimental errors) between the solid sampling values and the reference values obtained by means of PN-SF-ICPMS for all three samples. This proves that, under the conditions used, the ETV-ICPMS method provides accurate results when calibrating against aqueous standard solutions, even when the nature of the samples and their concentration levels are very different. Applicability of the method from the limit of quantification up to several hundreds of ppm can be granted for all the elements. This limit can be achieved by reducing the sample mass used for each sample (to a minimum limit of 0.1 mg) and by using the OmniRange settings for adjusting the sensitivity of the instrument.¹⁸

Concerning the reproducibility of the method, the RSD values obtained range between 9 and 13% for all of the analytes. These values, which are in the typical range expected for this technique, can be considered as acceptable for the fast and routine control of TiO₂ samples.

Detection limits in the ng g⁻¹ range were also calculated for all of the target elements with the SS-ETV-ICPMS method (3s definition); these LODs, also gathered in Table 3, are in the same range of other solid sampling methods reported in the literature^12,13 and at least one order of magnitude better than those attainable with digestion-based methodologies.^3,6 Taking into account the typical legal limits established for the different analytes by the different international regulations cited in the introduction it can be concluded that the LODs achieved with this method are really fit-for-purpose.

4. Conclusions

It has been demonstrated that solid sampling-electrothermal vaporization-inductively coupled plasma mass spectrometry with the use of aqueous standards for calibration is suitable for the direct multi-element analysis of TiO₂ samples with very different concentration levels for the target analytes (As, Cd, Hg, Pb, Sb and Zn). Although the technique requires careful optimization of the operating conditions (target nuclide selection, temperature program, use of a carrier agent and use of two internal standards to compensate for matrix effects) in order to obtain the best possible performance, once this optimization has been carried out in the proper way, the technique exhibits very attractive features, the most noteworthy being its excellent limits of detection and its ability to obtain reliable results at trace and ultratrace levels while maintaining the maximum level of straightforwardness for the whole procedure. It can be concluded then that SS-ETV-ICPMS can be a valuable tool for the analysis of TiO₂ samples in the context of routine control analysis.

Acknowledgements

This study was financially supported by the Fund for Scientific Research (FWO-Vlaanderen), the Spanish Ministry of Education and Science (Project CTQ2006-03649/BQU) and the Government of Aragón (DGA research projects PM013/2007). Maite Aramendía acknowledges Ghent University and FWO-Vlaanderen for her postdoctoral grants.

References

1. D. Wildhagen and V. Krivan, Fresenius J. Anal. Chem., 1997, 359, 230–238.
2. K. S. Park, N. B. Kim, K. Y. Lee, Y. Y. Yoon, S. K. Chun and J. H. Lee, J. Radioanal. Nucl. Chem., 1995, 192, 155–162.
3. D. Wildhagen, V. Krivan, B. Gercken and J. Pavel, J. Anal. At. Spectrom., 1996, 11, 371–377.
4. P. Tianyou, D. Pingwu, H. Bin and J. Zucheng, Anal. Chim. Acta, 2000, 421, 75–81.
5. Z. Wang, Z. Ni, D. Qiu, T. Chen, G. Tao and P. Yang, J. Anal. At. Spectrom., 2004, 19, 273–276.
6. M. d. G. A. Korn, A. C. Ferreira, A. C. S. Costa, J. A. Nóbrega and C. R. Silva, Microchem. Journal, 2002, 71, 41–48.
7. H. M. Dong and V. Krivan, Spectrochim. Acta Part B, 2001, 56, 1645–1656.
8. H. M. Dong, V. Krivan, B. Welz and G. Schlemmer, Spectrochim. Acta Part B, 1997, 52, 1747–1762.
9. European parliament and council directive 94/36/EC, O. J. Eur. Commun., 1994, p. L237/13.
10. http://www.cefic.org/sector/tdma/uses%26%3Bproperties.htm , April 21st 2008 .
11. L. Zhang, D. Ishi, K. Shitou, Y. Morita and A. Isozaki, Talanta, 2005, 68, 336–342.
12. B. Docekal and B. Vojtková, Spectrochim. Acta Part B, 2007, 62, 304–308.
13. G. Xiang, B. Hu, Z. Jiang and C. Gong, J. Mass Spectrom., 2006, 41, 1378–1385.
14. M. Hiraide, K. Namba and H. Kawaguchi, Anal. Sci., 1990, 6, 609–610.
15. K. Beckmann and G. Wunsch, Fresenius J. Anal. Chem., 1992, 342, 469–472.
16. M. Resano, M. Aramendía, W. Devos and F. Vanhaecke, J. Anal. At. Spectrom., 2006, 21, 891–898.
17. M. Resano, M. Verstraete, F. Vanhaecke, L. Moens, A. Van Alphen and E. R. Denoyer, J. Anal. At. Spectrom., 2000, 15, 389–395.
18. S. Boonen, F. Vanhaecke, L. Moens and R. Dams, Spectrochim. Acta Part B, 1996, 51, 271–278.
19. M. A. Belarra, M. Resano and J. R. Castillo, J. Anal. At. Spectrom., 1999, 14, 547–552.
20. M. Tibi and K. G. Heumann, J. Anal. At. Spectrom., 2003, 18, 1076–1081.
21. F. Vanhaecke, S. Boonen, L. Moens and R. Dams, J. Anal. At. Spectrom., 1995, 10, 81–87.
22. D. Pozebon, V. L. Dressler and A. J. Curtius, At. Spectrom., 1998, 13, 80–88.
23. J. M. Carey, E. H. Evans, J. A. Caruso and W. L. Shen, Spectrochim. Acta Part B, 1991, 46, 1711–1721.
24. J. S. Alvarado and M. D. Erickson, J. Anal. At. Spectrom., 1990, 11, 923–928.
25. M. Resano, F. Vanhaecke and M. T. C. de Loos-Vollebregt, J. Anal. At. Spectrom., 2008, 23, 1450–1475.
26. CRC Handbook of Chemistry and Physics 71^st Edition, 1990. Ed. David R. Lide. CRC Press.
27. P. P. Mahoney, S. J. Ray, G. Li and G. M. Hieftje, Anal. Chem., 1999, 71, 1378–1383.
28. M. Resano, M. Aramendía and F. Vanhaecke, J. Anal. At. Spectrom., 2006, 21, 1036–1044.
29. D. C. Grégoire and R. E. Sturgeon, Spectrochim. Acta Part B, 1999, 54, 773–786.
30. T. Kántor, Spectrochim. Acta Part B, 2001, 56, 1523–1563.
31. M. Resano, E. Garcia-Ruiz, L. Moens and F. Vanhaecke, J. Anal. At. Spectrom., 2005, 20, 81–87.
32. M. Resano, E. Garcia-Ruiz, F. Vanhaecke, C. Crespo and M. A. Belarra, J. Anal. At. Spectrom., 2004, 19, 958–965.
33. F. Vanhaecke, S. Boonen, L. Moens and R. Dams, J. Anal. At. Spectrom., 1995, 10, 81–87.
34. M. Resano, E. Garcia-Ruiz, F. Vanhaecke, C. Crespo and M. A. Belarra, J. Anal. At. Spectrom., 2004, 19, 958–965.
35. M. Resano, E. Garcia-Ruiz, L. Moens and F. Vanhaecke, J. Anal. At. Spectrom., 2005, 20, 81–87.
36. J. D. Norris, B. Preston and L. M. Ross, Analyst, 1992, 117, 3–7.
37. H. Biester and G. Nehrke, Fresenius J. Anal. Chem., 1997, 358, 446–452.
38. A. Krata and E. Bulska, Spectrochim. Acta Part B, 2005, 60, 345–350.

---