: Results of an interlaboratory comparison for characterization of Pt nanoparticles using single-particle ICP-TOFMS

This study describes an interlaboratory comparison (ILC) among nine (9) laboratories to evaluate and validate the standard operation procedure (SOP) for single-particle (sp) ICP-TOFMS developed within the context of the Horizon 2020 project ACEnano. The ILC was based on the characterization of two different Pt nanoparticle (NP) suspensions in terms of particle mass, particle number concentration, and isotopic composition. The two Pt NP suspensions were measured using icpTOF instruments (TOFWERK AG, Switzerland). Two Pt NP samples were characterized and mass equivalent spherical sizes (MESSs) of 40.4 ± 7 nm and 58.8 ± 8 nm were obtained, respectively. MESSs showed <16% relative standard deviation (RSD) among all participating labs and <4% RSD after exclusion of the two outliers. A good agreement was achieved between the different participating laboratories regarding particle mass, but the particle number concentration results were more scattered, with <53% RSD among all laboratories, which is consistent with results from previous ILC studies conducted using ICP-MS instrumentation equipped with a sequential mass spectrometer. Additionally, the capabilities of sp-ICP-TOFMS to determine masses on a particle basis are discussed with respect to the potential for particle density determination. Finally, because quasi-simultaneous multi-isotope and multi-element determinations are a strength of ICP-TOFMS instrumentation, the precision and trueness of isotope ratio determinations were assessed. The average of 1000 measured particles yielded a precision of below ±1% for intensity ratios of the most abundant Pt isotopes, i.e.194Pt and 195Pt, while the accuracy of isotope ratios with the lower abundant isotopes was limited by counting statistics.


Scope and application
Following the rapid development of nanotechnology and their widespread use, engineered nanoparticles (ENPs) have become part of our daily lives. Through the use and disposal of ENP-containing products, the ENPs are released into the environment. Hence, in order to understand implications for human health and the environment, specific analytical methods, able to detect, quantify, and characterize these nanomaterials, are required. For nanotoxicology purposes, the detection of nanomaterials (NMs) at low concentrations is another key point. In this context, inductively coupled plasma-mass spectrometry (ICP-MS) used in singleparticle mode (sp-ICP-MS) has become an established method for the detection and characterization of inorganic nanoparticles. Currently, sp-ICP-MS can be applied in routine analysis for the measurement of one or at most two isotopes at a time, hence limiting the available information. The monitoring of only one isotope per NP is particularly disadvantageous as it is the combined information of several elements or isotopes which makes it possible to understand the real composition of a particle and thereby its origin.
Indeed, the determination of the "fingerprint" elements or isotopes in NPs is one of the most promising ways to distinguish natural (NNPs) and engineered nanoparticles (ENPs) and is therefore of great importance to the study of NPs in the environment [1]. In order to conduct multi-element single particle analysis, a clear solution is to employ simultaneous full-spectrum mass analyzers such as time-of-flight mass spectrometers (TOFMS), which enable the monitoring of multiple isotopes simultaneously further allowing the distinction of ENPs and NNPs.
The aim of this SOP is to provide a quantitative method for the determination of the elemental composition, the particle number concentration (PNC) and the mass (respectively size 1 ) of inorganic NPs in aqueous solutions using an ICP-TOFMS (icpTOF, TOFWERK AG, Thun) with a conventional liquid sample introduction system. Although the data presented here were obtained using silver shelled gold core nanoparticles, the scope of this quantitative analysis of NPs can be extended to any inorganic NPs that produce a detectable signal by the icpTOF. The analysis guidelines are valid for multi-element NPs, consisting of metal and metal oxides with sizes ranging roughly from 20 to 250 nanometer 2 in aqueous suspensions, depending on the isotope and its corresponding sensitivity. 1 Assuming shape is known and equivalent density to bulk material 2 For larger particles, it is advisable to check for nonlinear detection effects. Non-linear response can be caused by incomplete atomization and ionization of large particles in the ICP or too intense signal at the detector of the MS.

Method principle
This SOP describes the quantitative analysis of silver shelled-gold core nanoparticles by sp-ICP-TOFMS as an example. The quantification is based on the method developed by Pace et al. [2] and requires the use of a nanoparticle standard, here 50 nm monodisperse Au NPs for the determination of the transport efficiency.
It is to be highlighted that sp-ICP-TOFMS is subject to the same matrix effects as standard ICP-MS and so unmatched solvent between ionic standards and particles samples may lead to erroneous results [3]. The procedure described in this SOP applies to the characterization of inorganic NPs in diluted aqueous suspensions. If the sample matrix is different, the accuracy of the results is not guaranteed. In such cases, a sample pre-treatment is recommended to produce an aqueous suspension.

Safety procedures and precautions
Standard personal protective clothing including lab coat, safety glasses and gloves, is required. When handling suspensions containing nanoparticles as well as preparing the solution standards with diluted acid, appropriate caution should be used.

Procedure
In this SOP, the analysis delivers information regarding the mass of each isotope in each single particle and the PNC per vial. The quantification is based on the method reported by Pace et al. where the transport efficiency is determined based on the known size of the reference particles [2]. The following substances and information are required: • Particle samples with recommended PNC < 10 5 particles/mL.
(If this concentration is unknown, measure your sample with the minimum available integration time -1 ms for icpTOF S2, 1.8 ms for icpTOF and 3 ms for icpTOF 2R -and dilute the sample until the number of particle signals for each analyte is < 15/second.) • Calibration standards containing known concentrations of all analytes present in the particles. The signal of the particles in the sample should fall in the linear range of the calibration curve.
• Single element reference nanoparticles of a known size and density to determine the transport efficiency of the sample introduction system. Au NPs are recommended with a PNC < 10 5 particles/mL.
• Calibration standards containing the same element as the reference nanoparticles at known concentrations. The signal of the reference particles should fall within the linear range of the calibration curve.
• Liquid flow (i.e nebulizer uptake) must be determined externally prior to the particle analysis. The nebulizer uptake can be measured by weighing a blank solution before and after a defined aspiration time, e.g. 10 min. At least 1 g of solution should have been aspirated to reduce weighing errors and using the mean nebulizer uptake from 3 runs is recommended.

Additional information to the particle number concentration:
The ideal PNC is highly dependent on the sample uptake rate and transport efficiency. If a too high PNC is introduced into the plasma, the probability of double events will also increase proportionally, which will lead to an overestimation of the particle size and an underestimation of PNC. For example, with a sample flow rate of 300 µl/min, a transport efficiency of 3%, a PNC of 10 5 particles/mL, 15 particles are introduced per second into the plasma. Under such conditions, and with an integration time of 1 ms, the likelihood of an event being caused by multiple, concurrent NPs will be only 0.75% (ratio of statistical occurrence of double events to single events).

Additional information to the preparation of calibration standards:
Different matrices induce different matrix-effects in both the ICP and the sample uptake. Differences in the matrix composition (i.e. acid content, buffers, dissolved salts, organics) affect the nebulization, the size of the droplets transported into the plasma, which in turn affect the plasma load, temperature and ionization efficiency. A compromise was sought out here by preparing all calibration standards and sample dilutions with MilliQ.
• Because Au(III) stock solutions are usually stabilized in hydrochloric acid, while other standards are stabilized in nitric acid, it is advised to prepare the Au calibration standards separately from the multielement calibration standards.
• Calibration standards must be prepared freshly since not all elements are stable in H 2 O.

Apparatus and equipment
• Inductively Coupled Plasma Time-of-Flight Mass Spectrometer (TOFWERK AG, icpTOF S2, icpTOF R or icpTOF 2R) with standard liquid sample introduction system (i.e. pneumatic nebulizer and cyclonic spray chamber)

• Micropipettes & tips
• Analytical balance (e.g. Mettler Toledo) • Ultrasonic bath It is important to clean the glassware and cones, as well as to replace old used tubings prior to measurements to guarantee minimum background signals.

Ionic calibration standards solutions
Calibration standards for the element of interest, i.e. Au and Ag in this case, as well as for the determination of the transport efficiency are required.
In the presented example, Ag and Au calibration series were prepared from single-element standard solutions (Fluka and Inorganic Ventures). All dilutions were made with ultra-high purity water (MilliQ) and were carried out gravimetrically. Final concentrations ranged from 30 pg/g to 3 ng/g.

Nanoparticle standard
A well-characterized monodisperse nanoparticle standard, in terms of elemental composition, density, shape and size is required for the determination of the transport efficiency using the size method. A typical choice is NIST reference material SRM 8013, which consists of nominal 60 nm Au spherical NPs. However, due to scarcity of the material, any monodisperse well-characterized NPs of similar density to bulk material may be used. Additionally, the particle signal has to be clearly distinguishable from the dissolved and instrumental background to yield an unbiased mean or median sensitivity. Here monodisperse 50 nm spherical Au NPs from NanoComposix were used as a particle standard.

Sample
Bimetallic spherical Ag-Au NPs were used as sample in this example.  Table 3 Example of a sample of known characteristics which was analyzed to evaluate the SOP described in this document.

Sample preparation
If the samples have been stored in the fridge, it is recommended that they reach room temperature before dilutions are performed. Vigorous shaking or sonication of the NPs suspensions is required prior to dilution.
Here the 50 nm monodisperse Au NPs were diluted by a factor of 10 6 and the 60 nm Ag/Au coreshell NPs were diluted by a factor of 10 7 in ultra-high purity water.

Caution during sample preparation and storage of Ag NPs:
Ag NPs are susceptible to oxidation, hence exposure to light should be minimized.
 Store Ag NPs in amber bottles or bottles covered with aluminium containers in the fridge.
 Protect samples in aluminium foil.

Operation of the equipment
Instrument Start: 1. Start TOFpilot. The TofDaqViewer application will start automatically. 2. Make sure the tubing on the peristaltic pump is undamaged, hooked in with the plastic collars, and secured with the clamps. Check for the flow direction of the liquids.
3. In TOFpilot, on the top toolbar, choose the appropriate Tune Setting for your experiment, and click Get Ready to start the plasma. Wait for the startup sequence to finish. The indicator to the right of Get Ready will turn green and the Get Ready button will change to Shut down.

Measurement description
This section describes the set-up of a particle workflow in TOFpilot with a detailed step-by-step procedure: These can all be modified in the particle processing settings. 5) Check the Files/Path, for the path for the data in the data folder and the path for the results in the Results folder.
Additional information to particle processing settings: • Thresolding: ICP-TOFMS noise is not gaussian-distributed, hence it is recommended to use the Poisson based thresholding rather than the sigma approach.
• Averaging window: A running average is used in the peak extraction step. The larger the window the more computational power is required, while with the smaller the window more artefacts due to edge effects will appear. Hence, 1000 is a good compromise. • Insert Liquid flow/(ml/min) -sample uptake rate for the introduction system measured externally prior to the particle analysis  • If using an autosampler indicate rack and vial position

Additional information regarding integration time:
The integration time/s defines how many single TOF extractions are integrated into one datapoint, given in seconds. Robust timing settings allowing 100% data transfer are key for accurate results. It is recommended to use: • > 0.1 s for standards (longer integration time result in better signal stability) • Minimum available integration time for particle samples and standads (see specifications of instrument, 1 ms for icpTOF S2, 1.8 ms for icpTOF and 3 ms for icpTOF 2R) 7) Select all isotopes required for the analysis and TE determination 8) Edit the concentrations for the ionic standards and select the analyte for the particle standard vial.
Note: it is recommended to work with gravimetric concentrations for improved accuracy, but these can be added in later with the reprocessing module. 9) For the Particle Standard, the particle mass in fg needs to be specified. The Mass Calculator can be used to calculate this value from the specified diameter of the NP standard.

12) Example of a running sequence
The example sequence started with the particle vials (standards and samples) to avoid long wash out times and elevated baseline from the ionic standards. Ultra-high purity water was used to rinse between vials. Ionic standards were measured from low to high concentration.
After the successful completion of the sequence, rinse the system with 1% HCl and 1% HNO 3 . When all analytes have successfully been washed out and the baseline has returned to normal, rinse with MilliQ then run dry for a few minutes before switching off.

Reprocessing module
The processing of data is done in TOFpilot either directly following the measurement or later using the dedicated reprocessing module as highlighted here below.
And can be used to insert exact concentrations:

Calibration curves
To build calibration curves, first the averages and standard deviations of all the vials marked as standards by the user need to be determined. The "raw" intensities (ions/extraction) are extracted from the peak data and converted into counts and cps. The average signal (A) and standard deviations (SD) are calculated for all analytes defined in the peak list as following: where, a i is the signal intensity of analyte a per time bin i, given in counts per second, and n is the number of single data measurements in the run.
where, a i is the signal intensity of analyte a per time bin i, given in counts per second, is the mean value of all a i and n is the number of single data measurements in the run.
A weighted least squares fit is used to determine the calibration curve using 1/SD 2 as weights [4].
The least squares fit gives a function on the form:

= +
with associated uncertainties for slopes α i (counts per second per ppb) and intercepts β I (counts per second)

Limits of detection
Limits of detection (LOD) are calculated with the following methods (both are reported): where SD is the standard deviation of the signal in blank in counts, α the slope of the calibration curve and τ the dwell time in s. α * τ converts the slope of the calibration curve into counts/ppb, yielding LODs in ppb.

Particle events extraction
The following steps described the procedure performed automatically by the software. The procedure for identifying particle events is the same for the particle standards vials and particle sample vials. The required parameters can be set in the "particle processing settings" (chapter 5.5, step 4).
1. The raw data profiles are read out in counts/datapoints for all analytes.
2. The profiles are split into data chunks corresponding to the averaging window defined earlier (per default 1000 datapoints) 3. An iterative signal/background separation is performed on each data chunk: 3. 3.3. Repeat for the remaining dataset until no more values above the threshold are found or the specified maximum number of iterations has been reached. 4. The particle dataset is corrected for split-events: Peaks neighbouring other peaks are binned together, as it is assumed that these stem from the same particles. The maximum number of data points binned is configurable (default is 2). The timestamp of the highest of the binned points is taken as the timestamp for the binned data. 5. Analytes with identical bin times stem from the same particle and are stored on the same rows in the output csv files (see Annexe 1).

Particle quantification
The particle mass quantification for mono-metallic NPs is based on the work of Pace et al. [2] with some adaptations to match the TOF. Instead of working with peak heights, peak areas are used. In an effort to facilitate comparison, the same nomenclature was used. Multi-metal NPs or NPs containing elements not detectable by ICP-TOFMS are quantified by the same principle but including correction for the respective mass fraction of the element detected.

Results of the quantification
In the automatically generated report, signal intensity and mass histograms are presented for each analyte of interest for each particle sample vial as well as for the particle standard vial. Additional statistics such as number of registered events, average, median and standard deviation are also provided.
Consequently, the following values can be extracted:

Size determination
TOFpilot stops at the determination of the particle masses as this workflow is not limited to spherical nanoparticles but may be used for any type of single entities (nanorods, nanocubes, cells…). However, if both the shape and density are known, then the determined masses in the .csv files can be further converted to volume and the corresponding particle diameters can be calculated using appropriate formulas. Table 7 presents the corresponding sizes for the Au core and Ag shell for the measured test sample. A detailed discussion regarding further data processing and quality assessment can be found in the Annexe 2.

Multi-elemental fingerprinting
Further data processing steps regarding the multi-element composition of the particles can be performed from the provided .csv files (see Annexe 1). For example, composition filtering can be used.

Quality control
In order to assure accurate results, verify the following points: • The shape of the signal intensity distributions for the gold standard NPs should be Gaussian or lognormal. A bimodal distribution would indicate insufficient dilution or particle agglomeration. In the case of a bimodal signal intensity distribution, the sample should be further diluted. If the shape of the signal intensity distribution does not change after dilution, then sample preparation needs to be optimized or the quality of the samples needs to be assessed by an alternative method because the sample could have aged and degraded The same observation applies for known-homogeneous samples.
• Using the standard liquid sample introduction system of the icpTOF (pneumatic nebulizer and cyclonic spray chamber with peltier cooling), a transport efficiency ranging from 2 to 10% is expected.
If this is not the case, apply the following measures: o Check clamps on peristaltic pump o Replace tubings o Check the performance of the nebulizer and spray chamber:  If the nebulizer is spraying correctly, a fine mist should be observed.
 If coarse droplets are accumulating on the wall of the spray chamber, then it probably needs cleaning (for example, rinse 15 minutes with the rinsing solution).

Reporting of results
Results are automatically saved in a folder entitled QuantificationResults_<date>_<time>, which includes the following files: • AnalysisReport <datetime>.pdf : Compilation of results • particle_intensities_<name>_<uid>.csv : conversion of h5 data into a more user friendly format.
The .csv files display the timestamp and intensity (in counts) for each particle event. Data has been corrected for split events and blank subtracted. Simultaneous events will be shown on shared rows. One file for each particle and particle standard vial.
• particle_masses_<name>_<uid>.csv : Equivalent structure for particle intensities, but with calculated masses with one individual file for each particle vial.

Validation status
Results from the interlaboratory comparison study showed that all laboratories could determine the particle mass, size and particle number concentration of the test samples using the developed SOP and workflow within TOFpilot. After in-house and external validation, it was concluded that this SOP is well adapted and validated for particle mass, size and particle number concentration.

ANNEXE 1: Multi-elemental fingerprinting
A major feature of sp-ICP-TOFMS relies on its capability to distinguish single-element from multi-element NPs. For each particle which is transported into the plasma and ionized, full elemental spectra are recorded that allow the determination of the composition of the particle. Hence, it is important to understand how to recognize multi-element particles.
As mentioned, a full mass spectrum is acquired for each TOF extraction. The data is subsequently saved as a .h5 file but can be represented in a 2-dimensional array, where each row corresponds to a new time bin and each column to a different analyte/isotope. Consequently, signals which occur at the same time, are concurrent and are assumed to originate from the same particle. Although concurrent signals are assumed to stem from a true multi-element single particle, it is always possible that they originate from multiple concurrent particles. Indeed, when two independent particles A synthetic dataset is presented here to illustrate single and multi- only contains analyte C. For P2, P4 and P6, analytes A, B and C are all detected concurrently, hence these are considered as multi-element particles. P5 is also a multi-element particle, but of a different kind than P2, P4 and P6, as no signal is detected for analyte B.
reach the plasma at the same time, concurrent signals will be detected although they are not correlated. The probability of such cases is influenced by the PNC in the sample as well as the integration time used for the measurement and can be determined by event concurrency analysis. It is generally not possible to a posteriori distinguish whether a specific event results from multi-element NPs or from concurrent detection of multiple NPs. It is only possible to assign a probability for concurrent events from two particle types to occur by multiplying the probability of the occurrence of each particle type during the analysis. E.g. if 10 6 data sets (time bins) were collected and particle type P1 occurred 100 times and particle type P2 200 times, the probability of P1 and P2 occurring together would be 100/10 6 x 200/10 6 x 10 6 = 2%.
In the data processing of the particle workflow, a split event correction procedure is applied it and sums successive signals together into one bin as illustrated below. The same artificial data set as presented above has been corrected here for split events. P2 and P6 both had signals which were split over two time bins. Consequently, for P2 and P6, all signals are found in the same time bin, indicating a multi-element particle composed of analytes A, B and C.

ANNEXE 2 : Further Data Processing & Discussion of the results
In this document, results from the further analysis of the processed data are discussed. Software: • Excel was used for composition filtering • Igor Pro 7 was used for data representation and fits.
• TOFDAQViewer was used for data visualization of the raw .h5 files.

Signal Intensity histograms
The signal intensity histograms can be directly plotted from the "particle intensities" file, as shown in Figure   A1, for 107 Ag and 197 Au, respectively.

Mass histograms
The mass histograms can be directly plotted from the particle masses file, as presented in Figure A2. It should be however noted that the masses are given in (g) and the conversion to (fg) leads to the artificial introduction of zeros into the dataset when multiplying blank cells and numbers using Excel 3 (blank cell*10 -15 = 0). These artificial zeros need to be filtered out, otherwise an artificial bin at position zero will appear. Alternatively, blank cells need to be ignored when performing calculations.

Size histograms -Core diameter and shell thickness
From the particle mass (g), the diameter of the gold core is calculated using the following formula, where is the diameter in cm, is radius in cm, is the calculated gold mass in g, is the density of gold in g/cm 3 . Results are displayed in Figure A3. The hollow sphere formula is used to calculate the thickness of the silver shell, where is the sum of the gold core radius and the shell thickness in cm, is the mass of silver in g, and the density of silver in g/cm3. The shell thickness R and total diameter are displayed in Figure A4 and A5, respectively.  An unexpected second hump of smaller magnitude can be observed in Figure A4 after conversion from mass to shell thickness. It should be highlighted that this hump was not previously observed in either the signal intensity histogram ( Figure A1) or the mass histogram ( Figure A2) and only appeared after calculations involving both the Ag signals and its "concurrent" Au response (Au signal on the same row). In Figure A5 displaying the total diameter of the coreshell NPs, a similar hump to that observed in Figure A4 can again be observed. In both cases, the hump is centred around 25-30 nm. A closer look into the .csv files revealed that Ag occurrences were not always associated with a gold signal but sometimes with a blank cell.
Because the composition of the test NPS is known, namely that they are composed of a gold core and a silver shell, the data can be filtered with respect to particles composed of both elements. Figure A6 shows the shell thickness distributions after filtering with respect to the simultaneous occurrence of Ag and Au signals. From Figure A6, it becomes clear that the smaller magnitude hump observed in Figure A4 and Figure A5, corresponds to Ag signals without any concurrent Au signals. Consequently, if the particle signals are filtered based on their dual composition of silver and gold, a monomodal histogram is obtained for the total diameter (see Figure A7). After composition filtering, results are in good agreement with the expected values and are summarized in table A1. Although care needs to be taken in the interpretation of the data, it should be noted that only TOF data allows for such in depth investigation of multi-element particles.

Measured (Guaussian fit)
Au Core Diameter (nm) 30 ± 3 29 ± 7 Ag Shell Thickness (nm) 14.5 ± 1.5 13 ± 5 Total diameter (nm) 59 ± 6 56 ± 9 15% of the 6850 observed Ag signals were not associated with gold signals. Hence in order to better understand why some Ag signals are not associated with Au signals, the raw .h5 file was analyzed using TOFDAQViewer. The time traces for 107 Ag (red), 109 Ag (pink) and 197 Au (blue) were monitored. The instances of Ag signals without concurrent Au signals (blank cells in the Au column of the processed .csv files) can be explained by three cases: 1) The particles are exclusively composed of silver (see Figure A8) 2) The particles have a silver shell and a small gold core, whose signal is below the thresholding limit. (see Figure A9) 3) The particles are composed of both a silver shell and a gold core, but their signals are split over multiple bins. The split event correction integrates the analyte signals into the maximum bin, but as the Au signal and Ag signal have different maximum bins, they are separated and appear as "two different" particles in the processed .csv file. (see Figure A10 and Annex 1) Based on the assumption that the test sample contained exclusively silver shelled gold core NPs and taking in to account that the sensitivity was sufficient to ensure that Ag and Au events were above LOD for all NPs, the number of Au and Ag events would have to be identical resulting in a ratio of PNC Au /PNC Ag =1. Even in the case of a single particle split in two events (case 3), the number of Au or Ag containing particles would not change. Hence, the higher PNC Ag compared to the PNC Au supports the occurrences of cases 1) and, eventually, 2). The presence of apparently pure Ag NPs was here unexpected and shows that the chosen test sample was either not as monodisperse or compositionally pure as initially assumed and reported.