Quantitative TEM Validation for Nanomaterials

The validation study is performed on a series of certified reference materials (CRMs) to test the capabilities of the method: colloidal silica in water: ERM-FD100 [29 ] and ERM-FD304 [30 ] (JRC, Geel, Belgium); representative test materials (RTMs): titanium dioxide: NM-100 and NM-103 [52 ], and cerium oxide: NM-212 [53 ] (JRC, Ispra, Italy); and, gold nanorods dispersed in aqueous medium (product number 46945, Lot number L05X007, Alfa Aesar, Thermofisher scientific, Karlsruhe, Germany) [54 ].
For the selected CRMs, ERM-FD100 and ERM-FD304, the homogeneity and stability of the ampouled, diluted raw material, as well as the characterization while using an interlaboratory comparison approach, are described in certification reports [29 ,30 ]. For the RTMs, NM-100, NM-103, and NM-212, it is either generally assumed or demonstrated that these materials are sufficiently homogenous and stable with respect to their constituent particle size [52 ,53 ,55 (link)]. ISO/TS 16195:2013 notes that such RTMs can be a useful tool in inter- or intra-laboratory developments of test methods for which reference materials were not (yet) produced [56 ]. The dispersion of gold nanorods in water is considered to be stable by the manufacturer under the recommended storage conditions [54 ].
All of the described materials are included in the intra-laboratory validation study, and four selected materials are included in the inter-laboratory validation study: ERM-FD100, NM-100, NM-212, and the gold nanorods dispersion.
The ERM-FD100, ERM-FD304, and gold nanorods dispersions were vortexed during 5s by a IKA Vortex Genius 3 (IKA®-Werke GmbH & Co. KG, Staufen, Germany) to ensure the homogeneous dispersion of the particles. These materials were not further diluted. The powdered materials NM-100, NM-103, and NM-212 were dispersed while using the ENPRA dispersion protocol for NANoREG [57 ,58 (link)].
The approach to characterize the selected materials by EM consists of a combination of three SOPs [34 ,35 ].
The SOP on EM specimen preparation: “Preparation of EM-grids containing a representative sample of a dispersed nanomaterial” describes how to bring a dispersed nanomaterial in contact with an EM-grid, and to select the appropriate concentration of the nanomaterial, and the type and charge of the grid [35 ]. These conditions have to be chosen, such that the fraction of nanoparticles attached to the grid optimally represents the dispersed nanomaterial, and that the particles of interest can be detected later by image analysis software. TEM specimens were prepared using Alcian blue treated positively charged pioloform- and carbon-coated, 400 mesh copper grids (Agar Scientific, Stansted, Essex, UK), by drop deposition in this study.
The SOP on TEM imaging: “Transmission electron microscopic imaging of nanomaterials” aims to record a set of calibrated transmission electron images that representatively show the nanomaterial on the TEM specimen [35 ]. The SOP foresees that the images are randomly and systematically recorded, at 10 positions that are pre-defined by the microscope stage and evenly distributed over the entire grid area to avoid subjectivity in the selection of particles by the analyst. The microscope used in this study was a Tecnai G2 Spirit TEM with BioTwin lens configuration (Thermo Fisher Scientific, Eindhoven, The Netherlands). Micrographs were recorded with a 4 × 4 k Eagle charge-coupled device (CCD) camera (Thermo Fisher Scientific, Eindhoven, the Netherlands) while using the TEM imaging and analysis (TIA) software (Version 3.2, Thermo Fisher Scientific, Eindhoven, The Netherlands). For each material, a suitable magnification allowing for measuring a high enough number of particles for descriptive and quantitative image analyses was selected (Table 1). The intra-laboratory validations of both CRMs ERM-FD100 and ERM-FD304 were performed at magnifications of 18,500× and 68,000× to determine the effect of the quantification limits imposed by the selected magnification on the precision and trueness of the method.
The SOP on image analysis: “Measurement of the minimal external dimension of the constituent particles of particulate materials from TEM images by the NanoDefine ParticleSizer software” describes the application of the ParticleSizer software [34 ,43 ]. The ParticleSizer software allows for selecting four image analysis modes (“Default”, “Irregular Watershed”, “Ellipse fitting”, or “Single particle mode”) to measure the constituent particle properties, depending on the type of particle (ellipsoidal or irregular) and type of overlap between particles (no overlap, touching, low degree of overlap, high degree of overlap) (Figure 1).
ERM-FD100 and ERM-FD304 are examples of stable aqueous colloids of non-aggregated particles and they were selected to validate the “Default” mode. The gold nanorods dispersion is an example of an agglomerated material with irregular touching or slightly overlapping constituent particles and it was selected to validate the “Irregular watershed” mode. NM-100 is an example of an aggregated/agglomerated material with spherical or ellipsoidal touching or slightly overlapping constituent particles and it was selected to validate the “Ellipse fitting” mode. NM-103 and NM-212 are examples of aggregated/agglomerated materials with highly overlapping constituent particles and they were selected to validate the “Single particle” mode.
The scope of this validation study was to validate the measurement of the median value of the number-based distribution of the minimal external particle dimension, being assessed as the minimal Feret diameter. In case the “Ellipse fitting” mode was selected, the minimal Feret diameter was estimated as the length of the short axis of the fitted ellipse. The measurement uncertainties that were associated with the quantitative TEM measurement of the median of the minimal Feret diameter distribution were estimated while using a top-down approach (Figure 2). For each material, a set of 150 images was generated by performing measurements on five days within one week. On each day, three TEM specimens (repetitions) were prepared from one vial and then imaged by TEM. From each TEM specimen, 10 images were systematically and randomly recorded over the grid surface.
For each set of 150 images (i.e. for each material), the image analysis settings were optimized on a representative set of images, while using the suitable image analysis mode (Default, Irregular Watershed, Ellipse fitting, and Single particle). Subsequently, these settings were applied on all 150 images, in sets of 10 images that originated from one TEM specimen, resulting in 15 minimal Feret diameter distributions. The median value of each minimal Feret diameter distribution was determined. One-way analysis of variance (ANOVA) was performed on these 15 median values to estimate the precision associated with the measurements. The uncertainty that was associated to repeatability, u_r, and the uncertainty associated to day-to-day variation, u_day, were estimated based on equations (1) and (2), respectively: $u_r=\frac{\sqrt{M{S}_{within}}}{C_m}$
$u_{day}=\{\begin{array}{c}\begin{array}{cc}\frac{\sqrt{\frac{M{S}_{between}-M{S}_{within}}{n_r}}}{C_m}& forM{S}_{between}>M{S}_{within}\end{array}\\ \begin{array}{cc}\frac{\sqrt{\frac{M{S}_{within}}{n_r}}\sqrt[4]{\frac{2}{{\nu }_{MSwithin}}}}{C_m}& forM{S}_{between}<M{S}_{within}\end{array}\end{array}$
With n_r the number of replicates per day (three replicates), MS_Within the mean squares within days, MS_Between the mean squares between days, ν_MSwithin the number of degrees of freedom within sample units and C_m the mean. The uncertainty that was associated to intermediate precision, u_IP, combines u_r and u_day (Equation (3)).
$u_{IP}=\sqrt{u_r^2+u_{day}^2}$
u_IP was combined with the uncertainty associated to calibration, u_cal, and the uncertainty associated to trueness, u_t to determine the full uncertainty budget of the approach.
u_cal was determined based on the variation on the calibration results. The lower magnifications (440 to 23,000) were calibrated while using the cross-grating method and the intermediate magnifications (30,000 to 180,000) were calibrated using the image shift method based on a 2160 lines/mm diffraction-cross grating (AGS106L, Agar Scientific, Stansted, Essex, UK). The calibration method was implemented by using the magnification calibration software, which is integrated in the Tecnai user interface software (Version 3.1.1, Thermo Fisher Scientific, Eindhoven, The Netherlands) [59 ]. Magnification calibration was further verified by comparing the measured values with the (certified) value of CRM, including ERM-FD100, ERM-FD101, and ERM-FD304, assuring SI-traceability.
To obtain the uncertainty that is associated to trueness (u_t), the uncertainty associated to trueness of a certified reference material (u_t,CRM) has to be combined with u_IP (Equation (4)): $u_t=\sqrt{u_{t,CRM}+u_{IP}^2}$
However, no (certified) reference values and associated uncertainties for the median of the minimal Feret diameter distributions are currently available for the tested materials. Therefore, the standard uncertainty (k = 1) of the certified modal equivalent circular diameter (ECD) value of ERM-FD100 and the indicative modal ECD value of ERM-FD304, both being obtained by EM, and referred to as u_CRM, were applied as an estimate for u_t,CRM [29 ,30 ]. The mean of the uncertainties associated to trueness of ERM-FD100 and ERM-FD304 was added to their uncertainty budget as u_t,CRM since no (certified) reference values were available for the other materials.
Assuming that all of the uncertainty contributions for the presented approach are covered by the uncertainty associated to repeatability, the uncertainty due to day-to-day variation, the uncertainty associated to calibration, and the uncertainty associated to trueness, the combined measurement uncertainty, u_c(x), was estimated by Equation (5): $u_c(x)=\sqrt{u_{IP}^2+u_t^2+u_{cal}^2}$
When assuming that the combined uncertainty is normally distributed and requiring a confidence level of approximately 95%, the combined uncertainty is multiplied by a coverage factor of 2 to obtain the expanded measurement uncertainty, Ucx (Equation (6)).
$Ucx=2u_c(x)$
Due to the lack of certified reference materials, the trueness of the approach could only be assessed based on the certified modal equivalent circular diameter (ECD) value, C_CRM, of ERM-FD100, and the indicative modal ECD value of ERM-FD304, with both being obtained by EM. For each particle, the ParticleSizer application measured the ECD parameter together with the minimal Feret diameter parameter. A histogram was constructed from the raw data for each set of 10 images in the validation studies of ERM-FD100 and ERM-FD304, and a normal distribution was fitted to the raw data to determine the mode. The mean modal ECD and the corresponding expanded measurement uncertainty were determined, as described above. The absolute difference between the mean measured value and the reference value (Equation (7)), Δ_m, and the combined uncertainty of result and certified value (Equation (8)), u_Δ, were calculated. To evaluate the method performance, Δ_m was compared with the expanded uncertainty U_Δ [60 ,61 ,62 ].
${\Delta }_m=\left|C_m-C_{CRM}\right|$
$u_{\Delta }=\sqrt{u_c{(x)}^2+u_{CRM}^2},\mathrm{and}{U}_{\Delta }=2u_{\Delta }$
In addition to determining the precision and trueness of the approach, method performance characteristics, including limit of detection, working range, selectivity, ruggedness, and robustness were assessed in the intra-laboratory validation study.
The image analysis part of the intra-laboratory validation study was performed independently on identical sets of images by three experienced and trained labs referred to as partner 1 (P₁), partner 2 (P₂), and partner 3 (P₃), and by two independent test persons (TP₁ and TP₂) that did not receive any training or explanation on optimization of settings to test the ruggedness of the ParticleSizer application against the operator. The u_IP obtained for the different materials by P₁, P₂, P₃, TP₁, and TP₂ were compared. In addition, the ruggedness of the approach was evaluated against variation in the number of measured particles by determining u_IP of the quantitative TEM analysis from subdatasets of measurements in function of the number of analyzed particles.
The robustness of the approach was evaluated against small variations in the image analysis settings and against classification of a material. Analyzing one image of each material by the four different modes tested the latter (Default, Irregular watershed, Ellipse fitting, and Single particle mode). A visual comparison of the particles detected by each mode was made, and the resulting median values of the minimal Feret diameter distributions were compared.
The image analysis part of the approach was evaluated in an inter-laboratory validation study. The SOP for image analysis, containing a comprehensive description of all operational procedures, and 150 images of each selected material (ERM-FD100 at a magnification of 68,000×, Gold Nanorods, NM-100 and NM-212), were distributed to all participating laboratories. The participants were requested to optimize image analysis settings themselves in a specific image analysis mode, as specified in Figure 1 for the respective materials. The same image analysis settings had to be applied on all images of a certain material, in sets of 10 images. Participants were requested to report 15 median values of the minimal Feret distributions per material. The participants were requested to strictly follow the SOP, since the interlaboratory comparison aimed at the validation of the method and not at assessing the proficiency of the laboratories. The results were reported on-line while using the JRC in-house developed MILC® interface (JRC, Geel, Belgium).
The statistical evaluation of the data was performed following the recommendations of the ISO 5725-2:1994 standard [61 ]. AOAC International harmonized guidelines for collaborative study procedures to validate the characteristics of analysis methods were also followed as a cross-validation for the data evaluation [63 ]. Outliers in the laboratory precision were checked by applying the Cochran test that compared the highest laboratory internal repeatability variance with the sum of reported variances from all of the participants. Laboratory outliers within the series of independent replicates were checked by applying the Grubbs-internal test (repeatability). Pairs of outliers were checked by applying the double-Grubbs’ test. The outliers in the laboratory mean were checked by applying the Grubbs test, checking for laboratory means significantly deviating from the total mean calculated from data reported from all participants.
The results were compared with their respective critical values at 1%cv (99% confidence level) and 5%cv (95% confidence level) for both statistical tests (Cochran and Grubbs), as foreseen in ISO 5725 [61 ,62 ].
The values for the target performance characteristics of the method, namely within-laboratory repeatability (RSD_r) and between-laboratory reproducibility (RSD_R) per each test material, were determined by ANOVA based on the remaining valid results after the exclusion of the non-valid results, including the non-compliant laboratories as well as the statistical outliers (Cochran and Grubbs tests).