First, we investigated the influence of substrate holder material on measurements by FP-XRF. The substrate holders supplied with the nano-MOUDI are composed of Al alloy with a magnet attached to the backside. Holders with the same dimensions as the supplied Al holders were machined from polyether ether ketone (PEEK). The PEEK holders were not fitted with a magnet on the backside. (Figure 1(a)).

The Al alloy and PEEK holders were analyzed by FP-XRF (Niton XL3t Ultra, Thermo Scientific, USA) with and without substrates (for each holder material, n = 3 without substrate and n = 9 with substrate, total 24 measurements). For substrates, PC filters (PCT0247100, pore size of 0.2 μm, diameter of 47 mm, Sterlitech, USA) were coated with silicone oil (Heavy-duty silicone spray, Part #07041, MSP Corp., Shoreview, MN, USA) to prevent particle bounce. The oil was applied using a foam stamp (diameter of 20 mm; see supplemental information) and then placed in an oven at 50°C for five hours to evaporate volatile components in the oil. We selected this substrate because it can be fully digested with a microwave-assisted acid digestion method as specified below. A 3D-printed support (see supplemental information) was used to keep a consistent distance between the FP-XRF and the surface of holder, to ensure consistency between analyses. The FP-XRF thin-film (standard filter) mode was used to measure metals in units of μg/cm². The measurement time was set to 120 seconds (main range: 40 seconds, low range: 40 seconds, and high range: 40 seconds).

The experimental setup is shown schematically in Figure 1(b). Dry and particle-free air, controlled by a mass flow controller (MFC; MPC20, Porter Instrument, USA), was delivered to a spark discharge system equipped with stainless steel electrodes (303 alloy; 2EXC7, Grainger, USA) to generate stainless steel particles (Park et al., 2015). Highly charged, <10 nm sized, and chain-like aggregated particles containing Cr and Fe, were generated from the spark discharge system. The test particles were passed through a polonium-210 aerosol neutralizer (2U500, Staticmaster, USA) to neutralize their charge to Boltzmann equilibrium, then passed through a coagulation chamber (200 L) to increase their size from <10 nm to larger than 60 nm, and finally delivered to a sampling chamber (22 L). The concentration of test particles was diluted by room air that was first passed through a high efficiency particulate air (HEPA) filter. An MFC and vacuum pump were used to control the airflow rate of dilution air.

The test particles were sampled using a nano-MOUDI (Model 125-R, MSP, USA) at an airflow rate of 10 L/min. Polycarbonate substrates were fixed to the PEEK holders and loaded into the nano-MOUDI stages for particle collection. A mixed cellulose ester (MCE) filter (FMCE847, pore size of 0.8 μm, diameter of 47 mm, Zefon International, Inc., FL, USA) was used as a backup filter. The sampling time was four hours (sampling volume of 2.4 m³) and sampling was performed in triplicate.

The particle-laden substrates were analyzed directly using a FP-XRF (Niton XL3t Ultra, Thermo Scientific, USA) as shown in Figure 1(a). After analyzing Fe and Cr content, and the mass per substrate of i-th stage (M_i) of nano-MOUDI was calculated as follows:

where M_Ai and M_Ab are the mass per unit area of i-th stage and blank substrates, respectively. To account for possible variations in the mass of the blank substrates in the calculations, the average value of nine black substrates was used for M_Ab. A_i is the sampled area of the nano-MOUDI at the i-th stage (A_1,3-10 = 1.8 cm², A_2,11 = 3.1 cm², A₁₂ = 4.9 cm², A₁₃ = 5.7 cm², and A₁₄ = 10.8 cm²).

Following the FP-XRF analysis, the particle-laden substrates were subjected to microwave-assisted acid digestion. The digestion protocol was a slightly modified version of the protocol used to dissolve TiO₂ nanoparticles in our previous study (Mudunkotuwa et al., 2016). Briefly, the particle laden substrates were carefully placed in Teflon digestion vessels, a combination of conc. H₂SO₄ (6 mL) and conc. HNO₃ (3 mL) was added, and the substrates were digested using a microwave digestion system (MARS 6, CEM Corporation, USA) for 45 minutes. The digestate was added to vials containing water (10 mL) and allowed to degas overnight in the fume hood. An additional dilution step (10 times dilution) was carried out to reduce the acidity of the medium. Finally, the solutions were topped up to 30 mL with water and analyzed by ICP-MS. The Fe and Cr content was analyzed and the values for mass per substrate were blank-subtracted.

For each metal, the mass per substrate measured by FP-XRF and ICP-MS were converted to mass concentration (C_i) as:

The total mass concentration was calculated as follows:

where Q and t are the sampling flow rate and time, respectively. Mass mean diameters (d_MMs) for fine (< 300 nm) and coarse (> 300 nm) modes were calculated using mass concentration from FP-XRF and ICP-MS as follows:

where d_i is the midpoint diameter at the i-th stage of the nano-MOUDI. Midpoint diameters and concentrations from 8^th - 14^th stages were used for fine mode, while midpoint diameters and concentrations from 1^st - 7^th stages were used for coarse mode.

The mass per substrate data produced by FP-XRF were compared to those from ICP-MS. The correlation between FP-XRF and ICP-MS was investigated using fitted linear regression models. The slope, y-intercept, and R² were calculated using Excel (14.0.7166.5000, Microsoft, USA).

With the results from FP-XRF compared with ICP-MS, respiratory deposition of Fe and Cr particles was calculated from the respiratory deposition curve presented by ICRP (1994). Fractional mass concentrations calculated as follows:

where X represents the fraction (inhalable, Inh; respirable, Resp; particles deposited in the head airways, HA; particles deposited in the tracheobronchial, TB; and particles deposited in the alveolar, AL). F_X,i is the fraction for the i-th stage of the nano-MOUDI. These fractions were calculated using equations from Hinds (1999).

The original Al alloy holder and the custom PEEK substrate holders (both with and without the PC substrate) were analyzed for metals (Fe, Cr, Mn, Cu, Zn, Ni, Ti, and Pb) using FP-XRF (Table 1). Al was excluded because XRF detection is generally limited to elements with an atomic number greater than 16 (Brouwer, 2006) and because of the Al in the holder itself. Eight metals were detected in Al alloy holders with the magnet. Substantially fewer metals (Cr, Cu, and Ti) were measured in the PEEK holder and at lower concentrations than the Al alloy holders.

The presence of the PC substrate had no effect on the concentrations of metals measured in the Al or PEEK holders (Table 1). Hence, we conclude that neither the PC substrate nor oil on the substrate contained measureable metals. Moreover, these data provide evidence that the x-ray beam from the FP-XRF penetrates the substrate and excites metals in the holder.

Metals in blank MCE filters were measured by FP-XRF. The blank MCE filter was positioned in a thin-film holder instead of original filter holder. No metals were detected in blank MCE filters.

Metals (Fe and Cr) in blank PC substrates and MEC filters measured by ICP-MS are shown in Table 2. Less than 0.5 μg of Fe and Cr were detected in PC substrates and MEC filters. The limit of detection of ICP-MS can be calculated as the mean plus three standard deviations of the elemental mass per substrate found in blank substrates (IUPAC, 2006). However, the limit of detection of FP-XRF is difficult to calculate since the FP-XRF detected only Cr in PC blank substrate with PEEK holder while it detected no Fe in PC substrate with PEEK holder and no metals in MCE filters. The limit of detection for Cr could be calculated from M_Ab (Table1) and varied from 5.9 to 19.1 μg since the sampling area of nano-MOUDI stages varied from 1.77 to 5.73 cm².

Particle size distributions measured by FP-XRF and the ICP-MS are shown in Figure 2. A bimodal distribution was observed with most particles in the fine mode and a small fraction of particles in a coarse mode. Whereas the fine mode was always visible for Fe and Cr when measured by FP-XRF and similar in magnitude to the ICP-MS, the second coarse mode was visible in only some of the size distributions measured with ICP-MS. The coarse mode was most prominent for Fe measured by ICP-MS. We believe that the coarse mode was visible only for ICP methods because ICP-MS can achieve limit of detections many orders of magnitude lower than FP-XRF with an instrumental limit of detection of 1-10 ppt. Therefore, the limit of detection of FP-XRF for Fe and Cr is expected to be higher than one for ICP-MS (Table 2). In fact for Cr, the calculated limit of detection of ICP-MS (0.14 μg) based on the blank substrate measurements were much smaller than one of FP-XRF (5.9 ∼ 19.1 μg). Fe in blank substrates was detected only with ICP-MS. Therefore using the instrumental limit of detection of the FP-XRF (standard 37-mm filter mode) for Fe and Cr were approximately 0.2 μg/cm² (Laohaudomchok et al., 2010) the method limit of detections by mass were estimated to be within 0.4 to 2.2 μg.

The mode diameters and total mass concentration are summarized in Table 3. Mode diameters for Fe and Cr obtained for the fine mode were similar between FP-XRF and ICP-MS. In contrast, coarse mode from FP-XRF were substantially smaller than those from ICP-MS. Fe and Cr collected on the substrates at the larger size stage in nano-MOUDI were detected more precisely by ICP-MS and this resulted in larger coarse mode diameter. Total mass concentrations for Fe and Cr from FP-XRF were similar to those from ICP-MS.

A scatter plot of mass detected on the substrate measured by FP-XRF relative to that from ICP-MS is shown in Figure 3. For both Fe and Cr, the mass detected by FP-XRF was linearly correlated with that measured by ICP-MS with most data near the 1:1 correlation line (illustrated by a dotted line). Fe was detected by both FP-XRF and ICP-MS at mass loadings higher than 2.5 μg but not by FP-XRF < 2.2 μg. Thus, at least 2.5 μg of Fe must be collected for detection by FP-XRF. Black dotted lines in Figure 2 identify these detectable Fe concentrations calculated from dividing 2.5 μg by actual sampling volumes. The detectable concentration was 1.0 μg/m³ for the comparison of FP-XRF and ICP-MS. Using these computed detection limit, differences between results from FP-XRF and ICP-MS at the fine and coarse modes could be distinguished. Concentrations from FP-XRF, below the dotted line at the coarse mode, did not match well with the concentrations from ICP-MS.

The FP-XRF detected 30% more Cr than the ICP-MS as shown in Figure 3. One reason for these differences could be that experiments were conducted at the low concentrations near the detection limit of FP-XRF. Moreover, FP-XRF technique involved a single measurement location (at center of substrate), whereas ICP-MS measured all the particles collected on the substrate surface. Therefore, the one point measurement with FP-XRF at the low concentration may not accurately represent the total deposited mass averaged over the substrate surface.

Three data points (blue dot line in Figure 3) were far from the 1:1 line (black dot). Those points represent data collected on the MCE filters used in the last stage of the nano-MOUDI. The fact that particles are distributed over a much larger area on the MCE filter (10.75 cm²) than the PC substrates (1.77 – 5.73 cm²) increases uncertainty in the measurement of mass by FP-XRF. Also, uncertainty about the spatial distribution of particles on the MCE filter combined with the single FP-XRF measurement location (center) may have resulted in differences between analytical measurements. These error values from MCE filter could be multiplied when the mass per substrate is calculated using Equation 1.

One application of this method is to estimate the quantity of material deposited in human body. Concentrations of Fe and Cr particles deposited in the respiratory system are documented in Table 4. Total mass concentration measured by FP-XRF was the same as concentrations of inhalable and respirable particles because most test particles were smaller than 300 nm. When workers are exposed to 8.6 μg/m³ of Fe and 4 μg/m³ of Cr, the mass of regional deposition can be predicted. Human inspiratory flow rate may vary from 0.39 to 3 m³/h (Hinds, 1999). When an inspiratory volume was assumed to be 1.5 m³ for one hour, total respirable depositions of Fe and Cr were is 13 μg and 6 μg, respectively. The estimated mass that would be collected in the alveolar region was 2.8 μg for Fe and 1.3 μg for Cr. A total of 22% of respirable Fe and Cr particles were estimated to be deposited in the alveolar region.

A limitation of this method is that at least one stage of the nano-MOUDI should collect more than 2.5 μg of Fe to distinguish a mode diameter. The threshold limit value (TLV) of Fe is 5000 μg/m³ (ACGIH, 2006). If particles are sampled at a concentration of 10% of the TLVs for 10 minutes, the expected mass collected by the nano-MOUDI is 50 μg. If the mass fraction at the mode diameter is 20%, 10 μg of Fe could be collected on the substrate at a stage with a cutoff diameter similar to mode diameter. Thus, sampling for at 10 minutes is required with our method to distinguish mode diameter at a mass concentration of 1/10^th TLV. Longer sampling time can allow for better correlation of FP-XRF and ICP-MS results. This study was also limited to fine-mode Fe and Cr particles. Future work will include testing coarse-mode and other metal-containing aerosol particles. The spark discharge system could be modified with different electrodes such as Mn, Cu, or Ti rods and the agglomeration chamber could be modified to increase the size of the test particles.

Removal of the magnet from the substrate holder did not affect uniformity of particle deposition. Uniform deposition was verified visually with new PEEK holders. Friction between the PEEK holder and the rotating plate was sufficient to keep the PEEK holder in place. However, care should be exercised to avoid tilting the nano-MOUDI and dislodging the holder from the rotating plate. Custom PEEK holders can be used for any kind of air sampling membrane filters with diameters of 47 mm.