Are there generalizable trends in the release of airborne synthetic 1 clay nanoparticles from a jet milling process?

13 14 15 Despite recent efforts to assess the release of nanoparticles to the workplace during different 16 nanotechnology activities, the existence of a generalizable trend in the particle release has yet to 17 be identified. This study aimed to characterize the release of synthetic clay nanoparticles from a 18 laboratory-based jet milling process by quantifying the variations arising from primary particle 19 size and surface treatment of the material used, as well as the feed rate of the machine. A broad 20 range of materials were used in this study, and the emitted particles mass (PM 2.5 ) and number 21 concentrations (PNC) were measured at the release source. Analysis of variance, followed by 22 linear mixed-effects modeling, was applied to quantify the variations in PM 2.5 and PNC of the 23 released particles caused by the abovementioned factors. The results confirmed that using 24 materials of different primary size and surface treatment affects the release of the particles from 25 the same process by causing statistically-significant variations in PM 2.5 and PNC. The interaction 26 of these two factors should also be taken into account as it resulted in variations in the measured 27 particles release properties. Furthermore, the feed rate of the milling machine was confirmed to 28 be another influencing parameter. Although this research does not identify a specific pattern in 29 the release of synthetic clay nanoparticles from the jet milling process generalizable to other 30 similar settings, it emphasizes that each tested case should be handled individually in terms 31 exposure


37
Due to their novel properties, the application of nanotechnology products has been growing 38 over the last two decades, and at the same time, their potential impacts on human health and 39 environment have also been extensively studied. For example, there has been a significant growth 40 in the number of published studies on workplace exposure, as the number of workers involved in 41 different stages of the nanotechnology process (including production, processing, handling, 42 bagging and shipping) is also increasing. Review papers (Brouwer 2010;Brouwer et al. 2009;43 Kuhlbusch et al. 2011) summarized the numerous studies conducted in various nanotechnology 44 workplaces, alongside the measurement instruments and characterization strategies which were 45 used. 46 In particular, characterizing the release of airborne nanoparticles during the abovementioned 47

Materials tested 109
Synthetic clays have certain advantages over natural clays and are produced through different 110 methods, including the ion exchange process by using various surfactants (Stoeffler et al. 2008). 111 In this experiment, a total number of 24 clay nanocomposites were tested (4 different sizes with 6 112 different surface treatments). The samples used and their primary sizes were: Lucentite® (25 113 nm), Laponite in two sizes of 80 and 120 nm (quoted as H80 and H120 in this paper), and 114 Cloisite® (300 nm). Each sample was available in plain or unmodified form (referred to as "-N" 115 in the sample's name throughout the text), as well as in five other surface treatments. These 116 surface treatments were obtained by using different mixing ratios of two surfactants: Choline 117 Chloride (CC) and Ethoquad O/12 PG (E-THO), with molecular masses of 139.6 and 406.1 g 118 mol -1 , respectively. The obtained surface treatments and surfactant mixing ratios used were: CC 119 (100% CC), C-MOD (75% CC, 25% ETHO), M-MOD (50% CC, 50% ETHO), E-MOD (25% 120 CC, 75% ETHO), and E-THO (100% ETHO). 121 7

Instrumentation 123
A range of real-time measurement devices were used for this nanotechnology process. 124 However, in this paper, we were only interested to study the variations in real-time particle mass 125 and number concentrations. 126 The instruments used in this study were: 127 -A condensation particle counter (TSI model 8525 P-Trak) to measure PNC; size range = 128 0.02-1 µm 129 -A TSI model 8520 Dust-Trak with a 2.5 µm impactor at its inlet to measure PM 2.5 130 -A scanning mobility particle sizer (SMPS) to measure particle number size distribution 131 comprised of a TSI long differential mobility analyzer model 3081 and a TSI 132 condensation particle counter model 3782 with measurement size from 10 nm, scan time 133 of 150 seconds (120 s up-scan followed by a 30 s down-scan), and the sheath and aerosol 134 air flows of 6 and 0.6 L min -1 , respectively. 135 -A TSI model 7545 indoor air quality meter (IAQ) to monitor the temperature and relative 136 humidity of the room where the jet milling machine was located 137 At the beginning of each measurement day, the instruments times were synchronized and 138 their sampling intervals were set to the same value. In order to be able to record all of the the shortest time possible (1 second). 141 142

Study design 143
Using the P-Trak, the source of airborne particle release from the jet milling machine was 144 identified to be at the connection point of the collection bag to the venturi outlet. The 145 measurement setup, including the Dust-Trak, was placed on a trolley, adjacent to the jet milling 146 machine. A black conductive rubber tube with the length of 20 cm was connected to the Dust-147 Trak aerosol inlet with its other end placed very close (~ 2 cm) to the particle release source, 148 where the P-Trak was placed on a stand and fixed by a clamp to ensure uniform measurement 149 throughout the experiment. Fig.1 shows a schematic of the jet milling machine and the locations 150 of the measurement instruments. 151 For each sample, the allocated experiment time was around 18 minutes. One test was 152 conducted per each combination of primary particle size and surface treatment, beginning with 153 the smallest size/lightest surface treatments (LUC-N) and ending with the largest size/heaviest 154 surface treatment (CLO-ETHO). The tests were carried out according to the below plan: 155 -Measurement of background particles at the release source while the milling machine was 156 on and before feeding the material (~ 3 min) -Measurement at the breathing zone of the operator during the milling process (~ 3 min) 159 -Measurement at the release source after turning the machine off and during dismantling 160 and cleaning (~ 3 min) 161 Since the source and breathing zone were located fairly close (only one meter apart), moving 162 the instruments from one location to the other was almost instantaneous. Yet, just before moving 163 the instruments, data logging was paused for a few seconds and recommenced after relocating 164 and reaching the steady reading. In this study, the entire data measured at the source during the 165 milling process are considered, however, to verify the release of the particles from this process, 166 taking into account the background particles due to the idling machinery, the data collected at the 167 source before feeding the material to the running machine were also used as a reference for 168 comparison. 169 Finally, the effect of the machine's feed rate on the release of particles was studied by selecting 170 two samples with different sizes and surface treatments (sample A; LUC-N, and sample B; H80-171 EMOD) and feeding them to the machine at three different rates (FR.1= 7.5, FR.2= 4.1 and 172 FR.3=2.1 g min -1 ). 173

Data analysis 175
All statistical analyses and plotting were conducted using the R programming and statistical 176 computing software (R.Core.Team 2013). Initially, and for each size/surface treatment 177 combination, source PM 2.5 and PNC were corrected for respective background values. 178 Explanatory data analysis with Analysis of Variance (ANOVA) and Tukey's Honest Significant 179 Difference (TukeyHSD) test were used to identify whether the surface treatment and particle 180 primary size were sources of variation in PNC and PM 2.5 . If so, a linear mixed-effects model 181 (LMM) was then fitted using the lme4 package in R (Bates et al. 2013) to examine the 182 relationships between each of PM 2.5 and PNC and the explanatory factor variables (i.e., primary 183 particle size, surface treatment and their interaction). This modeling approach has been 184 extensively used by researchers in different areas of science, due to its capability for the full and 185 simultaneous analysis of multiple random effects (Quené and van den Bergh 2008). More 186 detailed description of the applied method is presented in Appendix 1. 187 A similar approach was used to analyze the feed rate experiment data. The variations in the 188 particle mass and number concentration due to different feed rates, and also due to the interaction 189 of samples type and feed rate, were characterized in the same way as explained above. 190 191

Overall results 195
The effect of starting the milling process on raising the mass and particle number concentration 196 levels (up to two orders of magnitude) was observed at the particle release source for all tested 197 cases. As an example, 10-second averaged time-series of PM 2.5 and PNC for  combination are shown in Fig. 2 (complete averaged time-series are presented in Appendix 2). 199 In addition, the variations in PM 2.5 and PNC for different size/surface treatment combinations 200 are depicted in Fig.3. 201 As demonstrated in Fig.3, there were variations in both mass and number concentration of the 202 released particles due to different sizes and surface treatments. However, it is also clear that the 203 release was influenced by these factors differently, in terms of both trend and the extent of 204 variations. The trends seen are the increases or decreases in PM 2.5 and PNC respectively, for each 205 size between successive surface treatments in Fig.3, from N to ETHO. For example, while PM 2.5 206 and PNC trends for H120 and CLO were similar, the other two samples exhibited totally different 207 trends for PNC compared to PM 2.5 . Another example of these non-identical variations can be seen 208 in PNC graph, where variation in different sizes of the unmodified particles (N surface treatment) 209 is almost 10000 units more than that of MMOD. During all these experiments, the variations of 210 both temperature and relative humidity of the measurements location were very small (in the 12 range of 21.1-23.8 °C for temperature and 50.6-62.3 % for the relative humidity). Therefore, 212 these parameters were considered to be negligible in terms of their effect on the measurements 213 results. 214 The general implication of these findings is that the size and surface treatment of the material 215 can indeed affect the mass and number concentration of the released particles from the jet milling 216 process. Therefore, the following analyses focus on quantifying variations in the release due to 217 different sizes and surface treatments, and the extent to which each particular parameter and their 218 interaction affect particle release. 219 220 3.1.2. Initial analysis; pair-wise comparison 221 Fig. 4 shows the results of the two-way ANOVAs which were run separately for each factor 222 (i.e. size and surface treatment), followed by TukeyHSD analysis. The results of this so-called 223 pair-wise comparison are presented as 95% confidence intervals for the differences in the mean 224 levels of each factor. 225 Fig. 4 confirms that there were significant differences -to different extents -between levels 226 of surface treatment and size in several cases where the 95% confidence intervals did not include 227 zero. It can also be derived from Fig. 4 that "LUC" particles (25 nm) had the lowest means in 228 both PM 2.5 and PNC, and thus contributed the least to the particle release, while among surface 13 treatments, "N" had the highest mean values and therefore, contributed most to the release. Table  230 1 ranks the levels of each factor based on their mean mass and number concentrations. 231 According to Fig. 4, the distributions of the differences in PNC and PM 2.5 mean values are 232 more widely-spread due to different surface treatments rather than different particle sizes, 233 particularly in PNC. This is an indication of higher significance of surface treatment in causing 234 variations in the release of particles from this process. 235 236

Mixed-effects model's results 237
It should be noted that auto-correlation exists in time-series for each test, i.e., each 238 observation is not totally independent of those before and after. An Autoregressive Model of 239 order 1 (Box et al. 2013) was fitted for each size and surface treatment and the residuals were 240 found to be Independent and Identically Distributed (IID) white noise. This indicates that while 241 each time series exhibits auto-correlation, the stochastic processes generating the PNC and PM 2.5 242 measurements are stationary. Therefore, even though there is a relationship between successive 243 values of PNC and PM 2.5 , the mean is well characterized by the time-series. 244 The two separate models, one for PM 2.5 and one for PNC, in which both size and surface 245 treatment were considered as random-effects parameters, were fitted to the observed data. The 246 variance of the residuals (σ 2 ɛ ) and the estimate for the overall mean (β 0 ) are tabulated in Table 2. 248 The model results indicate that surface treatment-to-surface treatment variations made a 249 greater contribution than size-to-size variations, as its standard deviation was greater than that for 250 size. This conclusion is consistent with what was shown earlier in Fig. 4. The 95% prediction 251 intervals on the random-effects for the fitted models are shown in Fig. 5, where the conditional 252 distributions of the random-effects for "size" were confirmed to have less variability compared to 253 the conditional distribution of the random-effects for "surface treatment" for both PM 2.5 and PNC. 254 Yet, it cannot be inferred confidently from the above figure that the conditional distribution of the 255 random effect "size" for a particular level, say "LUC", had more or less variability than the 256 conditional distribution of the random effect "surface treatment" for a particular level, say 257 "MMOD". This is mainly because the number of responses that the conditional distribution of the 258 random effects "size" and "surface treatment" depend on -4 and 6, respectively -are very close. 259 As explained in the Data analysis section, for each measured value (i.e., PNC and PM 2.5 ), two 260 separate null models were fitted to assess the significance of size and surface treatment (one for 261 each). The p-values obtained from the Likelihood Ratio Test (LRT) were considerably smaller 262 than the test level (i.e. the typical value of 0.05). Therefore, the significance of both random-263 effects factors was confirmed. 264 another random-effects parameter. The purpose was to determine if the interaction of size and 266 surface treatment could also significantly affect the release, as each factor did separately. The 267 95% prediction intervals on the random effects for the interaction factor in both the PM 2.5 and 268 PNC models are plotted in Fig. 6. 269 It can be seen in Fig. 6 that zero is within several prediction intervals on the random effects 270 for the interaction of size and surface treatment, which means that in these cases, the prediction 271 intervals are not different from the overall means. Nevertheless, the LRT was performed for the 272 interaction of size and surface treatment, as it was for each of them separately. The test indicated 273 that the interaction affected PM 2.5 ( 545  Fig. 7 gives an overall view of how different sample/feed rate combinations affected the 280 particle release parameters. 281 same way. For example, while the general variation trends for PM 2.5 and PNC were similar across 283 the feed rates for each sample, the extents of these variations were not equal. According to this 284 figure, altering the feed rate of the jet milling machine could be a factor in the release of the 285 particles, but it is not clear whether this was the only factor. Interactions with sample type may 286 also affect the particle release. It was also a matter of interest to determine whether the variations 287 in PM 2.5 and PNC, arising from different feed rates and sample/feed rate interactions, were 288 significant or not. In order to address the abovementioned, firstly, differences in the mean values 289 of PM 2.5 and PNC for each level of the feed rate were determined by the same method that was 290 used for the main experiment (ANOVA followed by TukeyHSD analysis). Fig. 8 shows the result 291 plotted as 95% confidence intervals of the differences in mean. This figure indicates that the 292 difference between the effect of different levels of feed rate on both PM 2.5 and PNC were 293 significant, as zero was not included in any of the 95% CIs in the above plots (except for FR.3 294 and FR.2 in the PM 2.5 of sample "A"). The samples reacted similarly to different feed rates and 295 this was in agreement with the results shown in Fig. 7. By examining the plots for both samples, 296 it can be seen that sample "B" exhibited more variations compared to sample "A", in both mass 297 and number concentration. 298 assess the significance of sample type/feed rate interaction on particles release properties. Fig. 9  301 shows the 95% prediction intervals on the random effects for the sample/feed rate interaction in 302 both PM 2.5 and PNC models. 303 Similar to Fig. 6, it can be seen that most of the prediction intervals include zero, which could 304 imply that the mean values for interaction were not significantly different from the overall means. Pardhasaradhi at AIBN for his technical assistance during the aerosol measurements, as well as 332

Feed rate experiment 279
Ms. Rachael Appleby and Ms. Chantal Labbe for their administrative assistance. 333 Both of our covariates were categorical with 4 and 6 levels for size and surface treatment, 338 respectively. Parameters associated with the levels of the covariates are known as effects of those 339 levels which are either mixed or random effects (Bates 2010). Since we were interested in 340 characterizing the variations in PM 2.5 and PNC arising from the different levels of both 341 covariates, size and surface treatment were considered as random-effects terms. The overall mean 342 of samples were chosen as a fixed-effects parameter so that the random effects could be 343 considered as deviations from the "average" case. 344 The mixed-effects model was considered as below: 345

346
(1) 347 348 Where the subscripts i represents time of measurement, β represents the fixed effect of the 349 overall mean, γ represents the random effect of surface treatment, α represents the random 350 effect of particle size, and δ represents the random effect of their interaction. The errors are 351 assumed normally distributed, ∼ 0, . The residual term denotes the part of the 352 variability that comes from a source other than those specified in the model and therefore, cannot