6 AIMS OF THE STUDY
7.3 FPA-FTIR and data analysis in Publications II and III
Publications II and III utilized automatic FPA-FTIR imaging instead of manual selection and measurement of single particles as in publication I (Figure 7). After the UEPP process, sample was filtered to infrared refractive filter to a circle with 13 mm diameter. Gold-coated polycarbonate filters (Sterlitech Co, pore size 0.8 µm) were used in II and silver membrane filters (Sterlitech Co, pore size 5 µm) in III. The whole 13 mm circle was measured with FPA-FTIR equipped with 128x128 pixel FPA detector (Agilent Cary 670/620). Because the whole sample volume could be filtered, every particle could be analysed. Measurements were preformed from the refractive filters in reflection mode, using 15X cassegrain objective, pixel size 5.5 mm, spectral resolution 8 cm-1, spectral range 3800–800 cm-1 and number of scans 4/8 (publication III/II).
Figure 7. Development of methods for quantitation of microplastics from environmental samples.
The final step in the analytical process is data analysis. FPA-FTIR produces excessive amount of data – in this case, one 13x13 mm square equals 36 GB. The data is a so-called hyperspectral cube, where two dimensions specify the spatial location and third contains spectral information. One dataset of this size consists of 18x18 detector
“tiles”, which each contain 128x128 spectra, resulting in 5.3 million spectra. Spectra are identified as plastic or not by comparing them individually to a reference library composed of spectra of different reference plastic types.
In publications II and III, data was analysed with siMPle, which is a freeware designed for analysing MPs from imaging FTIR data (Primpke et al., 2020b). SiMPle and its older version MP Hunter have been used and validated previously in multiple studies (Liu et al., 2019; Rist et al., 2020; Simon et al., 2018; Vianello et al., 2019).
SiMPle allows relatively fast detection of MPs from large datasets. With 40 reference spectra, data analysis took approximately three to four hours per sample, depending on the performance of the computer used for calculations. The software calculates Pearson’s correlation coefficients between reference and sample spectra. The user can define the weights of raw, first derivative and second derivative of spectra used in the correlation calculations. In general, higher weight of the derivatives usually gives better results, because random noise in the baseline does not affect the derivatives as much as raw spectrum. When correlation coefficients are calculated, pixels in which spectra has higher correlation than the threshold value, for example >60-70% are identified as MPs. Results, such as counts, and sizes of MPs are calculated by composing neighbouring pixels with similar material to a MP particle. If a single
treated in plastic buckets after collection and filters were stored in plastic Petri dishes, though plastic equipment should be avoided.
In Publications II and III, Universal Enzymatic Purification Protocol (UEPP) (Löder et al., 2017) was adapted for the sample pre-treatments. The protocol contains multiple steps, and the choice of steps depends on the composition of samples.
Therefore, before each study the method must be optimized, and recovery tested for the specific sample matrix and type. Sea water samples were collected and analysed for publication II. SDS, H2O2 and chitinase were chosen for the decomposition and dissolution of organic biological polymers. In Publication III, MPs were analysed from GITs (gastrointestinal tracts) of fish, which is a complex sample matrix.
Processing required one more step in the UEPP, resulting in the following reagents:
SDS, protease, chitinase and H2O2.
More QC measures were implemented in publications II and III compared to I. Plastic equipment were avoided as far as possible, all treatments were done in laminar flow cabinets or fume hoods and water and reagents were pre-filtered prior to use. The validity of the analytical methods was verified with three control samples in publication II and 10+3 controls in publication III. In publication II, contamination was determined for the whole pre-treatment and measurement process with the same samples. However, in publication III, ten controls were used for both pre-treatment and measurement, and three for the final filtration and FPA-FTIR analysis.
Recovery tests were performed in publication II for pre-treatment method and in publication III for FPA-FTIR measurement, too.
7.3 FPA-FTIR AND DATA ANALYSIS IN PUBLICATIONS II AND III
Publications II and III utilized automatic FPA-FTIR imaging instead of manual selection and measurement of single particles as in publication I (Figure 7). After the UEPP process, sample was filtered to infrared refractive filter to a circle with 13 mm diameter. Gold-coated polycarbonate filters (Sterlitech Co, pore size 0.8 µm) were used in II and silver membrane filters (Sterlitech Co, pore size 5 µm) in III. The whole 13 mm circle was measured with FPA-FTIR equipped with 128x128 pixel FPA detector (Agilent Cary 670/620). Because the whole sample volume could be filtered, every particle could be analysed. Measurements were preformed from the refractive filters in reflection mode, using 15X cassegrain objective, pixel size 5.5 mm, spectral resolution 8 cm-1, spectral range 3800–800 cm-1 and number of scans 4/8 (publication III/II).
Figure 7. Development of methods for quantitation of microplastics from environmental samples.
The final step in the analytical process is data analysis. FPA-FTIR produces excessive amount of data – in this case, one 13x13 mm square equals 36 GB. The data is a so-called hyperspectral cube, where two dimensions specify the spatial location and third contains spectral information. One dataset of this size consists of 18x18 detector
“tiles”, which each contain 128x128 spectra, resulting in 5.3 million spectra. Spectra are identified as plastic or not by comparing them individually to a reference library composed of spectra of different reference plastic types.
In publications II and III, data was analysed with siMPle, which is a freeware designed for analysing MPs from imaging FTIR data (Primpke et al., 2020b). SiMPle and its older version MP Hunter have been used and validated previously in multiple studies (Liu et al., 2019; Rist et al., 2020; Simon et al., 2018; Vianello et al., 2019).
SiMPle allows relatively fast detection of MPs from large datasets. With 40 reference spectra, data analysis took approximately three to four hours per sample, depending on the performance of the computer used for calculations. The software calculates Pearson’s correlation coefficients between reference and sample spectra. The user can define the weights of raw, first derivative and second derivative of spectra used in the correlation calculations. In general, higher weight of the derivatives usually gives better results, because random noise in the baseline does not affect the derivatives as much as raw spectrum. When correlation coefficients are calculated, pixels in which spectra has higher correlation than the threshold value, for example >60-70% are identified as MPs. Results, such as counts, and sizes of MPs are calculated by composing neighbouring pixels with similar material to a MP particle. If a single
pixel outweighs thresholds for multiple plastic types, it is sorted according to the highest correlation value. Finally, the software produces a spectral map where MPs are marked with respective colours (Figure 8).
Figure 8. Example MP map from siMPle.
Additionally, siMPle measures the longest and smallest dimensions of particles and estimates the masses of particles. The dimensions can be used for categorizing particles by morphology as fragments or fibers (Vianello et al., 2019). Besides, volume of a particle is calculated from the dimensions with an assumption that particle is an ellipsoid, whose thickness depends on the longest dimension (Liu et al., 2019; Simon et al., 2018). The mass estimation is then calculated from the estimated volume and the density of the identified material.
Reference spectra were chosen for publications II and III based on previous studies from the Baltic Sea and Lake Kallavesi, mainly publication I. Only the most common plastic types were included, because other were very rare in publication I. Moreover, common natural polymers such as carbohydrate and protein materials were included to prevent mismatching of MPs and natural particles. Hence, the following plastic types were analysed: PE, PP, PET, acrylonitrile butadiene styrene (ABS), polystyrene (PS), polyacrylonitrile (PAN), polyamide (PA), polymethyl methacrylate (PMMA), polyurethane (PU), and polyvinyl chloride (PVC).
Before starting the library fitting, the thresholds of correlation coefficients for categorizing a particle as MP have to be established (Liu et al., 2019). Both composing a reference library and setting thresholds need tests with reference MPs with known material and particle size. Reference samples are measured, and data is analysed with different library spectra to acquire highest correlation with known samples.
Spectral data is manually examined and the thresholds for identifying MPs are set based on the interpretation of spectra. In this thesis, self-made MPs (size range approximately 300–500 µm) were used for preliminary tests before analysing the samples in publications II and III. NIST (National Institute of Standards and Technology) traceable size standard PS beads were used for the final recovery tests in publications II and III, and fluorescent PS beads in publication III. Based on the tests and manual examination of data, correlations >60% were considered as reliable identification for other materials than ABS that was easily misidentified as PS and therefore checked manually every time when present.
From particle numbers, polymer types, masses and sizes, SiMPle produces tables, which were used in further processing of the data. Statistical analyses and plotting of graphs were performed with Microsoft Excel and RStudio (R Core Team, 2019) in publications II and III.
pixel outweighs thresholds for multiple plastic types, it is sorted according to the highest correlation value. Finally, the software produces a spectral map where MPs are marked with respective colours (Figure 8).
Figure 8. Example MP map from siMPle.
Additionally, siMPle measures the longest and smallest dimensions of particles and estimates the masses of particles. The dimensions can be used for categorizing particles by morphology as fragments or fibers (Vianello et al., 2019). Besides, volume of a particle is calculated from the dimensions with an assumption that particle is an ellipsoid, whose thickness depends on the longest dimension (Liu et al., 2019; Simon et al., 2018). The mass estimation is then calculated from the estimated volume and the density of the identified material.
Reference spectra were chosen for publications II and III based on previous studies from the Baltic Sea and Lake Kallavesi, mainly publication I. Only the most common plastic types were included, because other were very rare in publication I. Moreover, common natural polymers such as carbohydrate and protein materials were included to prevent mismatching of MPs and natural particles. Hence, the following plastic types were analysed: PE, PP, PET, acrylonitrile butadiene styrene (ABS), polystyrene (PS), polyacrylonitrile (PAN), polyamide (PA), polymethyl methacrylate (PMMA), polyurethane (PU), and polyvinyl chloride (PVC).
Before starting the library fitting, the thresholds of correlation coefficients for categorizing a particle as MP have to be established (Liu et al., 2019). Both composing a reference library and setting thresholds need tests with reference MPs with known material and particle size. Reference samples are measured, and data is analysed with different library spectra to acquire highest correlation with known samples.
Spectral data is manually examined and the thresholds for identifying MPs are set based on the interpretation of spectra. In this thesis, self-made MPs (size range approximately 300–500 µm) were used for preliminary tests before analysing the samples in publications II and III. NIST (National Institute of Standards and Technology) traceable size standard PS beads were used for the final recovery tests in publications II and III, and fluorescent PS beads in publication III. Based on the tests and manual examination of data, correlations >60% were considered as reliable identification for other materials than ABS that was easily misidentified as PS and therefore checked manually every time when present.
From particle numbers, polymer types, masses and sizes, SiMPle produces tables, which were used in further processing of the data. Statistical analyses and plotting of graphs were performed with Microsoft Excel and RStudio (R Core Team, 2019) in publications II and III.