Investigating microplastics and nanoplastics released from food bag ziplock using SEM and Raman imaging

Microplastic contamination is a concern in our daily lives, such as being released from self-sealing ziplock (sliderless zipper) plastic bags that are commonly used for food storage. That is because during the closure and opening process, due to friction and deformation, the male rim inserting into or separating from the female rim can release debris as micro- and nanoplastics (MNP). Herein, we initially observed the released debris using scanning electron microscopy (SEM). Subsequently, Raman imaging was employed to directly visualise the debris, either scratched on the rim surface or fallen down from the rim, from molecular spectrum perspective. Raman imaging analyses MNP from hundreds to thousands of spectra rather than from a single spectrum or peak, enhancing the signal-to-noise ratio statistically and providing morphological information for quantification. The confocal Raman-based mapping of MNP may be susceptible to be false images, which can be improved through terrain 3D mapping. Additionally, the weak signal of nanoplastics can be enhanced by reducing scanning pixel size and deconvoluting with surface-fitting algorithm. Consequently, we estimated that approximately 5(±3) MNP per millimetre along the ziplock length may be released during each closure/opening process. Given the use of these plastic bags for food storage, this level of contamination is concerning, warranting careful risk assessment alongside other potential MNP sources of plastic items used in our kitchens. Overall, Raman imaging can be effectively analyse MNP and more broadly nanomaterials, with help of algorithms and SEM.


Introduction
Self-sealing ziplock (no slider) bags are commonly used in our daily lives.The zipper-like plastic fastener allows an easy closing (and opening) of the bags for a tight sealing.The sealing offers protection from moisture, heat and other atmospheric contaminants, which makes it a preferred choice for package, storage and transportation of food and household items [1,2].Its market was estimated at US$ ∼21 billion in 2021-2022, and likely reach US$ ∼25 billion by 2032 [3].While we appreciate the benefits of using those products, we are also wondering about the potential environmental contamination, such as from the emerging contaminant of microplastics point of view, which is the main objective of this study [4].
Plastics are ubiquitous in our daily lives because of convenience, durability and functionality [5][6][7][8].Their widespread use has raised concerns about their impact on environment including water, soil, air and almost all organisms [9][10][11][12].Efforts are underway to reduce plastic waste, by increasing recycling rates and developing more sustainable alternatives to the traditional plastics, such as bioplastics [13,14].On the other hand, once the big plastic items break down to small sizes as microplastics (<5 mm) that cannot be directly and easily seen by our naked eyes, the contamination might get much worse [10,15,16].More seriously, we might have been exposed to this emerging contamination in our daily lives [17][18][19].For example, due to the wear/tear and continuous use of the self-sealing plastic ziplock bag, the plastics can be broken down and release some debris as microplastics.This is because the ziplock typically consists of two interlocking plastic tracks or rims with protruding teeth, usually a male rim and a female rim.During the closing and opening processes, the friction between the male and female rims might release some debris that can be microplastics or even nanoplastics (<1000 nm), which is tested herein.
However, the test on microplastics and nanoplastics (MNP) can be a challenge due to small size, complicated components, weak signal and high background [6,15,[20][21][22].For environmental sample, for example, MNP can be similar in morphology to other environmental particles, such as sand or organic matters, meaning a high background.Molecular spectra are usually employed to identify MNP, including Fourier-transform infrared (FTIR), Raman, fluorescence and vis-UV [15,22].Among them, Raman has received increasing attention due to its non-destructive detection, no interference from water, spectrum specificity, high spatial resolution originating from the excitation laser, and versatility [20,21,23,24].The limitations include the fluorescent interference of background, weak signal and expensive setup [20,23,[25][26][27].
Once the laser scans the sample surface by combining Raman spectroscopy with microscopy (as micro-Raman), the scattered signal is collected at specific position and then mapped as a pixel for imaging.The scan thus generates an array of spectra as a hyperspectral matrix that contains hundreds to thousands of spectra, also termed as hyper spectrum [16,28,29].The image analysis can statistically enhance the sensitivity when compared to the traditional single-spectrum analysis that works as only a pixel in the image.That is, for imaging analysis, each pixel is mapped with a single spectrum; a hyperspectral matrix containing hundred-to-thousand spectra (depending on image/map resolution) can statistically and significantly increase the signal-to-noise ratio.The mapped image can also directly visualise the sample from the molecular spectrum point of view, or the chemical window/channel, with the identification capacity to provide the distribution and size/shape information as well [15].However, the accompanied challenge is the big-data (hundred-to-thousand spectra) analysis of the scanning hyperspectral matrix, the dataset size of which can be megabytes to gigabytes.The setup is also generally expensive and the test process is time-consuming.Although the hyperspectral matrix analysis has been well demonstrated in satellite mapping and medical application [30][31][32], the size-down application such as for MNP via Raman is still at the early stage.
In this study, we employ Raman imaging to capture the MNP released from self-sealing ziplock, to address the concern that the ziplock might directly contaminate our food in our daily lives.We advance the Raman imaging to overcome several possible drawbacks [33][34][35], by combining the Raman imaging with scanning electron microscope (SEM) to increase the resolution [29], using a 3D terrain map to guide the imaging and to distinguish the scratches on the male/female rim from the fallen down debris, deconvoluting the weak signal of nanoplastics to increase the certainty, and developing statistical analysis on the debris counting to increase the representativeness etc.The results will not only provide useful information about the potential source of MNP in our daily lives, but are also helpful for the further application of Raman imaging, hyperspectral matrix analysis and nanomaterials research.

Chemicals and sample preparation
All chemicals including ethanol and acetone were purchased from Sigma-Aldrich (Australia) and used as received.Super-pure water (>18 MΩ•cm or Milli Q water) was used for the analysis.Three different brands of ziplock bags were purchased from a local market (Woolworth, Australia), and shown in figure S1 (Supporting Information).
Three brands (Armada, Hercules and Ziploc) of new ziplock bag were tested first as control/reference (to help the assignment of the released debris), after being cleaned with ethanol, acetone and Milli Q water.Each test was duplicated or scanned to increase the results' representativeness.
To mimic the closure/open process of the ziplock, we intentionally close and open the plastic sliding fastener in the bag for 10 times, using fingers to press the ziplock to close, and tear to open the bag (figure S2, Supporting Information).During this process (duplicated too), a cleaned glass slide (with ethanol, acetone and Milli Q water) was localised on the bottom (with a vertical distance of ∼1 cm) to collect the fallen down debris.After that, we further collected the debris from the ziplock by gently touching the glass slide to the female/male rims (figure S2, Supporting Information).The nearly-dried but still wetted glass surface with Milli Q water is helpful to collect the debris.Finally, the rims was slightly washed by ethanol (∼1 ml) and the liquid was deposited onto the slide surface for drying.In this case, we pretended the glass slide to be our food materials to test the possible contamination arising from the debris released from the ziplock bag.This simplified mimicking process might be different from the real-time situation, and we did not survey all the ziplock bags on the market.The result variation can be thus expected.
The sample collected on the glass slide can be directly tested by Raman.For the SEM test, the ziplock was cut using scissors and knife, and fixed onto the SEM holder using a carbon tape.The sample was sputter-coated with a thin layer of platinum (∼10 nm) for the SEM test.We also touch the rim to the carbon tape surface to directly collect the debris for the SEM test.However, the strong sticking interaction of the carbon tape might peel off the debris from the rim surface, rather than falling down naturally, which should be pointed out.Furthermore, the debris on carbon tape surface is difficult to be tested by Raman, because the Raman background from the carbon tape is high and shield the plastic signal.We intend to test the sample on the glass surface using Raman to identify them, while sample on carbon tape surface using SEM to visualise/count them, as reported below.
There are many brands of zip bag on the market.For food-related applications, usually the material is polyethylene (PE), including the bag and the zipper, as shown in figure S1 (Supporting Information).The zipper also has many types of configurations, such as size and multi-layer/set of zipper as well, as shown in figure S1 (Supporting Information).All of these can lead to the variation on the test results.Herein the aim is to capture the potential MNP released from our daily lives, to confirm the potential source in our kitchens.

Testing protocols and data analysis algorithms
The testing protocols were adapted the previous report [36].In brief, Raman spectra were recorded using a confocal Raman microscope (DXRxi / ThermoFisher, USA) equipped with a 532 nm laser (<30 mW).A chargecoupled device (CCD) detector was cooled at −60 °C to collect Stokes Raman signals under an objective lens (10×, or others such as 20×, 50×, 100×) at room temperature (∼24 °C).Generally, the laser power was 9 mW, integration time was 0.1 s and the scan number was 1.The scanning pixel size was adjusted from 5 μm × 5 μm to 0.1 μm × 0.1 μm, to scan an area from 200 μm × 200 μm to 2 μm × 2 μm, to collect the spectrum array from 40 × 40 to 20 × 20 at each specific physical position (x-/yaxis) as the hyperspectral matrix, respectively.The collected spectrum of sample was compared with the standard spectrum of the virgin plastics.
Most of the scans were collected on the x-/y-axis plane at a fixed/focused z-axis.The target has a 3D size (x-/y-/zaxis) so the fixed z-axis might generate a false image that will be discussed later.In this case, terrain 3D map (x-/y-/zaxis) is recommended to guide the Raman scanning.The terrain map was previously generated by changing the z-axis step-by-step to collect the snapshots under the light illumination (not laser).The different wavelength between the illumination light and the laser should be noted, which might lead to the not-exact terrain map to match the tiny or nanostructure via Raman, even after alignment.Similarly, the different imaging approaches among photo, Raman and SEM can also lead to the possible position-mismatch, as discussed below.
To map the image via Raman, the excitation laser was scanned on the sample surface to collect the signal at each point or pixel (x/y), as a hyper spectrum or a hyperspectral matrix.The spectrum at each point was converted to red/green/blue (RGB) value as (pseudo)-z-axis and mapped as a pixel to generate an image (2D) via algorithm.A previous study has successfully demonstrated the capability of Raman imaging to analyse polystyrene nanoplastics, down to 100 nm [36].The methodology is validated and advanced herein to investigate the debris, particularly for those that are smaller than 1 μm.

Image deconvolution
For the image re-construction via deconvolution, the image surface fitting was performed using OriginPro (2022) software.After importing, the image data of 2D was fitted with a 2D Gaussian surface using optimised parameters like Levenberg-Marquardt, 1000 × 1000 points, 50-200 replica, local maximum, positive peak, 1% minimum peak height, 100-400 iterations etc.

Particle analysis
For particle analysis, ImageJ software was employed.Using ImageJ (1.53k) enables the recognition of individual particles in complex aggregates or clusters, leading to an effective particle estimation [37].After an image was opened in the software, the image background was subtracted using a suitable value (to clearly present and distinguish the to-be-targeted particles), and the random noise was removed by filter such as Gaussian blur.A colour threshold was adjusted to make almost every to-be-targeted particle identifiable; the image was then made binary, followed by filling hole and watershed, and lastly the implementation of the particle analysis function.The outlines of each particle can be extracted, and the particle area can be converted to size by rootsquaring for statistical analysis, aided by the OriginPro software.

Quality assurance and quality control
Before Raman test, the confocal setup was calibrated using a tools box provided by the setup, to correct the wavenumber and alignment.For each test, a blank control was conducted in parallel with the samples in a fume hood to experience the similar sample preparation process side-by-side.Only cotton lab coat was allowed, and no gloves were used.The samples were then transferred to Raman and SEM tests, along with the control.During the transferring process (from laboratory to laboratory), all samples were stored in a cleaned glass container covered with aluminium foil.No significant amount of particle was observed from the control as presented herein (<2 versus ∼49 for ∼30 tests, see below) [38].
For all Raman spectra (in the scanning matrix) and images, at least three typical spectra are presented, including a relatively strong one (Scan #1), a middle one (Scan #2) and a blank one (Scan #3), to cross-check with each other.In this case, the blank one can be used as an internal 'spectrum background control', which can be further mapped as an internal 'image background control'.Furthermore, Raman images were cross-checked with SEM images as well, by aligning the same sample's approximate same physical positions for both tests.

Results and discussion
3.1.SEM Figure 1 shows the SEM images collected from the ziplock.For comparison, (a), (b) are collected from the brand-new one, there is neither scratch nor deformation on the rim, no matter on the female rim (a) or the male one (b).After ten times of closure and open, the deformation and fractures can be observed on the female rim surface in (c), (d), and the scratch and debris on the male rim in (e)-(h).
If we have a close look, the image in figure 1(e) shows a scratch along the boundary, which might be due to the ziplock's closure and open practices leading to friction.The scratch can be clearly observed in (f) to generate debris.Once the SEM images are collected from the side of the rim, (g), (h) are obtained.The debris and scratch can be observed in (g) and zoomed in as (h).The scratch along the rim's side-boundary can be clearly visualised, echoing (e) (a top view).
The images in figures 1(i)-(l) show the morphologies of the fallen down debris.We touch the ziplock rim onto carbon tape surface to better transport the debris for analysis, as shown in (i), (j).The debris in (i) might be peeled off from the rim surface so that look like fibres.Even on the glass surface, some debris are also collected and presented in (k)-(l).There is no significant difference.However, for Raman analysis, the carbon tape yields a high background so that the sample on the glass surface should be focused, as discussed above.
In the following sections, using Raman, we will confirm whether or not these debris are plastics, or MNP, from the rim surface and the fallen down debris.We focus the test on the male rim, including the top view and  The average spectrum has less background noise or variation, suggesting the improvement of analysis sensitivity by the scanning hyperspectral matrix.The appearance of the characteristic peaks of PE (marked with dashed lines) leads us to assign the main materials in the scan area (from the average spectrum) and at the marked positions (Scans #1-#3) to PE.The non-flat baseline of the spectrum Scan #1 might be due to the co-formulated ingredient in the zipper rim.
To directly visualise the plastic of PE, we can map the characteristic peaks as images.Before that, we first map a blank window that is also marked with a dashed line in figure 2 Figure 2(d) is another version of (j).The 3D image can provide more details and the two particles are well patterned on the top part.However, herewith no matter the patterned particles on the top or the scratch in the middle part, they are seemingly attached on to the rim 'mother' surface as embryos of MNP.Therefore, these Raman images should get support from the SEM image in figure 1(f).The SEM image can suggest the independent particles or scratch to release independent 'daughter' MNP.We thus should combine the information from both to benefit each other.
Since images in figures 2(e)-(j) are similar, we select the strongest peak's images (d), (j) for the following analysis, the peak of which is arrowed in (b).However, once the assignment is of low certainty, we should come back to map multi-peaks to cross-check the patterns, to further confirm the assignment with an increased certainty or signal-to-noise ratio.

Zoom in
The images in figure 2 were generated from the scanning hyperspectral matrix that was collected by focusing the laser on the ziplock's rim surface.During the scanning process, the z-axis was fixed while the x-/y-axis was changed to collect the spectrum at each position as an array.The zipper and the scratch have the 3D structure of the x-/y-/z-axis, the fixed z-axis for scanning can generate a false image in potential.That is because the confocal Raman can collect signal effectively from the focused plane, at the fixed z-axis.Off-focal plane has the declined contribution significantly.To overcome this possible bias image, a terrain map of the x-/y-/z-axis (3D) should be collected previously to guide the subsequent Raman imaging, the results are shown in figure 3.  S3 (Supporting Information).The scratched particle in the middle part is mapped along with several small particles, suggesting the improvement using the terrain map for imaging.However, the 'released' particles are still localised onto the rim surface.In this case, the Raman image has difficulty to distinguish them because the background 'mother' yields a strong signal that can easily shield the weak signal of particles or 'daughters'.While the terrain mapping can overcome this issue to some degree, the fallen down particles should be tested, as presented below.

Side view
Before we test the fallen down particles, in this section we test the side view of the ziplock rim to see if any difference.The results are presented in figure 4. The SEM image is provided in figure 1(g), at the approximate position due to the difficulty in aligning the sample at the same position for Raman and for SEM, which have been conducted at two different setups independently.
Using the terrain map to guide the Raman imaging, the photo map is overlapped with the Raman image in figure 4(a), where the rainbow colour of Raman image is presented.Figure 4(b) shows the typical spectra, and the collection positions are marked in figure 4(d).Basically we can assign them to PE by comparing with the standard PE spectrum.The Raman images are shown in figures 4(c)-(f), via gradually zooming in by shrinking the scanning pixel size (to be discussed below, in Part 3.2.4).The layered structures in figures 4(c), (d) might be related to the rims closure/open positions or inserting degree of male into the female rim, which needs more research.Once zoomed in, the tiny structures or particles in figures 4(e), (f) are either the particle embryos that still attach onto the rim mother surface or independent daughter particles, as suggested by the SEM images in figures 1(g), (h).Again, the combination of two analytical instruments, SEM and Raman is recommended, the former can clearly visualises the detailed structures, while the latter can identify the plastics via molecular spectrum.

Fallen down particles and image deconvolution
In this section, we test the particles fallen down onto the glass slide surface.For this kind of sample prepared on the flat glass surface, the terrain map is not helpful, unless the particle size along the z-axis is big, such as beyond the spatial resolution of the laser spot (∼1300 nm if defined by 2λ/NA 2 , λ is the wavelength of laser at 532 nm, NA is the numerical aperture of the objective lens 0.9 for 100×) [39]., more details and an extra particle can be mapped.This extra particle is further scanned and zoomed in as figure 5(f), where we can see a nanoplastic is successfully mapped [33].Although the size of the pattern in (f) is 800-1000 nm (diameter), this pattern size ideally is the size of a nanoplastic (500-700 nm) plus the size of a laser spot (∼300 nm) (λ/2NA, to be discussed below, In figure 6) [39].The mapped image in figure 5(f) looks broken, which is either due to the laser-burn of the sample or the variation in the collected signal.To average the varied signal and smoothen the pattern, we can deconvolute the image.Herein we select a Gaussian surface to fit the pattern in figure 5(f).The reason to select the Gauss surface is because the laser spot has a size due to the diffraction, and the energy density distribution within the laser spot follows a 2D Gaussian surface.Ideally, the emitted Raman signal or intensity should also follow a similar distribution, a 2D Gaussian surface.The deconvoluted image is presented in figure 5(g), the broken pattern has been integrated and the bumpy surface has been smoothened, suggesting the benefit of the fitting or the deconvolution.Figure 5  ), the increased signal intensity is obvious.The shrink pixel size (l) means that a nanoplastic (dia) can be illuminated by laser for a longer time (t × πdia 2 /4l 2 , t is the integration time for each scanning pixel, taking a nanoplastic as a cycle with an area of πdia 2 /4 and a pixel as a square with an area of l 2 ) with a better excitation.The spectrum (figure 5(f)) looks weakened, which might be due to the laser-burn, as mentioned above, supported by the broken pattern in figure 5(f).Also as presented above, each image is mapped from a hyperspectral matrix the contain hundred-tothousand of spectra.The spectra can be averaged and shown in figure 6 5(f), the nanoplastic can be better illuminated (with energy density of 0.9 mWs / 0.4 μm × 0.4 μm, and 0.9 mWs / 0.1 μm × 0.1 μm, respectively) [33][34][35].Consequently, the signal gets stronger, as evidenced in figures 6(a), (b).Accordingly, a nanoplastic can be well patterned in figures 5(e), (f), and the pattern size (green) is the sum of the nanoplastic and the laser in figure 6(c).
In figure 5(f)/6(c), the pixel size is 0.1 μm × 0.1 μm, smaller than the recommended scan resolution of laser, the full width at half maximum (FWHM) of the Gaussian peak (∼300 nm, if defined by λ/2NA) [39].As shown in figure 6(c), by doing so, the centroid of the laser spot (via pixel of figure 5(f)) can be employed to excite the Raman scattering, the collected signal can be effectively patterned via pixel [34,35].Once fitting towards deconvolution, the location of the nanoplastic can be well predicted from the Gaussian 2D surface's centroid as well, as shown in figures 5(g), (h), to realise the super-resolution image [40][41][42].More research is needed here.

Particle estimation
In this section, we count the number of particles.We take the image in figure 5(a) as an example, because our test suggests that most of the particles can be assigned to the PE MNP, depending on their size.Figure 7(a) is an overview version after we correct the background.The line on the left top is a pattern that suggests the zip rim's direction or orientation, which is due to the scratch of the ziplock rim onto the glass surface.Figure 7(b) marks in red all the possible particles, including the pattern line.After being made binary, figure 7(c) highlights the red areas in figure 7(b).By carefully selecting the analysis parameters, particularly the circularity to the range of 0.75-1.0, the line pattern can be excluded and outlines of the particles are presented in figure 7(d).We then convert the areas of the particles to sizes by square-rooting, because we can roughly assume the particles are squares.In figure 7(e), we can see that along the zip rim, ∼49 particles/ 0.9 mm 2 can be estimated.If we assume the distribution of the particles is uniform, we can estimate that ∼50 particles/mm (along the fasterner line or the rim) from ∼10 times zipping closure/open, or ∼5 (±3, for ∼ 10 times tests) particles/mm from each zipping process.This result agrees with previous reports that MNP can be released from different types of food packaging, for which we should pay more attention [4,[43][44][45].
The resolution of the photo image in figure 7(a) is much lower than that of SEM images in figure 1, which restricts the effective analysis on nanoplastics.Taking a typical SEM image of figure 1(j) as an example, the similar particle analysis is demonstrated in figures 7(f)-(i).Around 22 particles can be counted, with ∼8 ones smaller than 1 μm that can be categorised as nanoplastics.This amount (∼22) is high, if the test area of ∼340 μm 2 (∼17 μm× ∼20 μm) is expanded to 1 mm 2 (∼65,000), and if the particle distribution is uniform.However, in figure 1, the uniform distribution is not observed.Anyway, the nanoplastics can be confirmed.The possible adverse effect of nanoplastics is unclear yet, but we should be cautious since the risk assessment is yet conducted and these nanoplastics can potentially contaminate our food [4,[46][47][48].
Furthermore, the results herein are just typical ones from multi-tests (∼10 tests for each brand).That is, this estimation has limitations and variations [49,50] including (i) a mimicked zip closure/open might be different from the real situation because no food is applied in the test.The temperature effect is not considered as well, such the frozen/heated food/bag might lead to the varied results.We also use a glass slide to collect the released debris, which differs from the real application too; (ii) the types and quality of ziplock materials can lead to variations.Some ziplock might have two or three zipper linings to complete the zipping; (iii) the closure/open skills (such as pressing/tearing forces or directions) are expected to yield different results, along with the age of the bag.In the latter case, it might be difficult to accurately estimate the release amount for each closure/open, because the aged or used one might accumulate the debris generated by all previous actions.This is the reason why we test ∼10 times closure/open actions to average the results.Perhaps we should not reuse them but that needs more research; (iv) the estimation in figure 7 also has some variations and depends on the colour threshold and many parameters.(v) some nanoparticles are not counted quantitatively due to the resolution issue, as discussed.One more reason is that the embryos in figure 1 might not all fall down nor be collected on the glass surface.(vi) it is also difficult to estimate the daily amount of MNP that human are exposed to, because the source is not clear yet and the ziplock herein is just an example of the kitchen items that can potentially release MNP [4].
The toxicity of MNP is still not clear at this moment, particularly for nanoplastics where the characterisation is generally more difficult than that of microplastics [48,51].While many research are focused on the environment contamination such as from food chain and even from air [18,52], there is rare report about the kitchen items that can contaminate our food directly without the involvement of environment [43].Although the current toxicity research commonly employs synthesised MNP that might be different from the real sample [53], MNP has been detected in our bodies including blood [54][55][56].We thus need to be cautious to use plastics items in our kitchen, particularly when they can contaminate our food or drink, by potentially releasing MNP.More research is needed to identify the possible source of MNP in our kitchen and more broadly in our daily lives toward toxicity.

Conclusion
Ziplock can release MNP, as confirmed here.Because we daily use this kind of bag to store or transport our food, the possibility for this emerging contaminant to present in our food is of paramount concern.From plastics contamination perspective, we recommend glass containers.However, the rubber O-ring used for sealing might be another concern, for which we need more test.
The characterisation of MNP is still very challenging.The reported approach herein particularly the image deconvolution is helpful to handle the weak signal of nanoplastics, especially for environmental sample that usually has a strong background.Also from the Raman signal perspective, small scanning pixel should be used to capture nanoplastics.From image resolution point of view, SEM should be employed.Their combination can provide robust results for MNP, and more broadly for nanomaterials.More research is needed to validate those approaches.

Figure 1 .
Figure 1.SEM images collected from a new ziplock (a), (b) and the used ones (rests) after 10 times of closure/open practices.(c), (d) show the female rim with deformation, fracture, scratch and debris.(e)-(h) present the male rim.(g), (h) are side views while the rest are top views.(i)-(l) are the debris fallen down onto the carbon tape (i), (j) or the glass surface (k), (l).The approximate test positions for the following figures are suggested on the images' bottom.

3. 2 .
Raman imaging MNP 3.2.1.Scratch on the male rim, top view Figure 2(a) shows the ziplock image under microscopy for the Raman test.The squared area is scanned and overlapped with the Raman image shown in (j) (to be discussed later).The typical Raman spectra among the 1600 spectra (40 × 40) are listed in (b), to compare with the PE standard spectrum.There are 3 typical spectra in figure 2(b), the collection positions are marked in (j), including a relatively intensity-strong spectrum of Scan #1, a middle one of Scan #2, and a weak one of Scan #3.The area in (a) was scanned and 1600 (40 × 40) spectra have been collected as a hyperspectral matrix, their average one is also listed in figure 2(b) for comparison.
(b), around 2000 cm −1 where PE has no signal.Image figure 2(c) is thus generated, only random noise can be mapped so that this image can work as an internal reference to visualise the image background.Once PE' peaks are mapped, images in figures 2(e)-(j) are patterned.Basically they look similar, some strong and some weak, which is related to their intrinsic peak intensities as marked in figure 2(b).The similar patterns in figures 2(e)-(j) can confirm the presence of PE from different peaks, or different chemical windows/channels, particularly the two particles on the top that can be assigned as PE MNP.The SEM image is shown in figure 1(f) for comparison.

Figure 2 .
Figure 2. Photo image (a), typical Raman spectra (b) and Raman images (c)-(j) of the scratch on the male rim.The area squared in (a) of 200 μm × 200 μm was scanned.Raman spectra were collected under an objective lens of 10×, integration time of 0.1 s for each pixel of 5 μm × 5 μm (to create a matrix of 40 × 40), under a laser intensity of 9 mW.(b) shows the Raman spectra of PE to compare with 3 typical scanning spectra collected from different positions marked in (j), and their average spectrum of 1600 (40 × 40) spectra.The intensity images (c)-(j) are mapped at peak positions shown in (b) and marked under each image, including a blank wavenumber window (c), the characteristic peaks of PE (e)-(j), after 10% colour off-setting.(d) is another version of (j), using a 3D presentation and white background.

Figure 3 (
a) shows the terrain map overlapped by the Raman image.The typical Raman image in figure 3(b) is mapped via the strongest peak at ∼2910 cm −1 , as discussed above.The 3D version is shown in figure 3(c), with more details presented.The rest images are provided in figure

Figure 5 (
a) squares the area to scan, the generated Raman image is presented in figure 5(b).Only several particles are successfully mapped.The area squared in figure 5(b) is zoomed in and scanned as figures 5(c), (d),

Figure 3 .
Figure 3. Photo image with terrain map (a) and the typical Raman images (b), (c).A terrain map was previously collected to guide the Raman imaging.(c) is another version of (b), using a 3D presentation and white background.The area squared in (a) of 120 μm × 120 μm was scanned.Raman spectra were collected under an objective lens of 20×, integration time of 0.1 s for each pixel of 3 μm × 3 μm (to create a matrix of 40 × 40), under a laser intensity of 9 mW.

Figure 4 .
Figure 4. Photo images with terrain map (a), typical Raman spectra (b) and images (c)-(f) under different magnifications (10×, 20×, 50×).In (b), the main characteristic peaks of PE are suggested but only the strong peak at ∼2910 cm −1 is mapped here.(d) is another version of (c).In (c)-(f), the terrain maps were previously collected to guide the imaging.
(h) merges images in figure 5(f), (g), to better represent the difference and the improvement, from different viewing angle and with different colour off-set.More parameter for the image deconvolution is provided in figure S4 (Supporting Information).The reason for the improvement from figures 5(b) to (d)-(f) is because of the shrinking size of the scanning pixel.Once shrink, the collected signal intensity can be increased, as suggested in figures 6(a), (b).

Figure 6 (
a) lists the typical strong signal on each scan area that we presented above.In figure 6(a), from the spectrum (figure 5(b)) to (figure 5(d)) to (figure 5(e) (b), where the spectrum (figure 5(b)) is dominated by the glass background and the PE signal is weak.However, from (figure 5(b)) to (figures 5(d)) to ((e), (f)), the PE signal increased gradually, particularly at the peak of ∼2910 cm −1 that has been employed to map the above images, due to the shrinking size of the scanning pixel too.Taking the pattern in figure 5(g) as an image background, figure 6(c) schematically shows the effect of the scanning pixel size.Roughly, the pixel used in figure 5(d) (marked as square 'figure 5(d)' in figure 6(c)) is too big to localise the nanoplastic, and the 0.1 s (t) of 9 mW laser (P) is scanned and uniformly distributed in this big area (with energy density of tP/l 2 , 0.1 s × 9 mW / 15 μm × 15 μm, if taking the nanoplastic area of πdia 2 /4 as a constant).Consequently, the excited and emitted Raman signal is relatively weak, which is reason why this particle is not patterned in figures 5(b), (d).Once the scanning pixel shrinks from 15 μm × 15 μm in figures 5(b) and 5 μm × 5 μm in figure 5(c), to 0.4 μm × 0.4 μm in figure 5(e) and 0.1 μm × 0.1 μm in figure

Figure 6 .
Figure 6.Effect of scanning pixel size on the typical Raman spectra (a) and average spectra (b) of the whole areas scanned in figures 5(b), (d)-(f), as marked.(c) is a schematical drawing.The size (l) of pixel square, nanoplastic (blue cycle with a diameter or dia) and laser spot (red cycle with full width at half maximum or FWHM) are roughly estimated on the background of Raman image mapped in figure 5(g).

Figure 7 .
Figure 7. Particle analysis with help of ImageJ.(a) is the photo image shown in figure 5(a) after background correction, towards analysis in (b)-(e).Similarly, (f)-(i) analyse a SEM image in figure 1(j), but with different colour to mark the analysis process.After colour-thresold adjusted, (b), (f) mark the to-be-analysed particles by red.(c), (g) are the binary image and (d), (h) present the outlines of the particles.(e), (i) are the statistic analysis.