A rapid-screening approach to detect and quantify microplastics based on fluorescent tagging with Nile Red

A new approach is presented for analysis of microplastics in environmental samples, based on selective fluorescent staining using Nile Red (NR), followed by density-based extraction and filtration. The dye adsorbs onto plastic surfaces and renders them fluorescent when irradiated with blue light. Fluorescence emission is detected using simple photography through an orange filter. Image-analysis allows fluorescent particles to be identified and counted. Magnified images can be recorded and tiled to cover the whole filter area, allowing particles down to a few micrometres to be detected. The solvatochromic nature of Nile Red also offers the possibility of plastic categorisation based on surface polarity characteristics of identified particles. This article details the development of this staining method and its initial cross-validation by comparison with infrared (IR) microscopy. Microplastics of different sizes could be detected and counted in marine sediment samples. The fluorescence staining identified the same particles as those found by scanning a filter area with IR-microscopy.

Raman spectra of small fluorescent dots extracted from marine sediments WITHOUT density floatation step. 5 Fluorescence spectra of Nile Red in solvents of different polarity and photographs of polymers of similar chemical structure to the solvents showing the correspondence of the solvatochromic shift. 6 Enlarged IR spectra from the FT-IR microscopy 7 Fluorescence images of algae and microplastics stained with NR and comments about other possible false positives. 8 Example g-code listing for X-Y motion of milling machine 1

Design and construction of the automated imaging rig
This was based on a Sanven CNC3020T three-axis micro-milling/engraving machine. Similar systems are available from many online retailers and it is a very cost-effective system for automated x,y,z motion control with reasonable positional accuracy (claimed repeatability of 10 µm). Our system was shipped with a test version of Mach 3 machine control software, but a full version was purchased from Arcsoft to run the instrument.
The mill was modified by removing the milling spindle motor (cylindrical) and replacing it in the clamp with an aluminium cylinder of the same diameter. At the top of this, two studs and wing-nuts were inserted to allow a PVC frame to be clamped in place. The frame was machined to clamp the head of a Motic trinocular microscope firmly in place. This is best explained by reference to photograph SI2. A commercial photomicroscope adaptor was used to connect the imaging port of the trinocular head to the Canon EOS 600D camera. The eyepieces were blanked off with black caps to minimise stray light.
Note: While this optical arrangement works effectively and provides a useful zoom range for this work, a lighter and more elegant solution would be to use an appropriate microscopic imaging lens connected directly to the camera and adapting the mounting arrangements appropriately. This would be smaller, lighter and easier to work with and align. Our system was built as described due to the availability of the parts.
An aluminium holder was machined to hold 47 mm diameter filters in place, with a thin ring machined to weight down the filter around the edges and ensure it is held flat for imaging, while not obscuring any of the actual sample area (the edge of the filter is clamped under the glass flange of the filtration apparatus, so the outermost 3-5 mm of the filter is blank). This holder would probably be better sprayed matt black to avoid any stray reflections, but for fluorescent imaging it works well as it is.
X-y motion was controlled by a G-code script, which moved the camera in a back and forth rectangular grid motion to scan the entire filter area using either a 7 X 5 or a 9 X 6 array. This is a good compromise between maximum speed of scanning and ease of image reconstruction. Staggered arrays could reduce the number of tiles required to cover the circular area, but would potentially be more complex to reassemble. The speed of the scanning motion and the acceleration/deceleration ramps can be adjusted in Mach 3 software to minimise jerking and vibration. This was adjusted by trial and error until a good compromise was found. A delay was also programmed before each image was acquired to allow for any vibrations to be damped, so that images were sharp and focused. The microscope zoom was set to 1 for the 7 X 5 array, but could then be zoomed (up to 4 X for this microscope head) to higher magnification if required.
The rectangular photographic image was set square to the x-y motion by aligning the edge of the field of view with one of the T-slots on the milling machine base by rotating the camera adaptor in the 3 rd port of the microscope head.
Images were acquired using the Canon EOS remote shooting software run on a different PC (it is recommended that Mach 3 is run on a dedicated computer, to prevent unexpected interrupts from disrupting the timing of G-code events). Focus was set by manual adjustment of the z-axis of the mill using the manual movement controls of the Mach 3 software, using white light to illuminate the filter. Once focused, the orange filter was set in place at the base of the microscope head and the sample filter illuminated with blue light from the Crimelite, which was clamped in place to give a uniform lighting across the filter. Exposure for fluorescent imaging was set manually by trial and error prior to imaging the whole filter.
Images were acquired manually, after each movement of the head, by activating the remote shutter button in the software. This action could almost certainly be automated by programming a suitable contact closure signal to the remote shutter input of the camera, but this has not yet been implemented. Note the notch in the edge that allows tweezers to be inserted to lift the filters in and out.

ZnCl2 solutions and densities.
Varying densities of ZnCl2 solution were prepared gravimetrically from a freshly opened bottle of ZnCl2 (this salt is very hygroscopic) using the weights indicated in the table. The densities were measured at 20 °C by weighing 100 mL class A volumetric flasks empty and filled with the solutions. All solutions were vacuum filtered through 0.45 µm membrane filters prior to use. They were stored in bottles/flasks with ground glass stoppers.

Separation of chitin using density differences.
Chitin-based exoskeleton material from marine organisms is quite heavily mineralised with calcium carbonate and hence has a density significantly higher than the vast majority of common plastics. This makes it relatively easy to separate such material from microplastic fragments in marine sediment samples using simple density separation. For most work we have used ZnCl2 solutions with a density of about 1.35-1.38. This floats the vast majority of plastics but allows the mineralised chitin to sediment, as seen in the picture below. Fig SI6. Mixture of NR stained PE Microplastics and crab shell fragments suspended in ZnCl2 and then centrifuged at 3900g for 10 minutes in a glass centrifuge tube. The tube was then photographed in blue light through an orange filter.

4.
False positives from chitin in the ABSENCE of density separation.
NR staining of sediment samples led to the visualisation and extraction of a number of small fluorescent fragments, which were picked from filters using moistened cocktail sticks. To identify the material from which they were made, they were analysed by Raman microspectrometry using an Ar ion laser at 532 nm as source. For comparison, NR-dyed reference polymer samples were also measured. The spectra are shown below: The Raman study confirmed that NR-staining fragments containing high levels of calcite could be found in samples NOT treated by density separation. After density separation, these fragments were not found, suggesting that the density separation is an important step in removing these potential false positives from the sample.

(E) (F)
Note: All images have the same green fluorescent colour due to the use of a 525/50 nm bandpass emission filter in this (Zeiss GFP) filter set. This is why the nylon is rather faint and needs the same exposure as the algae, since, due to the red shift of NR on the nylon surface, only the tail of the emission light is captured. The filter matches the yellow emission spectrum of PE and PP much better, hence the much brighter images. NR stained neutral lipids in algae have a yellow emission colour very similar to PE and PP, so this provides a good comparison of relative visibility under this illumination regime. Note that no stained algae can be seen even though they were present in much higher numbers.

Possible false positives from other natural organic material.
We have checked a wide range of materials in addition to the chitin and algae discussed in more detail in the text. This includes wood fragments of various types, seaweeds of different types, feathers, common whelk egg cases and a variety of different shells as well as chitin materials such as shrimp and crab claw. Most of these materials stain either very weakly (too little to show up under our imaging conditions) or not at all. The exception to this is proteinaceous material, such as the whelk egg cases and fragments of crab claw that still contained crab meat. Since proteinaceous material is biodegraded rather rapidly and is generally buoyant, we did not consider it likely that there would be significant pools of such materials in marine sediments, but for measurements of other types of samples this may need to be taken into account and additional steps added to protocols, such as treatment with protease enzymes, as is generally used for processing biota.
Carbohydrate polymers do not seem to stain. We believe that this is due to the quenching effect of hydroxyl groups and the fact that the dye does not adsorb readily to such hydrophilic surfaces. The solvatochromism data presented in ESI shows that fluorescence of NR in water and alcohols is VERY weak, compared with non-polar solvents and we believe that this is the explanation for the nonstaining of typical OH-rich carbohydrates. We have tested cotton cellulose and many types of paper in other work and these materials do not stain significantly. The heavy acetylation of chitin makes it much more hydrophobic than more typical carbohydrate polymers and this is probably the explanation for its staining. Although we have not investigated other carbohydrates in detail, we have observed many spectra from non-fluorescing particles in FT-IR microscopy that closely resemble carbohydrate spectra and may have been e.g. from various seaweed carbohydrates (alginates, carageenens etc.). Many non-staining "black carbon" particles are also probably of lignocellulosic origin and give IR-spectral signatures consistent with this.
Wax-like materials (e.g. from cuticles of terrestrial plants) are an interesting problem, since they will have spectra very similar to polyethylene, distinguishable only by subtle differences in the CH2 and CH3 intensities, which will be difficult to pick up in the generally low-resolution and truncated spectra imaged on Anapore filters, hence they would very likely be mis-identified as polyethylene using current FT-IR screening methods currently in use, leading to false positives. They probably would stain (though we have not specifically checked this) and hence could also generate false positives (if present) using our staining method, hence either approach would be sub-optimal if large numbers of such fragments were present. We are not aware that this possibility has been discussed more widely in the microplastic literature, so it is probably an issue that should be further investigated, but it is beyond the scope of this proof-of-concept paper.
8 Example G-code programme for controlling milling machine motion.