1. Introduction
Plastic debris can be found nearly everywhere, even to the point where we can define a geological epoch by its presence [
1]. As plastic weathers and degrades in the environment, there has been an intensive effort in the last decade to better understand the impacts of microplastic debris in a variety of settings, including along waterways [
2,
3,
4], in the air [
5], in wildlife [
6], and in humans [
7,
8]. Plastics themselves are heterogenous materials, with a wide range of complex chemistries, many of which can trigger toxicological endpoints, such as oxidative stress and cytotoxicity [
9]. The implications of this persistent debris for human health are not yet known, but there is evidence that plastic nanoparticles can cross the blood–brain barrier in fish, causing behavioral changes [
10]. Additionally, polystyrene microspheres (5 and 20 µm in diameter) were found in the liver, kidney, and gut of mice after 28 days of exposure, and 0.1 mg/day of exposure was enough to induce metabolomic alterations [
11]. One estimate suggests the average American consumes between 39,000–52,000 microplastic particles annually [
12].
A recent review of microplastic assessments in fresh water sources documents a range of reported microplastic concentrations spanning ten orders of magnitude [
13]. It is likely that both the variable measurement techniques and the true variation in environmental pollution levels help explain this remarkable range of answers in the literature and underscores the urgent need for standard definitions and practices. These challenges are confounded at the sub-millimeter size scale, where microparticulate debris has the potential to be absorbed by the human body [
8]. In this range, it becomes more difficult to isolate microplastics in environmental samples efficiently, increasing time and cost, ultimately making an assessment of human exposure prohibitively difficult. Manual-inspections that pull out fibers or size plastic ‘nurdles’ are not scalable methods for quantifying the amount of small debris (<20 µm) in the environment, as many millions of microparticles could be produced from a single millimeter-sized fiber [
14]. Methods for inspecting plastic debris rely heavily on the sequestration and planarization of the debris to make it compatible with different forms of metrology (e.g., fluorescence microscopy, Raman spectroscopy, Fourier-transform infrared spectroscopy (FTIR), energy-dispersive X-ray spectroscopy (EDS)). An ideal way to prepare samples for these techniques would be to utilize a direct filtration method to concentrate the particulate mass for imaging and characterization. Effective machine learning techniques can help categorize particles given clear imaging criteria [
15]. Simply drying a drop of liquid and immobilizing particles on a surface can create samples for characterization on a smaller scale, but this method cannot interrogate larger liquid volumes for sparse numbers of particles efficiently.
Here, we demonstrate that ultrathin silicon nitride membranes [
16,
17] provide a robust way to rapidly filter municipal water samples and capture a size-specific range of debris for a variety of inspection techniques. Silicon nanomembrane technology has been used in a variety of biosensing applications since their introduction over a decade ago [
18,
19,
20,
21,
22,
23,
24]. Nanomembrane filters are extraordinarily efficient at separations due to their thinness (<400 nm thick), providing separations at low pressures. Over the last decade, the silicon nanomembrane platform has transitioned into silicon nitride materials that are more robust while still maintaining the ability to fabricate a variety of pore shapes and sizes (50–30,000 nm). In the context of detecting microplastics, silicon nitride nanomembranes also provide a plastic-free background for direct measurements on the filter after filtration, are inert to harsh chemical treatments (4M KOH, 3:1 H
2SO
4:H
2O
2 Piranha Etch) used to dissolve biological contaminants, and create low-background signals to common spectroscopic and conventional imaging modalities. Thus silicon ‘nanomembranes’ offer significant value to the study of microplastic contamination, particularly in the smallest size range.
Our study introduces a novel silicon nanomembrane-based filtration and processing assay to track the existence of microplastic debris along the route from the City of Rochester’s water production facility at Hemlock Lake, NY to the University of Rochester’s Goergen Hall tap and drinking fountain. The ability to rapidly process water samples to visualize debris, including plastics, provides new context to earlier reports of microplastic contamination in municipal drinking water. We found that the water produced by the Hemlock Lake facility was relatively free of debris (10 particles/mL), however, some samples along the water transport route measured >1500 particles/mL (with microplastics accounting for as much as 9% of all the debris). Debris, including microplastics, was particularly concentrated at the end of long interrupted stretch of pipes including the entrance to the University’s Goergen Hall, suggesting they are entrained en route to the taps of homes and businesses.
2. Materials and Methods
2.1. Water Processing and Sample Collection
Samples were taken at various points along the water transport route from the Hemlock Lake water production facility to the Goergen Hall at the University of Rochester (
Figure 1). Samples were collected on 2 July 2019 (between the Plant Output and Reservoir Exit samples) and 2 August 2019 (between the Campus Entrance and 5 min Drink samples). The Goergen Tap sample was collected on 16 December 2019. To ensure the sampling containers were not contaminating any of the sample collected, the outside and inside of containers were thoroughly rinsed with 100% ethanol stored in glass bottles, which we have previously found to contain no particulate. The containers and caps were sonicated for 1 h, then rinsed again with ethanol. After drying, 5 mL of ethanol was left in the container to confirm that the interior was airtight while being stored. As a control, we stored and processed 50 mL of ultrapure water (Invitrogen) in the same manner.
2.2. Sonication and Cleaning Protocol
All glassware, pipette tips, 1.5 mL conical Eppendorfs, and customized nanomembrane holder (SEPCON) components were sonicated for 15 min in 100% ethanol prior to use, then dried in a 70 °C oven. Glassware were covered with ethanol rinsed aluminum foil and all other items were stored in sterile, covered tissue culture petri dishes until use.
2.3. Graduated Cylinder Gravity Filtration
Lithographically patterned silicon nitride nanomembrane filters (5.4 × 5.4 mm silicon chip, three 0.7 × 3.0 mm rectangular windows, 8 µm slit widths 400 nm thick membrane, 6.3 mm
2 active area) were produced by SiMPore Inc. (
www.simpore.com, West Henrietta, NY, USA) [
25]. These filters were then placed in SEPCON™ (SiMPore Inc., West Henrietta, NY, USA) units and sealed with silicone gaskets. The bottoms of 100 mL glass graduated cylinders were drilled out using a 3 mm diameter glass-drilling bit, then pressure-sensitive adhesive (3M) was used to attach a SEPCON device containing a silicon nanomembrane to the bottom of the cylinder, allowing gravity filtration to occur through both the cylinder and SEPCON device. Sample water (50 mL) was filtered through the silicon nitride nanomembranes, then the SEPCON units were removed and dried in a 70 °C oven. Membranes and debris were imaged under white light using a brightfield microscope (Nomarski DIC, 2.5625 pixels/µm), simultaneously in reflection and transmission mode at uniform lighting, or under white light using a dissection microscope (1.96 pixels/µm).
2.4. Dissolution and Washing Protocol
Eppendorf tubes (1.5 mL) were filled with 0.75 mL of 0.125 M Tris HCl. The SEPCONs containing filtered debris were then placed in the filled tubes. Next, 200 µL of 10% w/v Sodium Dodecyl Sulfate (SDS) and an additional 300 µL of 0.125 M Tris HCl were placed in the SEPCON basket. The Eppendorf-SEPCON unit was then heated to 95 °C and stabilized for 5 min at that temperature. An additional 200 µL of 14.6 M 2-mercaptoethanol was added to the SEPCON basket and the filters were left to process for 1 h inside of a fume hood. SEPCONs were removed and dried on a cleaned petri dish. Ultrapure water (Invitrogen) was heated to 90 °C. The heated water (750 µL) was used to load the SEPCON basket and left for five minutes. The SEPCONs were then gravity drained by lifting the SEPCON unit slightly out of the 1.5 mL tube, and this process repeated for 3 repetitions. After washing, the membranes were dried in a 70 °C oven, then imaged under white light using a brightfield microscope (Nomarski DIC, 2.5625 pixels/µm), simultaneously in reflection and transmission mode.
2.5. Nile Red Staining
Substrates were stained in situ with a lipophilic dye Nile Red (Abcam ab228553, 20 µL, 1 µg/mL), then imaged under an epifluorescent microscope (0.6107 pixels/µm). Stained debris was assumed to be plastic as described elsewhere [
26].
2.6. Image Processing, Segmentation, and Quantification
Images were processed on a MacBook Pro (15-inch, 2019, 2.3 GHz 8-Core Intel Core i9, 16 GB Ram). The number of images analyzed for each replicate ranged between 9–36 images. Automatic segmentation of particulate from DIC imaging was achieved through the use of Trainable WEKA Segmentation [
27], a plugin found in the Fiji [
28] distribution of ImageJ [
29]. A separate classifier was trained using 5–10 examples of 5 categories (defined as particle, slot, residue, membrane, and edge) identified on each sample. The classifiers were used to categorize images from each sample, then the probability map for each particle category was automatically generated (Minimum, Auto-Threshold, ImageJ) to create a binary map of debris. Fibers were extracted from the debris map, using an algorithm found in DiameterJ [
30] (path 29), leaving only a sparse particulate map. The map was then processed with ImageJ’s watershed algorithm to separate aggregates. The particles were subsequently quantified using FiJi’s count particles plugin, capturing the individual particulate dimensions, which were then used to calculate a variety of physical information, such as particle volume with a simple model of an oblate spheroid.
Fluorescent images were cropped manually to the entire active area of the nanomembrane filter, then automatically thresholded (Minimum, Auto-Threshold, ImageJ) quantified using ImageJ’s watershedding algorithm to separate particles. Some outliers were manually removed from the data set (12/765 images) due to orders of magnitude changes in particulate concentration. Separately, fibers in images were counted by eye (2–3 independent volunteer counters) and particles of noticeable length were sorted into ‘large’ (>8 µm) or ‘small’ (≤8 µm) bins based on the relative size of the diameter of the fiber to the size of the slots in the silicon nanofilter (
Supplementary Materials,
Figure S1).
2.7. Elemental Analysis
We performed EDS (EDAX) on selected substrates using a Zeiss Auriga SEM. Substrates were sputter-coated with a 7 nm Au layer for improved backscatter imaging.
4. Conclusions
We used silicon microslit filter technology to concentrate and purify microplastic debris from a variety of water samples taken along the water production and treatment route for our drinking water. This platform simplifies the sample preparation and quantification of small microparticulate through direct filtration, (5–60 min dependent on debris profile), staining (10 min), drying (10 min), imaging (5–15 min) and semi-automated machine learning for image analysis and particle classification (30–120 s/image). While the water leaves the producing plant relatively devoid of small debris particulate (sub 20 µm), transport through many miles of piping appears to entrain debris, while settling points and end filtration along this route reduce the amount of debris a person might consume. A majority of these particulates are not plastic although a significant fraction of them (up to 8.6%) were identified as microplastics using a lipophilic stain. This platform could be used in a variety of other contexts for environmental sampling of microdebris, such as a rapid quality control or a debris sensing mechanism. Furthermore, the usage of silicon nanomembranes enables many other sample preparation methodologies and complex metrologies that are unavailable to conventional membrane solutions. Lastly, the combination of small input volumes and membrane surface areas can allow for parallelization of sample processing and imaging in the future, leading to higher throughput analysis of small water volumes.