Can RO Membranes Shed Nanoplastic In Drinking Water?

I have been asking the same question since I first became interested in testing drinking water for nanoplastics: Could a reverse osmosis membrane shed plastic particles into the water it is supposed to purify?

That question is the primary reason I started pursuing this testing.

I use reverse osmosis because it is designed to remove an enormous range of contaminants. At the same time, the membrane doing that work is made from layers of polymers. I kept coming back to the possibility that a filter could remove contaminants while adding nanoplastic particles from its own materials.

It is why I kept looking for a laboratory, a method, and reference materials capable of seeing particles this small. This lab session did not create the concern. It was one step in a much longer attempt to answer it.

For this pilot, I compared my home RO water with glass-bottled spring water and water sold in a single-use plastic bottle. We measured unstained samples in ordinary scattering mode, then ran separately stained samples in fluorescence mode to look for Nile Red-responsive particles.

The results did not identify the source of any individual particle. They did, however, give me my first useful comparison of raw scatter and fluorescence-mode results across these three water sources.

How We Tested for Plastic-Like Particles

Along with the three drinking-water samples, we ran ultrapure-water controls, a no-dye control, and a known polystyrene particle standard as a positive control.

The instrument used nanoparticle tracking analysis, commonly called NTA. A laser illuminates particles suspended in water, a camera records their movement, and software estimates their size and concentration based on how they move.

The important limitation is that ordinary NTA sees particles, not polymer identity. A mineral particle, a bit of organic material, and a plastic particle can all scatter light. The first part of the experiment could tell us how many trackable particles were present. It could not tell us which ones were plastic.

That is why we added Nile Red, a fluorescent dye that becomes much brighter when it associates with hydrophobic surfaces, including many plastics. After staining, we used fluorescence mode with a 550 nanometer long-pass filter. In simple terms, the instrument was looking for tracked events that emitted redder light after excitation.

The scattering and fluorescence measurements were separate runs, so the instrument was not following the same individual particle from one mode into the other. What we could compare was the overall reported concentration and the size distribution of tracked events in each run.

For shorthand, I describe fluorescence-mode events as plastic-like or Nile Red-responsive. I do not call them confirmed plastic, and the control results show why that distinction matters.

Nile Red can also associate with some nonplastic hydrophobic material. Different polymers stain with different brightness. Very small plastic particles may not pick up enough dye to cross the instrument's detection threshold. The solvent, dye concentration, incubation time, excitation wavelength, and optical filter can all change the result.

This is a screening method. It is another piece of the puzzle, not chemical identification.

Before comparing the drinking-water samples, we checked whether known polystyrene particles could be detected in fluorescence mode. The archived reports show 440,000 particles per milliliter from 11 tracked events in the earlier scatter run and 320,000 particles per milliliter from eight tracked events in the fluorescence run. The raw fluorescence concentration was therefore about 73 percent of the scatter concentration. The size distributions overlapped, but they were not identical and the fluorescence distribution shifted toward smaller sizes.

This confirmed that the polystyrene control produced a detectable fluorescence response. It was not a formal recovery measurement because these were sparse, separate runs rather than replicated, blank-corrected measurements. It also did not establish equal response for every polymer, especially the materials used in an RO membrane.

The background control is equally important. The dye-stained ultrapure-water control produced a raw fluorescence-mode concentration of 240,000 particles per milliliter from six tracked events. A no-dye methanol control produced no fluorescence events. The instrument reports did not apply a stored blank correction, so all fluorescence concentrations below are raw values. This background prevents me from treating the reported fluorescence concentrations as clean counts of sample-derived plastic particles.

What the Water Samples Showed

During the session, the laboratory technician reviewed the preliminary distributions and relative concentrations. The archived files now let me state the raw outputs more exactly. Each result came from one run of 10 videos, or 3,000 processed frames. The concentration estimates were based on only one to 19 tracked events per run.

The spring water is the clearest example of why total particle count cannot be treated as plastic count. It had the highest raw scatter concentration, at 760,000 particles per milliliter, yet the lowest raw fluorescence concentration, at 40,000 particles per milliliter. Many of the particles detected in scatter mode may have been naturally occurring minerals or other nonplastic material. At the same time, its fluorescence value came from only one tracked event, so the exact number is extremely uncertain.

The reverse osmosis water had the lowest raw scatter concentration, at 80,000 particles per milliliter from two tracked events. Its fluorescence-mode concentration was 120,000 particles per milliliter from three events. Because the two modes were separate runs and the fluorescence result included an uncorrected background, I do not interpret the higher fluorescence number as evidence that staining created a larger true particle population.

The water sold in a plastic bottle had the highest raw fluorescence-mode concentration among the three drinking-water samples. Its reported 480,000 particles per milliliter was four times the 120,000 particles per milliliter reported for the RO sample. That comparison describes the two raw instrument outputs. It is not a fourfold blank-corrected difference in confirmed plastic particles.

That result matched my intuition, but matching intuition is not the same as proving the cause. We tested one sample from each source, on one day, with an exploratory protocol. We did not chemically identify the individual fluorescent particles. The bottle itself may be one possible source, but processing equipment, source water, caps, tubing, filtration systems, and other contact materials are also possible sources.

The narrow conclusion is still meaningful: among these three water runs, the plastic-bottled sample produced the highest raw fluorescence-mode concentration, while the glass-bottled spring water produced the lowest. Because the dye-stained ultrapure control produced 240,000 particles per milliliter, this pilot cannot convert that ranking into reliable blank-corrected counts of sample-derived plastic particles.

The Question Behind This Entire Project

Concern about the RO membrane is not a new question raised by these results. It is the question I brought into the laboratory from the beginning.

My RO water had the lowest raw scatter concentration of the three drinking-water samples. Its raw fluorescence result was lower than the plastic-bottled sample and higher than the spring-water sample, but it was also below the dye-stained ultrapure control. The scatter result was encouraging. The fluorescence result was unresolved. To understand what either result means for membrane shedding, I still need to know whether debris from the membrane would fluoresce under our exact test conditions.

Most modern RO membranes use a thin-film composite structure. A typical membrane has an extremely thin aromatic polyamide separation layer, a thicker porous polysulfone support, and a nonwoven polyester backing. Some systems use cellulose acetate or related materials instead. Academic reviews describe the polyamide layer as roughly 0.2 micrometers thick, supported by much thicker polysulfone and polyester layers.[1]

This matters because a polystyrene control does not validate every one of those materials.

Research shows that Nile Red can stain polyamide and polyester, but the result depends heavily on the protocol. In one fluorescence spectroscopy study, Nile Red-stained nylon and polyester emitted most strongly around 600 to 620 nanometers when excited near 465 nanometers.[2] Our 550 nanometer long-pass filter should be capable of passing much of that red-shifted emission.

That is encouraging, but there are two important cautions.

First, the aromatic, highly cross-linked polyamide used on an RO membrane is not the same thing as the nylon samples commonly used in staining studies. Nylon is a helpful comparison, not a perfect substitute.

Second, published results are inconsistent because the staining conditions are inconsistent. One laboratory study using Nile Red in n-hexane stained several common plastics but did not successfully stain polyamide or polyester.[3] Other work using acetone-based conditions produced fluorescence from both materials.[2] A more recent comparison across multiple carrier solvents found that fluorescence performance and polymer damage varied substantially with the solvent, and concluded that the method still needs better standardization.[4]

The single-particle evidence is also cautionary. In a flow-cytometry experiment using 10 micrograms per milliliter of Nile Red in diluted DMSO, polystyrene and polyester were identified more cleanly, while polyamide was partly obscured by dye aggregation and precipitation. The reported lower particle-size limit was about 200 nanometers.[5]

Researchers have detected Nile Red-stained polystyrene particles down to about 45 nanometers using highly sensitive single-particle tracking, but they also found that less hydrophobic polymers produced weaker fluorescence.[6] That means we cannot take the small-particle performance of polystyrene and assume that RO-membrane polyamide will behave the same way.

My best reading of the evidence is this: at least some debris from a typical RO membrane would probably fluoresce under an appropriate Nile Red protocol, especially polyester debris and larger fragments. The polyamide surface layer may also fluoresce, but likely with less consistent recovery than polystyrene. Polysulfone remains less well validated under conditions like ours.

This is how I have to interpret the RO result.

A positive fluorescence signal in RO water could include membrane-related particles, but it could also come from dye-related background, tubing, housings, a storage tank, plumbing, source water, or nonplastic hydrophobic material. A low raw fluorescence result cannot prove that the membrane shed nothing. Some membrane debris might stain weakly or remain below the fluorescence detection threshold.

Why Nanoplastics Are So Much Harder to Identify

It helps to pause here and remember that microplastics and nanoplastics are not simply the same testing problem at different sizes. For this discussion, I am using microplastics to mean particles from about 1 micrometer up to 5 millimeters, and nanoplastics to mean particles below 1 micrometer, or 1,000 nanometers. The dividing line varies across studies, but the practical problem is the same: as particles move into the nanometer range, the methods that work for larger fragments begin to lose their usefulness.

A larger microplastic can often be captured on a filter, seen under a microscope, and chemically examined as an individual particle using infrared or Raman spectroscopy. A nanoplastic may be too small for ordinary optical imaging and too small for reliable particle-by-particle chemical identification. NTA can follow the motion of very small particles and estimate their size and concentration, but it does not tell me which polymer each particle is made from. Nile Red fluorescence narrows the field toward dye-responsive particulate matter, but it still does not provide definitive polymer identification.

The other problem is mass, and particle mass falls with the cube of diameter. At the same density, one 100 nanometer sphere has only one millionth of the mass of a 10 micrometer sphere. This is why a water sample can contain a very large number of nanoparticles and still contain almost no recoverable polymer mass.

I wanted a rough sense of what that could mean for a GC/MS path. The following calculation uses the raw RO fluorescence concentration from this pilot, 120,000 events per milliliter, and a planning target of 25 micrograms of recovered polymer. That 25 microgram figure is not a universal GC/MS detection limit. The actual amount required depends on the instrument, the polymer, the method, and the laboratory.

For a deliberately simple estimate, I assumed that every fluorescence event was one spherical plastic particle, every particle had the same diameter, the polymer density was 1 gram per cubic centimeter, and recovery was 100 percent. The calculation is:

mass per particle = density × π/6 × diameter³

At an assumed diameter of 50 nanometers, one particle would weigh about 0.0000000000654 micrograms. At 120 million particles per liter, that becomes only about 0.00785 micrograms of polymer per liter. Reaching 25 micrograms would require about 3,200 liters.

This table is a scale estimate, not a measurement of plastic mass in my RO water. It makes assumptions that are almost certainly too favorable. The fluorescence events were not chemically confirmed as plastic, the stained ultrapure control produced a higher raw concentration than the RO sample, the particle sizes were not uniform, and no real concentration process captures 100 percent of the particles. If recovery were only 50 percent, the required water volume would double. If the true sample-derived particle concentration is lower after background correction, it would rise again. A laboratory willing to work with less than 25 micrograms could reduce the volume in direct proportion.

Even with all those qualifications, the math explains why nanoplastic confirmation is so difficult. Counting small moving particles in a fraction of a milliliter and recovering enough clean polymer for chemical identification are two very different tasks. Concentrating tens, hundreds, thousands, or even tens of thousands of liters without adding plastic from filters, seals, tubing, containers, airborne fibers, or handling becomes an experiment of its own.

What I Think I Know Now

I do not think this pilot is strong enough to rank every type of drinking water, every bottled-water product, or every RO system. It was not designed as a peer-reviewed study. It did not include enough replicate samples, enough brands, blind analysis, or polymer-specific chemical confirmation.

But it did answer several useful questions for my own investigation:

• My RO water produced the lowest raw scatter concentration of the three drinking-water samples, although that result was based on only two tracked events.
• Glass-bottled spring water produced the highest raw scatter concentration and the lowest raw fluorescence concentration among the three water samples.
• The water sold in a plastic bottle produced the highest raw fluorescence-mode concentration among the three water samples.
• The polystyrene control produced a detectable fluorescence response, with eight tracked events and a raw concentration about 73 percent of the earlier scatter run. That was not a formal recovery measurement.
• The dye-stained ultrapure control also produced fluorescence events, and no blank correction was applied to the reported water results.
• The plastics used in RO membranes are likely stainable to some degree, but the polystyrene control does not establish equal recovery for those materials.
• The raw fluorescence result in my RO water does not establish where those events came from or rule out membrane shedding.

For my own daily choices, I still see value in minimizing contact with single-use plastic when practical. I am encouraged that my RO water produced the lowest raw scatter concentration. I am not treating that result as proof that the membrane shed nothing, and I am not using the uncorrected fluorescence numbers as a clean measure of sample-derived plastic. I also see glass-bottled spring water differently now. It was not particle-free, but its scatter and fluorescence runs looked very different.

That is a personal decision based on three specific samples. It is not a health recommendation, a product endorsement, or proof that one category of water is always safer than another.

The experiment did not tell us whether the observed particles cause harm. It did not establish that every fluorescent particle was plastic. It did not tell us whether particle number, polymer mass, particle chemistry, or some combination of those factors matters most for health.

Those are the questions that remain.

The Two Difficult Paths I Am Considering

I am not at a point where there is one obvious next test. I keep coming back to two paths. The first tries to chemically identify polymer collected from actual RO water. The second stays with NTA and asks whether material from the RO membrane itself can be obtained, prepared, stained, and detected under our test conditions. Each path would answer a different part of the question, and neither looks easy.

Path 1: Furnace or Pyrolysis GC/MS

The first path is to concentrate particles from a much larger volume of RO water and use a furnace or another pyrolysis step coupled with GC/MS to analyze the captured material. If enough clean material could be recovered, this approach could identify polymer types and estimate their total mass. Finding chemical markers consistent with polyamide, polysulfone, or polyester would be important evidence, although it still would not prove that the membrane was the source. Those materials could also be present elsewhere in the RO system or introduced during collection and preparation.

The exact workflow would depend on what a laboratory can accommodate. It might mean capturing particles on a thermally compatible filter and placing the filter, or a section of it, into a furnace or pyrolysis cup. It might mean concentrating the water first and repeatedly loading and drying the concentrate into a small cup. Either way, I would need to work backward from the laboratory's equipment to determine the minimum target mass, the collection material it can accept, and how the entire captured sample could be transferred without leaving most of it behind.

The volume calculation above shows the scale of this problem. Even the 400 liter example is about 106 gallons, while 3,200 liters is about 845 gallons. The true requirement could be smaller or much larger depending on the actual particle sizes, background-corrected concentration, polymer density, laboratory detection limit, and recovery rate.

Collecting that much water is only the beginning. The system would have to retain particles below 1 micrometer without adding plastic of its own. Particles could be lost on vessel walls, tubing, seals, filters, and transfer tools. Dissolved minerals would also be concentrated, while the amount of polymer might remain extremely small. Collection blanks, process blanks, recovery spikes, and a carefully characterized filter or collection substrate would be essential.

This path has the advantage of providing chemical identification that NTA and Nile Red cannot. Its limitations are just as important. GC/MS would report polymer type and mass in the material that made it through the collection and preparation process. It would not report the original particle count or size distribution, and it might miss polymers that were lost before they reached the furnace or pyrolysis cup.

Path 2: More NTA Using Material from the RO Membrane

The second path is to continue with NTA, but this time use material from the same type of RO membrane installed in my system. The goal would be to learn whether that membrane material, under our exact preparation and staining conditions, can produce a fluorescence signal that the instrument can detect.

At first, this sounds easier than concentrating hundreds or thousands of liters of water. I am not sure that it is. A modern RO membrane is an extremely thin, layered composite. The aromatic polyamide surface is bonded to a much thicker polysulfone support and polyester backing. Obtaining the exact membrane element is one challenge. Separating or sampling those layers without mixing them, contaminating them, or changing their surfaces is another.

Creating a representative NTA sample may be harder still. Cutting, scraping, grinding, or sonicating the membrane could produce particles, but those laboratory-made fragments might not resemble material released during normal filtration. I would need enough particles in the nanometer range for the instrument to track, but not so much material that the suspension overwhelmed the measurement. Large fragments would have to be removed without also losing the smallest particles. Any filter used during preparation could shed its own material, capture the particles of interest, or change the size distribution.

Only after solving those problems could I move into staining and measurement. I would want to compare unused and used membrane material if both were available, and test the polyamide surface, polysulfone support, and polyester backing separately if they could be isolated. The controls would include unstained and stained ultrapure water, dye-only blanks, the known polystyrene reference, prepared membrane material by itself, and membrane material spiked into actual RO water.

This path would answer a narrower but necessary question: if particles from this exact membrane material are present, can our preparation, staining, and NTA method actually see them? It would not prove that the membrane sheds during ordinary use, and it would not tell me the shedding rate. It would also require confidence that the material I prepared was a fair representation of what might enter the water.

That is the real choice I am considering. The furnace or pyrolysis GC/MS path moves closer to chemical confirmation in the water, but it requires a credible way to concentrate and transfer enough material from a very large sample. The NTA path stays closer to the method that worked in this pilot, but it requires a credible way to obtain and prepare membrane particles without manufacturing an artificial answer. Obtaining and preparing the RO membrane material may be every bit as difficult as concentrating many gallons of water.

I do not yet know which path should come first. This testing did not create the membrane-shedding question. It moved a question that has driven this project from the beginning one step forward. We were able to detect known plastic particles in fluorescence mode, compare raw outputs across real water samples, expose the importance of the dye-stained background, and define the two difficult directions I am now weighing. For a method still being worked out at the edge of what these instruments can measure, I consider that a successful day.

Sources and Further Reading

1. Lejarazu-Larrañaga, A. et al. Thin Film Composite Polyamide Reverse Osmosis Membrane Technology towards a Circular Economy. Membranes, 2022.
2. Prasad, S., Bennett, A., and Triantafyllou, M. Characterization of Nile Red-Stained Microplastics through Fluorescence Spectroscopy. Journal of Marine Science and Engineering, 2024.
3. Shim, W. J. et al. Identification and Quantification of Microplastics Using Nile Red Staining. Marine Pollution Bulletin, 2016.
4. Ho, D. and Masura, J. Dyeing to Know: Harmonizing Nile Red Staining Protocols for Microplastic Identification. Colorants, 2025.
5. Kaile, N. et al. Preliminary Results From Detection of Microplastics in Liquid Samples Using Flow Cytometry. Frontiers in Marine Science, 2020.
6. Molenaar, R. et al. Nanoplastic Sizes and Numbers: Quantification by Single Particle Tracking. Environmental Science: Nano, 2021.