The Microplastic Data Extrapolation Challenge: How do we Approximate True Concentrations from Subsamples?

There is no way around it. When conducting microplastic analyses in environmental samples, with the aim of quantifying the smallest microplastics (from 1 µm to 50 µm), we are forced to subsample due to time restraints. Generally, current Fourier-transform infrared (FTIR) and Raman microspectroscopy subsampling strategies usually measure about 1 to 10% of the filter area (Brandt et al., 2021).

It is generally accepted that particle spread on a filter surface is inhomogeneous. For that reason, microplastic researchers have developped a wide range of creative ways to avoid faulty extrapolation. Most-often, these techniques rely on sampling a statistically relevant fraction of the total filter. This can be done by sampling in a wedge-shaped (pizza model) formation, in a cross, a spiral, or completely random area-selection (Fig. 1).

Figure 1 - Different subsampling strategies to correct for inhomogeneous particle spread on a filter membrane. Figure reproduced from Huppertsberg and Knepper (2018):

While these methods have shown a general improvement in microplastic extrapolation, decreasing error by 20-80%, they are only available to research teams with specialized software and equipment. Generally, most FTIR and Raman software packages can only subsample in a circuclar or rectangular field of view (FOV).

During my PhD, I determined (based on a compromise between analysis time and data volume) that a 2x2 mm rectangular FOV was a feasible subsample area for automated Raman microspectroscopy of particles from 1 or 2 µm in diameter. Fig. 2 illustrates a 2x2 mm subsample from a true environmental sample.

Figure 2 - 2x2 mm subsample. The is mosaic is conctructed from several hundreds of micrographs at a spatial resolution on the order of 0.16 µm/pixel.

Upon closer inspection, we realize that the relatively small 2x2 mm area is rich in particles (Fig. 3). Depending on the sample type, an area of this size would usually host from 10,000 to 30,000 individual particles if you were applying a limit of detection (LOD) of either 1 or 2 µm in area-equivalent diameter. Note that each of these particles would be analyzed by automated Raman microspectroscopy.

Fig. 3 - Makro view of a FOV measuring approximately 160x90 µm, within a 2x2 mm subsample. The FOV demonstrates the particle abundancy within the subsample.

While each subsample represented thousands of individual particles, we knew that our extrapolation might yield bewteen ± 20-80% error (due to particle spread inhomogeneity), had we simply extrapolated based on the area of filter investigated. Intelligent extrapolation was critical as the subsample area only corresponded to 2.6% of the total filter area (Fig. 4).

Figure 4 - Illustration of a rectangular subsample area measuring 2x2 mm on a Ø = 14 mm circular filter area, in true relative scale. The subsample corresponds to only 2.6% of the filter area, yet contains chemical information on 10,000 to 30,000 individual particles

To correct for iinhomogeneous particle spread on the filter membrane, we determined to investigate the ratio between microplastics and generic particles. We knew that the particles were not homogenously spread on the filter but what if the ratio between microplastics and generic particles was stable? In that case, all we had to do was determine the total area of particles investigated in the subsample and the microplastic-to-generic-particle-ratio, and extrapolate based on the total area of particles on the filter membrane.

The area of generic particles in the subsample was already provided by the Raman software package, so to determine the area of particles on the whole filter, we employed CADFLI (critical angle darkfield illumination) microscopy (Fig. 5). We later learned that this could also be achieved by switching to a lower magnification obejctive on the Raman microscope (or any other automated microscope). The latter appraoch was particularly useful in the case of lightly charged filter surface.

Figure 5 - Isolated particles on an entire filter membrane (Ø = 14 mm). The graphic is reworked from a CADFLI-captured micrograph mosaic. The data was used to extrapolate microplastic data based on the microplastic-to-generic-particle-ratio.

Using this particle-area based extrapolation approach, we found that subsampled areas in different samples, diverged from occupying 1.5% to 6% of the total area of particles, although the subsampled area was always placed in the center of the filter. To determine if this technique was valid, we staged a homogeneity test in which we subsampled four different areas on the filter membrane, and compared total particle area, microplastic-to-generic-particle-ratio, polymer type and size distribution, etc. (Fig. 6).

Figure 6 - Visualization of homogeneity test to determine if the particle-area based extrapolation approach was valid. Four subsamples were placed in the current pattern, on two individual filters.

Our preliminary results indicate that the total number and area of particles diverged by more than 100% in individual 2x2 mm rectangular subsamples, clearly demonstrating inhomogeneous particle spread. However, we also observed that the microplastic-to-generic-particle-ratio remained close to constant! The same applied to the particle size and polymer type distribution.

At the moment, I am in the process of preparing a manuscript for publication in a scientific journal. I am confident that this approach has the potential to enhance extrapolation results without requiring specialized sampling techniques.

If you are intersted in learning more about microplastic extrapolation, reach out at oskar@microplasticsolution.com. Thank you for reading this blog post. If you found it interesting, don't hesitate to subscribe to our newsletter.