In Situ Lung Dust Analysis by Automated Field Emission Scanning Electron Microscopy With Energy Dispersive X-ray Spectroscopy: A Method for Assessing Inorganic Particles in Lung Tissue From Coal Miners Open Access

Lung tissue specimens from 10 coal miners and 4 control cases were included in this work. The miner specimens were obtained as part of a larger project led by the University of Illinois Chicago (UIC) and with approval from its institutional review board (protocol 2016-0767). For that project, UIC collaborated with the National Institute for Occupational Safety and Health (NIOSH), Federal Black Lung Program clinics, and attorneys to obtain tissue specimens and related case histories for coal miners with PMF.²⁷  Each case had been evaluated by 7 pulmonary pathologists using optical (brightfield) microscope images of hematoxylin-eosin (H&E)–stained tissue sections to confirm PMF (defined as lesion >1 cm in length with collagen indicative of dust-related fibrosis); and 10 of 62 confirmed PMF cases were selected for the current work from a sampling of the histopathologic subtypes of PMF observed, which included coal-type, mixed dust–type, and silica-type PMF.²⁷  Nine of the cases originated from NIOSH’s National Coal Workers’ Autopsy Study, which was established to support scientific research and as a service program to provide miners’ families with medical findings that may support a federal benefits claim.⁴⁰  The tenth case was selected from those recruited through clinics.

The 4 control cases were obtained from lungs of accident victims donated to the Lung Tissue Research Consortium at National Jewish Health (Denver, Colorado); this donor program is approved by the Biomedical Research Alliance of New York Institutional Review Board under protocols HS-3209 and HS-3429 (New Hyde Park, New York). Two individuals had known smoking histories and the other 2 had no known lung disease, based on details collected from next of kin. Control cases were selected on the basis of demographic characteristics, selecting those most similar in sex, age, and smoking status to the coal miner cases.

A 5-µm-thick section of the formalin-fixed and paraffin-embedded lung tissue was cut and mounted on a glass slide and stained with H&E. Brightfield images were captured at ×40 magnification using Aperio XT (ImageScope v. 8.2, Leica Biosystems, Deer Park, Illinois). We used ImageJ v1.53 software (National Institutes of Health, Bethesda, Maryland) and the Bio-Formats plugin to calculate the total tissue area for each sample, based on the brightfield images,⁴¹  after creating a binary mask to distinguish tissue from background. A pulmonary pathologist digitally outlined the PMF lesions on the brightfield images for the miner cases. Additional sections from each case were cut from the same tissue blocks as the H&E-stained sections, mounted on carbon planchettes (either 25.4 mm or 19.1 mm in diameter, depending on availability of supplies at the time), and deparaffinized for use in the analysis by FESEM-EDS. Figure 1 provides examples of the annotated brightfield images and corresponding photographs of the tissue sections mounted for SEM analysis. (The images and photographs for all miner and control tissue sections are provided in the supplemental digital content [containing 7 figures and 8 tables], Supplemental Figures 1 and 2, respectively).

In situ FESEM-EDS analysis was performed with an FEI Quanta 600 field emission environmental SEM (FEI, Hillsboro, Oregon) equipped with a Bruker Quantax 400 EDS spectroscope (Bruker, Ewing, New Jersey) and Esprit software (version 1.9.4). An automated procedure was set up to scan and collect data from a total of 89 fields across the 25.4-mm diameter carbon planchettes, or 69 fields across the 19.1-mm planchettes. As shown in Figure 2, A, the fields were spaced in a regular square grid 2 mm apart in both the x and y directions. The number of fields was chosen to maximize data collection within a reasonable analysis time (ie, about 1–3 hours per tissue section), and the field pattern was chosen to ensure uniform spatial distribution of data collection areas across the planchette. The procedure was run at ×4000 magnification, such that each field was approximately 874 µm² in area. The instrument was operated at 15 kV in high vacuum mode, with a working distance of 12.5 mm, spot size of 5.5, brightness of 92.5%, and contrast of 60% to 70%. The planchettes were analyzed as received, and no sputter coating was applied.

In each field, the automated process first collected a backscatter electron image from which it generated a binary mask to distinguish particles from the background. In backscatter electron images, features associated with heavier (ie, higher atomic number) elements produce a more intense signal. Thus, for in situ tissue analysis, the binary mask can be used to identify mineral particles with higher intensity than the carbon-rich tissue background (Figure 2, B). After particles were identified and labeled by their x-y coordinates in a given field, the process collected size (length, width) and elemental data on up to 50 particles in the field (moving from left to right, top to bottom). (Notably, across all tissue sections, a total of 593 fields were scanned that fell within the tissue area, and only 16 of these fields [<3%] had 50 or more particles.) Data collection was limited to particles with length (ie, longest axis) in the range 0.2 to 50 µm, based on the pixel size at ×4000 magnification and our desire to capture agglomerates, if present. For each particle, the EDS spectrum was captured for 4 µs per pixel live time, and the normalized atomic percentages were recorded for 9 elements: sodium (Na), magnesium (Mg), aluminum (Al), silicon (Si), phosphorus (P), potassium (K), calcium (Ca), titanium (Ti), and iron (Fe). These elements are characteristic of most mineral dust particles present in coal mines (eg, a range of silicates, silica)^22,42–46  or suggestive of endogenous deposition (ie, with expected enrichment in Ca, P, Na, K, and/or Fe).^35,47,48  Carbon (C) and oxygen (O) were not included in the normalized atomic data for the in situ analysis given the inherent C and O content within human tissue. Following data collection on each particle in a field up to the limit of 50 particles, or if no other particles were identified, the automated process moved on to the next field in the numbered sequence per Figure 2, A through D.

Because the control cases had relatively smaller total tissue areas and fewer particles than the miner cases, the automated process was performed a second time on each control tissue section and results averaged. Between the 2 runs on the control cases, the carbon planchette was removed from the SEM and rotated to ensure that different fields were scanned each time, and results from both runs were combined. For the miner cases where 2 tissue sections were analyzed (cases 1 and 5), the data were also combined to represent the total number of fields across both available planchettes.

At the end of each automated procedure run, images were also captured manually at ×20 magnification with the secondary electron detector to maximize resolution of tissue and planchette boundaries (Figure 2, C). These images were used for spatial referencing so that each field could be labeled as either outside or within the tissue area, based on the corresponding brightfield image (Figure 2, D). Fields outside the tissue area were excluded from the in situ particle analysis. (However, these fields were evaluated for contamination per Supplemental Table 1.) For the miner cases, fields within the tissue area were sublabeled as either within a PMF lesion (PMF) or outside the PMF in lung parenchyma (non-PMF), such that the FESEM-EDS data could be correlated respectively.

All data analysis was performed with JMP Pro 16 (SAS Institute Inc, Cary, North Carolina).

For each case, data from the in situ analysis were summarized by particle number density and particle size and class distributions. Particle density was computed on an area basis as the total number of particles identified per tissue area analyzed (No./cm²). For fields with fewer than 50 particles, the analyzed area was the total field area; for any field with more than 50 particles, the analyzed area included only the portion of the field that contained those 50 particles. For particle sizes, median and mean length were computed. For all statistical tests, particle length values were log-transformed (ie, log[length + 1]) because the data were not normally distributed.

To enable broad analysis of particle chemistry—from which mineralogy and possible origins can be inferred—we established a classification scheme. Briefly, we pooled all particles from all 10 miner cases. From the normalized percentage values for the 9 elements recorded during the automated FESEM-EDS data collection, we identified and temporarily excluded outlier particles as those having values 3× higher than the 90% quantile for the entire dataset. For the remaining particles, we used a statistical clustering approach (K means⁴⁹ ) to identify common particle chemistries. For this, we iteratively explored between 3 and 10 clusters, and used the cubic cluster criterion (CCC) value for each iteration to inform selection of an optimal number of clusters. Then, we set up particle classes such that there was 1 class for each cluster and also for each major group of outlier particles. Finally, we developed criteria (ie, elemental thresholds) for each class and used these to bin all particles identified in the miner and control cases.

Comparison of mean particle length across cases was done by using Tukey-Kramer honestly significant difference (HSD) tests. When comparing mean particle length across particle classes, a linear mixed model (unbounded variance components) was used, specifying class as a fixed effect and case as a random effect, to account for possible correlation within subjects. Pairwise comparisons were made by using a Tukey-Kramer HSD test as part of the mixed model.

Comparison of mean particle number density or length between the miner and control groups, or between PMF lesions and non-PMF lung parenchyma areas observed in the miner tissues, was done with linear mixed models, with case included as a random effect to account for possible correlation within subjects. All statistical tests were performed at α = .05.

For the miner cases, the results of the in situ analysis were compared with results previously obtained with particles recovered from digested lung tissue; the “digested analysis” results were published in 2023 in an open dataset by Lowers.⁵⁰  It is noted that the dataset from Lowers⁵⁰  represents a total of 95 miner cases, which were acquisitioned as part of the larger project mentioned above.²⁷  However, only data for the 10 miner cases included in the current work were investigated here.

For each case, the digested analysis was completed by using a 20-µm-thick section of tissue, cut adjacent to the 5-µm sections used in the brightfield microscopy and in situ particle analysis. The sample preparation and particle analysis procedures have been detailed elsewhere.^29,38  Briefly, each tissue section was deparaffinized in xylene and then digested in bleach (7% sodium hypochlorite solution). The resulting suspension was passed through a 25-mm polycarbonate filter (0.1-µm nominal pore size), which was rinsed with deionized water and dried before sputter coating with carbon. The particles were analyzed by FEI Quanta 450 FESEM operated at 15 kV, spot 5, working distance of 11 mm, and ×3000 for a horizontal field width of 48 µm. Automated particle analysis was completed with Oxford Instruments Aztec software Feature Module using the x-Max 50 mm² silicon drift detector. Image thresholds were set to ignore the filter and collect elemental data (weight percentage for 29 elements, including C and O) from particle features for 5 seconds and record the length, breadth (width), aspect ratio, area, and perimeter of the feature. The stopping criteria were set to 2000 particles or 110 fields of view per sample, which typically resulted in 200 to 2000 particles per case with a lower size limit of about 0.1 µm in length. The elemental data were used to bin particles into 1 of 67 different phases representing known minerals, or dominant element groups, with manual confirmation or correction. Blank samples were also prepared and analyzed.

For each miner case, we compared the in situ and digested analysis results as based on particle number density, particle length, and class distributions; all data were log-transformed (ie, log[value + 1]). Group means were compared by using linear mixed models (unbounded variance components) with case as a random effect to account for within-subject correlation.

Comparison of particle size distributions between the 2 analysis methods was done on the basis of median and mean values per case. To enable comparisons between particle class distributions, we carefully reviewed the particle data from the digested analysis⁵⁰  and collapsed the original 67. For this, we first excluded phases with particles that were not expected to be identified in the in situ analysis (eg, because they were dominated by elements not included in the in situ analysis or appeared to be primarily due to contamination). Then, each of the remaining digested analysis phases was aligned based on dominant elements with one of the classes established for the in situ analysis.

All particles identified in the 10 miner cases (n = 6030) were pooled and compared on the basis of chemistry, using normalized atomic percentage values for the 9 elements recorded during the automated FESEM-EDS data collection. Outlier particles (n = 337) were identified as having values 3× higher than the 90% quantile for the entire dataset; these particles were typically high in Ti (n = 222), Fe (n = 72), or Na (n = 29).

The outlier particles were temporarily excluded from the dataset, and clustering analysis was performed on the remaining particles (n = 5693) to look for natural groups based on chemistry. Between 3 and 10 clusters were explored iteratively. Supplemental Table 2 shows the CCC value and the characteristic elements for each cluster per iteration. The CCC value for 7 clusters was found to be markedly higher than for any number of fewer clusters, indicating improved fit of the clusters to the data. Additional clusters (8 to 10) did not substantially increase the CCC value, and these additional clusters would merely enable finer grouping of particles rich in Si and Al. For these reasons and the practicality of having fewer groups to adequately describe most of the particles, including their possible origins, we deemed the results for 7 clusters as optimal.

We developed classification criteria for a total of 10 classes (Table 1). These included 7 classes to correspond to the 7 statistical groups identified by the clustering analysis, plus 3 classes to capture the Ti-, Fe- or Na-rich particles identified as outliers. The criteria were applied as a series of “if, then” rules in the order shown in Table 1. If a particle did not meet the criteria for any of the defined classes, it was placed in an 11th class called “others.” Initially, the criteria for each class were set as based on the corresponding cluster mean plus or minus the standard deviation (or the outlier group threshold for Ti, Fe, or Na), with care to prevent the possibility of overlap between the criteria for multiple classes. Then, the criteria were iteratively adjusted to minimize the overall misclassification rate for the entire dataset of particles identified in situ from the 10 miner cases (n = 6030). Ultimately, the criteria shown in Table 1 produced a misclassification rate of only 9.4%, meaning that 90.6% of particles were classified into their statistical cluster or outlier group (Supplemental Table 3). These criteria were also applied to classify particles identified by the automated FESEM-EDS procedure in the 4 control cases. (Supplemental Figure 3 shows example SEM images with particle classifications indicated.)

Table 2 shows a summary of the results from the in situ particle analysis for all miner and control cases. (For results per tissue section for miner cases 1 and 5, which each had 2 available sections, and results per automated FESEM-EDS procedure run for the control cases, which were each run twice, see Supplemental Table 4.) The number density of particles for the miner cases was about 3.3× to nearly 2200× greater than that of the controls, and a t test on the group means showed the difference was statistically significant (P < .001). Miner case 10 had a particularly high density of particles (about 4.5× greater than the next highest case) and accounted for 38% of all particles identified in the 10 miner cases.

With respect to the types of particles found in situ by the automated FESEM-EDS analysis, Figure 3 shows the distribution by class for each case. Overall, the miner cases were dominated by exogenous silica and silicate particles (ie, including all Sil-Al types). In cases 1 through 9, these particle types accounted for 70% or more of the total particles; in case 10, silica and silicates accounted for 48% of the total particles. Most of the miner cases exhibited small percentages (7% or less) of Ti-rich or Fe-rich particles. Ti-rich particles (n = 155) are also attributed to environmental exposures and appeared to include both titanium oxides (particle dominated by Ti) and Ti-rich particles likely in clusters with silicates (particle with substantial Si and Al, but insufficient to be classified as Sil-Al,Ti) (see Supplemental Figure 4). Particles in the Fe-rich class could potentially be either exogenous or endogenous. Further inspection of the composition of such particles (n = 68) in the miner cases studied here suggests they were mostly endogenous: About two-thirds (66%) of these particles (n = 45) had Fe content less than 60% and Si and/or Al content less than 15%, which suggests they were generally not high purity iron oxides or iron silicates. Rather, 74% of the Fe-rich particles (n = 50) had Na, P, and/or Ca content greater than 30% (Supplemental Figure 4), which suggests endogenous deposition.

Other particles with likely endogenous origins observed in the miner cases included those in the Ca,P-rich and Na-rich classes. Cases 1, 6, and 10 exhibited moderate percentages (16%–26%) of Ca,P-rich particles, and these were accompanied by lesser percentages (4%–23%) of Sil-Al,Ca,P particles. This might be due to endogenous deposition on or adjacent to dust particles. Small percentages of Ca,P-rich and/or Sil-Al,Ca,P (totaling 4% or less) were also found in other cases. Na-rich particles were only observed in case 4 (3%) and case 7 (2%). These particles tended to have minor amounts of P, Fe, and K, and generally lacked the other elements included in the in situ analysis.

In cases 1 and 4, “other” particles accounted for 5% of the total particle count. Upon further inspection of the elemental data for these particles, most were found to be 1 of 2 types: aluminum silicates with such a mix of additional cations (ie, K, Na, Ca, P) as to preclude their classification into one of the defined silicates classes (likely exogenous or hybrid); or particles rich in both Na and P with smaller abundances of Ca or Mg (likely endogenous).

Results of in situ analysis for the control cases were in stark contrast to the miner cases (Figure 3). The 2 controls with very few total particles (C1 and C2) only exhibited Na-rich and Ca,P-rich particles, which were interpreted as endogenous deposits. Interestingly, these 2 control cases were individuals with known smoking histories. On the other hand, C3 and C4 did not have a known history of smoking, and the in situ analysis indicated particles of both endogenous (Na-rich, Ca,P-rich, Fe-rich) and exogenous origins, including silica and silicates, as well as Ti-rich particles. While C4 had the highest observed percentage of silica particles (31%) in any of the cases analyzed in situ, the overall particle number density was relatively low. Elemental data for the particles classified as “others” in C3 (18%) and C4 (8%) showed they were typically rich in Na and P (ie, consistent with some of the “other” particles in miner case 4, and likely of endogenous origin).

Figure 4, A, presents particle size distributions (based on length, µm) by particle class; and Figure 4, B, presents the size distributions by case, for the in situ analysis. When considering all miner and control tissues, particles in the silica class were significantly larger than all other particle types except those in the Na-rich class (P values given in Supplemental Table 5). It is possible that silica particles to which these miners were exposed were somewhat larger than other dust particles (eg, silicates). However, this finding could also be related to the capability of the in situ automated process to classify silica particles, since the EDS signals for smaller particles might be more susceptible to interference from surrounding particles (eg, in clusters). Sil-Al (general) and Ca,P-rich particles were significantly larger than other Sil-Al types. Notably, when the statistical analysis was performed using only particles from the miner tissues, the same significant differences were observed. When the analysis was run using only particles from the control tissues, no significant differences in mean size were detected between particle classes owing to low particle counts.

When the log-transformed mean particle size (irrespective of class) was compared between cases, particles identified in miner cases 1, 4, and 10 were significantly finer than in cases 3 and/or 7 (P values given in Supplemental Table 6). While the mean particle size for control C2 is quite fine (Figure 4, B), no statistical difference could be found owing to the very low particle count in C2. Moreover, comparing the log-transformed mean particle sizes between the miner and control groups indicated no significant difference (P= . 35).

Figure 5 shows the distribution of particles observed within and outside of PMF lesion(s) as demarcated on the corresponding brightfield image of each tissue section from a miner case. The results are scaled to number density of particles in either the PMF or non-PMF areas. For all 10 miner cases, the density of total particles within the PMF lesion(s) was higher than in the non-PMF lung parenchyma, and a t test indicated the difference was statistically significant (P = .03). Nevertheless, dust loading in the non-PMF areas was still very high in some cases—especially in comparison to the controls. The brightfield microscope images for most of the miner cases clearly show that many areas outside of the PMF lesions have substantial collagen consistent with dust-related fibrosis (see Supplemental Figure 1), and thus observation of high dust loading in the non-PMF areas is unsurprising. Where present, particles classified as “others” were always more concentrated within the PMF area. Otherwise, consistent trends were not observed in particle distributions with respect to PMF versus non-PMF areas. Similarly, no consistent trends were observed in log-transformed particle length with respect to PMF versus non-PMF areas (data not shown), and there was no statistically significant difference (P = .30).

To compare results for the 10 miner cases between the in situ analysis conducted here and the digested analysis from Lowers,⁵⁰  the digested particle data were first collapsed as shown in Figure 6. (For reference, the results are summarized in Supplemental Table 7, and Supplemental Figures 5 and 6.) Of 11,061 total particles in 67 classes across all 10 miner cases, 54% (n = 5996) were binned into 29 classes that overlap with the classes established for the in situ analysis (ie, the 10 classes with criteria defined in Table 1, plus an “others” class for particles like magnesium silicates that should be observable in situ but are unlikely to be binned into a defined class based on their elemental composition). These particles were included in the comparison of results between the in situ and digested analysis.

The other 46% of particles from the digested analysis (left side of Figure 6) were excluded because they were not expected to be identified by the in situ analysis; the excluded particles were of 4 main types. For 3 types, their characteristic elements were not recorded by the automated in situ procedure (ie, carbonaceous type, characterized by C and O; sulfur-rich, characterized by S; and metal rich, most often characterized by Cr, Ni, Zn, Cu, Zr, Sn, or Pb). The fourth type included aluminum, calcium, and iron oxides; while such particles should be identifiable by the in situ procedure if present, these particles in the digested results primarily appeared to be contamination, based on analysis of blank sample data. (It is noted that, for the 10 miner cases included in this work, no particles were binned into 9 of the original 67 classes considered by Lowers.⁵⁰  These classes are shown as “zero count” in Figure 6.)

Following collapse of the digested particle data, Figure 7 compares the in situ and digested results. Figure 7, A, shows the relative distribution of particles by class for each analysis method. For all 10 miner cases, both methods show that the overall particle count was dominated by silica and silicates—with relatively higher percentages of particles classified as Sil-Al types (58%–94% of total count) and lesser percentages classified as silica (5%–31% of total count). Figure 7, B, illustrates a correlation between the 2 methods in terms of the ratio of silica to the sum of silica plus all Sil-Al types. However, the ratio was typically higher for the digested analysis (ie, silica represented a greater proportion of the total when particles were recovered and then analyzed). The difference in mean silica was statistically significant between in situ and digested analyses (Supplemental Table 8).

Moreover, there were significant differences between the mean percentage of particles in some individual classes. While the digested analysis found higher percentages of silica and Sil-Al particles, the in situ analysis found higher percentages of Sil-Al,Na and Sil-Al,Ti. Significant differences were not detected for the Sil-Al,Fe or Sil-Al,Ca,P classes—or for the Fe-rich, Ti-rich, or Na-rich classes, which typically accounted for only a small (less than 7%) percentage of the overall particle count in each case. (All P values are given in Supplemental Table 8.)

Notably, the in situ analysis found significantly more Ca,P-rich particles than the digested analysis. In fact, while these particles were found in situ in 6 of the 10 cases (ie, 2%–26% in cases 1, 4, 6, 7, 8, and 10), they were virtually absent in the digested results for all cases (ie, less than 1%). Likewise, Sil-Al,Ca,P particles were virtually absent from the digested results for all cases.

The in situ and digested results were also compared with respect to the relative size and number density of particles identified in each of the 10 miner cases. Figure 7, C, shows that the median length of particles identified by the in situ analysis was typically larger than that of the particles in the digested analysis; this finding was statistically significant (Supplemental Table 8). This trend is unsurprising and can be attributed to clustering and embedding of particles in situ versus liberation and dispersion of particles for the digested analysis. These factors also influenced the particle densities observed in each analysis method. Figure 7, C, shows that the particle density estimated from the digested analysis was typically much higher than that estimated from the in situ analysis; this difference was statistically significant (Supplemental Table 8). For 7 cases (ie, cases 1, 3, 5–7, 9, and 10), the digested analysis particle density was higher than the in situ density by about an order of magnitude (ie, a factor of about 8× to 27×); and for case 8, the difference was even greater (ie, factor of about 51×). However, for cases 2 and 4, the density estimates were more similar (ie, factor of just 2× to 3×). Incidentally, these were the only 2 cases in which slightly smaller particles were observed in situ than in the digested analysis. One possible explanation is atypical loss of relatively small particles during the lung tissue digestion and particle recovery/redeposition procedures for these 2 cases.

SEM-EDS analysis of lung particles in situ can provide important insights into the pathogenesis of pneumoconioses and might serve as a valuable tool for development and evaluation of treatment for these often progressive and fatal diseases. However, manual in situ analysis can be time and cost intensive, and so it is not routinely applied. Automated analysis might overcome these limitations, while also reducing the potential for user bias, but efforts in this direction are quite new. At the time of publication, the 2023 study by Hayden et al³⁹  was the only report of in situ automated FESEM-EDS analysis for lung particle characterization that could be found in the archived literature.

For the purpose of discussing method development, FESEM-EDS analysis of particles can be conceptualized in 2 major steps: data collection and data interpretation. The approach to automated data collection presented in the current work is similar to that of Hayden et al,³⁹  though it is acknowledged that far less area was analyzed per tissue section (ie, an average of 0.032 mm² per each of the 10 miner cases here versus 1 mm² per each of 38 military service member cases in Hayden et al³⁹ ). Even so, it seems the particle density in the miner cases we report was considerably higher than that of the cases analyzed by Hayden et al³⁹ ; using average values for particle counts and analyzed tissue areas in each study, the particle density for the miner cases would be about 5× greater than that of the military service members.

Clearly, development of any analytical method should consider the overall representativeness of results, and balance available time and resources against data collection. The general agreement between particle distributions observed with the in situ versus digested analysis suggests that numbers of fields and grid spacing used here were sufficient for the miner tissues studied. That said, the per-section results for cases 1 and 5 reveal some nuances (Supplemental Table 4). For these cases, 2 sections were analyzed by the in situ method, but their particle number density and distribution (classification and size) varied somewhat. The per-run results for the control cases highlight the challenge for analyzing tissues with relatively low particle loading. For each of these cases, the tissue section was analyzed twice by the in situ method, but multiple particles were only identified by both runs for cases C3 and C4. Future work should consider if, when, and how to analyze multiple tissue sections per case to achieve a more representative understanding of the entire lung region of interest, as well as when and how to increase the scan areas for tissues with low particle density. Moving forward, other improvements in data collection may also be afforded by advancements in SEM instrumentation. For example, combining a high-scan-rate SEM with software that supports user-defined scan areas could enable analysis of an entire tissue section. At the same time, such an approach might eliminate the need for post processing of data to exclude fields outside of the tissue area—though we note that having the “blank” fields available here was actually valuable for assessing potential contamination. Advanced image processing tools might also allow more automated referencing of FESEM-EDS data to features of interest identified via correlative analysis. These could include pathologic features such as the PMF lesions identified on brightfield images and manually linked to the FESEM-EDS data here. Such an approach might even enable identification of coal dust particles in situ by using correlative analysis conducted with light microscopy.

Regarding interpretation of FESEM-EDS data, a well-considered approach to particle classification is of primary importance. For samples where particles are relatively dispersed, such as those prepared for digested-lung particle analysis, a common approach is to compare the particle chemistry to a database of known materials and identify the best match. This approach can enable very granular classification and may be favorable in cases where the objective is to describe even rarely observed particle types. For in situ analysis of inorganic particles in lung tissue, the 2018 study by Lowers et al³⁸  took this approach when using a manual data collection procedure, and the 2023 study by Hayden et al³⁹  attempted it with limited success using an automated procedure. Both studies noted the challenges of clustered particles, which necessitated time-intensive manual review of EDS spectra to arrive at a classification. With the automated procedure, Hayden et al³⁹  reported that many particles were ultimately still classified as “mixed.” In cases where mixed particle types can occur or where observed particle chemistry might be influenced by the sample substrate, an alternative approach to classification is to establish criteria by busing ranges of allowable elemental content, which can account for deviations from theoretical mineral composition.⁵¹  For in situ analysis of lung particles, this has been done empirically in numerous studies.^33,35,52 

Our focus in the current study was analysis of coal miner lung tissues, which can be characterized by especially high particle number density, often due to many years of exposure, and frequent particle clustering, which might yield mixed EDS spectra. (Supplemental Figure 7 shows example SEM images and elemental maps.) Additionally, we expected that exogenous particles in such tissues would be dominated by a limited number of particle types (eg, silica, aluminum silicates, titanium-bearing particles). This expectation was based on previous manual FESEM-EDS analysis of coal miner lung tissues,²⁷  as well as studies of RCMD, which shows that the mineral content of the dust is often dominated by silica and silicates.^22,43  Moreover, we expected that some of the tissues may exhibit endogenous deposition from the severity of lung injury visible under brightfield microscopy. Endogenous deposits are generally understood to include certain elemental enrichments, such as Ca and P associated with calcification and ossification in response to lung injury,^53,54  or Fe associated with hemosiderin deposition⁵⁵ ; and enrichments of other elements such as Na and K can also be associated with endogenous processes.⁴⁷  However, any of these elements might also be associated with exogenous particles, which can complicate classification. Considering all the above factors, we took a modified approach to establish classification criteria, which was based on statistical clusters of the observed particle chemistries and is an approach commonly used in environmental health research to understand multiple exposures.⁵⁶  Our approach yielded reasonable classes in terms of likely particle origins (Table 1), while accommodating mixed particles owing either to agglomerates or hybrid origins (eg, dust with endogenous deposition). Notably, although just 9 elements were included in our criteria, additional elements could be included in future work to enable classification of less common particle types (eg, rich in metals other than Fe, Ti, or Al)—or for application to tissues from individuals with variable or unknown exposure history. Further, as more is learned about the ways in which biological processes affect elemental composition of lung particulates, classification schemes could also be improved to differentiate between endogenous and exogenous particles more definitively.

For the coal miner cases studied here, our automated FESEM-EDS procedure and particle classification criteria yielded results that generally fit with the understanding of mineral contents of RCMD. Cases 1 through 9 were dominated by aluminum silicate and silica particles; and while case 10 was unique in terms of exhibiting very high percentages of endogenous particles, the exogenous particles were also dominated by silicates and silica. Respirable silica has been widely reported as a key factor in the most severe forms of CWP, including PMF.^2,27  The role of other silicate minerals has received considerably less attention, but may nevertheless be important in the pathology of mixed-dust pneumoconiosis.⁵⁷  The digested-lung particle analysis also showed the predominance of silicates and silica in all cases, though it often indicated higher ratios of silica to total silicates (Figure 7, B). This might be related to interference between particles during the in situ analysis; because silica particles are classified by their characteristic absence of elemental content other than Si, they might be misclassified when clustered with other particle types (eg, Sil-Al). The per-section results for cases 1 and 5 support this hypothesis. As mentioned, for both cases, the section with higher particle number density appeared to have a relatively lower percentage of silica particles but a higher percentage of Sil-Al particles (Supplemental Table 4). Clustering of particles in lung tissue is often observed in situ^38,39  and may be attributed to biological mechanisms (eg, phagocytosis)—although it is possible that some clusters represent agglomerates that were present in the coal-mining environment.⁵⁸  Animal studies have shown that exogenous mineral particles of varying size introduced into the alveolar lumen form large aggregates that can be taken up whole into alveolar epithelial cells and alveolar macrophages by caveolin-initiated micropinocytosis.⁵⁹  The process of aggregation in the alveolar lumen is facilitated by entanglement in the tubular myelin of alveolar surfactant, which seems to facilitate the endocytosis.⁶⁰ 

Particle recovery and redeposition procedures used for the digested analysis are expected to disperse particles, which thereby reduce interference from adjacent particles during FESEM-EDS work. Particle clustering (versus dispersion) can also explain some of the other differences observed between the in situ and digested analysis results—most notably, differences in particle size and number density (Figure 7, C and D, respectively). The tendency of the in situ results to show higher percentages of Sil-Al,Ti particles could be due to interference between clustered silicate and Ti-rich particles.

We interpret the Ca,P-rich and Na-rich particles observed during in situ analysis of the miner and control cases to be primarily endogenous in nature. While environmental exposure to Ca,P-rich mineral dust (eg, apatite) is possible, the co-occurrence of Ca,P-rich and Sil-Al,Ca,P particles in the miner lung tissues suggests endogenous deposition of Ca and P. Such deposition might form Ca,P-rich particles³⁵ ; it might also form a Ca,P-rich layer on dust particles in the lung tissue, or occur in clusters with dust particles, resulting in a hybrid classification (eg, Sil-Al,Ca,P). The Ca,P-rich particles we observed generally had a characteristically round shape relative to other particle types, which is consistent with previous observations using manual FESEM-EDS to investigate lung particles in situ.³⁸  The fact that neither Ca,P-rich nor Sil-Al,Ca,P particles were observed in the digested-lung particle analysis is notable. A likely explanation is that endogenously deposited Ca,P species were removed during the tissue digestion.⁶¹  Indeed, Ca,P-rich particles were virtually absent from the digested analysis results for all coal miner cases (n = 95) included in the larger dataset published in 2023 by Lowers⁵⁰  that we drew from for the current study.

Similarly, Na-rich particles were virtually absent from the digested analysis results, whereas they were identified in situ in both miner and control tissues. The in situ analysis also showed higher percentages of Sil-Al,Na particles than the digested analysis for the miner cases (Figure 7, A), and particles classified as “others” in some control cases that were rich in both Na and P. Taking these observations together, it seems that Na content found in situ often represents endogenous enrichment, which can occur either as localized deposition on its own, or in association with exogenous or endogenous particles. But, like deposits of Ca and P, endogenous Na is probably removed during the tissue digestion procedure.

Particles identified in situ with considerable Fe content might also have endogenous or hybrid origins. As noted, particles we classified as Fe-rich often exhibited relatively high content of Na, P, and/or K (see Supplemental Figure 4), which might indicate endogenous deposition. Other Fe-rich particles exhibited Si and Al content, but in ratios that suggest the particles may be exogenous (eg, silicate with a layer of Fe-rich deposition).

Also relevant to interpretation of the results is the possibility of contamination—which could occur during sample preparation for either the digested-lung particle analysis or the in situ analysis. For the former, inclusion of blank samples in the original data collection by Lowers⁵⁰  allowed us to identify and minimize the effects of contamination on the digested particle data investigated here. For the in situ analysis, we were able to use the data associated with FESEM-EDS fields outside the tissue area to assess contamination. Results confirmed that relatively few particles were identified on “blank” areas (Supplemental Table 1). Indeed, a total of just 84 particles were identified in 40 unique fields across all 20 runs of the automated FESEM-EDS procedure (ie, a single run on each of the 12 planchettes with miner tissues, and 2 runs on all 4 planchettes with control tissues.) Based on the EDS data, these particles were most commonly dominated by Na, Ca, and/or K (n = 30), which might indicate salt precipitates following deparaffinization. Other particles appeared to be either mineral dust (silicates or silica) that might have been dislodged from the tissue during planchette preparation, or Fe- or Ti-dominant particles that might be due to contamination from labware. Notably, one of the blank fields included in the second run of case C2 included 14 particles dominated by Si; but these were morphologically dissimilar to silica dust particles observed within the lung tissue and appeared to be Si-based fibers. Nevertheless, our results generally indicate that contamination, while possible, can be minimized with careful preparation of samples for either analysis method—and that inclusion of blanks (digested analysis) or blank areas (in situ analysis) is valuable for data quality assurance.

A key strength of the work presented here is the approach to particle classification using the data derived from the automated in situ analysis, as discussed above. Our approach was tailored specifically for analysis of coal miner lung tissues, and the classification criteria established can be applied to additional cases in the future. Moreover, the approach could be applied to develop different or expanded criteria for evaluation of tissues from individuals with other types of exposures. For example, dust exposures for hard-rock miners could include silica and aluminum silicates, as well as additional minerals like metal oxides and sulfides. Welders and metal workers, engineered-stone workers, and ceramics workers might also be expected to have predictable dust exposures related to their specific occupation. For individuals with more varied or unknown exposure histories, an approach that includes EDS data collection across a wider range of elements would be preferred.

Another strength of this study is the inclusion of control cases, which provide important reference data in terms of particle densities and characteristics for the miner cases. Nevertheless, in future studies, it may be possible to select controls with particular characteristics (eg, individuals from the same geographic region as occupational cases) to support more specific conclusions about exposure sources and effects. Likewise, using a standardized methodology, future work could also seek to draw comparisons between coal miners from different regions, work eras, or conditions; or between coal miners and workers in other dusty occupations.

The case-by-case comparison of the miner tissue results between the automated in situ analysis and the digested analysis is an additional strength of this study. This aspect was critical both for interpreting and validating the in situ analysis results. The case-by-case comparison highlights some important differences in the capabilities of each method as well. Most notably, the in situ analysis enables referencing of results back to specific areas of interest on the tissue specimen (eg, PMF versus non-PMF regions in the current work), but this capability does not exist (or may be more limited) for methods that rely on recovery of particles from tissue before analysis.

Additionally, the in situ analysis can enable observation of endogenous deposits including those enriched in Ca, P, Na, and Fe, whereas evidence of such deposits may be lost during sample preparation for the digested analysis. These endogenous deposits can provide insights into the pathogenesis of the pneumoconioses, particularly on the roles of Fe and Ca in inflammation, autophagy, apoptosis, and disease progression. On the other hand, the digested analysis can enable evaluation of carbonaceous (eg, coal dust) particles, but this might not be possible with the automated in situ analysis and was not attempted here. Importantly, both the in situ and digested methods used automated FESEM-EDS procedures, which included a predefined grid of fields for data collection, thus avoiding user bias in selection of the fields for analysis.

Limitations of the current study include the relatively small sample size (10 coal miner cases) and the limited range of case types included (ie, only miners with confirmed PMF). Indeed, most of the tissue sections evaluated in the study exhibited such severe disease that even areas demarcated as non-PMF (in the lung parenchyma) had clear collagen formation and appeared to contain both immature and mature silicotic nodules; this likely confounded comparisons between PMF and non-PMF areas, which showed differences in particle number density but typically not mineral distributions (Figure 5). This underscores the importance of the control cases for reference in this study. We also acknowledge that the study was not designed for a rigorous assessment of method precision. While it includes limited results for instances where 2 tissue sections were analyzed for the same case (miner cases 1 and 5) or a single section was analyzed twice (control cases C1–C4), this work should be expanded moving forward in order to better evaluate precision and set targets for total analysis area based on particle density or other factors.

The current work demonstrates that automated FESEM-EDS can be conducted in situ for analysis of inorganic particles in lung tissue. For particle classification, an approach that accounts for particle clustering (and the resultant mixed EDS spectra) may be advantageous, especially for analysis of tissues with relatively high particle number density. Using such an approach here, the in situ results for the coal miner tissues—which had much higher densities than controls—indicated that particle counts were typically dominated by aluminum silicates with varying percentages of silica. This finding is consistent with expectations based on knowledge of coal mine dust, and is also in agreement with results from separate analysis of particles recovered from digested tissue sections that had been serially cut from the same cases.

A primary advantage of the in situ method over those requiring recovery of lung dust particles for analysis is that the in situ results have the potential to be correlated with specific lung tissue areas or features of interest (eg, the PMF lesions identified on brightfield microscopy images in the current work). Moreover, this approach provides insight into processes that reflect the lung’s complex and still poorly understood response to particle-induced injury. This has important implications for investigation of the histopathogenesis of dust-related lung diseases in general, and could provide new opportunities to better understand exposure-related PMF in miners in particular.

The in situ FESEM-EDS analysis was performed at the Nanoscale Characterization and Fabrication Laboratory, which is supported by the Virginia Tech National Center for Earth and Environmental Nanotechnology Infrastructure (NanoEarth), a member of the National Nanotechnology Coordinated Infrastructure (NNCI), supported by the National Science Foundation (Electrical, Communication and Cyber Sytems Award numbers 1542100 and 2025151).