Linking Ambient Fire Emissions to Respiratory Effects Using Gelatin Filter Sampling

Abstract

On November 8, 2023, a fire occurred at a chemical plant near Shepherd, Texas. Ambient air monitoring using a novel, biocompatible, gelatin filter alongside proton transfer reaction time-of-flight mass spectrometry (PTR-ToF-MS) with both targeted and non-targeted analysis (NTA) was conducted on November 10, 2023, with seasonally matched monitoring on November 12, 2025. Concentrations of benzene, toluene and ethylbenzene/xylenes were higher near the incident site (Shepherd) compared to downwind (Livingston) following the fire. NTA revealed greater numbers of VOCs, and unique site-specific VOCs, at the incident site versus downwind just after the fire in 2023. In 2025 the number and types of VOCs did not vary across sites, suggesting an apparent site-specific VOC enrichment during the monitoring period in Shepherd post-fire. The gelatin filter, positioned upstream of the PTR-ToF-MS, enabled direct capture of airborne mixtures and subsequent dissolution into cell culture media without additional extraction steps. Trace metal analysis of these filters showed a higher burden of metals, as well as unique (barium, thulium, and samarium) and elevated metals (sodium, magnesium, iron, copper, gallium, strontium, tin, and cesium) in filters collected at Shepherd. The filtrates were then applied to human bronchial epithelial cells. Exposure to incident site filters significantly upregulated the oxidative stress-responsive gene HMOX1. These findings demonstrate that short-term exposure to real-world air pollutant mixtures induces biological responses in airway cells and highlight the utility of integrating air monitoring and novel sampling strategies with in vitro toxicity/biological activity assessment following episodic emission events.

1. Introduction

On November 8, 2023 the chemical plant Sound Resource Solutions caught fire outside of Shepherd, Texas. The Environmental Protection Agency (EPA) Region 6 responded the same day, deploying air monitoring and imposing a five-mile shelter-in-place (EPA, 2023). Handheld multi-gas meters, PM_2.5 measurements, and the airborne spectral photometric environmental collection technology (ASPECT) aircraft all showed ambient air pollutant levels below the site-specific action levels, and the shelter-in-place was lifted at 4:00 pm on the same day (EPA, 2023). A detailed sampling map can be seen within the cited EPA report (EPA, 2023).

On November 10, as shown in Figure 1, the Texas A&M University Mobile Responding to Air Pollution in Disasters (mRAPiD) van was deployed to the Sound Resource Solutions plant fire to conduct ambient air monitoring using proton transfer reaction time-of-flight mass spectroscopy (PTR-ToF-MS). A novel gelatin filter trap was placed upstream of the PTR-ToF-MS to capture pollutants emitted from the fire and evaluate their ability to induce biological effects in human bronchial epithelial cells in a laboratory setting. The condition of the incident site in Shepherd at the time of the mRAPiD deployment is shown in the supplement (Figure S1).

Figure 1.

Timeline of the Sound Resource Solutions chemical plant fire in Shepherd, Texas, and the coordinated mRAPiD rapid-response ambient air monitoring campaign. The timeline depicts the date of the fire (November 8, 2023), initial deployment of ambient air sampling by Texas A&M University (November 10, 2023), and follow-up sampling conducted two years post-incident (November 12, 2025). Key response activities, including site selection, gel membrane filter deployment, and sample collection for chemical analysis and in vitro toxicity assessment, are indicated.

A handful of research groups have previously reported direct ambient exposures in the field during non-disaster times to determine in vitro biological effects on human lung cells e.g., (Bisig et al., 2018; Vizuete et al., 2015). Vizuete et al. demonstrated that four hours of ambient exposure near the Houston Ship Channel yielded robust changes in gene expression in alveolar epithelial cells. Similarly, Bisig et al. demonstrated that 12 hours of ambient exposure in Fribourg, Switzerland resulted in the upregulation of oxidative stress or pro-inflammatory responses. Thus, it is feasible that short-term ambient exposures yield discernable biological effects. However, implementing direct ambient exposure approaches during acute disaster events is inherently challenging due to rapidly evolving emission sources, constrained access to impacted areas, and the absence of pre-positioned infrastructure to carry out monitoring of direct exposure studies.

Gelatin filters similar to those used in this study are traditionally deployed to collect bacterial or viral aerosols in the field (Burton et al., 2007; Jaschhof, 1992; Lewandowski et al., 2013). A key element of these filters is their high-water solubility, which allows them to be easily dissolved in cell culture medium immediately following ambient air sample collection. As such, the gelatin filter offers the practical advantage of less labor and materials needed for the sampling and processing of the filter before cellular applications. Still, it is unclear how effective these filters are at collecting ambient VOCs, trace metals, and particulate matter. Generally, volatile organic compound (VOC) and fine particulate matter levels are known to increase following environmental disasters (Huang & Skidmore, 2025; Saitas et al., 2025). However, there is limited information about the composition of VOCs and other air pollutants following unique, short-term disasters such as the Sound Resource Solutions plant fire (Kulkarni, Chellam, & Fraser, 2007; Sloop et al., 2023). Accordingly, the primary objectives of this research were to define the acute inhalation hazard posed by a real-world disaster-related air pollution mixture and to demonstrate the utility of a novel in vitro platform for rapidly assessing biological responses to corresponding ambient air samples.

2. Methods

2.1. Sampling Strategy

Air monitoring was conducted on November 10, 2023 and November 12, 2025 at three sites: Oakhurst, Shepherd (incident site), and Livingston as shown in Figure 2. Mobile air monitoring was conducted at both Oakhurst and Livingston, whereas stationary air monitoring was conducted at the incident site (Shepherd). Oakhurst was selected as a background site due to its off-wind geographic location and similar population density to Shepherd. Livingston (15 miles from the incident site) was selected due to its downwind location of the incident site, and it was outside of the 5-mile evacuation zone.

Figure 2.

A visual of the sampling route taken from Google maps. Mobile sampling (shown as a red line) was conducted with the PTR-ToF-MS at the background and downwind sites. Stationary monitoring was conducted at the incident site. This route was conducted 11/12/2025 and is believed to be accurate to the 11/10/23 route which was confirmed with iPhone location data and field notebooks of the original mRAPiD team members since GPS data was unavailable for the 11/10/23 route. Wind direction is only for 11/12/2025; wind direction for 11/10/23 was primarily from the north during sampling times as confirmed by TCEQ Conroe air monitor station and NOAA HYSPLIT satellite model (Supplemental Figure #).

On November 10, 2023, all sampling occurred between 12:11 pm and 2:35 pm CST. On November 12, 2025, all sampling occurred between 11:08 am and 1:38 pm. Humidity, temperature, and GPS data were only collected on November 12, 2025 due to file corruption on the Magellan software from November 10, 2023. However, for November 10, 2023 wind direction was modeled using the National Oceanic and Atmospheric Administration (NOAA) HYSPLIT software, and temperature data was pulled from the TCEQ Conroe (CAMS: 0078) air monitoring station (Figure S2. and Table S3.). Extensive details on the mRAPiD van and associated characterization of VOC emissions are found in Vitucci et al. (Vitucci et al., 2025).

2.2. VOC Emissions Measurements using PTR-ToF-MS

In short, the mRAPiD van is fitted with a Teflon sampling tubing inlet about 3.5 meters from the van. A pump on board the mRAPiD van pulls air through the exterior sampling inlet into a sampling line (tubing with a 0.5” outer diameter) that is directed to an exhaust port that enables the exterior air to exit the mRAPiD van in a closed system (Figure 3A). Additional tubing connects this main sampling line to the filter tubing apparatus shown in Figure 3B. This attaches to the sample inlet of the PTR-ToF-MS 4000 trace gas analyzer (IONICON, Inc.), which consists of 1/16” outer diameter PolyEtherEtherKetone (PEEK) tubing. A mechanical backing pump pulls ambient air through the PEEK tubing, allowing for ambient air to pull into the PTR-ToF-MS for VOC monitoring. Detection of VOCs was done at an inlet flow rate of 200 sccm in hydronium ([H₃O⁺]) mode (Td = 120) with 1 second sampling interval. A mixture (1,3-butadiene, acrolein, benzene, toluene, styrene, ethylbenzene, xylenes) calibration gas tank was used for calibration along with Ultra Grade Zero Air (Airgas, Inc). Calibration curves were used for targeted analysis of benzene, toluene, and ethylbenzene/xylenes. The calibration curves can be found in Table S1 and S2.

Figure 3.

(A) A pump on board the mRAPiD van pulls air through the exterior sampling inlet into a sampling line (tubing with a 0.5” outer diameter) that is directed to an exhaust port that enables the exterior air to exit the mRAPiD van in a closed system. (B) A photograph of the tubing unattached to the PTR-ToF-MS. Air continuously flows from the inlet into the PTR; however, during filter collection an additional air flow path (marked with red arrows) is made available to expose a contained gelatin filter to ambient air.

2.3. mRAPiD Sampling of Ambient Emissions by Gelatin Filters

The gelatin filters (Sartorius 12602–47-ALK) are individually packaged, sterile, and marketed for collection of airborne microbes and viruses. The filters were placed inline of the PTR-ToF-MS while on board the mRAPiD van (described above), and ambient air was sampled through the filters for 30–45 minutes at each site location. Immediately after sample collection, the filters were removed from the holder with clean tweezers and dissolved in 50 mL of basal Dulbecco’s minimum essential medium (DMEM) in a 50 mL conical tube that was capped with minimal headspace. The 50 mL conical tubes were then wrapped in Parafilm and stored in a Styrofoam packing container on ice for travel. At the end of the sampling day, the 50 mL conical tubes were separated into 2 mL aliquots and stored at −80 °C until further analysis. Note, gelatin filters were dissolved in 50 mL of DMEM as this volume resulted in complete dissolution of the gelatin filter while minimizing the dilution of any environmentally bound agents on the filters. This volume also minimized the amount of headspace in the tube, decreasing the likelihood that highly volatile VOCs would volatilize out of solution.

2. 4. VOC Quantification: Non-targeted Analysis

A complete description of the non-targeted analysis can be found in a previous manuscript (Vitucci et al., 2025). In short, the ion signals of compounds with counts per second (cps) thresholds of 5 or above were converted to final VOC concentrations in parts per billion by volume (ppbv) in the PTR-MS Viewer software using the following Equation 1:

Equation 1:

“X” is the reaction time between each analyte and hydronium (H₃O) was constant. “k” is a default rate constant of 2×10⁻⁹ cm⁻³ molecule⁻¹ s⁻¹ for each analyte as a known k value was not available for every detected analyte. This constant has been widely accepted in atmospheric chemistry literature when k is not directly available because it results in measurements of uncalibrated compounds with an accuracy of ±30 % (Pagonis et al., 2019; Rupert Holzinger, 2019; Vitucci et al., 2025).

$$\left(d\left[X{H}^{+}\right]\right)/dt=k\left[H3O\right]\left[X\right]$$

2.5. VOC Characterization and Quantification: Targeted Analysis

The calculated concentrations for benzene, toluene, and ethylbenzene/ xylenes were quantified as shown in Equation 1 then multiplied by a predetermined calibration factor that was derived from a calibration gas to account for the instrument-specific variability. (Tables S1 and S2).

2.6. VOC Hazard Assessment

A hazard assessment was generated from the unique VOCs detected at each sampling site according to our previously reported protocol using the EPA CompTox Chemicals Dashboard (Vitucci et al., 2025). Briefly, predicted chemicals were identified using the CompTox Dashboard, and any chemical not identified by the CompTox Dashboard was excluded from the hazard assessment. All identified compounds were added to the EPA hazard comparison module, which provides summarized hazard data for each compound in relation to human health effects. None of the unique VOCs had known acute mammalian inhalation or dermal toxicity according to the EPA hazard comparison module. The full table is shown in the supplement (Table S4).

2.7. Trace Metal Analysis

Aliquots (1 mL), field blanks, and laboratory blanks were processed using a three-stage closed-vessel digestion protocol in a microwave oven (MARS 6, CEM Corporation) (Kulkarni, Chellam, et al., 2007a; Kulkarni, Chellam, & Mittlefehldt, 2007). Samples were first digested with 3 mL HNO₃ (~70%, TraceMetal^™ grade, FisherScientific) and 0.02 mL HF (~48%, TraceMetal^™ grade, FisherScientific) and heated to 130 °C for 26 minutes to avoid rapid pressure buildup. Next, the contents were ramped to 200 °C with a dwell time of 20 minutes to dissolve the siliceous matrix. Finally, any remaining HF was masked, and fluoride precipitates were redissolved by adding H₃BO₃ (5% m/v, ≥99.5% purity, Sigma Aldrich) at 8-times the HF volume (i.e., 0.16 mL) and heating to 200 °C with a dwell time of 20 minutes. After cooling to room temperature, the resulting digestate was diluted with ultrapure water to a final HNO₃ concentration of 2% prior to inductively coupled plasma – mass spectrometry (ICP-MS, PerkinElmer^® NexION^® 300) analysis. Fifty-three elements (Al, As, Ba, Be, Ca, Cd, Ce, Co, Cr, Cs, Cu, Dy, Er, Eu, Fe, Ga, Gd, Hf, Ho, K, La, Li, Lu, Mg, Mn, Mo, Na, Nb, Nd, Ni, Pb, Pr, Rb, Sb, Sc, Se, Si, Sm, Sn, Sr, Ta, Tb, Th, Ti, Tl, Tm, U, V, W, Y, Yb, Zn, and Zr) were quantified under optimized operating conditions. Instrumental settings and operating parameters are provided in Table S5. Plasma-based isobaric interferences and spectral overlaps from polyatomic species to key aerosol metals such as Al, V, Cr, Fe, Ni, Cu, and Zn were suppressed by exploiting ion-molecule reactions in a collision cell with NH₃ as the cell gas.

The accuracy and precision of this procedure were evaluated through analytical recoveries of certified and uncertified elements in Standard Reference Materials SRM 1633b (Coal Fly Ash) and SRM 1648a (Urban Particulate Matter). Independently prepared calibration check standards (1 and 50 μg/L) were analyzed periodically throughout each run to verify calibration accuracy, instrumental response stability, and reproducibility across the entire analytical range. Recoveries for SRM 1633b and SRM 1648a were within 80–120% (Tables S6 and S7), while calibration check standards remained within 90–110% (Table S8). Prior to each run, autotuning was performed with a 10 ng/mL multielement solution (Be, In, U, Ce) to span low- to high-mass regions. Internal standards (⁷²Ge, ¹¹⁵In, and ²⁰⁹Bi at 10 μg/L) were used to track instrumental drift and sensitivity and to correct non-spectral interferences.

Method detection limits (Thomas, 2008) for the targeted elements are summarized in Table S9. Additionally, laboratory blanks were analyzed at regular intervals to check any potential carryover and baseline intensity of each targeted element (Das, McEwen, et al., 2023; Das, Prospero, & Chellam, 2023; Kulkarni, Chellam, et al., 2007b). Additionally, mid-level calibration standards were analyzed at regular intervals throughout each run to monitor instrumental drift and measurement uncertainty, with a ±10% acceptance window applied for drift checks (Danadurai et al., 2011). Elemental concentrations were converted from μg/L to μg/m³ using Equation 2.

Equation 2.

Formula for converting concentration ($\mathrm{C}$) from μg/L to μg/m³.

$$\begin{array}{l}\hfill C_{\mu g/m^3}=\frac{C_{\mu g/L}V}{Qt}\hfill \\ \hfill \mathrm{Q}=0.012{\mathrm{m}}^3/\mathrm{h}\hfill \\ \hfill \mathrm{t}=\text{sampling}\text{time}(\mathrm{h})\hfill \\ \hfill \mathrm{V}=\text{volume}\text{of}\text{media}\text{used}\text{to}\text{dissolve}\text{gelatin}\text{filter}(\mathrm{L})\hfill \end{array}$$

2.8. Cell Culture Methods

Human bronchial epithelial cells (16HBEs) were seeded onto 24-well transwell inserts at a seeding density of approximately 60,650 cells/cm². After 48 hours, cells reached 100% confluency and were exposed to 500 μL of either sample filtrate (field blank, Oakhurst, Shepherd, or Livingston) or a positive control, 100 μg/ml of diesel exhaust particulate matter (DEPM SRM 2975 NIST). The field blank was a gelatin filter dissolved in cell culture media in a 50 mL conical tube that was not placed in line of the sampling setup but was packaged and carried with the field samples for the duration of the field campaign from set-up to storage.

Cells were conditioned in their assigned treatment for 24 or 48 hours before lysis with TRIzol for further analysis. Additionally, exposure media was collected for the lactate dehydrogenase (LDH) assay and stored at −80 °C until further analysis.

2.8.1. Cell Culture Materials

Dulbecco’s Minimum Essential Medium Eagle Media, phosphate buffered saline (PBS), and penicillin streptomycin solution were purchased from Sigma-Aldrich. 12-well transwell insert plates (#230615) were purchased from Celltreat Scientific Products. Trypsin (1:250), fetal bovine serum (FBS), and GlutaMAX (100X) were purchased from Gibco ThermoFisher Scientific. The immortalized human bronchial epithelial cell line 16HBE was purchased from Sigma-Aldrich (Catalog # SCC150). The Abcam colorimetric lactate dehydrogenase (LDH) assay kit was purchased for measuring cytotoxicity (ab102526). TRIzol reagent was purchased from Invitrogen. The QuantiTect reverse transcription kit was purchased from Qiagen (Catalog No. 205313).

2.8.2. Taqman RT-qPCR

At 24- or 48-hours post-exposure, cells were lysed with TRIzol reagent, and the lysate was collected and stored at −80 °C until further analysis. Next, the samples were thawed on ice, and the Qiagen QuantiTect reverse transcription kit was used to extract mRNA from the lysate and synthesize cDNA according to manufacturer instructions. Taqman RT-qPCR was then conducted to determine the relative expression of interleukin-8 (IL-8), tumor necrosis factor-alpha (TNFα), heme oxygenase-1 (HMOX-1) and NAD(P)H quinone dehydrogenase-1 (NQO1). The list of primers and probes can be found in the supplement (Table S10).

2.8.3. Statistical Analysis of Gene Expression Data

Gene expression data was analyzed using the PRISM software (Version 10.6.1). An ordinary Gaussian one-way ANOVA was used to compare the relative expression of a given gene in cells exposed to condition media containing a dissolved filter collected at a single sampling site or the field blank. A p-value < 0.05 was considered significant.

3. Results

3.1. Targeted Detection of Benzene, Toluene, Ethylbenzene/Xylenes (BTEX)

On November 10^th 2023, levels of benzene were elevated at the incident site (Shepherd) compared to Livingston following the fire. The median concentration of benzene at both sites (4.47 ppbv and 1.03 ppbv, respectively) fell below the ATSDR acute inhalation minimal risk level (MRL) of 9 ppbv as shown in Figure 4 (ATSDR, 2026). Benzene was not detected at any of the sites sampled on November 12^th 2025. Toluene was detected at the incident site (Shepherd) but not downwind (Livingston) following the fire on November 10, 2023. Toluene was detected at concentrations below the ATSDR acute inhalation MRL of 2000 ppbv on November 10, 2023, as well as November 12, 2025. In 2025, the Shepherd and Livingston sites had slightly higher detected levels of toluene compared to the background site.

Figure 4.

Concentrations of BTEX detected at sampling sites November 10, 2023 and November 12, 2025. ND = below level of detection. U = unavailable. **** = p < 0.0001. A Welch’s t-test was used for the 2023 data set whereas a Brown-Forsythe and Welch ANOVA test was used for the 2025 data set. Error bars represent standard deviation of the mean. Ethylbenzene/xylenes represents the compound C₈H₁₀ which can include ethylbenzene or mixed xylenes (o – xylene, m – xylene, p – xylene) since the PTR-ToF-MS cannot distinguish between compounds of an identical mass. The blue dotted line represents the June 2026 ATSDR acute inhalation minimum risk level (MRL) for o/m/p-xylene (2 ppm) and the black dotted line represents the June 2026 ATSDR acute inhalation MRL for ethylbenzene (5 ppm).

Similarly, ethylbenzene/xylenes were detected at elevated levels at the incident site (Shepherd) compared to Livingston following the fire (46.10 ppbv and 5.75 ppbv, respectively). Again, the median concentrations of these VOCs fell below the ATSDR acute MRL for either compound. On November 12, 2025, all three sites had similar concentrations of ethylbenzene/xylenes, all below the ATSDR MRLs.

3.2. Non-targeted Analysis of VOCs

Following the Sound Resource Solutions plant fire in November 2023, a total of 29 VOCs were detected only at Shepherd and 8 VOCs only at Livingston relative to the available 2023 datasets, as shown in Figure 5. Because the Oakhurst background site was not available for NTA analysis in 2023, these VOCs represent compounds unique only within the Shepherd–Livingston comparison and may not fully exclude presence at broader regional background locations. In 2025, there were 2 unique VOCs detected in the background site. Overall, detected VOCs were nearly double at all sites during the follow-up sampling compared to the samples taken immediately after the fire, which was likely due to instrumental cleaning (described in detail in the discussion).

Figure 5.

A bar chart showing the number of unique, duplicate or common VOCs between sampling sites. Unique VOCs (black dotted box) were only detected at one site, duplicate VOCs (vertical lines) were detected at two sites, and common VOCs (white box) were detected at all three sites. The November 10^th 2023 graph does not show any shared VOCs because the Oakhurst PTR-ToF data is unavailable (denoted by X).

3.3. Evaluation of Filter Capture of Trace Metals

The total concentration of metals detected at the incident site (Shepherd) in 2023 was higher than either the Oakhurst or Livingston sites. As shown in Figure 6, eight different metals had more than a 2-fold greater concentration at the incident site (Shepherd) than the background site, including sodium (Na), magnesium (Mg), iron (Fe), copper (Cu), gallium (Ga), strontium (Sr), tin (Sn), and cesium (Cs). In comparison, at Livingston, none of the metals presented at more than a 2-fold greater concentration than the background site. However, two metals, Mg and Cs, had a more than 1.5-fold greater concentration in the filter collected at Livingston. During follow-up sampling in 2025, total metal concentrations were lower across all three sites compared to the 2023 concentrations (Figure 6). Additionally, the total concentrations at Shepherd were similar to the background site at Oakhurst; these sites shared 13 elements in common that were not found in the 2025 Livingston samples, including Be, K, Co, Ga, Sr, Mo, Cd, Sb, Eu, Gd, Tb, Ho, and Er.

Figure 6.

(A) Total elemental mass between sites and years is displayed on a bar graph to illustrate the increased incidence of metals in the filter collected at the Shepherd site compared to the Oakhurst and Livingston sites in 2023. Total mass at all sites decreases from 2023 to 2025. (B) Heat maps displaying the fold-change in metal concentration (μg/m³) in filters collected at Shepherd and Livingston compared to the background site of Oakhurst, TX on November 10, 2023 and November 12, 2025.

A handful of metals were unique between sampling sites as shown in Figure 7 and Table S11. In 2023, the incident site (Shepherd) had the highest number of unique metals detected with barium (Ba), thulium (Tm) and samarium (Sm) only being detected at Shepherd. The concentration of Ba was 190.00 ± 1.294 μg/m³. The concentration of Tm was 0.86 ± 0.029 μg/m³, and the concentration of Sm was 0.91 ± 0.004 μg/m³. Neodymium (Nd) and dysprosium (Dy) were only detected at the Shepherd and Livingston sites, and copper (Cu) was only detected at the Shepherd and Oakhurst sites. Chromium (Cr) was only detected at the background site at a concentration of 8.36 ± 0.127 μg/m³.

Figure 7.

A Venn diagram showing unique metals found between sampling sites. Metals prevalent in all three sites are not included (black center): Al, Na, Mg, Si, Ca, Sc, Mn, Fe, Co, Ni, Ga, As, Rb, Sr, Y, Zr, Nb, Mo, Sn, Cs, La, Ce, Pr, Eu, Gd, Tb, Ho, Er, Yb, Lu, Hf, Ta, W, Tl, Pb, Th, and U.

To investigate the potential for the gelatin filters to potentially also capture VOCs at environmentally relevant concentrations during sampling the quantity of detected, non-targeted compounds and the concentration of targeted compounds were compared during sampling through and not through the filter flow path on the second visit (Figure S3A – S3C). The potential of the filters to bind VOCs at elevated, non-ambient concentrations was also evaluated in a laboratory setting using the common VOC, benzene (Supplementary Method 1) (Figure S4A and S4B). Non-targeted analysis revealed a similar number of VOCs detected across sampling sites, irrespective of sampling through or not through the filter flow path. The concentration of the targeted compound, toluene, significantly decreased by 60% and 35% when air was sampled through the flow path containing the filter compared to air sampled from a similar location without the filter. This trend was repeated for ethylbenzene/xylenes when air was sampled from the background site, but this trend was not repeated at the Shepherd location. In the laboratory setting, benzene concentrations measured by the PTR-ToF-MS decreased by at least 35% when a low dose, approximately 30 ppb benzene, was sampled through the flow path containing the gelatin filter versus the flow path without the filter. In contrast, no significant decreases in benzene concentrations were detected when a high dose, approximately 170 ppb benzene, was sampled through the flow path containing the gelatin filter compared to no filter. Importantly, average background concentrations of benzene in the Ultra Grade Zero Air with and without the gelatin filter were under 1.5 ppb, suggesting the filters are not a large source of benzene. As the number of VOCs detected stayed relatively constant when air was sampled through the filter path, it is likely that the gelatin filters do not result in complete sorption of VOCs, and thus blockage of read-out from the PTR-ToF-MS. Instead, as BTEX decreased when low and/or environmentally relevant concentrations were sampled through the filter path, these data suggest the gelatin filters have the potential for partial VOC sorption at low concentrations. However, binding may become saturated and ineffective at higher concentrations.

3.4. Biological Effects of Air Samples

16HBEs were exposed to condition media that contained a dissolved gelatin filter collected from a single sampling site (Oakhurst, Shepherd, or Livingston) or the field blank for 24- or 48-hours. Cell lysate and culture media were collected after each exposure. Cytotoxicity was assessed using the LDH assay. No significant changes to cell viability were observed at 24- or 48-hours. The LDH assay results are shown in the supplement (Figure S5).

3.4.1. Gene Expression – Taqman RT-qPCR

16HBEs exposed to condition media containing the dissolved filter collected from the incident site (Shepherd) and the downwind site (Livingston) showed increased expression of the oxidative stress marker, HMOX-1, at 24 hours compared to the background site (Oakhurst). At 48 hours post-exposure, the Shepherd samples returned to baseline whereas the Livingston samples actually showed a slight downregulation of HMOX-1. At either timepoint, there did not seem to be significant changes to the relative expression of IL-8, TNFα, or NQO1 in the Livingston, Shepherd, and Oakhurst samples.

4. Discussion

The Sound Resource Solution plant fire resulted in the burning of approximately 6.62 million pounds of chemicals which were stored onsite (TCEQ, 2024). The company is involved in petroleum processing and specializes in blending and distributing oil-based products, as well as the production of solvents for glue and paint remover (Figure S6). Additionally, the company packages and transports petroleum-based products such as oil tank bottoms, grease and off-specification oil. Due to the multi-functionality of this facility, a complex mixture of potentially hazardous air pollutants was likely emitted following the fire. As such, there was a need for the timely quantification and identification of the complex mixture of air pollutants that would be emitted following the fire of such a multi-functional facility. Furthermore, there is a gap in our understanding of the effects that inhalation of complex pollutant mixtures has on human health. Thus, there is an unmet need to rapidly collect ambient air samples that can be easily applied to existing in vitro models for biological effect screening. To quantify and identify this complex mixture in a timely manner, mobile and stationary air monitoring were conducted using a PTR-ToF-MS to detect and characterize emitted VOCs surrounding this disaster. In addition, a novel filter apparatus was developed and employed during mobile and stationary monitoring to evaluate its potential as a tool that could capture real-world pollutant mixtures and for application to in vitro models for exposure assessments. Gelatin derived from mammalian collagen is a biocompatible matrix for the 16HBE cell line, so by using gelatin filters to capture air samples, intensive extraction methods are noy needed, as the case for traditional matrices likes activated carbon or polytetrafluoroethylene (PTFE) filters (Su & Wang, 2015). Additionally, those extraction methods also require multiple wash steps utilizing solvents, like methanol or acetonitrile, that are highly toxic to living cells. In contrast, the gelatin filters can be easily dissolved in water or cell culture media for ready application to mammalian cells.

At ~48 hours after the fire, a targeted monitoring of BTEX revealed higher levels at the incident site (Shepherd) compared to the background site as shown in Figure 4. The concentrations of BTEX at the incident site (4.47 ppb) were higher compared to concentrations of BTEX measured at other environmental disaster sites using the mRAPiD platform, including the Richmond, Indiana plastics recycling facility fire, where near-source benzene levels were generally at or below ~1 ppb (Vitucci et al., 2025). It is important to note that sampling by Vitucci et al. occurred approximately four days after fire initiation, when active combustion had largely subsided and evacuation orders had been lifted, which likely contributed to lower concentrations relative to peak emission periods.

Published measurements from large industrial fires (e.g., refinery, chemical storage, and tank fires) indicate that benzene can reach very high short-term concentrations, including tens to hundreds of ppb, with episodic peaks exceeding 1000 ppb near source during acute combustion phases (An Han et al., 2020). In contrast, wildfire studies generally report substantially lower concentrations, with most plume-impacted ambient measurements in the range of ~0.1–4 ppb and occasional short-term elevations up to ~10–25 ppb in fresh smoke or near plume cores (Dickinson et al., 2022). Overall, the benzene concentrations measured at the Shepherd incident site (4.47 ppb) fall below acute industrial fire peaks but within the upper range of wildfire smoke observations, consistent with measurements collected during a post-acute combustion phase when emissions are declining. In contrast, benzene was not detected at any site when follow-up air monitoring was conducted on November 12^th 2025. In addition, toluene and ethylbenzene /xylenes concentrations were comparable between all sites during the follow-up. As such, it appears that the elevated BTEX concentrations measured during the 2023 air monitoring may be related to potential emissions from the Sound Resource Solution plant fire.

In addition to the targeted BTEX assessment, a non-targeted analysis (NTA) was simultaneously conducted during mobile and stationary sampling ~48 hours after the incident. As disasters such as the Sound Resource Solution plant fire likely result in the release of a wide array of VOCs, NTA is a viable approach for identifying and characterizing complex mixtures (McCord et al., 2022; Phillips et al., 2022; Vitucci et al., 2025). Indeed, 33 individual VOCs were detected during the stationary sampling closest to the fire at the Shepherd location. This was a notably higher number of VOCs than those detected during the mobile sampling at a location several miles downwind from the incident site, along the Livingston route. With only four VOCs shared between Shepherd and Livingston, the Shepherd incident site exhibited a greater number of site-enriched compounds relative to the Livingston route. This higher number of compounds detected only at Shepherd (within the available 2023 comparison dataset) suggests localized enrichment of VOCs in closer proximity to the incident site and is consistent with a potential influence of fire-related emissions. However, given the absence of the 2023 Oakhurst background NTA dataset, these compounds cannot be definitively attributed as fire-specific and should be interpreted as unique only within the available spatial comparison framework. During follow-up air monitoring on November 12^th 2025, targeted (BTEX) and non-targeted VOC levels were comparable between sites. Notably, the number of detected masses were higher in the follow-up air monitoring when compared to the initial 2023 monitoring. This may primarily be due to a recent cleaning that removed carbon and other contamination build up from the ion source before the 2025 field campaign. Carbon and other contaminates act as nucleation sights for heavier compounds preventing passage to the mass spectrum analysis (Kang et al., 2017). While cleaning of the ion source is not mandatory before each use of the PTR-ToF-MS, the increase in compound detection in the 2025 datasets suggests this may have improved the sensitivity of the instrument and suggests that the 2023 dataset may be underestimating the number of detected masses across sampling sites, particularly those with higher masses. Additionally, all three sites in the 2025 dataset had a similar distribution of shared and duplicate VOCs with only the background site possessing unique VOCs during the follow-up sampling trip. This similarity in VOC profiles between sites in 2025 as opposed to immediately after the fire in 2023 further supports that the increase in VOCs detected in Shepherd may have been associated with the fire and that NTA is a valuable tool to investigate disaster related VOC emissions.

As demonstrated with the hazard assessment of the VOCs detected by NTA in the 2023 sampling campaign, as well as by others (Saitas et al., 2025; Vitucci et al., 2025), our understanding of the acute inhalation toxicity risk from exposure to a large number of VOCs is lacking (Supplemental Figure S4). This data gap is further emphasized when considering exposure to VOC mixtures as well as more complex environmental mixtures encompassing VOCs and additional disaster-emitted pollutants, such as particulate matter or metals. Indeed, the difficulty of recreating real-world complex mixtures in a laboratory setting can hinder these types of investigations (Bein et al., 2020). The gelatin filters used in this study were thus deployed to bridge mobile air monitoring in the field with in vitro toxicity testing since previous delivery applications of complex mixtures have multiple limitations, including poor solubility, compatibility with small testing volumes, and preferential binding of mixture components to cell culture media (Cordova et al., 2022).

While these filters are traditionally utilized for sampling of airborne bacteria and viruses, we investigated their utility in capturing the disaster-emitted pollutants during the 2023 sampling campaign and serving as an ex-vivo exposure source to investigate the exposure effects in human bronchial epithelial cells (Canpolat et al., 2022; Jaschhof, 1992). When the dissolved filtrates from each sampling site were applied to the bronchial epithelial cell line, 16HBEs, increased expression of the antioxidant gene HMOX-1 was observed only in cells exposed to filtrates collected from the Shepherd and Livingston sites at 24 hours. By 48 hours, the cells had returned to baseline or, in the case of the Livingston site, showed decreased expression of HMOX-1. Oxidative stress is a cellular stress response that can cause cellular dysfunction after prolonged redox disequilibrium. Cellular redox is regulated through numerous signaling pathways that included antioxidants and ROS scavengers like HMOX-1 and glutathione (Li et al., 2003). HMOX-1 is a cytoprotective enzyme that protects against oxidative stress-induced lung injury by catalyzing the degradation of oxidants (Fredenburgh et al., 2007). VOCs can induce oxidative stress and produce reactive oxygen species (ROS), which could activate HMOX-1 expression in the lung (Yan et al., 2023). Numerous air pollutants are known to induce oxidative stress and show an upregulation of HMOX-1 expression, including but not limited to PM_2.5, BTEX, and heavy metals (Kolawole et al., 2025; Merk et al., 2020; Sarma et al., 2011). Our investigation of the ability of the gelatin filters to sorb VOCs suggests that these filters may partially sorb VOCs at environmentally relevant concentrations. Thus, the elevated HMOX-1 expression in the cells may be a result of dissolved VOCs into the media that were originally sorbed onto the gelatin filter. Interestingly, this sorption efficiency was not reproduced at higher VOC concentrations in a laboratory setting, suggesting these filters have a binding capacity or adsorption limit. These filters largely rely on entrapment of sampled material in the pore structures within the gelatin. However, these filters have not been screened for compound-specific binding efficiency (e.g., polar vs nonpolar) nor the retention of a given VOC during storage. As such, the delivered dose to cells remains uncertain.

As previously stated, other air pollutants like metals can also induce oxidative stress (Li et al., 2008). Consequently, trace metal analysis was conducted on the filters to investigate the presence of any heavy metals. There is very little research assessing the trace metal composition in ambient air following environmental disasters. There are studies evaluating the metal concentration in the air around wildfires and electronic waste burn sites, but these studies primarily focus on particulate matter or the classical heavy metal toxicants rather than total trace element characterization (Baliaka et al., 2025; Boaggio et al., 2022; Gangwar et al., 2019; Kwarteng et al., 2020).

Cadmium (Cd) and selenium (Se) were not detected at any of the site, and lead (Pb) and manganese (Mn) levels at Shepherd and Livingston were less than a 1.5-fold change in concentration from the background site. Exposure to particulate matter and anthropogenic emissions often contain heavy metals that have been associated with increased oxidative stress and inflammatory responses in the human airways (Jin et al., 2025). However, there is limited research on whether many of the metals detected at elevated levels in this study are linked to adverse respiratory effects.

There were eight metals that were detected at the incident site at concentrations greater than a 2-fold change from the background site: Na, Mg, Fe, Cu, Ga, Sr, Sn, and Cs. The detection of many of these metals, Na, Mg, Fe, Cu and Sr, could be attributed to salt spray from the ocean, dust, and anthropogenic sources like waste incinerators (ATSDR, 2024; Ito & Miyakawa, 2023; Jahnen-Dechent & Ketteler, 2012; Ooki et al., 2002). Potentially, these metal concentrations were higher at Shepherd and Livingston than the background site due to increased human activity and/or less tree cover at those sites. Similarly, atmospheric sources of tin and cesium are usually due to dust or the burning of fossil fuels (ATSDR, 2004, 2005). Naturally occurring cesium is not radioactive and generally considered nontoxic. Gallium does not exist as a free element in nature, and is primarily found within bauxite deposits, which is a major source of aluminum metal (Schulte & Foley, 2014). As such, atmospheric sources of gallium are usually emissions from industrial smelting of aluminum products, electronic waste, and coal combustion (Erkmen et al., 2025; Rudnik, 2024).

The incident site, Shepherd, had more unique metals and a higher total metal concentration compared to both Oakhurst and Livingston sites as shown in Figure 6. Interestingly, many of these unique metals were rare earth elements (REEs). Tm and Sm, which were only detected at Shepherd, are both REEs that can be found naturally in the soil, but these metals are also commonly used as catalysts for chemical refining and found in electronics (Klinger, 2015). Similarly, Dy and Nd, which were only detected at the incident and Livingston site, are REEs that are used inside electronics, magnets, ceramics, and as catalysts for chemical refining (Klinger, 2015; Niinistö, 1987). It is possible Shepherd and Livingston, TX have a higher concentration of REEs in the soil than Oakhurst, which would be emitted into the air as dust from vehicle traffic or wind transported REEs from the Houston Ship Channel inland (Kulkarni, Chellam, & Fraser, 2007). However, it is also possible an increased incidence of REEs is indicative of the combustion of electronics and chemical manufacturing equipment. Inhalation exposure is the most common route of exposure to REEs (Han et al., 2018). Several REEs, including Dy, Sm, and Nd, are associated with adverse respiratory and/or cardiovascular effects (Censi et al., 2011; Wang et al., 2024). However, there is a significant gap in in vitro studies investigating the effects of Dy, Sm, Nd, and Tm as individual components or as a mixture on the human bronchial epithelium.

Barium was the other metal uniquely detected at Shepherd in 2023. Barium is not normally detected at elevated levels following wildfires when comparing smoke days to non-smoke day concentrations, and therefore the prevalence of barium at Shepherd might be unique to the Sound Resource Solutions Plant fire (Baliaka et al., 2025; Li et al., 2023). Barium compounds, such as barite (barium sulfate) and barium hydroxide, are used by industry in drilling muds which keeps the drill bit lubricated during oil and gas drilling operations (Johnson et al., 2017). Sound Resource Solutions blends products for the petroleum industry and transport waste, which might contain these barium compounds. Barium itself is not a free element in nature due to its reactive nature, and therefore its toxicity is greatly dependent on the solubility of the specific barium compound (ATSDR, 2007a). Barium sulfate is relatively insoluble and therefore considered nontoxic because there is no dissolution and biological activity of the Ba²⁺ ion. In contrast, soluble barium compounds like barium hydroxide are considered highly toxic to human and animal life (Gad, 2014). Even so, there remains limited research on the effects of inhaled barium compounds. The majority of studies explore the adverse effects of ingested barium salts which results in serious cardiovascular, gastrointestinal, and musculoskeletal effects (Diengott et al., 1964; Tardiff et al., 1980). Ultimately, the identity of which barium salts the gelatin filters captured remains undetermined.

Finally, the human respiratory system is divided into numerous sections each with multiple cell types with the overall objective to filter debris, prevent inflammation, and facilitate gas-exchange. Using a bronchial epithelial cell line like 16HBEs does not capture every cell-to-cell interaction that occurs in the lung which would require a significant immune component, but the bronchial epithelium is one of the first lines of defense against inhaled irritants such as PM_2.5 and VOCs, which have been shown to reach the middle and lower airways (ATSDR, 2007b; Sturm, 2020). There are also multiple respiratory diseases that primarily affect the bronchial epithelium including asthma, bronchiectasis, COPD, and bronchitis, which highlights that the bronchial epithelium is a viable target of interest for inhalation toxicity studies (Leung et al., 2020). Submerged cell culture exposure systems are a widely used and well-established approach for assessing the biological effects of volatile and semi-volatile organic compounds in vitro (Gałęzowska et al. 2016). As such, the 16HBE cell line possesses the key characteristics needed for respiratory toxicity screening. However, for more comprehensive mechanistic investigations, incorporating a co-culture model with immune cells (e.g., macrophages) or using primary human bronchial epithelial cells would provide a more physiologically relevant system and offer deeper insight into how individuals or populations respond to air pollution exposure.

A key overall limitation of this study was incomplete data coverage across background, ancillary, and processing dimensions during the 2023 sampling period. Specifically, the absence of the Oakhurst background dataset limited full characterization of regional VOC background composition and constrained attribution of compounds identified at Shepherd as definitively fire-related or regionally unique. In addition, humidity, temperature, and GPS data for November 10, 2023 were unavailable due to file corruption, introducing uncertainty in fine-scale plume interpretation, spatial comparability, and route reconstruction. Although NOAA-HYSPLIT back trajectories, nearby TCEQ station data, iPhone location metadata, and field notes were used to reconstruct general conditions, these sources do not resolve micro-scale variability. Finally, the dissolution of gelatin filters in 50 mL of media, while necessary to minimize headspace-related volatilization losses, may have introduced additional dilution of environmentally associated constituents, potentially reducing biological sensitivity. Ongoing studies are evaluating the use of smaller dissolution volumes to improve exposure concentration for cellular assays. Collectively, these limitations underscore the importance of preparedness for disaster-response research.

5. Conclusion

The Sound Resource Solutions plant fire in Shepherd, Texas, illustrates how even a relatively small industrial fire can generate a transient air pollution mixture. Mobile and stationary air monitoring conducted immediately following the fire identified elevated concentrations of VOCs and trace metals at the incident site compared with downwind and background locations, whereas seasonally matched monitoring two years later demonstrated that these differences had resolved, indicating pollutant patterns enriched at Shepherd relative to available comparison sites. The use of a novel biocompatible gelatin filter enabled direct collection of ambient pollutant mixtures for subsequent in vitro toxicity testing without additional extraction procedures. Exposure of human bronchial epithelial cells to filtrates collected near the incident site significantly increased expression of the oxidative stress-responsive gene HMOX1, providing evidence that real-world air pollutant mixtures generated during episodic emission events can elicit biologically relevant cellular responses.

Although the specific constituents responsible for this response remain unclear, the findings suggest that the observed biological effects likely reflect the combined influence of complex pollutant mixtures rather than a single contaminant. Collectively, this study demonstrates the value of integrating rapid-response air monitoring, advanced chemical characterization, and biologically compatible sampling with in vitro respiratory models to characterize both environmental exposures and their potential health impacts following industrial disaster events.

Figure 8.

Relative gene expression of IL-8, TNFα, HMOX-1 and NQO1 in 16HBEs are shown after 24- and 48-hours exposure to condition media of filters collected at sampling sites. * = p < 0.05 and ** = p < 0.01 and *** = p < 0.001 and **** = p < 0.0001. A normal one-way ANOVA was used to compare responses between filter samples and the background site (Oakhurst). DEPM = 100 μg/ml diesel exhaust particulate matter which was used as a positive control.

Environmental Implications:

Environmental incidents such as industrial fires release complex mixtures of hazardous pollutants that are challenging to characterize and assess for health risk. This study links real-world air monitoring with toxicological evaluation by combining VOC and metals measurements, gelatin filter sampling, and exposure of human bronchial epithelial cells. Pollutant mixtures collected near a chemical plant fire induced a transient oxidative stress response, demonstrating a short-term cellular response associated with the emission event. This integrated approach provides a practical framework for rapidly evaluating the health hazards of episodic environmental pollution and informing post-incident environmental monitoring strategies.

Glossary

ATSDR

Agency for Toxic Substances and Disease Registry

16HBEs

human bronchial epithelial cells

BTEX

benzene, toluene, ethylbenzene, xylenes

EPA

environmental protection agency

GC-MS

gas chromatography mass spectroscopy

HF

hydrofluoric acid

HMOX-1

heme oxygenase-1

HNO₃

nitric acid

ICP-MS

inductively coupled plasma mass spectroscopy

IL-8

interleukin-8

LDH

lactate dehydrogenase

MEM

minimum essential medium

mRAPiD

Texas A&M Mobile Responding to Air Pollution in Disasters

MRL

minimal risk level

NOAA

National Oceanic and Atmospheric Administration

NQO1

NAD(P)H quinone dehydrogenase

NTA

nontargeted analysis

PBS

phosphate buffered saline

PTR-ToF-MS

proton transfer reaction time-of-flight mass spectroscopy

REEs

rare earth elements

ROS

reactive oxygen species

RT-qPCR

real time quantitative polymerase chain reaction

TNFα

tumor necrosis factor alpha

VOC

volatile organic compound

Availability

The PTR-ToF-MS data generated and analyzed for this study is provided in the supplemental information. These data can also be accessed in the Texas Data Repository, separated for 2023 (https://doi.org/10.18738/T8/EE2SM5) and 2025 (https://doi.org/10.18738/T8/ZUTESE) datasets. Trace metal analysis data is available upon request.