Air Pollutant Patterns and Human Health Risk following the East Palestine, Ohio, Train Derailment

On February 3, 2023, a train carrying numerous hazardous chemicals derailed in East Palestine, OH, spurring temporary evacuation of residents and a controlled burn of some of the hazardous cargo. Residents reported health symptoms, including headaches and respiratory, skin, and eye irritation. Initial data from U.S. Environmental Protection Agency (EPA) stationary air monitors indicated levels of potential concern for air toxics based on hazard quotient calculations. To provide complementary data, we conducted mobile air quality sampling on February 20 and 21 using proton transfer reaction-mass spectrometry. Measurements were taken at 1 s intervals along routes designed to sample both close to and farther from the derailment. Mobile air monitoring indicated that average concentrations of benzene, toluene, xylenes, and vinyl chloride were below minimal risk levels for intermediate and chronic exposures, similar to EPA stationary monitoring data. Levels of acrolein were high relative to those of other volatile organic compounds, with spatial analyses showing levels in East Palestine up to 6 times higher than the local rural background. Nontargeted analyses identified levels of additional unique compounds above background levels, some displaying spatiotemporal patterns similar to that of acrolein and others exhibiting distinct hot spots. These initial findings warrant follow-up mobile air quality monitoring to characterize longitudinal exposure and risk levels.


Proton transfer reaction time-of-flight reaction (PTR-ToF) mass spectrometer operation
The PTR-ToF detects volatile organic compounds (VOCs) through "soft" ionization. Thus, most analytes are detected as the parent ion with minimal fragmentation. The Ionicon, Inc. 4000 time-of-flight mass spectrometer has a mass resolution of 10,000. This means that the PTR-ToF-MS can separate ions that have the same nominal mass but different amounts of C, H, O, and heteroatoms.
The PTR-ToF was operated in two modes. In the traditional hydronium mode, ionization is achieved via proton transfer (hence the instrument name): Here, X is the analyte of interest. This mode is sensitive to species that have a higher proton affinity than water, which includes a wide array of atmospheric VOCs but crucially does not include small aliphatic molecules like methane and ethane that would otherwise dominate the signal.
The PTR-ToF can also be operated in oxygen mode. Here the charge carrier is O 2 + , and ionization is via direct charge transfer.
Oxygen mode is significantly more sensitive to chlorinated organic compounds that have low proton affinities than hydronium mode.
At the start and end of each day of sampling, we sampled zero air with the instrument. This was used to determine the instrument baseline, which was subsequently subtracted during analysis. Additionally, a multi-point calibration was conducted each night with a 16-component calibration mixture ( Table 1). The calibration standard was dynamically diluted with zero air during each nightly calibration.

Proton transfer reaction time-of-flight reaction (PTR-ToF) mass spectrometer data analysis
Ion abundance was determined via peak integration in the PTR-MS Viewer 3.22 software ( Figure 1). Target analysis uses a predefined peak table to select ions for integration. In cases of isobaric ions, a multi-peak fit was applied to separate the (clearly distinct) peaks. Ion abundance was also adjusted for ion transmission efficiency in the PTR-ToF. The transmission efficiency curve was generated daily using our calibration mixture. Figure S1. Peak integration of raw data. There are two distinct peaks with a nominal mass of m/z 57: C 3 H 4 OH + and C 4 H 9 + . The oxygenated peak has a slight negative mass defect (actual mass < nominal mass) because of the presence of oxygen atoms, whereas the unoxygenated ion has a positive mass defect. The cyan and green traces show integration of these separate ions.
Two methods were used to convert ion abundances to species concentration. The first directly considered the kinetics of the proton transfer reaction: For a fixed reaction time (which exists in the PTR-ToF), the above reduces to an algebraic expression. If k is known, then the concentration of analyte [X] can be computed. We used this approach for compounds measured in hydronium mode. Many measured k values are published in the literature and therefore readily available. 2 The reaction rate k can also be calculated from molecular properties following the method of Sekimoto et al. 3 When k is not published, a default value of 2x10 -9 cm -3 molecule s -1 is commonly used.
The other method applied was to compute [X] using a calibration factor. These calibration factors are expressed in units of ppb per normalized ion counts (ppb/ncps). 4 In this approach, the raw ion signal for each ion is normalized by the primary ion signal (either H 3 O + or O 2 + depending on the instrument mode). We use this approach to determine concentrations of chlorinated species, such as vinyl chloride, measured in oxygen mode. Table 2 shows species identified during target analysis. For each target species, we defined the minimum detection limit (MDL) as three times the standard deviation of the signal measured while sampling zero air.

Quantification of acrolein
Acrolein was identified as a priority component because of high concentrations in the canister samples collected by the EPA. Several previous studies have used PTR-ToF to quantify acrolein, primarily in emissions from biomass burning. There is wide variation in the published calibration factors for acrolein.
Brilli et al. 5 used the traditional default k of 2x10 -9 cm -3 molecule -1 s -1 . They did not directly calibrate for acrolein. Sekimoto et al. 3 presented a method for calculating k. They also performed experimental measurements to evaluate their predictions. The calculated k for acrolein to be 3.1x10 -9 cm -3 molecule -1 s -1 . However, they also observed that the sensitivity for acrolein and other small, oxygenated molecules was only ~40% of the expected sensitivity based on k alone. They attributed this to delocalization of electrons in the conjugated doubled bonds in acrolein. Schieweck et al. 6 measured acrolein in an indoor environment and determined a k of 3.55x10 -9 cm -3 molecule -1 s -1 .
Two other papers reported acrolein sensitivity instead of k. Lastly, Koss et al. 7 reported a sensitivity of 38.73 ncps/ppb for acrolein. However, those experiments used a slightly different version of the PTR-ToF. Stockwell et al. 8 reported a sensitivity nearly an order of magnitude lower.
We did not have a calibration standard for acrolein. Given the wide range in published acrolein calibrations, we do not report acrolein concentration. Instead, we report in Table 2 the minimum detection limit in terms of counts per second (cps). Analysis of acrolein data compares signals (e.g., between Pittsburgh and East Palestine) but does not report concentrations directly.

Calibration checks
We evaluated instrument performance each day by comparison to a calibration mixture. We evaluated the slope of the linear correlation between the supplied and measured concentrations, as well as the linearity of the calibration. Our calibrations demonstrate excellent linearity (R 2 > 0.99).  Table 2, underestimates the true concentration by about 18% (slope of 0.822). Thus, benzene data for this day are adjusted by the best-fit slope. This process was repeated daily for each species present in our calibration standard. Figure S2. (left) Calibration of vinyl chloride. The resulting sensitivity (3.04 ncps/ppb) was used to convert raw signals to concentration. (right) Calibration check for benzene. There was good linearity, but the reported concentration was biased low by 18%. The resulting data were therefore adjusted based on this bias.

Acrolein
Because acrolein was not quantified on an absolute scale, we used relative comparison with sampling that was performed in Pittsburgh on 2/16/2023. Specifically, sampling in hydronium mode in "downtown" Pittsburgh occurred from 10:54 am until 11:07 am, and a separate "urban" area of Pittsburgh from 1:22 pm until 1:33 pm. The average value for "downtown" was used as a reference point data from East Palestine data. For mapping, acrolein data (ratio to Pittsburgh mean) within 3 miles of East Palestine (defined as latitude=40.833951 and longitude= -80.540347) were included. Data were averaged over a grid with grid sizes of 150 ft x 150 ft, and log-transformed for color scaling.

Benzene, Toluene, Vinyl Chloride, and Xylenes
Data for benzene, toluene, vinyl chloride, and xylenes were visualized using violin plots, and compared to available reference levels (

Contextual Information
For context, the train manifest is shown in Figure S3, and the meteorological data on the day of sampling is summarized in Figure S4.  Figure S3. Train manifest, as provided by Norfolk Southern, with chemicals of concern highlighted. Figure S4. Summary of temperature, precipitation, wind speed, and wind direction at nearest weather station to East Palestine (Pittsburgh International Airport Station) on the day of sampling (February 20, 2023). Source: Weather Underground.