Assessing the Oxidative Potential of Outdoor PM2.5 in Wintertime Fairbanks, Alaska

The oxidative potential (OP) of outdoor PM2.5 in wintertime Fairbanks, Alaska, is investigated and compared to those in wintertime Atlanta and Los Angeles. Approximately 40 filter samples collected in January–February 2022 at a Fairbanks residential site were analyzed for OP utilizing dithiothreitol-depletion (OPDTT) and hydroxyl-generation (OPOH) assays. The study-average PM2.5 mass concentration was 12.8 μg/m3, with a 1 h average maximum of 89.0 μg/m3. Regression analysis, correlations with source tracers, and contrast between cold and warmer events indicated that OPDTT was mainly sensitive to copper, elemental carbon, and organic aerosol from residential wood burning, and OPOH to iron and organic aerosol from vehicles. Despite low photochemically-driven oxidation rates, the water-soluble fraction of OPDTT was unusually high at 77%, mainly from wood burning emissions. In contrast to other locations, the Fairbanks average PM2.5 mass concentration was higher than Atlanta and Los Angeles, whereas OPDTT in Fairbanks and Atlanta were similar, and Los Angeles had the highest OPDTT and OPOH. Site differences were observed in OP when normalized by both the volume of air sampled and the particle mass concentration, corresponding to exposure and the intrinsic health-related properties of PM2.5, respectively. The sensitivity of OP assays to specific aerosol components and sources can provide insights beyond the PM2.5 mass concentration when assessing air quality.

S.1 Acellular oxidative potential measurement S.1.1 Water-soluble and total DTT measurement The water-soluble DTT assay was conducted using a semi-automated system developed by Fang et al. (2015), ( 1 ) following the protocol described by Cho et al. (2005).( 2) To perform the assay, the filtered PM extract (3.5 mL) was incubated with DTT solution (1 mM; 0.5 mL) and potassium phosphate buffer (1 mL; purified by passing thru Chelex 100 Resin column to remove possible binding polyvalent metal ions before adjusting pH to ~7.4) at 37 °C and shaken at a rotational frequency of 400 rpm using a ThermoMixer (Eppendorf North America, Inc., Hauppauge, NY, USA).At designated times (4,13,23, 31 and 41 min), a small aliquot (100 µL) of the mixture was withdrawn and mixed with trichloroacetic acid (TCA, 1% w/v; 1 mL) to quench the DTT reaction.
The quantification of OP total DTT was carried out using a semi-automated system developed by Gao et al. (2017), ( 3 ) following the same protocol as the OP WS DTT measurement with the exception that the insoluble PM components suspended in the extracts or attached to the filter surface could also participate in DTT oxidation by performing the assay with the filter in the reagents.
The DTT consumption rate was determined by the slope of the linear regression of DTT residual versus reaction time.A field blank and a positive control (9,10-phenanthraquinone) were analyzed with all sample batches.The PM OP measured by this DTT assay was blank-corrected and normalized by the air volume that passed through the extracted filter (volume-normalized, OPv DTT )    or the PM mass loading on the extracted filter (mass-normalized, OPm DTT ).The assay was performed at an approximate particle concentration of 10 µg/mL for the analysis of Fairbanks and LA samples, but not for Atlanta samples.

S.1.2 Hydroxyl radical generation in surrogate lung fluid
The OH assay was performed following the protocol described by Yu et al. (2021).( 4) The PM extract was mixed with surrogate lung fluid (SLF) consisting of a mixture of 2.5 mM ascorbic acid (AA), 1.25 mM reduced glutathione (GSH), and uric acid (UA), which was dissolved in phosphatebuffered saline (PBS, PH ~7.4) along with potassium phosphate-buffered disodium terephthalate (TPT) (50 mM; pH ~7.4).TPT captures OH radicals generated in the reaction and forms 2hydroxyterephthalic acid (2-OHTA), a fluorescent product that can be measured at an excitation/emission wavelength of 310/427 nm.The mixing solution was incubated at 37 °C and shaken at a rotational frequency of 400 rpm.At selected time intervals (30, 60, 90 and 120 min), aliquots of this reaction mixture (2 mL) were withdrawn and mixed with dimethyl sulfoxide (DMSO, 100 mM; 1 mL) to quench the reaction between •OH and TPT.The final mixture was analyzed by a Shimadzu Spectro fluorophotometer (RF-5301PC).The excitation and emission slit widths were set at 5 nm and 10 nm, respectively.The concentration of 2-OHTA was determined by calibrating with 10 different concentrations (0-500 nM) of 2-OHTA standards, and the generation rate of •OH was determined as the formation rate of 2-OHTA divided by a yield factor of 0.35.( 5) The PM OP measured by this OH assay was blank-corrected and normalized by the air volume that passed through the extracted filter (volume-normalized, OPv OH ) or the PM mass loading on the extracted filter (mass-normalized, OPm OH ).The assay was performed at an approximate particle concentration of 25 µg/mL for the analysis of the samples from all three cities.

S.2 Aerosol mass spectrometry measurements
A High-Resolution Time-of-Flight Aerosol Mass Spectrometer (HR-ToF-AMS, Aerodyne Research, Inc., USA) was used for online measurement of non-refractory sub-micron PM (NR-PM1) composition and mass concentrations.Positive matrix factorization (PMF) was applied to AMS organic aerosol mass spectra and a set of factors corresponding to specific sources or types of compounds were determined.The overall data set included mass concentrations of NH4 + , NO3 − , SO4 2− , Cl − , and OA; OA components from PMF; polycyclic aromatic hydrocarbon (PAH) concentrations; and oxygen-to-carbon (O:C) ratio for the OA.The OA components include factors from a three-factor PMF solution (described in more detail below): hydrocarbon-like OA (HOA), biomass-burning OA (BBOA), and another primary OA (POA2).The instrument was connected to a sampling line (built with 3/8 inch-diameter copper tubing) fitted with a PM1 inlet positioned outside the house at 1.5 m above ground.The total sampling flow was 5.5 L/min.Prior to the AMS measurements, the aerosol was dried with a Nafion dryer (MD-070, Perma Pure LLC., USA) to reduce the relative humidity (RH) below 30%.
The three-factor PMF solution we present here was run on the entire AMS dataset, which included alternating outdoor and indoor sampling at equal time intervals.There was a series of indoor experiments conducted in the house to explore interactions between outdoor-infiltrating and indoor-generated aerosols, but these periods were excluded from this PMF analysis.The HOA and BBOA factors identified in this PMF solution have mass spectra very similar to canonical examples from the literature.We performed a correlation analysis and found our HOA to have an average Pearson correlation coefficient of 0.97 with spectra from Elser et al ( 6) and Mohr et al ( 7 ).
Similarly, our BBOA factor has an average Pearson correlation coefficient of 0.75 with Elser et al ( 6).This correlation coefficient is lower than the respective value for HOA compared to literature, but BBOA mass spectra are typically highly variable, and so in that context, the correlation coefficient is still high.We also see characteristic mass spectral fragments in our BBOA factor (e.g., C2H4O2 + , a tracer fragment for levoglucosan) and expected diurnal pattern.The remaining mass was apportioned to a factor that we are calling "POA2", though we are not sure exactly what source(s) may contribute to it.We name it "POA2" because it does not resemble the mass spectrum of any canonical secondary OA or oxidized OA.Rather, it has strong contributions from CxHy fragments and high signal at m/z 55, both of which are characteristic of a variety of primary OA spectra (e.g., cooking, vehicles, and others).
We quantified total PAHs, per the method introduced by Dzepina et al ( 8).This approach uses unit-mass resolution fragmentation patterns established by Dzepina et al., which are built into the SQUIRREL (v.1.65C)AMS analysis software.We were unable to resolve our PAH signal into sub-classes per the methods of Herring et al. ( 9), as our mass spectra lacked the proper resolution at high (>200) m/z to differentiate between possible candidate ion fragments.High m/z is where most PAH fragments are located due to their resistance to fragmentation.We operated the instrument in V-mode using a mass range up to m/z 405.
O:C was calculated using the PIKA (v.1.25C)analysis software, using the "improved ambient" method introduced by Canagaratna et al ( 10 ).

S.3 Elemental Analysis
To measure the water-soluble elements, one (1-inch diameter, 5.06 cm 2 ) punch was extracted in 5 mL of DI water in a sterile polypropylene centrifuge tube via a 60-min ultrasonic treatment.The extracts were then filtered using a 0.45 µm PTFE syringe filter and the filtered extract was acidpreserved with concentrated nitric acid (70%) to a final concentration of 2% (v/v) to maintain the metals in solution.Metals in the filtrate are defined as water-soluble metals, which encompass all dissolved forms as well as any colloidal particles with a diameter of less than 0.45 µm.( 12) One 1.5 cm 2 filter punch from the Hi-Vol quartz filter was acid-digested using aqua regia (HNO3+3HCl) to quantify the total metal concentration.The filter was incubated in the acid at 99 ℃ while shaken at a rotational frequency of 400 rpm using a ThermoMixer for 24 h.The aciddigested sample was then diluted in DI water and filtered through a 0.45 µm PTFE syringe filter.( 12) Both total and water-soluble elements were measured by inductively coupled plasma mass spectrometry (ICP-MS, Agilent 7500a series, Agilent Technologies, Inc., CA, USA) using EPA method 6020.( 13) Calibrations were performed at the beginning of every measurement day using a dilution of commercially available standard stock solutions (1000 ppm in 2% HCl), and the data were blank-corrected.The concentrations of various elements including magnesium, aluminum, potassium, manganese, iron, copper, zinc and lead are included in the analysis.Other elements were measured but deemed less useful or were mostly below detection limits.

S.4 Organic carbon and elemental carbon and brown carbon analysis
To determine the organic carbon (OC) and elemental carbon (EC) content of fine particulate matter, a 1.5 cm 2 filter punch was analyzed with a benchtop Sunset OC/EC analyzer (Sunset Laboratory Inc. OR, USA).The measurement was carried out following the National Institute for Occupational Safety and Health (NIOSH) 5040 analysis protocol.( 14) For the determination of brown carbon (BrC), a 1.5 cm 2 filter punch from the Hi-Vol filter was extracted in 15 mL DI or methanol by 60 min sonication.The light absorption of soluble BrC species was measured with a liquid-based spectrophotometric method in each of these solvent extracts that was filtered (0.22 µm pore size syringe filter) and then injected into a 2.5 m path LWCC, which was coupled with a broadband UV-Vis-NIR light source and a spectrometer.As a measure of BrC levels (i.e., chromophores), the light absorbance at 365 nm (average of 360-370 nm) wavelength was determined for each solvent (  365 ) and the absorption coefficient was calculated by ( 15): •  ln (10) where  700 is the average light absorbance at 695-705 nm,   the volume of water or methanol the filter was extracted into, which is 15 mL,   the volume of sample air that passed through the filter punch (based on overall flow rate and the ratio of punch area to overall filter collection area), and  the length of the waveguide, which is 2.5 m.

S.5 Description of stepwise regression
Stepwise regression involves a systematic process of adding and removing variables from the model, based on their statistical significance, until a final model with the best set of predictors is identified.Before regression analysis, extreme outliers (data points beyond an outer fence, which are lower than the lower quartile (Q1; 25th percentile) -3*interquartile range (IQ; Q3-Q1) or higher than the upper quartile (Q3; 75th percentile) + 3*interquartile range; 2 data points in total) were removed.The inclusion or exclusion of variables is based on the Akaike information criterion (AIC) and Bayesian information criterion (BIC), which are widely accepted criteria for selecting the optimal model.There is no difference between the models selected by AIC and those selected by BIC.
The standardized MLR method involved rescaling the variables using a linear transformation such that all variables had a mean of 0 and variance of 1 (e.g., normalize each variable  by ( − ̅ )/), allowing for a more effective comparison of their relative importance, where ̅ is the mean and  the standard deviation of that variable.The standardized coefficients were used to interpret the change in OP that would occur for a one-unit change in the predictor variable.The absolute values of these standardized regression coefficients represented the relative importance of the variables.

S.6 Multiple Linear Regression results with the selection of different groups of species
The unstandardized and standardized regression results for the selection of different groups of species selected from those that were correlated are shown below:

Unstandardized Regression with different independent variables that were correlated:
The first equation in each group in italics, is the regression included in the main text of the paper.The selection of different groups of species resulted in a consistent pattern of regression results, exhibiting similar coefficients, intercept, and overall model performance, as evidenced by comparable values of r 2 and MSE.This consistency implies that the choice of species grouping did not significantly impact the models' predictive ability and explanatory power in most cases.
An exception was observed in the regression analysis of OPv total DTT , where an alternative combination of species, namely EC and BBOA or BrC, proved to be as effective as including PAHs.It is important to note that this observation is in line with the known characteristics of PAHs as hydrophobic organic compounds predominantly emitted from combustion processes such as biomass burning and fossil fuel combustion.Therefore, OPv total DTT could be considered as a measure reflecting the combined impact of residential heating and vehicular emissions in Fairbanks.

S.7 Additional notes about multivariate linear regression
To explore the impact of synergistic and antagonistic interactions among various PM species on oxidative potential, interaction terms were included for fitting the MLR models.In the MLR model for OPv WS DTT , no interaction terms were found to be significant based on the stepwise regression analysis.Whereas, for OPv total DTT and OPv OH , significant interaction term(s) were identified.

Unstandardized Regression with interaction terms:
OPv Considering the intercept, r 2 and MSE of the regression models, the inclusion of interaction terms had limited improvement in model fitting performance, except for OPv OH , which may indicate a synergistic effect of HOA and Fe on hydroxyl radical production rate.Therefore, the inclusion of interaction terms was not pursued.Additionally, due to the limited number of data points (~40), introducing higher-order terms or employing splines for nonlinear regression may lead to overfitting issues.Thus, MLR models were constructed without including any higher-order or nonlinear terms, assuming the effects of each predictor were additive.However, it is worth noting that this additive assumption may not always hold true in all cases.

Figure S2 .
Figure S2.Comparison between total and water-soluble (a) volume-normalized and (b) mass-

Table S2 .
Correlation between various PM2.5 components measured at the House site in Fairbanks.

Table S3 .
Statistical summary of different PM2.5 chemical components measured in Fairbanks (N

Table S6 .
Mean and standard deviation of volume-normalized OP and various components of ambient PM2.5 in Fairbanks (Jan-Feb), Atlanta and Los Angeles in winter.Unless stated separately, ambient PM2.5 data in Atlanta is obtained from Gao et al. (2020).( 16 ) b. Unless stated separately, ambient PM2.5 data in Los Angeles is obtained from Shen et al. (2022).

Table S7 .
Mean and standard deviation of mass-normalized OP and various components of ambient PM2.5 in Fairbanks (Jan-Feb), Atlanta and Los Angeles in winter.Unless stated separately, ambient PM2.5 data in Atlanta is obtained from Gao et al.