Atmospheric conditions and composition that influence PM2.5 oxidative potential in Beijing, China

Epidemiological studies have consistently linked exposure to PM2.5 with adverse health effects. The oxidative
potential (OP) of aerosol particles has been widely suggested as a measure of their potential toxicity. Several acellular chemical
assays are now readily employed to measure OP, however, uncertainty remains regarding the atmospheric conditions and
specific chemical components of PM2.5 that drive OP. A limited number of studies have simultaneously utilised multiple OP
assays with a wide range of concurrent measurements and investigated the seasonality of PM2.5 OP. In this work, filter samples
were collected in winter 2016 and summer 2017 during the atmospheric pollution and human health in a Chinese megacity
(APHH-Beijing) campaign, and PM2.5 OP was analysed using four acellular methods; ascorbic acid (AA), dithiothreitol (DTT),
2-7-dichlorofluoroscin/hydrogen peroxidase (DCFH) and electron paramagnetic resonance spectroscopy (EPR). Positive
correlations of OP normalised per volume of air of all four assays with overall PM2.5 mass was observed, with stronger
correlations in the winter compared to the summer. In contrast, when OP assay values were normalised for particle mass, days
with higher PM2.5 mass concentrations (μg m-3) were found to have lower intrinsic mass-normalised OP values as measured
by AA and DTT. This indicates that total PM2.5 mass concentrations alone might not always be the best indicator for particle
toxicity. Univariate analysis of OP values and an extensive range of additional measurements, 107 in total, including PM2.5
composition, gas phase composition and meteorological data, provides detailed insight into chemical components or
atmospheric processes that determine PM2.5 OP variability. Multivariate statistical analyses highlighted associations of OP
assay responses with varying chemical components in PM2.5 for both mass- and volume-normalised data. Variable selection
was used to produce subsets of measurements indicative of PM2.5 sources, and used to model OP response; AA and DTT assays
were well predicted by small panels of measurements, and indicated fossil fuel combustion processes, vehicle emissions and
biogenic SOA as most influential in the assay response. Through comparative analysis of both mass- and volume-normalised
data we also demonstrate the importance of also considering mass-normalised OP when correlating with particle composition
measurements, which provides a more nuanced picture of compositional drivers and sources of OP compared to volumenormalised analysis, and which may be more useful in temporal and site comparative contexts