Oxidative potential of atmospheric particles in Europe and exposure scenarios

Sites and data description

In this work, the data come from a large number of European sites: permanent monitoring station or punctual site of campaign. The main value of these data lies in the notable diversity of site types (five different identified types) and the long periods of sampling (2011–2024). The database contains PM mass concentration and OP measurements (AA and DTT assays) from 43 different locations. Most sites are located in France, but thanks to collaborations with several European research infrastructures, data from nine other countries are also available. The database includes sampling periods ranging from 2 months to 9.5 years for various PM size fractions (PM₁, PM_2.5, PM₁₀). The location and description of the sites are shown in Extended Data Fig. 1 and Extended Data Table 1.

Part of this data is the result of collaborations between the Institut des Géosciences de l’Environnement (IGE) in Grenoble and various European research institutions. These collaborations involved sampling campaigns as part of national and European research programmes aimed at evaluating air pollution. Some data considered in this work have already been described and interpreted in the literature. At the European level, certain stations are part of the European Monitoring and Evaluation Programme (EMEP), ACTRIS, a pan-European research infrastructure on atmosphere, or the European air quality monitoring network (EUROAIRNET)^49,50. At the national level, the five stations in Switzerland are managed by the Swiss Federal Office for the Environment and EMPA (Swiss Federal Laboratories for Materials Science and Technology) within the framework of the National Air Pollution Monitoring Network (NABEL) project³⁰. Research into PM sources and OP in samples from locations such as Kraków (Poland) was conducted under projects led by the Paul Scherrer Institute (PSI) in Switzerland^40,42. French stations are administrated by the Associations Agréées Surveillance Qualité de l’Air (AASQA), regional monitoring networks, in the framework of various research programmes^{29,37,44,51,52}. SArajevo AEROsol Experiment (SAAERO) at the headquarters of the Federal Hydrometeorological Institute of Bosnia and Herzegovina provided Sarajevo samples⁵³, whereas a municipality-funded project, led by University of Nova Gorica, provided Kanal samples⁵⁴. Stations in Barcelona and Montseny are managed by CSIC-IDEA as described in ref. ³¹. The three sampling sites in Hungary, described in ref. ⁵⁵, had similar OP and PM mass results and were therefore combined into one dataset.

Following the criteria of classifications of the European Environment Agency (EEA), all stations have been assigned a category based on their urbanization levels in the area of the site (urban, rural or suburban areas) and nearby types of pollution source (such as industries or main roads). In this study, we referred to both ‘air quality station area’ and ‘air quality station type’ as ‘site type’. The stations were categorized into five site types: traffic (T), urban (U), industrial (I), suburban (SU) and rural (R). Also, we identified some sites with special topographies (valleys (V) and coastal sites), which could influence the pollution dispersion. For instance, the Kanal and Passy stations, located in the Soča (southwestern Slovenia) and Arve (eastern France) valleys, respectively, share topographic characteristics typical of valleys, such as temperature inversion events during winter. It is worth noting that all industrial sites of the dataset, except for Kanal, are also coastal areas that are characterized by high PM mass but with low reactivity particles such as sea salts. The aim was to cluster the stations with similar air pollution characteristics; however, some sites are influenced by different specific sources and are difficult to classify among these categories. Passy is a suburban site but also close to high-emission industrial sources, resulting in higher PM levels throughout the year compared with other suburban sites and valleys. Likewise, the Bern station is close to main highways and a busy railway station. Like the other traffic stations, this site is influenced by several PM sources including typical urban sources (biomass burning, exhaust and non-exhaust vehicles, mineral dust and so on) but it is also affected by road and rail traffic. Although Bern is included in the traffic sites, its PM and OP levels are higher than other traffic sites probably because of the street canyon micrometeorology but are also affected by the influence of rail traffic. The Courmayeur station is located alongside a main international traffic road between Italy and France, within a predominantly rural setting. However, a large portion of the samples was collected during the recent closure of the Mont Blanc tunnel (year 2023), when traffic-related emissions were minimal, further justifying the classification of this site as rural.

Measurements at most stations were performed on daily filters (24-h sampling) but the dates and durations of the campaigns vary across the dataset (start and end date). All sites are characterized by at least 2 months of measurements and a minimum of 45 samples except for FSM (Fos-sur-Mer, France), with 28 samples covering an 8-month period.

Sampling and analyses

The samples were analysed in the IGE lab using the same protocol, which allows to compare the data of all of the stations.

OP analysis

OP was analysed offline on PM filters (PM₁, PM_2.5 or PM₁₀ depending on the campaign and the location) according to the protocol established in ref. ⁹ and refined later^37,39. After sampling, the filters were stored at −20 °C in the IGE lab until OP analysis. Then, punches of PM filters were extracted in a simulated lung fluid to approach the physiological conditions. This fluid is a mixture of Gamble and 1,2-dipalmitoylphosphatidylcholine (DPPC), as described in ref. ⁵⁶. Samples were extracted at a concentration of 10 µg ml⁻¹ or 25 µg ml⁻¹, depending on the loading of the filters or their period of analysis (<2015 or after; see Supplementary Figs. 18 and 19). After extraction, the samples were vortexed at 37 °C for 1 h 15 min and the OP was measured with the AA and DTT assays. For both assays, a solution of reactants (AA or DTT) was added to the extracts in a 96-well plate (CELLSTAR, Greiner-Bio). For the DTT assay, the DTT was titrated by dithionitrobenzoic acid (DTNB) every 10 min for 30 min. During every titration, the absorbance of the solution was measured at 412 nm (TECAN spectrophotometer Infinite M200 Pro) owing to the formation of coloured 2-nitro-5-thiobenzoic (TNB). AA consumption was determined by absorbance measurements at 265 nm every 4 min for 30 min. These protocols allowed to obtain OP measurements as consumption rate in nanomoles per min. These results were either divided by the total mass of PM on the filter punches to obtain the intrinsic OP (OP_m in units of nmol min⁻¹ µg⁻¹), that is, the toxicity of 1 µg of PM, or normalized by the air volume (by multiplying OP_m by PM mass concentration in the air) to obtain the volumetric OP (OP_v in units of nmol min⁻¹ m⁻³). OP_v better reflects OP exposure and integrates both OP_m and PM mass. Final results correspond to the mean values of two to three repeated measurements with a coefficient of variation usually less than 10% for each sample. Positive and negative control tests were performed each time respectively with a solution of 1,4-naphthoquinone (1,4-NQ) and an extract of blank filter in Gamble solution.

It is important to point out that artefacts can occur after filters have been cold-stored for several months. Offline methods probably result in an underestimation of the PM OP owing to the volatilization or degradation of certain short-lived chemical species. To address this limitation, online OP prototypes are being developed but have not yet surpassed offline measurement methods in terms of detection limits, measurement time step, operational and technical constraints or reliability. All the more, no commercial devices exist.

OP assays sensitivity

OP acellular assays use probes (AA and DTT here) to simulate the consumption of antioxidants by ROS in a simulated lung fluid. Depending on the probe, OP tests are specific to some ROS induced by particular redox-active PM chemical compounds. Thus, each test individually cannot estimate the overall potential of PM to trigger oxidative stress. For this reason, all samples were analysed simultaneously with two different OP tests as recommended in ref. ¹⁰. DTT and AA OP assays are part of the most commonly applied acellular tests and can be carried out together to give a better assessment of the capacity of PM to generate ROS^9,10. DTT is an organosulfur molecule with thiol groups, similar to glutathione (GSH), responsible for superoxide anion formation in the presence of metals and quinones⁵⁷. This probe is sensitive to a wide and balanced range of chemical compounds, such as organic species (OC, PAHs, quinones, HULIS) and many transition metals^8,9. On the other hand, AA is a natural redox-active antioxidant sensitive to specific species, notably OC and some metals such as Cu and Pb (refs. ^8,9). In the presence of transition-metal ions, AA is capable of transferring electrons to oxygen, generating a superoxide anion and leading to the formation of other ROS, such as H₂O₂ (ref. ⁵⁸). It is worth noting that the intrinsic OP measured with the two assays are not highly correlated in our data (Pearson’s correlation of 0.45 between \({{\rm{OP}}}_{{\rm{m}}}^{{\rm{AA}}}\) and \({{\rm{OP}}}_{{\rm{m}}}^{{\rm{DTT}}}\) for PM₁₀). The characteristics of the probes also give them specificity to certain PM sources. Some studies compared different OP assays (DTT, AA, GSH, electron spin resonance (ESR), dichlorofluorescein (DCFH), hydroxyl radicals ·OH, ferrous-orange xylenol (FOX), respiratory tract lining fluid (RTLF)). Overall, for all of these tests, the main sources associated with ROS formation in winter are biomass burning and traffic emissions and in summer they are traffic emissions, primary and secondary biogenic aerosols, mineral dusts and industrial^8,9,39. Furthermore, the winter time seems to be the most favourable for the generation of these ROS for all assays. Industrial emissions also play an important role in winter according to some assays (DCFH and ·OH)³⁹.

Analysis of chemical compounds

For many sites, filters were also analysed for their chemical speciation. This includes major ions quantified by ionic chromatography⁴¹. Levoglucosan was analysed with high-performance liquid chromatography with pulsed amperometric detection (HPLC-PAD)⁵⁹, metals were characterized by inductively coupled plasma mass spectrometry (ICP-MS) analysis following a protocol close to that in ref. ³⁹ and OC/EC thermo-optical analyses were conducted with a Sunset Lab. Instrument⁶⁰ using the EUSAAR2 protocol. All of these measurements have been achieved in accordance with the present and corresponding European standards. These data allowed to perform positive matrix factorization (PMF) and multiple linear regression (MLR) processing to identify the main sources of PM₁₀ and OP_m and to calculate the contribution of these sources to the PM mass concentration of each observation and to the OP_m (refs. ^43,51,54). These results were used in the emission reduction matrix section. Moreover, we chose to study certain compounds for the correlation matrix in Extended Data Fig. 4: elemental carbon (EC), organic carbon (OC), transition metals (Fe, Cu, Ni, Cr, Pb, Mn, Zn, V, Al, Sb, Sn), water-soluble anions and cations (\({{\rm{SO}}}_{4}^{2-}\), \({{\rm{NH}}}_{4}^{+}\), \({{\rm{NO}}}_{3}^{-}\)), benzo(a)pyrene (polycyclic aromatic hydrocarbons (PAHs)) and levoglucosan (an anhydrous monosaccharide). It may also be of interest in future studies to investigate associations between PM OP and further air pollutants, including NO_x (NO, NO₂) and ozone.

Average OP and PM mass computation

As specified previously, the dates and durations of the campaigns vary across the dataset depending on the site (start and end dates). These between-site temporal variations may introduce bias when it comes to comparing OP and PM average levels between sites. To account for seasonal variability, that is, the fact that the proportion of samples collected in cold (Nov-Dec-Jan-Feb-Mar), warm (May-Jun-Jul-Aug-Sep) and intermediate (Apr and Oct) seasons is not homogeneously distributed across sites, we calculated seasonality-weighted OP and PM mass concentration means (shown in Figs. 1 and 2, Extended Data Figs. 3 and 5, Supplementary Figs. 1 and 3 and Supplementary Table 2) using relative weights calculated as follows (using DescrStatsW() function of the statsmodels package in Python):

in which p_cold, p_warm and p_interm are the relative weights attributed to the three seasons (sum of p_i = 1) and N_cold, N_warm and N_interm correspond to the numbers of observations for each season.

The DescrStatsW() function was also used to obtain the weighted standard deviations.

For Fig. 2a,b, Supplementary Fig. 1 and Supplementary Table 2 (total data), the season-weighted means of all sites within each type of site were then averaged so that each site was given an equivalent weighting. Thus, the error bars on these plots are the standard errors of the means of groups by site type.

Owing to the nature of the data, sampling years vary from site to site. Consequently, annual changes in OP and PM mass levels may affect comparisons between sites. The absence of long-term time series over several sites, covering the geographical variability and types of site presented in the data, prevents statistical correction of this potential bias. This possible bias should be considered when reading and interpreting the results.

Linear regression models

Robust linear regression models were performed on daily observations in Python using the Huber’s T function for M-estimation with the statsmodels package (Fig. 2). We focused on the two PM size fractions considered in the European regulation (PM₁₀ and PM_2.5) within the site types in which people are more likely to live or spend a lot of time (traffic, urban, suburban and rural areas). We have also excluded some valleys that are influenced by specific meteorological phenomena linked to their topography and which are not representative of the site type in which they are classified, particularly for suburban and rural sites. The excluded sites were the valley sites significantly different (t-tests, P values < 0.01) from other sites of the same type in terms of mean PM mass, mean \({{\rm{OP}}}_{{\rm{v}}}^{{\rm{DTT}}}\) and mean \({{\rm{OP}}}_{{\rm{v}}}^{{\rm{AA}}}\) (Passy, Magadino, Marnaz). It worth noting that around 45% of the daily samples used in Fig. 2c,d were collected during the cold season, whereas almost 40% were collected during the warm season. Tests were conducted to check the normality of distributions and assess the influence of the heteroscedasticity (Supplementary Figs. 14–17). On the basis of these tests, robust linear regressions were chosen to reduce the influence of outliers and partially correct for heteroscedasticity.

OP_v exposure scenarios and HIA methodology

The reflexion on OP exposure scenarios is based on the HIA methodology (in French: EQIS – Evaluation Quantitative des Impacts Sanitaires), as described in ref. ⁶¹. To assess the health benefits associated with reducing exposure to a pollutant, these studies explore different scenarios of pollution reduction depending on the local context. These hypotheses are often based on the reduction of a given concentration of pollutant up to a reference threshold (such as WHO guidelines) or the impact reduction on a specific health outcome (such as mortality). When there is no reference threshold, it is usual to use standard levels in areas with little or no anthropogenic activities (such as rural stations) to estimate the total burden of air pollution associated with human activities. Because no threshold exists for \({{\rm{OP}}}_{{\rm{v}}}^{{\rm{DTT}}}\), we chose to set OP exposure scenarios based on levels in rural areas or mildly polluted urban areas.

This methodology is also similar to the rationale adopted by the WHO, owing to the linear dose–response relationship between PM mass and mortality, with no lower threshold below which no effect is observed. Instead of aiming for a zero target, the WHO’s recommended values are based on PM mass concentrations observed in regions with the lowest recorded exposures, such as low-density areas in Manitoba, Canada, or parts of Northern Europe⁴⁸.

A comparison with reference values obtained using \({{\rm{OP}}}_{{\rm{v}}}^{{\rm{AA}}}\) or PM mass values instead of \({{\rm{OP}}}_{{\rm{v}}}^{{\rm{DTT}}}\) can be found in Supplementary Information Section 6.

PM reduction matrix

Concept and data

PM reduction matrices (Fig. 3 and Supplementary Figs. 4–13) have been designed on the basis of the approach in ref. ⁶² with PM_2.5 exposure in Grenoble. For Fig. 3 and Supplementary Fig. 5, the arrays contain OP_v values depending on the percentage of emissions reduction from two PM₁₀ sources (traffic and biomass burning) in four urban sites (Grenoble, Talence, Nice and Barcelona). These reductions refer to a multisite, multiyear reference level based on average source contributions across the four studied sites (2012–2023). Data gather 551 samples collected between March 2012 and April 2013 for Talence, between February 2017 and March 2018 for Grenoble, between July 2014 and May 2015 for Nice and between March 2022 and March 2023 for Barcelona. Other matrices were produced for the French traffic site of Roubaix (156 samples), the French valley of Chamonix (115 samples), the French rural site of the Observatoire Pérenne de l’Environnement (OPE; 438 samples) and the two industrial sites of Kanal and Port-de-Bouc (300 samples). To calculate all matrices, we used the results in ref. ⁵¹ for the OPE rural site, ref. ⁵⁴ for Kanal, RI-URBANS project for Barcelona and refs. ^43,45 for the French urban sites.

In the four previously cited studies, a SA method was performed using PMF. The methodology of PMF with all of the selected parameters has been described in detail in ref. ⁵⁴. Following the PM SA, OP SA was performed in these studies, aimed to enlighten the redox activities of PM sources using MLR. The PMF and MLR methodologies are detailed in Supplementary Information Section 8.

Matrix construction

The PM and OP SA results of these three studies were used to calculate OP_v values for each observation by multiplying the daily contribution of each source to the average OP_m of this source (estimated from the MLR technique with 500 iterations). By reducing the contributions of certain sources to PM₁₀, we can recalculate an average OP_v and create the PM reduction matrices. The SA results for the four urban sites (Grenoble, Talence, Nice and Barcelona) could be investigated together because they were obtained with harmonized PMFs using a single set of constraints. Traffic source was not identified in Port-de-Bouc; consequently, the two sources studied were biomass burning and industrial sources for the industrial sites.

Targets and projections

Direct results from PMF in the urban sites were used to create a parallel matrix of PM₁₀ mass values (Supplementary Fig. 4). On the basis of this matrix of PM₁₀ mass values, the boundary position for PM mass concentration thresholds on the OP_v matrices was calculated (Fig. 3 and Supplementary Fig. 5). The value in the first box at the top left of the OP_v matrix corresponds to the mean OP_v level in the four urban sites for a reference period of March 2012 to March 2023 (measurement periods for the PMF of the three sites). The target for 2030 considered here is based on the NECD⁴⁶, that is, a 49% reduction in PM_2.5 emissions relative to 2005 levels. In our projections, we assume an equivalent 49% reduction in PM₁₀ emissions from traffic and biomass burning sources. On the basis of historical emissions in Europe (EU27), such a reduction relative to 2005 would require a further approximately 35% reduction in PM₁₀ emissions compared with our reference period (2012–2022). To estimate this PM₁₀ emission reduction, we used the historical data of anthropogenic emissions in Europe (EU27) from the EU emission inventory report 1990–2022 under the UNECE Convention on Long-Range Transboundary Air Pollution. The target for 2040 corresponds to the OP_v level achievable with a 60% reduction in PM₁₀ emissions from traffic and biomass burning sources, which aligns approximately with a 90% reduction in greenhouse gas emissions relative to 1990 (based on the European Commission’s impact assessment of February 2024 (ref. ⁴⁷)). In these projections, we considered a reduction in the two main anthropogenic sources (traffic and biomass burning) and the contribution from other anthropogenic sources was not accounted for.