9-year trends of PM 10 sources and oxidative potential in a rural background site in France

. Long-term monitoring at sites with relatively low particulate pollution could provide an opportunity to identify changes in pollutant concentration and potential effects of current air quality policies. In this study, a 9-year sampling of PM 10 (particles with an aerodynamic diameter below 10 µm) was performed in a rural background site in France (Observatoire 15 Pérenne de l’Environnement or OPE) from February 28, 2012 to December 22, 2020. The Positive Matrix Factorization (PMF) method was used to apportion sources of PM 10 based on quantified chemical constituents and specific chemical tracers analysed on collected filters. Oxidative potential (OP), an emerging health-metric that measures PM capability to potentially cause antioxidant imbalance in the lung, was also measured using two acellular assays: dithiothreitol (DTT) and ascorbic acid (AA). The sources of OP were also estimated using multiple linear regression (MLR) analysis. In terms of mass contribution, the dominant 20 sources are secondary aerosols (nitrate- and sulphate-rich) associated with long-range transport (LRT). However, in terms of OP contributions, the main drivers are traffic, mineral dust, and biomass burning factors. There is also some OP contribution apportioned to the sulphate- and nitrate-rich sources influenced by processes and aging during LRT that could have encouraged mixing with other anthropogenic sources. The study indicates much lower OP values than in urban areas. A substantial decrease (58% reduction from year 2012 to 2020) in the mass contributions from the traffic factor was found, even though this is not 25 clearly reflected in its OP contribution. Nevertheless, the findings in this long-term study in the OPE site could indicate effectiveness of implemented emission control policies, as also seen in other long-term studies conducted in Europe, mainly for urban areas.

areas, where most emissions originate from and where populations are impacted. Further work has also been carried out in more specific areas to understand particular processes of aerosol formation and transport, as well as specific sources such as in the boreal forest (Yan et al., 2016), polar environments (Barrie and Hoff, 1985;Moroni et al., 2016), high altitude (Rinaldi et al., 2015), or marine sites (Scerri et al., 2016). Rural sites are of great interest as well because they can represent the regional 35 background of the atmosphere and potential influence from long-range transport (LRT) of pollutants. Studies at such sites provide more understanding of large-scale and mesoscale processes (Anenberg et al., 2010;Mues et al., 2013;Konovalov et al., 2009), which can be useful in chemical transport models.. The continuing can lead to the identification of long-term trends and the effect of recent changes in the source emissions. Indeed, several programmes have been set to monitor atmospheric composition in a harmonised way for background areas throughout Europe and North America. Among these are ACTRIS 40 (Aerosol, Clouds and Trace Gases Research Infrastructure) (Pappalardo, 2018), EMEP (European Monitoring and Evaluation Programme) (Aas et al., 2012;Alastuey et al., 2016), IMPROVE (Interagency Monitoring of Protected Visual Environments) (Hand et al., 2012), and CAPMoN (Canadian Air and Precipitation Monitoring Network) (Nejedlý et al., 1998). However, only few sites provide long-term in-depth series of PM chemical speciation data.
Further, although current air quality standards are based on ambient mass concentration of PM, there is also a growing interest 45 in new types of characterization that take into account not only particle composition but also its capability to generate health impacts (Park et al., 2018;Crobeddu et al., 2017;Møller et al., 2010;Bates et al., 2019). This is the case of PM oxidative potential (OP) (Nel, 2005;Conte et al., 2017;Yang et al., 2014;Verma et al., 2014), the ability of PM to generate reactive oxygen species (ROS) leading to PM-induced oxidative stress in the lungs. In France, several studies have reported about OP of ambient PM in different urban environments (Weber et al., 2018(Weber et al., , 2021Borlaza et al., 2021;Calas et al., 2019;Daellenbach 50 et al., 2020), but there are still limited studies performed in rural areas. The characterization of PM sources and OP in a rural site will enable us to investigate the large-scale effects of mitigation policies that target reduction of PM mass concentrations.
This will also provide knowledge of the efficiency of current air quality guidelines in terms of other emerging health-based metrics of PM exposure.
The understanding of trends of PM are essential to evaluate the effects of mitigation policies on air pollution levels. A reference 55 background site offers a good opportunity to gauge the broad effects of certain improvements in the transportation fleet and other regulations aimed at reducing vehicular emissions in large cities. Thus, in this study, an extensive dataset of PM over a 9-year period (n = 434), obtained from a French national background site, was investigated to: (1) provide insights on the longterm trends of PM sources and other emerging health-based metrics of PM exposure, such as OP of PM, (2) quantify the temporal evolution of the contributions of these sources, particularly focusing on vehicular emissions that have already been 60 shown to decrease in urban environments in Europe during the last decades.

Site description and sampling parameters
The OPE (Observatoire Pérenne de l'Environnement) sampling site is managed by the French national radioactive waste management agency (ANDRA). It is located in a remote area in the north-eastern part of France (48. 5°N, 5.5°E) at an altitude 65 of 390 m above sea level, in a large agricultural area without any residential areas within several kilometres (Figure 1). The mean annual temperature between 2011 and 2018 in the area was 10.5°C [minimum, maximum: -15.2°C, 36.4°C], average cumulated yearly precipitation was 829 mm, and the predominant local wind regimes are south-westerly and east-northeasterly winds . This site, being far from any local anthropogenic sources, is considered representative of the rural atmospheric background of north-eastern France. Golly et al. (2019) also demonstrated that the PM chemistry at this 70 site is very close to that in several other background rural sites in France.
All filters were preheated at 500 °C for 12 hours before use to avoid organic contamination. Blank filters (about 10% by number of the actual filters) were also collected to quantify detection limits and to secure the absence of contamination during sample transport, setup, and recovery. After collection, all filter samples were wrapped in aluminium foil, sealed in zipper 80 plastic bags, and stored at <4 °C until further chemical analysis. After collection, samples were subjected to various chemical analyses to perform the quantification of the major constituents by mass and specific chemical tracers of sources needed for the PMF model. These analyses were performed in the same laboratory for all samples (n=434) during the entire sampling duration (February 28, 2012to December 22, 2020. The carbonaceous components (organic carbon (OC) and elemental carbon (EC)) were analysed using a thermo-optical method on a Sunset Lab analyser (Birch and Cary, 1996), using the EUSAAR2 temperature program. Total organic matter (OM) in 90 daily ambient PM10 were estimated by multiplying the OC mass concentration by a factor 1.8. Yazdani et al. (2021) showed that this is consistent with the range estimated for rural samples from the IMPROVE network, that are generally higher than for urban samples.
A set of other chemical analyses was performed on a single water extraction of each filter. This extraction was performed in a 10 ml of ultra-pure water under vortex agitation for 20 minutes. The extract was then filtered with a 0.22 µm porosity 95 Nucleopore filter. The major ionic components (Na + , NH4 + , K + , Mg 2+ , Ca 2+ , Cl -, NO3 -, SO4 2-) and methane sulfonic acid (MSA) were measured by ion chromatography (IC, Thermo Fisher ICS 3000) following a standard protocol described in Jaffrezo, et al. (1998) and Waked, et al. (2014). An ICS300 (Thermo-Fisher) with AS11 HC column for the anions and CS16 for the cations was used.
The analyses of anhydro-sugars and primary saccharides were achieved using a high-performance liquid chromatography with 100 pulsed amperometric detection (HPLC-PAD). The samples collected before year 2017 was analysed using a set of Metrohm columns (MetroSep A Supp 15 and Metrosep Carb1) on Dionex DX500 HPLC. The samples collected after year 2017 were analysed with a Thermo-Fisher ICS 5000+ HPLC equipped with 4 mm diameter Metrosep Carb 2×150 mm column and 50 mm pre-column. The analytical run is isocratic with 15% of an eluent of sodium hydroxide (200 mM) and sodium acetate (4 mM) and 85% water, at 1 ml min -1 . These methods allowed the quantification of anhydrous saccharides (levoglucosan and 105 mannosan) and polyols (sum of arabitol and mannitol) as described in detail in Waked et al. (2014) and Samaké et al. (2019).
All procedures have been performed following the related EN standards (i.e., EN 12341, EN 14902, EN 16909, EN 16913). A quality control of the chemical analyses including a mass closure test, available in the supplementary information (SI, see S1).
A summary of the quantification limits (QL) on each chemical specie measured in the OPE site is also provided in Table S1.
Finally, our group successfully and regularly participates in inter-laboratory comparison exercises for OC and EC within 115 ACTRIS, and in EMEP (European Monitoring and Evaluation Programme) for ions analysis. The PM10 measurements from the tapered element oscillating microbalance (TEOM) are all in a daily (24-hour, 09:00 to 09:00) resolution, while the reconstructed PM10 were obtained from chemical analysis performed on filters collected on a weekly (7 days, 09:00 to 09:00) or daily (24-hour, 09:00 to 09:00) basis. A total of 299 out of 434 (69%) TEOM measurements were paired with reconstructed PM10 data, due to many interruptions in the TEOM functioning, in order to evaluate the semi volatile mass missing in the 120 mass reconstruction with filter chemistry.

Oxidative potential (OP) analysis
The OP analysis was performed on PM10 extracts from collected filter samples using a simulated lung fluid (SLF) solution composed of a Gamble + DPPC (dipalmitoylphosphatidylcholine) at 25 µg ml -1 iso-mass concentration (Calas et al., 2017).
This methodology facilitates particle extraction in conditions closer to lung physiology. The OP analysis only started on 125 samples collected from June 13, 2017 to December 22, 2020, amounting to a total of 191 samples.
Two assays were used to characterize OP activity: (1) dithiothreitol (DTT) and (2) ascorbic acid (AA) assays, as briefly described in the following sections. The volume-normalized OP activity ( ) is the OP consumption (nmol min -1 ) normalized by the sampled air volume (m 3 ), representing the OP exposure in each sample. All samples analysed were subjected to triplicate analysis and each sample results in the mean of such triplicate. The common coefficient of variation (CV) is between 0 and 130 10% for each assay.
DTT is used as a chemical surrogate to mimic in vivo interaction of PM with biological reducing agents, such as adenine dinucleotide (NADH) and nicotinamide adenine dinucleotide phosphate (NADPH), in the DTT assay. The consumption of DTT in the assay represents the ability of PM to generate ROS (i.e., superoxide radical formation) (Cho et al., 2005). The PM10 extract is mixed with the DTT solution. Afterwards, the remaining DTT that did not react with PM10 is reacted with 5,50-135 dithiobis-(2-nitrobenzoic acid) (DTNB). This reaction produces 5-mercapto-2-nitrobenzoic acid or TNB. The TNB is measured by absorbance at 412 nm wavelength using a plate-reader (TECAN spectrophotometer Infinite M200 Pro) with 96well plates (CELLSTAR, Greiner-Bio) in a 10-minute time step interval for a total of 30 minutes of analysis time.
AA is a known antioxidant used in AA assays using a respiratory tract lining fluid (RTFL) (Kelly and Mudway, 2003). This antioxidant prevents the oxidation of lipids and proteins in the lung lining fluid (Valko et al., 2005). The consumption of AA 140 also represents PM-induced depletion of a chemical proxy (i.e. cellular AA antioxidant). The mixture (PM10 extracts reacted with AA) is injected into a 96-well multiwall plate UV-transparent (CELLSTAR, Greiner-Bio) and measured at 265 nm absorbance using a plate-reader (TECAN spectrophotometer Infinite M200 Pro) in a 4-minute time step interval for a total of 30 minutes of analysis time.
Both DTT and AA assays measure OP by depletion of specific chemical proxies, cellular reductants (for DTT) and antioxidants 145 (for AA). Studies have well-identified a large number of PM constituents that influence OP concentrations. At least, OP assays are known to be associated with some metals (Cu, Fe, Mn among others) and some organic species (especially photochemically sensitive species such as quinones) (Calas et al., 2017, Charrier et al., 2014, Pietrogrande et al., 2019. However, in ambient air, each assay reports its own associations that may vary according to the local context (emission sources, local transport leading to various ageing processes and spatiotemporal variations) (Gao et al., 2020). Hence, a synergetic approach 150 using multiple OP assays, to capture the most complete information regarding PM reactivity, is commonly suggested. (Bates et al., 2019, Calas et al., 2017, Borlaza et al., 2021).
In every experiment, a positive control test is performed to ensure the accuracy and precision of measurements. A 1,4naphthoquinone (1,4-NQ) solution was used for both DTT (40 μl of 24.7 μM stock solution) and AA (80 μl of 24.7 μM 1,4-NQ solution) assays. The CV of the positive controls were <3% for the 2 assays. Additionally, an ambient filter collected from 155 the lab roof (with an expected constant OP value) was added on each microplate to ensure precision of OP measurements.

PMF model and input variables
The United States Environmental Protection Agency Positive Matrix Factorization (EPA PMF 5.0) software (Norris et al., 2014) was used to identify and quantify the major sources of PM10. PMF is a receptor model fully described by Paatero and 160 Tapper (1994) and is now widely used for source apportionment around the world. Additional information about the model description is provided in the SI (S2).
In this study, 23 chemical species were used as input variables, namely OC, EC, ions (Na + , NH4 + , Mg 2+ , Ca 2+ , NO3 -, SO4 2-), trace metals (Al, Cu, Fe, Rb, Sb, Se, Sn, Ti, Zn) and organic markers (MSA, levoglucosan, polyols (sum of arabitol and mannitol)). We assumed that arabitol and mannitol came from a similar source and, hence, combined them into one component 165 named as "polyols" (Samaké et al., 2019). The uncertainties of the input variables were calculated based on Gianini et al. (2012). Finally, the species displaying a signal-to-noise ratio (S/N) lower than 0.2 were discarded and those with S/N between 0.2 and 2 were classified as "weak", consequently multiplying the uncertainties by a factor 3.

Criteria for a valid solution
Solutions with a total number of factors from 6 to 11 were tested for the baseline models. Following the recommendations of 170 the European guide on air pollution source apportionment with receptor models (Belis, 2019), the Q/Qexp ratio (<1.5), the geochemical interpretation of the factors, the weighted residual distribution, and the total reconstructed mass were evaluated during factor selection. Finally, the factor chemical profiles obtained during this study were compared with those from previous studies in France, using the PD-SID method (Belis, 2019;, in order to validate their proper similarity.

Appropriate constraints in the PMF model
A set of constraints were applied on a basic model solution, in order to refine the results of the mathematical model by providing sound geochemical knowledge. Hence, the usual constraints as discussed in  and some constraints 185 corresponding to the traffic source following Charron et al. (2019) were also tested on the model (Table S3). However, these set of constraints were tested with caution as most of them have been previously applied only on sites with different typologies (i.e., urban or roadside sites), questioning their applicability in a rural site such as the OPE site. Finally, only a limited set was applied to generate the final solution (S1). After application of the constraints, a BS method was re-applied to verify the stability of the model. 190

Similarity assessment of chemical profiles
To investigate further any differences in the chemical profiles at the OPE site compared to those obtained at other French sites, a test of similarity was performed using the Pearson distance (PD) and standardized identity distance (SID) metric. This is calculated using Eq. S5 and Eq. S6 in the SI (S2) (Belis et al., 2015), closely following a previous work by our group . This comparison is based on the source relative mass composition, which allows the evaluation of the variability 195 of PMF solutions across different sites. In this case, the chemical profiles obtained for the OPE site were compared against 15 different other sites over France. A "homogenous source" tends to have a similar profile over different site types and should have PD<0.4 and SID<1.0 (Pernigotti and Belis, 2018). Conversely, the sources with PD and SD values outside of this range are considered as "heterogeneous sources".

OP source contribution estimation 200
The OP contribution of each PM10 source was determined by performing an OP deconvolution method using multiple linear regression (MLR) analysis. This methodology is based on the procedure proposed in Weber et al. (2018).. Briefly, the OP activity in nmol/min/m³ was used as the dependent variable, while the PMF-resolved source PM10 mass contributions in µg/m³ are the independent variables, as shown in Eq. 1: where is the observed daily (nmolreactant min⁻¹ m⁻³) with matrix size d×1, G is the PMF-resolved source contribution (µg m⁻³) of size d×n, and β is the regression coefficient representative of the intrinsic OP ( ) (nmol min⁻¹ µg⁻¹) of each n source. Finally, ε is the residual between the observed and modeled OP (nmolreactant min⁻¹ m⁻³). The source-specific OP contribution is calculated by multiplying the regression coefficient of each source by the mass contribution of the source to PM10 (Gk × βk). This methodology is essentially based on that in Weber et al. (2018). 210

Season-trend (STL) deconvolution method
The STL (Season-trend deconvolution using locally estimated scatterplot smoothing) model is a versatile and robust statistical method allowing the decomposition of a time-series dataset into three components including trend, seasonality, and residual.
The trend provides a general direction of the over-all data; the seasonality is a repeating pattern that recur over a fixed period of time; finally, residual is the random fluctuation or unpredictable change in the dataset. The seasonal component allows to 215 eliminate seasonal variation from the time series, resulting to a smoothed trend line that shows the tendency of the time-series dataset. This method somehow takes into account the changes in seasonal cycles from year to year which could also delineate part of the effect of meteorology on the long-term trend of PM10.
To investigate the long-term trends of sources or species concentrations, the STL model (Cleveland et al., 1990) was applied on the monthly mean concentrations, as described by Eq. 2: 220 where : ( ) is the time series observed in monthly average, ( ) is the trend component of the signal, ( ) is the seasonal component, and ( ) is the residual part not explained by the trend and seasonal part. The frequency was set to 13 (i.e., 6 months before and after the current month) to account for yearly seasonality. This model uses an iterative algorithm that constantly minimizes the residual ( ) by successively adjusting the trend and seasonal components. It has to be noted that the 225 resulting ( ) and ( ) do not represent concentrations, but a statistical deconvolution of the input signal ( ). ( ) could then be negative and the trend ( ) should be interpreted as an elaborated "moving average" of the concentrations. To account for extreme events or outliers in the data, the impact of data points with very high residuals were given less weight in the estimation of the trend and seasonal components, using the "robust" option of the algorithm. The presence and strength of tendency was evaluated thanks to the ordinary least square (OLS) linear fit of the ( ) component against time, removing the 230 first and last 6 month of the time-series to avoid edge-effects. Note that due to the lack of PM10 measurements in July 2019, the concentrations for that month were arbitrarily set to the August 2019 values. This model was implemented in Python 3.8 making use of the statsmodels module (Seabold and Perktold, 2010).

Results and discussion
The sections 3.1 to 3.5 below discuss the concentrations, sources, and trends of PM10, while the sections 3.6 and 3.7 discuss 235 the OP measurements and sources.

PM10 and its major chemical components
The reconstructed mass of PM10 in the OPE site was calculated following Eq. S1 in the SI and is presented in Figure 2. The mass concentration of the reconstructed daily PM10 samples ranged from 2 to 51 µg m -3 with an overall average of 9 ± 7 µg m -3 a few times (n = 3) in the entire measurement period. These values are in the lower range of the concentrations reported for rural areas in Europe, ranging from 3 to 35 µg m -3 (Putaud et al., 2004), and are relatively close to the values found at a remote site in Revin (France, located 165 km away), as described in the SOURCES programme (average of 13 µg m -3 ) . Some changes in the concentration can be observed in the PM10 mass concentration, but there are no drastic changes in the major chemical components at the OPE, even with the lockdown restrictions during year 2020. The yearly average volatile 245 mass (i.e., unaccounted by chemical analysis), deduced from the difference between TEOM-FDMS measurements and reconstructed PM10, ranges from 9% to 44% with an average of 22% (of the yearly median) and is well within range generally found in a rural environment (Pey et al., 2009).

250
Accounting for 37% to 45% (based on year) of the reconstructed PM10 mass concentrations, organic matter (OM) is the largest contributor. The other main contributors are inorganic secondary species (NO3 -, NH4 + , nss-sulphate), suggesting a strong influence from LRT of pollutants. There are also contributions coming from dust and sea salt. Although all of these components are often dominated by specific emissions, they can be derived from a wide range of sources. For example, vehicular emissions are usually composed of both carbonaceous and metals species, while road dust are usually minerals and some metal species. 255 Understanding the sources (as with the PMF methodology) and transformation processes of PM proves to be an essential step for efficient air quality policies.

Statistical stability of the PMF solution
The final retained solution includes 9 factors as described in section 3.3. Only 71 out of 100 baseline runs (without constraints) converged for this solution, but most factors were 100% correctly mapped, except for the traffic factor (93%, 66 out of 71 260 converged solutions) and sulphate-rich factor (99%, 70 out of 71 converged solutions). Applying the constraints greatly improved the BS mapping to 100% on all factors. Adding constraints in the base model allowed refining of the model through addition of expert knowledge on the profiles, which lead to the increased model stability. In previous source apportionment studies, specifically by our group, there are common constraints used depending on the site type such as presented in Borlaza et al. (2020). 265 Particularly, the constraints for traffic-related factors reported in Charron et al. (2019) have been optimized for traffic and urban background sites in our previous works. However, these constraints appeared restrictive when applied in a rural typology such as the OPE site. In fact, the Cu-to-Sb ratios appeared unsuitable as this ratio was 6.3 in our final solution compared to 12.6 in Charron et al. (2019). Based on literature, the Cu-to-Sb ratio can range from 1.6 (Handler et al., 2008) to 12.6 (Charron et al., 2019) depending on site typology. The addition of this constraint resulted in or led to a non-convergent solution in the 270 OPE site. To avoid inconsistencies, the Cu/Sb constraint was excluded in the optimal solution. The OC-to-EC ratio in the traffic profile was also too restrictive for the model, as this ratio was 3.9 in our baseline solution against 0.44 in Charron et al. (2019). The OC/EC levels calculated in this profile also suggests a strong influence of secondary organic aerosols (SOA) (Johnson et al., 2006;Pio et al., 2011;Rodríguez González et al., 2003;Viana et al., 2006), instead of primary traffic emissions.
As OC in a rural site can undergo multiple re-transformations in the atmosphere from the emissions sources, this has led to a 275 wide range of OC-to-EC ratios as similarly found in , hence this constraint was excluded.
In the final model, some constraints were used as summarized in Table S3, which resulted to all factors being correctly mapped and all BS runs converged, suggesting a good improvement in the traffic (from 93% to 100%) and sulphate-rich (from 99% to 100%) factors as well as the overall statistical robustness of the model. The other constraints either resulted in a non-convergent constrained model and/or less robust BS results. This implies that in sites with strong influence of LRT, the appropriate 280 constraints tend to vary and an optimal PMF solution can be more difficult to achieve.
The challenge in adding the constraints may also be linked to the inherent nature of the PMF algorithm since it assumes chemical profiles are identical for the whole period of analysis. However, during the 9 years of this study, some chemical source profiles may have changed, notably the traffic factor. Indeed, an evolution of the car fleet in France and Europe could lead to the changes in the OC-to-EC ratio emitted by the vehicle, so this profile may also have changed during this period. For 285 this specific case, a rolling PMF approach (Canonaco et al., 2021) with statistically-mapped PMF profile could be useful to investigate the time-variability of a given profile, slightly evolving with time.

PMF solution description and PM10 contributions
The 9 resolved sources of PM10 in the OPE site include nitrate-rich (25% of average contribution to PM10 for the full period), sulphate-rich (15%), traffic (12%), mineral dust (16%), biomass burning (9%), fresh sea salt (4%), aged sea salt (6%), primary 290 biogenic (7%), and MSA-rich (7%). These factors were identified based on their chemical profiles and the mass loading of specific tracers, as summarised in Table S2 in the SI. The error estimations, chemical profiles, and temporal evolutions of the PMF-resolved sources are available in the SI (S3). Figure 3 represents the repartition of the chemical species in the different factors. The summed PM10 contributions from all sources showed very good mass closure (r=0.95) with PM10 mass reconstructed with Eq. S1, indicating very good model results. 295 can be apparent, such as for the biomass burning and nitrate-rich factors, which are more prominent in winter and spring, respectively, and the primary biogenic and MSA-rich factors, increasing in summer due to greater photochemical and biological activities. Figure 4 depicts the seasonal average contributions of the PM10 sources at the OPE site from years 2012 310 to 2020.
The nitrate-rich factor, identified by high loadings of NO3and NH4 + , has a strong seasonal pattern with maximum contribution to PM10 mass concentration especially in the months of March and April.
The sulphate-rich factor is identified by high loadings of SO4 2and NH4 + . There are also contributions from some metal species (Se, Zn, Cu, and Sb) in this factor, suggesting potential influence from road dust and/or non-tail pipe vehicular 315 emissions. A small portion of OC (5% of OC mass) is also observed in this factor. The presence of these metals remained, even when the number of factors was increased (up to 11 factors) during the PMF optimization process.
The aged sea salt factor is characterised by high loadings of Na + and Mg 2+ , with a certain amount of species originating from potentially anthropogenic sources such as nitrates (6% of NO3mass) and sulphates (19% of SO4 2mass) that can be attributed to mixing and transformation processes in the atmosphere. Interestingly, there are some contributions from EC (8% of EC 320 mass), Cu (11% of Cu mass), Sb (13% of Sb mass), and Se (19% of Se mass). This could imply potential mixing of aged sea salt with other anthropogenic source linked to these species (e.g., traffic, shipping). The minimal loadings observed in the contributions of Clin this factor can be a likely result of ageing processes occurring between sea salt and acidic particulate compounds such as nitric and sulfuric acid (Seinfeld and Pandis, 2016). This factor could also be associated to road salting in the winter, however there is no clear seasonality in the contributions to support this hypothesis. There was no added constraint 325 in this factor as our solution shows a Mg 2+ to Na + ratio at 0.06 while this ratio is usually found around 0.12 in sea-salt emissions (Henderson and Henderson, 2010).
The fresh sea salt factor is characterised by high loadings of Cl -(91% of Clmass) and some contributions from Na + (35% of Na + mass) and Mg 2+ (25% of Mg 2+ mass). This factor contributes 4% to total PM10 mass and, unlike the aged sea salt factor, it is less likely influenced by anthropogenic sources with extremely low contributions from carbonaceous and metal species. 330 The MSA-rich factor identified by high loadings of MSA (methanesulfonic acid), a known product of oxidation of dimethylsulfide (DMS) commonly from marine phytoplankton emissions Li et al., 1993). A small mass fraction of SO4 2-(7% of SO4 2mass) is also found in this factor, that may be due to the co-emission of DMS and non-sea salt 335 sulphates, but also resulting from the production of biogenic sulphate from DMS oxidation. Hence, MSA-rich could potentially be mixed with secondary inorganic aerosols as well. The measured MSA mass concentration showed weak correlations with specific ionic species from marine aerosols such as Na + (r<0.01) and Mg 2+ (r<0.01). This could indicate that marine biogenic emissions may not be the only source of this factor. Instead, this factor could be influenced by sources with terrestrial origins and/or from forest biota, as previously reported in other studies (Bozzetti et al., 2017;Golly et al., 2019;Jardine et al., 2015;340 Miyazaki et al., 2012). This factor also presents a clear seasonal pattern with maximum contribution from May to July due to higher photochemical activity and algae / microbial activity. Golly et al. (2019) reported a very coherent seasonal cycle for MSA concentrations over a large portion of the French territory, including at the OPE site.
The primary biogenic factor is characterised entirely by polyols. These species are emitted by fungal spores which partly explains the high loadings of OC found in this factor. This factor has a higher contribution to PM10 during the summer season, 345 consistent with the observations at other rural and urban sites (Samaké et al., 2019;Borlaza et al., 2021).
More details about the characteristics of primary biogenic aerosols can be found in Samaké et al. (2019). Briefly, meteorological conditions, such as high temperature and relative humidity, could facilitate the increase in their formation. This factor can also include some fraction of plant debris, identified by cellulose measurements, as discussed in Samaké et al. The biomass burning factor, a major contributor to PM10 during the winter season, includes most of levoglucosan and mannosan. This factor contains around 25% of the total EC mass, consistent with a combustion chemical profile. Trace elements like Rb and Sn are also found in this factor, rubidium being the major trace element with 21% of its mass being reconstructed in this factor. Due to the distance of any residential areas from the OPE site, the contributions of this factor to PM10 at 20% on winter average is much less than the contributions generally observed in most sites in France in winter (e.g., 355 urban, sub-urban or countryside sites), which are mainly in the range of 25 -70 % with an average of about 35% . The contribution at the OPE site most probably represents the average winter loading of the French national background of the atmosphere.
The traffic factor is the second factor where EC is a major contributor (52% of EC mass). The major trace elements found in the traffic factor are Cu, Sb, Sn, and Zn, and most of their masses are reconstructed in this factor. There is no seasonality 360 associated with this factor. However, its contribution to PM10 presents a decreasing trend over the sampling period from 2012 to 2020. This is also consistent with the decreasing trend found in EC mass concentrations over the same period, as presented in Figure S11. These findings present an opportunity to explore the potential decrease in traffic emissions observed in the OPE site, taken as a good proxy of the national background burden of the rural atmosphere. This is further discussed in section 3.5.
The mineral dust factor is mainly composed of Ca 2+ (78% of Ca 2+ mass), Al (84% of Al mass), and Ti (86% of Ti mass). 365 There are also contributions from other trace elements that could be originating from re-suspended road dust or non-tail pipe emissions such as Fe (65% of Fe mass), Cu (27% of Cu mass), Rb (51% of Rb mass), and Zn (16% of Zn mass). A fraction of it could therefore be of anthropogenic origin. Figure 5 presents the similarity plot (PD-SID distances) obtained for the 9 factors found in the OPE site compared to the French 370 sites included in the SOURCES programme . It is striking that most factors remained homogeneous within France, including both rural and urban sites. The most stable factors are nitrate-rich (lowest PD) and mineral dust (lowest SID).

Comparison of the source chemical profiles with previous results in France
The traffic factor also appears relatively stable, but presents some dissimilarities according to the high variation in the PD metric. A high PD variation generally indicates a difference in the chemical species that identify the main mass contribution of the profile. In fact, in the traffic factor, this variation between the OPE and other sites can be attributed to the variations in 375 the OC to EC levels, similar to the findings in . Compared to the other French sites, the OC to EC levels of this factor in the OPE site are much higher, which highlights a strong influence from LRT processes with SOA formation.
The aged sea salt and MSA-rich factors are the only ones positioned outside of the shaded box in Figure 5, indicative of heterogeneous profiles between the OPE and the other sites. The heterogeneity found in the aged sea salt profile can be attributed to the contributions of EC and some metals in this factor. These were not typically found in other sites in France and 380 could also be due to the mixing of this sea salt profile with other anthropogenic contributions as a result of LRT processes, as well as different aging processes. Similarly, the MSA-rich factor has previously shown site-to-site variations and a wide variation in the PD-SID metric , mostly attributed to the variability of the contribution of OC in some sites. This is also the case in the OPE site, with a large contribution of about 11% of OC in this MSA-rich factor. Despite these few differences, the very large similarity of the chemical profiles at OPE compared to those at all other sites in France will be 385 essential to the comparison of the intrinsic OP of sources in section 3.7.   Figure S10 in the SI presents the long-term trend of the observed PM10 in the OPE site. The PM10 levels appear to be consistent from 2012 to 2020 and there is no clear increasing or decreasing trend found in PM10 (r 2 =0.2, Table 1). However, there is a clear decline found in EC mass concentrations (r 2 =0.9) with a reduction of 22 ng m -3 year -1 (p≤0.01) ( Figure S11 in the SI).

Long-term trends of PM10 sources
This could indicate that the mass contribution of one or more sources contributing to EC should also be decreasing. Following 395 Germany and Italy, France places third on highly impacted countries in Europe from vehicular exhaust emissions (Anenberg et al., 2010). Through the years, a variety of vehicular regulations have been adopted to reduce traffic-related emissions, not only in France (Bernard et al., 2020), but also across Europe (Wappelhorst and Muncrief, 2019). The data obtained in the OPE site poses an interesting opportunity as it covers 9 years of sampling in a rural area, making it possible to investigate emission trends over a long period of time in a site representing a background atmosphere. 400 Using the model described in section 2.6 (Eq. 2), the STL deconvolution of the PMF-resolved sources in the OPE site were also investigated. It is extremely significant that the contributions from the traffic factor in the OPE site also decreased substantially, as presented in Figure 6. A very large reduction of 58% from year 2012 to 2020 based on average mass contribution and an overall yearly average reduction of 104 ng m -3 y -1 (p≤0.01) were found. In parallel, there is also a reduction observed in the sulphate-rich factor, proposed as a highly anthropogenic-derived factor (see Figure S12 in the SI), with 66% 405 reduction of average mass contribution and an over-all yearly average reduction of 72 ng m -3 y -1 (p≤0.01). Indeed, several other sources have shown lower but significant decrease of their mass contribution over the years (Table 1) in the OPE site, except for the fresh sea salt, nitrate-rich, and MSA-rich factors.
These findings allowed the unravelling of the decreasing trend in terms of source contributions by the STL model. The STL deconvolution was applied on all the identified sources, which clearly showed that the traffic source has the highest tendency 410 with a decreasing trend. The other major sources of PM, such as biomass burning, mineral dust, nitrate-rich sources, do not have as much decreasing tendency as the traffic factor. The internal annual variabilities of weather/climate conditions might not be the leading factors explaining these trends, as they would have affected PM sources in the same way. The downward trends found in our study are well consistent with other existing studies in Europe (Li et al., 2018;Sun et al., 2020;Salvador et al., 2012;Pandolfi et al., 2016;Gama et al., 2018;Amato et al., 2014), nearly all of them conducted in urban areas. Pandolfi et al. (2016) found a significant long-term decrease of the contributions from anthropogenic emissions (specifically a mixed industrial/traffic factor, -0.11 µg m -3 year -1 , 56% total reduction) in a regional background site in altitude 420 in northeast of Spain (Montseny, Spain) from 2004 to 2014. This is also consistent with a similar study in the metropolitan area of Madrid, Spain (Salvador et al., 2012) which showed a reduction of 32.7% attributed to traffic emissions, alongside the decrease of the carbonaceous and SO4 2in PM. In a southern Spain area (Andalusia), the same group also found a consistent decreasing trend of PM at some traffic and urban sites in the region (Amato et al., 2014).
Another long-term study in Central Europe (Sun et al., 2020) focusing on equivalent black carbon (eBC) concentrations found 425 decreasing trends in high-altitude Alpine sites located in Germany (-3.88% year -1 , [-10.15%, 0.56%]) and Switzerland (-3.36% year -1 , [-8.71%, -0.28%]). These findings are also consistent with results from other parts of Europe, with the largest decrease found in OC up to -48% (Cusack et al., 2012) and the decrease in PM has been associated to non-meteorological factors (Barmpadimos et al., 2012). Other studies with pluri-annual series of data on PM chemistry in rural environments in Europe It should be noted that the role of meteorology on the observed decrease in PM in these studies (including ours) cannot be 435 totally ruled out (Hou and Wu, 2016;Czernecki et al., 2017;Kim, 2019) and is generally not fully considered. In most cases, there is a complex interplay between PM and meteorological conditions that could increase or decrease PM mass concentration (Chen et al., 2020). Indeed, there are some studies at high-altitude or regional background sites that highlighted a concurrent role of changing meteorology and changes in frequency of Saharan dust advections to Europe (Brattich et al., 2020) in modulating the dust concentrations in the atmosphere. The study at Melpitz , despite an in-depth work 440 on the wind sector classification, does not address the impact of possible changing in the air mass origin on long-term changing origins.
The evolution of the absolute concentration of the traffic factor in µg m -3 at the OPE site was also compared to an evaluation of black carbon (BC) emissions in kilotons by the transport sector for overall France, provided by the CITEPA, the official agency in charge of the emissions inventory in France (https://www.citepa.org/fr/2021-bc/). Both series were converted in an 445 arbitrary level starting from 100, using 2012 as the base year (Figure 7). This figure shows an excellent agreement in the trend and in the total percentage decrease (%) for estimated BC emissions from traffic (-64%) and the traffic source contributions observed at the OPE site (-52%), between the years 2012 to 2020. While local or regional changes in meteorology may be a factor in the evolution of the concentrations observed, this is unlikely to be the dominant one in the evolution of the concentrations of chemical species related to traffic emissions, in light of the strong correlation observed with the national 450 emissions inventory in France.
The interesting points of our study is that it pertains directly to a specific source, identified with a long term and robust PMF study. Further, as there are minimal local anthropogenic sources expected in the OPE site, it may be safe to assume that these contributions to PM mass in this area are influenced by LRT of pollutants. Our results indicate that the implementation of emission control policies are also playing a role in the consistent decrease in traffic emissions in rural sites far away from 455 direct emissions.   Figure 8 presents the observed average values of OP of PM10 in the OPE site compared to the other sites in France (Calas et 465 al., 2018(Calas et 465 al., , 2019Weber et al., 2021Weber et al., , 2018. All series cover at least one year of sampling. As expected, the OP level in a rural background is much lower (about 2 to 8 times) than other typologies including traffic, urban, and urban alpine sites, for both AA and DTT assays. Further, the average ratio between urban sites and the rural OPE site is generally much higher for OP's than for PM mass, an indication that the nature of the particles at the rural site make them less oxidant than in urban areas, as already pointed out in Daellenbach et al. (2020). 470 Figure 9 presents the daily and monthly mean distributions of observed PM10 and OP activity ( and ) from June 13, 2017 to December 22, 2020. There is an observed seasonality, where PM10 and appear to be similar with relatively higher levels during warmer months. On the contrary, has slightly higher levels during colder months.

Temporal trends of observed OP of PM10
In many European studies, the seasonality in PM10 mass concentration can be usually explained by higher contributions from biomass burning during winter (Bessagnet et al., 2020;Tomaz et al., 2017), especially in alpine valleys (Calas et al., 2019b;475 Favez et al., 2010;Herich et al., 2014;Srivastava et al., 2018;Tomaz et al., 2016Tomaz et al., , 2017Weber et al., 2018Borlaza et al., 2021). Similarly, this seasonal pattern has been observed in OP as well (Borlaza et al., 2021;Weber et al., 2018;Calas et al., 2019b;Weber et al., 2021). However, the typology (i.e., rural) of OPE site could be associated with a different type of OP temporal profile as it is far from direct anthropogenic emission sources (but not from vegetation and soil biogenic emissions). With the large influence of LRT on the sources of PM in the OPE site, this also poses an opportunity to investigate on the impact of LRT on OP. In fact, few studies have looked into the effects of aerosol aging to OP properties (Guascito et al., 2021;Bates et al., 2019). Pietrogrande et al. (2019) reported association of OP to redox-active organics linked to photo-oxidative 485 aging. Using backward trajectory analysis,  found strong effects of LRT on OP of fine PM. This is also consistent with the findings in Paraskevopoulou et al. (2019) which revealed highly oxygenated secondary aerosols as one of the main drivers of OP in fine PM, further highlighting the importance of combustion and aging processes in OP. In a shipborne measurement study in South Korea, a higher intrinsic OP has also been found in samples where secondary aerosol formation is more dominant, also suggesting strong impacts of long-range transported PM (Oh et al., 2021). Cesari et al. (2019) found 490 negligible contribution from secondary sulphates, but have found relevant OP contributions from a factor identified as a combination of vehicular traffic and secondary nitrates. All these studies used DTT assay to measure the OP of PM.
Generally, current literature poses importance of the role of secondary/aged aerosols on OP of PM, especially those with influence from anthropogenic sources. However, to our knowledge, this is the first long-term OP study, especially dealing with ambient samples far from direct emissions at a rural background site. 500

Sources of OP in PM10
The sources of OP in PM10 was apportioned following an OP deconvolution method proposed by Weber et al. (2018) using the source contributions (µg m -3 ) obtained in the PMF and the measured OP (nmol min -1 m -3 ) at the OPE site. Generally, the modelled OP ( ) is within range of the observed OP, with a reasonable reconstruction ( (r=0.76) and (r=0.76)).
The of each PM source is given in Table 2, where can range from -0.01±0.02 to 0.10±0.03 nmol min -1 µg -1 and 505 can range from -0.001±0.02 to 0.16±0.03 nmol min -1 µg -1 . Generally, higher indicates higher redox-activity associated with the factor. There are some differences in the based on the type of assay used and this can be attributed to the sensitivity of the assay towards certain redox-active species in PM (Borlaza et al., 2021;Xiong et al., 2017;Charrier and Anastasio, 2012).
In terms of overall daily mean contribution, as presented in Figure 10, the main contributors to PM10 mass are the nitrate-rich, 510 mineral dust, and sulphate-rich factors in the OPE site. However, in terms of , the mineral dust factor showed the highest average contribution (0.15 nmol min⁻¹ m⁻ 3 ), followed by the sulphate-rich (0.11 nmol min⁻¹ m⁻ 3 ) and traffic (0.07 nmol min⁻¹ m⁻ 3 ) factors. For , the biomass burning factor showed the highest contribution (0.12 nmol min⁻¹ m⁻ 3 ), followed by the traffic (0.07 nmol min⁻¹ m⁻ 3 ) and nitrate-rich (0.06 nmol min⁻¹ m⁻ 3 ) factors. Although lower in magnitude, the OP contribution of mineral dust, traffic, and biomass burning (only in ) are also prominent in the OPE site, similar to other sites in France (Weber et al., 2021). These sources are commonly composed of species that are highly redox-active, hence it is not surprising that they are one of the main drivers of OP even in a rural site. 520 Both mineral dust and biomass burning are also sources that can be associated with LRT and aging, respectively, which are atmospheric processes linked to increased OP (Pietrogrande et al., 2019;Paraskevopoulou et al., 2019;Oh et al., 2021). Further, while the mineral dust profile in OPE is considered homogeneous with those determined in other parts of France, as discussed above, its chemical composition includes slightly larger fractions of some metals, particularly Fe and Cu, possibly making it more redox-active. 525 As observed in other studies (Daellenbach et al., 2020;Borlaza et al., 2021;Weber et al., 2021), there is also a clear difference 530 in source ranking when considering the PM mass or OP, highlighting that the sources driving PM mass are not the same as the ones driving OP activity. The mass contributions of the nitrate-rich factor can be twice that of the traffic factor. However, in terms of OP (both and ), the traffic factor contribution can be twice as much as that of the nitrate-rich factor. The biomass burning factor, with only <1 µg m -3 mass contribution on annual average, appears to have the highest contribution in . 535 Some previous studies associated secondary inorganic sources (SIA) with minimal contributions on PM toxicity (Cassee et al., 2013;Daellenbach et al., 2020;Park et al., 2018). However, the nitrate-rich factor apportioned in the OPE site showed contributions to both and , and the sulphate-rich factor to . In the sulphate-rich factor, a fraction of OC (5%) and metals (Se (42%), Zn (44%), Cu (27%), and Sb (25%)) were apportioned, while in the nitrate-rich factor there are contributions from OC (6%), EC (4%), and metals (Sb (6%) and Sn (11%)). These species are commonly anthropogenic-540 derived, signalling that the sulphate-and nitrate-rich factors could be influenced by these types of emissions as well. In fact, a similar study considered that both SIA factors can be associated to anthropogenic SOA sources (Borlaza et al., 2020).
Although both DTT and AA assays represent potential PM-induced oxidative stress, through in vivo interactions between redox active components in PM10 and biological oxidants, it can be observed that they differ in terms of source impacts. This can be attributed to the sensitivity of each assay to specific species and/or emission sources of PM. Nevertheless, most of the 545 sources of PM suggested to be anthropogenic-derived or impacted, such as traffic, mineral dust, nitrate-rich, sulphate-rich (only in ), and biomass burning (only in ), were all usually in upper half of the scale (Figure 8) in terms of contributions. The knowledge of source-specific contributions provide useful information on the main drivers of , even in a rural area such as the OPE site.
There is an interesting seasonality observed in the OP of PM, as previously shown in Figure 9. With the OP source 550 deconvolution method, this seasonality has been revealed in terms of PM sources, as presented in Figure 11. During colder months, the biomass burning factor clearly dominated the contributions. During warmer months, the contributions were dominated by the mineral dust factor. However, there is a relatively consistent monthly contribution for both assays coming from the traffic factor. 555 It is also interesting to note the negative contributions from some sources. This negative contribution is brought by a negative intrinsic OP ( obtained in the OP deconvolution method (Table 2)). This can be broadly interpreted as follows: for every 560 similar interpretation can be done for the primary biogenic factor and its contributions ( =-0.01±0.02, p=0.68).
However, it is important to note that both MSA-rich and primary biogenic factors do not always present a negative in every site investigated in our group. One cannot completely assume that these two factors always act as suppressors of OP of PM. In fact, these two factors have shown high variabilities across different sites in France (Weber et al., 2021). With the 565 use of fit-for-purpose organic tracers, possible mixing issues in these factors can be minimized (Borlaza et al., 2021). However, these supplementary tracers were not available in the OPE site, making it difficult to eliminate potential influence from other factors or species in PM.  Figure 12 presents the yearly contribution of each factor to the OP of PM10 for the 3 years investigated in this study. There is no clear decreasing trend in the total OP reconstructed by MLR analysis. Although there is a decreasing trend found in the mass contributions of the traffic factor, this was not clearly reflected in its OP contributions. The other sources seem to have comparatively consistent OP contributions from 2017 to 2020, and no notable tendency can be found in the total OP 575 contribution as opposed to the contributions to PM10 mass showing decreasing trends. This may be explained by the limited subset of samples for the OP assay (OP data spanning 4 years against 9 years for the PMF data), the shorter time range being insufficient to reach significance and robustness in the trend assessment of OP levels.

Limitations of the study
In spite of the advantages offered by the long-term monitoring in the OPE site, there were a few limitations encountered during the investigation of the dataset. Each limitation is discussed as follows:  There was a change in sampling duration between the collection performed in year 2012 to 2016 (7-day sampling) 585 and 2016 to 2020 (24-hour sampling). A 7-day filter sample includes both weekdays and weekends, whereas a 24hour sample will either be a weekday or weekend, depending on the sampling interval. This implies that the weekly collected samples may contain particles that are not fully captured in a daily sample. However, since the OPE site is quite distant from direct emissions, the expected difference in the weekday and weekend levels should be relatively small. Further, PMF source apportionment were conducted separately (i.e., 7-day samples versus 24-hour samples), 590 leading to very similar results for the chemical profiles and source contributions, justifying the coupled analysis. (see SI, Table S4 and Figures S13 to S21)  The lack of fit-for-purpose tracers to fully elucidate the influence of SOA in a site with relevant distance from direct emissions (or rural typology) such as OPE. In Borlaza et al. (2021), a secondary biogenic oxidation source was additionally identified using organic tracers (3-MBTCA and pinic acid), while anthropogenic influence was 595 supported by contributions of phthalic acid, even in secondary aerosol sources. With the typology in the OPE site, this would have been useful.
 The use of a single chemical profile for long-term PM source apportionment could be limiting for the PMF model.
As we have found consistent decrease in some species, particularly EC, perhaps a rolling PMF (e.g., yearly PMF) could better capture possible changes in the source profiles. 600  The absence of samples for OP analysis from years prior to 2017 have limited the investigation of long-term OP in the OPE site. Consequently, it was not easy to capture the decrease in OP contributions from the traffic factor as similarly captured in the mass contributions. Perhaps a hindcasting method on the years without OP data could have been performed, however that would heavily rely on the modelled from years 2017 to 2020, which can lead to bias in the results. 605  In the year 2020, a series of lockdown restrictions were placed nationally as a response to the coronavirus disease  pandemic. In the OPE site, there is no clear decrease in average PM10 mass concentration in 2020 ( Figure   12) that could have greatly affected the results of this study. In fact, excluding all samples from year 2020, the traffic factor contributions to PM10 still has a reduction of 39% from year 2012 to 2019 and an overall yearly average reduction of 135 ng m -3 y -1 (p≤0.01). 610

Conclusions
Over the 9-year analysis in a rural background site in France (OPE), the observed PM10 mass concentration and OP were found to be much lower than other sites in France. The sources of PM10 mass and OP were apportioned using PMF and MLR analysis, respectively. The nine identified factors relevant for PM10 include secondary inorganics (nitrate-and sulphate-rich), traffic, mineral dust, biomass burning, sea salts (fresh and aged), primary biogenic, and MSA-rich. 615 A redistribution of the factor impacts between mass and OP contributions was observed, underlining the importance of taking into account the redox activity of PM when considering their potential health effects. Based on PM mass, the major contributors are nitrate-and sulphate-rich factors, both factors being associated with secondary inorganics formed during long range transport (LRT). On the other hand, based on OP activity, the main contributors are mostly anthropogenic-derived sources such as traffic, mineral dust, and biomass burning factors. 620 As the OPE site is located far from direct anthropogenic emissions, the influence of LRT processes was noted in some sources.
Sources such as sulphate-and nitrate-rich, MSA-rich, and aged sea salt factors have shown potential mixing with other anthropogenic sources, most probably due to the transit time. These potential mixing and aging processes were reflected in the chemical mass profile of each factor as well as in their OP contributions.
Thanks to the long-term dataset in the OPE site, it was observed that the traffic factor contribution to total PM10 has decreased 625 over the years for this site that may well represent the French national background PM. This decrease is much larger than any change observed for the other PM sources and is in excellent agreement with estimations in the decrease in BC emissions from the transport sector all over France from the national inventory. This effect may be attributed to improvement of the exhaust emission of terrestrial transportation fleet, and/or to regulations restricting vehicular emissions in bigger cities and/or other regional-scale. However, persistent changes in meteorological conditions influencing the transport of air masses to OPE or 630 formation of PM during this transport cannot be totally ruled out.