Introduction

The negative impact that aerosols have on human health (Liu et al. 2019), climate and the environment (Ramanathan et al. 2001) is now well established. Aerosols consist of a complex mixture of chemical components that vary depending on the sources and their consequent atmospheric processes (Perrone et al. 2013). Several studies have identified particulate matter (PM), especially its fine fraction (PM2.5), as a critical parameter for adverse health effects (Delfino et al. 2013; Canha et al. 2016). However, many questions remain about the underlying drivers of PM toxicity.

Oxidative stress induced by PM is proposed as the primary mechanism responsible for the adverse health effects associated with PM inhalation and its toxicity (Valavanidis et al. 2008). Oxidative stress can occur when the body’s antioxidant capacity is unable to counteract or detoxify the harmful effects of reactive oxygen species (ROS) due to an excess presence of ROS introduced into the body by PM inhalation (Bates et al. 2019). Recently, several studies have linked measurements of oxidative potential (OP) of aerosols to adverse health effects (Fang et al. 2016; Abrams et al. 2017).

There are several cellular and acellular methods for studying the OP of ambient particles, each with its own characteristics. To fully represent the toxicity of ambient PM, a combination of chemical and cellular assays is required to account for all aerosol components and mechanisms that may contribute to its toxicity. These methods include ascorbic acid (AA) (Fang et al. 2016), antioxidant-reduced glutathione (GSH) (Weichenthal et al. 2016), electron spin resonance (ESR) (Yang et al. 2015) and dithiothreitol (DTT) (Wang et al. 2019). The DTT method (OPDTT) is widely used and has been associated with airway inflammation markers (Delfino et al. 2013), cellular oxidative stress markers (Li et al. 2003), cellular cytotoxicity (Steenhof et al. 2011) and cardiorespiratory health endpoints in epidemiological studies (Bates et al. 2015; Yang et al. 2016). These findings support OP as a highly health-relevant air quality parameter.

However, the specific chemical species, aerosol sources and atmospheric processes that influence the OP of PM are not yet fully established (Paraskevopoulou et al. 2019). In Europe, most of the published studies on OP include a limited sampling for periods not exceeding a few months (Weber et al. 2018; Paraskevopoulou et al. 2019), and for Portugal, there are no studies on OP assessment of aerosols in the country. Therefore, the establishment of a relationship between PM toxicity (derived from OP evaluation), aerosol composition and emission sources remains to be defined in many European countries and, in particular, in Portugal.

The present study aimed to contribute to fulfil this lack of knowledge, being an exploratory study to characterize OP levels of PM2.5 sampled in an urban-industrial area of the Lisbon Metropolitan Area (Seixal, Portugal). For this purpose, fine aerosols were sampled for one year (December 2019 - November 2020, with a total of 128 sampling days) and their chemical composition was assessed to perform a source apportionment study using Positive Matrix Factorisation (Gamelas et al. 2023). A total of 7 different sources were identified, namely: soil, secondary sulphate, fuel-oil combustion, sea, vehicle non-exhaust, vehicle exhaust and industry (Gamelas et al. 2023).

From the total dataset, a subset of 30 samples was selected (considering the samples with the highest load for each source, both in mass and in %) and their OP was determined for their water-soluble fraction by the dithiothreitol method. The OP levels were then evaluated taking into account the contribution of the different pollution sources, assessed by Positive Matrix Factorisation (PMF), to the PM2.5 levels in order to understand their potential influence on PM toxicity.

Materials and methods

Study site

Fine aerosols were sampled at Aldeia de Paio Pires (38.617885º, − 9.080055º), located in the municipality of Seixal, a suburban and industrial area in the Lisbon Metropolitan Area (Portugal) which is one of the most densely populated municipalities in Portugal (with 167,294 inhabitants in 95.5 km2). The sampling site was located in the vicinity of busy motorways, several small and medium sized industries (including steelworks, lime factory, and metal waste management and treatment factory), a shipyard and the ferry terminal that connects the two river banks for commuting (Seixal – Lisbon). A detailed description of the sampling site and the study area is available elsewhere (Abecasis et al. 2022; Gamelas et al. 2023).

PM characterization, source apportionment and selection of samples

A full description of the PM characterisation strategy and methodology can be found elsewhere (Gamelas et al. 2023). Sampling was conducted from December 2019 to November 2020, and a total of 128 samples were obtained. Briefly, mass concentration of PM2.5 sampled over 24 h (using 47 mm diameter quartz filters) was determined by gravimetry and its chemical characterisation for 24 chemical elements was performed by Particle Induced X-Ray Emission (PIXE), together with the determination of three water-soluble inorganic ions by ion chromatography and the black carbon content by light absorption at 639 nm.

To perform the source apportionment, the receptor model PMF was used to identify the main emission sources and to estimate their contribution to the PM2.5 mass. The original data matrix was composed by a total of 128 samples and 24 PM species (only those variables with more than 70% of the data above the detection limit were considered). A total of 7 different pollution sources were identified as contributing to the PM2.5 mass, namely, soil (3.6%), secondary sulphate (18.4%), fuel-oil combustion (11.5%), sea (11.0%), vehicle non-exhaust (2.3%), vehicle exhaust (40.8%) and industry (12.6%). A description of the methodology and a detailed description of the results can be found elsewhere (Gamelas et al. 2023).

For the present study, thirty samples were selected from the full dataset and their selection was based on the highest load for each source (both in mass and %). Table 1 shows the selected samples, including the mass contribution of the different seven sources assessed to their PM2.5 levels.

Table 1 Selected PM2.5 samples with identification of mass contribution from 7 different pollution sources
Full size table

Oxidative potential analysis using the DTT assay

The oxidative potential of the water-soluble fraction of the sampled PM2.5 was assessed following the DTT assay, as fully described elsewhere (Chirizzi et al. 2017). The water-soluble fraction was extracted from 1/4 of each filter in 15 mL deionized water (DI, Milli-Q; > 18 MΩ) by sonication in a water bath for 30 min. The extracts were then filtered through 0.45 μm pore PTFE (polytetrafluoroethylene) syringe filters to remove insoluble materials and residual fibres. Sample extracts were then incubated at 37 °C with DTT (100 µM) in 0.1 M potassium phosphate buffer at pH 7.4 (5 mL total volume) for a period of 90 min. At specified times (namely 5, 10, 15, 20, 30, 45, 60, and 90 min), a 0.5 mL aliquot of the incubation mixture was picked up and 0.5 mL of 10% trichloroacetic acid was added to stop the reaction. This reaction mixture was then mixed with 2 mL of 0.4 M Tris–HCl, pH 8.9 containing 20 mM EDTA and 25 µL of 10 mM DTNB. The concentration of 5-mercapto-2-nitrobenzoic acid formed was measured by its optical density absorption at 412 nm, using a Cary 50 UV–Vis spectrophotometer (Varian Inc.). The rate of DTT consumption (δDTT, pmol/min) was determined from the slope and intercept of linear regression of measured absorbance versus time, where N0 is the initial moles of DTT used (Cho et al. 2005). Two replicates were made for 7 random samples and the variability (standard deviation) between replicates was taken as the uncertainty of the measured DTT activity.

The average DTT consumption on the blanks was 0.947 ± 0.119 nmol/min (total of 4 different filter blanks, with 2 replicates each). The final DTT activity for a sample was calculated by subtracting the average value of the blanks from the value of each sample. The final DTT activity was normalised with respect to (1) the sampled air volume (V), which provides DTTV and depends on the emission rates and dilutions and characterises the human exposure to particles at a given location, and it indicates the oxidative potential per unit volume of air of the aerosol component; and (2) the collected aerosol mass (M), which provides DDTM and represents an intrinsic property of particles linked to sources (Chirizzi et al. 2017), which reflects the oxidative potential of PM per unit mass (Farahani et al. 2022).

Statistical analysis

Data handling and statistical analyses were performed using Excel and XLSTAT 2023.1.6 software programmes. The data were analysed using non-parametric statistics with a significance level of 0.050. Associations between parameters were obtained using Spearman correlations. Origin version 7.5 (OriginLab Corporation) was used to plot the results.

A multiple linear regression (MLR) analysis was performed between the contribution of the sources that showed a significant influence on \({OP}_{\text{V}}^{DTT}\) (from the analysis of the Spearman correlations) and the \({OP}_{\text{V}}^{DTT}\) values (as the dependent variable), in order to have an independent estimate of the source contributions to OP. For this, OP was assumed to be linearly linked to PM2.5 due to the different sources, according to the formula (Weber et al. 2018):

$${OP}_{\text{V}}^{DTT} = {m}_{PM}\times \beta +\epsilon$$

Where \({OP}_{\text{V}}^{DTT}\) is the dependent variable, i.e., a matrix (\(n\times 1\)) in nmol min− 1 m− 3, \({m}_{PM}\) is the (\(n\times p\)) matrix with the PM contributions attributed to each source by the PMF study in µg m− 3, \(\epsilon\)is (\(n\times 1\)) uncertainty matrix in nmol min− 1 m− 3, \(n\) is the number of samples and \(p\) is the number of sources. The estimator \(\beta\) matrix (\(p\times 1\)) represents the intrinsic OP of the sources (i.e., the OP per mass unit of PM attributed to a given source, expressed in nmol min− 1 µg− 1) and the intercept is expressed in nmol min− 1 m− 3.

The MLR was performed using the XLSTAT tool, taking into account the replacement of the source’s contributions by the overall mean when no value was available and imposing the intercept equal to zero.

Results

Oxidative potential of PM2.5

Figure 1 presents the oxidative potential (normalised to the mass) together with the PM2.5 levels of the studied fine aerosol filters. The samples had a mean DTT activity (normalised to the mass, \({\text{O}\text{P}}_{\text{M}}^{\text{D}\text{T}\text{T}}\)) of 12.9 ± 6.6 pmol min− 1 µg− 1, ranging from 3.5 to 31.8 pmol min− 1 µg− 1. The mean PM2.5 levels of the selected samples was 18.8 ± 15.9 µg m− 3, ranging from 4.5 µg m− 3 to 67.8 µg m− 3. Figure 1 also highlights the great variability of \({OP}_{M}^{DTT}\) levels between the analysed samples, and also that they clearly do not follow the trend of PM2.5 concentrations. It is noteworthy to highlight that the selection of samples was based on the most representative samples for each of the sources identified by the PMF study (due to their higher contribution in mass and % to the PM2.5 levels). In addition, Fig. 1 shows that samples having comparable PM2.5 concentrations may have significantly different \({\text{O}\text{P}}_{\text{M}}^{\text{D}\text{T}\text{T}}\) and \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) according to the sources influencing those samples. For example, sample Q97 influenced mainly by sulphate has significantly lower OP compared to sample Q108 mainly influenced by industry source. This is in agreement with previous works (Molina et al. 2020; Daellenbach et al. 2020; Clemente et al. 2023) that suggested that a linear relationship between OP and PM concentrations is not always observed. This is the reason why OP could be a metric for toxicity complementary to mass concentration of PM.

Fig. 1
figure 1

Oxidative potential, normalized to the mass (\({OP}_{M}^{DTT}\)) – mean and standard deviation when two replicates were done, and PM2.5 levels of the selected samples

Full size image

The DTT activity (normalised to the sampled volume, \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\)) of the selected samples showed mean levels of 0.21 ± 0.17 nmol min− 1 m− 3, ranging from 0.02 to 0.85 nmol min− 1 m− 3. As seen for \({\text{O}\text{P}}_{\text{M}}^{\text{D}\text{T}\text{T}}\), the analysed samples also show a high variability for the \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels, taking into account the selection of samples done considering the highest load (in mass and %) for each identified source. Table 2 provides an overview of the \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels of PM2.5 from different types of environments and countries. The results of the present study are similar to others studies carried out in other countries, such as in an urban-industrial area in France (Dunkerque − 0.36 ± 0.24 nmol min− 1 m− 3 (Moufarrej et al. 2020), urban backgrounds in Italy (Sarno − 0.19 ± 0.10 nmol min− 1 m− 3 (Cesari et al. 2019), Lecce − 0.40 ± 0.26 nmol min− 1 m− 3 (Chirizzi et al. 2017) and 0.29 ± 0.19 nmol min− 1 m− 3 (Giannossa et al. 2022), urban environments in Greece (Athens − 0.33 ± 0.20 nmol min− 1 m− 3 (Paraskevopoulou et al. 2019) and in India (Mumbai - median of 0.12 nmol min− 1 m− 3 in a low-traffic slum and median of 0.20 nmol min− 1 m− 3 in a high-traffic slum (Anand et al. 2022); and in different types of environments (urban, roadside, near-road and rural) especially during summer and fall in Atlanta, USA (Fang et al. 2015). Our results were found to be lower than those obtained in different types of urban environments in China (Liu et al. 2018; Chen et al. 2022; Wu et al. 2022), India (Patel and Rastogi 2018), Iran (Khoshnamvand et al. 2023), Brazil (Serafeim et al. 2023), Spain (in ’t Veld et al. 2023) and different types of environments in Switzerland (Grange et al. 2022).

Table 2 Comparison of \({OP}_{V}^{DTT}\) levels of PM2.5 from different types of environments and countries
Full size table

However, it is important to highlight that some of the above-mentioned studies had applied a slightly different extraction method of the PM2.5 filters, which can be translated into different levels of OP, such as the use of a 1:1 mixture of methanol and Milli Q water (Patel and Rastogi 2018) and the use of a simulated lung fluid (SLF) solution consisting of a mixture of Gamble’s solution and dipalmitoylphosphatidylcholine (DPPC) (Grange et al. 2022; in ’t Veld et al. 2023). However, \({\text{O}\text{P}}_{V}^{\text{D}\text{T}\text{T}}\) measured in Milli-Q water, Gamble and Gamble + DPPC solution have been shown to have no statistically significant differences (Calas et al. 2017). This highlights the need for a standardised OP procedure to allow direct comparisons of OP metrics between studies in different countries and performed by different laboratories.

Even limiting to cases with similar protocols, there is still a large spatial and seasonal variability, likely associated not only to PM2.5 mass concentrations but to the differences in chemical composition due to the sources acting on the different sites. It is therefore useful to mention that the values found here are comparable with average values observed in Europe and USA but significantly lower than those observed in China and India, likely as a consequence of both a higher PM2.5 concentration in these countries and to the difference on sources contributing to the observed concentrations.

\({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels were found to increase linearly with PM2.5 concentrations, with a R2 of 0.714 (p-value < 0.0001), as shown in Fig. 2. A similar trend in the correlation between PM2.5 and \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) has already been reported in other studies. In Shanghai (China), a study in urban environments found a high correlation between PM2.5 and \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) (ρ = 0.69) (Wu et al. 2022). In the USA, \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) was also found to be fairly well correlated with PM2.5 (with Pearson’s correlation coefficient, r, ranging from 0.49 to 0.86) sampled at seven sites in the southeastern USA (Fang et al. 2016). Similar results were also observed in Lecce (Italy), where a significant correlation (r = 0.57) was found between \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) and PM2.5 levels (Chirizzi et al. 2017); in Patiala (India), \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels were also found to increase linearly with PM2.5 concentrations (R2 = 0.44, p < 0.001) (Patel and Rastogi 2018), as well as in Tehran (Iran) for the summer dataset (R2 = 0.63) (Khoshnamvand et al. 2023). However, in the particular case of the present study, the high linear correlation found is highly dependent of the 3 samples with higher PM2.5 levels. Without those samples, the R2 decreases to a value of 0.24.

Fig. 2
figure 2

Correlation between PM2.5 levels and their associated oxidative potential, normalized to the volume (\({OP}_{V}^{DTT}\))

Full size image

Influence of pollution sources on oxidative potential of PM2.5

Considering that the mass contribution of the different sources to the PM2.5 levels was known from the previous source apportionment study by PMF, Spearman correlations between \({OP}_{\text{V}}^{DTT}\) and the contribution of each source to the PM2.5 levels were assessed and are described in Table 3. Three different sources showed a significant positive correlation with \({OP}_{\text{V}}^{DTT}\), namely fuel-oil combustion (ρ = 0.523, p-value = 0.012), vehicle exhaust (ρ = 0.647, p-value = 0.001) and industry (ρ = 0.463, p-value = 0.018), indicating their influence on OP levels and, consequently, their potential impact on the health of the citizens of the study area.

Table 3 Spearman correlations between \({OP}_{V}^{DTT}\) and the contribution of the different assessed pollution sources for PM2.5 levels
Full size table

To estimate the contributions of the significant PM sources to \({OP}_{\text{V}}^{DTT}\), a multiple linear regression (MLR) was applied considering \({OP}_{\text{V}}^{DTT}\) as the dependent variable and the contributions of fuel-oil combustion, vehicle exhaust and industry as the independent variables, following the methodology described elsewhere (Weber et al. 2018; Cesari et al. 2019; Giannossa et al. 2022). The results of the MLR methodology are characterised by a R2 = 0.824 (R2 correct = 0.804) and a RMSE (root mean squared error) of 120. In addition, the F-test results (at a 95% confidence level) indicated that the model provided a good fit of the three variables considered. Considering the R2, 82% of the variability of the dependent variable \({OP}_{\text{V}}^{DTT}\) is explained by the contribution of the three sources and, based on the Type III sum of squares, the variable vehicle exhaust was the most influential independent variable.

Table 4 presents the results of the MLR analysis in terms of coefficients of β, standard errors, p-values and 95% confidence intervals. The model provided a good fit for all the sources considered (p-values below 0.050). Results showed that the different sources contributed differently to \({OP}_{\text{V}}^{DTT}\) of PM2.5 levels in the study area. The source with the highest contribution for the intrinsic \({OP}_{\text{V}}^{DTT}\) of the studied aerosols was vehicle exhaust (standardized β coefficient = 0.471), followed by fuel-oil combustion (standardized β coefficient = 0.366) and, finally, by industry (standardized β coefficient = 0.222).

Table 4 Multiple linear regression (MLR) analysis results and parameters regarding the goodness of the fit produced by the model for each source
Full size table

Figure 3 presents the correlation between the measured and the reconstructed \({OP}_{\text{V}}^{DTT}\) using the MLR, which shows a good correlation (R2 = 0.60), but with larger scatter on the comparable data, as also observed in other studies (Cesari et al. 2019; Giannossa et al. 2022).

Fig. 3
figure 3

MLR reconstructed vs. measured \({OP}_{V}^{DTT}\) of fine aerosol

Full size image

Figure 4 presents the relative contributions of the different sources to the PM2.5 and \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels. For the \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels, only the three identified sources used for the MRL methodology were considered. A high standard deviation in both cases is found since there is a high variability on the type of selected samples (which had influence of different sources). In average, fuel-oil combustion contributed 15.6% to the PM2.5 levels and 41.9% to the \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels, while vehicle exhaust showed a relative contribution of 27.0% to PM2.5 levels and 30.4% to the \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels. The industry source had a relative contribution of 19.3% to PM2.5 levels and 25.7% to the \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels. Overall, the contribution of sources to the PM2.5 levels were, in decreasing order, vehicle exhaust (27.0%), secondary sulphate (24.2%), sea (23.1%), industry (19.3%), fuel-oil combustion (15.6%), soil (8.7%) and vehicle non-exhaust (5.1%). Regarding the \({\text{O}\text{P}}_{\text{V}}^{\text{D}\text{T}\text{T}}\) levels of the PM2.5 sampled in the studied urban-industrial area, the highest contribution was from fuel-oil combustion (41.9 ± 37.3%), followed by vehicle exhaust (30.4 ± 34.7%) and, then, by industry (25.7 ± 30.0%).

Fig. 4
figure 4

Relative contributions of the different sources to PM2.5 levels and to \({OP}_{V}^{DTT}\) levels

Full size image

The influence of these three specific sources on the \({OP}_{\text{V}}^{DTT}\) of PM2.5 levels obtained in the present study is consistent with several studies around the world that have also identified their contribution to the OP levels of fine aerosols. A study in the urban environments of Shanghai (China) found that the sources contributing to the \({OP}_{\text{V}}^{DTT}\) of PM2.5 levels (using PMF) were mainly biomass burning (42%), followed by vehicle emissions (28.9%), road dust (17.4%) and incomplete combustion (11.8%) (Wu et al. 2022). In the coastal city of Ningbo (China), a PMF study identified five sources (out of six contributing to PM2.5 levels) that contributed to \({OP}_{\text{V}}^{DTT}\), namely, road dust (71%), marine vessels (9%), industry emissions (8%), vehicle emissions (7%) and secondary inorganic aerosol (5%) (Chen et al. 2022). In Italy, several works have investigated the main sources contributing to the oxidative potential of fine aerosols. A study in Sarno (Italy) identified three major sources contributing to the \({OP}_{\text{V}}^{DTT}\) of PM2.5 levels (using PMF), namely, vehicle traffic with secondary nitrate (≅ 50%), biomass burning (≅ 35%), and industrial emissions (≅ 12%) (Cesari et al. 2019). In Milan, a study using an MLR analysis identified four main sources contributing to the measured DTT activity in PM2.5 levels, namely, biomass burning (41%), secondary organic aerosols (20%), resuspended road dust (18%), and vehicular emissions (16%) (Hakimzadeh et al. 2020). In Lecce, the majority of \({OP}_{\text{V}}^{DTT}\) (about 50.6%) in PM2.5 was due to combustion sources (biomass burning and traffic), while natural and soil sources (sea spray, crustal and carbonates) contributed only 13.6% (Giannossa et al. 2022).

In Switzerland, road traffic (especially non-exhaust emissions) and wood burning were identified as the most important drivers of \({OP}_{\text{V}}^{DTT}\) of fine aerosols (Grange et al. 2022). In the USA, the main sources contributing to \({OP}_{\text{V}}^{DTT}\) of PM2.5 levels (using PMF) were biomass burning (35%), secondary water-soluble organic carbon (WSOC) (19%), vehicle emissions (both from mechanically generated – brake/tire wear – and from combustion – tail pipe emissions) (16%), secondary sulfate (12%), road dust (9%) and residual (9%) (Fang et al. 2016). A study in São Paulo (Brazil) identified four major sources contributing to \({OP}_{\text{V}}^{DTT}\) by using a PMF-MLR approach, namely biomass burning (40.7 ± 26.3%), followed by industrial emissions (24.3 ± 22.5%), vehicle emissions (20.5 ± 15.2%), and secondary aerosol (14.5 ± 15.0%) (Serafeim et al. 2023).

Considerations

The present study provides the first insights into the OP of fine aerosols in Portugal, using a limited number of samples (30) and focusing only on samples with the higher contributions from the different assessed pollution sources (by higher mass and %). Despite the lower number of samples, it was still possible to identify three pollution sources that significantly influence the OP levels of PM2.5. Future work will aim to assess the OP of the remaining samples (total dataset of 128 PM2.5 filters, corresponding to a one year of sampling done between December 2019 and November 2020) to confirm the influence of the sources identified in this study on OP levels of the fine aerosols in the study area. Moreover, the evaluation of the complete dataset (corresponding to one year of sampling) will also allow to assess the seasonal variability of the OP and to provide a more complete characterisation of the OP of the fine aerosols of this Portuguese urban-industrial area.

It is important to highlight the relevant role of wood burning as one of the main sources contributing to the oxidative potential of aerosols, as shown by the various studies cited above and others (Borlaza et al. 2021). Naturally, the nature of the sources affecting a particular region may have different origins, from local to regional sources. The previous source apportionment study (Gamelas et al. 2023) carried out on the PM2.5 samples analysed in this study did not identify wood burning as a source in the study area. However, specific tracers of this source were not evaluated during the PM characterisation (as levoglucosan)(Calvo et al. 2013), and this fact may have contributed to the non-identification of this source for the PM levels in the study area. Considering the known impact of wood burning to the OP levels of aerosols, future studies in the study area should include the evaluation of tracers of this source to assess whether its contribution to the OP levels is significant or not.

Conclusions

To the best of our knowledge, this is the first study providing information on the OP levels of fine aerosols in Portugal. The values of \({OP}_{\text{V}}^{DTT}\) in the studied urban-industrial area (with a mean of 209 ± 174 pmol min− 1 m− 3) were in agreement with others evaluated in European countries, India and the USA, and lower than those registered in some Asian countries, Brazil and some European countries (namely, Spain and Switzerland, where the extraction method of PM2.5 used was slightly different).

\({OP}_{\text{V}}^{DTT}\) was found to be significantly correlated with PM2.5 levels. Furthermore, based on the information on the pollution sources contributing to the PM2.5 levels, it was possible to identify three pollution sources that had a significant influence on the OP of these fine aerosols, namely, fuel-oil combustion, vehicle exhaust and industry. Using an MLR approach, it was found that these three sources contributed to 82% of the \({OP}_{\text{V}}^{DTT}\) of the studied fine aerosols. This information is useful to define targeted mitigation measures to tackle these sources of pollution, in order to minimise the health impacts due to their exposure.

Future work should aim at evaluating of the seasonal OP variability in the studied area and in other types of environments in Portugal, in order to strengthen the information on this indicator of the health potential of aerosols. In addition, special efforts should be made to understand which are the drivers of OP and the influence of the different pollution sources.