Particulate pollutants in the Brazilian city of São Paulo : One-year investigation for the chemical composition and source apportionment

São Paulo in Brazil has relatively relaxed regulations for ambient air pollution standards and often experiences 20 high air pollution levels due to emissions of particulate pollutants from local sources and long-range transport of biomass burning-impacted air masses. In order to evaluate the sources of particulate air pollution and related health risks, a yearround sampling was done at the University of São Paulo campus (20 m above ground level), a green area near an important expressway. The sampling was performed for PM2.5 (≤ 2.5 μm) and PM10 (≤ 10 μm) in 2014 through intensive (every day sampling in wintertime) and extensive campaigns (once a week for the whole year) with 24 h of sampling. This year was 25 characterized to have lower average precipitation comparing to meteorological data, and high pollution episodes were observed all year round, with a significant increase of pollution level in the intensive campaign, which was performed during wintertime. Different chemical constituents, such as carbonaceous species, polycyclic aromatic hydrocarbons (PAHs) and derivatives, water-soluble ions and biomass burning tracers were identified in order to evaluate health risks and to apportion sources. The species such as PAHs, inorganic and organic ions and monosaccharides were determined by chromatographic 30 techniques and carbonaceous species by thermal-optical analysis. Trace elements were determined by inductively coupled plasma mass spectrometry. The associated risks to particulate matter exposure based on PAH concentrations were also assessed, along with indexes such as the benzo[a]pyrene equivalent (BaPE) and lung cancer risk (LCR). High BaPE and LCR were observed in most of the samples, rising to critical values in the wintertime. Also, biomass burning tracers and PAHs were higher in this season, while secondarily formed ions presented low variation throughout the year. Meanwhile, 35 vehicular tracer species were also higher in the intensive campaign suggesting the influence of lower dispersion conditions in that period. Source apportionment was done by Positive Matrix Factorization (PMF), which indicated five different factors: road dust, industrial emissions, vehicular exhaust, biomass burning and secondary processes. The results highlighted the contribution of vehicular emissions and the significant input from biomass combustion in wintertime, suggesting that most of the particulate matter is due to local sources, besides the influence of pre-harvest sugarcane burning. 40


Introduction
summertime. The main sources for PM 2.5 were automobile traffic and soil dust. However, biomass burning was not considered as a potential source by the authors.
In order to determine the water-soluble ions (Cl − , NO 3 − , SO 4 2− , C 2 O 4 2-, methylsulfonate, Na + , K + , NH 4 + ) 10 mL of deionized water was used to extract the sample aliquots, with 10 min of gentle rotation. The ions were determined using two ion 165 chromatography systems (ICS 2000 system, Dionex) simultaneously; cations were analyzed using a CG12A/CS12A column with an electrochemical suppressor (CSRS ULTRA II, 4 mm) and anions using an AG11/AS11 column with an electrochemical suppressor (ASRS ULTRA II, 4 mm).
Finally, trace elements in the samples were extracted using a microwave digestion system (MLS-1200 mega, Milestone Inc., Italy) at National University of Singapore. Punches of the filters were cut into small pieces and added into PTFE vessels with 170 4 mL HNO 3 (Merck), 2 mL H 2 O 2 (Merck) and 0.2 mL HF (Merck). The vessels were then subjected to a three-stage digestion inside the microwave digester (250 W for 5 min, 400 W for 5 min, and 600 W for 2 min). Following the digestion procedure, extracts were filtered with 0.45 µm PTFE syringe filters, diluted 8 times and stored in the 4 ˚C cold room. The concentrations of trace elements were quantified by ICP-MS (Agilent 7700, USA) in triplicates. The instrumental parameters maintained during sample runs using the ICP-MS analysis were: plasma gas (15.0 L min -1 ), auxiliary gas (1.0 L min -1 ), and 175 nebulizer gas (1.0 L min -1 ). Clean ceramic scissors and forceps were used to handle all PM samples. ICP-MS standards (purchased from High-Purity Standards, USA) were used for calibration.

Statistical analysis and receptor model
Pearson coefficients were calculated to verify the correlation between all the species (software STATISTICA). It determines the extent to which values of the variables are linearly correlated. The coefficients (r) were considered significant when p < 180 0.05. Two-tailed t-tests were also employed in order to evaluate equal and unequal variances (p < 0.05). Polar plots considered the mass concentrations as a function of wind speed and direction (software R x64 3.3.2).
The widely used source apportionment model, positive matrix factorization (PMF), was applied to all samples dataset (Paatero and Tapper, 1994). In this study, specifically, the EPA PMF5.0 software was used. Variables were classified as strong, weak and bad according to the signal-to-noise ratio (S/N), number of samples below the detection limit (Amato et al., 185 2016;Contini et al., 2016;Paatero and Hopke, 2003) and thermal stability of the species. The species were categorized as bad when the S/N ratios were less than 0.2 and weak when the S/N ratios were greater than 0.2 but less than 2 (Lang et al., 2015). Accordingly, species with S/N ratios higher than 2 were considered strong. Bad variables were excluded from the model and the weak ones had their uncertainty increased by a factor of 3, as described in the EPA PMF Fundamentals and User Guide (Norris et al., 2014).

190
When concentrations were below the detection limits, they were substituted by half the detection limit (DL). Missing data were replaced by the median (M) of the whole dataset for that species. Uncertainties were calculated by Eq. (1) according to Norris et al. (2014), when the concentrations were below the detection limits: (1) Uncertainty for missing data (Brown et al., 2015) is given by Eq. (2):

195
(2) When the concentrations were above the detection limit, uncertainty is determined from Eq. (3): ( Where EF is the error fractions and C is the element concentration. Q robust value (Q R ) is the goodness-of-fit parameter computed with the exclusion of points not fitted by the model. To 200 evaluate the number of factors, Q R was compared to Q T (Q theoretical value). At the point when changes in the ratio Q R /Q T become smaller with the increase of the number of factors, it can be demonstrative that there might be an excessive number of factors being fitted (Brown et al., 2015). Q T was estimated as in Lang et al. (2015), given by Eq. (4): Where n s is the number of samples, n e is the number of strong elements, and n f is the number of factors.

Results and discussions
3.1 Concentrations of PM 2.5 and PM 10 during extensive campaigns The extensive campaigns (Ext 2.5 and Ext 10 ) were carried out over a whole year, during which the meteorological conditions varied largely. The average temperature during the sampling days in all campaigns ranged from 14 to 26 °C and the wind 210 speed varied between 0.6 and 2.6 m s -1 ; most of the sampling was carried out on days without rainfall. In Fig. 2 the meteorological variables, PM 10 and PM 2.5 concentrations for all analyzed days are presented.
There were moderate negative correlations between PM 10 and average wind speed, average and minimum relative humidity; and between PM 2.5 and average wind speed and minimum relative humidity (Table S1). This observation is in agreement with the fact that days with lower relative humidity and lower wind speed present higher PM 2.5 and PM 10 levels than in more 215 humid and with more ventilation conditions.
In the extensive campaign, the PM mass concentrations exhibited a wide range of concentrations. For example, Ext 2.5 ranged from 8 to 78 µg m -3 (average 30 µg m -3 ), whereas Ext 10 values varied between 12 and 113 µg m -3 (average 44 µg m -3 ) (Fig.   3). The World Health Organization (WHO) recommends a daily limit for PM 10 of 50 µg m -3 and of 25 µg m -3 for PM 2.5 , (WHO, 2006) while the Brazilian Environmental Agency (CONAMA) recommends a threshold of 150 µg m -3 for PM 10 220 (CONAMA, 1990;Pacheco et al., 2017). When considering the CONAMA standards, only one day in the extensive campaign was near the target limit. The Ext 10 campaign was divided into two periods: dry (April to September) and rainy (October to March). It was observed that the average PM 10 was 52 µg m -3 in the dry period and of 35 µg m -3 in the rainy period.
A study done by Vasconcellos et al. (2011b)  since an annual mean of 40 µg m -3 is established for PM 10 and a limit value of 25 µg m -3 is imposed for PM 2.5 (Kassomenos et al., 2014). However, these averages in São Paulo are lower than the observed in year-round studies performed in Chinese megacities, such as Shanghai (83 µg m -3 for PM 2.5 and 123 µg m -3 for PM 10 ) and Nanjing (222 µg m -3 for PM 2.5 and 316 µg m -3 for PM 10 ) (Shi et al., 2015;Wang et al., 2003Wang et al., , 2013a

Concentrations of PM 2.5 during intensive campaign
The winter campaign began with high PM 2.5 concentrations (a maximum of 88 µg m -3 on 02 July) and low relative humidity 245 (minimum of 21 %). The average temperatures ranged from 15 to 21 °C and the wind speed ranged between 0.6 and 2.2 km h -1 . The concentrations of PM 2.5 in the intensive campaign ranged from 15 to 88 µg m -3 (average 45 µg m -3 ), with a similar average to that obtained in another intensive study in 2008, 47 µg m -3 (Souza et al., 2014a). The levels of PM 2.5 in this campaign were above the recommended by WHO in 90 % of the sampling days.
The average concentration of PM 2.5 was higher in Int 2.5 than in Ext 2.5 , which can be explained by the fact that the campaign 250 took place in the dry season (winter). In winter, the meteorological conditions are more unfavorable to the dispersion of pollutants and also due to the predominance of sugarcane burning (da Rocha et al., 2005(da Rocha et al., , 2012; Sánchez-Ccoyllo and Andrade, 2002;Vasconcellos et al., 2010).

WSI and trace elements
The WSI represent a large fraction in the aerosol mass and have already been suggested to present ability to form CCN

255
(cloud condensation nuclei) and fog (Rastogi et al., 2014). The secondary inorganic components, sulfate, nitrate and ammonium (SNA) were the most abundant ions in all campaigns (

270
although older vehicles are still allowed to use S-500 (500 ppm) diesel (CETESB, 2015). During the studies done in several urban sites in China, sulfate concentrations varied between 4200 and 23000 ng m -3 . These values are higher than those of this study and also 5 to 10 times higher than the measured concentrations in Europe and United States (Hidy, 2009;Putaud et al., 2004;Zheng et al., 2016).
The SO 4 2-/NO 3 ratio was nearly twice higher in the Ext 2.5 campaign than in Int 2.5 . This pattern has already been observed, in warmer ambient conditions the fine NO 3 aerosols can be volatilized, increasing the ratio between these species (Rastogi and Sarin, 2009;Souza et al., 2014a). NH 4 NO 3 exists in a reversible equilibrium between HNO 3 and NH 3 (Tang et al., 2016).
However, potassium ions can also come from soil resuspension (Ram et al., 2010;Tiwari et al., 2016), which becomes important in PM 2.5-10 . Higher concentrations of chloride in fine particles (964 and 330 ng m -3 for Int 2.5 and Ext 2.5 , 285 respectively) were observed in campaign Int 2.5 (although the value of p was slightly above 0.05), probably due to a higher influence of biomass burning (Allen et al., 2004). On the other hand, chloride in coarse particles is mostly attributed to marine aerosols. Cl -/Na + ratios were below 1.8 in Ext 2.5 and Ext 10 and higher in Int 2.5 ; although Cl -/Na + ratios are attributed to increased sea salt contribution (Souza et al., 2014a), the higher contribution in the intensive campaign may be explained by a higher contribution of other sources of chloride in that period, such as biomass burning.
Pearson correlations were obtained for all determined species in Ext 2.5 and Ext 2.5-10 (coarse mode), including meteorological data such as temperature, relative humidity and wind speed, some gaseous species such NO x and CO, were obtained from CETESB database and were also included. NH 4 + was from moderately to strongly correlated with C 2 O 4 2-(oxalate), Cl -, NO 3 and SO 4 2in Ext 2.5 (R = 0.66, 0.62, 0.85 and 0.79, respectively) suggesting the neutralization of oxalic, hydrochloric, nitric and sulfuric acids by NH 3 (Table S2). The formation of (NH 4 ) 2 SO 4 , a non-volatile species, could represent a gas-to-particle 295 process and can account for the formation of new particles through nucleation (Mkoma et al., 2014;da Rocha et al., 2005) and can lead to CCN (cloud condensation nuclei) formation. NH 4 NO 3 and NH 4 Cl also have an important influence on Earth's acid deposition (Tang et al., 2016).
These species are often associated with marine aerosol, which is mainly in the coarse mode ( Average, maximum and minimum trace element concentrations are presented in Table 4. Mg, Al, K, Ca, Fe, Cu and Zn were the most abundant elements in all campaigns, similar to those observed by Vasconcellos et al. (2011a) for the intensive campaign in 2008. All of them had higher concentrations in Int 2.5 than in Ext 2.5 . However, crustal elements were significantly higher; Al and Ca had concentrations nearly 3 times higher in intensive campaign (p < 0.05). A similar trend was observed 310 between wintertime and summertime campaigns by Castanho and Artaxo (2001). They reported higher concentrations of soil resuspension elements during wintertime. An increase in soil resuspension is expected in drier conditions.
As observed for nss-K + , elemental K average concentration was more than twice higher in Int 2.5 than Ext 2.5 (p < 0.05). This may be explained by a higher biomass burning contribution during the intensive campaign since sugar cane burning is significantly increased in this time of the year. Cu has been attributed to vehicular emissions in São Paulo (Castanho and 315 Artaxo, 2001), because it may be present in the ethanol, which is mixed with gasoline and used in light-duty vehicles in Brazil. Cu has also been related to wear emissions of road traffic (Pio et al., 2013). This element was approximately 70 % higher in Int 2.5 than Ext 2.5 . Although there is no significant difference in vehicular emissions all year round, the meteorological conditions are more unfavorable to pollutant dispersion in winter season.
Enrichment factor (EF) is an approximation often used in order to identify the degree to which an element in an aerosol is 320 enriched or depleted regarding a specific source. EFs are calculated based on a reference metal (Al as a soil tracer in this study), considering crustal element composition (Lee, 1999). A convention often adopted is considering that when elements have EFs below 10 they have significant crustal source and are often called non-enriched elements (NEEs) and when the elements have EFs above 10 they have a higher non-crustal character and are referred as anomalously enriched elements (AEEs) (Pereira et al., 2007). Values were higher than 10 for Cr (except for Int 2.5 ), Cu, Zn, As, Se, Cd, Sn, Tl, Pb and Bi, 325 meaning that they can be attributed to anthropogenic sources as vehicular and industries emissions (Table S3) (Rao et al., 2016). Mg, Al, K, Sr and Fe were negatively correlated with relative humidity (R ≤ -0.60), suggesting strong influence of drier conditions over these species.

Carbonaceous species and mass balance
Higher concentrations of OC and EC were observed in Int 2.5 than in Ext 2.5 with average values of 10.2 µg m -3 for OC and 7.0 335 µg m -3 for EC ( Fig. 4 and Table S4). However, the difference of carbonaceous species concentrations was not considered statistically significant between the campaigns (p ~0.1). The OC/EC ratios were 1.5, 1.7, and 1.8 for Int 2.5 , Ext 2.5 , and Ext 10 , respectively. Since the ratio values were similar, as well as the absolute OC and EC concentrations were higher in intensive than extensive campaigns, this may indicate similar sources of OC and EC are contributing all year long but with higher concentrations during Int 2.5 . Ratios lower than 1 are constantly observed in roadway tunnels and are assumed to describe the OC and EC were well correlated in Ext 2.5 , with values above 0.8. This suggests that a large amount of OC is emitted by a 350 dominant primary source at this site (Aurela et al., 2011;Kumar and Attri, 2016). The studied site is strongly affected by vehicle emissions and, during the winter months biomass burning also contributes to these species (Pereira et al., 2017).
Correlations were strong between the carbonaceous species with vehicular emitted gases such as NO x and CO (R > 0.85).
OC also had good correlations with soil elements (Mg and Al) and also nss-K + (R > 0.8), suggesting association with the resuspension of road dust and also a significant biomass burning contribution.

Polycyclic aromatic hydrocarbons and derivatives
The PAH and derivatives concentrations are presented in Table 5. The total PAHs were higher in Int 2.5 than in Ext 2.5 , 23.3 ng m -3 and 18.4 ng m -3 , respectively (although not significantly different, with p > 0.05). The total PAH concentration for the Ext 10 was 24.3 ng m -3 . The lowest total PAH concentration of 2.6 ng m -3 was observed in Ext 2.5 , while the maximum of 115.3 ng m -3 was observed in Ext 10 . These levels were similar to those obtained in past studies at the same site, as 25.9 ng m -3 in 360 PM 10 samples during the intensive campaign in the winter of 2008 (Vasconcellos et al., 2011a) and 27.4 ng m -3 for PM 10 in the winter of 2003 (Vasconcellos et al., 2011b). In addition, the total PAH levels from the present study is higher than in 2013 and 2012 intensive campaigns (8.7 ng m −3 and 8.2 ng m −3 in PM 10 ) (Pereira et al., 2017). Total PAHs represented 0.23, 0.27 and 0.31 % of OC for Int 2.5 , Ext 2.5, and Ext 10 , respectively. In spite of accounting for a small fraction of organic carbon, it is important to observe that PAHs are among the pollutants of major concern due to their carcinogenic and mutagenic 365 effects.
BbF was the most abundant PAH (the BbF percentages in relation to total PAHs were 13, 12, 12 % for Int 2.5 , Ext 2.5 and Ext 10 , respectively) in all the campaigns. This compound has carcinogenic properties already reported in other studies (Ravindra et al., 2008). Its concentrations reached the values of 6.1, 6.4 and 13.3 ng m -3 in Int 2.5 , Ext 2.5 and Ext 10 . BbF was also the most abundant PAH in 2013 intensive campaign (Pereira et al., 2017). This species was also among the most 370 abundant PAHs in the study performed at Jânio Quadros tunnel, with a predominance of light-duty vehicles (Brito et al., 2013). Correlations were strong between all PAHs heavier than Flt (R > 0.8); suggesting different sources from the PAHs with lower molecular weight at this site. Most of the heavier PAHs appeared to have negative correlations with temperature; the condensation of organic compounds in the aerosol is influenced by lower temperatures (Bandowe et al., 2014).
Coronene, a PAH often used as a vehicular fuel marker (Ravindra et al., 2006) was correlated to vehicular related species as 375 Cu and Pb (R > 0.7).
BaP, the PAH most studied due to its proven carcinogenic potential, was considerably higher in Int 2.5 than in Ext 2.5 . It reached the mean values of 5.5, 7.6 and 12.5 ng m -3 in Int 2.5 , Ext 2.5 and Ext 10 , respectively. In the tunnels, its presence was associated with the higher contribution of LDV emissions (Brito et al., 2013). 2-NFlu and 2-NBP were among the nitro-PAHs with highest concentrations. 2-NFlu is a major component of diesel exhaust particles, such as the nitropyrenes, and is 380 known as a carcinogenic nitro-PAH (Draper, 1986;Fujimoto et al., 2003). 2-NFlt was moderately correlated with Flt (R = 0.4); this species is produced from reactions between Flt and NO 2 (Albinet et al., 2008). The ratios 2-NFlt/1-NPyr were close to 1; ratios lower than 5 indicate a predominance of primary emissions of nitro-PAHs (Ringuet et al., 2012). The compound 9,10-AQ was the most abundant oxy-PAH found in this study. It can be either primarily emitted or secondarily formed. A recent study showed that it can be formed from the heterogeneous reaction between NO 2 and Ant adsorbed on NaCl particles 385 (sea salt) (Chen and Zhu, 2014). A moderate correlation was found between 9,10-AQ and Ant (R = 0.54).

PAH diagnostic ratios
The PAH diagnostic ratios (Table 5) were obtained for all the campaigns, since they can point to some emission sources, such as oil products, fossil fuels, coal or biomass combustion. However, these ratios should be used with caution, due to the peculiarity of fuel compositions in Brazilian's car fleet. The values of PAH ratios can also be affected by changes of phase, 390 transport and degradation (Tobiszewski and Namieśnik, 2012).The ratio BaP/(BaP+BeP) is related to the aerosol photolysis.
Most of the local PAH emissions contain equal concentrations of BeP and BaP. However, BaP is more likely to undergo photolysis or oxidation (Oliveira et al., 2011). The average BaP/(BaP+BeP) was close to 0.4 for the three campaigns. This ratio was slightly lower than the ratio obtained in the 2013 intensive campaign, although still very close to 0.5 (Pereira et al., 2017); it is suggested that the PAHs found in the site are mostly emitted locally.

395
The Flt/(Flt+Pyr) and InP/(InP+BPe) ratios were reported to be the most conservative by Tobiszewski and Namieśnik winter, just as the ratios were lower, corroborating the results of some previous studies (Chen et al., 2016;Teixeira et al., 2013). In turn, HMW PAHs are more likely to be retained in particles due to its lower vapor pressure than LMW PAHs.
2004) and also found in a campaign performed in São Paulo (1.11) (de Martinis et al., 2002). This may be a characteristic fingerprint for local vehicular emissions.

PAHs risk assessment
BaPE is a parameter introduced to quantify the aerosol carcinogenicity related to all carcinogenic PAHs instead of BaP solely. BaPE values above 1.0 ng m -3 represent an increased cancer risk. The carcinogenic nitro-PAHs (1-NPyr, 4-NPyr and 6-NChr) were below the detection limit in most part of the extensive campaign samples, so they were not considered in the risk assessment. BaPE is calculated according to Eq. (5) 435 DBA had the largest contribution to carcinogenic potential and BaP for mutagenic potential, in studies performed in Italian urban areas BaP was the compound that most contributed to total carcinogenicity in PM, although the TEF used for DBA was lower in that cases (Cincinelli et al., 2007;Gregoris et al., 2014). LCR from exposure to atmospheric PAH was estimated by multiplying BaP-TEQ and BaP-MEQ by the unit risk (87×10 -6 (ng m -3 ) -1 ) for exposure to BaP established by WHO (de Oliveira Alves et al., 2015;WHO, 2000) (Fig. 5) and was possible to observe an increase during the intensive 440 campaign. In all campaigns, the values observed were higher than the observed in studies done in the Amazon during dry season with events of biomass burning (de Oliveira Alves et al., 2015); studies done at different seasons in other urban areas as New York and Madrid pointed carcinogenic risks within the recommended by environmental and health agencies (Jung et al., 2010;Mirante et al., 2013).

445
The highest concentrations of biomass burning tracers (levoglucosan, means 509 ng m -3 ; mannosan 45 ng m -3 ; galactosan 33 ng m -3 ) were observed in the Int 2.5 (p ~ 0.05), during the biomass burning period (Fig. 6 -Table S5). In the intensive The Lev/Man ratios are characteristic of each type of biomass. The ratios were similar to that obtained in a chamber study with sugarcane burning (Lev/Man = 10) (Hall et al., 2012), and also to that reported for the 2013 intensive campaign (Lev/Man = 12) (Pereira et al., 2017). Nss-K + /Lev ratios were 1.6, 1.4 and 1.3 for Int 2.5 , Ext 2.5 and Ext 10 , respectively. These

470
of São Paulo and the back trajectories revealed air masses crossing the West and Northwest of the state (Fig. 7a), where the fire spots were observed. In this same sampling day, local fire spots were observed, possibly landfill burning.
On 12 th of July, the air masses travelled through the Atlantic Ocean before reaching the site. In the same period, the PM 2.5 and biomass burning tracers concentrations dropped. Figure 7b shows the trajectories for 13 th of July. Some of the lowest concentrations of levoglucosan (80 and 74 ng m -3 ) and PM 2.5 (28 and 26 µg m -3 ) were observed on 12 th and 13 th of July, 475 respectively. Figure 8 shows the mass percentage of tracers in fine (PM 2.5 ) and coarse particles (PM 2.5-10 ) in the extensive campaign; their values are presented in SI (Table S6). The biomass burning tracers, levoglucosan and mannosan were present mostly in PM 2.5 mass fractions (over 75 %). In this study, 73 % of nss-K + mass was in PM 2.5 . This species may also be attributed to 480 biomass burning, although coarse potassium may be from soil dust resuspension (Souza et al., 2014a;Vasconcellos et al., 2011a).

Distribution of species in fine and coarse particles during extensive campaigns
Species related to vehicular emissions as coronene and Cu (Brito et al., 2013;Ravindra et al., 2006) were also predominantly found in PM 2.5 (73 and 61 %, respectively). More than 50 % of Fe and Ca, crustal elements, was found in PM 2.5-10 . A previous source apportionment study in Southern European cities (AIRUSE-LIFE+ project) also pointed out soil dust as a 485 significant source, accounting for 2-7 % of PM 2.5 at suburban and urban background sites and 15 % at a traffic impacted station. In the case of PM 10 , these percentages increased to 7-12 % and 19 %, respectively (Amato et al., 2016). Sulfate and ammonium were predominant in PM 2.5 (over 65% and 80%). Sulfate was also predominant in PM 2.5 in a previous study done in São Paulo between 1997 and 1998; it was attributed to the gas-to-particle conversion of vehicular SO 2 (Castanho and Artaxo, 2001). Both ions may be present as (NH 4 ) 2 SO 4 in the fine mode. Nitrate is well distributed in both phases, likely resulting from reactions of HNO 3 with soil species (Tang et al., 2016). In turn, fine mode nitrate is often present in the form of ammonium nitrate, which is a thermally unstable species (Maenhaut et al., 2008).
Other ions, such as sodium and chloride were halved in each mode. These species are related to sea salt aerosols and are more often present in the coarse mode, as observed in the study done at urban sites in Rio de Janeiro (Godoy et al., 2009).
Chloride in PM 2.5 can also be originated by biomass burning emissions (Allen et al., 2004). OC and EC, which are mainly related to vehicular emissions in São Paulo (Castanho and Artaxo, 2001), were mostly in the fine particles. OC and EC were 500 also associated with biomass burning in a recent study (Pereira et al., 2017).
BaP was found mainly in the PM 2.5 (over 80 %), which are able to be deposited in the tracheobronchial region of the human respiratory tract, representing an increased health risk (Sarigiannis et al., 2015). In turn, As, Cd and Pb, identified as elements that can cause carcinogenic health effects (Behera et al., 2015), were also found predominantly in the PM 2.5 (over 75 %). In this way, they may be indicative of higher carcinogenicity of fine over coarse particles.

Source apportionment by PMF and polar plots
Source apportionment was performed with PMF including all data. Then, the factor contributions were separated for each campaign (n = 78

510
Considering the limited number of samples, a restricted number of species had to be chosen. Elements already studied and attributed to sources in São Paulo were preferred. In some of the base model runs it was possible to observe a sea salt profile with Na + and Cl -, but after they were removed, other profiles were clearly improved. PAHs were firstly included in the model but it created a factor associated with temperature conditions, increasing in the dry season since the lower dispersion conditions in the period favor the accumulation of HMW-PAHs in suspended particles (Agudelo-Castañeda and Teixeira, 515 2014; Ravindra et al., 2006) Lev and man were set as weak due to their organic character, and NO 3 and NH 4 + due to their thermal instability. Ca and Cu also had to be set as weak in order to have a convergent base model run.
Solutions with three to eight factors were tested. The ratio of robust to theoretical parameters (Q R /Q T ) reduced between simulations when increasing the number of factors. A solution with five factors was found to have more meaningful results; Q R and Q T values were 367 (Table S7). The source profiles obtained in the PMF analysis and the contribution of each factor to PM 10 concentrations are found in Fig. 9. Constraints were applied, Cu was pulled up maximally in the vehicular factor and Lev and Man were pulled up maximally in the biomass burning factor in order to have a better separation between both factors, relative change in Q was of 0.4 %. The PMF result charts are presented in Fig. S3.
Factor 1 presented higher loadings for Mg, Ca and Fe, elements associated with soil resuspension in previous studies (da Rocha et al., 2012). The factor was also mixed with vehicular related species, such as Cu and OC, which can be attributed to 525 the resuspension of road dust by traffic. Accounting for 24.3, 12.5 and 25.7 % of Int 2.5 , Ext 2.5 and Ext 10 , respectively, it was more relevant for the PM 10 campaign. In some runs, it was possible to observe Li and Tl in this factor, but these species were not considered in the final model. This soil contribution was similar to that obtained for PM 10 in a year round inventory in the city (CETESB, 2015). High loadings for ions, such as nss-K + and NO 3 -, were also present in the factor. Gaseous HNO 3 can interact with soil particles and form coarse nitrates (Tang et al., 2016). The factor contribution appeared to increase with 530 wind speed from NW and decrease with SE winds (Fig. 10). Soil dust and vegetation sources also tended to reduce with SE winds, as observed previously by Sánchez-Ccoyllo and Andrade (2002).
Factor 2 shows high loads for Ni, Pb and Cr, which are often attributed to industrial emissions (Bourotte et al., 2011;Castanho and Artaxo, 2001). This factor had some of the lowest contributions, 10.5, 9.7 and 9.5 % for Int 2.5 , Ext 2.5 and Ext 10 and appeared to increase with SE winds, passing through nearby industrial regions (southeast of the city). The growth of industries have been limited in the last years and vehicle fleet is expected to be a main source of atmospheric pollutants in the area (Kumar et al., 2016).
Factor 3 showed high loadings for vehicular related tracers, such as Cu, Fe, OC and EC (with a higher load on EC and Cu).
Cu and Fe were found in the LDV impacted tunnel study in São Paulo and Cu is emitted from brake pads, in stop-and-go driving in the expressways (Andrade et al., 2012b;Brito et al., 2013) and also are present in ethanol after the processing of 540 copper tanks. Loadings for levoglucosan and mannosan were observed in this factor, which precluded the total separation from the biomass burning factor. On days with NW winds, both source contributions tended to increase as observed in the polar plots. This factor represented 30.9, 39.1 and 39.2 % contribution for Int 2.5 , Ext 2.5 and Ext 10 , and had a constant contribution comparing dry and wet period in the Ext 10 campaign. Vehicular source seemed to increase with winds coming from the North and Northwest, passing by the expressway, but decreased with SE winds, as observed previously (Sánchez-

545
Ccoyllo and Andrade, 2002). The polar plots profiles of VE and RD factors presented a different pattern, since the aerosol from road dust suspension has a larger aerodynamic diameter (Karanasiou et al., 2009) and tends to increase with wind speed.
Factor 4 was associated with biomass burning due to the loadings for levoglucosan, mannosan and non-sea-salt potassium, OC and EC. It is also noteworthy the loading of Cd in this factor; wood burning was pointed out as a possible source of this 550 metal in a previous study in Belgium (Maenhaut et al., 2016), but more studies are necessary in order to explain the biomass burning contribution to this species in São Paulo. This factor represented 18.3, 11.6 and 7.6 % for Int 2.5 , Ext 2.5 and Ext 10 , respectively. The contributions of this factor were higher in the intensive campaign (sugarcane burning period), but were also present in the other periods, suggesting other biomass burning sources in the city, such as waste burning and wood stoves (Kumar et al., 2016). Several fire spots were registered in São Paulo state in the intensive campaign, some of them in 555 the neighboring towns (INPE, 2014). The polar plot showed that this factor tended to increase with NW winds, passing through the inland of São Paulo state, and decrease with SE (more humid) winds from the sea. High correlations (R > 0.8) were observed between the gases CO and NOx and the primary sources factors VE and BB (Table S8). These gases are related to vehicular emissions (Alonso et al., 2010) and the correlations with the biomass burning factors may be due to the fact that it increases with the same wind direction as the vehicular factor. No correlations were found between these gases 560 and SP factor.
Factor 5 was attributed to the secondary inorganic aerosol formation processes (as seen by high mass loadings for NO 3 -, SO 4 2and NH 4 + ) and also OC (secondary organic carbon). The contributions were 15.9, 27.1 and 17.9 % for Int 2.5 , Ext 2.5 and Ext 10 , respectively. The contributions of this profile did not follow any seasonal trend (Fig. S3a). In 2014, 78 % of NO x and 43 % of SO x emissions in Greater São Paulo were attributed to the vehicle fleet (CETESB, 2015). Taking into account that in 565 São Paulo SO x and NO x concentrations are similar all year round, this could explain the lack of seasonality of this factor. The polar plot showed a centralized profile, increasing with lower wind speed, which suggests a local secondary process.
Other polar plots were obtained for individual species and are presented in Fig. 11. It is possible to see that Na + tended to increase with stronger winds coming from the sea, while Clhad a different pattern. Chloride in the marine aerosol can be depleted after atmospheric reactions with acids (Calvo et al., 2013;White, 2008). It is noteworthy that MSA was associated 570 with NW winds, This species is often associated with the decomposition of DMS, emitted by the sea (Bardouki et al., 2003).
More studies are needed in order to identify MSA sources on this site. Similarly, as for biomass burning factor, levoglucosan tended to increase with NW winds. However, it is also possible to observe local sources for this species due to its high levels, even with lower speed wind. Secondarily formed species such as NO 3 and SO 4 2had a much-centered profile and tended to increase with lower wind speed. EC, Chr and Cor seemed to be emitted by local sources, likely vehicular emissions 575 Ravindra et al., 2008). On the other hand, Flt (a light molecular mass PAH), seemed to be influenced by different air masses, suggesting different sources.
for the help with the PMF analysis.