PM_2.5 was collected at two different locations, Milan-Pascal (45°28^′42.0348° N, 9°13^′54.03° E) and Schivenoglia (45°1^′0.336° N, 11°4^′34.0032° E), both belonging to the ARPA Lombardia (the regional environmental protection authority) air quality network. The first one (hereinafter “MI”) is an urban background site, located in a small park next to a low-traffic street of the University of Milan campus. Milan is the most populated city in the Northern part of Italy and is located in the center of the Po Valley. The second site (hereinafter “SKI”) is a rural background site about 150 km far from MI, surrounded by agricultural fields and few animal husbandries. A map of Northern Italy showing the sampling sites is provided in Fig. S1 in the Supplement. At both sites, low volume samplers (SKYpost PM, TCR-Tecora, or Lifetek PMS, Megasystem, 16.67 L min⁻¹) were used to collect the PM_2.5 samples on not pre-treated quartz fiber filters (Pall TissueQuartz, Ø = 47 mm) for 24 h since 2014. On a weekly basis, the filters were collected and stored in the darkness at low temperature (about 4 °C). For this work, we chose to investigate samples collected in 2021. Therefore, a total of 132 samples for MI site and 117 samples for SKI site were used for the chemical and molecular characterization. The covered period ranges from the end of January 2021 to the end of the same year. Moreover, additional data, such as meteorological parameters (air temperature, global radiation, relative humidity) and gaseous concentration of NO_x (NO + NO₂), benzene and toluene, were provided by ARPA Lombardia.

Water-soluble inorganic compounds and levoglucosan were quantified by high-performance anion-exchange chromatography with pulsed amperometric detection by extracting 1.5 cm² quartz filter punches in ultrapure water (10 mL volume) for 20 min in an ultrasonic bath. After filtration (Nylon or PTFE Syringe Filter, pore size 0.45 µm), the resulting solution was injected in the analytical system for the determination of anions (Metrohm 930), cations, and levoglucosan (Metrohm 881). A second punch was used to quantify EC and OC through thermo-optical analysis (Sunset Laboratory Inc., Tigard, OR, USA) based on the EUSAAR2 protocol. Both analytical procedures are also described in . For the organic phase investigation, a 12 mm diameter round punch was used for OA extraction in 180 µL of a 10 % acetonitrile (ACN) solution (Merck KGaA and Optima LC/MS Grade, Thermo Fisher Scientific Inc. and Milli-Q water Reference A+). The extraction was carried out using an orbital shaker (300 rpm) for 20 min. To the extracted and filtered (non-sterile PTFE Syringe Filter, Thermo Fisher Scientific Inc.) solution of 80 µL, an internal standard (10 % of the final extracted volume) consisting of isotopically labeled benzoic acid (C6H513CO2H, 99 at. % 13C, SigmaAldrich) and caffeine (13C3C5H10N4O2, 99 at. % 13C SigmaAldrich) was added. The analytes were separated by injecting 10 µL of each sample into a UHPLC with a reverse phase column (CORTECS T3, 120 Å, 2.7 µm, 3 mm × 150 mm, Waters, C-18) system. The mobile phase, ultrapure water and ACN, was maintained at a constant flow rate of 0.4 mL min⁻¹ through the C-18 column heated at a constant temperature (40 °C). The gradient program was set as follows: 0–1 min at 1 % (v/v) ACN, 1–16 min with a linear increase to 99 % ACN, 16–16.5 min at a constant 99 % ACN, 16.5–17.5 min with a linear decrease back to 1 % ACN, and 17.5–20 min at 1 % ACN to re-equilibrate the column. A heated electrospray ion source (HESI-II Probe, Thermo Fisher Scientific Inc., operating in both positive and negative modes) was used to ionize the analytes before entering the mass spectrometer (high-resolution hybrid quadrupole-Orbitrap mass spectrometer, Q-Exactive Focus, Thermo Fisher Scientific Inc.). Ions in 75–750 m/z range were detected in both positive and negative ionization mode in full-scan MS with a resolution of 70 000 at m/z 200. Data-dependent tandem mass spectrometry (ddMS²) with a resolution of 17 500 was used at higher-energy collision energies (15, 41/30 (negative/positive mode), 50 eV). As quality assurance, the sample analysis was periodically supplemented with mixed standards to evaluate the performance of the analytical system. A previous evaluation was also performed in .

A subset of sample extracts was used to assess the light-absorbance to compare samples, which were collected at the two sites on the same days and exhibit very similar OC concentrations and contribution to PM_2.5 (within a tolerance of 2 µg m⁻³ and 5 %, respectively). For this purpose, 10 µL of sample extracts were directly injected without chromatographic separation into a diode array detector (DAD, Thermo Scientific, Vanquish, VH-D10-A) with a liquid carrier (99 % ACN, 300 µL min⁻¹). The light absorbance was measured with 20 Hz collection rate over the wavelength range of 230–650 nm. Solvent blank with the same amount of the internal standard injected in the sample extracts was used to remove the absorbance background. Finally, to account for the absence of a quantification of the OC mass extracted, we calculated the Absorbance Angstrom Exponent (AAE) for three wavelength ranges, that are 230–250, 250–400, and 400–650 nm. To this end, a linear interpolation was performed between the logarithm of absorbance and that of wavelength. The absolute value of the resulting slope represents the light-spectral behavior of the extracts.

The calculation of the normalized total signal intensity (nTSI, Eq. S2), i.e. the sum of the nSI of individual compounds in each sample, allowed us to compare the seasonal variability at each site. Both time series sites exhibit peak values during winter and summer (Fig. ). A similar trend is also visible in the OC concentration quantified in the PM_2.5 samples from both sites, although the increase during the summer period, mainly due to aliphatic CHO and CHOS compounds, is not fully reflected in the OC values. This leads to a poor correlation (R2 = 0.280 and 0.466 for MI and SKI, respectively) between nTSI and OC on an individual samples basis. Highly oxygenated compounds dominate during summer, with an average OS_c of 0.91 ± 0.06 for MI and 0.89 ± 0.04 for SKI in July. Figure  indicates that nTSI has a stronger contribution in aromatic compounds, especially belonging to CHNO and CHO families, during winter. In this regards, both DBE and Xc index show minimum values during summer (2.54 ± 0.14 and 0.37 ± 0.09, respectively in July) and maxima in winter (3.40 ± 0.34 and 0.87 ± 0.15, in February). While no clear temporal trend is observed for P-containing compounds in Milan, two periods of notable contribution from this family to the nTSI are identified at the agricultural site. The annual pattern shows that this compounds contribution increases from March to early May and from late September to the beginning of December, peaking in samples collected in April and November.

Retrieving light-absorption information via a molecular-derived proxy, we use the DBE/#C ratio to identify compounds of light-absorbing aerosols (LAAs), setting limits for this range between 0.5 (for polyenes) and 0.9 (for fullerene-like hydrocarbons) e.g.,. Using these criteria, we selected molecules from our dataset that likely contribute to BrC (Fig. ). Organic LAAs show maxima nTSI values during winter (0.73 ± 0.36 at MI and 0.53 ± 0.24 at SKI). This is in agreement with and describing that emission from wildfires, peat fires, agricultural residue burning, and residential combustion substantially contribute to BrC. While some of these sources may remain active during summertime, photobleaching and ventilation of the Po Valley are more efficient during summertime . Consistently, we observe lower values of the BrC-related molecules in summer (0.27 ± 0.11 at MI and 0.16 ± 0.04 at SKI). The higher values in nTSI observed at the urban site in the warm season indicate either more active emission sources or air-quality conditions that more efficiently promote chromophores formation. Indeed, the plots in Fig.  show that the urban site (left) exhibits a larger number of BrC-related compounds (+409) compared to the rural site (right). From a qualitative perspective, the seasonal BrC plots for the two sites (Fig. S2) confirm a different composition of LAAs. Seasonally averaged BrC compound intensities were aggregated by molecular-family and normalized to the total BrC intensity (Fig. S3), revealing that the relative contributions of these families differ between the two sites across seasons. During them (summer and winter) the urban site exhibits higher contributions due to the CHN family compared to the agricultural site (8.8 % vs. 3.7 % at MI and SKI, respectively, in summer; 17.1 % vs. 7.9 % in winter), whereas CHO family contributes to the higher nTSI values (Fig. ) in respect to SKI during summer only. In contrast, CHNO species contribute more at the agricultural site in both seasons (23.9 % vs. 33.9 % in summer and 46.2 % vs. 51.5 % in winter, respectively). Contributions from the other families are marginal, accounting for less than 2 % at both sites.. Similarly to the temporal pattern shown in Fig. , at both sites the contribution of wintertime BrC is mainly due to N-containing compounds. At the same time, the nSI-weighted average shows that in the coldest months BrC has the lowest O/C ratio, suggesting a low oxidation regime affecting BrC molecular composition (Fig. S4). On the contrary, the lower summertime peak in BrC nTSI is mainly ascribable to highly oxidized compounds (O/C > 0.5) with a probable secondary origin .

In addition to the seasonality of elemental-composition groups and the BrC-related molecules, we focus on few individual tracers to confirm our hypothesis concerning their seasonal patterns. For instance, the later-generation product of monoterpene oxidation, that is 3-methyl-1,2,3-butane-tricarboxylic acid (MBTCA), peaks in summer. This is consistent with the higher monoterpene emission and stronger oxidative conditions from May to September. On the opposite, the CHNO compound nitrocatechol, a well-known combustion product from BB , shows the opposite trend. The time series of these compounds, whose identification was allowed by the comparison with analytical standards (identification level 1, L1, based on ), are shown in Fig. S5.

Based on the blank-corrected annual average SI, we ranked the compounds with the highest values for the most representative molecular families (that are CHO and CHNO), which alone represent 60 % of the nTSI. Within the CHO(-) family, the formulae C4H6O5, C8H12O6, and C6H8O7 exhibit the highest average intensities. The first two compounds show a distinct seasonal trend, with intensities peaking in spring and reaching a maximum during the summer, followed by a decline during autumn. This pattern suggests that the conditions favorable for the emission or formation of these compounds are most prominent in summer. Together with the aforementioned MBTCA (L1, C8H12O6), the compound with the formula C4H6O5 and its pronounced increase during summer suggest a likely monoterpene-derived biogenic origin (L4, as suggested by , , and the literature within). In contrast, the third-highest intensity compound, C6H8O7, appears to be an anthropogenic SOA tracer (ASOA) derived from the oxidation of benzene (L4, ). For the CHNO(-) molecular family the highest signals occur during wintertime. At both sites, nitrocatechol (L1, C6H5NO4) ranks highest. Additionally, at the urban site, two isomers of C7H7NO4 were detected, likely representing methyl-nitrocatechol and nitroguaiacol (L4, ). Finally, SKI samples reveal a significant signal for C8H6N2O2, which can be attributed to nitroindole (L4), known as a BrC-related compound from the oxidation of indole in presence of NO2 . The presence of this gaseous amine in the atmosphere has been attributed to many sources, such as vegetation, biomass burning, industrial activities and animal husbandry (literature within ). We observe a similar annual pattern for nitroindole at urban (MI) and rural-agricultural (SKI) sites. However, SKI exhibits higher signal intensities, indicating a closer proximity to the precursor sources, i.e. animal husbandry. Since the DBE/#C of nitroindole is 0.875, it is also highlighted in Fig.  as the blue circle with coordinates (7; 8). In positive ionization mode, the highest intensities for CHO(+) compounds are measured in MI as C11H22O5 and C12H26O5, attributed to ASOA from alkane photooxidation , and more specifically to n-dodecane SOA (L4, ). The third highest signal at MI corresponds to the highest one at SKI, i.e. C10H22O4, although no correspondence was found in the literature. At SKI site, C8H16O2 and C20H26O3, the second and third highest signals, are attributed to BBOA by and . For the nitrogen-containing compounds (CHNO) detected in positive mode, the formulae corresponding to the m/z values with the highest signals,are C9H11NO2, C12H17NO and C6H15NO for MI site and C9H18N2O, C12H23NO and C12H25NO for SKI, tentatively assigned to 1-[[(2R)-oxolan-2-yl]methyl]piperazine, a plant protection product (L4, ), a compound found in agricultural residue burning (L4, ) or to n-octyl pyrrolidone (L4, ) also used as co-solvent in crop protection formulations, and N,N-dimethyl capramide, an additive in pesticide formulations (L4, ). As these last mentioned compounds are anthropogenic chemicals, an unambiguous molecular identification can be achieved via target analysis.

In the following, we describe the hierarchical cluster analysis of the complete dataset. We conducted two independent cluster analyses for the two sites to ensure that the differences between the sites are maximized, enabling an unbiased assignment of a compound to a cluster regardless of its temporal behavior at the other site. It is important to clarify that the presence of a formula within a cluster does not necessarily indicate that the corresponding compound is emitted by the source after which the cluster is named. Rather, it may suggest a shared emission driver or reactive pathway that causes the compounds in the same cluster to exhibit a similar temporal pattern. However, many anthropogenic sources are known to be active year-round, with their influence often varying over time. In addition to the seasonality of emission strength, the meteorological variability can drive different reaction pathways. An increased oxidation of anthropogenic VOCs during the summer months might consequently lead to the assignment to a biogenic cluster. Even more difficult to assign to an anthropogenic cluster are SOA compounds that originate from the oxidation of volatile chemical products (VCP), such as limonene. On the other hand, such products may experience an increase in their concentration during wintertime due to a reduced mixing layer height. As such, the following discussion does not intend to definitively attribute a source to each detected compound, but rather to provide insights into their temporal trends and which other compounds show similar behavior over the course of the year and assign them to a representative molecular or sector-specific cluster. Furthermore, the use of highly resolved chromatographic and spectrometric techniques yields a large number of isomers, which cannot be precisely identified in the absence of specific reference standards of characteristic fragmentation spectra. As a result, comparisons with data from literature may also be subject to this inherent uncertainty. At each site we examined the compounds that were detected in more than 10 % of the analyzed samples. This leads to investigate via hierarchical cluster analysis (HCA) 6487 and 5499 parameters for MI and SKI, respectively. We also used external data from meteorological factors (such as global radiation, daily temperature, and relative humidity), gaseous species concentrations (NO_x, O3, benzene and toluene), ion chromatographic and thermo-optical transmittance data (i.e., EC and OC) to improve our interpretation of the results. The HCA revealed seven main clusters for the urban site (Fig. S6) and five for the rural site (Fig. S7). According to the temporal patterns, three clusters at both sites showed a clear increase in intensity during the colder season. The higher signals observed during winter may arise from multiple contributing factors, acting either independently or in combination. Seasonal variations in emissions may lead to the presence of sources that are active or more intense in winter than in other periods of the year. In addition, meteorological conditions typical of winter, such as reduced atmospheric mixing and dispersion, can enhance aerosol concentrations. Partitioning processes may also play a role, with gas–particle equilibrium shifting toward the particulate phase under colder conditions. Furthermore, the absence or low abundance of certain compounds in summer does not necessarily imply a lack of emissions; rather, enhanced photochemical degradation may lead to their rapid transformation into other species. As a result, compounds that are prevalent in winter may be replaced in summer by secondary products that are only present at low levels during the colder season. In either case, this behavior suggests that the compounds in these clusters have a likely anthropogenic origin. Additionally, at MI, two clusters show an increase during the warmer season, whereas only one cluster exhibits this behavior at the rural site. The remaining clusters do not show a clear temporal pattern, which could be due to a sporadic rise or activation of the sources. The contribution of the clusters, as monthly average, is shown in Fig. . In the Supplement, we show the fingerprints of each cluster using the retention time versus the molecular mass of each compound, the Van Krevelen diagram, and the Kroll plot to infer the cluster's chemical signature. Finally, the molecular formulae with tentative identifications and cluster attribution are listed in Additional Data (Table A1).

The use of the HCA technique allowed us to characterize cluster of compounds with similar temporal pattern. Between the two sampling sites, similarities and differences in molecular composition of the interpreted clusters are discussed in Sect. . Their contribution to the annual nTSI is also diverse: winter-peaking clusters in Milan, linked to traffic, biomass burning, and a mix of both, account for about 16.1 %, 10.9 % and 8.1 % to annual nTSI, respectively. At Schivenoglia, mixed traffic and biomass burning sources account for 14.9 %, the cluster linked to biomass burning with ammonium nitrate formation account for 20.2 %, and the third cluster containing aliphatic amines possibly from agricultural reside burning, accounts for 2.6 % to the annual nTSI. The compounds attributed to biogenic-related sources, peaking during summertime, have a higher impact at MI (31.4 %) rather than at SKI (26.4 %): in addition, a fraction of MI-Summer B/ASOA cluster (11.0 %) might also be due to biogenic sources or stem from VCP (e.g. limonene) oxidation. At the urban site, anthropogenic sources have an additional 14.1 % contribution to the annual nTSI due to MI-other anthr-OA cluster, and 2.1 % due to MI-Aliphatic OrgNO₃ one, even though no clear attribution is performed. At the rural site, anthropogenic activities linked to agricultural practices are clearly observed and quantified to 27.8 %. Notably, some clusters of MI and SKI show a similar molecular composition. For example, 61 % of compounds clustered in MI-BSOA are in common with SKI-BSOA, and in turn SKI-BSOA shares 56 % of its compounds with MI-BSOA and 12 % with MI-Summer B/ASOA. Both biogenic clusters show a number of compounds that are not classified at the other site: at MI 350 compounds are not detected (or detected in less than 10 % of the analyzed samples) in SKI, whereas a higher amount (436) are observed at the rural site but not at the urban one. This provides a first insight of the molecular variability between the two environments. For SKI, the cluster that contains the largest number of these site-specific compounds is SKI-Agricultural activities (645, i.e. 46 % of compounds falling into it). On the contrary, every cluster found in MI contains a comparable amount of site-specific compounds. The chord chart (adapted from ) in Fig. , graphically assists the assessment of the site differences, where the width of the flow-lines is proportional to the number of compounds that are common between the two connected clusters. The chart suggests that combustion sources have a different profile at the rural compared to the urban site. MI-BBOA, MI-TrOA and MI-Other combustion share 45 % (n=412), 19 % (n=175), and 33 % (n=306) compounds only with SKI-BBOA/TrOA, and 11 %, 30 % and 15 % with SKI-Nitro Aromatics, respectively. This leads to a percentage of site-specific combustion-related compounds from this MI clusters of 36 %, 40 %, and 30 % respectively.

We found a total of 2513 and 1525 compounds that are site-specific for MI and SKI, respectively, confirming the higher molecular-diversity at the urban site. The site-specific fingerprints, shown in Figs. S20 and S21, highlight that the urban site is richer in nonpolar compounds with high molecular mass, and in CHO-containing aromatics. Among aliphatics, N-containing compounds are the most prominent compounds present at the urban site. Figure  illustrates the contribution to the monthly nTSI of each cluster at both sites. As expected, during wintertime the nTSI explained by the site-specific compounds (in darker colors) is lower than in the other seasons. The reason is explainable by the lack of both weak oxidant conditions, leading to a longer lifetime of organic molecules, and meteorological events able to abate PM concentrations: as a result, aerosols disperse, mix and age in the lower layers of the troposphere in a vast area of the Po basin. Limiting the discussion on the site-specific features only, during winter in both sampling sites the anthropogenic clusters explain between 10 % and 15 % of the nTSI, whereas site-specific BSOA compounds are almost negligible. On the opposite, from May to September this cluster provides the highest molecular variability at MI in respect of SKI. The rural site, in fact, exhibits site-specific compound profiles predominantly originating from agricultural activities throughout the year, with only modest contributions from the SKI-Nitro-Aromatics cluster in winter and from source SKI-BSOA in summer. Between May and September, both sampling sites exhibit a significant molecular variability, which rises from 15 % to 25 %. It is worth noting that, at both sites, the lowest concentrations of OC were measured in May (2.0 ± 0.6 and 1.6 ± 0.3 µg m⁻³, Fig. ). At the same time, the highest plant protection product intensities at the SKI site were detected in May, suggesting that, despite theoretically indicating a reduced health impact that month, the aerosols could have a non-negligible toxicological effect during that period at SKI. Simultaneously, the MI-Summer B/ASOA cluster, predominantly composed of compounds detected in positive mode, with a significant fraction unknown at the SKI site, shows its highest contribution to nTSI (7 % accounting for site-specifice compounds only) once again in May. These results highlight the need to assess the health effects of aerosols not only based on the mass quantification metrics but also, and perhaps most importantly, based on chemical composition and evaluate health effects of emissions from agricultural practices.

Finally, we select sample pairs with similar OC concentrations (within a tolerance of 2.0 µg m⁻³) and contribution to PM_2.5 concentration (within a tolerance of 5 %) collected on the same day at both sites (hereinafter referred as “similar samples”). Considering the full year, 116 sample pairs meet this criterion, although they are unevenly distributed over time (Fig. S22). In particular, the number of pairs is the lowest in June (#5), September (#1), and December (#1). When restricting the analysis to the subset of filters used in this study, the number of valid pairs decreases to 54, and no pairs meet the above criterion in June and December. Consequently, these months are not represented in the figures showing similar sample pairs (Figs.  and ). These conditions lead us to select 54 sample pairs across all months but June and December 2021. The differences of the nSI highlight the dissimilarities between sample pairs of similar OC concentrations. Once again, we take advantage from the HCA performed (Sect. ) and classify the enrichment in compounds based on their source attribution. The bar chart in Fig.  shows the compound membership clusters with residuals, averaged on a monthly scale, favorable to MI (upper semi-axis) or SKI (lower semi-axis) for the selected samples as contribution to nTSI (see Sect. S1.2, “Residuals contribution calculation”). Both semi-axes are plotted as absolute values to display only the magnitude of the residuals. At MI site, the molecular variability is mostly due to compounds detected at both sites (lighter colors) for the majority of the sources with a seasonal pattern. The residuals of common compounds contribute about 38 ± 7 % to nTSI (reported as an annual average), whereas the residuals of site-specific compounds contribute 18 ± 5 %. At SKI residual contributions of common compounds account to 26 ± 6 %, slightly higher than for site-specific ones (21 ± 11 %). In particular, Fig.  shows that the agricultural activities drive both the site-specific features and the high residual contributions. Moreover, the highest differences are observed in May, when the OC concentrations are the lowest. This reinforces the assertion that the sources at the two sites have distinct emission profiles and formation pathways.