Introduction

Particulate nitrogenous organics (ONs) comprise a large fraction of urban atmospheric organic aerosols1. They exert a great role in particle hygroscopicity2, particle light-absorption3, and atmospheric nitrogen cycle4. Atmosphere particulate ONs originate from a wide range of primary sources5,6,7,8 (i.e., biomass burnings, vehicle emissions, industrial emissions, coal combustion, and soil dust emission) and secondary formations9,10 (i.e., photooxidation and heterogeneous reactions), resulting in complex mixtures. Previous studies have reported that ONs with low oxygen content mostly contain amine groups11, while those with high oxygen content may contain nitro (-NO2) or nitrooxy (-ONO2) group12,13. Some ONs containing sulfur, generally recognized as nitrooxy-organosulfates14, are proposed to be formed via the heterogeneous oxidation under high NOx and SO2 levels15,16,17. However, the characteristics and evolution of ONs vary across cities and seasons18,19,20,21, indicating the significant differences in sources and atmospheric oxidations. Digging into the molecular-level chemical components of ONs can provide useful information on their primary sources, secondary formations, and atmospheric impacts.

The explorations of molecular-level information on ONs have been facilitated with the employment of ultrahigh-resolution mass spectrometers (UHRMS), such as Orbitrap and Fourier-transform ion cyclotron resonance mass spectrometry22,23. As hundreds of ONs were usually identified in the ambient atmosphere, various parameters have been proposed as representatives to study the bulk characteristics of ONs, such as double bond equivalence (DBE) and oxygen to carbon element number ratios (O/C)19,20,21. However, such parameters can only represent the bulk characteristic of ONs, they cannot distinguish isomers and functional groups, and thus provide very limited information for mechanistic studies. Alternatively, tandem mass analysis can provide the information of functional group information. Some studies used high performance liquid chromatography coupled with tandem mass spectrometry (HPLC-MS/MS) to differentiate isomers and determine their structures24,25. HPLC allows the isomers to be separated and enter the MS at different times. Then MS/MS uses a certain amount of energy to break up the molecules, and the fragments are used to distinguish the structural composition26,27,28,29. Previously, it has been used to characterize the functional groups of atmospheric organic aerosol21,30.

Functional groups can provide useful information on the origins and evolution of compounds. For example, nitroaromatic groups can be emitted directly through diesel exhaust, wood combustion, and consumer product solvent emissions25,31. They can also be formed chemically, e.g., by nighttime aqueous-phase nitrification of catechols with HNO232 or gas-phase oxidation of phenolic compounds by OH•/NO3• followed by further nitration with NO233. These compounds can undergo further reactions in the atmosphere. Photolysis and radical oxidation are the main modes of nitrophenol degradation, which may form carboxylic acids and other oxygen-containing compounds34,35,36,37. Organonitrate groups are typical secondary products, they are formed by the oxidation of volatile organic compounds (VOCs) by NO3• during nighttime or by reaction with OH•/O3 and subsequent NO during daytime38,39. Almost all secondary ONs have oxygen-containing functional groups such as alcohol, carbonyl, carboxylic acid, ether, and ester40. Different nitrogen-containing groups may indicate different routes of secondary aerosol formation30,41,42.

The megacity of Wuhan is characterized by intense primary emissions, highly-humid air conditions, and frequent transport influences. Although the air quality in Wuhan has improved a lot during last 20 years43,44, severe haze pollution still happens occasionally45,46, and the mechanisms are not entirely clear. Firstly, as the most populous city, an important industrial base, and a comprehensive transportation hub in central China47,48, Wuhan holds many heavy industries including steel, glass, cement, and chemical industry, and has a large number of motor vehicles, boats, and ships, leading to a vast amount of primary pollutant emissions49. Intensive agricultural activities in the Jianghan Plain (especially rice and rapeseed cultivation) result in a large amount of straw residue, and open burning is the main treatment method50,51. Hubei Province has achieved some success in strictly controlling straw burning through satellite monitoring and administrative penalties, but due to the lack of alternative means of utilization (e.g., high cost of straw recycling, immature market for organic fertilizers), burning is still the main method of disposal52,53. Thus, Wuhan is also influenced significantly by the biomass burning. Secondly, Wuhan is characterized by a humid subtropical monsoon climate and densely dotted with rivers and lakes, leading to highly humid air conditions with yearly-average RH > 70% and facilitating the secondary heterogeneous or aqueous reactions. Thirdly, Wuhan and its surrounding cities and provinces are surrounded by mountains to the east, west, and south, resulting in weak horizontal atmospheric flow and slow dispersion of pollutants54. Due to its terrain, Wuhan serves as a conduit for the intrusion of cold northern air and moist eastern air, facilitating the transport of pollutants to this area. Previous studies on PM2.5 in Wuhan have been extensive, but the molecular-level particulate organic compositions in Wuhan have not been reported, especially during the haze evolution.

In this study, for a 12-day long-lasting high-humid haze pollution process in Wuhan, we sampled PM2.5 and analyzed molecular-level organic aerosols using HPLC-Orbitrap MS/MS. Four samples were collected each day. This provided a more-detailed time series of compound variations during the haze episode. We found that ONs constituted a large proportion of organic aerosol in Wuhan. The functional groups of ONs were analyzed. Combining the backward trajectory analysis and chemical tracers, we revealed that biomass burning and its evolution contributed greatly to the haze pollution in Wuhan.

Results and discussion

Molecular-level organic composition and functional groups in particles

Using HPLC-Orbitrap MS/MS, we identified approximately 2200 and 2100 compounds (isomers counted as distinct compounds) in ESI+ and ESI– modes, respectively, with total signal intensities of 2.85E9 and 3.04E8. These compounds were selected due to their high signal intensity, high signal to noise ratio, as well as good peak shape. It is noteworthy that, as methanol was chosen as the extraction solvent in this study, only methanol-soluble organic compounds were investigated. Comparative experiments using three commonly-used solvents, methanol, acetonitrile, and acetonitrile/water (8:2, v/v), revealed that methanol extracted the largest amount of organic species (as detailed in Methods section, Table S1). The component distributions extracted by methanol and acetonitrile were similar, while those extracted by acetonitrile/water (8:2, v/v) were quite different (Fig. S1). For instance, methanol and acetonitrile showed higher selectivity toward CHON species in ESI+ mode and lower sensitivity to CHOS/CHONS species in ESI- mode compared to acetonitrile/water. Detailed comparison can be found in Method Section and the SI.

The compounds were categorized according to their elemental composition, as illustrated in different colors in the mass defect plots (Fig. 1a, d). Clear homologous sequences can be observed for different categories. In ESI+ mode, the species with the highest peak area were CHON (72.8%), followed by CHO (11.0%), CHN (10.2%), CHONS (4.6%), and others (1.4%, including CHOS, CH, CHS, CHNS) (Fig. 1b). In ESI- mode, CHON also represented the highest peak area (50.3%), followed by CHO (27.1%), CHOS (22.6%), and CHONS (2.2%) (Fig. 1e). In summary, nitrogenous organics (ONs) exhibited dominant proportions in both ESI+ (~87.6%) and ESI- mode (~52.5%).

Fig. 1: Species and functional group composition of PM2.5 in Wuhan, measured by HPLC-Orbitrap MS/MS.
Fig. 1: Species and functional group composition of PM2.5 in Wuhan, measured by HPLC-Orbitrap MS/MS.
Full size image

a, d Mass defect plot of the detected compounds. The diameter of the point is proportional to the peak area. b, e Peak area fractions of different species and homologous sequences, represented by the inner and outer circle, respectively. c, f Functional group peak area for different categories. ONs + /ONs- represents the nitrogenous organics detected in the positive/negative electrospray ionization mode and the peak area mentioned above refers to the average peak area of all collected PM2.5 samples.

In the ESI+ mode, ONs were predominantly characterized by homologous sequences with a lower oxygen content (≤2 oxygen atoms) and lower degree of unsaturation (averagely ~3.9), mostly containing amine/amide groups. For CHON, CnH2n+1ON was the most abundant (accounting for 19.1% in total signal intensity), followed by CnH2n-1ON (15.9%), CnH2n+3O2N (11.5%), and CnH2n+3ON (3.3%). For CHN, CnH2n-13N was the most abundant (18.4%), followed by CnH2n+3N (14.8%), and CnH2n-2N2 (14.1%). For CHONS, the most abundant homologous sequence was CnH2nO2N2S2 (1.3%). In terms of functional groups, ONs in ESI+ mode mainly contained amine (75.5%), amide (16.0%), azole (2.8%), pyridine (2.4%), N-N (1.5%), nitrile (1.4%), isocyanate (0.3%), and N = N (0.2%). Most ONs in ESI+ mode (65.2%) contain no ring, 20.8% of ONs contain aromatic ring, and 14.1% contain non-aromatic ring.

In the ESI- mode, ONs were predominantly represented by homologous sequences with higher oxygen content (≥3 oxygen atoms) and higher degree of unsaturation (averagely ~4.7), mostly containing nitroaromatic group. For CHON, CnH2n-7O3N (18.7%) was the most abundant, followed by CnH2n-7O4N (8.7%), CnH2nO3N3 (3.0%), CnH2n-1O3N (2.4%), CnH2n-8O5N2 (2.2%), and CnH2n-9O5N (2.1%). For CHONS, the most abundant homologous sequence was CnH2n-3O7NS (2.0%). In terms of functional groups, ONs in ESI- mode mainly contained nitroaromatic (72.0%), followed by amine (15.6%), amide (9.1%), azole (1.4%), pyridine (1.3%), and isocyanate (0.7%). Most ONs in ESI- mode (75.9%) contain aromatic ring, 13.3% of ONs contain non-aromatic ring, and 10.8% contain no ring.

To sum up, although ESI+ and ESI- mode could detect similar categories, their functional groups are entirely different, resulting in different oxygen contents and degree of unsaturation. ONs in ESI+ mode were dominated by aliphatic amines and amides, while those in ESI- mode were dominated by nitroaromatics. Besides ONs, CHO compounds represented the second-largest proportion in both modes. They were mainly ethers and alcohols in ESI+ mode, while mainly carboxylic acids (of which multi-carboxylic acids account for 54.0%) in ESI- mode. Sulfur-containing organics in ESI- mode were dominated by organosulfates, while those in ESI+ mode were mainly sulfides.

Compared with other urban cites in China, ON fraction, especially CHON fraction, in Wuhan was at a relatively high level. For instance, the CHON fraction in Wangdu55, Shanghai19,56, Changchun, Guangzhou56, and Beijing21, varied between 9.0–46.5% in ESI+ mode and 6.0–55.0% in ESI- mode. In contrast, CHON fraction in Wuhan reached 72.8% in ESI+ mode and 50.3% in ESI- mode, respectively. Seasonal variations in previous studies showed that CHON fraction typically increased during wintertime19,20, but even considering this, the CHON fraction in Wuhan was still relatively higher. A possible reason was the high NO2 concentration in Wuhan, which was averagely 71 μg/m3 during the sampling period. We compared the ON fraction (including CHON, CHONS, and CHN) with NO2 concentration in different studies which used methanol or acetonitrile as solvent (Fig. 2). We found that the ON fraction generally showed an increasing trend as the NO2 concentration increased. And it is notably that even considering the higher NO2 concentration, the fraction of CHON in ESI+ mode in Wuhan was still significantly higher. The high ON fraction in Wuhan may also be attributed to the sources and will be explored in the following analysis.

Fig. 2: ON fraction with NO2 concentration in different cities.
Fig. 2: ON fraction with NO2 concentration in different cities.
Full size image

The cities include Shanghai19,56, Wangdu55 and Wuhan(this article). “*” indicates methanol as the extraction solvent while others used acetonitrile.

The functional group composition in Wuhan shared some similarities with other cities, while exhibiting its distinct characteristics. The main functional groups in previous studies included alcohol (15–49%), carboxylic acid (19–59%), ether (10–23%), etc30,40,57. The fraction of alcohol (28%), carboxylic acid (21%), and ether (11%) in Wuhan were within the reported ranges. But there were some differences. We measured higher levels of amine (32%) than the previously reported 11–28%30,40,58,59. In addition, organonitrates were observed at Atlanta (1–11%)40, 5–12% over the southeastern United State57, and 3–8% in a previous study in Wuhan30. However, we hardly detected organonitrates (the appearance time was less than 10, and thus not further analyzed) in this study, which may result from the hydrolysis of organonitrate during the high humidity in this sampling period or the degradation on the filters during sampling60,61.

Source apportionment of particulate organics

To explore the sources of organic pollutants, we applied K-Means clustering to classify the identified organics into 3 classes in ESI+ and ESI- modes respectively, based on their similarities in timeseries (Fig. 3a, b). Additionally, to explore the meteorological effects for the haze formation, we divided the haze event into 4 periods based on analysis of backward trajectories, fluctuations in temperature and humidity, and changes in wind speed and direction (Fig. 3c, d). We also compared the classified sources with SOR (ratio of particulate sulfate to the sum of particulate sulfate and gaseous SO2, calculated as Eq. S1) and NOR (ratio of particulate nitrate to the sum of particulate nitrate and gaseous NO2, calculated as Eq. S2) to explore the role of secondary formations (Fig. 3e). From the combined analysis, we preliminarily classified Class 1, Class 2, and Class 3 as primary organic aerosol-like (POA-like) class, biomass burning organic aerosol-like (BBOA-like) class, and secondary organic aerosol-like (SOA-like) class, respectively.

Fig. 3: K-means clustering with compounds time series, back trajectory with fire points and parameter variation in haze event.
Fig. 3: K-means clustering with compounds time series, back trajectory with fire points and parameter variation in haze event.
Full size image

a, b Average time series of each class with standard deviation error area and representative compounds related to each class in two modes. POA-like, BBOA-like, and SOA-like refer to primary organic aerosol-like, biomass burning organic aerosol-like, and secondary organic aerosol-like, respectively. c Back trajectory and fire spot map with corresponding clustered air masses. d Variation of temperature, RH, wind speed, and wind direction. e Variation of Sulfur oxidation ratio (SOR), nitrogen oxidation ratio (NOR), and PM2.5 concentration.

Class 1 mainly originated from high local emissions with influences from long-range transported dust. As shown in Fig. 3a, b, the time series of Class 1+ and Class 1- both exhibited relatively high concentrations at the beginning, followed by decreases of ~56.1% and ~68.4%, respectively, until the end of the study period. Class 1+ and Class 1- showed significant positive correlations with NO2 and SO2 (r = 0.49–0.78, p < 0.05, Figs. S2-S3), which were typical primary gases from combustion emissions62,63,64. SOR and NOR remained below 0.2 during Periods 1–2, suggesting particle dominance by primary emissions (Fig. 3e). Class 1+ and Class 1- both peaked during Period 1, accompanied by air masses showing long-range transport from the high-attitude northwest. The strong correlation between Class 1+ and Ca2+ (r = 0.67, p < 0.05, Fig. S3) suggested that the northwestern air masses may contain dust.

Class 2 mainly originated from primary biomass burning in southeastern Hubei and northern Jiangxi. Class 2+ and Class 2- both exhibited a sharp peak between Jan.15th and Jan. 16th, and remained low at other periods. This trend showed a significant positive correlation with Na+ (r = 0.33-0.54, p < 0.05, Fig. S3), which could be released from wood and straw combustion65 and originated from similar sources with commonly-used biomass burning tracer K+ in inland cities66,67. Class 2+ and Class 2- both reached the highest during Period 2, with dominant air trajectories of short-range transport from the south and southeast. The air masses started at lower altitudes, passing through the northern Jiangxi and the southeastern part of Hubei, where high densities of fire hotspots were observed.

Class 3 mainly represented the secondary transformations occurring in highly-humid environment. In contrast to the high-to-low trend observed in the first two classes, Class 3+ and Class 3- exhibited relatively low concentrations initially, followed by a gradual increase and a significant rise on Jan.18th. They showed strong positive correlations with inorganic salts including SO42- (r = 0.34–0.57, p < 0.05, Figs. S2–S3) and NO3- (r = 0.41, p < 0.05 with Class 3-, Figs. S2–S3) and, which were mainly formed via secondary transformation from NO2, and SO2 respectively68,69. Class 3 peaked during the highly humid Period 3, where the air masses were mainly from low-altitude local origins. RH reached ~86.8%, much higher than Period 1 (~46.8%) and Period 2 ( ~ 66.8%). Simultaneously, PM2.5, SOR, and NOR continued to increase, while primary pollutants (e.g., NO2 and SO2) decreased. During Period 4, under the influence of cold air, Wuhan experienced heavy snowfall process, and the PM2.5 pollution was cleared.

Composition of particulate organics from different sources

The ON composition significantly differed among the three classes. In ESI+ mode (Fig. 4a), CHON compounds were the most abundant in the SOA-like Class 3+ (83.7%), especially for CHON>275Da, which accounted for 70.3% in Class 3 + , far exceeding the fractions of 1.8% and 18.4% in the other two classes. The CHON>275Da appearing in Class 3+ were mainly amines with oxygen-containing groups, including alcohol, ether, carboxylic acid and ketone (Fig. 4c). For comparison, CHN compounds were the most abundant in BBOA-like Class 2 + . Especially for CHN<180Da (mainly amines), which accounted for 25.5% in Class 2 + , but their signals nearly disappeared in the secondary-like Class 3+, with only 0.7% contribution. CHN<180Da were mainly amines with no oxygen-containing groups. CHONS compounds were the most abundant in Class 1+ (9.9%), while lower in Class 2+ (2.2%) and Class 3+ (0.2%), indicating their origin from primary sources.

Fig. 4: Chemical compositional and functional group distribution.
Fig. 4: Chemical compositional and functional group distribution.
Full size image

a, b Mass distribution and species intensity fraction of each class. The chemical formulas of the most abundant substances are labeled in the diagram. CHN<180 Da refers to CHN with molecular weight less than 180 Da, while CHON>275 Da refers to CHON with molecular weight greater than 275 Da. c, d Functional group distribution of each class and fraction of three classes in ONs. ONs + /ONs- represents the nitrogenous organics detected in the positive/negative mode. POA-like, BBOA-like, and SOA-like refer to primary organic aerosol-like, biomass burning organic aerosol-like, and secondary organic aerosol-like, respectively.

In ESI- mode (Fig. 4b), CHON compounds were the most abundant in the BBOA-like Class 2- (71.8%), far exceeding the fractions of 21.7% and 9.7% in the other two classes. The CHON compounds in Class 2- mainly consisted of aromatic ring, phenol, and nitro-aromatic compounds (Fig. 4d), indicating the dominance of nitrophenols. For comparison, CHO compounds accounted for 64.9% of SOA-like Class 3-, significantly higher than the other two classes. Specifically, CHO<150Da were observed almost exclusively in SOA-like Class 3-. In sight of the functional groups, multi-carboxylic acid groups were dominantly distributed in SOA-like Class 3-. These observations may be associated with secondary formation processes, which will be further investigated in the following sections.

Base on the above analysis, we proposed that the high ONs in Wuhan might be related to (1) high biomass burning contribution. As 78.2% of ONs (mostly nitrophenols) in ESI- mode and 21.2% of ONs (mostly amines) in ESI+ mode are associated with biomass burning (Fig. 4c, d), Since backward trajectory analysis indicates both inter-provincial and intra-provincial biomass burning influences, regional joint prevention and control is crucial for mitigating severe haze. (2) high NOx concentration and high humidity in Wuhan, enhancing N-containing aqueous/heterogeneous reactions. As 71.2% ONs in ESI+ mode and 18.7% ONs in ESI- mode are associated with secondary evolutions under high RH. (Fig. 4c, d). Given the large contribution of secondary ONs, emission controls should not only target primary sources (e.g., biomass burning) but also focus on precursors (e.g., amines, nitroaromatics) that drive SOA formation.

From the above analysis for Fig.4, we identified a series of high-intensity representative compounds which can potentially act as sources markers (Table 1). (1) CHN<180Da compounds in ESI+ mode (No. 1-17) may serve as tracers for biomass burning. They are mainly composed of C6-C12 amines. Previous studies have reported that biomass burning is an important source of amines in Yangtze River Delta (Shanghai and Nanjing, China), Pearl River Delta (Guangzhou)70, northwest Atlantic Ocean71 and southeastern US72. (2) CHON>275Da compounds in ESI+ mode (No. 18–20) may serve as tracers for heterogeneous/aqueous reactions. They are mainly composed of C17-C19 amines containing alcohol or carboxylic acid groups. (3) CHON compounds containing nitroaromatic groups in ESI- mode (No. 21-36) may serve as tracers for biomass burning. They are mainly composed of C6-C9 nitrophenols. Many of them are established tracers of biomass burning in previous studies73,74,75,76,77,78, such as C6H5NO4 and C7H7NO4. (4) Dicarboxylic acids in ESI- mode, including CHO compounds (No. 37–43) and CHON compounds (No. 44–46) may serve as tracers for heterogeneous/aqueous reactions. Overall, some of the above tracers have been previously reported while others are not. The list needs to be verified under broader atmospheric environments, with the help of MS/MS analysis to identify the functional groups.

Table 1 Summary of molecular formulas potentially act as markers for biomass burning (Class 2+ and Class 2-) and heterogeneous/aqueous reactions (Class 3+ and Class 3–)
Full size table

Evolution of the organic components under the highly-humid atmosphere

As previously discussed, abundant ONs from biomass burning were identified in Period 2. During the transition from Period 2 to Period 3, under an average ambient RH of ~87%, the biomass burning-related ONs diminished while the secondary ON concentration increased. In Period 3, aerosol liquid water (ALW), as simulated by ISORROPIA II (as detailed in the Methods), averagely reached up to ~15.7 times of PM2.5 dry volume (Fig. S4), indicating a significant extent of heterogeneous/aqueous-phase reactions79,80. Under such high RH and high ALW, the gas phase oxidation would be suppressed due to the uptake of gas-phase oxidants to the particle phase81 and reduced radiation under the high wet aerosol mass82,83. The rise of secondary ONs was accompanied by increased SOR and NOR (Fig. 3). Previous studies have indicated that under high RH levels (≥60%), particulate nitrate and sulfate are predominantly formed via heterogeneous/aqueous-phase reactions84,85. In summary, we hypothesize that the secondary ONs observed from Period 2 to Period 3 were mainly related to heterogeneous-/aqueous-phase reactions, while the possibility of gas-phase oxidation followed by partitioning was not excluded. Three potential evolution pathways are proposed, as illustrated in Fig. 5.

  1. (1)

    Pathway 1 of ON evolution is the polymerization of amines with an oxygen-containing molecule. In Fig. 5a, with the decrease of CHN<180Da, the intensity of CHON>275Da gradually increases (r = −0.20, p = 0.18, Fig. S5). They both contain amine groups, and carbon numbers ranges were C6-C12 and C17-C19 for CHN<180Da and CHON>275Da, respectively. It is reasonable to hypothesize that CHON>275Da were formed by polymerization of C6-C12 amines with other compounds. Given that C17-C19 amines typically contain alcohol or carboxylic acid groups, the polymerization likely involves additional oxygenated molecules. Previous studies have shown that bimolecular reactions can occur on aqueous surface between amine-carbonyl2,86, amine-carboxylic acid2, and amine-Criegee87. For example, Duporté et al.86 reported that condensed-phase reaction between pinonaldehyde (C10H16O2) and dimethylamine (DMA, C2H7N) could generate oligomers including C12H21ON, C22H37O3N, and C32H53O5N. Furthermore, they pointed out that water could promote dimerization hydrate formation in the heterogeneous reactions, suggesting that the high-humidity conditions observed in our study may be more conducive to heterogeneous amine polymerization reactions.

  2. (2)

    Pathway 2 of ON evolution is the ring-opening fragmentation of nitrophenols to form CHO dicarboxylic acids. In Fig. 5b, with the decrease of nitroaromatics, the intensity of CHO dicarboxylic acids increases (r = –0.20, p = 0.19, Fig. S5). Most CHO dicarboxylic acids are non-ring structures. This suggests that nitroaromatics may undergo ring-opening fragmentation during heterogeneous reactions, yielding lower molecular weight but more oxidized CHO compounds. A study conducted by Wang et al.88 revealed that, the aqueous oxidation of primary aromatics at high RH lead to ring-breaking and functionalization to form carbonyls and carboxylic acids. Similarly, Hems et al.37 found that, with aqueous phase photo-oxidation of nitrophenols, OH• would attack hydroxyl-substitute carbon at an already substituted position, leading to the formation of dicarboxylic acid molecules. For instance, in Hem’s study, photo-oxidation fragmentation of nitrocatechol (C6H5O4N) would produce maleic acid (C4H4O4), a dicarboxylic acid. Similarly, in our study, C6H8O4 may be formed by the aqueous phase photo-degradation of C8H9O4N. Their functional groups (Table 1) support this hypothesis.

  3. (3)

    Pathway 3 of ON evolution is the ring-retaining functionalization of nitroaromatics to form nitroaromatics carboxylic acids. In Fig. 5c, with the decrease of nitroaromatics intensities, the fraction of nitroaromatics containing carboxylic acid groups increase significantly (r = -0.45, p < 0.05, Fig. S5), indicating the functionalization of nitroaromatics. Previous studies have reported that oxidation of aromatics can generate aromatic carboxylic acids, such as toluene to generate benzoic acid89. However, such reactions have not been previously reported for nitroaromatics, deserving further studies. The nitroaromatics carboxylic acids may also be generated from the photo-degradation of polycyclic aromatic hydrocarbons90,91.

Fig. 5: Secondary evolution of ONs through heterogeneous/aqueous-phase reactions in the haze episode.
Fig. 5: Secondary evolution of ONs through heterogeneous/aqueous-phase reactions in the haze episode.
Full size image

a Timeseries of CHN<180 Da and CHON>275 Da. CHN<180 Da refers to CHN with molecular weight less than 180 Da, while CHON>275 Da refers to CHON with molecular weight greater than 275 Da. b Timeseries of nitrophenol and CHO with dicarboxylic acids. c Timeseries of nitroaromatics and ratio variation of nitroaromatics with -COOH group.

Figure 6 illustrates a potential heterogeneous/aqueous-phase oxidation mechanism for Pathway 3. MS/MS analysis identified three molecular, including 2,5-dimethyl-4-nitroaniline (C8H9O3N), 2-hydroxy-4-methyl-5-nitrobenzoic acid (C8H7O5N), and 2-hydroxy-5-nitrobenzene-1,4-dicarboxylic acid (C8H5O7N), respectively. Based on the characteristics of their timeseries, it is proposed that C8H9O3N was primarily from biomass burning. With increases in air humidity, heterogeneous/aqueous-phase reactions facilitate C8H9O3N to form particulate C8H7O5N and subsequently to C8H5O7N.

Fig. 6: Heterogeneous/ aqueous-phase oxidation of a biomass burning-derived nitrophenol.
Fig. 6: Heterogeneous/ aqueous-phase oxidation of a biomass burning-derived nitrophenol.
Full size image

a Timeseries of C8H9O3N, C8H7O5N, and C8H5O7N. b The corresponding MS/MS fragmentation patterns of three compounds. c Proposed oxidation reaction pathway.

While we have investigated the possible pathways in secondary ONs formation from amines and nitroaromatics with illustrative cases, there is still a lack of convincing evidences of the evolutionary mechanisms. Some of the pathways, such as ring-retaining oxidation of nitrophenols to form nitroaromatic carboxylic acids, have been rarely reported in previous atmospheric studies. This may be related to the high humidity (~86.8% during the secondary evolution) in Wuhan. Further experiments could simulate the observed secondary oxidation pathways under high humidity to quantify reaction kinetics and secondary organic aerosol (SOA) yields.

Isomers of the particulate organics

During the classification analysis, we found that isomers may exhibit distinct temporal trends and be distributed across different sources. Overall, ~39.3% of the ONs contained isomers and ~13.4% had three or more isomers. Among all ON isomers, ~43.7% were distributed across different classes. Specifically, in ESI+ mode, CHN compounds exhibited the highest proportion of isomers (44.8%), followed by CHON (41.4%), CHONS (37.8%), CHO (29.0%), and others (7.6%) (Fig. 7a). Among them, ~14.0% organics had three or more isomers, with 44.1% of the isomers distributed across different classes. In ESI- mode, CHONS exhibited the highest proportion of isomers (35.3%), followed by CHON (33.6%), CHOS (25.2%), and CHO (25.1%) (Fig. 7d). Among them, ~7.0% organics had three or more isomers, with 43.9% of the isomers distributed across different classes.

Fig. 7: Distribution of isomer numbers and sources with representative compounds.
Fig. 7: Distribution of isomer numbers and sources with representative compounds.
Full size image

a, d The distribution of isomer numbers among various species in ESI+ and ESI-. NO represents no isomers; 2-SC represents two isomers distributed in the same class; 2-DC represents two isomers distributed across different classes, and so on. b, e Examples of isomers with significantly different time series. c, f MS/MS fragmentation analysis of the corresponding isomers.

Furthermore, combining the timeseries and MS/MS data, we present two representative examples of isomers with distinct timeseries. For ESI+ mode, C13H17ON exhibits two isomers classified to POA-like Class 1+ and SOA-like Class 3 + , respectively (Fig. 7b). Through MS/MS (Fig. 7c), we observed that the two isomers exhibit significantly different neutral losses, indicating distinct fragmentation patterns. Isomer 1 exhibits neutral losses of C2H2O and C2H4, corresponding to ester/amide group and ethyl group30, respectively, whereas these neutral losses are absent in isomer 2. For ESI- mode, C9H9O3N exhibits two isomers belonging to BBOA-like Class 2- and Class 3-, respectively (Fig. 7e). Isomer 1 exhibits neutral losses of HCN and H2O, while isomer 2 exhibits a more complex fragmentation pattern (Fig. 7f). HCN, CO2, and CO are characteristic neutral losses corresponding to amine/nitrile, carboxyl, and carbonyl groups, respectively. CHNO and C2H3NO are typical neutral losses associated with amide groups30. It might be presumed from the molecular formula C9H9O3N that the compound is a nitrophenol. However, neither isomer exhibits characteristic fragments such as •NO, •NO2, HNO, HNO2, CO, and CHO (a carbon directly bonded to benzene ring)30. Therefore, HPLC and MS/MS structural analyses are essential for accurate species identification. This is particularly important when compounds are used as source tracers, as isomerism can introduce substantial uncertainty21.

Methods

Sample collection and basic analysis

The sampling site was located at Wuhan Ziyang National Ambient Air Quality Automatic Monitoring Station (30.53°N, 114.30°E), and there was no obvious building shading or pollution source around the sampling site (Fig. S6). The sampling period was from Jan. 13th to 24th, 2018, covering 12 days. PM2.5 concentration showed an increasing and then decreasing trend during this period, indicating a complete pollution process. Four samples were collected each day from 06:00–11:00, 11:30–16:30, 17:00–22:00, and 22:30–05:30 for an overall of 46 samples. Since the duration of the first three periods was 5 hours, while the last period was 7 h, the total sampling flow rates and total mass of PM2.5 on the filters were different. The differences in sampling time were accounted for by normalizing the signals to the same sampling volume.

The PM2.5 samples were collected by a high-flow particulate sampler (TH-1000C II, 1.05 m3 min–1, Tianhong, China) using the quartz filters (203 mm × 254 mm, Whatman, England), and a medium-flow sampler (TH-150F, 100 L min–1, Tianhong, China) using the Teflon filters (Diameter 90 mm, Munktell, Switzerland). The collected PM2.5 samples were firstly weighed. The organic and elemental carbon (OC and EC) in the quartz filters were respectively analyzed by the thermal optical reflectance method (Thermal/Optical Carbon Analyzer, DRI Model 2001A; Desert Research Institute, USA). The water-soluble ions (Cl-, NO3-, SO42-, and NH4+) in the quartz filters were analyzed by ion chromatography (883 Basic IC Plus, Metrohm, Switzerland) after extraction. The trace metals in the Teflon filters were analyzed by inductively coupled plasma-mass spectrometry (ICP-MS, ElanDRCII; Perkin-Elmer, USA)92. The meteorological parameters (temperature, relative humidity, wind speed, and wind direction) were measured with a meteorology station. Gaseous pollutant concentrations (SO2, NO2, O3, and CO) were from the nearest National Air Quality Station (Ziyang station) to our sampling site.

Measurement of the molecular-level organic composition

We compared three most frequently used solvents for LC–MS untargeted analysis including methanol (MeOH), acetonitrile (ACN) and acetonitrile/water (8:2, v/v, ACN/water) (Table S1), and found the superiority of MeOH in the extraction organic species numbers. For nitrogenous organics (ONs), the three solvents showed comparable ON fractions in ESI+ mode (93.2%, 86.8%, and 85.7% in MeOH, ACN, and ACN/water, respectively). There was a higher CHON fraction in MeOH (72.2%) and ACN (65.6%) than in ACN/water (49.5%), but a lower CHN in MeOH (15.1%) and ACN (14.7%) compared to ACN/water (31.0%). However, in ESI- mode, ACN/water showed a higher ON fraction (50.4%) than MeOH (29.1%) and ACN (24.3%), which may result from the higher CHONS fraction (16.3%, 17.2%, and 43.4%, in MeOH, ACN, and ACN/water, respectively), while CHON were comparable (12.8%, 6.1%, and 7.0%, respectively). Moreover, the CHOS fraction was significantly higher in ACN/water (31.7%) than in MeOH (12.8%) and acetonitrile (9.5%) (Fig. S1). Overall, methanol extraction yields a broader range of substances but may selectively overestimate nitrogen-containing components while underestimating sulfur-containing constituents.

Quartz filters with a diameter of 50 mm were cut and shredded using methanol-rubbed ceramic scissors. 5 ml of methanol (chromatographically pure) was added to the filters and the extraction was put in an ultrasonic bath for 15 minutes. Ice was added to the ultrasonic bath water to prevent thermal decomposition or volatilization of organic matter. This procedure was repeated three times with same piece of filter and the three extracts were mixed together and then filtered through a 0.22 µm Teflon filter to remove insoluble substances. The filtered sample was purged to almost dryness at 30 °C under a gentle stream of nitrogen and redissolved in 100 μl of methanol. The mean value of the concentration of OC in the sampled PM2.5 was 183.87 μg/m3. The mean value of the solution concentration finally prepared for the MS test by extraction was 1838.73 ng/μl, with a range of 634.3-3352.67 ng/μl, and the time series was shown in Fig. S7. The sensitivity of compounds may vary greatly for different species. For example, signal intensity in ESI+ mode is usually several time higher than those in ESI- mode20,93, which also happened in our study. However, the quantification is challenging due to the lack of calibration standards.

2 µl of the extracted solution was analyzed by HPLC (Dionex 3000, Thermo Scientific, USA) coupled to a Q-Exactive Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo scientific, USA). The Thermo Hypersil column (C18, 100 mm × 2.1 mm, 1.9 µm) was used. The mobile phases were A: 0.1% formic acid in water, B: 0.1% formic acid in acetonitrile, and the gradient elution was performed using a mixture of A and B phases at a flow rate of 300 µl/min. The mass spectrometer was equipped with a heated electrospray ionization source in ESI- and ESI+ modes. All samples were analyzed by MS mode and MS/MS mode in both ESI- and ESI + . The mass resolutions of MS were 140,000 at 200 Da, and the scanning ranges of MS were 50–750 Da. For MS/MS experiment, we employed a Data-Dependent Acquisition (DDA) mode to obtain MS/MS spectra. Precursor ions of intensities ≥10, 000 were selected and then fragmented using collision energy of 20 eV, 40 eV. Dynamic exclusion time was set to 10 s.

Before the HPLC analysis, we have optimized the liquid chromatography gradient elution procedures which varied greatly in different references. To determine the best procedures for our samples, we selected four representative chromatographic conditions from previous references for pre-analysis19,21,30,94, and the same actual atmospheric samples in Wuhan was analyzed under the four conditions. The peak times and peak areas of substances with reasonable molecular formulae were compared. As shown in Fig. S8, for all the four conditions, many substances peaked at the same time in the 98% B platform stage, and even more substances peaked after the beginning of the decline. Based on this test, we determined to shorten the gradient rise time and extend the 98% B platform time to improve the separation performance. The improved condition was as follows: the gradient elution conditions were: 5% B for 4 min, increased linearly to 98% B in 31 min and hold for 15 min, back to 5% B in the next 1 min and hold for 9 min. The results after the improvement are shown in Fig. S9, which shows that the peaks of the optimized conditions were better separated at the 98% B platform stage, more substances were detected in the rising stage of the previous gradient, and the peaks of the optimized conditions were reduced after the descending stage. Linearity tests were performed on samples of different concentrations, and most of the compound peak areas and substance concentrations conformed to a linear pattern, so the variation of peak area in different samples can represent the variation in its concentration. Overall, the optimized conditions showed a significant improvement in the separation performance and were used for subsequent sample analysis.

It should be noted that the measured organics in our study cannot represent all the particulate organic compounds, due to the selectivity of extraction and measuring methods. Firstly, we used methanol to extract organics. Although it is the most-widely used solvent for HPLC detection (Table S2) and are shown to extract more molecules than acetonitrile95 and acetonitrile/water, it may still miss a lot of less polar compounds. Secondly, the ability of HPLC to separate compounds is highly dependent on the solvent and the column. The Thermo Hypersil column (C18, 100 mm × 2.1 mm, 1.9 μm) and the mobile phases of water and acetonitrile that we chose are more suitable for the detection of polar compounds96. Thirdly, the ESI source can also only detect compounds with medium to high polarity, and hydrocarbons such as alkanes and olefins cannot be detected97.

Data processing of the molecular-level organic composition

Xcalibur (V4.2, Thermo Scientific, USA) software, the non-targeted analysis software MZmine (V3.4.16, Mzio, Germany), and MATLAB (R2023a, The MathWorks, USA) were used for data processing of MS. The HPLC-MS spectra was first processed using Xcalibur to subtract the background. Then the peaks were processed and the molecular formula was fitted by MZmine, and the detailed processing steps and parameter settings of MZmine software were referred to the previous literature19,20,30. The final molecular formula was expressed as CcHhOoNnSs, where c, h, o, n, and s correspond to the numbers of C, H, O, N, and S in the molecular formula, respectively. The molecular formulas were batch processed using MATLAB software programming to derive the number of each atom. In order to ensure the accuracy of the peak list and obtain a more rational molecular formula, we selected species that fulfill the following conditions: (1) signal intensity higher than 1000 for ESI+ and higher than 100 for ESI-; (2) signal-to-noise ratios greater than 10; (3) the corresponding chromatographic peaks of the compounds were checked one by one for plausibility and isomerism using the Xcalibur software, and only substances with good peak shapes were selected; (4) the elemental ratios were limited to 0.3–3.0 H/C, 0–3.0 O/C, 0–1.3 N/C, and 0–0.8 S/C. After manual check, about 2200 and 2100 compounds were identified in ESI+ and ESI- modes, respectively.

For the extraction of MS/MS spectra and functional group analysis. Firstly, we used MZmine software to match product ions with corresponding precursor ions to extract MS/MS spectra, and then imported it into SIRIUS (V6.0.7, San Diego Supercomputer Center, USA) for molecular structure identification. SIRIUS merging CSI: FingerID search services has excellent identification accuracy at small-molecule identification98 and has been used in atmospherically relevant studies21,30,40. In SIRIUS, we limited ionization adducts to [M + H]+ and [M − H]−, mass tolerance to 3 ppm. The top scoring candidate structure’s SMILES (Simplified Molecular Input Line Entry System) which can explicitly describe molecular structures using ASCII strings were exported from SIRIUS for further analysis. Finally, we applied APRL Substructure Search Program99 to enumerate atmospherically relevant functional groups, and if the number of compounds containing a certain functional group is less than 10, we will not further analyze that functional group.

Statistical analysis

Through MATLAB batch processing, each organic compound was given a time series showing its peak area variation during the 12 days. We found that the time series for different organics showed different trends. Therefore, we clustered the time series using K-Means clustering algorithm100. K-Means clustering algorithm is a common clustering method that classifies the time series to several classes. Time series of substances with more than or equal to 30 data points were selected for clustering to ensure reliability.

The sulfur oxidation ratio (SOR) and the nitrogen oxidation rate (NOR) calculated by Eq. S1 and Eq. S2 can represent the oxidation fraction of primary SO2 and NOx to secondary particulate sulfate and nitrate, respectively. Aerosol liquid water (ALW), were modeled by ISORROPIA II79,101,102 and calculated by Eq. S3 and Eq. S4, to indicate the extent of aqueous/heterogeneous reactions. The concentration of SO42-, NO3-, Cl-, NH4+, particulate metals, ambient temperature, and relative humidity were input to the ISORROPIA II model. The reverse mode and metastable mode were selected as previously described79.

Meteoinfo software was used for the backward trajectory analysis and air mass clustering. The input meteorological data during Jan. 2018 was obtained from the National Oceanic and Atmospheric Administration (NOAA). To facilitate specific analysis of air mass trajectories during heavy pollution periods and to trace pollution sources, a -72h backward trajectory was generated every hour. Since the main range of human activity is within 200 m of the surface, the starting height of the air masses was set to 200 meters. Fire spot data was downloaded from NASA Firms Web Fire Mapper and could be accessible at https://firms.modaps.eosdis.nasa.gov/103 (last accessed: 13 June 2025).