Introduction

Chemicals of emerging concern (CECs) are known to pose serious threats to the health of organisms worldwide and are thus subject to the Stockholm Convention and the European REACH (Registration, Evaluation, Authorization and Restriction of Chemicals) framework1,2. These hazardous chemicals have physicochemical properties that allow them to persist in the environment and undergo transformation processes, which leads to various transformation products (TPs), some even more toxic than their parent compounds2. Apart from pesticides, a number of toxic TPs are not considered as part of chemical risk assessment in the European REACH regulation, despite their widespread presence worldwide3. However, the formation and transformation mechanisms of CECs in the complex real atmosphere remain poorly understood, hampering sufficient consideration given to the atmospheric TPs generated from their parent CECs, in the existing framework of risk assessment and regulation.

Organophosphate esters (OPEs), as a group of CECs, are of increasing concern following the ban on polybrominated diphenyl ether flame retardants (PBDEs) by the Stockholm Convention in 2009. OPEs have become replacements for PBDEs, with the global consumption of OPEs increasing nearly a hundredfold from 2011 to 20194,5. OPEs in the atmosphere occur in gaseous and particulate adsorbed states, and the gas-to-particle partitioning determines the environmental fate of OPEs, including transport, deposition, and degradation6. Levels of OPEs in the air have been reported to be increasing in recent years, preferentially partitioning to the particle phase7,8. Several empirical models based on vapor pressure (\({{\mbox{p}}}_{{\mbox{L}}}^{\circ }\)) and the octanol-air partition coefficient (KOA) have been used to predict the partitioning of CECs, but differences in physicochemical properties complicate the interpretation of their results. For instance, most non-polar or less polar chemicals (e.g., PBDEs) are poorly water-soluble and can be reasonably estimated by all models9. However, the majority of highly polar OPEs have been observed to become very enriched on particles, contradicting predictions that suggest them to be mostly in the gas phase10,11.

Particle-bound OPEs are generally believed to be dominated by the partitioning of primary OPEs, which originate from the volatilization, abrasion, and leaching of OPE-containing products during their production, use, and waste disposal12. However, recent studies have detected OPEs as TPs formed by oxidation of organophosphate antioxidants (OPAs) in indoor dust and airborne particles, suggesting a possible secondary formation of OPEs13,14. Meanwhile, laboratory and modeling studies have demonstrated a large contribution of OPA oxidation to OPE contamination15,16. In the atmosphere, numerous oxidants and particulate components have complex influences on the uptake kinetics of organic gases and secondary organic aerosol (SOA) formation17. Yet, the occurrence and contribution of secondary formation of OPEs in the ambient atmosphere remain largely unknown. As with many other CECs, such gaps impede a complete understanding of the atmospheric fate of OPEs and hinder effective policy-making to reduce ecosystem and health risks.

In this study, intensive field observations in the Pearl River Delta (PRD) are combined with theoretical predictions (detailed in the Methods) to elucidate the partitioning behavior and formation pathway of OPEs. We report that the significant gap between the measured and estimated particle-phase fractions of OPEs partially results from their aqueous-phase processes in both summer and winter. Moreover, we reveal the potential impact of the ‘salting-in’ effect on the aqueous partitioning of OPEs in summer. We provide observational evidence for enhanced aqueous formation of highly polar OPEs (with higher solubility) in winter. Finally, we perform source apportionment of OPEs and analyze δ13C isotopic fingerprinting of molecular tracers in four of China’s megacities, which further confirms the increasing importance of aqueous secondary formation for OPEs in the atmosphere.

Results and discussion

Dynamics of OPE chemical composition over the PRD in China

We analyzed 11 OPEs in both gas and particle phases at nine urban sites of the PRD during summer (June to July) and winter (January to February) in 2018 (Supplementary Fig. 1). The OPE compounds are classified as alkyl-OPEs (i.e., triethyl phosphate (TEP), tris(2-butoxyethyl) phosphate (TBEP), tris(2-ethylhexyl) phosphate (TEHP), tripropyl phosphate (TPrP)), aryl-OPEs (i.e., triphenyl phosphate (TPhP), tricresyl phosphate (TCP), 2-ethylhexyl diphenyl phosphate (EHDPP)), and halo-OPEs (i.e., tris(1,3-dichloro-2-propyl) phosphate (TDCPP), tris(2-chloroethyl) phosphate (TCEP), tris(2-chloroisopropyl) phosphate (TCPP), tris(2,3-dibromopropyl) phosphate (TDBPP)). Detailed descriptions of the measurements are provided in the Methods.

The campaign-mean concentration of total atmospheric OPEs (∑11OPE in the gas and particle phases) was 16 ± 12 ng m−3 in the PRD region, which was significantly higher than those reported in other regions (Supplementary Table 1)11,18,19. The halo-OPEs comprised the major fraction of total OPEs in the atmosphere, followed by the aryl-OPEs and the alkyl-OPEs (Fig. 1a). TDBPP, which has been phased out of markets in many countries due to its high toxicity, accounted for only 11% of ∑11OPE. Notably, three chlorinated OPEs (Cl-OPEs, i.e., TDCPP, TCEP, and TCPP) comprised, on average, 73% of ∑11OPE in the PRD region, which was comparable to other urban areas in China (Supplementary Fig. 2)11,18,19. Such high Cl-OPE fractions in China are probably explained by their high usage and resistance to degradation20.

Fig. 1: Composition profiles of OPEs and variation of OPEs as a function of PM2.5.
figure 1

a Concentrations and relative contributions of alkyl-OPEs (i.e., TEP, TBEP, TEHP, and TPrP), aryl-OPEs (i.e., TPhP, TCP, and EHDPP), and halo-OPEs (i.e, TCEP, TCPP, TDCPP, and TDBPP) in gas phase and particle phase during summer and winter. Relationships between the mass concentrations of PM2.5 and ∑11OPEs in (b) gas phase and (c) particle phase in summer and winter. The square boxes and the error bars represent the mean values and standard deviation, respectively. The gray dots represent the concentration of total OPEs in the whole sampling period.

Full size image

We observed distinct seasonal differences in both OPE concentration and composition (Fig. 1a and Supplementary Table. 2). First, the concentrations of gaseous and particulate OPEs in summer were much higher than those in winter, signifying that higher air temperature favors the release of OPEs from surface sources, such as the OPE-containing products, soils, water bodies and landfills21. Second, larger amounts of OPEs tend to accumulate in the particle phase than the gas phase, especially in winter. Third, in summer, the alkyl-OPEs, aryl-OPEs and halo-OPEs accounted for, on average, 1.1%, 15.4% and 83.6%, respectively, of the total OPE mass in the gas phase, and a similar pattern was observed in the particle phase (1.7%, 9.7% and 88.6%, respectively). However, in winter, while the mass fractions of alkyl-OPEs (16.2%) and aryl-OPEs (46.7%) in the gas phase were nearly twice as high as those in the particle phase (7.7% and 20.9%, respectively), the halo-OPE fraction in the gas phase (37.1%) was half that in the particle phase (71.4%).

We further investigated the relationship between particulate matter ≤ 2.5 μm (PM2.5) and OPEs (Fig. 1b, c). The concentration of gaseous OPEs decreased with PM2.5 loading (R = − 0.95, p < 0.05), and the particle-bound OPEs were positively correlated with PM2.5 (R = 0.89, p < 0.05) in summer, implying that the uptake of gaseous OPEs by aerosol is an important pathway for the formation of particle-bound OPEs. This is supported by the similar pattern of composition in the gas phase and particle phase, as mentioned earlier. However, no significant correlation was found between the gaseous OPEs and PM2.5 in winter (R = − 0.04, p > 0.05), even though the particle-bound OPEs remained correlated well with PM2.5 (R = 0.97, p < 0.05). The nitrate (NO3−) fraction in winter accounted for nearly half of the water-soluble inorganic constituents, which was two times higher than that in summer (Supplementary Fig. 3). It can be inferred that SOA formation in winter would be driven by the increasing aerosol liquid water (ALW), since NO3− is more hygroscopic than sulfate (SO42−)22. This was further supported by the higher secondary organic carbon (SOC) mass fraction observed in winter (55%) than that in summer (42%). As shown in Supplementary Fig. 4, SO42− was positively correlated with relative humidity (RH) in winter (R = 0.63, p < 0.05), whereas in summer, it displayed a negative correlation with RH (R = − 0.56, p < 0.05) and positively correlation with atmospheric oxidants (Ox, the total concentration of ozone (O3) and nitrogen dioxide (NO2), i.e., Ox = O3 + NO2) (R = 0.79, p < 0.05). These different seasonal characteristics lead us to hypothesize that winter PM2.5, containing a large fraction of hygroscopic ions, may have enhanced the formation of secondary species, which is consistent with previous studies23,24,25. We, therefore, investigated in detail the role that gas-to-particle partitioning and formation processes play in the fate of OPEs in the atmosphere.

Gas-to-particle partitioning of OPEs

As shown in Supplementary Table 3, most highly polar OPEs (\({{\mbox{p}}}_{{\mbox{L}}}^{\circ }\) ≥ 10−7 atm and log KOA < 8.5), like TEP, TPrP, TCEP, and TCPP, have high water solubility, ranging from 52 to 1.12 × 104 mg L−1, which are considered to be hydrophilic OPEs. Comparatively, the water solubility of less polar OPEs (\({{\mbox{p}}}_{{\mbox{L}}}^{\circ }\) < 10−7 atm and log KOA ≥ 8.5), like TBEP, TEHP, aryl-OPEs, and TDCPP, is 3 to 6 orders of magnitude lower than that of hydrophilic OPEs, and they can thus be regarded as hydrophobic OPEs. Herein, we applied the equilibrium absorptive partitioning theory to model the absorptive partitioning into the organic phase26. The predicted particle-phase fraction of OPEs (Fp,t) was calculated from the corresponding gas-to-particle partitioning coefficient (KOM) and the mass concentration of organic matter (OM) in aerosols (detailed in the Methods). The measured particle-phase fraction of OPEs (Fp,m) was calculated using the concentrations of OPEs in gas (Cg) and particle phase (Cp) (detailed in the Methods). Fp,t with different \({{\mbox{p}}}_{{\mbox{L}}}^{\circ }\) datasets spans the gray-shaded region in Fig. 2a, b, of which the upper and lower limits are calculated using activity coefficients (ξ) corresponding to 0.3 and 3.0, respectively27.

Fig. 2: Modeling and observation of the gas-to-particle partitioning of hydrophilic and hydrophobic OPEs.
figure 2

a, b Comparison between the measured (Fp,m) and the calculated particle-phase fractions (Fp,t) of hydrophilic and hydrophobic OPEs by Pankow’s absorptive theory during summer and winter; Fp,t as a function of \({{\mbox{p}}}_{{\mbox{L}}}^{\circ }\) spans the gray-shaded region, where the upper and lower limits are calculated using activity coefficients (ξ) corresponding to 0.3 and 3.0, respectively. Measured particle-phase fractions of (c, d) hydrophilic (e, f) hydrophobic OPEs as a function of aerosol liquid water mass fraction (FALW) and organic matter mass fraction (FOM) in PM2.5. The error bars in (a–f) represent the standard deviation of the measured fraction values.

Full size image

We found that the diversity of partitioning behaviors of OPEs was not determined by their functional groups, but was largely influenced by different physicochemical properties (Fig. 2a, b). The average Fp,m of hydrophobic OPEs in both seasons was almost close to thermodynamic equilibrium partitioning, within the uncertainties of the measurements and calculations. This suggests that hydrophobic species have a greater tendency to associate with OM in PM2.5. However, the mean Fp,m of hydrophilic OPEs was much higher than their Fp,t, indicating that they did not preferentially partition into the organic phase of aerosols or gas phase, consistent with the previous results obtained from KOA-based models7,10. This phenomenon was previously explained, in part, by the influence of equilibrium sampling artifacts such as gas-phase absorption onto filters9. However, the deviation caused by measurement uncertainties has been suggested to be much lower than the gap between theoretical and observed values28, suggesting that OPEs may undergo other unknown mechanisms besides the sorption of OPEs onto a filter’s water film.

The absorbing phase in PM2.5 is generally not a single organic phase, as it contains a substantial fraction of compounds that are highly hydrophilic and could form a concentrated solution in particles. Assuming that aerosols have a single phase, the total concentration of aerosols could be underestimated by 4 to 22% or even more29. The aqueous-phase absorption of gaseous organics has been suggested to be an important formation pathway for SOA, which depends on their bulk solubility30,31. To investigate the different phase partitioning of OPEs, we estimated the concentrations of ALW and OM in particles (“Methods”). A substantial increase in the Fp,m of hydrophobic OPEs was associated with an elevated OM mass fraction in aerosols (FOM) rather than the ALW loading (Fig. 2e, f). The mass fraction of nonpolar organics partitioning into the aqueous phase is negligible compared to that in the organic phase, despite the ALW being 2 to 3 times higher than the total aerosol dry mass32,33. In contrast, the Fp,m of hydrophilic OPEs positively responded to the increase in ALW mass fraction (FALW), but showed insignificant relationships with FOM (Fig. 2c, d). This signifies that the partitioning of hydrophilic OPEs is very likely driven by the aqueous phase rather than the organic phase; this is consistent with an observation in the southeastern U.S., where the partitioning of water-soluble organics was highly dependent on ALW34. It has also been suggested that gaseous water-soluble organics could have a stronger potential to partition in the aqueous phase compared to the organic phase33.

To further investigate the aqueous absorption of OPEs, we calculated the field-derived effective Henry’s law coefficient (KH,e, mol m−3 atm−1) from measurement-based data (details in the Methods), which can be employed to describe the partitioning of gases to the aqueous phase. The KH,e of the OPEs in both seasons (1012–1014 mol m−3 atm−1) was 5–12 orders of magnitude larger than the estimated Henry’s law coefficient in highly dilute solution (KH,w, mol m−3 atm−1) obtained from the Estimation Programs Interface Suite (Supplementary Table 4). The difference (∆log KH) between the measured log KH,e and the recommended log KH,w of hydrophobic OPEs has a strong dependence on log KOA (R = 0.81, p < 0.05) and log KOW (R = 0.93, p < 0.05) describing the partitioning capability of organics from gas phase to organic phase and from aqueous phase to organic phase, respectively (Supplementary Fig. 5). This further suggests that the organic phase may be more favorable for the absorption of hydrophobic OPEs. By contrast, the ∆log KH of hydrophilic OPEs was not significantly correlated with both log KOA and log KOW (p > 0.05), indicating little effect of organic-phase absorption on the partitioning of hydrophilic OPEs. The ∆log KH of hydrophilic OPEs was stable at approximately 6 and was comparable to or even greater than that of the carbonyl compounds (3–7) reported by Shen et al35. Since the aerosol containing water-soluble compounds is more hypersaline than pure water, the interactions among water, organic, and inorganic species in aerosol have been identified to be partially responsible for the enhanced partitioning to the condensed phase35,36. Nevertheless, although there was no seasonal difference in the ∆log KH of hydrophilic OPEs, the Fp,m values of most hydrophilic OPEs in winter were highly underestimated compared to those in summer (Fig. 2a, b).

Enhanced aqueous secondary formation of hydrophilic OPEs in winter

As discussed before, aqueous-phase processing might have imposed significant impacts on the formation of hydrophilic OPEs with a mean mass fraction of 72% in the total OPEs during the field campaign. Compared to TEP and TPrP, TCEP and TCPP have been demonstrated to be more stable OPEs in the atmosphere37,38. Therefore, we selected TCEP and TCPP to further investigate the aqueous-phase formation pathways of hydrophilic OPEs. The concentrations of TCEP and TCPP in particle phase exhibited significant correlations with ALW (R = 0.80, p < 0.05) in winter (Supplementary Fig. 6a). They also tended to increase with the increase of 2-methylglyceric acid (2-MGA) serving as a tracer used to investigate the multiphase formation mechanisms of SOA39. However, no significant correlation was observed in summer (p > 0.05) (Supplementary Fig. 6b), although the mass fractions of TCEP and TCPP in the hydrophilic OPEs exceeded 90% both in summer and winter. Therefore, the secondary formation of hydrophilic OPEs seems to take place preferentially in winter. A plausible mechanism for OPE formation in the atmosphere may be the oxidation of OPAs15. We observed a significant correlation between Ox and particle-bound TCEP and TCPP (R = 0.42–0.79, p < 0.05) in winter but no significant correlation in summer (Fig. 3a, b). This is consistent with previous laboratory studies on O3 oxidation or photo-transformation of tris(2-chloroethyl) phosphite (TCEPi) on cover glass and dust15,40. Although reports on the environmental occurrence and transformation behaviors of tris(2-chloropropyl) phosphite (TCPPi) are scarce, the correlation coefficient between TCPP and TCEP (R2 = 0.61, p < 0.05) was higher than that of other OPEs (Supplementary Fig. 7), indicating that these two hydrophilic OPEs, with similar structures and physicochemical properties, may originate from similar sources or formation pathways. However, a higher concentration of ALW was observed in winter, particularly at the Guangzhou site, while the Ox concentration in winter was lower than that in summer (Supplementary Fig. 8), implying that the enhanced secondary OPE formation in winter may be influenced not only by oxidants but also by aqueous-phase processes in aerosols.

Fig. 3: Relationships between hydrophilic OPEs in PM2.5 and secondary formation tracers.
figure 3

Correlations between the total concentration of TCEP and TCPP in particle phase and the concentrations of (a, b) Ox and (c, d) succinic acid in winter and summer. The spots in (c, d) are colored according to the secondary organic carbon (SOC) concentrations.

Full size image

Aqueous-phase processes, including the absorption of precursors and aqueous oxidation, may play a key role in the secondary formation of hydrophilic OPEs in winter. Firstly, the concentration of succinic acid, which can be formed in the aqueous phase from lower molecular weight species via radical-radical reactions of carboxylic acids41, was higher in winter than that in summer (Supplementary Fig. 8). A highly positive correlation between the concentrations of hydrophilic OPEs and succinic acid in the particle phase was also evident in winter (R = 0.82, p < 0.05) (Fig. 3c). Secondly, the total concentrations of particle-bound TCEP and TCPP in winter tended to elevate with increasing SOC mass, whereas they showed much weaker relationships in summer (Fig. 3d). The SNA (comprising SO42−, NO3− and ammonium (NH4+)) levels in winter were higher (22 ± 8.1 μg m−3) than those in summer (8.5 ± 5.7 μg m−3). At high SNA concentrations and increasing RH, aerosols may contain more water and become less viscous (Supplementary Fig. 9), favoring the accumulation of precursors and oxidants on the aerosol surface and ultimately facilitating the aqueous-phase reaction to form OPEs. Thirdly, the contribution of the condensation of gas-phase oxidation products could be excluded in winter, as the total mass of TCEP and TCPP in the gas phase did not show significant correlation with Ox (R ≤ 0.30, p > 0.05), and their PM2.5-bound mass was not significantly correlated with Fp,m (R = 0.11, p > 0.05) (Supplementary Fig. 10a, c). Moreover, their precursors (e.g., TCEPi and TCPPi) exhibit physicochemical properties (e.g., log KOA, log KOW, and water solubility) similar to those of OPEs (Supplementary Table 3 and Table 5), implying that hydrophilic precursors may be as dominated by aqueous partitioning as hydrophilic OPEs. The water solubility of their precursors is 3 to 6 orders of magnitude higher than that of hydrophobic OPAs, indicating that these hydrophilic precursors are more likely to partition into the aerosol aqueous phase, followed by aqueous oxidation. Recently, laboratory studies have suggested that the transformation of OPAs may be influenced by a number of key factors, including O3 level, reactive oxygen species (ROS), and RH. For example, Liu et al. provided experimental evidence that OPA concentrations decreased significantly with increasing O3 levels, accompanied by a high yield of the corresponding OPEs15. Similarly, Zhang et al. revealed that the heterogeneous transformation of OPAs in dust was accelerated in the presence of ROS and elevated RH40. Consequently, the aqueous oxidation of precursors is likely the dominant process responsible for the secondary formation of hydrophilic OPEs in winter.

Transition metal elements (e.g., Fe and Mn) dissolved in the aqueous phase might play a crucial role in SOA formation, as they are expected to enhance the oxidative potential of O3 by producing the ROS via Fenton chemistry42. The concentrations of Fe and Mn in winter were approximately twice as high as those in summer (Supplementary Fig. 8). The particle-bound TCEP and TCPP in winter were highly correlated with Fe and Mn (R = 0.53–0.79, p < 0.05), especially at the Guangzhou site with higher ALW mass, while the weaker correlations were observed in summer (R ≤ 0.42, p < 0.05) (Supplementary Fig. 6c–f), consistent with the aforementioned relationships between hydrophilic OPEs and Ox. The increased levels of transition metal elements and ALW possibly led to an enhanced conversion of gaseous oxidants to the condensed phase43, thereby promoting the aqueous secondary formation of hydrophilic OPEs. This finding was in line with the conclusion of an experimental study that the oxidation of trimethyl phosphite with hydrogen peroxide (H2O2) in the presence of Fe salts occurred much faster than in the absence of an Fe-ions catalyst44. Further laboratory investigations into the role of Fe-salts and Mn-salts aqueous chemistry in the secondary formation of OPEs, along with additional field measurements, are therefore suggested.

Gas-to-particle partitioning and salting-in effect in summer

In contrast to traditional persistent organic pollutants, aqueous-phase partitioning of gaseous hydrophilic OPEs may contribute more to the particle-bound OPEs in summer. The PM2.5-bound TCEP and TCPP in summer were highly correlated with Fp,m (R = 0.62, p < 0.05), while the gaseous mass was not significantly correlated with Ox (p > 0.05) (Supplementary Fig. 10b, d). This was consistent with the similar fraction characteristics of OPEs in gas and particle phases during summer (Fig. 1a), pointing to the partitioning of gaseous OPEs onto particles. The Henry’s law-type partitioning of hydrophilic organics can be enhanced by the ‘salting-in’ effect36, which is defined as the negative slope (KS) of the linear relationship between the salt concentration in aqueous phase and the logarithm of the ratio of KH,w and KH,e (log (KH,w/KH,e)). In Supplementary Fig. 11, the log (KH,w/KH,e) values of TCEP and TCPP were regressed against the molality of SO42− in ALW (\({{\mbox{C}}}_{{{\mbox{SO}}}_{4}^{2-}}\), mol kg−1 ALW), since SO42− has been recognized as a major factor for the salting effect of polar species28,35,36. The KS for TCEP and TCPP were observed to be significantly negative in summer (p < 0.05), with values of −0.06 and −0.09, respectively, which fall within a similar range to that of carbonyl compounds and polyol tracers28,35. The ‘salting-in’ effect was associated with the hydrophilic interactions, such as hydrogen bonding and water replacement, resulting in a decrease in Gibbs free energy or an increase in entropy36,45. The enhanced partitioning of gaseous hydrophilic OPEs in summer might therefore be partially attributed to the ‘salting-in’ effect. However, the positive KS values and weak correlations indicated that such an effect was not obvious in winter, likely due to competition from other formation pathways, such as aqueous oxidation.

Contribution of secondary formation to hydrophilic OPEs

To further confirm the importance of secondary formation to hydrophilic OPEs, we performed positive matrix factorization (PMF) to investigate the potential sources of hydrophilic OPEs in four megacities in China with active aqueous oxidation. Figure 4 showed the locations of the sampling sites in a time-averaged map of SO42− production from aqueous sulfur dioxide (SO2) oxidation. A six-factor solution was achieved (Methods and Supplementary Fig. 12, 13). The first factor was characterized by a high abundance of TEHP (55%) and TPhP (83%), which was attributed to emissions from the electronic and lubricant products5. The second factor was dominated by TCEP (80%) and was identified as building construction emissions11. The third factor showed a high abundance of TCPP (86%) and TDCPP (62%), and was identified as plastics and interior decoration emissions5,46. Herein, the first, second, and third factors were collectively referred to as the production and use of OPEs. The fourth factor contained high loadings of metals (i.e., Sb, Pb, Cd, and As; 40–72%) that were associated with traffic and industrial sources47. The fifth factor was dominated by the abundance of TDBPP (80%), phthalic acid (72%), and levoglucosan (82%). TDBPP, a widely used flame retardant additive for textiles, became obsolete in many countries and has been considered unused since 1970 due to its carcinogenicity and genotoxicity7. Previous studies suggested phthalic acid as an indicator of waste disposal emissions and levoglucosan as an indicator of combustion emission48. Thus, the fifth factor was identified as waste disposal and incineration sources. Given that high fractions of secondary formation tracers (i.e., NO3−, SO42−, NH4+, and 2-MGA; 53–81%) were distinguished in the sixth factor, it was assigned as a secondary formation source39,49.

Fig. 4: Sources of hydrophilic OPEs in PM2.5.
figure 4

Relative contributions from production and use of OPEs, traffic and industrial emissions, waste disposal and incineration, and secondary formation sources to the mass concentrations of hydrophilic OPEs in (a) Chengdu, (b) Wuhan, (c) Shanghai and (d) the Pearl River Delta region (PRD) during summer and winter. Markers on the map show the measurement locations. The map is color-coded for the annual mean (January to December 2018) of sulfate (SO42−) production from aqueous sulfur dioxide (SO2) oxidation. The data were obtained from Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA-2) observations (https://giovanni.gsfc.nasa.gov/giovanni/). The base map of China and coastline boundaries in the map are originated from Meteolnfo free vector map data (http://meteothink.org/downloads/index.html).

Full size image

As shown in Fig. 4, the emissions from production and use of OPEs were the dominant contributors to hydrophilic OPEs in summer (36–87%) across all urban sites, likely due to enhanced OPE release from product surfaces under warmer conditions. The mixed source of traffic and industrial emissions contributed 10–42% and 16–27% to hydrophilic OPEs in summer and winter, respectively, highlighting their substantial impact on hydrophilic OPE levels. There were no significant seasonal differences in Wuhan and the PRD region, whereas Chengdu and Shanghai showed the opposite trend between summer and winter. Moreover, waste disposal and incineration emissions explained 3–32% of hydrophilic OPEs in winter, which was 3 to 10 times higher than that in summer (1–8%). The comparison among cities illustrates that the PRD and Chengdu are significantly affected by waste disposal and incineration emissions in winter.

More evidence supports the enhanced aqueous secondary formation of hydrophilic OPEs in winter. Firstly, the contribution of secondary formation showed a significant increase from summer (1–14%) to winter (25–50%). This indicated that an enhanced secondary formation of OPEs likely occurred in winter across China. Secondly, the sulfur oxidation ratio (SOR, SOR = nSO42−/(nSO42− + nSO2), where n represents the molar concentration), an indicator of atmospheric oxidation processes50, was highly correlated with secondary hydrophilic OPEs to Ox ratio ((hydrophilic-OPEs)sec/Ox) in winter (R = 0.62, p < 0.05), while a weaker correlation was observed in summer (R = 0.35, p < 0.05) (Fig. 5a, b). Thirdly, we examined the relationship between δ13C fingerprints of oxalic acid and (hydrophilic-OPEs)sec/Ox in Guangzhou, Chengdu, Wuhan, and Shanghai. The compound-specific δ13C of oxalic acid is generally used as a tracer to differentiate various atmospheric chemical processes51. As illustrated in Fig. 5c, with increasing ALW, the higher generation of hydrophilic OPEs occurred with oxalic acid becoming more depleted in δ13C in winter. On the contrary, (hydrophilic-OPEs)sec/Ox remained low regardless of the δ13C reduction during summer. This was consistent with previous in situ observations of substantial water-soluble organic aerosols formed through secondary aqueous processes in megacities of China, especially in winter51. Taken together, these observations provide robust evidence for the growing importance of aqueous oxidation processes triggered by the elevated ALW in winter.

Fig. 5: Variation of the secondary hydrophilic OPEs to Ox ratio ((hydrophilic-OPEs)sec/Ox) as a function of sulfur oxidation ratio (SOR) and δ13C of oxalic acid.
figure 5

a, b Correlations between (hydrophilic-OPEs)sec/Ox and SOR in winter and summer. c Relationships between (hydrophilic-OPEs)sec/Ox and δ13C of oxalic acid in winter and summer for Guangzhou, Chengdu, Wuhan, and Shanghai. The spots in (c) are colored by aerosol liquid water (ALW) concentrations. The mass concentrations of secondary hydrophilic OPEs were obtained from the positive matrix factorization (PMF) model results.

Full size image

Atmospheric implications

The large contribution of aqueous-phase processes, including aqueous uptake and secondary formation of atmospheric OPEs, emphasized in this study, has important implications for accurate risk assessment and effective control of OPEs. Previous field and model studies focused on the partitioning behavior of OPEs between the gas and organic phase, but ignored the abundant aqueous phase in aerosols, resulting in the failure of KOA-based models to accurately estimate the partitioning of hydrophilic OPEs (e.g., TCEP and TCPP)9,21. Based on our observations, we reveal two distinct partitioning mechanisms of OPEs, characterized by the partitioning of hydrophobic species to the organic phase and the aqueous uptake of hydrophilic species, respectively. As for the sources of OPEs, current regulatory frameworks for OPEs generally pay more attention to the direct emission from OPEs used in commercial products, but does not consider the indirect contribution from OPAs15,16,52. Our findings support the hypothesis that aqueous-phase oxidation of OPAs is a key pathway for hydrophilic OPE formation in the atmosphere.

As shown in Fig. 6, we illustrate two prominent effects of aqueous-phase processing on OPE formation in the atmosphere. First, particle-bound OPEs in summer are mainly driven by the gas-to-particle partitioning of gaseous contaminants emitted from primary sources. The high salt concentration (e.g., SO42−) in the aqueous phase can promote the Henry’s law-type partitioning process of OPEs via the ‘salting-in’ effect. Second, aqueous-phase reactions of OPAs and oxidants dominated TCEP and TCPP formation under polluted conditions during winter. Specifically, OPE precursors, such as TCEPi, first uptake into the aerosol aqueous phase through gas-to-particle partitioning, and subsequently undergo aqueous oxidation with O3/O2 to form TCEP. In the presence of abundant SNA in aerosols and increasing RH, enhanced ALW formation can facilitate the aqueous uptake of OPAs and oxidants, leading to greater OPE production in winter. In addition, the high fractions of Fe and Mn dissolved in the aqueous phase can enhance the oxidative potential of O3 by generating ROS via Fenton chemistry, which may also be important for OPE formation. Therefore, these aqueous-phase processes can form an important, yet overlooked, driving force for OPE formation, especially with global warming and changes in atmospheric pollutants. Numerous studies have suggested that global warming can accelerate photochemical reactions to enhance O3 formation53,54. An increasing surface O3 level has been observed in the PRD, YRD, and BTH regions, resulting in enhanced SOA formation55,56. In addition, the ammonia-rich atmosphere in China, together with the slower decline rate of NOx emissions and the effective sulfur emission control, has led to an increase in ALW and decreased aerosol viscosity, facilitating the aqueous-phase chemistry processes across China17,22. Moreover, the rapid transition of energy structure from coal to natural gas in China was also expected to drive aerosols to be more hygroscopic, since natural gas combustion can generate three times more water vapor than coal burning51.

Fig. 6: Conceptual diagram depicting the secondary formation of hydrophilic OPEs with increasing relative humidity (RH) and aerosol liquid water (ALW).
figure 6

The left formation scheme illustrates the potential formation pathway of hydrophilic OPEs in summer, which is dominated by the partitioning of gas-phase OPEs emitted from primary sources. The right reaction scheme shows the secondary formation of hydrophilic OPEs via aqueous-phase oxidation of organophosphite antioxidants (OPAs), which is driven by high ALW triggered by increasing RH and SNA (comprising SO42−, NO3−, and NH4+) concentrations in winter.

Full size image

In addition to aerosol water, cloud/fog droplets are also important media for aqueous-phase reactions, where dissolved inorganic and organic gases are oxidized by various oxidants (e.g., singlet oxygen, O3, and hydroxyl radical)30,31. These cloud/fog processes are generally accompanied by a liquid water content several orders of magnitude higher than ALW content, which has been demonstrated to enhance the formation of secondary species through accelerating aqueous-phase oxidation24,57. Given that clouds cover nearly 70% of the Earth and the cloud/fog events frequently occur in winter58,59, the role of cloud/fog chemistry in OPE formation should be undertaken for a better understanding of the atmospheric fate of OPEs. Note that the aqueous-phase processes in aerosol water or cloud/fog droplets are likely influenced by many factors. For example, transition metal elements (e.g., Fe, Mn) may catalyze OPE precursor oxidation via Fenton-like reactions in the aquatic environment60. Recent evidence showed that O3-driven Fenton chemistry occurred four orders of magnitude faster on aqueous microdroplets than in bulk water42. A field study further suggested that Mn catalysis contributed up to 92.5% of SO42− yields in North China’s haze episodes61, highlighting the importance of metal-mediated pathways. While our results provide field evidence for the aerosol aqueous-phase formation of OPEs and its dependence on aerosol compositions, more laboratory-controlled studies are needed to better understand the factors influencing these aqueous-phase processes.

This study is limited by the scarcity of data on OPA production and use, as well as the technical challenges in measuring OPAs in aerosols, especially for TCPPi. Despite the high global production volume of total OPAs (up to 100,000 tons per year)15, data on the production and application of individual OPAs (e.g., TCPPi) remain unavailable. Moreover, the absence of reported environmental concentrations of TCPPi is a key limitation, which may reflect analytical challenges rather than its absence in the environment. TCEPi, including but not limited to it, has been detected in dust and soil but not in airborne fine particles, likely due to its rapid transformation in the atmosphere13,14,20. To address these gaps, future studies urgently need to establish a comprehensive database of OPA production and application and refine environmental monitoring methods for OPAs to further understand the secondary formation mechanism of OPEs in the atmosphere. While previous laboratory studies and our field observations provide insights into OPE formation mechanisms, upcoming research should prioritize integrating field measurements and laboratory simulations to better investigate aqueous-phase secondary formation pathways.

Our work provides important insights into the atmospheric fate of CECs and relevant air pollution control strategies, including but not limited to OPEs. First, CECs with different physicochemical properties show diverse gas-to-particle partitioning behaviors (e.g., organic- and aqueous-phase uptake), which should be considered in modeling studies to improve prediction accuracy. Second, failing to account for the aqueous secondary formation of CECs from parent chemicals could lead to an underestimation of the contribution of indirect sources and introduce uncertainties in risk assessment. Finally, a comprehensive and mechanistic understanding of CEC transformation processes would assist in improving TPs risk assessment and effectively addressing their hazards in chemical development and production, motivating refinements in future chemical management policies.

Methods

Sampling information

Sampling campaigns were conducted in nine cities in the PRD region (Supplementary Fig. 1): Guangzhou, Zhuhai, Shenzhen, Dongguan, Huizhou, Zhongshan, Jiangmen, Foshan, and Zhaoqing. The PRD region, as one of the representative megacity clusters, is the base of electronics and manufacturing industries in China, where high OPE concentrations have been observed62. In addition, active atmospheric chemical processing results in frequent O3 pollution in the PRD region55. Thus, these emission hotspots and strong atmospheric oxidation capacity provide a representative environment for the atmospheric transformation of OPEs. A total of 162 paired air samples were collected across 12 cities in China, comprising 126 samples from nine cities in the PRD region and 36 samples from three additional cities (Chengdu, Wuhan, and Shanghai). Each paired sample consisted of a quartz fiber filter (QFF, Whatman, 203 mm × 254 mm) for particle-phase collection and a polyurethane foam (PUF, 14 cm in diameter, 7.5 cm in thickness, and 0.02 g cm−3 in density) plug for gas-phase collection, using a high-volume active air sampler (Minya Instruments Co., Guangzhou, China). Sampling was conducted over 24-hour periods during summer (June–July 2018) and winter (January–February 2018), with 28 samples collected per city in the PRD region and 24 samples per city in the other three locations. To ensure data quality, four field blanks (24 PUF blanks and 24 QFF blanks) were collected at each sampling site to account for potential contamination during sampling.

Each QFF was baked at 450 °C for 8 h, and PUF was precleaned separately with acetone and dichloromethane before sampling. All samples were freeze-dried, packed, and stored at − 20 °C. The concentrations of PM2.5 were calculated by the gravimetric method. The meteorological parameters (RH and temperature) and the trace gas concentrations of O3, NOx, and SO2 were obtained from the corresponding official environmental monitoring stations at each sampling site, where trace gases were monitored using Thermo Scientific Environmental Instruments (models 49i, 42i, and 43i-TLE, respectively).

Chemical analysis and quality control

OPEs in gas- and particle-phase samples were determined using Soxhlet extraction combined with multi-segment column purification, and instrumental analysis7. QFFs and PUFs were spiked with TnBP-d27 and TDCPP-d15 as recovery surrogates and Soxhlet extracted for 24 h with dichloromethane (DCM). The extracts were reduced and concentrated in a volume of 0.5-1 mL through nitrogen evaporation and solvent-exchanged into hexane. Multilayer silica gel was applied to purify the sample extracts and then eluted with 20 mL of DCM and acetone (1:1, v/v). The eluted solution was concentrated under nitrogen to a volume of 200 μL and spiked with TCEP-d12 as an internal standard prior to instrumental analysis. Analysis of OPEs was performed using an gas chromatograph coupled to a mass selective detector (Shimadzu GCMS-QP2010), operated in negative chemical ionization under selected ion monitoring (SIM) mode with a DB-5MS column (30 m × 0.25 mm × 0.25 μm). The bulk components in PM2.5 including water soluble ions (i.e., SO42−, NO3−, Cl−, NH4+, K+, Na+, Ca2+ and Mg2+), organic tracers (e.g., levoglucosan, phthalic acid, succinic acid and 2-methylglyceric acid), metal elements (e.g., Fe, Sb, Cu, Pb and As), organic carbon (OC) and elemental carbon (EC), were measured for each filter sample. Detailed analytical methods of these compositions are given in the Supplementary Method 1.

For quality control, field blanks, procedural blanks, and surrogate-spiked recoveries were analyzed with each batch of samples to assess potential background contamination and instrumental performance. The reference ions for 11 OPEs and the corresponding labeled standards used in SIM mode on a Shimadzu GCMS-QP2010 and retention times (RT, min) are given in Supplementary Table 6. The instrumental detection limits (IDLs) and method detection limits (MDLs) of the target OPEs are provided in Supplementary Table 7. The average recovery rates of surrogates in samples were 86 ± 21% for TnBP-d27, and 51 ± 11% for TDCPP-d15. The detailed blank concentrations, including field and procedural blanks, were shown in Supplementary Table 8. The concentrations reported here were corrected for field blanks and recoveries.

Compound-specific δ13C of oxalic acid

The PM2.5 filter was extracted with ultrapure water for three times and then concentrated to dryness. Subsequently, the extracts were derivatized with 10% boron trifluoride (BF3) in 1-butanol in a water bath at 100 °C for about 1 hour, and then extracted with n-hexane. The δ13C values of oxalic acid derivatives were analyzed using a gas chromatograph connected to an isotope ratio mass spectrometry (GC ISO-Link2/IRMS, Thermo Fisher Scientific). The δ13C values of free oxalic acid were estimated by an isotopic mass balance method, based on the measured values and the δ13C of the derivatizing agent (1-butanol, δ13C = − 30.21‰). During derivatization, the isotopic fraction was found to be less than 0.73‰ in our previous studies63. The difference in δ13C of oxalic acid in replicate analyses was less than 1.0‰.

Calculations of partitioning coefficients

For each OPE compound, the theoretical partition coefficient (KOM) for absorptive uptake into the organic phase can be calculated according to ref. 26:

$${{\mbox{K}}}_{{\mbox{OM}}}=\frac{{\mbox{R}}\times {\mbox{T}}}{{{10}^{6}\times {\mbox{W}}}_{{\mbox{OM}}}\times {\xi }\times {{\mbox{p}}}_{{\mbox{L}}}^{{\mbox{o}}}}$$
(1)

where R (8.2 × 10−5 m3 atm mol−1 K−1) is the ideal gas constant and T (K) is the ambient temperature. WOM (set at 200 g mol−1 for all samples64) denotes the average molecular weight of the absorbing phase. ξ is the activity coefficient of the compound in the absorbing phase. \({{\mbox{p}}}_{{\mbox{L}}}^{{\mbox{o}}}\) (atm) is the vapor pressure of the pure compound at the average ambient temperature. The particle-phase fraction of OPEs (Fp,t) based on Pankow’s absorptive theory can be determined as ref. 26:

$${{\mbox{F}}}_{{\mbox{p}},{\mbox{t}}}={\left(1+\frac{1}{{{\mbox{K}}}_{{\mbox{OM}}}\times {{\mbox{C}}}_{{\mbox{OM}}}}\right)}^{-1}$$
(2)

where KOM was calculated by the Eq. 1 and COM was the mass concentration of OM in aerosol (COM = OC × 1.865). The measured particle-phase fraction (Fp,m) was calculated by the mass concentrations of OPEs in gas and particle phases (Cg and Cp, ng m−3) as follows:

$${{\mbox{F}}}_{{\mbox{p}},{\mbox{m}}}=\frac{{{\mbox{C}}}_{{\mbox{p}}}}{{{\mbox{C}}}_{{\mbox{g}}}+{{\mbox{C}}}_{{\mbox{p}}}}$$
(3)

The measurement-based effective Henry’s law coefficient (KH,e, mol m−3 atm−1) was calculated according to Qin et al.28:

$${{\mbox{K}}}_{{\mbox{H}},{\mbox{e}}}=\frac{{{{\rm{\rho }}}}_{{\mbox{w}}}\times {{\mbox{C}}}_{{\mbox{p}}}}{{{\mbox{C}}}_{{\mbox{g}}}\times {\mbox{R}}\times {\mbox{T}}\times [{\mbox{ALW}}]}$$
(4)

where Cg (ng m−3) and Cp (ng m−3) are gas- and particle-phase concentrations of OPEs, respectively. ρw (g cm−3) is the density of water. R and T are the ideal gas constant and ambient temperature, respectively. ALW (μg m−3) is the aerosol water content driven by inorganic (ALWi) and organic components (ALWo). The ALWi was estimated using the ISORROPIA-II model, which was run in the metastable mode with meteorological parameters (T and RH) and the concentrations of inorganic ions (Na+, K+, NH4+, Ca2+, Mg2+, Cl−, NO3−, and SO42−)17. We excluded the data corresponding to RH > 95% and RH < 40%, as extreme RH values can introduce high uncertainties in ALWi prediction66,67. The ALWo was calculated using the following equation17:

$${{\mbox{ALW}}}_{{\mbox{o}}}=\frac{{{{\mbox{C}}}_{{\mbox{OM}}}{{\rm{\rho }}}}_{{\mbox{w}}}}{{{{\rm{\rho }}}}_{{\mbox{org}}}}\frac{{\kappa }_{{\mbox{org}}}}{\frac{1}{{\mbox{RH}}}-1}$$
(5)

where COM (μg m−3) is the mass concentration of OM; ρw and ρorg are the density of water (ρw = 1.0 g cm−3) and OM (ρorg = 1.4 g cm−3), respectively; and κorg is the hygroscopicity parameter for organic aerosol composition (κorg = 0.06)68.

The salting effects have been suggested to be an important factor for the Henry’s law-type partitioning process of organic compounds28,35. The salting effects can be described using the following equation35:

$$\log \left(\frac{{{\mbox{K}}}_{{\mbox{H}},{\mbox{w}}}}{{{\mbox{K}}}_{{\mbox{H}},{\mbox{e}}}}\right)={{\mbox{K}}}_{{\mbox{s}}}\times {{\mbox{c}}}_{{\mbox{salt}}}$$
(6)

where KH,w, and KH,e are the Henry’s law coefficients of the OPEs in pure water and in aerosol aqueous phase containing salts, respectively. KS is the salting constant or Setschenow constant, where a negative value indicates a ‘salting-in’ effect, while a positive value means a ‘salting-out’ effect. Csalt is the concentration of salt in mol kg−1 ALW.

PMF model

Source apportionment of OPEs at the study sites was performed using a positive matrix factorization (PMF) model based on multivariate factor analysis. PMF has been widely used for source apportionment research of atmospheric pollutants. In this study, the mass concentrations of the measured organic species and inorganic ions, along with their associated uncertainty, were analyzed by PMF v5.0. The basic calculation formula of PMF is as follows:

$${{\mbox{X}}}_{{\mbox{ij}}}={\sum }_{{\mbox{m}}=1}^{{\mbox{n}}}{{\mbox{G}}}_{{\mbox{im}}}{{\mbox{F}}}_{{\mbox{jm}}}+{{\mbox{E}}}_{{\mbox{ij}}}$$
(7)

where Xij represents the concentration of the j-th component of the i-th sample, n represents the number of factors, Gim represents the contribution rate of the m-th factor for the i-th sample, Fjm represents the fingerprint matrix of the main sources, that is, the content of the j-th component in the m-th factor, and Eij is the residual matrix, representing the difference between the measured values and the model-estimated values.

Twenty-two species were input into PMF, including hydrophilic OPEs (the total of TCEP and TCPP), seven OPE species (i.e., TBEP, TEHP, TPhP, TCEP, TCPP, TDCPP and TDBPP), phthalic acid (PA) and levoglucosan (Levo.), eight metals (i.e., Zn, Sb, Pb, Cd, As, Cu, V and Mn), and four secondary formation tracers (2-MGA, NO3-, SO42− and NH4+). Firstly, hydrophilic OPEs was set as a total variable and species weights were adjusted according to S/N, residual analysis, observed/predicted concentration correlation, and the Q values (Qtrue/Qexp) to determine the optimal source apportionment solution.

After iterative testing from two to eight factors in the preliminary runs, the six-factor solution was selected, as it yielded a Qtrue/Qexp ratio of 5.4, corresponding to a relatively stable variation in d(Qtrue/Qexp)/dN (Supplementary Fig. 12a). The scaled residuals for almost all species ranged between -3 and 3, indicating that outliers did not influence the solution. As shown in Supplementary Fig. 12b, the observed and PMF-predicted concentrations of hydrophilic OPEs were in good agreement (R2 = 0.96, p < 0.05). The factor profiles for the six-factor constrained run of the PMF analysis were shown in Supplementary Fig. 13. To evaluate the stability of the six-factor solution, bootstrapping (BS), displacement (DISP), and bootstrap combined with displacement (BS-DISP) were performed. As shown in Supplementary Table 9, 10, BS results of each factor were all above 90, and the DISP summary showed no observed drop of Q, and no error, and no swaps for dQmax = 4 either. BS-DISP is a hybrid approach with a combination of the BS and DISP methods, showing 96% cases accepted. These uncertainty assessments confirm that the six-factor solution is both stable and reliable.