Frankfurt Airport is one of the largest airports in Europe, with more than 500 000 flight operations in 2019, shared over four runways. It is located in the Rhine–Main metropolitan area within a distance of around 12 km to the city centre of Frankfurt. In 2019 more than 70.5 million passengers and 2.1×106 t of cargo were transported, with a kerosine consumption of around 5.5×106 m3 (Fraport AG, 2020). The coronavirus pandemic caused a decline in flight movements by 58.7 % in 2020 compared to 2019. The transported cargo decreased by 8.3 % and passenger numbers by 73.4 % (Fraport AG, 2021).

Since September 2017 the size distributions of ambient particles, including the ultrafine fraction, have been monitored at an air quality monitoring station of the Hessian Agency for Nature Conservation, Environment and Geology 4 km north of Frankfurt Airport (Fig. 1). It is located in a forested area with no highly frequented streets within 1 km distance. This sampling station is considered representative of the outflow emissions of highly frequented airports.

We used a Micro Orifice Uniform Deposition Impactor (Nano-MOUDI; Model 115, MSP, Minneapolis, MN, USA), which consists of a 13-stage impactor system that is able to collect size-resolved UFPs on the last four stages. However, we only sampled on the last three stages (aerodynamic diameters between 0.010–0.018, 0.018–0.032, and 0.032-0.056 µm) since particle-size-distribution measurements show the highest number concentrations for particles smaller than 50 nm (Rose and Jacobi, 2018). This type of low-pressure cascade impactor was already used by other research groups for analysis of atmospheric and industrial particulate matter (Cena et al., 2014; Chen et al., 2016; Fushimi et al., 2019).

The Nano-MOUDI was installed at the air quality monitoring site in Frankfurt Schwanheim (Fig. 1). We sampled ambient air without a pre-impactor inlet through a stainless-steel tube (ID: 1 in.; length: 2 m) towards the Nano-MOUDI, with a total sampling flow rate of 1.8 m3 h-1. The sampling on the last three stages was conducted with a flow rate of 0.6 m3 h-1. The deposition of particles > 0.056 µm on the upper 10 stages and the sampling of ambient UFPs on the last three stages were carried out using aluminium foils (TSI; Ø × THK: 47×0.015 mm). To limit a bounce-off of larger particles to the lower stages, the foils for the upper 10 stages were coated with Apiezon^® L grease. Therefore, after baking out the filters for 30 min at 300 ∘C, a 1 cm section of the Apiezon^® grease was dissolved in 100 mL n-hexane, and each foil was evenly coated by adding six droplets of the mixture. After drying for 24 h in a fume hood, each single foil was stored in a glass petri dish. The foils for the UFP sampling were heated at 300 ∘C for 30 min and then stored in metal boxes until sampling.

Sampling took place from August to October 2019 exclusively during a southerly wind direction and during the airport operating hours (05:00–23:00 CET). Sampling times varied between 18 and 54 h (Table S1 in the Supplement) in order to accumulate sufficient mass for analysis. Shortly after sampling, the filters were stored in metal boxes at -20 ∘C to prevent subsequent evaporation or transformation processes. The final filter loadings depend on sampling time, meteorological conditions, and operating state of the airport. Although the particle emissions of aircraft engines depend on the actual thrust (Timko et al., 2014; Lobo et al., 2015; Yu et al., 2017), we assume that we sampled a mixture of UFPs that represent different aircraft engines under various operating states. Hence, the collected samples reveal the average UFP emissions of the whole airport and not of individual engines or operating conditions. Due to no strict compliance with the night flight ban between 23:00–05:00 CET and further urban and industrial sources in the other wind directions, it was not feasible to collect isolated background samples. We collected field blanks on the last three stages for 115 h without active sampling air flow and processed them analogously to the UFP samples.

As a consequence of the low mass of the sampled UFPs (sub-microgram range of total particle mass after 30–50 h of sampling), the extraction method was developed in order to achieve the highest possible extract concentration. Simultaneously, no pre-concentration step through solvent evaporation was conducted in order to prevent volatilization of target compounds. Owing to the strongly adsorptive behaviour of the jet engine oil constituents, no filtration step of the solvent extracts was implemented. Various solvents were tested regarding their ability to dissolve the jet engine oil target compounds and the overall particle mass. Pure organic solvents provided a higher extraction efficiency than mixtures with water and similar efficiencies to mixtures of acetonitrile and methanol. The tests were based on UFP filter extractions with different solvents (100 % methanol; 50 % acetonitrile, 50 % methanol; 60 % methanol, 40 % water; 60 % acetonitrile, 40 % water) and evaluated based on the non-target-analysis-generated identifications and corresponding signal intensities. Due to the use of methanol (Optima^® LC/MS grade, Thermo Fisher Scientific) and water (Milli-Q, Merck) as the UHPLC solvents, pure methanol was used for the filter extraction.

A circular section with a diameter of 2.5 cm was cut out of each foil sample, which were located below the array of the nozzles of each impactor stage. It is worth mentioning that the region of deposited UFPs on the foil surface was generally not visible. The resulting sections were cut in small pieces (approximately 2.5 mm × 2.5 mm) with ceramic scissors and extracted two times with 100 and 50 µL, respectively, for 20 min each on an orbital shaker (300 rpm). We avoided ultrasonic extraction due to a possible formation of free radicals, which can lead to chemical transformation of the original sample composition (Riesz et al., 1985). The extraction was conducted in glass vials equipped with flat-bottom micro inserts with a maximal volume of 200 µL (LLG Labware; Ø × H: 6×31 mm) to ensure a complete filter surface covering and then pipetted into 100 µL micro inserts with conical bottoms (VWR; Ø × H: 6×31 mm) to guarantee a complete UHPLC injection of 5 µL in the autosampler. Altogether 22 filter samples and 3 field blanks were analysed.

Jet engine oils are made of a base stock material like pentaerythritol esters or trimethylolpropane esters and different additives. Organophosphate additives are used as anti-wear agents and metal deactivators (Wyman et al., 1987; Du Han and Masuko, 1998), whereas amine constituents serve as stabilizers (Wu et al., 2013) (Table S2).

In order to characterize UFPs and screen for the molecular composition of jet engine oil, we analysed five different jet engine oils of various brands. The oils were selected regarding the recommended oil brand of engine manufacturers. The Mobil Jet Oil II (ExxonMobil, Irving, TX, USA) has the highest market share (Winder and Balouet, 2002), whereas the market share of the other oils (Mobil Jet Oil 254 (ExxonMobil, Irving, TX, USA), Aeroshell 500 (Royal Dutch Shell, The Hague, Netherlands), Turbo Oil 2197 and 2380 (Eastman, TN, USA)) is unknown.

The chemical characterization of the airport filter samples was realized by using ultra-high-performance liquid chromatography (UHPLC) (Vanquish Flex, Thermo Fisher Scientific) and heated electrospray ionization (HESI) coupled to an Orbitrap high-resolution mass spectrometer (HRMS) (Q Exactive Focus Hybrid Quadrupole-Orbitrap, Thermo Fisher Scientific) as a detector. The chromatographic separation was achieved by using a reversed-phase column (Accucore C18, 150×2.1 mm, 2.6 µm particle size, Thermo Fisher Scientific), operated in gradient mode and thermostated at 40 ∘C (still air). Ultrapure water (18.2 MΩ cm; Millipak^® Express 40: 0.22 µm, Millipore; Milli-Q^® Reference A+, Merck) with 0.1 % formic acid (by volume; solvent A) and methanol (Optima^® LC/MS grade, Thermo Fisher Scientific) with 0.1 % formic acid (solvent B) were used as the UHPLC solvents. Formic acid (LiChropur^®, Merck) with a purity of 98 %–100 % was used to improve the chromatographic separation and ionization in the positive mode. Due to the high fraction of nonpolar compounds in the samples, the UHPLC chromatography started with 60 % solvent B (0–0.5 min), was then increased to 90 % (0.5–11 min), and then was ramped up to 99 % (11–16 min). The high-organic-solvent starting conditions allow the injection of large sample volumes of organic solvents without compromising the chromatographic separation. In the end, solvent B was reduced to 60 % (16–17 min), and the system was allowed to equilibrate for the subsequent measurement within 3 min. The overall method duration was 20 min, with a flow rate of 400 µL min-1 and an injection volume of 5 µL. The ionization for the mass spectrometric detection was conducted by HESI in positive and negative mode. The HESI settings in positive mode were 3.5 kV spray voltage, 40 psi sheath gas (nitrogen), 8 psi auxiliary gas (nitrogen), and 350 ∘C gas temperature. The negative mode was operated with the same settings but with a spray voltage of 2.5 kV. The scan range in the positive mode was from 150 to 750 and in the negative mode from 50 to 700 mass-to-charge ratio (m/z) with a resolution of ∼70 k at m/z 200. In negative-mode molecular ions ([M - H]-) are produced by deprotonation, whereas in positive-mode molecular ions ([M + H]+, [M + Na]+) are generated by adduct formation. In order to gain structural information of the single molecules, a data-dependent MS2 (dd-MS2) method was used with MS data recorded in profile mode.

The limit of detection (LOD) and limit of quantification (LOQ) of tri-o-cresyl phosphate (TCP) and pentaerythritol tetrahexanoate (C29H52O8) were determined by external calibration according to the standard DIN 32645 of the German Institute for Standardization (DIN). For tri-o-cresyl phosphate (≥97.0 %; Sigma-Aldrich) and for pentaerythritol tetrahexanoate (95 %; Carbosynth Ltd), we calibrated with five points in the range of 0.001–1 ng µL-1. Each calibration point was measured three times in succession. We determined the LOD of pentaerythritol tetrahexanoate as 0.007 ng µL-1 and the LOQ as 0.018 ng µL-1. For tricresyl phosphate we achieved an LOD of 0.021 ng µL-1 and an LOQ of 0.060 ng µL-1. Both calibration curves show only a small linear-response range, which is likely due to their adsorptive behaviour on glass surfaces. Therefore, we used only three concentrations in the range of 0.001–0.1 ng µL-1 for the determination of the LOD and LOQ. Presumably, all the detected pentaerythritol esters and trimethylolpropane esters show a similar adsorption tendency like the used ester standard. Due to the adsorptive behaviour, quantification of the UFP constituents in ambient samples should preferentially be carried out by standard addition. We translate the results of the calibration into ambient-concentration LODs: an average sampling duration of 40 h with the Nano-MOUDI flow rate of 0.6 m3 h-1 results in a sampled air volume of 24 m3. Reaching the LOD of pentaerythritol tetrahexanoate and tricresyl phosphate in a solvent extraction volume of 150 µL would require ambient-air concentrations of 44 and 131 pg m-3 (for each sampled size interval), respectively. However, we can project this calculation on ambient samples only under the assumption that no losses occur during sampling.

The full-scan MS spectra analysis was carried out by the non-target-analysis software Compound Discoverer (CD) (version 3.2.0.261, Thermo Fisher Scientific), which was already validated regarding the qualitative and quantitative results by comparison with the non-target-analysis software MZmine2 (Vogel et al., 2019). The Compound Discoverer software identifies substances and determines their molecular formula depending on their exact mass, isotopic signature, and MS2 fragmentation pattern and by database matching of MS2 fragmentation spectra (database: https://www.mzcloud.org/, last access: 11 July 2020).

We iteratively optimized the non-target-analysis workflow (see Supplement Fig. S1 and Table S3 for the detailed settings of the CD workflow) in order to achieve the best possible characterization of the particles' chemical composition. The described jet engine oils were diluted to a concentration of 1 µg mL-1 in methanol (Optima^® LC/MS grade, Thermo Fisher Scientific) and analysed with an identical non-target approach. The chromatograms of the jet engine oils, the UFP samples, and the field blanks were processed in one experiment in order to improve the software-based identification procedure. We filtered the non-target-analysis results with the following criteria: the sample signal must be more than 5 times larger than the blank signal, the minimum area is 1.0×105, and the minimum retention time is 0.7 min (column dead time peak at 0.6 min). A large signal at m/z 399.2506±4 ppm ([M + H]+ of tris(2-butoxyethyl) phosphate, C18H39O7P) was manually filtered out prior to the graphical illustration because of its ubiquitous use as a plasticizer, flame retardant, and floor finish (Lewis et al., 2016). The peak areas of the compounds in the samples were background-corrected by subtracting the corresponding peak areas of the field blank. For the detection of the compounds, a mass tolerance of 5 ppm was allowed, whereas we reduced the mass tolerance for the molecular-formula prediction to 2 ppm. The allowed elemental combinations for the molecular formulae assignments were constrained to the limits of C1-90, H1-190, O0-20, N0-4, S0-3, P0-1, Cl0-4, and Br0-3. Finally, we classified the assigned molecular formulas into composition groups (CHO, CHN, CHNO, CHOS, CHNOS, CHOP, other), which helped to graphically illustrate the complex chemical composition of UFPs.

In order to gain structural information of the chemical species found by the non-target analysis, we conducted successive data-dependent full-scan and dd-MS2 measurements in the discovery and confirmation mode, respectively. In a first measurement, we used the discovery mode to obtain the exact mass and possibly MS2 spectra of abundant signals. In this mode, the mass spectrometer estimates a peak apex during acquisition of the chromatogram and triggers the recording of a fragmentation spectrum by higher-energy collisional dissociation (HCD) of the highly abundant signals. Subsequently, we analysed the discovery-mode data with CD, and we obtained a list of compounds that are significantly different from the blank. This list was loaded as an inclusion list for a second measurement of the samples. By this approach, we recorded high-quality MS2 fragmentation spectra of the listed compounds of interest (even of the low-intensity compounds), allowing us to confirm the molecular structure and improve the accuracy of the database identification process. The MS2 experiments were conducted with a normalized collision energy (NCE) of 30.

We used the different jet engine lubrication oils as qualitative identification standards with regard to the material safety data sheets (MSDSs) declaring the composition of the oils. With the help of the exact mass, retention time, and MS2 fragmentation pattern of the jet oil constituents, we were able to unambiguously identify the oil-related compounds in ambient-UFP samples. In the case of TCP, the different structural isomers do not show different MS2 fragmentation patterns. Here, we make use of different retention times for the unambiguous identification of tri-o-cresyl phosphate, enabling us to examine whether the ortho-isomer can be found in the jet oils and the ambient-UFP samples.

The non-target screening detected almost 1000 compounds in the positive-ionization mode. However, we observed the majority of these compounds at similar intensities also in the field blanks and hence excluded from the dataset. We find approximately 200 organic compounds at a sample-to-blank ratio > 5 in the largest size fraction (0.032–0.056 µm) based on the average of six samples. For the two smaller size fractions, we determined the sample-to-blank ratios based on the average of 2×8 samples (see Table S1). The two smaller size fractions yielded more than 100 organic compounds with a sample-to-blank ratio > 5 (Fig. 2). Averaged over all filters of the largest size fraction, we obtained sample-to-blank ratios > 100 for approximately 30 compounds in the positive-ionization mode. The particles in the size fraction 0.032–0.056 µm show a 5- to 10-times-higher signal intensity of the most abundant compounds as well as a larger number of compounds due to the larger mass concentration compared to the smaller size bins. The negative-ionization mode revealed only 16 compounds with the same data processing filter criteria.

For each size fraction, we use the molecular fingerprints (retention time vs. molecular weight (MW), Van Krevelen diagram, Kroll diagram, and Kendrick mass defect vs. MW) for illustration of the chemical composition of the organic fraction, which is shown in Fig. 3 for 0.032–0.056 µm particles. Molecular fingerprints of the size stages 0.010–0.018 and 0.018–0.032 µm are shown in Figs. S2 and S3, respectively. Each circle represents a compound, and the colouring describes the molecular-composition group. Signals to which a molecular formula could not be assigned are classified as “other”. The area of the circles is proportional to the measured signal intensity. The measured mass-to-charge ratios (m/z) are converted into MW.

Figure 3a (retention time vs. MW) shows primarily CHO compounds. Group 1 represents mainly the homologous series of pentaerythritol esters (C27-38H48-70O8). We found that Compound Discoverer falsely interprets ion signals as [M + H]+ of the measured ions [M + Na]+, [M + K]+, and [M + NH4]+ of the pentaerythritol esters and hence suggests up to three likely false molecular formulas per ester molecule. These artefacts were manually removed based on the same retention time and exact mass difference to the molecular ions of the pentaerythritol esters. The native molecular fingerprint is displayed for comparison in Fig. S4. Pentaerythritol esters are used as a common base stock material for synthetic jet engine lubrication oils (Winder and Balouet, 2002) and therefore can be attributed to jet engine oil emissions. These synthetic oils are particularly used in aviation, while the base stock of automotive oils is made of a crude-oil fraction that consists mostly of petroleum hydrocarbons (Vazquez-Duhalt, 1989). Another jet engine oil base stock material of trimethylolpropane esters (C27-34H50-64O6) (Wright, 1996) forms a homologous series in group 2. Group 3 can be mainly attributed to fragmentation, probably formed during the ionization of the pentaerythritol esters in the HESI source. The compounds in group 3 have the same retention times as the original pentaerythritol esters of group 1 but differ in their molecular weight, which can be explained by the neutral loss of fatty-acid fragments (e.g. C5H10O2, C7H14O2, C8H16O2, and C10H20O2). Although electrospray ionization is known as a very soft ionization technique, we observe different fragmentation products of the pentaerythritol esters, depending on the side chain length. In order to minimize fragmentation of the pentaerythritol esters, we tested different settings of the HESI source: the auxiliary gas heater temperature was reduced from 350 to 200 ∘C in 50 ∘C steps, and the spray voltage was reduced from 3.5 to 2.0 kV in 0.5 kV steps. Variation in both parameters, which were identified as the most important drivers of the ionization process, did not reduce fragmentation of the pentaerythritol esters. Besides the large CHO compounds at 5.2 min, which can likely be attributed to decanedioic acid-1,10-diethyl ester (C14H26O4) and its fragments, we do not detect large contributions of other organic species in this size fraction. Only a small number of signals remain unidentified, appearing in the “other” composition group. Hence, the results of the non-target analysis suggest that the sampled UFPs are mainly composed of lubrication oils. However, other techniques might reveal the presence of additional compounds (e.g. metals, black carbon, inorganic compounds) which are not detected by the presented technique.

The Van Krevelen diagram describes the different compounds regarding their hydrogen-to-carbon (H/C) and oxygen-to-carbon (O/C) ratios (Fig. 3b). Compounds that consist mainly of single-ring aromatic moieties have an H/C ratio around 1. Tricresyl phosphate, an anti-wear additive in synthetic jet oil, contains three single-ring aromatic (cresol) moieties. TCP appears in Fig. 3b as a magenta circle at H/C=1 and O/C=0.2. The majority of the detected compounds, however, are located in the upper part of the diagram (H/C>1.5), which implies that most of the airport-related UFP constituents do not have an aromatic character. With O/C ratios below 0.6, it can also be stated that the UFPs do not become oxidized during the 4 km transport distance between the emission source and the monitoring station. The analysed jet engine lubrication oils show O/C ratios within the same range (0.2–0.4; Figs. S6–S10).

The average carbon oxidation state (avg. OSC) versus the number of carbon atoms (Fig. 3c) is a way of characterizing the oxidation state of complex organic-compound compositions and is calculated according to Kroll et al. (2011). The diagram shows that the synthetic esters (CHO) are large molecules containing 27–37 carbon atoms, which implies a low vapour pressure and supposedly the ability to reside in the UFP fraction. The average carbon oxidation state was not corrected for oxygen atoms that are attached to non-carbon atoms. Thus, compounds with oxygen-containing functional groups (e.g. oxygen atoms of the phosphate group in tricresyl phosphate) are shifted in the diagram towards seemingly higher OSC values, but in fact the carbon atoms exhibit a lower average oxidation state. Also in this illustration, we find no evidence for oxidation products of airport-related lubrication oil emissions. These results are consistent with the objective to design lubrication oils with a stability against oxidation. For this purpose antioxidants such as N-phenyl-1-naphthylamine (C16H13N) and alkylated diphenyl amine (C28H43N) are added, which serve as radical scavengers (Wu et al., 2013). Even though tricresyl phosphate serves as an anti-wear agent in lubrication oils, it is also reported that this additive is able to enhance the oxidation inhibition ability of antioxidant mixtures (Duangkaewmanee and Petsom, 2011).

We illustrate the Kendrick mass defect (CH2 base) vs. MW in order to identify homologous series of hydrocarbons (Fig. 3d). The Kendrick mass defect (KMD) is calculated by subtracting the Kendrick mass of the nominal mass (Equation 1). The Kendrick mass is defined as the IUPAC mass standardized on 12CH2 with an exact mass of 14 Da (Kendrick, 1963). 1KMD=nominalmass-IUPACmass⋅14.0000014.01565 Thereby, compounds of a homologous series differing in the number of CH2 groups align on a horizontal line. We observe the homologous series of pentaerythritol- and trimethylolpropane esters and their fragments as horizontal lines in Fig. 3d.

We analysed the molecular pattern of the synthetic esters by comparison between the UFP samples and the five different jet engine oils. Figure 4 shows the contribution of each individual ester molecule to the sum of all esters. The boxes illustrate the spread of the ester fraction in all 22 measured ambient-UFP samples, and the coloured symbols indicate the composition of the oil standards. We observe that the ester composition of four out of the five jet engine lubrication oils matches the ester composition in the UFP samples well, which confirms that these esters can be attributed to the specific source of jet engine oils. Furthermore, the Mobil Jet Oil II follows most closely the ambient pattern of the median level of the pentaerythritol esters (eight oxygen atoms), which is in line with the highest market share of this oil. The Eastman Turbo Oil 2380 consists of trimethylolpropane esters (six oxygen atoms) as the base stock, which are not a constituent of the other four jet engine oils. We observe a small but significant fraction of the trimethylolpropane esters in the ambient UFPs, indicated by the consistently higher values of the boxes (and the median) compared to the pentaerythritol-ester-based oils. This is presumably a result of a minor utilization of trimethylolpropane ester containing jet engine lubrication oils in air traffic. Chromatograms of the various jet engine lubrication oils are shown in Fig. S5 and the individual molecular fingerprints as in Figs. S6–S10, showing the different base stock materials.

We used fragmentation experiments of the homologous series of pentaerythritol esters (C27-38H48-70O8), N-phenyl-1-naphthylamine (C16H13N), alkylated diphenyl amine (bis(4-(1,1,3,3-tetramethylbutyl)phenyl)amine; C28H43N), tricresyl phosphate (C21H21O4P), and the homologous series of trimethylolpropane esters (C27-34H50-64O6) in order to verify the identity of the molecules in the ambient-UFP samples. These compounds have been described as molecular markers for jet engine lubrication oil emissions and are not present in automotive lubrication oils (Fushimi et al., 2019). Figure 5 shows the comparison between the fragmentation patterns of tricresyl phosphate (Fig. 5a), the alkylated diphenyl amine (Fig. 5b), and one of the pentaerythritol esters (Fig. 5c) in the UFP samples and the Mobil Jet Oil II by ExxonMobil, respectively.

The fragment at m/z 261.0674 (C14H14O3P) is a fragment ion of tricresyl phosphate with only two aromatic moieties, whereas the fragment at m/z 91.0543 (C7H7) is the corresponding aromatic structure without the oxygen atom related to the phosphate group. Fragmentation of the alkylated diphenyl amine leads only to a slight formation of fragment ions, e.g. m/z 134.0964 (C9H12N), which is one of the two aromatic rings connected to the nitrogen atom and parts of one alkyl side chain. Hence it seems that the alkylated diphenyl amine additive features a higher stability compared to the other oil constituents with the used fragmentation settings. The pentaerythritol ester (C29H52O8), exemplary of the ester base stock material, shows a characteristic fragmentation pattern. The fragment ion at m/z 399.2737 (C22H39O6) is an ester molecule after the loss of one side chain (C7H13O2), and the fragment ion at m/z 113.0960 (C7H13O) is an aliphatic side chain after ester bond cleavage. We were not able to obtain a clear fragmentation spectrum of N-phenyl-1-naphthylamine in the UFP samples as the detected signal intensity was not sufficient. This could be due to a minor usage of the additive in jet engine lubrication oils and a faster atmospheric degradation rate as it is utilized as a radical scavenger (Wu et al., 2013). Furthermore, the screening of the jet oils showed a matching MS2 pattern of tricresyl phosphate and N-phenyl-1-naphthylamine with the mzcloud database.

Recent studies have reported that tricresyl phosphate exposure provokes detrimental health effects as it exhibits a neurotoxic potential functioning as an acetylcholinesterase inhibitor and also by acting as an endocrine disruptor (Chang et al., 2020; Ji et al., 2020). Different TCP isomers vary regarding their potential health effects, for which reason it is important to gain information about the isomer composition of tricresyl phosphate in UFPs. Therefore, we performed a targeted screening on tricresyl phosphate isomers using selected-ion monitoring (SIM). The tri-ortho isomer shows a peak at 7.82 min, whereas the TCP peak of the UFP sample elutes after 8.06 min, likely an overlapping peak of a mix of different meta- and para-isomers. The tri-ortho-isomer of tricresyl phosphate was not detected in the airport samples. This is in accordance with the reduction in the ortho-isomer fraction of tricresyl phosphate used as an additive for lubrication oils (Winder and Balouet, 2002; de Nola et al., 2008). Other studies have also reported that they were not able to identify the tri-ortho isomer of TCP in various sample types (Solbu et al., 2010, 2011). Although no tri-ortho isomer of TCP was detected, it is still worthwhile to consider that isomers with only one ortho-methyl group feature possibly a higher toxicity than isomers having methyl groups only in meta- and para-position (Hanhela et al., 2005).

We detect further organophosphate compounds like triphenyl phosphate (TPP) in the UFP samples, which likely originate from hydraulic oils. TPP is a high-production-volume chemical used as a flame retardant and likely has further sources other than airport operations. However, Solbu et al. (2010) have reported that ground maintenance of aircraft is also a possible source of organophosphates in the environment.

Besides the common use of pentaerythritol esters as the base stock of lubrication oils, also trimethylolpropane esters (TMPEs) are in use. It is known that a combination of TMPEs with a phosphate additive proportion of only 2 % can result in the formation of the neurotoxin trimethylolpropane phosphate (TMP-P) (Callahan et al., 1989; Centers, 1992). The combination of pentaerythritol esters with phosphate additives does not lead to the formation of TMP-P (Wyman et al., 1987). Studies have shown that the formation of TMP-P occurs possibly with all phosphates, regardless of the structure, at temperatures between 250–750 ∘C. The reaction is probably not highly temperature-dependent due to the low activation energy (Wright, 1996). At the upper temperature limit, the formed TMP-P starts to decompose, implying that the actual yield of TMP-P depends on the balance between formation and degradation reactions (Callahan et al., 1989). TMP-P affects the nervous system by binding to γ-aminobutyric acid (GABA) receptors, especially at the picrotoxinin site, and therefore affects neurotransmission processes (Bowery et al., 1976, 1977; Mattsson, 1980; Simmonds, 1982; Ticku and Ramanjaneyulu, 1984).

In contrast to other studies (van Netten and Leung, 2000; Solbu et al., 2011), we detected TMP-P in the UFP samples, which could be the result of the continued use of TMPE-based lubrication oils. TMP-P was identified via fragmentation leading to a characteristic phosphate fragment. The analysed Eastman Turbo Oil 2380 shows a different pattern regarding the jet engine oil marker compounds compared to the other tested oils (Fig. 4). This oil is not based on a pentaerythritol but on a trimethylolpropane ester formulation. Callahan et al. (1989) identified the Exxon 2380 oil as a source of TMP-P formation, which is possibly the forerunner oil of the Eastman Turbo Oil 2380 tested here. In order to confirm the identified TMP-P, we simulated the thermal decomposition in a temperature experiment using the Eastman Turbo Oil 2380. The jet engine oil was heated to temperatures of 250–450 ∘C in 50 ∘C steps for 4 h. Analysing the oil after the temperature treatment, we were able to detect an emerging signal at m/z 179.0467 ([M + H]+, TMP-P, C6H11O4P) at temperatures starting from 400 ∘C. The appearing compound features the same retention time as detected in the UFP samples, which speaks for the unambiguous identification of TMP-P. To our knowledge, this is the first detection of TMP-P in ambient UFPs. With increasing temperature, we observe the fraction of trimethylolpropane esters decreasing, whereas the fraction of TMP-P rises. The relative fraction of tricresyl phosphate also increases as it is temperature-stable in contrast to the esters. The fraction of the two amines (C16H13N, C28H43N) decreases between 250–350 ∘C, and from 400 ∘C onwards both compounds are not detectable anymore (Fig. 6).