Teflon filters (47 mm, 2.0 µm pores, Tisch Scientific) were used for particle-phase sampling. Filters were collected at 20 Lmin-1 for 8 h during the day (09:00–17:00 LT) and at night (21:00–05:00 LT). Samples were extracted in methanol and analyzed via liquid chromatography (LC) using an Agilent 1260 Infinity LC and an Agilent Poroshell 120 SB-Aq reverse-phase column (2.1×50mm, 2.7 µm particle size). The LC was coupled to an electrospray ionization (ESI) source and a high-resolution mass spectrometer (Agilent 6550 Q-TOF) and operated following previously described methods (Ditto et al., 2018, 2020). The mass resolution (M/ΔM) of the Q-TOF used in this work was ≥25000–40 000, and the mass accuracy was 1–2 ppm. Our use of LC (or gas chromatography – GC) to separate compounds prior to their ionization and detection by the mass spectrometer reduced mass spectral interferences and thus enabled accurate molecular formula assignments beyond what would be possible by relying on the Q-TOF's mass resolution alone.

Filter extracts were run with MS (i.e., TOF-only, to identify molecular formulas) and MS/MS (i.e., tandem mass spectrometry, to identify functional groups) data acquisition, using both positive and negative mode electrospray ionization. These methods are hereafter referred to as “LC-ESI-MS” and “LC-ESI-MS/MS”, respectively. Acquisition and non-targeted analysis methods, including data quality assurance and quality control (QA/QC), are discussed in past work (Ditto et al., 2018, 2020). Briefly, for LC-ESI-MS analyses, any ion mass appearing in both a sample and its corresponding blank (matching ion mass with a tolerance of 5 ppm and matching ion retention time with a tolerance of 0.25 min – both tolerances were chosen to be quite conservative) was removed if its abundance in the sample was less than 5 times its abundance in the blank. Ions with greater sample : blank ratios were retained, and the abundance of the blank peaks was subtracted from the sample peaks. Positive and negative ionization mode data were combined, and any ions appearing in both modes were flagged; abundances were averaged, and the compound was only counted once. Ions from m/z 50 to 600 were assigned formulas assuming hydrogen or sodium adducts in positive mode and acetate adducts or deprotonation in negative mode. We also allowed for the neutral loss of water. Only peaks that well surpassed instrument noise and that had strong peak quality scores (based on both liquid chromatography and mass spectrometry data) were selected for formula identification according to thresholds detailed in Ditto et al. (2018). Formulas were assigned with the following elemental constraints in Agilent's Mass Hunter software, C3-60H4-122O0-20N0-3S0-3, minimizing the parts-per-million mass difference between the observed and proposed ion mass and accounting for isotope distribution. Prior to non-targeted analysis, further QA/QC was performed on these formula identifications using custom R code. As discussed by Kind and Fiehn (2007), the number of elements was further constrained to 39 carbons and 72 hydrogens, and H/C ratios were checked to ensure they fell within expected limits (0.2<H/C<3.1) (Kind and Fiehn, 2007). Formulas were then screened to ensure they agreed with the Nitrogen rule, to ensure that all double bond equivalent values were integers and to flag any large mass differences (>7 ppm) between the observed and proposed mass for a given molecular formula.

For MS/MS, any ions from the LC-ESI-MS analyses that passed these QA/QC steps were targeted for MS/MS fragmentation at 5, 10, 20, 30, and 40 V. We used SIRIUS with CSI:FingerID for functional group identification with a subset of compounds from MS analysis (Dührkop et al., 2015, 2019), as detailed in past work (Ditto et al., 2020). We assumed the same ionization behavior as discussed above, with the same elemental composition constraints and a conservative 7 ppm mass tolerance. Functional groups for the top-scoring candidate structure for each ion were tallied with the APRL Substructure Search Program (Ruggeri and Takahama, 2016). The exact position of each functional group was not considered, as the focus of our work was instead to assess the presence or absence of atmospherically relevant functional groups and their combinations across a large number of multifunctional compounds.

After stringent QA/QC for peak shape and accurate molecular formula determination, non-targeted compound identification from LC-ESI-MS identified an average of 200±56 and 167±47 compounds per sample analyzed in summer and winter, respectively, across 34 samples in summer and 15 in winter.

We note that filter ion abundance data are presented as combined positive and negative ionization mode data, which treat the compounds equally without corrections for ionization efficiency. As mentioned above, compounds were not double counted; any ion appearing in both positive and negative mode was flagged, its average abundance was computed, and it was tallied only once. While ionization efficiency differences between compound types exist, their exact effects for multifunctional compounds present in a complex mixture are uncertain. Thus, similarly to other studies and to our past work, we treat the intercomparison across compounds without adjusting for ionization efficiency differences (Ditto et al., 2018). We note that the figures in the main text are displayed as fractions of total observed ion abundance to consider variations in atmospheric abundance across the complex mixture of functionalized species. However, due to uncertainty in exact ionization efficiency, these are not intended to directly represent mass concentration. For comparison, identical figures represented by occurrence (i.e., unweighted by abundance) are presented in the Supplement (Figs. S4–S6, S8–S9, S11, and S13); general observations remain similar between abundance-weighted and occurrence results.

Gas-phase samples were collected on glass adsorbent tubes (6.35 mm OD, 88.9 mm long) packed with quartz wool, glass beads, Tenax TA, and Carbopack X (Sheu et al., 2018). Samples were collected at 200 mLmin-1 for 2 h and sub-sampled off of the 20 Lmin-1 filter flow during the day (02:00–16:00 LT) and at night (02:00–04:00 LT). Adsorbent tubes were analyzed using a GERSTEL TD3.5+ thermal desorption unit and an Agilent 7890B gas chromatograph (GC) with a DB5-MS UI column (30m×320µm×0.25µm). The GC was coupled to an atmospheric pressure chemical ionization (APCI) source and the same Q-TOF as above, operated with MS (i.e., TOF-only) data acquisition and a positive ionization mode only. These methods are hereafter called “GC-APCI-MS”, and acquisition and analysis methods are discussed in past work (Ditto et al., 2021; Khare et al., 2019). After QA/QC (as detailed in Sect. 2.1.1), this non-targeted analysis yielded an average of 388±201 and 612±133 compounds per sample in summer and winter, respectively, across 34 samples in summer and 14 samples in winter.

Finally, as a supplemental analysis to probe the composition of functionalized gases that were not GC amenable and thus not measured using the adsorbent tube and thermal desorption–gas chromatography techniques mentioned above, we used PEEK-based sample collectors and liquid chromatography to trap and speciate oxygen-, nitrogen-, and/or sulfur-containing gases without thermal desorption. This method was designed to target functionalized gases, which represent important precursors, intermediates, and by-products in the atmospheric processing of emitted organic compounds but which are often challenging to speciate with traditional GC techniques due to their chemical functionality, reactivity, and/or thermal lability. Additionally, in many gas-phase measurement systems, primary emissions (i.e., hydrocarbons) can overwhelm the signal of more functionalized analytes, adding to the challenge of speciating these lower-abundance compounds.

Thus, to probe the chemical composition of these functionalized gases, we used a sampling approach, desorption method, separation method, and ionization technique that leveraged their relatively lower volatility and higher polarity. This included adsorptive sampling onto cooled PEEK tubing followed by direct inline desorption into the LC mobile phase for LC-ESI-MS analysis. ESI was specifically chosen here because it is sensitive to functionalized compounds. Testing was performed in positive and negative ionization mode, but field samples were run in positive mode only. Further details and discussion of this method, including method development and evaluation, can be found in Sect. S1 and Tables S1–S3 in the Supplement. Briefly, PEEK tubing was cooled to 2 ∘C and used as an adsorptive collector, with a Teflon filter positioned upstream of the PEEK tubing to remove particles. PEEK was selected due to its inert behavior, thus reducing the possibility of surface–analyte interactions that might inhibit effective inline solvent desorption and dissolution. PEEK is also compatible with the solvents used in the LC system and is frequently used in LC instruments. Field samples were collected on cooled PEEK tubing during the subsequent winter (5–6 March 2020) for 2 h each between 08:00 and 14:00 LT. For these 2 h (∼2.6 L) field samples, functionalized gases in a typical 100–250 gmol-1 molecular weight range were resolvable at ∼25–60 ppt in the atmosphere, based on instrument detection limits (Ditto et al., 2018). For analysis, each PEEK collector was installed in the LC system flow path, and analytes were directly desorbed using the LC mobile phase solvents and then trapped and focused on the LC column for 20 min before being analyzed using the same LC-ESI-MS system in positive ionization mode (Fig. 1). This inline mobile phase desorption step gently mobilized potentially fragile analytes from the PEEK collector and trapped and focused them on the LC column prior to chromatographic separation and mass spectral analysis. Additionally, this preconcentration step allows for the detection and characterization of lower-concentration species. Data were processed and QA/QC was performed as detailed in Sect. 2.1.1.

We note that there are other existing approaches for offline collection of highly functionalized organic gases and particles that are compatible with LC analysis, such as spray chambers, particle-into-liquid samplers, coated denuders, and polyurethane foam (PUF) sampling. This PEEK sampling method with inline desorption into the LC mobile phase was pursued to reduce sample preparation steps and thus possibilities of losses (e.g., during solvent extraction or evaporative preconcentration) as well as for its direct similarity to the filter-based particle-phase LC-ESI-MS analysis.

We also note that for all filter collection and LC analyses (filters and PEEK collectors), it is possible that some functional groups of interest may have undergone hydrolysis on the filter during 8 h filter collection periods or in the LC mobile phase, which was primarily water at the beginning of the LC solvent gradient. For example, organonitrates may be susceptible to hydrolysis depending on their structure; tertiary organonitrates can undergo hydrolysis on the timescale of minutes–hours depending on pH, while primary/secondary organonitrates are relatively stable. Hydrolysis occurs more quickly at low pH. The pH of the LC mobile phase (pH∼2) and the pH of the sampled aerosol (pH<∼5, Pye et al., 2020) are both acidic; alpha-pinene-derived organonitrates, for example, could have a lifetime of as low as roughly 8 min to 1.5 h across this pH range (Rindelaub et al., 2016). If hydrolysis occurred, some of the observed compounds could be by-products of other functionalized species. While we did not observe any of our nitrogen-containing test standards to hydrolyze over these timescales, standards were not available to reflect every functional group observed in these datasets.

During this period of active photochemistry, the observed distribution of particle-phase compounds in summertime samples spanned the intermediate-volatility (IVOC) to ultra-low-volatility organic compound (ULVOC) range, with a predominance of semivolatile (SVOC), low-volatility (LVOC), and extremely low-volatility organic compounds (ELVOC) as shown in Fig. 2a as a function of compound class. To assess differences in summer and winter volatility distributions, we used individual molecular formulas and the Li et al. (2016) parameterization to estimate the saturation mass concentration (log⁡(C0)) of each observed compound (Li et al., 2016). Compounds were then classified into volatility bins following these definitions: VOC>3×106µgm-3; 3×106µgm-3>IVOC≥300µgm-3; 300µgm-3>SVOC≥0.3µgm-3; 0.3µgm-3>LVOC≥3×10-5µgm-3; 3×10-5µgm-3>ELVOC≥3×10-9µgm-3; 3×10-9µgm-3>ULVOC (Donahue et al., 2011; Schervish and Donahue, 2020).

Due to elevated summertime O3 mixing ratios at the site (shown in Fig. S3, 8 h maximum mixing ratio in summer: 57±20 ppb vs. winter: 46±5 ppb, including day and night sampling periods), O3 may have influenced the photochemical processing of emitted volatile species, especially unsaturated biogenic VOCs which readily undergo ozonolysis due to their chemical structure. However, we did not observe a correlation between 8 h maximum (or 8 h average) O3 mixing ratios with average particle-phase volatility (as saturation mass concentration), carbon number, or O/C (nor did we observe such relationships for gas-phase organic compounds). There were, however, weak relationships between NOx mixing ratios and each of these particle-phase characteristics in the summer. While average NOx mixing ratios were slightly lower during the summer (as shown in Fig. S3, 2.3±1.5 ppb in summer vs. 3.7±2.7 ppb in winter), NOx mixing ratios trended weakly with particle-phase O/C (r∼0.45) and volatility (as saturation mass concentration, r∼0.49) and inversely with carbon number (r∼-0.56) in summer.

While our correlations and conclusions are somewhat limited by the 8 h filter sampling duration and the resulting highly regionally mixed samples, one possible hypothesis is that the presence of NOx could have promoted more fragmentation reactions in the gas phase (Loza et al., 2014) that decreased average carbon number and correspondingly increased volatility and O/C. In fact, we observed highly oxidized C3–C6 compounds in the gas phase (from adsorbent tube measurements with GC-APCI, Sect. S2) that were possibly products of these fragmentation reactions of larger compounds. These trends of NOx mixing ratios with O/C, volatility, and carbon number were not apparent for the observed complex mixture of gas-phase organic compounds. However, these highly oxidized gases may not have persisted in the gas phase and could have been taken up by the condensed/aqueous phase due to their water solubility, where they would have instead contributed to the observed trends of NOx with carbon number, volatility, and O/C in the particle phase. We note that if there was significant uptake of gas-phase NOz to the particle phase, this may have in part contributed to the particle-phase correlations with NOx given that the chemiluminescence NOx analyzer used in this study is known to also respond to gas-phase NOz (Dunlea et al., 2007).

Additionally, NOx could have been involved in heterogeneous chemistry, promoting oxidation and/or nitrogen addition reactions, such as interaction with NO3⚫ to yield organonitrates (Lim et al., 2016), formation and interaction with ambient nitrous acid (HONO) to yield nitrophenols (Vidović et al., 2018), or other pathways.

In the winter, these same relationships between NOx and particle-phase characteristics were not observed. This is possibly due to the decreased role of photochemistry in the winter and the increased role of other competing physical and chemical processes, such as aqueous-phase chemistry. In the discussion of our results, we note that aqueous-phase chemistry is meant to be inclusive of aqueous processing in aerosols, in cloud water, and/or in fog water, all of which may have occurred upwind of the site during the 8 h sampling periods under variable local and regional weather conditions.

In the winter, we observed a generally higher average saturation mass concentration (summer: log⁡(C0)=-3.7±3.9µgm-3 vs. winter: log⁡(C0)=-0.7±4.0µgm-3). We note that this comparison of saturation mass concentrations was performed at a reference temperature of 300 K, and we discuss the expected wintertime volatility shift below. The wintertime O/C was also slightly lower than summer (summer: O/C=0.5±0.4 vs. winter: O/C=0.4±0.4). In the winter, the observed chemical composition of the particle phase – in terms of both volatility and functional group distribution – suggests a relatively greater role for aqueous-phase processing. Our observations were similar to those made in past studies of higher-volatility products from fragmentation reactions in the aqueous phase, e.g., Brege et al. (2018), where they observed that aged fog water samples contained organic compounds with smaller carbon backbone structures than non-aqueous aged particles and linked this difference to aqueous-phase fragmentation reactions, the uptake of smaller water-soluble gases to the aqueous phase, and/or less oligomerization (Brege et al., 2018). Also, Yu et al. (2016), discussed the role of fragmentation in aging aqueous phenolic secondary organic aerosol (Yu et al., 2016), and Schurman et al. (2018) discussed the role of fragmentation and evaporation in cloud water (Schurman et al., 2018). Similarly, here we observed a shifted compound distribution that included smaller molecular weight and generally higher-volatility particle-phase species in winter compared to summer along with notably different functional group distributions, both of which could be attributed to aqueous chemistry.

We note that, for direct comparison, volatility bins in Fig. 2a and b were defined for the same reference temperature (i.e., 300 K, the average summertime sampling period temperature), though wintertime saturation mass concentrations for the observed compounds would shift approximately 2 orders of magnitude lower due to lower temperatures (i.e., 270 K). The dotted black line in Fig. 2b shows the shift in bins expected at 270 K. In the winter, compounds defining IVOCs or SVOCs at 300 K will expectedly exhibit a greater degree of partitioning to the particle phase, though the effect of this temperature shift on partitioning was likely more pronounced for the SVOCs than the IVOCs (Table S4 in the Supplement). Even when accounting for this shift, the mean saturation mass concentration of wintertime samples was log⁡(C0)=-2.7±3.9µgm-3, which is still higher than the mean summertime saturation mass concentration of log⁡(C0)=3.7±3.9µgm-3 and thus still demonstrates a volatility difference between summer and winter, with higher-volatility species in winter. This shift is also reflected in the carbon number distribution observed via the LC-ESI-MS/MS analysis of this sample set shown in Ditto et al. (2020), Fig. S5. In addition to this shift in molecular size and volatility, there was a distinct change in functional group composition from summer to winter discussed below.

To assess the potential contribution of aqueous-phase chemistry, we also estimated aerosol liquid water concentrations based on available data in Sect. S2.1 in the Supplement. We estimated a lower but still appreciable aerosol liquid water content in winter relative to summer, but with fewer photochemical processes in winter along with generally cloudier/foggier local weather (i.e., 44 % of summer sampling periods with partly cloudy or cloudy weather conditions vs. 67 % of winter sampling periods, from the Weather Underground archive), aqueous-phase processing likely remains an important pathway. We note that the compounds discussed here could have been formed locally or regionally, and thus the role of conditions at the site (aerosol liquid water, cloud cover, fog cover) is just as important as the conditions in the surrounding upwind region. As a result, it is challenging to pinpoint the exact contributions of aerosol liquid water, in-cloud, or in-fog processing, and we consider that all three may be occurring upwind or near the site.

Furthermore, from MS/MS analysis, we observed functional groups that were possible indicators of aqueous-phase processing, including the presence of nitrophenols during the winter, which may have formed via dark aqueous-phase reactions with HONO (Vidović et al., 2018), and relatively low contributions from carbonyls across seasons, possibly linked to carbonyl hydrolysis (Ditto et al., 2020). Based on laboratory studies, the presence of azole functional groups and other heterocyclic nitrogen species could also indicate aqueous-phase processing and may be formed from small carbonyl precursors such as glyoxal (DeHaan et al., 2009; Grace et al., 2019) and biacetyl (Grace et al., 2020) reacting with atmospheric ammonia or small amines. Many of the N-only-containing azoles observed here had similar substructures to those formed in the aqueous-phase reactions of small carbonyls with ammonia/amines (DeHaan et al., 2009; Grace et al., 2019). In addition, as discussed above, we observed many small gas-phase C3–C6 compounds at the site in the summer, which likely included multifunctional isoprene oxidation products (e.g., glycolaldehyde, hydroxyacetone, and isomers); these potential precursors could have reacted with atmospheric ammonia or species containing amino groups to form the observed azole-containing reaction products. We observed more azoles during the summer (Ditto et al., 2020), perhaps due to the increased prevalence of the C3–C6 precursors and overall prominence of atmospheric water (e.g., aerosol liquid water, cloud water, or fog water).

Lastly, the role of aqueous-phase chemistry in the region is further supported by prior summertime observations at Brookhaven National Laboratory (on the opposite side of the Long Island Sound), which examined a low-volatility oxygenated organic aerosol factor in the source apportionment of aerosol mass spectrometry measurements and showed a strong contribution from carboxylic acids and other ELVOCs that were attributed to aqueous-phase processing (Zhou et al., 2016).

The distribution of compound classes observed at the YCFS was significantly different from observations at a range of field sites discussed in past studies (Fig. 3), including a remote forested site (i.e., the PROPHET site in northern Michigan), an urban inland site (i.e., near downtown Atlanta) across two seasons, and New York City (Ditto et al., 2018, 2019). We perform a direct comparison to our past studies here, because the same sampling and analytical methods were used, and thus we can compare the distribution of ions observed without any biases due to differences and uncertainties resulting from variations in sampling or ionization chemistry between instruments. While a more detailed site-to-site comparison is outside the scope of this work, the proximity of the YCFS to the ocean and thus the impact of marine emissions and over-water chemistry likely contributed to the differences between the YCFS and inland locations. In particular, at the YCFS, we observed notably smaller relative contributions from compounds containing carbon, hydrogen, and oxygen (i.e., CHO, 11 %–16 % of observed functionalized compounds here vs. 34 %–50 % at other sites), and the contributions from nitrogen-containing particle-phase compounds at the YCFS were in stark contrast to other sites. Here, 85 % of compounds (by ion abundance) in summer and 68 % of compounds in winter contained at least one nitrogen atom compared to 38 %–51 % at the other previously studied sites (Fig. 3). These nitrogen-containing species were comprised of compounds with various reduced and oxidized nitrogen-containing functional groups with varying oxygen-to-nitrogen ratios (O/N), which are broadly classified and discussed below as compounds containing carbon, hydrogen, and nitrogen (i.e., CHN) and compounds containing carbon, hydrogen, oxygen, and nitrogen (i.e., CHON with O/N ratio <3, and CHON, O/N ratio ≥3). There were notably greater contributions at the site from nitrogen-containing compounds that also contained at least one oxygen atom, including CHON compounds with O/N<3 (19 %–20 % here vs. 10 %–15 % at other sites), CHON compounds with O/N≥3 (24 %–44 % here vs. 14 %–19 % at other sites), as well as compounds containing oxygen, nitrogen, and sulfur (i.e., CHONS, 20 %–21 % here vs. 9 %–10 % at other sites) (Ditto et al., 2018).

We note that while these measurements were of PM10 aerosols, the observations of high nitrogen content were not biased by the inclusion of larger, primary (possibly biological) particles. Quartz filter samples collected with a cascade impactor at the site during the summer and analyzed with the same LC-ESI-MS methods did not show any significant differences between any of these nitrogen-containing compound classes as a function of particle size, across particles ranging from 0.4 to 10 µm (i.e., 69 %–71 % of ion abundance for PM≤2.2 µm and 69 %–73 % of ion abundance for PM ranging from 2.2 to 10 µm were nitrogen-containing species, summarized in Table S5 in the Supplement). This is consistent with past studies which have demonstrated that amines, as one example of a prominent nitrogen-containing functional group, are ubiquitous in size-resolved aerosol samples in urban and rural locations (VandenBoer et al., 2011).

The prevalence of nitrogen-containing species at the YCFS is consistent with the study at Brookhaven National Laboratory discussed above, where a dedicated nitrogen-enriched aerosol mass spectrometry factor was identified and contained prevalent signals from aliphatic amines and amides. However, in the Brookhaven study, the nitrogen-enriched factor was associated with industrial amine emissions that were enhanced during periods of southern/southwestern backward trajectory influence and that had correlations with tracers linked to industrial processes. In our study, there was no correlation between backward trajectory direction and the contribution of nitrogen-containing species. Also, wintertime air parcels arrived predominantly from directions other than south/southwest, suggesting that the nitrogen-containing species observed in our study were the result of mixed anthropogenic, biogenic, and marine precursors and their transformation products. This high nitrogen content at the YCFS, where aqueous-phase chemistry is expected to be important, is also consistent with the cloud water composition discussed in Zhao et al. (2013), which reported roughly 65 % of detected ions in their cloud water samples as containing a nitrogen atom, and the primary marine aerosol composition discussed in Wozniak et al. (2014), where 61 % of their observed compounds contained nitrogen and 54 % were CHON species.

While nitrogen-containing compounds in general were very prominent at the site (Fig. 3a and b), CHN compounds were relatively less abundant in these samples of functionalized organic aerosol. Particle-phase CHN compounds represented just 1 % and 3 % of observed functionalized organic aerosol abundance in summer and winter, respectively, which was similar to observations at other ambient sites (∼5 % CHN) (Ditto et al., 2018).

In the summertime LC-ESI-MS/MS measurements, CHN particle-phase compounds were comprised primarily of amines (72 % of CHN species contained an amine group) and nitriles (28 % of CHN species contained a nitrile group), as shown in Fig. 4. In the winter, these compounds were nearly exclusively amines (present in 99 % of CHN species). Amines have many primary land-based sources (e.g., biogenic emissions, Kieloaho et al., 2013; agricultural activity, Ge et al., 2011; emissions from decomposing organic matter, Ge et al., 2011, Sintermann and Neftel, 2015; biomass burning, Ge et al., 2011; emissions from port activity, Gaston et al., 2013; chemical products, Khare and Gentner, 2018; vehicle exhaust, Sodeman et al., 2005), but their presence on the coast could also indicate marine contributions. Amines have been detected in bulk ocean water, the surface microlayer, and sea spray aerosol, and their emissions and chemical transformations in the marine environment have been the topic of many recent studies (e.g., Brean et al., 2021; Dall'Osto et al., 2019; Decesari et al., 2020; Di Lorenzo et al., 2018; van Pinxteren et al., 2012, 2019; Quinn et al., 2015; Wu et al., 2020). In the summer, biogenic and marine sources likely dominated the amine distribution, while in the winter, anthropogenic amine sources likely became more important.

Recent studies have also evaluated amine phase partitioning or formation in cloud/fog water (e.g., Chen et al., 2018; Youn et al., 2015) as well as condensed-phase or aqueous-phase pathways that may transform emitted amines (e.g., Ge et al., 2016; Lim et al., 2019; Tao et al., 2021). Interestingly, the observed amines at this site, as well as other reduced nitrogen groups like nitriles, imines, and enamines, were not present exclusively in CHN species and thus were a mix of both direct emissions and chemically processed compounds. Reduced nitrogen groups were often paired with hydroxyl groups, carboxylic acids, carbonyls, ethers, and esters as part of nitrogen- and oxygen-containing compounds with a range of O/N ratios. This is consistent with other studies observing reduced nitrogen contributions to CHON compound classes, such as Zhao et al. (2013), LeClair et al. (2012), and Altieri et al. (2009) discussed above. As such, we discuss CHON species as a function of O/N ratio to focus on differences between less-oxygenated (O/N<3) and more-oxygenated (O/N≥3, e.g., organonitrates) species, using a ratio of 3 to distinguish between the two as informed by the O/N ratio of the organonitrate functional group.

CHON (O/N<3) compounds were notably more important at this site than other sites, representing 20 % and 19 % of observed functionalized organic aerosol abundance in summer and winter, respectively (Fig. 3a and b), compared to ∼13 % at other sites (from predominantly summer measurements). These CHON compounds included some functional groups that contained both oxygen and nitrogen, such as amide groups (12 % of this compound class's nitrogen content in summer vs. 1 % in winter, Fig. 4) and nitro groups (15 % of this nitrogen content in summer vs. 6 % in winter, Fig. 4). However, most CHON (O/N<3) compounds were comprised of a combination of nitrogen- or oxygen-containing groups rather than a functional group containing both nitrogen and oxygen. This included large contributions from hydroxyls and ethers across both seasons as well as important contributions from amines, isocyanates, and heterocyclic nitrogen, as shown in Ditto et al. (2020) (Fig. 1). The presence of these functional groups in the winter could be indicative of wood burning emissions in the region, which has been observed in the wintertime in past ambient sampling in the northeastern US (Sullivan et al., 2019). Isocyanates contributed notably to this compound class during the winter, which could similarly be linked to burning wood, other biomass, or building materials (Leslie et al., 2019; Priestley et al., 2018; Roberts et al., 2014) or could be photochemically produced via the oxidation of amines and amides (Borduas et al., 2015; Leslie et al., 2019). Importantly, levoglucosan, a common biomass burning tracer, was observed across nearly all daytime and nighttime winter particle-phase samples (verified with an authentic standard), supporting the influence of biomass burning compounds at the site. Together, the overall high prevalence of reduced nitrogen at this site could be influenced by the mixing of aged biomass burning plumes with marine air, which is consistent with past observations of very high alkylamine concentrations in biomass burning particles that mixed with marine air prior to sampling (Di Lorenzo et al., 2018).

CHON (O/N≥3) compounds were the dominant compound class in the observed summertime distribution and played an important role in the wintertime distribution as well, comprising 44 % of observed functionalized organic aerosol abundance in summer vs. 24 % in winter (Fig. 3a and b). These contributions were far greater than the contributions of CHON (O/N≥3) species at other sites, which typically ranged from 14 % to 19 % (predominantly from summertime measurement, Fig. 3c).

Similarly to CHON (O/N<3), we observed some CHON (O/N≥3) compounds with functional groups containing three oxygen atoms and one nitrogen atom, e.g., nitrophenols and organonitrates (Fig. 4), but also contributions from nitrogen-only functional groups paired with oxygen-containing groups. Notably, in the summer, there were important contributions from amines (47 % of this compound class's nitrogen content), imines (19 %), organonitrates (10 %), and azoles (16 %) (Fig. 4). In contrast, in the winter, nitrogen content in the CHON (O/N≥3) compound class was dominated by IVOC/SVOC nitrophenols, comprising 64 % of the CHON (O/N≥3) ion abundance.

NOx mixing ratios were typically low in both summer and winter (2.3±1.5 ppb in summer vs. 3.7±2.7 ppb in winter) but were slightly higher during winter. In the winter, CHON (O/N≥3) compounds showed a weak positive relationship with NOx mixing ratios (r∼0.58) and a stronger correlation with NO mixing ratios (r∼0.81). This relationship between CHON (O/N≥3) and NO (and NOx) suggests that many of these oxidized nitrogen species were products of NOx-related chemistry (i.e., NOz compounds). The enhancement in nitrophenols serves as one example of this, as NO mixing ratios also correlated with the contribution of nitrophenols in the winter (r∼0.69).

In past work, we discussed nitrophenol nighttime enhancements during winter and noted their reported aqueous formation pathways mentioned in prior laboratory studies (Ditto et al., 2020). Here, we demonstrate that nitrophenols were important contributors to the CHON (O/N≥3) compound class and highlight their role as examples of NOz due to their possible formation via dark aqueous-phase nitration pathways of oxygenated aromatics with HONO (Vidović et al., 2018). While nitrophenols may have other sources (e.g., diesel exhaust), our observations of a clear nighttime enhancement during the winter suggest that these functional groups were most likely formed by secondary chemistry related to NOx oxidation, as this field site was removed from major roadways. Our wintertime observations suggest that HONO could have been derived from local wood burning and could have reacted away as the smoke plume aged to form stable products like nitrophenols, similarly to HONO transformation chemistry into other forms of oxidized nitrogen (e.g., particulate nitrates, PANs, organic nitrates) that has recently been observed in wildfire smoke (Juncosa Calahorrano et al., 2021).

Furthermore, the correlation between NO and CHON (O/N≥3) could also be influenced by the daytime formation of organonitrates via reaction with OH⚫ and NO (i.e., RO2⚫+NO) (Liebmann et al., 2019; Ng et al., 2017; Perring et al., 2013; Takeuchi and Ng, 2018), though organonitrates contributed a smaller fraction of CHON (O/N≥3) species (i.e., 10 % of this compound class's nitrogen content across seasons).

In the summer and winter, contributions from reduced nitrogen groups (e.g., groups shown in black/grey in Fig. 4) rivaled those of oxidized nitrogen groups in CHON compounds across a range of O/N ratios. In the summer, reduced nitrogen groups contributed 50 % of all detected CHON (O/N<3) compounds by ion abundance, while in the winter they contributed 47 % (Fig. 4). For CHON compounds with O/N≥3, reduced nitrogen groups contributed 68 % of compound ion abundance in the summer (possibly related to marine influences, Wozniak et al., 2014), while in the winter they contributed just 13 %. Interestingly, 90 % of the dominant reduced nitrogen functional groups observed (amines and imines) were present in acyclic rather than cyclic structures, which may have been the result of either direct emissions or formation via reactions with ammonia or other small amines.

In contrast, possible NOz products (e.g., groups shown in blue in Fig. 4) were present in 18 % and 7 % of CHON (O/N<3) compounds in the summer and winter, respectively. For CHON (O/N≥3) compounds, they were present in 18 % and 86 % in the summer and winter, respectively, with the latter wintertime increase in oxidized nitrogen groups largely driven by the presence of nitrophenols at night (Ditto et al., 2020). The remaining fraction of nitrogen-containing groups also contained oxygen but with a reduced nitrogen atom (e.g., amide, isocyanate, nitrogen-/oxygen-containing azole; shown in brown in Fig. 4). We note that CHONS compounds also represented a sizable fraction of observed organic nitrogen (Fig. 3) and contained a mix of reduced and oxidized functional groups (Sect. S3 and Figs. S7–S8 in the Supplement).

The importance of reduced nitrogen functional groups in CHON compounds highlights that not all oxygen- and nitrogen-containing species in the CHON (O/N≥3) compound class were NOz, despite their apparent molecular formulas and the observed correlation observed between CHON (O/N≥3) species with NO and NOx mixing ratios. For instance, many of the observed reduced nitrogen-containing functional groups co-occurred with several oxygen-containing groups like hydroxyls, carboxylic acids, esters, ethers, and carbonyls and thus had molecular formulas with O/N≥3, which could incorrectly be assumed to be an organonitrate or similar structure based on molecular formula alone.

We note that the relative distribution of reduced and oxidized nitrogen-containing groups shown here is subject to sampling and ionization conditions. While the electrospray ionization source used for the particle-phase analysis discussed here effectively ionized these nitrogen-containing groups, their relative sensitivity may differ because many of these functional groups were present in multifunctional compounds whose other features may also contribute to ionization behavior. Also, other aspects of the sample collection and extraction process could cause variability in the observed signal (e.g., PM size cut, organonitrate stability over long-duration samples). Thus, we emphasize that the observed relative abundances here are valuable because they suggest that fully reduced nitrogen-containing groups are important contributors to multifunctional CHON species, but their exact mass contributions remain uncertain.