The Observatoire Pérenne de l'Environnement (OPE) is a continental rural background observatory located about 230 km east of Paris at an altitude of 392 m (Fig. 1). This French critical zone observatory (CZO) is part of a long-term multidisciplinary project monitoring the state of environmental variables, including among others fluxes, abiotic and biotic variables, and their functions and dynamics (http://ope.andra.fr/index.php?lang=en, last access: 10 December 2019). It is largely impacted by agricultural activities. It is also characterized by a low population density (less than 22 per square kilometer within an area of 900 km2) with no industrial activities or surrounding major transport roads. The air monitoring site itself lies in a “reference sector” of 240 km2 in the middle of a field crop area (tens of kilometers in all directions). This reference sector is composed of vast farmlands interspersed with wooded areas. The area is further defined by a homogeneous soil type with a predominantly superficial clay–limestone composition. The daily agricultural practices and meteorological data (including wind speed and direction, temperature, rainfall level, and relative humidity) within the reference sector are recorded and made available by ANDRA (Agence nationale pour la gestion des déchets radioactifs). The agricultural fields of the area are generally submitted to a 3-year crop-rotation system. The major crops during the campaign period were pea and oilseed rape.

An intensive field campaign was conducted at this site for the sake of the present study. The aerosol sampling campaign period lasted from 12 June to 21 August 2017, covering the summer period in France. During this period, ambient PM10 was collected daily (from 09:00 UTC to 09:00 UTC the next day) onto prebaked quartz fiber filters (Pall Tissuquartz 2500QAT-UP, Ø=150 mm) using high-volume samplers (Aerosol Sampler DHA-80, DIGITEL; 24 h at 30 m3 h-1). After collection, all filter samples were wrapped in aluminum foil, sealed in zipper plastic bags, and stored at <4 ∘C until further analysis. More details on the preparation, storage, and handling of these filter samples can be found in Samaké et al. (2019b). A total of 69 samples and 6 field blanks were collected.

Surface soil samples (0–5 cm depth, 15 cm×15 cm area) were simultaneously collected from two fields within the pea and oilseed rape growing areas. The fields are located in the immediate vicinity of the PM10 sampler and under the prevailing wind directions (Fig. 1). To represent as closely as possible the local soil microbial communities, we randomly collected five subsamples (about 100 g per sampling unit) within each parcel and pooled them. Topsoil sampling took place on a weekly basis during the campaign period. After collection and homogenization, 15 g of each subsample was stored in airtight containers (sterile bottles, Schott, GL45, 100 mL) containing the same weight of sterile silica gel (around 15 g). Such a soil desiccation method is a straightforward approach to prevent any microbial growth and change in community over time at room temperature (Taberlet et al., 2018). A total of eight topsoil samples were collected for each parcel.

Finally, leaf samples were collected from the major types of vegetation within the reference sector. These include leaves of oilseed rape, pea, oak, maple, beech, and herbs (Fig. 1). A total of eight leaf samples were analyzed. These samples were also stored in airtight containers (sterile bottles, Schott, GL45, 100 mL) containing 15 g of silica gel. It should be noted that leaf samples were collected only once, 4 weeks after the end of PM and soil sampling, while the major crops were still on site.

Daily PM10 samples were analyzed for various chemical species using subsampled fractions of the collection filters and a large array of analytical methods. Detailed information on all the chemical-analysis procedures has been reported previously (Golly et al., 2018; Samaké et al., 2019b; Waked et al., 2014). Briefly, SCs (i.e., polyols and saccharides) and water-soluble ions (including Ca2+) were systematically analyzed in all samples, using respectively high-performance liquid chromatography with pulsed amperometric detection (HPLC-PAD) and ionic chromatography (IC; Thermo Fisher ICS 3000, USA). Free-cellulose concentrations were determined using an optimized enzymatic hydrolysis (Samaké et al., 2019a) and the subsequent analysis method of the resultant glucose units using HPLC-PAD (Golly et al., 2018; Samaké et al., 2019b; Waked et al., 2014). Organic and elemental carbon (OC, EC) were analyzed using a Sunset thermal-optic instrument and the EUSAAR2 protocol (Cavalli et al., 2010). This analytical method requires a high temperature, thereby constraining the choice of quartz as the sampling filter material. OM content in PM10 samples was then estimated using an OM-to-OC conversion factor of 1.8: OM=1.8×OC (Samaké et al., 2019a, b). This value of 1.8 for the OM/OC ratio was chosen on the basis of previous studies carried out in France (Samaké et al., 2019b, and references therein)

Aerosol samples typically contain very low DNA concentrations, and the DNA-binding properties of the quartz fibers of aerosol collection filters make its extraction with traditional protocols challenging (Dommergue et al., 2019; Jiang et al., 2015; Luhung et al., 2015). In the present study, we were also constrained by the limited daily collection filter surface for the simultaneous chemical and microbiological analyses of the same filters. To circumvent issues of low efficiency during genomic DNA extraction, several technical improvements have been made to optimize the extraction of high-quality DNA from PM10 samples (Dommergue et al., 2019; Jiang et al., 2015; Luhung et al., 2015). This includes thermal water bath sonication helping the lysis process in thick cell walls (e.g., fungal spores and Gram-positive bacteria), which might not be effectively lysed solely by means of bead beating (Luhung et al., 2015). Some consecutive (2 d at maximum) quartz filter samples with low OM concentrations were also pooled when necessary. Detailed information regarding the resultant composite samples (labeled A1 to A36) is presented in Table S1 in the Supplement. Figure S1 in the Supplement presents the average concentration levels of SC species in each sample. The results clearly show that air samples can be categorized from low (background, from A1 to A4 and A21 to A36) to high (peak, from A5 to A20) PM10-SC concentration levels.

In terms of DNA extraction, one-quarter (about 38.5 cm2) of each filter sample was used. First, filter aliquots were aseptically inserted into individual 50 mL Falcon tubes filed with sterilized saturated phosphate buffer (Na2HPO4, NaH2PO4, 0.12 M; pH≈8). PM10 was desorbed from the filter samples by gentle shaking for 10 min at 250 rpm. This pretreatment allows the separation of the collected particles from the quartz filters thanks to the high competing interaction between the saturated phosphate buffer and the charged biological materials (Jiang et al., 2015; Taberlet et al., 2018). After gentle vortex mixing, the subsequent resuspension was filtered with a polyethersulfone membrane disk filter (Supor^® PES 47 mm 200, 0.2 µm, Pall, USA). We repeated this desorbing step three times to enhance the recovery of the biological material from the quartz filters. Each collection PES membrane was then shredded into small pieces and used for DNA extractions using the DNeasy PowerWater kit (Qiagen, Germantown, MD, USA). The standard protocol of the supplier was followed with only minor modifications: 30 min of thermal water bath sonication at 65 ∘C (EMAG, Emmi-60 HC, Germany; 50 % efficiency) and 5 min of bead beating before and after sonication were added. Finally, DNA was eluted in 50 µL of EB buffer. Such an optimized protocol has been recently shown to produce a 10-fold increase in DNA extraction efficiency (Dommergue et al., 2019; Luhung et al., 2015), thereby allowing high-throughput sequencing of air samples. Note that all the steps mentioned above were performed under laminar flow hoods and that materials (filter funnels, forceps, and scissors) were sterilized prior to use.

Unless specified otherwise, all exploratory statistical analyses were achieved with R. Rarefaction and extrapolation curves were obtained with the iNEXT 2.0-12 package (Hsieh et al., 2016) to investigate the gain in species richness as we increased the sequencing depth for each sample. Alpha diversity estimators including Shannon and Chao1 were calculated with the phyloseq 1.22-3 package (McMurdie and Holmes, 2013) on data rarefied to the same sequencing depth per sample type (see Table S2 for details on the rarefaction depths). Nonmetric multidimensional scaling (NMDS) ordination analysis was performed to decipher the temporal patterns in airborne microbial community structures (phylum or class taxonomic group) in air samples. These analyses were achieved with the metaMDS function within the vegan package (Oksanen et al., 2019) with the number of random starts set to 500. The NMDS ordinations were obtained using pairwise dissimilarity matrices based on Bray–Curtis index. The envfit function implemented in vegan was used to assess the airborne microbial communities, which could explain the temporal dynamics of ambient SC species concentrations. Pairwise analysis of similarity (ANOSIM) was performed to assess similarity between groups of PM10 aerosol samples. This was achieved using the anosim function of vegan (Oksanen et al., 2019) with the number of permutations set to 999. Spearman's rank correlation analysis was used to investigate further the relationship between airborne microbial communities and SC species.

To gain further insight into the dominant source of SC-associated microbial communities, NMDS analysis based on Horn distance was performed to compare the microbial community composition similarities between PM10 aerosol, soil, and leaf samples.

Temporal dynamics of daily PM10 carbonaceous components (e.g., primary sugar compounds, cellulose, and OM) are presented in Fig. 2. Nine SCs including seven polyols and two saccharide compounds have been quantified in all ambient PM10 collected at the study site. Ambient SC concentration levels peaked on 8 August 2017 in excellent agreement with the daily harvest activities around the study site (Fig. 2a). The average concentration (average ± SD) of total SCs during the campaign is 259.8±253.8 ng m-3 with a range of 26.6 to 1679.5 ng m-3, contributing on average to 5.7±3.2 % of total OM mass in PM10 with a range of 0.8–13.5 % (Fig. 2b). The total measured polyols present an average concentration of 26.3±54.4 ng m-3. Among all the measured polyols, arabitol (67.4±83.1 ng m-3) and mannitol (68.1±75.3 ng m-3) are the predominant species, followed by lesser amounts of sorbitol (10.9±7.6 ng m-3), erythritol (7.0±8.8 ng m-3), inositol (2.3±2.0 ng m-3), and xylitol (2.3±3.0 ng m-3). Glycerol was also observed in our samples but with concentrations frequently below the quantification limit. The average concentration of saccharide compounds is 51.2±45.0 ng m-3. Trehalose (55.8±51.9 ng m-3) is the most abundant saccharide species, followed by glucose (46.9±37.1 ng m-3). The average concentration of calcium is 251.1±248.4 ng m-3.

A Spearman's rank correlation analysis based on the daily dynamics was used to examine the relationships between SC species. As shown in Table 1, sorbitol and inositol are well-correlated linearly (R=0.57, p<0.001). Herein, sorbitol (R=0.59, p<0.001) and inositol (R=0.64, p<0.001) are significantly correlated to Ca2+. It can also be noted that all other SC species are highly correlated with each other (p<0.001) and that they are weakly correlated to the temporal dynamics of sorbitol and inositol (Table 1).

The structures of bacterial and fungal communities were generated for the 62 collected samples, consisting of 36 aerosol, 18 surface soil, and 8 leaf samples. After paired-end assembly of sequence reads, sample assignment, filtering based on sequence length and quality, and discarding rare sequences, we are left with 2 575 857 and 1 647 000 reads respectively for fungi and bacteria, corresponding to 4762 and 5852 unique sequences, respectively. After the clustering of high-quality sequences, potential contaminants, and chimeras, the final data sets (all samples pooled) consist respectively of 597 and 944 MOTUs for fungi and bacteria, with 1 959 549 and 901 539 reads. The average number of reads (average ± SE) per sample is 31607±2072 and 14563±1221 for fungi and bacteria, respectively. The rarefaction curves of MOTU diversity showed common logarithmic shapes approaching a plateau in all cases (Fig. S2). This indicates an overall sufficient sequencing depth to capture the diversity of sequences occurring in the different types of samples. To compare the microbial community diversity and species richness, data normalization was performed by randomly selecting from each sample 4287 fungal sequence reads and 2 865 bacterial sequence reads. The Chao1 values of fungi are higher for aerosol samples than for soil and leaf samples (p<0.05), indicating higher richness in airborne PM10 (Fig. S3a). In contrast, PM10 and soil samples showed higher values of Shannon index (p<0.05), indicating a higher fungal diversity in these ecosystems. The soil harbors higher bacterial richness and diversity than PM10 (p<0.05), which in turn harbors greater richness and diversity compared to leaf samples (p<0.05) (Fig. S3b).

The NMDS (nonmetric multidimensional scaling) ordination exploring the temporal dynamics of microbial community beta diversity among all PM10 aerosol samples revealed significant temporal shifts in community structure for both fungi and bacteria (Fig. 4).

An NMDS (two dimensions, stress = 0.15) based on fungal class-level compositions (Fig. 4a) results in three distinct clusters of PM10 samples. With one exception (A23), all air samples with higher SC concentration levels (A5 to A20, see Table S2 and Fig. S1) are clustered together and are distinct from those with background levels of atmospheric SC concentrations. This pattern is further confirmed with the analysis of similarity, which shows a significant separation of clusters of samples (ANOSIM, R=0.31, p<0.01). As evidenced in Fig. 4a, this difference is mainly explained by the NMDS1 axis, which results from the predominance of only a few class-level fungi in PM10 samples, including Dothideomycetes, Tremellomycetes, Microbotryomycetes, and Exobasidiomycetes. Vector fitting of chemical time series data to the NMDS ordination plot indicates that the latter four fungal community assemblages best correlate with individual SC species. Mannitol (R2=0.37, p<0.01), arabitol (R2=0.36, p<0.01), trehalose (R2=0.41, p<0.01), glucose (R2=0.33, p<0.01), xylitol (R2=0.45, p<0.01), erythritol (R2=0.40, p<0.01), and inositol (R2=0.24, p=0.01) are significantly positively correlated to the fungal assemblage ordination solution.

For bacterial phylum-level compositions (Fig. 4b), an NMDS ordination (two dimensions, stress = 0.07) analysis differentiates the PM10 samples into two distinct clusters according to their SC concentration levels. All air samples with higher SC concentration levels except two (A23 and A24) are clustered separately from those with ambient background concentration levels. ANOSIM analysis (R=0.69, p<0.01) further confirms the significant difference between the two clusters of samples. Proteobacteria constitute the most dominant bacterial phylum during the SC peak over the sampling period. Interestingly, changes in individual SC profiles are significantly correlated with bacterial community temporal shifts (Fig. 4b). Mannitol (R2=0.25, p<0.01), arabitol (R2=0.24, p<0.01), trehalose (R2=0.32, p<0.01), glucose (R2=0.32, p<0.01), xylitol (R2=0.38, p<0.01), and erythritol (R2=0.27, p<0.01) are mainly positively correlated to the bacterial community dissimilarity.

Given the distinct clustering patterns of airborne PM10 microbial beta diversity structures according to SC concentration levels, a Pearson's rank correlation analysis has been performed to further examine the relationship between individual SC profiles and airborne microbial community abundance at phylum and class levels. This analysis reveals that for class-level fungi, the abundances of Dothideomycetes, Tremellomycetes, and Microbotryomycetes are highly positively correlated (p<0.05) to the temporal evolutions of the individual SC species concentration levels (Fig. S5a). Likewise, ambient SC species concentration levels are significantly correlated (p<0.05) to the Proteobacteria phylum (Fig. S5b). To gain further insight into the airborne microbial fingerprints associated with ambient SC species, correlation analyses were also performed at a finer taxonomic level. These analyses show that the temporal dynamics of SC species primarily correlates best (p<0.05) with the Cladosporium, Alternaria, Sporobolomyces, and Dioszegia fungal genera (Fig. 5a). Similarly, the time series of SC species are primarily positively correlated (p<0.05) with Massilia, Pseudomonas, Frigoribacterium, and to a lesser degree (nonsignificant) with the Sphingomonas bacterial genus (Fig. 5b).

As shown in Fig. 6, the airborne microbial genera most positively correlated with SC species are also distributed in the surrounding environmental samples of surface soils and leaves. In addition, microbial taxa of PM10 associated with SC species are generally more abundant in the leaf than in the topsoil samples (Fig. 6). In order to further explore and visualize the similarity of species compositions across local environment types, we conducted an NMDS ordination analysis (Fig. 7). As evidenced in Fig. 7, the beta diversities of fungal and bacterial MOTUs are more similar within the same habitat (PM10, plant, or soil) and are grouped across habitats as expected. Interestingly, the beta diversities of fungal and bacterial MOTUs in leaf samples and those in airborne PM10 are generally not readily distinguishable, with similarity becoming more prominent during atmospheric peaks of SC concentration levels (Fig. 6). However, the overall beta diversities in airborne PM10 and in leaf samples are significantly different from those from topsoil samples (ANOSIM, R=0.89 and 0.80 and p<0.01 for fungal and bacterial communities, respectively), without any overlap regardless of whether or not harvesting activities are performed around the sampling site.

This observation is also confirmed by an unsupervised hierarchical cluster analysis, which reveals a pattern similar to that observed in the NMDS ordination, where taxa from leaf samples and airborne PM10 are clustered together regardless of whether ambient concentration levels of SC peaked or not. They are clustered separately from those of topsoil samples (Fig. S7).

SC species have recently been reported to be ubiquitous in PM10 collected in several areas in France (Golly et al., 2018; Samaké et al., 2019b). In this study, the total SCs presented an average concentration of 259.8±253.8 ng m-3 with a range of 26.6 to 1679.5 ng m-3 in all air samples. These concentration values are on average 5 times higher than those typically observed in urban areas in France (average values during summer 48.5±43.6 ng m-3) (Golly et al., 2018; Samaké et al., 2019a, b). However, these concentration levels are in agreement with a previous study conducted in a similar environment, i.e., continental rural sites located in large crop fields (Yan et al., 2019).

The total concentrations of SCs quantified in atmospheric PM10 over our study site accounted for 0.8 % to 13.5 % of the daily OM mass. This is remarkable considering that less than 20 % of total particulate OM mass can generally be identified at the molecular level. Hence, our results for a 9-week-long period indicate that SC could be a major identified molecular fraction of OM for agricultural areas during summer, which is in agreement with several previous studies conducted worldwide (Jia et al., 2010b; Verma et al., 2018; Yan et al., 2019). Further, it has been shown (Samaké et al., 2019a) that the identified polyols probably represent only a small fraction of the emission flux from this PBOA source and that a large fraction of the co-emitted organic material remains unknown. Hence, the PBOA source can potentially represent, for part of the year, a major source of atmospheric OM unaccounted for in CTMs.

In this study, 597 (39–132 MOTUs per sample) and 944 (31–129 MOTUs per sample) MOTUs were obtained for the fungal and bacterial libraries, respectively, reflecting the high richness of airborne microbial communities associated with ambient PM10 in this rural agricultural zone in France. Airborne fungi were dominated by Ascomycota (AMC) followed by Basidiomycota (BMC) phyla, consistent with the natural feature of many Ascomycota, whose single-celled or hyphal forms are small enough to be rapidly aerosolized, in contrast to many Basidiomycota that are typically too large to be easily aerosolized (Moore et al., 2011; Womack et al., 2015). Many members of AMC and BMC are known to actively eject ascospores and basidiospores, as well as aqueous jets and droplets containing a mixture of carbohydrates and inorganic solutes, into the atmosphere (Elbert et al., 2007; Womack et al., 2015). The prevalence of Ascomycota and Basidiomycota is consistent with results from previous studies also indicating that the Dikarya subkingdom (Ascomycota and Basidiomycota) represents about 98 % of known species in the biological kingdom of Eumycota (i.e., fungi) in the atmosphere (Elbert et al., 2007; James et al., 2006; Womack et al., 2015; Xu et al., 2017).

Airborne bacteria in this study belonged mainly to the Proteobacteria, Bacteroidetes, Actinobacteria, and Firmicutes phyla, consistent with previous studies (Liu et al., 2019; Maron et al., 2005; Wei et al., 2019b). Gram-negative Proteobacteria constitute a major taxonomic group among prokaryotes (Itävaara et al., 2016; Yadav et al., 2018), including bacterial taxa which are very diverse, important in agriculture, and capable of fixing nitrogen in symbiosis with plants (Itävaara et al., 2016; Yadav et al., 2018). Proteobacteria can survive under conditions with very low nutrient content, which explains their atmospheric versatility (Itävaara et al., 2016; Yadav et al., 2018). These results are similar to those observed in previous studies conducted in different environments around the world, where Proteobacteria, Actinobacteria, and Firmicutes have also been reported as dominant bacterial phyla (Liu et al., 2019; Maron et al., 2005; Wei et al., 2019a). In particular, the most frequent Gram-negative (Proteobacteria and Bacteroidetes) and Gram-positive (Actinobacteria and Firmicutes) bacteria and filamentous fungi (Ascomycota and Basidiomycota) have been previously linked to raw straw handling activities. For instance, it has been suggested that straw combustion during agricultural activities could be a major source of airborne microorganisms in PM2.5 in the northern plains of China (Wei et al., 2019a, b). However, in our study, SC species are not correlated (R=-0.09, p=0.46; Fig. S7) with levoglucosan during the campaign period, confirming that biomass burning is not an important source of airborne microbial taxa associated with SCs in our PM10 series. Bubble bursting associated with sea spray could also potentially be a source of bacteria, fungi, and water-soluble organic species, along with sea salts, to PM10 (Prather et al., 2013; Zhu et al., 2015). However, SC species were not found to be significantly related to Cl- (R=-0.14, p=0.28) or Na+ (R=-0.18, p=0.16), which are two inorganic tracers typical of marine sources, nor did they correlate with methanesulfonic acid (R=-0.05, p=0.69), a well-known tracer of biogenic marine activity (Arndt et al., 2017; Gaston et al., 2010). It therefore seems unlikely that the sources of SCs from marine environments were significant at this site. This point is further discussed in Sect. 4.4.

SCs are widely produced in large quantities by many microorganisms to cope with environmental stress conditions (Medeiros et al., 2006). SC species are known to accumulate in high concentrations in microorganisms at low water availability to reduce intracellular water activity and prevent enzyme inhibition due to dehydration (Hrynkiewicz et al., 2010). In addition, the temporal dynamics of ambient polyol concentrations has been suggested as an indicator to follow the general seasonal trend in airborne fungal spore counts (Bauer et al., 2008; Gosselin et al., 2016). Although this strategy has allowed the introduction of conversion ratios between specific polyol species (i.e., arabitol and mannitol) and airborne fungal spores in general (Bauer et al., 2008), the structure of the airborne microbial community associated with SC species has not yet been studied. Our results provide culture-independent evidence that the airborne microbiome structure and the combined bacterial and fungal communities largely determine the SC species concentration levels in PM10.

Temporal fluctuations in the abundance of only a few specific fungal and bacterial genera reflect the temporal dynamics of ambient SC concentrations. For fungi, genera that show a significant positive correlation (p<0.05) with SC species include Cladosporium, Alternaria, Sporobolomyces, and Dioszegia. Cladosporium and Alternaria are fungal genera that contribute on average to 47.9 % of total fungal sequence reads in our air samples series. These are asexual fungal genera that produce spores by dry-discharge mechanisms, wherein spores are detached from their parent colonies and easily dispersed by the ambient airflow or other external forces (e.g., raindrops, elevated temperature), as opposed to actively discharged spores with liquid jets or droplets in the air (Elbert et al., 2007; Wei et al., 2019b; Womack et al., 2015). Our results are consistent with the well-known seasonal behavior of airborne fungal spores with levels of Cladosporium and Alternaria which have been shown to reach their maximum from early to midsummer in a rural agricultural area of Portugal (Oliveira et al., 2009).

Similarly, bacterial genera positively correlated with SC species are Massilia, Pseudomonas, Frigoribacterium, and Sphingomonas. Although it is the prevalent bacterial genus at the study site, Sphingomonas is indeed not significantly positively correlated with SC species. The genus Sphingomonas is known to include numerous metabolically versatile species capable of using carbon compounds usually present in the atmosphere (Cáliz et al., 2018). The atmospheric abundance of species affiliated with Massilia has already been linked to the change in the stage of plant development (Ofek et al., 2012), which can be attributed to the capacity of Massilia to promote plant growth through the production of indoleacetic acid (Kuffner et al., 2010) and siderophores (Hrynkiewizc et al., 2010), which makes it antagonistic towards Phytophthora infestans (Weinert et al., 2010).

As far as we know, this is the first study evaluating microbial fingerprints with SC species in atmospheric PM; hence, it is not possible to compare our correlation results with those of previous works. However, it has already been suggested that types and quantities of SC species produced by fungi under culture conditions are specific to microbial species and external conditions such as carbon source, drought, heat, etc. (Hrynkiewicz et al., 2010). In future studies, we intend to apply a culture-dependent method to directly characterize the SC contents of some species amongst the dominant microbial taxa identified in this study after growth under several laboratory chambers, reproducing controlled environmental conditions in terms of temperature, water vapor, and carbon sources.

There are still many challenging questions on the emission processes leading to fungi and bacteria being introduced into the atmosphere together with their chemical components. In particular, the potential influence of soil and vegetation and their respective roles in structuring airborne microbial communities are still debated (Lymperopoulou et al., 2016; Rathnayake et al., 2016; Womack et al., 2015), especially since this knowledge is particularly essential for the precise modeling of PBOA emissions processes to the atmosphere within chemical transport models.

The characterization of the temporal dynamics of SC species concentrations could provide important information on PBOA sources in terms of composition, environmental drivers, and impacts. The results obtained over a 9-week period of daily PM10 SC measurements clearly show that the temporal dynamics of sorbitol (R=0.59, p<0.001) and inositol (R=0.64, p<0.001) are well-correlated linearly with that of calcium, a typical inorganic water-soluble ion from crustal material. This indicates a common atmospheric origin for these chemicals. Sorbitol and inositol are well-known reduced sugars that serve as a carbon source for microorganisms when other carbon sources are limited (Ng et al., 2018; Xue et al., 2010). In microorganisms, sorbitol and inositol are mainly produced by the reduction of intracellular glucose by aldose reductase in the cytoplasm (Ng et al., 2018; Welsh, 2000; Xue et al., 2010). Moreover, significant concentrations of both sorbitol and inositol have already been measured in surface-soil samples from five cultivated fields in the San Joaquin Valley, USA (Jia et al., 2010b; Medeiros et al., 2006). Therefore, sorbitol and inositol are most likely associated with microorganisms from soil resuspension.

With the exception of sorbitol and inositol, all other SC species measured in air samples at our sampling site are strongly correlated with each other, indicating a common origin. Daily calcium concentration peaks are not systematically associated with those of these other SC species. Interestingly, the highest atmospheric levels of these SC species occurred on 8 August 2017, coinciding well with daily harvesting activities around the site. This is also consistent with the multi-year monitoring of the dominant SCs in PM10 at this site, where ambient SCs showed a clear seasonal trend with higher values recorded in early August and in good agreement with harvesting activities around the study area every year from 2012 to 2017 (Samaké et al., 2019a). This suggests that the processes responsible for the dynamics of atmospheric concentrations of SCs are replicated annually and are most likely effective over large areas of field crop (Golly et al., 2018; Samaké et al., 2019a). Interestingly, glucose – the most common monosaccharide present in vascular plants and microorganism – has already been proposed as a molecular indicator of biota emitted into the atmosphere by vascular plants and/or by the resuspension of soil from agricultural land (Jia et al., 2010b; Pietrogrande et al., 2014). Therefore, all other SC species measured in our series can be considered to be most likely the result of the mechanical resuspension of crop residues (e.g., leaf debris) and microorganisms attached to them. Other confirmations of this interpretation stem from the excellent daily covariations observed in the PM10 between SC species levels and ambient cellulose, which is widely considered a reliable indicator of plant debris sources in PM studies (Bozzetti et al., 2016; Hiranuma et al., 2019).

Microbial abundance and community structure in samples from the surrounding environment can provide further useful information on source apportionment and importance. Our data indicate that the airborne microbial genera most positively correlated to SC species are also distributed in the surrounding environmental samples of both surface soils and leaves, suggesting a dominant influence of the local environment for microbial taxa associated with SC species as opposed to long-range transport. This observation makes sense since actively discharged ascospores and basidiospores are generally relatively large airborne particles with a short atmospheric residence time (Elbert et al., 2007; Womack et al., 2015), limiting the possibilities of long-range dissemination. Accordingly, the majority of previous studies investigating the potential sources of air microbes identified local surface environments (e.g., leaves, soils) as having more important effects on the airborne microbiome structure in field crop areas (Bowers et al., 2011; Wei et al., 2019b; Womack et al., 2015). This is all the more the case in our study with homogeneous crop activities for tens to hundreds of kilometers around the site.

In the present study, microbial diversity and richness observed in the surface soils are generally higher than those on leaf surfaces. Microbial taxa which most positively correlated with PM10 SC species are generally more abundant in leaf than in topsoil samples. These results were unexpected and show the possible importance of leaf surfaces in structuring the airborne taxa associated with SC species. Considering the general grouping of leaf samples and airborne PM10 regardless of harvesting activities around the study site and in addition to the separate assemblies of rarefied MOTUs in airborne PM10 and topsoil samples, it can be argued that the aerial parts of plants are the major source of microbial taxa associated with SC species. Such an observation is most likely related to increased vegetative surfaces (e.g., leaves) in summer that provides sufficient nutrient resources for microbial growth (Rathnayake et al., 2016). By reviewing previous studies, Alternaria and Epicoccum, which made up 30 % of total fungal sequence reads in all air samples in this study, have been shown to be common saprobes or weak pathogens of leaf surfaces (Andersen et al., 2009). Similarly, Cladosporium, which accounted for 32.9 % of total fungal genera in all air samples, has also been shown to be a common saprotrophic fungus, inhabiting decayed tree or plant debris (Wei et al., 2019b). The high relative abundance of Sphingomonas and Massilia, accounting for 28.4 % of total bacterial genera in all air samples, is also noticeable. These two phyllosphere-inhabiting bacterial genera are well-known for their plant protective potential against phytopathogens (Aydogan et al., 2018; Rastogi et al., 2013).

Altogether, these observations support our interpretation that leaves are the major direct source of airborne fungi and bacteria during the summer months at this site of high agricultural activity. Endophytes and epiphytes can be dispersed in the air and transported vertically as particles by air currents, much faster and more widely than by other mechanisms, such as direct dissemination from surface soil, which is generally controlled by soil moisture (Jocteur Monrozier et al., 1993). The most wind-dispersible soil constituents are indeed the smallest soil particles (i.e., clay-size fraction), which contain the largest number of microorganisms (Jocteur Monrozier et al., 1993) and can only be released into the atmosphere under conditions of prolonged drought. This interpretation is also consistent with previous studies (Bowers et al., 2011; Liu et al., 2019; Lymperopoulou et al., 2016; Mhuireach et al., 2016), which also show the extent to which endophytes and epiphytes can serve as quantitatively important sources of airborne microbes during summertime when vegetation density is highest. For example, Lymperopoulou et al. (2016) observed that bacteria and fungi suspended in the air are generally 2 to more than 10 times more abundant in air that passed over 50 m of a vegetative surface than in air that is immediately upwind of the same vegetative surface. However, the relative abundance of taxa associated with SCs in surface soils in this study could also be indicative of a feedback loop, in which the soil may serve as a source of microbial endophytes and epiphytes for plants, while the local vegetation in turn may serve as a source and sink of microbes for local soils during leaf senescence.