Aerosol sampling was performed during the INUIT-BACCHUS-ACTRIS campaign in April 2016 at the Cyprus Atmospheric Observatory, which is a rural background station that operates under the co-operative program for monitoring and evaluation of the long-range transmission of air pollutants in Europe (EMEP) and the European research infrastructure for the observation of Aerosols, Clouds and Trace Gases Research Infrastructure (ACTRIS) networks while, at the same time, it is a designated regional Global Atmosphere Watch (GAW) station. The station (35.038692∘ N, 33.057850∘ E) is located close to the villages of Agia Marina (∼ 630 inhabitants) and Xyliatos (∼ 150 inhabitants) and has an elevation of 532 m above sea level. The site lies at the northeastern foothills of the Troodos Mountains. The nearest main urban agglomeration is at least 35 km away. The vegetation at the measurement site is dominated by evergreen shrubs and small trees, blending at higher elevations of the Troodos Mountains into oak and pine forests .

Total suspended particle samples were collected on glass-fiber filters (Type MN 85/90 BF, 150 mm diameter, retention capacity of 0.5 µm; Macherey-Nagel GmbH & Co. KG, Germany) with a DHA-80 high-volume aerosol sampler (DIGITEL Elektronik AG, Hegnau, Switzerland; volumetric flow rate 1000 L min-1, sampling time 24 h) positioned at the rooftop of the Cyprus Atmospheric Observatory station (∼ 5 m above the ground). Blank samples were taken at regular intervals (∼ 3 d intervals) to detect potential contamination by the aerosol sampler or by filter mounting. For this, filters were mounted in the sampler as for regular sampling, but the sampler was turned on either not at all (mounting blank) or only for 5 s (start-up blank).

Prior to sampling, all glass-fiber filters were pre-baked at 330 ∘C for 8 to 10 h in aluminum bags to remove any biological material. The filter holders were disinfected with Bacillol^® AF (Bode Chemie GmbH, Hamburg, Germany) before use. After sampling, the loaded filters were removed from the filter holders, placed into baked (330 ∘C, 8 to 10 h) aluminum bags, stored at -20 ∘C, and stored at -80 ∘C after international transport on dry ice. The aerosol filter samples were cut into 16 equal pieces using a self-made cutting tool, made of sharp guided blades stuck to a round aluminum frame (140 mm diameter and approximately 1 cm thickness). Prior to cutting, the cutting tool was sterilized by dipping it into ethanol (96 %) and flaming. The filter sample aliquots were transferred into sterile 50 mL tubes (Greiner Bio-One, Kremsmünster, Austria) and stored at -80 ∘C until analysis. To avoid contamination, filter handling and processing were carried out under sterile conditions on a clean bench.

Based on the monitored environmental conditions as detailed below, nine aerosol samples were selected for metagenomic sequencing and ice nucleation measurements. A list of the investigated aerosol samples with sampling details is given in Table . Note, that the measurements refer to atmospheric particles that were not embedded in or scavenged by raindrops.

The meteorological conditions during the sampling campaign in April 2016 were measured at the Cyprus Atmospheric Observatory by the Department of Labour Inspection of Cyprus. The data, comprised of hourly mean values for temperature, relative humidity, rainfall, wind speed, and regulatory mass concentrations for PM10 and PM2.5, were imported into R version 3.6.3; and the coordinated universal time (UTC) timestamps were converted to the local eastern European time (EET, UTC+2).

Figure  shows the meteorological conditions and aerosol mass concentrations during collection of the aerosol samples. Temperature and relative humidity (RH) exhibited some variation throughout the campaign. A strong increase in RH is noticeable on 9 April 2016, followed by a period of high RH (up to 80 %) and a temperature drop until 14 April 2016 corresponding to samples CY19–CY26. Except for this 5 d period, the daily temperature peaks were in the range of 25 to 30 ∘C. A period of higher RH was again registered starting on 25 April 2016 (CY48).

April 2016 was mainly dry, with only three rain events. One rain event was registered on 12 April during the last 3 h of collection of sample CY20, and two small rain events occurred during collection of CY23 on 13 April. Increased PM concentrations, corresponding to dust events , were measured from 9 to 12 April (CY18 and CY19) and from 15 to 18 April (CY31 and CY32).

Two filter sample aliquots from opposite sides of each air filter sample were extracted with the ChargeSwitch™ Forensic DNA Purification Kit (Invitrogen, Thermo Fisher Scientific Inc., Waltham, MA, USA) according to the manufacturer's instructions with the following modifications: incubation at 55 ∘C for 60 min and elution of DNA with 100 µL elution buffer. Moreover, the DynaMag™-2 magnet instead of MagnaRack™ was used. DNA extracts of the two filter pieces of each sample were pooled and purified with the PowerClean Pro DNA Clean-Up Kit (MO BIO Laboratories, Inc., QIAGEN, Hilden, Germany) according to the manufacturer's instructions. To detect potential contamination, four mounting blanks and four start-up blanks were included in the DNA extraction. Additionally, four baked filters were included as filter blanks, and four extractions without any filter served as extraction blanks. DNA concentrations were determined with the DeNovix dsDNA Ultra High Sensitivity Assay (DeNovix Inc., DE, USA) and the Qubit^® 3.0 Fluorometer (Thermo Fisher Scientific, USA). DNA was not detected in the mounting, start-up, filter, and extraction blanks, indicating that no contamination occurred during sample handling and extraction.

From each DNA extract, 1 ng of DNA was sequenced on Illumina HiSeq 3000 (output 2 × 150 bp paired-end sequencing reads) and Illumina HiSeq 2500 sequencers (output 2 × 150 bp paired-end sequencing reads) at the Max Planck Genome Centre in Cologne, Germany. Illumina TruSeq adapters were used for the sequencing library. The data output per run was 17.5 GB reads. In total 4.14 × 108 reads were obtained from all samples after quality filtering, with an average of 4.6 × 107 reads per sample.

To identify ribosomal small subunit (SSU, 16S and 18S) and large subunit (LSU, 23S and 28S) reads and to assign taxonomy and operational taxonomic units (OTUs), the MGnify pipeline 4.1 with MAPseq (1.2.2) and the SILVA database (SSU and LSU version 132) remapped to an eight-level taxonomy was used . In total, 8.41 × 105 SSU and 1.09 × 106 LSU reads were obtained with averages of 9.34 × 104 SSU and 1.21 × 105 LSU reads per sample. Reads identified as chloroplast and mitochondrial ribosomal gene fragments were excluded from downstream analysis. Preliminary taxonomic analysis displayed unusually high numbers of SSU reads and inconsistencies between SSU and LSU results for CY26. Analysis of read abundance revealed that, in contrast to the other samples, the data of CY26 contained a high number of reads starting with sequences of degenerated random 16S primers, which are usually used in fingerprinting techniques . Thus, all 16S rRNA reads from sample CY26 were excluded.

All downstream analyses were performed in R version 3.6.3 . The base R functions were used for data processing and plotting. The SSU and LSU files were imported into R using the biomformat package . Additionally, the InterPro matches provided by the MGnify pipeline were specifically analyzed for viral, allergenic, and bacterial ice-nucleating proteins.

For ice nucleation measurements, aqueous extracts of two pieces of each air filter sample were analyzed using the Twin-plate Ice Nucleation Assay (TINA) . The extracts were prepared with pure water, which was obtained from a Barnstead™ GenPure™ xCAD Plus water purification system (Thermo Fisher Scientific, Braunschweig, Germany), autoclaved at 121 ∘C for 20 min, and filtered through a sterile 0.1 µm pore diameter polyethersulfone (PES) vacuum filter unit (VWR International, Radnor, PA, USA). For extract preparation, 20 mL of the pure water was added to each filter piece in sterile 50 mL tubes (SPL Life Sciences Co. Ltd, South Korea). The tubes were vortexed at 2700 rpm for 15 min and another 15 min upside down (Vortex-Genie 2, Scientific Industries, Inc., USA). Afterwards, the aqueous extracts were transferred into new sterile 50 mL tubes. For IN characterization, 1 mL aliquots of the aqueous extracts were treated as follows: (i) filtration through a 5 µm pore diameter filter (Acrodisc, PES, Pall GmbH, Dreieich, Germany), (ii) filtration through a 0.1 µm pore diameter filter (Acrodisc), (iii) incubation at 98 ∘C for 1 h, (iv) filtration through a 5 µm pore diameter filter and incubation at 98 ∘C for 1 h, and (v) filtration through a 0.1 µm pore diameter filter and incubation at 98 ∘C for 1 h.

Of each untreated and treated aqueous extract, 96 aliquots of 3 µL were pipetted by a liquid handling station (epMotion 5073, Eppendorf, Hamburg, Germany) into 384 well plates (Eppendorf). The plates were placed in the TINA instrument, and, after 1 min of equilibration at 0 ∘C, they were cooled to -28 ∘C with a continuous cooling rate of 1 ∘C min-1. Pure-water samples (autoclaved and 0.1 µm filtered) served as negative controls in the individual experiments. Additionally, aqueous extracts of two filter pieces of one filter blank, one mounting blank, and one start-up blank served as negative controls for the aerosol sampling process (Sect. 2.1). As shown in Fig. B1, the blank samples showed freezing events starting around -20 ∘C with most droplets freezing at around -25 ∘C. Thus, the observed temperature range for the aerosol samples was 0 to -20 ∘C.

The freezing temperature data of the individual measurements were binned into 0.2 ∘C intervals, which corresponds to the temperature uncertainty in the measurement . For each aerosol sample, the data of the 96 droplets of the two filter pieces were combined to make 192 droplets for further calculations. All calculations and plotting were performed with R version 3.6.3 . The fraction of frozen droplets (fice), the cumulative number of IN (Nv) per liter of air, and corresponding errors were calculated as described in and .

The composition of the airborne community was determined by taxonomic assignment of ribosomal RNA gene SSU (small subunit) and LSU (large subunit) reads. In total, we detected 3000 operational taxonomic units (OTUs) for SSU and 4167 OTUs for LSU reads. Note that the SSU OTUs do not include data from sample CY26 (see Sect. 2.5). The obtained OTUs, however, can be seen as lower estimates of species richness, as MGnify assigns OTUs based on a reference database to different taxonomic levels. This leads to numerous OTUs at higher taxonomic levels, which comprise reads from multiple species that cannot be identified to lower taxonomic levels. Thus, for downstream analysis and inter-sample comparison, read proportions were used.

Figure  shows the read proportions and taxonomic assignments on the domain and kingdom level for the individual samples. The taxonomic composition, in particular the read proportion of fungi and bacteria, changed with sample CY23, indicating a change in the airborne community. Sample CY23 was collected after a 3 h rain event, which corresponds to the collection end of CY20 (Fig. ). Moreover, two additional rain events occurred during sampling of CY23. This sample shows the highest proportion of fungal reads (> 60 %) and lowest read proportion of Bacteria (< 20 %) and Viridiplantae (< 10 %). Samples collected up to 4 d after the rain events (CY26 and CY31) also show higher proportions of fungal reads (up to 50 %) and lower read proportions for Bacteria and Viridiplantae than the other samples. Metazoa and Archaea read proportions were < 7 % and < 1 %, respectively, in all samples (Fig. ).

The high proportion of fungal reads in CY23 may result from increased concentrations of fungal spores following rainfall, as many fungal spores are dispersed by rain splash, rain tap and puff, or other spore discharge mechanisms that require water e.g.,. Bacteria are often associated with larger particles such as soil dust and plant fragments or form aggregates with each other . As larger particles are more efficiently removed by precipitation than smaller particles, because the latter tend to follow the airstream around the raindrops , an efficient washout is assumed for bacteria and larger particles. This facilitates the return of airborne bacteria and other biological particles to the Earth's surface as well as deposition in new habitats , which possibly explains the reduced read proportions for Bacteria, Viridiplantae, and unidentified Eukaryota in samples CY23 and CY26.

Figure  also shows, for the individual samples, that the proportions of SSU and LSU reads assigned to Archaea, Bacteria, Fungi, and Metazoa are comparable, indicating that both genes can be used as representative markers for these groups. In contrast, Viridiplantae and unidentified Eukaryota show differences in the SSU and LSU read proportions in samples CY14–CY20 and more equal read proportions in all other samples. This indicates a change in the airborne community due to the rain events and a lack of plant LSU sequences in the SILVA reference database.

Unlike pro- and eukaryotes, for which ribosomal RNA genes can be used for identification, viruses do not contain common genetic markers . Analysis of the InterPro matches, however, revealed the presence of various double- and single-stranded DNA and RNA viruses in the aerosol samples. Of the virus-related InterPro matches ∼ 44 % were assigned to viruses that infect bacteria and archaea (e.g., Siphoviridae, Podoviridae, Myoviridae, and Microviridae), which cover ∼ 80 % of the reads that were assigned to virus-related matches. Viruses infecting animals and humans (e.g., Adenoviridae, Baculoviridae, Flaviviridae, Herpesviridae, Poxviridae) also exhibited ∼ 44 % of the matches corresponding to ∼ 4 % of reads. Plant viruses such as Geminiviridae, Caulimoviridae, and Potyviridae were less abundant with ∼ 7 % of the matches and < 1 % of reads. The remaining ∼ 4 % of the virus-related InterPro matches (∼ 15 % reads) either could not be assigned to hosts or were from fungi- and amoeba-infecting viruses. The detection of RNA viruses (e.g., Flaviviridae, Reoviridae, Retroviridae) can be explained by the integration of these viruses into the genomes of their host cells and their persistence as proviruses .

Figure  provides an overview of the freezing abilities for the individual samples. Figure a shows the fraction of frozen droplets; Fig. b shows the median temperatures T50 (i.e., the temperature at which 50 % of the droplets were frozen); Fig. c shows the ice nuclei (IN) concentrations calculated per liter volume of sampled air (from the data shown in Fig. a). The individual samples initiated freezing between -6.3 and -11.3 ∘C. The frozen fraction curves show differences in shapes and slopes as well as different T50 values. The curve of sample CY23, which was collected after and during the rain events, approximates a straight line with T50=-16.7 ∘C. Contrastingly, the samples CY26–CY32, which were collected after the rain events, all show a similar convex curve shape and higher median freezing temperatures (T50 is -13.9 to -15.5 ∘C). All other samples, including sample CY48, which was collected 13 d after the last rain event, show concave curves and lower T50 values (-17.3 to -18.3 ∘C).

Figure c shows that the individual samples differ in their IN concentrations at given temperatures. For example, at -15 ∘C, values between 10-1 and 10-2 IN per liter of sampled air were found for the investigated samples. Overall, the obtained IN concentrations span about 2 orders of magnitude for temperatures > -15 ∘C and about 1 order of magnitude for temperatures < -15 ∘C. These results are in good agreement with , who reported IN concentrations on the same order of magnitude for samples collected in Cyprus in April 2017.

The results indicate heterogeneity of the IN compositions in the samples. Higher initial freezing temperatures and T50 values can be an indication of the presence of bio-IN, as many bio-IN are known to nucleate at temperatures > -15 ∘C while mineral IN dominate at < -15 ∘C e.g.,.

To further characterize the IN of the different aerosol samples, filtration and heat treatment experiments were performed with the aqueous extracts. Figure  shows that filtration through a 5 µm and a 0.1 µm filter reduced the ice nucleation activity in all samples, indicating that all samples contained IN larger than 5 µm, IN between 0.1 and 5 µm, and IN smaller than 0.1 µm. For samples CY14, CY23, CY26, CY31, and CY48, however, the results show only a small fraction of IN between 0.1 and 5 µm while many IN were smaller than 0.1 µm. Of all samples, CY18 and CY19 show the lowest percentage of frozen droplets after 0.1 µm filtration and, thus, the lowest numbers of IN smaller than 0.1 µm in the observed temperature range.

Heat sensitivity can indicate biological IN, as heat can reduce ice nucleation activity of many bio-IN . Certain bio-IN, however, such as Lysinibacillus, a heat-resistant isolate of Xanthomonas, and various plant pollen, as well as inorganic particles are heat resistant to temperatures > 90 ∘C . Heat treatment at 98 ∘C for 1 h reduced the ice nucleation activity of the individual samples to a different extent (Fig. ). For example, the ice nucleation activity of samples CY20–CY32 was significantly reduced in the different size fractions after heat treatment (Fig. d–h), indicating the presence of heat-sensitive IN and thus presumably bio-IN of different sizes. Many fungal spores and bacterial aggregates are larger than 5 µm, whereas single bacterial cells or cell fragments can be smaller than 5 µm, and ice-nucleating macromolecules are typically in the submicrometer range . The samples CY23 and CY26, which were collected during and after the rain events, appear to have the largest fraction of heat-sensitive IN larger than 5 µm. In contrast, heat only slightly affected samples CY18 and CY19 (Fig. b–c), indicating a significant fraction of heat-resistant IN in these samples, possibly due to particles from the dust events which were registered during the collection of these samples (Fig. ). For samples CY14 and CY48, the ice nucleation activity of 5 and 0.1 µm filtrates was reduced by heat, while the unfiltered extract was only slightly affected (Fig. a, i). The results suggest that, in these samples, many heat-resistant IN were larger than 5 µm, whereas many heat-sensitive IN were smaller than 0.1 µm or, for CY14, also between 0.1 and 5 µm.

The results of the freezing experiments indicate the presence of bio-IN in different size ranges. These bio-IN may be present as living or dead cells, fungal hyphae and spores, pollen, cell fragments, or detached macromolecules or associated with plant particles or soil organic matter . Note that not all of these IN contain DNA, for example ice-nucleating macromolecules or cell membrane fragments with attached IN proteins, and they are thus not all covered by DNA analysis. The largest fraction of bio-IN > 5 µm was detected in CY23 and CY26. For these samples, the sequencing results show a higher proportion of reads assigned to fungi, in particular Agaricomycetes (Figs. , ), suggesting a possible contribution of these fungi to the increase in IN. Systematic surveys suggest that fungal ice nucleation activity, in particular at temperatures > -15 ∘C, is restricted to a limited number of fungal species . The class Agaricomycetes contains approximately 21 000 fungal species , but only a handful has been tested for ice nucleation activity so far . Further surveys of ice nucleation properties of Agaricomycetes and other fungi are necessary.

To identify possible bio-IN sources, the SSU and LSU data were analyzed for known IN fungi and bacteria that are active at temperatures > -15 ∘C. In total, 4 fungal genera and 13 bacterial genera that contain IN-active species were detected, but all exhibited read proportions < 1 % and species identification was partly not possible. Note that not all species and strains of the identified genera are IN-active, and, especially in bacteria, not all cells express the IN gene . Moreover, some recently isolated IN-active fungi and bacteria such as Cryptococcus sp. and Brevibacterium sp. are not identified to the species level . Thus, the following microorganisms represent possible sources of IN but provide no proof of to what extent they were IN-active in the air.

Among the fungi were Cryptococcus sp., Fusarium sp., the soil fungus Mortierella alpina, and five species of the plant-pathogen rust fungi Puccinia (P. arachidis, P. hordei, P. horiana, P. striiformis, P. triticina). Spores of Puccinia striiformis (stripe rust or yellow rust) and P. triticinacan have initial freezing temperatures of -7 and -9 ∘C, respectively . Although several other rust species are known as ice nucleators, for P. arachidis (peanut rust), P. hordei (barley leaf rust), and P. horiana (chrysanthemum white rust), ice nucleation activity has not yet been reported. In contrast to Puccinia spp., for which the spores are IN-active, the IN of M. alpina and Fusarium spp. are cell-free proteinous macromolecules and can trigger freezing at temperatures as high as -5 ∘C or -1 ∘C, respectively . The macromolecules might be released into the surrounding environment and contribute to atmospheric IN.

Among the potential IN-active bacteria were the well-known genera and species Erwinia, Pantoea agglomerans, Pseudomonas, and Xanthomonas as well as the recently identified Brevibacterium, Idiomarina, Lysinibacillus, Planococcus maritimus, and Psychrobacter . The genus Pseudomonas, which includes a range of IN-active species, was detected in all samples. Most of the reads (∼ 94 %) assigned to the genus Pseudomonas, however, could not be identified to the species level. The remaining 58 reads were distributed over 12 species (1 to 14 reads per species) but did not include P. syringae, a common bacterial ice nucleator. Analysis of the InterPro matches for IN-active proteins, however, revealed one InterPro match (IPR000258, bacterial ice nucleation, octamer repeat) for in total 24 reads from six of the nine samples (1 to 12 reads per sample), indicating the presence of a bacterial IN-active species. Recently, ice nucleation activity was found in Halococcus morrhuae and Haloferax sulfurifontis, two species of Haloarchaea, which initiated the freezing of water droplets at temperatures as high as -18 ∘C . In our data set, however, only one read from CY48 was assigned to the genus Halococcus and Haloferax was not detected.