The Biogeography of Dust in the Eastern Mediterranean

The atmospheric microbiome, the aerobiome, represents a complex mixture of suspended microorganisms originating from different environments, e.g., soil, marine environments, and plants. Aerobiomes sampled around the globe show distinct compositions, yet some common features might be observed. De�ning these similarities and differences will contribute to our understanding of this understudied environment. In this study, we sequenced the metagenomes of atmospheric dust collected in Israel, representing different particulate matter (PM) concentrations, dust origins and atmospheric trajectories. The metagenomes were compared to publicly available atmospheric metagenomes, as well as to marine and soil metagenomes from the Eastern Mediterranean region. This study helps elucidate the role of dust storms in replenishing the atmosphere with bacteria, fungi and archaea, which facilitate essential biogeochemical processes, by investigating microbiomes and functional genes from dust metagenomes. Our results reveal a high level of similarity between metagenomes of dust sampled in Israel and metagenomes of dust collected in Saudi Arabia asynchronically. Moreover, by identifying taxa and genes that are correlated with PM concentrations, we suggest a possible "core" atmospheric microbiome composed mainly of fungi. Application of the SourceTracker algorithm revealed an inverse correlation between the "core" aerobiome and the fraction of soil-derived bacteria and archaea in the dust metagenomes. Marine microbiomes contributed very little to the dust microbiome, even in dust that traversed the Mediterranean and Red Seas. The dust metagenomes showed a high relative abundance of genes associated with antibiotic resistance, sporulation and the ability to degrade aromatic pollutants, which could not be traced to any of the natural microbiomes that were examined. This study represents an important step toward integrating the knowledge accumulated on the global aerobiome and reveals the important role of dust storms in contributing to the presence of various environmentally relevant microbial functions.


Introduction
For the past decade, data on global airborne microbial communities have slowly accumulated [1][2][3][4][5][6][7] .In these studies, aerosols and airborne bacteria were collected using various sampling methods, covering different aerosol sizes and types, and were genetically analyzed by different DNA sequencing platforms.This effort has created a growing knowledge base on the microbial community in the atmosphere and has provided a few clues regarding the variability, viability and functionality of this community.However, many questions remain unanswered; although the studies were performed at different locations around the globe (e.g., Europe and the Alps, 8-10 Asia, 2,11−13 the Mediterranean, 7,14 North America, 15,16 South America, 17 the Amazon rainforest, 18 and Antarctica 19 ), little is known about the residence time of microorganisms in the atmosphere; the connectivity of microbial communities between regions; the presence of a "core" microbiome in the atmosphere; anthropogenic effects on the atmospheric microbiome; and the role played by the atmosphere in the introduction of new bacteria and functions to different environments, and therefore about the global dissemination of pathogens and antibiotic resistance genes.Finally, the viability and functionality of airborne microorganisms in different niches within the atmosphere, such as clouds vs. dust storms, remain unknown.These questions are of great importance for understanding the global microbiome and could be essential for global ecosystem functioning, human health and the climate. 20,21 date, most atmospheric microbiome studies have applied amplicon sequencing of a taxonomic proxy, such as the gene encoding the small subunit (SSU) of the ribosome, i.e., 16S rRNA or 18S rRNA for prokaryotes and eukaryotes, respectively.This approach enabled sequencing despite the relatively low biomass collected from air samples and provided important insight into the microbial community structure.It also provided some hints about the possible functionality of these microorganisms.However, only a limited number of studies have succeeded in collecting su cient material to obtain the DNA concentrations required for metagenomic and metatranscriptomic sequencing, including aerosols from anthropogenic environments such as farms and cities 13,22−24 and natural environments such as clouds 8 and dust aerosols 3 .Each of these studies explored different aspects of the atmospheric microbiome, including metabolic processes, environmental epidemiology and microbial viability, in addition to documenting the microbial community structure and functional gene abundance and distribution.
However, these rare studies provided only a glimpse into the complexity of the airborne microbiome, i.e., the aerobiome, and its potential functionality.
Israel is at a crossroads of air masses from different sources situated in the Eastern Mediterranean basin in the vicinity of the Sahara, Arabian and Syrian deserts. 25As such, it experiences dust events originating from all these sources and exhibits a relatively high daily average loading of particulate matter (PM), with the concentration of particles under 10 µm (PM 10 ) averaging approximately 30-50 µg•m − 3 . 26Dust sampled in Israel has been shown to display signi cant source variability, with distinct signatures for local aerosol and desert dust microbiomes. 4,7This variability enabled us to differentiate among the contributions of different aerobiome sources: terrestrial, marine and anthropogenic. 4We assume that dust storms in the Middle East, as well as a signi cant portion of the general aerosols in this region, originate from soil and are perhaps affected to some extent by the Mediterranean Sea and the Red Sea and by anthropogenic sources such as farms, industrial sites and cities.Under this assumption, we conducted a comparative analysis of shotgun metagenomes of dust samples collected at a single location in Israel, covering different atmospheric conditions, with publicly available atmospheric, soil and marine metagenomes, addressing differences and similarities in both taxonomic and functional compositions.

Methods And Materials
Aerosol sampling: Particulate matter was collected on the roof of a four-story building at the Weizmann Institute campus in Israel (31.9070N, 34.8102E) using a high-volume sampler (Ecotech, model: HiVol 3000, PM 10 inlet, ow rate: 67.8 m 3 •h − 1 ) onto prebaked quartz ber lters (Whatman, 203 mm × 254 mm) for 6 to 48 h at a time, depending on the load of particulate matter with an aerodynamic diameter under 10 µm (PM 10 ).The lters were then wrapped in sterile aluminum foil and kept at -20°C until DNA was extracted.The samples are described in Table 1.The source of each sample was determined using the hybrid single-particle Lagrangian integrated trajectory model (HYSPLIT) via a web-based interface (READY, http://ready.arl.noaa.gov/HYSPLIT_traj.php).The dust sources were also veri ed using the Giovanni online data system, developed and maintained by the NASA GES DISC. 27Back trajectories for all sampling events detailed in Table 1 are presented in Table S1.

DNA extraction:
DNA was extracted using a DNA PowerSoil Kit (Qiagen, Germany), with modi cations to the manufacturer's protocol, as follows: four strips, 30 cm 2 each, were cut from each lter using a surgical scalpel, and the upper layer of each strip was scratched off and inserted into a single PowerSoil bead tube, to which 250 µl of PBS and 150 µl of a phenol:chloroform:isoamyl solution (ratio 25:24:1) were added.Next, the tubes were shaken in a Mini-Beadbeater-16 (Biospec, USA) for 2 minutes, removed and placed on ice for 15 seconds and then shaken for an additional 2 minutes.Extraction was continued according to the protocol up to the membrane binding step.In this step, the contents of the four tubes were loaded onto two spin lter tubes.Next, the samples were washed twice with C5 solution to rid the DNA of all traces of phenol and chloroform.Elution was conducted by loading 50 µl of solution C6 on one tube, leaving it for 5 minutes at room temperature, centrifuging for 1 minute at 16,000 ×g, loading the eluent onto the second spin lter, and repeating the described elution step.
Following extraction, samples with similar characteristics (PM 10 concentration and source of aerosols according to back-trajectories and satellite images) were pooled together (see Table 1), and DNA was precipitated using 100% ethanol and 5 M NaCl solution.The precipitated DNA was resuspended in 25 µl of TE buffer.We thus created a set of nine samples, ve representing two major dust events originating in Syria and in the Arabian Desert, three composite samples representing minor to moderate dust events from the Sahara, Sinai and Jordanian Deserts and a single composite sample representing days with low PM 10 concentrations.Pooling of samples was necessary due to low extracted DNA concentrations.performed to ensure an even distribution of sequencing output.The samples were repooled, followed by repeating Pippin-prep size selection.The samples were sequenced using a HiSeq2500 (Illumina) with 2x250 cycles using a single lane.

Metagenome dataset composition and analysis
Table 2 shows the reference metagenomes that were compared to the dust metagenome collected in this study.Raw and assembled reads were uploaded to the MG-RAST metagenomic analysis server 28 and were analyzed using the default parameters for shotgun metagenomic sequencing.Taxonomic data were acquired by comparison to the RefSeq database (NCBI [29][30][31] ), and functional genes were annotated against the KO database (KEGG [32][33][34] ), both available on the MG-RAST server.Taxonomic data were ltered to exclude multicellular organisms as well as viruses.Functional annotation data were ltered to exclude organismal system and human disease categories.All counts were transformed using the center log-ratio (clr) transformation method prior to data analysis.Downstream analysis, statistical analysis and gure preparation were conducted using R v.4.0.3 35 and mothur v.1.44.0. 36General data formatting was conducted with tidyverse; 37 statistical signi cance tests for differentially abundant genes and taxa, as well as nonmetric multidimensional scaling (NMDS) ordination, were conducted using Vegan, 38 broom 39 and purr; 40 correlation matrices were produced using Hmisc; 41 gene and taxon counts were clrtransformed using the compositions package; 42 analysis of molecular variance (AMOVA) 43 tests were conducted using the amova command in mothur; tting of functional categories, taxa and NMDS ordination axes was conducted using the corr.axescommand in mothur with the method set to Spearman; plots were produced using the R packages ggplot2, 44 and Superheat. 45The presented qvalues are p-values tted according to Benjamini-Hochberg correction for multiple comparisons.
To compare dust to probable sources, as well as to other atmospheric metagenomes, we divided the different datasets according to their type: "Dust" included the dust samples collected in this study (Table 1) and dust samples collected over the Red Sea by Aalismail et al. 3 ; "Smog" represents samples collected during a smog event in Beijing 13 ; "Air" includes cloud condensation samples collected in France 8 and urban aerosol samples from two locations in North America 22 ; "Soil" represents different soils collected over a precipitation gradient in Israel 46 ; and "Sea" includes samples from the Eastern and Western basins of the Mediterranean Sea 47 and various locations along the Red Sea. 48All datasets and sample types are described in Table 2.

Results
Community structure and functional pro le Figure 1 presents NMDS ordination based on taxonomy (1a) and functional genes (1b) correlated with microbial genera or KEGG pathways.We divided the samples into three categories: atmosphere, including all atmospheric sample types, i.e., dust, air and smog; marine, for the sea sample type; and terrestrial, for the soil sample type.All presented taxa and pathways were signi cantly correlated (Spearman's correlation, p-value < 0.05) with either the MDS1 or MDS2 axis and displayed a vector length greater than 0.8.AMOVA results indicated that all ve sample types presented, i.e., air, dust, sea, smog and soil, differed signi cantly from one another (q-value < 0.0005), both taxonomically and functionally.However, when comparing the different datasets, urban air and clouds did not signi cantly differ in their functional gene pro les according to the AMOVA test (q-value = 0.058).All other datasets differed signi cantly in their community composition as well as functional gene pro le (Tables S2 and S3).
According to Fig. 1a, the variance along the MDS1 axis was correlated with the abundances of the fungal phylum Ascomycota, the bacterial phylum Proteobacteria and the archaeal phylum Euryarchaeota.The variance along axis MDS2 was affected by the abundances of Actinobacteria and Firmicutes.
NMDS functional ordination, based on KEGG annotated genes, did not differ much from that based on taxonomy, yet it displayed a greater variance between the different sample types in the marine and atmospheric environments (Fig. 1b).Screening against the KEGG pathway database revealed that the variance along MDS1 was correlated with the abundance of metabolic pathways compared with the abundance of genetic information processes.
In both ordinations, dust and smog samples clustered between soil and air sample types, suggesting that these microbiomes are composed of microbial components from both air and soil.Marine microbiomes did not seem to contribute microbial components to the dust, smog, soil and air samples.To test this hypothesis, we assessed the source apportionment of soil, air and sea microbiomes in the dust samples using SourceTracker 49 (Fig. 2).Beijing smog was not included in this analysis since it did not form an end member on either taxonomic or functional ordination.As expected, over 60% of the dust storm samples, i.e., Dust_6 to Dust_10, were predicted to be soil derived.The highest air contribution was observed for samples Dust_2 to Dust_5 and samples Dust_9 and Dust_10, with a range of 3.8-12.0%microorganisms resembling those in air samples.In all samples, including those sampled over the Red Sea, the contribution of sea microorganisms was at most 3.1%.

Functions and taxa correlated with PM 10 concentrations in dust
To further understand the contribution of soil microorganisms to the dust microbiome, we performed correlation analysis (Spearman's correlation 50 ) between PM 10 concentrations and taxa or functional genes with clr-transformed counts of Dust IL samples.Assuming that taxa and genes that were positively correlated with PM 10 are likely to originate from soil, especially in the Mediterranean area, whereas negatively correlated pathways and genera originated from unknown sources, we could distinguish between the local, terrestrial contribution to the dust aerobiome and the taxa and functional genes that were common to all atmospheric samples examined here.The taxa and functional genes that displayed a signi cant correlation with PM 10 concentrations, Spearman's |r| >0.75, and q-value < 0.05 were then further clustered (k-means) based on clr-transformed counts in all datasets examined in this study, with the optimal number of clusters determined by the total within-sum value of squares.The list of signi cantly correlated taxa and genes is presented in Tables S4a and b, along with their clusters.
PM 10 -correlated genera A total of 137 genera were negatively correlated with PM 10 concentrations.Most of these genera were bacterial (80 genera), a signi cant amount were fungal (56 genera) and only one was archaeal.The genera that were negatively correlated with PM 10 concentrations were of the phyla Proteobacteria, Firmicutes, Bacteroidetes and Ascomycota.These genera formed 6 clusters, as presented in Figs. 3 and   S1.Clusters 1, 2 and 6 represent fungal genera that were more abundant in atmospheric samples, i.e., clouds, urban air, Beijing smog, Israeli dust and Red Sea dust.They consisted entirely of fungi, mostly Ascomycota.Cluster 1 represents genera that were found only in atmospheric samples, and very often in all of them; these genera were absent or rarely found in most soil and marine samples, suggesting that their main source was not local soils or seas.This cluster consisted of the following 13 fungal genera: Allomyces, Cordyceps, Dissoconium, Epidermophyton, Glomus, Hypocrea, Lecanicillium, Metarhizium, Millerozyma, Mycosphaerella, Phycomyces, Rhizopus and Tilletia.Cluster 2 genera were found in some marine samples, as well as atmospheric samples; accordingly, they included some marine fungi, e.g., Debaryomyces and Meyerozyma.The fungi included in cluster 6 were found in all examined environments but were relatively more abundant in atmospheric samples than in marine and soil samples.These included Neosartorya, Aspergillus, Gibberella, Phaeosphaeria and Penicillium (Table S4a).Some of these pathogens are animal and human pathogens (e.g., Neosartorya and Aspergillus), and some are plant pathogens (e.g., Gibberella).
Cluster 5 included Bacteroidetes, Firmicutes and Gammaproteobacteria taxa that are ubiquitous in soil and aquatic ecosystems, and clusters 3 and 4 were composed almost entirely of Firmicutes, which were less abundant in urban air and clouds.The latter three clusters included almost solely bacterial genera.A single archaeon, Methanobrevibacter, a methanogen often found in the digestive systems of a diverse array of animals, ranging from ruminants to termites 51 , was found in cluster 4.Many of the bacteria included in these clusters were human or animal commensals or pathogens, such as Escherichia, Bacteroides, Helicobacter, Histophilus, Salmonella and Enterobacter.In general, more human-associated bacteria and more fungi were linked to low PM 10 concentrations than to high PM 10 concentrations, suggesting that they were from local sources.
Most of the 321 genera that were positively correlated with PM 10 concentrations were bacteria (260 genera), some were archaea (52 genera), 6 were algae and 3 were fungi (Figs. 4 and S1).Proteobacteria was the most prevalent phylum of bacteria (126 genera), with genera such as Burkholderia, Bradyrhizobium, Rhodopseudomonas, Sinorhizobium and Roseobacter that are ubiquitous in soil and marine environments.9][60] This was also true for speci c extremophile archaeal genera such as Aeropyrum, Methanothermococcus and Desulfurococcus.
Genera that were positively correlated with PM 10 concentrations in Dust IL formed three distinct clusters (Figs. 4 and S1).Most of the positively correlated genera were less abundant in cloud and urban air samples than in all other samples.Cluster 1, consisting of 188 genera, represents genera that were generally highly abundant in all samples but to a lesser extent in cloud and urban air.This cluster included only bacteria, mostly from the Proteobacteria and Actinobacteria phyla, e.g., Streptomyces, Mycobacterium, Burkholderia, Frankia and Arthrobacter.Many of these bacteria are soil bacteria, and some are extremophiles.
Cluster 2, consisting of 124 genera, included a majority of the archaea that were positively correlated with PM 10 concentrations, i.e., 50 archaea out of 52 archaeal genera.Many of these species were extremophiles, e.g., thermophilic Aeropyrum and Thermoplasma and halophilic Halalkalicoccus.
Methanogens such as Methanopyrus were also found.Bacteria in this cluster were mostly of the Proteobacteria and Firmicutes phyla.This cluster represents genera that were found in dust, soil and marine samples but were rare in smog, cloud and urban air samples.Thus, this cluster might best represent the contribution of desert soil microbiomes to the dust aerobiome.
Last, cluster 3 represents 9 genera that were rarely found in all samples.Only three of these genera were bacteria: Fluoribacter (Legionella), which contains species that often reside in protozoa, and some that can cause Legionellosis 61 ; the freshwater cyanobacterium Leptolyngbya; and Candidatus Hodgkinia, a cicada endosymbiont.Three other genera were of the algal phylum Chlorophyta, and three were fungi.
PM 10 -correlated functional genes The same analysis was applied to functional annotation, i.e., KEGG annotation.The results are presented in Figs. 5, 6 and S2.A total of 1516 genes were positively correlated with PM 10 concentrations, compared with only 565 genes that were negatively correlated with PM 10 concentrations.Most of the positively correlated genes, i.e., 985 genes, were related to metabolism, whereas most of the negatively correlated genes were related to genetic information processing (278 genes).The negatively correlated genes included many eukaryotic genes and pathways, such as ribosome biogenesis in eukaryotes (ko03008), RNA transport (ko03013), mRNA surveillance pathway (ko03015), spliceosome (ko03040) and cell cycleyeast (ko04111).
In general, the functional pro le of genes correlated with PM 10 concentrations was in accordance with the community composition: low PM 10 concentrations were accompanied by increased fungal abundance and eukaryotic functional genes, whereas increasing PM 10 concentrations associated with desert dust storms resulted in increasing bacterial and archaeal taxon and gene abundance.
Functional genes that were negatively correlated with PM 10 concentrations formed four distinct clusters (Figs. 5 and S2); clusters 1 and 2, comprising 377 genes, represent genes that were usually more abundant in clouds, urban air, Beijing smog and most Israeli dust samples, except for Dust_6 to Dust_8, and less abundant in Israeli soils and Mediterranean and Red Sea samples.These included mostly eukaryote-speci c genes, such as those related to cell cycle -yeast (ko04111) and meiosis -yeast (ko04113), as well as prokaryotic genes encoding the transport of trehalose, sucrose, maltose galactitol, lactose and cellobiose via the phosphotransferase system pathway (ko02060).Stress adaptation genes were also found in these clusters: SSK1 (K11233), SOH1 (K11246) and YPD1 (K11232) were related to osmotic stress in fungi; bceA and bceB (K11631, K11632) were associated with bacterial resistance to bacitracin; and vraS (K07681) was associated with bacterial resistance to the antibiotic vancomycin.
Clusters 3 and 4 represent genes that were found in all samples but were usually more abundant in clouds, urban air, Beijing smog and most samples of Israel dust (except for high PM 10 samples Dust_6 to Dust_8).Cluster 3 was dominated by metabolism-associated genes, whereas cluster 4 was dominated by genetic information processing-associated genes.These clusters included eukaryotic genes related to pathways such as endocytosis, phagosome and RNA transport.In addition, prokaryotic genes encoding the transport of mannose, cellobiose, and beta-glycoside via the prokaryotic phosphotransferase system pathway (ko02060).The spo0A gene, which is associated with sporulation, especially in Bacillus, and the SLN1 fungal gene, which is associated with osmotic stress, were also found in cluster 3.
Genes that were positively correlated with the PM 10 level formed 3 clusters (Fig. 6 and S2).Cluster 1 comprised 717 genes that were highly abundant in all samples but less abundant in the cloud and urban air samples.Many of these genes were associated with general functions, e.g., ribosome (ko03010), DNA replication (ko03030), and aminoacyl-tRNA biosynthesis (ko00970).They also included many bacterialand archaeal-speci c genes, such as genes involved in the bacterial secretion system (ko03070), agellar assembly (ko02040), bacterial chemotaxis (ko02030), and peptidoglycan biosynthesis (ko00550).
Cluster 3 comprised 604 genes that were found in most samples, excluding clouds and urban air, but to a lesser extent than genes included in cluster 1.Many of these genes were associated with xenobiotics biodegradation and metabolism e.g., chlorocyclohexane and chlorobenzene degradation (ko00361) and benzoate degradation (ko00362).ABC transporter genes and two-component system genes were also abundant in this cluster, with 85 and 45 genes, respectively.
Finally, cluster 2, comprising a total of 197 genes, represented genes that were rare in most samples.These included mostly metabolism-associated genes, e.g., amino acid metabolism and carbohydrate metabolism genes, as well as genes associated with environmental information processing, e.g., twocomponent system and ABC transporter genes.
The genes that were positively correlated with PM 10 were usually similarly abundant in most samples, excluding cloud and urban air; thus, pathways that were highly abundant in Israeli dust were also highly abundant in Israeli soils, Mediterranean and Red Sea water, Red Sea dust and Beijing smog.This implies that these pathways are common to bacteria everywhere and that their low abundance in clouds and urban air is in accordance with the signi cantly higher abundance of fungi than of bacteria and archaea in these samples.No unique contribution of the marine microbiome to dust collected in Israel or over the Red Sea was observed in this analysis.

Functional differences between dust and other environments
We assigned KEGG gene annotations to environmentally relevant functional categories (Table S5) 62,63 and calculated their relative abundances as percentages of the total gene counts.These functional categories were speci cally chosen since they do not include general housekeeping genes but are usually unique to speci c environmental functions such as metabolism and adaptation to stress.A full description of the abundances of the selected genes and functional categories can be found in Table S6.
Previous studies have suggested that the adaptation of airborne microorganisms to atmospheric conditions would bene t from the presence of functions such as UV resistance, osmotic stress resistance, envelope stress resistance, sporulation, methane oxidation, formate degradation, acetic acid degradation and oxidative stress resistance. 3,8,64,65Thus, these studies targeted these functions in metagenomic as well as metatranscriptomic analyses of different atmospheric samples.Consistent with this, we focused on these functional genes, as well as many other environmentally relevant functional genes, in our analysis to better de ne the contribution of mineral dust to the potential functionality of the aerobiome.
According to our results, dust samples often differed from soil samples in the composition of the selected functions; out of a total of 58 examined functions, 36 showed signi cant differences in abundance between soil and dust samples (Kruskal-Wallis test, q-value < 0.05).Dust also differed from air samples, i.e., urban air and clouds displayed 35 functions that were differentially abundant between the dust and air sample types (Kruskal-Wallis test, q-value < 0.05).
Figure 7 shows the relative abundances of chosen functions in the different sample types.Dust samples showed a high relative abundance of antibiotic resistance genes, speci cally genes conferring resistance to beta-lactam antibiotics, bacitracin and vancomycin, compared with sea and soil samples.Moreover, dust showed the highest relative abundance of these genes compared with all other atmospheric samples.These genes are often found in Bacillus species and other spore-forming bacteria, along with genes associated with sporulation.
Comparison of all atmospheric sample types, i.e., dust, air and smog, suggested that the Beijing smog metagenome had the highest abundance of functions linked with microbial survival in the atmosphere, including temperature, oxidative stress and UV resistance, as well as formate degradation.Dust also exhibited a high relative abundance of relevant functions, such as osmotic stress and envelope stress resistance and sporulation-associated genes.In addition, dust displayed a high relative abundance of antibiotic resistance genes and genes for the biodegradation of aromatic compounds.Clouds and urban air samples exhibited a higher abundance of fungal resistance genes for cell wall stress and starvation, in accordance with their taxonomic composition.
Comparison of the list of functions with PM 10 -correlated genes revealed that most stress resistanceassociated genes were positively correlated with PM 10 concentrations, as were genes associated with the degradation of aromatic compounds, whereas fungal genes associated with stress resistance were negatively correlated with PM 10 concentrations.Antibiotic resistance genes displayed a mixed trend, where bacitracin resistance genes were negatively correlated with PM 10 concentrations and most vancomycin resistance genes were positively correlated with PM 10 concentrations.

Discussion
Dust is a well-recognized vector for the long-range microbial transport of microorganisms and pollution. 12,64,66,67Recent studies suggest that aerosol microbiomes are most affected by the microbiomes of underlying environments. 19,64,68Variations over seasons and in airmass origin were also linked to changes in aerosol microbial community composition. 1,2,6,9,69,70The aerobiome community composition displays mixtures of sources, and detecting a constant and stable indigenous microbial community was not yet reported.As such, the identi ed microorganisms may serve as tracers for the sources contributing to the aerobiome.Previous studies comparing aerobiomes to possible sources have shown that soils, leaf surfaces, chicken coops, humans and farm animals contribute to the composition of the aerobiome to varying extents.However, most of these studies relied on amplicon sequencing of taxonomic genetic markers, e.g., the 16S rRNA-encoding gene; focused on a single kingdom, e.g., bacteria; and overlooked functional differences between closely related taxa.
In this study, we sampled dust under different atmospheric conditions, representing various sources, transport trajectories and different concentrations of mineral dust expressed by varying concentrations of PM 10 .The collected dust metagenomes were compared to metagenomes from three major environments: atmospheric (air and smog sample types), terrestrial (soils) and marine.This comparison, based on taxonomic classi cation as well as functional gene annotation of metagenomes, was conducted to better understand the similarities and differences among the different microbiomes, infer their functional capacities, and potentially elucidate the contribution of different sources to the atmospheric microbiomes.
Our results show that the metagenomes of dust sampled in Israel in this study most closely resembled dust metagenomes sampled in Saudi Arabia, close to the Red Sea, over 1000 km away, on different sampling dates 3 .A common source that is attributed to soils contributed to the two dust datasets, according to the SourceTracker analysis.However, variability within Dust IL and Dust Red Sea samples was observed.This variability in Dust IL was correlated with PM 10 concentrations, i.e., a higher relative contribution of the soil microbiome was observed in samples with higher PM 10 concentrations.This is expected for this region, since a major source of PM 10 is mineral dust originating in desert soils.Dust IL samples with low PM 10 loading showed higher similarity to atmospheric microbiomes of cloud condensation samples collected in France (cloud samples) and of urban areas in the USA (urban air samples).Sea sample metagenomes contributed very little to all the dust samples, i.e., both Dust IL and Dust Red Sea, despite the proximity to the Mediterranean and Red Seas, respectively.Some of the trajectories of the dust sampled in Israel included a marine component, which was previously linked to an effect on the aerobiome composition 7 .Moreover, all Dust Red Sea samples were collected in a location adjacent to the Red Sea, and some were even collected on a ship in the Red Sea.Nonetheless, the contribution of the Mediterranean and Red Seas to the dust microbiomes was always below 3.1%.For reference, the contribution of the air microbiomes sampled in France (cloud dataset) and North America (urban air dataset) to the dust microbiome ranged between 3.8% and 12% (Fig. 2).This suggests that the marine environment does not contribute much to transported dust.This conclusion is consistent with another study showing that the marine environment did not signi cantly contribute to the airborne microbiome over the Atlantic and Paci c Oceans. 71−76 Since elevated PM 10 concentrations in Israel usually indicate dust events, we looked for correlations between PM 10 concentrations and taxonomic and functional gene composition in Dust IL samples.The results indicated that dust storms changed the aerobiome community composition and the functional pro le from a fungal-dominated environment to a bacterial-dominated environment.A signi cant part of the functions that were previously suggested to be linked to microbial survival in the atmosphere, e.g., UV, oxidative and envelope stress resistance, was positively correlated with PM 10 concentrations, suggesting that dust-borne bacteria might be a global source of viable or even active bacteria in the atmosphere.
Genes associated with the biodegradation of organic contaminants such as aromatic compounds were relatively more abundant in dust than in any other environment examined and positively correlated with PM 10 concentrations.Most of these genes were also common to soils.Previous studies indicated the cometabolism of organic contaminants along with methane and ammonia, due to the presence of di-or monooxygenases with a broad substrate range 77 .Accordingly, we found that along with genes associated with the biodegradation of organic contaminants, methane metabolism genes were also positively correlated with PM 10 concentration, as well as genes associated with glyoxylate and dicarboxylate metabolism, which includes formate metabolism.All or some of these substances might be relevant to metabolic activity in the atmosphere due to their biodegradation, as suggested in other studies, 65,78 as well as in dust-affected terrestrial and marine regions.
We found that genes encoding beta-lactam and bacitracin resistance were more abundant in transported dust than in other atmospheric samples.Beta-lactam antibiotics such as penicillins and cephalosporins are commonly administered to humans as well as to food-producing animals such as cattle 79 , and the ndings suggest that an anthropogenic source contributes to the dust microbiome, as was suggested previously by Mazar et al and Gat et al, who showed that antibiotic resistance genes are found in aerosols on dusty and nondusty days and that their source is more likely anthropogenic than natural. 4,7.
Bacitracin is also a common antibiotic that is administered mostly as an ointment to prevent skin infections, yet resistance to bacitracin is commonly found in Bacillus species and other Firmicutes, suggesting that its source might be natural.The presence of resistance to these two antibiotics in dust metagenomes should raise concerns regarding the spread of this resistance and the possible effects on public health.
Overall, fewer KEGG-characterized functional genes were negatively correlated with PM 10 concentrations than were positively correlated with PM 10 (566 vs. 1518, respectively).This observation emphasizes the role played by dust storms in changing the microbial community composition from a fungal-dominated community to a bacterial-dominated community in the Mediterranean, as well as in the dispersal of environmental functions.Given the changes in dust storm frequency and severity due to global climatic changes, 26,80 the possible implications of such transport should be further investigated.
Interestingly, the SourceTracker analysis showed that low-PM 10 dust sample microbiomes signi cantly overlapped with cloud and urban air sample microbiomes.These common taxa and associated functional genes were also negatively correlated with PM 10 concentrations.They were more abundant in the aerobiome than in the soil and sea microbiomes and were mostly fungal.We suggest that these fungal taxa constitute a "core microbiome" for the atmosphere and provide insights into their possible functioning.We found that in general, the relative abundance of fungi, especially those of the phylum Ascomycota, increased when PM 10 concentrations decreased, with 13 fungal genera found to be common to all atmospheric datasets and only to these datasets.If we extend our de nition of "core microbiome" and include taxa that were negatively correlated with PM 10 and were rare in soils and marine samples (clusters 1, 2 and 6, Fig. 3), we obtain 56 fungal genera, mostly belonging to Ascomycota, with highly varied environments and functions.Some of these fungi might be carried by long-range transport, such as Cordyceps, Paracoccidioides and Allomyces, which are often found in tropical regions.Other fungi are animal and plant commensals or pathogens, such as Malassezia, Tilletia, Epidermophyton, Mycosphaerella, Arthroderma, Botryotinia and Gibberella.Studies indicate that fungal spores are actively emitted and dispersed into the atmosphere, 21,81−86 which explains their ubiquity in atmospheric samples.
Another source of fungal spores in the atmosphere is biomass burning, 87,88 as re smoke is transported great distances in the atmosphere. 89,90Both processes could explain the presence of Ascomycota in all atmospheric metagenomes, as well as their negative correlation with high PM 10 concentrations.
Often overlooked members of the aerobiome are archaea.In this study, we found a clear correlation between PM 10 concentrations and archaeal abundance in Dust IL samples.Many of these archaea were anaerobic methanogens; accordingly, some methanogenesis-associated genes were positively correlated with PM 10 concentrations.Methanogenesis-associated genes were also more abundant in dust than in other atmospheric samples.Methanogenesis is carried out solely by archaea under strictly anaerobic conditions, usually in highly reduced soils or sediments but also in the digestive system and feces of ruminants as well as chicken and other farm animals. 91,92Ruminant feces found in farms or grazing grounds or chicken feces found in chicken coops could have contributed to the dust metagenome.4][95] This nding may suggest that atmospheric conditions that result in high PM 10 concentrations may also result in the suspension of particles other than soil and mineral dust.It also suggests that some methanogens in the atmosphere can mark anthropogenic source contributions.The results from this study, showing the enrichment of antibiotic resistance genes in the aerobiome, corroborate this suggestion.Previous studies have suggested that a dust storm collects material over its course, and its microbial community composition is affected accordingly. 7,96In this manner, a dust storm that passes over an urban, agricultural or industrialized area is likely to increase anthropogenic emissions of microorganisms into the atmosphere and increase their potential impact on aerobiome functionality and composition.This is consistent with the presence of anthropogenic organic contaminants that adhere to transported dust, as shown by Falkovich et al. 97,98 The importance of anthropogenic in uence on the atmospheric microbiome is only now gaining attention. 99It is predicted that the aerobiome will be in uenced by the changing chemical and physical characteristics of the atmosphere due to anthropogenic activities such as cooking, agricultural practices and land use change.Climatic changes such as increased temperatures, res and drought are projected to increase the amount of dust emitted to the atmosphere, 26 thus increasing the contribution of soil microbiomes to the composition of the aerobiome.The increasing frequency and intensity of wild res 100,101 also increases the concentration of particles in the atmosphere.These serve as nuclei for bacterial transport in the atmosphere, 99 thereby increasing their dispersal and potentially the functional scope of the aerobiome.Recent evidence on the possible metabolic activity of bacteria in the atmosphere suggests that speci c carbon sources are likely to sustain bacteria, e.g., formate, acetic acid and ethanol. 8,65,78Moreover, it was suggested that bacterial metabolism might also affect cloud or even atmospheric chemistry. 65,102Our results suggest that if dust storms carry viable bacteria, they may be metabolically relevant for atmospheric chemical cycles and may spread these bacteria globally.
In this study, we investigated the links between the surface and atmospheric microbiomes, as well as between different atmospheric microbiomes.We showed that different atmospheric niches display different functional pro les and community structures and that they share a common background dominated by fungi.We also showed that two similar niches, i.e., the dust collected in Israel and over the Red Sea on different, nonsynchronized occasions, were strikingly similar, emphasizing the importance of the source of the dust in determining its composition on a regional and possible global scale.As expected, dust showed high resemblance to soils both in community structure and functional pro le, yet the dust samples harbored functional genes that are likely to have been derived from an anthropogenic source.

Figure 6 K
Figure 6

Table 1
Sample description of dust collected in Israel.Samples Dust_2 to Dust_5 were created by pooling DNA from several sampling events.

Table 2
Metagenomes used for comparison with the dust collected in this study.