Characterization of active and total fungal communities in the atmosphere over the Amazon rainforest

Abstract. Fungi are ubiquitous in the atmosphere and may play an important role in atmospheric processes. We investigated the composition and diversity of fungal communities over the Amazon rainforest canopy and compared these communities to fungal communities found in terrestrial environments. We characterized the total fungal community and the metabolically active portion of the community using high-throughput DNA and RNA sequencing and compared these data to predictions generated by a mass-balance model. We found that the total community was primarily comprised of fungi from the phylum Basidiomycota. In contrast, the active community was primarily composed of members of the phylum Ascomycota and included a high relative abundance of lichen fungi, which were not detected in the total community. The relative abundance of Basidiomycota and Ascomycota in the total and active communities was consistent with our model predictions, suggesting that this result was driven by the relative size and number of spores produced by these groups. When compared to other environments, fungal communities in the atmosphere were most similar to communities found in tropical soils and leaf surfaces. Our results demonstrate that there are significant differences in the composition of the total and active fungal communities in the atmosphere, and that lichen fungi, which have been shown to be efficient ice nucleators, may be abundant members of active atmospheric fungal communities over the forest canopy.


Introduction
Fungi are critical to the functioning of terrestrial ecosystems and may also play an important role in the functioning of the atmosphere. Fungi are abundant and ubiquitous in the atmosphere, with an estimated global land surface emission rate of 50 Tg year not. The ice nucleation efficiency of fungal cells also likely depends on their physiological state; vegetative cells derived from potentially active fungi are more efficient ice nucleators than spores. Vegetative forms of Fusarium (a filamentous fungi) as well as several lichen fungi have been shown to nucleate ice at temperatures as warm as −1 • C (Després et al., 2012) ( Fig. S1 in the Supplement), and ice nucleation by hyphae has 15 been observed at −2.5 • C (Pouleur et al., 1992). In contrast, dormant spores -particularly those with surface hydrophobins -are generally poor ice nucleators. For example, ice nucleation of rust (Puccinia) spores requires temperatures lower than −10 • C ( Morris et al., 2013), and Penicillium spores nucleate ice at temperatures of approximately −22 • C (Iannone et al., 2011). 20 Despite its importance, we know relatively little about the physiological state of fungal cells in the atmosphere. Specifically we know little about the taxonomic composition of metabolically active airborne fungi and how this compares to the composition of the total fungal community. One way to survey the total and active communities is to measure community composition from rDNA (i.e. rRNA genes) and rRNA in ribosomes. Introduction gal communities in soils and on decaying plant material (Baldrian et al., 2012;Barnard et al., 2013Barnard et al., , 2014Rajala et al., 2011) but has not been applied to fungal communities in the atmosphere. Information about the taxonomic composition of airborne fungi that are present in different physiological states could be used to advance atmospheric science in multiple 5 ways. For example, such data could be used to improve estimates of the ice nucleating capacity of fungal bioaerosols. Recent estimates of the ice nucleating capacity of fungal bioaerosols based on culture-based approaches -the abundance of colony forming units (CFUs) -have led some scientists to conclude that atmospheric fungi have a low ice nucleation efficiency (Iannone et al., 2011). However, these culturebased data may be misleading, as the vast majority of fungi require identification using culture-independent approaches (Borneman and Hartin, 2000). Culture-independent identification of active fungal taxa sampled from the atmosphere could be used to direct selective culturing of potentially important fungi to test their ice nucleation efficiencies and their metabolic capabilities in the laboratory. 15 In this study, we used culture-independent approaches to measure the composition of total and active atmospheric fungal communities in situ and a mass-balance model to aid in the interpretation of our results. Our study system is the atmosphere above the Amazon rainforest canopy. We chose this system because fungal bioaerosols make up a substantial proportion of aerosol particulate matter over the Amazon (Elbert et al., 20 2007;Heald and Spracklen, 2009) and are estimated to be a dominant force responsible for cloud formation over the Amazon . We used a combined approach of DNA and RNA sequencing to address the following questions: (1) What is the composition of total airborne fungal communities? (2) What is the composition of active airborne fungal communities? (3) What likely drives differences in the com-25 position of the total and active airborne fungal communities? (4) Is the diversity and structure of fungal communities in the atmosphere similar to that found in terrestrial environments?  -5) and eluted into 50 µL. cDNA was synthesized from the total RNA extract using the SuperScript II First-Strand Synthesis System (Invitrogen, Life Technologies Corporation) with random hexamers. All RNA was converted into cDNA in six synthesis reactions and one reverse transcriptase 5 negative control reaction.

Library preparation and sequencing
To increase the concentration of cDNA to levels required for sequencing, we used multiple displacement amplification (GenomiPhi V2, GE Healthcare) according to the protocol described in Gilbert et al. (2010) including second-stand synthesis, amplifica-10 tion, and de-branching of amplification products. The fully de-branched products were sheared by sonication (24 cycles, 30 s each) using the Bioruptor sonication system (Diagenode The D1-D2 region of the large subunit (LSU) rRNA gene was targeted using PCR with the primers LR0R (5'-ACCCGCTGAACTTAAGC-3') and LR3 (5'-CCGTGTTTCAAGACGGG-3') (http://sites.biology.duke.edu/fungi/mycolab/primers. htm). LSU amplicon libraries were prepared using a two-stage PCR procedure as described in (Kembel and Mueller, 2014) using unique combinatorial barcodes (Gloor et al., 2010) to identify samples (Table S2).

Metatranscriptome
Overlapping paired end reads were aligned and joined using fastq-join (https://code. google.com/p/ea-utils/wiki/FastqJoin). Joined reads and non-overlapping single-end reads were trimmed and filtered using PrinSeq (Schmieder and Edwards, 2011). Sequences < 75 bp, > 2 % Ns, and/or mean quality score < 20 were removed. Sequence 20 artifacts defined as exact duplicates with > 5000 sequences were removed. Sequences in the 10 December sample were primarily artifacts, so this metatranscriptome sample was excluded from further analysis. Putative rRNAs in the remaining sequences were identified using SortMeRNA (Kopylova et al., 2012) with the non-redundant version of the following databases: rfam 5.8S (version 11.0) (Burge et al., 2013)  . Of 5 165 185 quality-filtered reads, 1 915 994 with an average length of 137.5 bp were identified as putative rRNAs (Table S3).

LSU amplicons
Forward and reverse barcodes were combined to make a 12 bp barcode on the forward read. Only forward reads derived from the LR3 region were used for analysis. This re-5 gion has been shown to have high species-level resolution even with short read lengths (Liu et al., 2012).

Multi-environment sequences
LSU sequences from four soil studies (Barnard et al., 2013;Kerekes et al., 2013;Penton et al., 2013Penton et al., , 2014 and one phyllosphere study (Kembel and Mueller, 2014) were compared to air samples collected for this study (Table S4). Raw sequence data and associated metadata were downloaded from publically available databases. 12 bp barcodes were added to all sequences to identify each sample in downstream analysis.

LSU amplicon and metatranscriptome sequence processing
All sequences were processed in QIIME version 1.7 (Caporaso et al., 2010). Briefly, libraries were individually demultiplexed and filtered for quality. Sequences with an average quality score less than 20, shorter than 150 bp and with greater than 2 primer mismatches were discarded. The same parameters were used across all samples except the metatranscriptome rRNAs were a size cut off of greater than 75 bp was used. Sequences from Kembel and Mueller (2014) and Penton et al. (2014) were randomly 20 subsampled to 25 and 60 % respectively. Sequences were clustered into operational taxonomic units (OTUs) at 97 % sequence similarity using closed reference BLAST (Altschul et al., 1990)  Following sequence processing and quality filtering, a total of 55 414 amplicon and 1 915 994 metatranscriptome LSU sequences generated for this study and 1 577 458 LSU sequences from soil and phyllosphere studies were retained (Table S3). For analyses using only samples from this study, the data were rarefied to 5300 sequences per sample. For analyses that compare samples in this study to samples from other 5 studies, the data were rarefied to 500 sequences per sample.

Statistical analyses and data availability
All statistical analyses were conducted in R (R Core Team, 2014) primarily using the vegan (Oksanen et al., 2013) package for ecological statistics and the ggplot2 (Wickham, 2009) package for visualizations. 10 Sequence files and metadata have been deposited in Figshare (doi:10.6084/m9.figshare.1335851). Data from other studies used for cross environment analyses are available using the databases and identifiers referenced in the respective manuscripts. 15 We use a global, well-mixed, one-box material-balance model to predict the relative abundances of fungal cells measured as gene copies sampled in the active and total portions of atmospheric bioaerosols. Model description and details are available in the Appendix.

Basidiomycota dominate total airborne fungal communities
Measurements of airborne fungi using culture-based methods such as quantifying spore and colony-forming unit counts have been conducted for centuries (Després In comparison, there have been few culture-independent studies of the fungal composition of atmospheric samples (e.g. Boreson et al., 2004;Bowers et al., 2013;Fierer et al., 2008;Fröhlich-Nowoisky et al., 2009Pashley et al., 2012;Yamamoto et al., 2012). Using a culture-independent approach, we found the composition of total airborne fungal communities primarily included taxa belonging to the phyla 5 Ascomycota and Basidiomycota (Fig. 1). This result is similar to what is observed in environments on the Earth's surface (James et al., 2006) and what has been reported in other studies of fungi in the atmosphere (Bowers et al., 2013;Fröhlich-Nowoisky et al., 2009Yamamoto et al., 2012). Basidiomycota dominated the total airborne community in our air samples (mean relative abundance = 90.2 ± 6.9 %) ( Fig. 1). Within the phylum Basidiomycota, Agaricomycetes were the most abundant class in our samples. Agaricomycetes have been previously detected in air samples (Fröhlich-Nowoisky et al., 2012;Woo et al., 2013;Yamamoto et al., 2012) and are common in tropical soils (Tedersoo et al., 2014) and leaf surfaces (Kembel and Mueller, 2014). Within the Agaricomycetes, the most abundant order was the Polyporales (mean = 55.7±2.3 %). Polyporales have been detected in culture-independent studies of urban aerosols (Yamamoto et al., 2012) and culturable representatives have been isolated from cloud water (Amato et al., 2007). Given that these are largely saprotrophic (i.e. wood-decay) fungi (Binder et al., 2013;Larsson et al., 2007), it is parsimonious to assume there is a significant local source of 20 Polyporales on the forest floor. The presence of Agaricomycetes may have implications for atmospheric processes. Ice nucleation efficiency within the Agaricomycetes is variable, with some taxa capable of nucleating ice at temperatures as warm as −17 • C (Haga et al., 2014) (Fig. S1).
These temperatures are warmer than what has been measured for Penicillium spores 25 (Iannone et al., 2011) although not as warm as what has been measured for other spore types (Morris et al., 2013), hyphal fragments (Pouleur et al., 1992), and lichen fungi (Després et al., 2012). Despite the low ice nucleation efficiency of some taxa in this group, given the high abundance of Agaricomycetes over the forest canopy, this group could still have a significant impact on cloud formation and precipitation in the tropics.

Ascomycota dominate active airborne fungal communities
The composition of total and active fungal communities over the Amazon rainforest canopy significantly differed (ADONIS, R 2 = 0.342, p = 0.029). The active community 5 in the atmosphere over the forest canopy was dominated by Ascomycota (mean relative abundance = 80.4 ± 20 %) (Fig. 1). Basidiomycota comprised a smaller fraction of the sampled genes (mean = 7.3 ± 6.8 %) with the remainder of identified sequences belonging to the phyla Chytridiomycota and Glomeromycota. This result makes sense in light of the natural histories of many of the Ascomycota, which have single-celled 10 or filamentous vegetative growth forms that are small enough to become aerosolized, while many of the Basidiomycota are too large to be easily aerosolized, other than in the form of metabolically inactive spores. The most abundant classes of Ascomycota detected were Sordariomycetes (mean = 27.1 ± 6.6 %), and Lecanoromycetes (mean = 17.5 ± 7.6 %). Sordariomycetes 15 have been detected in air samples (Fröhlich-Nowoisky et al., 2009Yamamoto et al., 2012) and have been shown to be abundant on tropical tree leaves (Kembel and Mueller, 2014) and tropical soils . In most ecosystems, Sordariomycetes are endophytes, pathogens, and saprotophs (Zhang et al., 2007). Xylariales, which includes both endophytes and plant pathogens (Zhang et al., 2007), was the 20 most abundant order within the Sordariomycetes in our samples.
Lecanoromycetes were the second most abundant class of Ascomycota detected over the rainforest canopy. This group has been detected in other culture-independent studies of fungi in the atmosphere (Fröhlich-Nowoisky et al., 2012;Yamamoto et al., 2012). The Lecanoromycetes contain 90 % of the lichen-associated fungi (Miad-25 likowska et al., 2007). Lichens are a symbiosis between a fungus and a photosynthetic partner such as eukaryotic algae or cyanobacteria. Lichens are known to be hardy and may be particularly well-adapted for long distance transport and metabolic activity in 7187 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the atmosphere. Lichens are often the dominant life forms in environments that have conditions similar to those found in the atmosphere, including low water (Kranner et al., 2008) and nutrient availability, wide temperature variations, and high UV irradiance (e.g. Solhaug et al., 2003;Onofri et al., 2004). Another notable trait of lichens is their efficient ice nucleation capacity. Although  (Kieft, 1988). These studies have demonstrated that lichens are among the most efficient biological ice nucleators. Therefore their presence in the atmosphere may have a significant impact on cloud formation and precipitation. This ice nucleation capacity may also enable lichens to control the extent of their dispersal through the atmosphere. It 15 is possible that lichens achieve this by nucleating ice formation, which leads to precipitation and ultimately deposition. This phenomenon has been shown to occur in some phytopathogenic bacteria (Morris et al., 2008(Morris et al., , 2010 and potentially fungi as well (Morris et al., 2013).

active communities is consistent with mass-balance predictions
Our mass balance model (Appendix) predicted Basidiomycota would dominate the total community because they produce orders of magnitude more spores and have smaller aerodynamic diameters compared to Ascomycota. Consistent with this prediction, the total airborne community was dominated by Basidiomycota in our air samples (mean 25 relative abundance = 90.2 ± 6.9 %) (Fig. 1). There have been some empirical studies reporting the opposite pattern, with a higher relative abundance of Ascomycota compared to Basidiomycota (Bowers et al., 2013;Fierer et al., 2008;Pashley et al., 2012 There has been one study focused on airborne fungal communities in the Amazon Basin (Fröhlich-Nowoisky et al., 2012). Although the site of this study was the atmosphere above a rural pasture (vs. a tropical rainforest, as in our study) these investigators also found that Basidiomycota dominate airborne fungal communities Our mass-balance model explains the differences in composition between the total 5 and active air communities. However some of the differences we observed may be partially attributable to the use of different approaches in characterizing the total and active communities. In this study, the total community was characterized by PCR-based amplification and sequencing of LSU genes, whereas the active community was characterized through random sequencing of all the RNA present in the samples. Shotgun metatranscriptome sequencing and PCR-based community characterization approaches each have their own biases (Hong et al., 2009;Morgan et al., 2010). Our data suggest that the selection of LSU primers led to biased results. For example, the high relative abundance of lichen fungi in the active community was unexpected because this group was not present in the total community and has only been detected 15 in low abundance in other PCR-based studies of fungi in the atmosphere (Fröhlich-Nowoisky et al., 2012). We tested the LR0R-LR3 primer pair using the SILVA Test-Prime tool (Klindworth et al., 2013) and found coverage of the Lecanoromycetes with this primer pair was 71.4 %. Importantly, the order Teloschistales, which contains the most abundant species in the active community, would not be detected with this primer 20 pair. However the general pattern that Ascomycota were much less abundant than Basidiomycota in the total community is not likely due to primer bias as overage of the phylum Ascomycota by the LR0R-LR3 primer pair is 85.5 % according to TestPrime. Our findings underscore the value of using a combination of PCR-based and shotgunbased sequencing approaches, particularly in environments that are understudied and

Fungal air communities above the forest canopy are most similar in composition to tropical phyllosphere and soil communities
We compared total and active fungal air communities to communities from tropical, temperate, and tundra soils and from the surfaces of tropical tree leaves. Community composition significantly differed across environment types (ADONIS, R 2 = 0.167, 5 p = 0.001), and fungal communities in the atmosphere were compositionally distinct from communities in other environments (Fig. 2). Ascomycota was the most abundant phylum across all soil and phyllosphere samples (soil mean relative abundance = 78.4 ± 14.9 %, phyllosphere = 90.9 ± 4.9 %) followed by Basidiomycota (soil mean relative abundance = 19.0 ± 14.9 %, phyllosphere = 7.4 ± 4.5 %) (Fig. 3). We ex-10 pected communities to be distinct across habitat types because environmental conditions may differ across the habitat types and select for different communities. However, in the atmosphere, dispersal and mixing of fungi from multiple habitat types may be driving the observed community composition differences instead of environmental selection. 15 The diversity of fungal communities in the atmosphere is within the range of diversities reported for terrestrial environments, including those of tropical leaf surfaces, tropical soils, temperate grassland soils, and tundra soils. Overall taxonomic richness, defined as the number of OTUs, significantly varied among environment types (ANOVA, F (5237) = 66.89, p < 0.001) (Fig. S2). Tukey's HSD post-hoc comparisons indicated 20 that the richness of air communities, both total and active, was greater than tundra soil communities and did not significantly differ from temperate grassland soil communities. In general, air communities were less diverse than tropical forest phyllosphere and soil communities with the exception of tropical forest soils and active air communities, which did not significantly differ. Similar patterns have been observed in soil 25 communities where taxonomic richness in arctic soils was significantly lower than soils from temperate and tropical ecosystems (Fierer et al., 2012). Total air communities were most similar to tropical phyllosphere communities (mean Sørensen similarity = 0.015 ± 0.009; statistic) and active air communities were most similar to tropical soil communities (mean Sørensen similarity = 0.010 ± 0.007) (Fig. S3). These results suggest that inputs of fungi into the atmosphere over the canopy are derived from local, as opposed to long-distance, sources. This sugges-5 tion makes sense since fungal spores and hyphae are relatively large aerosol particles with short residence times in the atmosphere, limiting opportunities for long-distance dispersal. While these results are suggestive, detailed information is lacking regarding the potential influence of terrestrial source environments and their role in structuring airborne fungal communities.

Conclusion
Fungi in the atmosphere play an important role in atmospheric processes including precipitation development through ice nucleation. This is of particular significance in the atmosphere over the Amazon rainforest canopy where fungi constitute a large fraction of the total aerosol content (Elbert et al., 2007;Heald and Spracklen, 2009) and pre-15 cipitation is aerosol-limited . Our study represents the first cultureindependent survey of fungal communities over the Amazon rainforest canopy. It is also the first to measure metabolically active microbial communities in the atmosphere using an RNA-based approach. Using this RNA-based approach, we found evidence for the presence of potentially active fungi in the atmosphere, including lichen fungi. While 20 an understanding of the structure of fungal communities in the atmosphere is beginning to emerge, studies on the function of these communities have lagged behind. We suggest that future research focus on understanding the functional capacity of airborne microbes with traits particularly well-suited for survival and metabolic activity in extreme environments. As with any environment, understanding both the structure and function of microbial communities in the atmosphere is needed to assess their potential impact on ecosystem processes such as water and carbon cycling. This study opens Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the door for future investigations of the diversity and function of fungal communities in the atmosphere.

Appendix: Mass-balance model
We use a global, well-mixed, one-box material-balance model to explain the relative abundances of fungal cells measured as gene copies sampled in the active and total 5 portions of atmospheric bioaerosols. By material-balance, for any taxon i within a biological community, the change in time in the abundance of fungal gene copies, N i , must be equal to the difference in source and sinks: Here we assume sources are equal to the emission of fungal gene copies from the 10 Earth's surface into the atmosphere, E i (gene copies hour −1 ). We assume sinks are equal to deposition of fungal gene copies out of the atmosphere back to the Earth's surface, D i = N i k i , (gene copies hour −1 ), where k i (1 hour −1 ) represents a first order deposition coefficient. We can rewrite Eq. (A1) as: 15 We expect the terms E i and k i to vary as a function of life history traits including the method of cell release into the atmosphere, the physiological state of sampled cells, and the aerodynamic diameter of fungal taxa. In this case, Eq. (A2) does not directly represent the entire airborne fungal gene copy abundance. We assume that a first order approximation of fungal bioaerosol behavior may be obtained by subdividing the 20 particle distribution into two modes: vegetative cells, N i ,veg , and spores, N i ,spores . We thus model fungal gene copy abundance as: We can then write and solve parallel versions of Eq. (A2) for each mode. At steady state, the expected gene copy abundance taxa i in each mode is: Our interest lies in the two most common fungal phyla sampled in the atmosphere: 5 Ascomycota, N A , and Basidiomycota, N B . To make predictions about the expected relative abundance of gene copies in these two groups, we make informed assumptions about the relative magnitude of their respective emission and deposition rates. We begin by considering fungal spores. Although a few empirical studies have suggested that Ascomycota are more abundant than Basidiomycota in likely source environments 10 including tropical soils (Kerekes et al., 2013) and leaf surfaces (Kembel and Mueller, 2014), Basidiomycota (e.g. Agaricomycetes, the most abundant class of Basidiomycota in our samples) produce orders of magnitude more spores per individual than Ascomycota (Elbert et al., 2007;Pringle, 2013). For this reason, we assume the emission rate of Basidiomycota spores is much greater than that of Ascomycota spores: Culture-based microscopy data suggests that spores of Ascomycota are typically larger than spores of Basidiomycota (Elbert et al., 2007;Ingold, 2001;Yamamoto et al., 2014). Owing to the difference in spore size, we expect deposition rate of Ascomycota spores to be greater than that of Basidiomycota spores: 12,2015 Characterization of active and total fungal communities Based on these assumptions, it follows that the expected number of Ascomycota spores in the atmosphere will be less than the number of Basidiomycota spores: , spore or N A, spores N B, spores 5 We next consider fungal vegetative cells. Vegetative forms of Ascomycota are generally smaller than vegetative forms of Basidiomycota (Moore et al., 2011). Many Ascomycota grow as filaments or single cells which are small enough to be aerosolized (Després et al., 2012). In contrast, many Basidiomycota grow as mushrooms, which are too large to be aerosolized (although debris from mushrooms and their mycelia can be 10 aerosolized). Due to this difference in the vegetative forms of each group, we expect emission rate of vegetative Ascomycota to be greater than Basidiomycota: , veg No comparative data currently exists on the relative deposition rate of vegetative cells across fungal taxa. Research has shown that at the phylum level, the aerodynamic 15 diameter of Ascomycota is greater than that of Basidiomycota, resulting in a greater deposition rate overall for Ascomycota (Yamamoto et al., 2014). However this work did not differentiate between vegetative cells and spores, and there is no a priori reason to assume that the deposition rate of Ascomycota vegetative cells are less than or greater to that of Basidiomycota cells. For this reason, we make the null assumption that the 20 deposition rate of each group is equal: We expect Eq. (A4) to hold due to the likelihood that spores greatly out number vegetative cells in the atmosphere in both phyla. Spores can be actively discharged into the air, whereas vegetative cells are not actively propelled into the atmosphere and require aerosolization by mechanical forces like wind. Furthermore, empirical data suggests that vegetative cell fragments constitute a small fraction (0.2-16 %, Green et al., 2011) 5 of the total fungal biomass in the atmosphere. For these reasons, we predict that

BGD
Based on the conclusions of this model, we expect Basidiomycota will dominate the total community, and Ascomycota will dominate the active community in the atmosphere. We note there are many limitations to our model. First, we model fungal gene copy 10 abundances assuming a well-mixed atmosphere at steady state. Yet the atmosphere is a highly heterogeneous and dynamic environment; the sampled air volume was likely neither well mixed nor at steady state over the time intervals we measured. Second, we use a global model with emission and deposition as the sole input and output, whereas a local model that incorporated site-specific environmental fate and transport terms 15 would likely provide more accurate expectations. Third, due to a paucity of data, our estimates of fungal gene abundance levels rely on assumptions about the emission and deposition rates of vegetative cells and spores across fungal taxonomic groups. Empirically derived estimates of these model parameters would significantly improve our approach. Fourth, we do not know to what extent vegetative cells and spores are 20 associated with other particulate matter and how this affects their deposition and emission rates. Adopting an approach to empirically estimate the aerodynamic diameter of these fungal cell types across taxonomic groups would allow for improved estimates of deposition rates (Yamamoto et al., 2014).
The Supplement related to this article is available online at