The microbial diversity of a storm cloud as assessed by hailstones


Correspondence: Ulrich Gosewinkel Karlson, Department of Environmental Science, Aarhus University, Frederiksborgvej 399, 4000 Roskilde, Denmark. Tel.: +45 871 58617; fax: +45 871 55010, e-mail:


Being an extreme environment, the atmosphere may act as a selective barrier for bacterial dispersal, where only most robust organisms survive. By remaining viable during atmospheric transport, these cells affect the patterns of microbial distribution and modify the chemical composition of the atmosphere. The species evenness and richness, and the community composition of a storm cloud were studied applying cultivation-dependent and cultivation-independent techniques to a collection of hailstones. In toto 231 OTUs were identified, and the total species richness was estimated to be about 1800 OTUs. The diversity indices – species richness and evenness – suggest a functionally stable community, capable of resisting environmental stress. A broad substrate spectrum of the isolates with epiphytic origin (genus Methylobacterium) implied opportunistic ecologic strategy with high growth rates and fast growth responses. These may grow in situ despite their short residence times in cloud droplets. In addition, epiphytic isolates utilized many atmospheric organic compounds, including a variety of carboxylic acids. In summary, the highly diverse bacterial community, within which the opportunistic bacteria may be particularly important in terms of atmospheric chemistry, is likely to remain functional under stressful conditions. Overall our study adds important details to the growing evidence of active microbial life in clouds.


Atmospheric dispersal may be important for bacterial biogeography patterns and for the expansion of bacterial fitness by colonizing novel environments (Morris et al., 2008). The budget and the composition of the pool of atmospheric chemicals may even be modified by bacterial activity, as some chemical constituents of the atmosphere support bacterial metabolism (Delort et al., 2010). The mean global emissions of bacteria from terrestrial surfaces have been estimated to be between 140 and 380 CFU per m2 per s (Burrows et al., 2009). Only a small fraction of these aerosolized bacteria leave the boundary layer and enter the free atmosphere, which allows for their long distance transport (Kellogg & Griffin, 2006). Particularly important phenomena in terms of the vertical mixing of troposphere are storm clouds, which suck a large amount of boundary layer air (typically 1011 m3) from below the cloud base into the cloud, often propelling it high up into the troposphere in strong vertical air currents (Markowski & Richardson, 2010), thereby mixing troposphere aerosols. Owing to their inaccessibility and extremely short lifetimes, storm clouds are one of the least studied compartments of the atmosphere.

Fog and cloud droplets, which provide bacteria with dissolved nutrients and protection against desiccation and UV radiation, have been proposed as an atmospheric niche for bacterial growth. Diverse bacterial strains have been identified in atmospheric liquids (Amato et al., 2005, 2007a; Ahern et al., 2007; Kourtev et al., 2011), hosting between 1500 and 430 000 of bacterial cells per mL (Sattler et al., 2001; Hill et al., 2007), which is less than is typical for other aquatic environments (e.g. Graneli et al., 2004; Liu et al., 2009). Viability and metabolic activity of bacteria in cloud water have been demonstrated by a few studies (Bauer et al., 2002; Hill et al., 2007), while others have shown growth of indigenous bacterial communities on cloud-borne organic compounds (Herlihy et al., 1987; Sattler et al., 2001; Amato et al., 2007b). In addition, the ability of atmospheric isolates to grow on organic compounds abundant in cloud water has been confirmed by several studies (Ariya et al., 2002; Amato et al., 2005, 2007ab). Recently, Vaïtilingom et al. (2010, 2011) provided evidence that biodegradation in clouds could be of similar quantitative importance as the photooxidation of atmospheric organic compounds.

As aerosolized bacterial cells are exposed to numerous stress factors including desiccation, high UV radiation and reactive oxygen species, not all microorganisms may be capable of surviving aerosolization, that is, remaining viable or even active during atmospheric transport. Consequently, atmospheric transport would function as a selective barrier that only allows the most robust strains to spread by aerosolization. The diversity of the total bacterial community in the free atmosphere, which can be assessed by measuring species richness and evenness, reflects the presence of bacterial groups that can get uplifted from their terrestrial habitats and escape the boundary layer of the atmosphere. However, only the viable members of the airborne bacterial community, which can at least in part be studied using the cultivation methods, undoubtedly have the potential to resist atmospheric stress and may play a role in biodegradation of atmospheric chemicals as well as bacterial dispersion. Species evenness has been related to the functional stability of a community, in particular to its capacity to resist environmental stress (Wittebolle et al., 2008, 2009). Communities with medium species evenness were proposed to be most balanced (Wittebolle et al., 2009), preserving their functionality by efficiently dealing with environmental stress. Species richness, which is related to metabolic potential of microbial communities, was estimated to be between 1800 and 45 000 bacterial species in diverse terrestrial and marine environments (Torsvik et al., 1996; Venter et al., 2004; Schloss & Handelsman, 2005). The bacterial diversity of the atmosphere and the clouds has so far been addressed mainly based on the community composition (Bowers et al., 2009; Kourtev et al., 2011) of nonreplicated samples. However, only through replication, the relevance of which has recently been stressed by Prosser (2010), a precise description of bacterial diversity of an environment can be achieved. We propose the use of hailstones, which grow by stochastic collection of a large number (typically > 109, Ahrens, 2009) of super-cooled storm cloud droplets, as independent replicate samples of storm clouds, and assert that a precise description of its bacterial diversity can be obtained thereby.

Materials and methods

Meteorological description of storm cloud conditions

Development of the thunderstorm over Ljubljana, Slovenia, in the late afternoon on 25th May, 2009 took place close to a nearly stationary upper-tropospheric ridge (high pressure) axis. At the base of the storm cloud, which was about 1.5 km above ground, the estimated temperature was 13 °C and at the top of the cloud, which was at 12 km, the estimated temperature was −60 °C. Roughly, 3 lower km of the cloud (27%) had an estimated temperature higher than 0 °C. Observations point to deep moist convection of the single-cell type. Such storm clouds typically have a short lifetime of 30–60 min (Markowski & Richardson, 2010), during which they go through the towering cumulus stage, followed by the mature and dissipation stages. The towering cumulus stage has no downdraft and a single updraft (air moving upward from the surface into the cloud), which sucks a large amount of boundary layer air from below the cloud base into the cloud. The specific humidity at the cloud base was 11 g kg−1 and was almost identical to the specific humidity at ground level, which points to a well-mixed boundary layer below the cloud base. Single-cell systems usually have rather low precipitation efficiency (~20%, Lin, 2007), defined as the ratio of the total precipitation to total available moisture of the cloud system. In many single-cell storms, a fraction of the precipitation falls as hail, which forms close to the updraft core.

The paths individual hailstones take through the cloud may differ considerably, and the portion of boundary layer aerosols, which is sampled via cloud droplets by different hailstones, is therefore usually not the same. This is a consequence of the fact that the backward trajectories would most likely originate at widely different ‘departure’ points in the boundary layer.

Collection and cleaning of the hailstones

Hailstones were collected no later than 5 min after they fell on ground during the hail event. Hailstones were stored at −20 °C immediately after collection. We measured mass and volume of 24 individual hail pieces prior to sterilizing their surface (Table 1). The volume was measured by briefly submerging each hailstone into sterile particle free water, which had been cooled to 4 °C to prevent loss of volume by melting. The surface of hailstones was sterilized by rinsing it with a mixture of 1% benzalkonium chloride and 62% ethanol in deionized water three times followed by rinsing it with sterile deionized water to remove any of the remaining sterilization liquid. Ice cubes of autoclaved deionized water, with their surface contaminated by soil and grass, were sterilized as a control for the sterilization procedure and analysed with flow cytometry and by plating on nutrient agar plates (data not presented). Aliquots from 17 hailstones were stored at −20 °C and later used for ion analysis. Meltwater of 12 hailstones was stored at −20 °C and was used for total community analysis. The remaining 12 hailstones were examined for their cultivable bacterial community.

Table 1. Mass and volume of 24 hailstones
Hail no.H1H2H3H4H5H6H7H8H9H10H11H12
m (g)29.311.113.113.310.520.523.425.22117.410.111.7
V (mL)35.515.016.517.513.
Hail no.H13H14H15H16H17H18H19H20H21H22H23H24
m (g)18.312.123.316.519.610.622.916.98.914.410.433.9
V (mL)21.513.526.020.023.513.026.019.512.517.013.542.5

Extraction of total DNA and construction of a clone library

Total genomic DNA was successfully extracted from the meltwater of 9 of 12 hailstones. Triplicate clone libraries were constructed from the DNA extract of two hailstones, while single libraries were constructed from the DNA extract of the remaining seven hailstones. To extract DNA, 500 μL of hailstone water was centrifuged for 99 min at 20 000  g and 4 °C. The supernatant was carefully pipetted away, and the pellet was resuspended in 10 μL of TE buffer. The resulting cell suspension underwent three freeze-thaw cycles, with sudden temperature shifts between approximately −65 °C and +65 °C, to break the cells and extract total genomic DNA. To reduce loss of DNA, no purification step was used. DNA was stored at −20 °C prior to use. Three negative controls, where TE was treated with three freeze-thaw cycles, and seven negative controls, where UV-treated water was added as a template for PCR, were analysed along with 13 DNA samples.

Semi-nested PCR (Nichols et al., 2010) was used to amplify the 16S ribosomal genes. Four microlitres of total genomic DNA was used for amplification of a nearly full length segment of the 16S rRNA gene in the first PCR reaction, using the universal bacterial primers 27f (5′-AGAGTTTGATCMTGGCTCAG-3′) and 1492r (5′-GGYTACCTTGTTACGACTT-3′). Four microlitres of the product from this PCR was used as a template in the second, semi-nested PCR reaction with universal bacterial primers 27f and 518r (5′-GTATTACCGCGGCTGCTG-3′) targeting a 500 bp segment of the 16S rRNA gene.

A multistrategy DNA decontamination procedure (Champlot et al., 2010) was carried out on laboratory surfaces, gloves, labware and reagents used in the first PCR reaction, to reduce the amount of contaminating DNA. The laboratory surfaces, gloves and labware were cleaned with LookOut DNA Erase (Sigma-Aldrich). For decontamination of the PCR reagents, three separate PCR master mixes were prepared. The first master mix, which contained dNTP and the primers (27f and 1492r), was supplemented with 10 mM MgSO4, 6.25 mM Tris–HCl, 1 mM CaCl2 and 1 mM DTT and treated with 0.1 U μL−1 of heat-labile, double-strand-specific DNase (hl-ds-Dnase; Biotec Marine Biochemicals). The second master mix, which contained 1.25 U of high fidelity Pwo DNA polymerase (Roche), was supplemented with 12.5 mM Tris–HCl, 12.5 mM MgSO4, 1 mM CaCl2 and 1 mM DTT and treated with 0.125 U μL−1 hl-ds-Dnase. The hl-ds-Dnase treatment was performed at 37 °C for 30 min; the hl-ds-Dnase was afterwards deactivated at 60 °C for 30 min. Aliquots of the third master mix that contained water and PCR reaction buffer were decontaminated by UV irradiation in a UV Crosslinker device (Stratalinker UV Crosslinker) at 254 nm and at 1 cm from the UV bulbs for 15 min. Following the decontamination, all master mixes were combined and 4 μL of total genomic DNA was added. The final concentrations of the PCR reagents were 3.75 mM for MgSO4, 400 nM for dNTP and 600 nM for primers. The PCR reaction was performed with initial denaturation at 95 °C for 5 min, followed by 30 cycles of denaturation at 95 °C for 1 min, annealing at 52 °C for 30 s and elongation at 72 °C for 2 min. There was a 10 min final extension step at 72 °C. The second, semi-nested PCR was performed in the presence of 1.25 U of Pwo, 3.5 mM MgSO4, 400 nM dNTP and 600 nM primers (27f and 518r). After initial denaturation at 95 °C for 5 min, there were 30 cycles of denaturation at 95 °C for 1 min, annealing at 55 °C for 30 s and elongation at 72 °C for 1 min. There was a 10 min final extension step at 72 °C. PCR products were checked on 1.5% agarose gel.

The resulting PCR products were purified with a Montage DNA gel extraction kit (Millipore). 30 μL of the PCR product was loaded on a 1.2% agarose gel prepared with 1× modified TAE buffer. The 500-bp bands were cut out and purified. After purification, the PCR products were purity checked on a 1.5% agarose gel. Then, A-overhangs were added to purified PCR products by adding dATP (0.2 mM) and 1.5 U of Taq (Fermentas) polymerase in 1× PCR buffer solution. The mixture was incubated at 72 °C for 20 min. PCR products were cloned into pCR 2.1 vector (TOPO TA Cloning Kit; Invitrogen) and transformed into chemically competent Escherichia coli cells TOP10 (TOPO TA Cloning Kit; Invitrogen). The presence of vector with insert was confirmed on 96 white colonies by running a colony PCR using M13 forward and M13 reverse primers according to the instructions of the manufacturer. Purification and sequencing of the PCR products was performed by Macrogen (Seoul, South Korea). Primer M13 forward was used for sequencing.

Removal of contaminant and chimerical sequences

The sequence quality was checked manually, and of 1591 sequences (1206 sequences from hailstones and 385 sequences from negative controls), 363 were discarded because of low quality. The vector and both 16S rRNA gene universal bacterial primers were removed using DNA Baser Sequence Assembler (Heracle BioSoft, Pitesti, Romania). The sequences that were likely chimerical were determined by analysis with Bellerophon (Huber et al., 2004). If necessary, this was repeated several times until all chimeras were identified. As a result, 85 chimerical sequences were excluded from further analysis. Operational taxonomic units (OTUs) were created by 99% similarity using the CD-HIT Suite: Biological Sequence Clustering and Comparison (Weizhong CD-HIT Suite). Three hundred and fifty-seven sequences from hailstone clone libraries that shared 99% of sequence similarity with sequences from negative controls were removed from further analysis. The remaining 485 sequences from hailstones were deemed suitable for diversity analysis.

Analysis of bacterial diversity

The Ribosomal Database Project (RDP) classifier was used for naive Bayesian classification of sequences (Wang et al., 2007). Alignment and clustering of the sequences were carried out with Pyrosequencing Pipeline of the RDP. EstimateS SWin820 (Colwell, 2006) was used for calculating shared species statistics (e.g. Bray–Curtis similarity) and diversity statistics (e.g. Chao 2 richness estimator, number of singletons, doubletons, uniques and duplicates). Chao 2 richness estimator was taken as a measure of expected bacterial richness, as it is useful for estimating population size for data obtained by resampling (Chao, 1987) and is recommended for sample sets with many rare species. Diversity statistics was calculated with EstimateS SWin820 using 50 runs of randomizations without replacement. Biased corrected formula was employed to calculate Chao 2 in case the CV for abundance distribution was < 0.5, and the classic formula was used additionally when the CV was > 0.5. Then, values resulting from both calculations were compared, and the larger values were reported. For estimating species evenness, we used Lorenz curves together with Gini coefficients (ranging from 0 – completely even community to 1 – completely uneven community, Wittebolle et al., 2009).

Cultivation, fingerprinting and 16S rRNA gene sequencing of isolates

Twelve melted hailstone samples were plated together with the negative controls in three replicates of 500 μL on solid low nutrient medium R2A described by Reasoner & Geldreich (1985). The medium contained 0.05 μg mL−1 natamycin to prevent fungal growth. After 2 weeks of incubation at 15 °C under aerobic conditions, colony forming units (CFUs) were counted. Colonies were also purified on R2A plates, grown in liquid R2 medium (prepared as R2A Difco medium without the addition of agar) and stored with 23% glycerol at −80 °C. For the isolation of genomic DNA from isolates, a small amount of colony from each isolate was resuspended in 200 μL of TE buffer. The cell suspensions were either boiled at 102 °C for 15 min or subjected to three rounds of freeze-thaw cycles (approximate temperature span, −65 °C to +65 °C). Cell debris was pelleted by centrifugation at 15 000  g and 4 °C for 15 min, and the supernatant containing the genomic DNA was transferred to a new Eppendorf tube. DNA was stored at 4 °C prior to use. A nearly full length segment of 16S rRNA gene was amplified by the use of universal bacterial primers 27f (5′-AGAGTTTGATCMTGGCTCAG-3′) and 1492r (5′-GGYTACCTTGTTACGACTT-3′). The PCR reaction was performed with initial denaturation at 95 °C for 3 min, followed by 35 cycles of denaturation at 95 °C for 1 min, annealing at 52 °C for 30 s and elongation at 72 °C for 90 s. There was a 10 min final extension step at 72 °C. PCR products were checked on 1.5% agarose gel. Amplified rDNA restriction analysis (ARDRA) was used to cluster the isolates into groups sharing the same fingerprint. Restriction of PCR products was carried out with Hae III (FastDigest Hae III, Fermentas) according to manufacturer's instructions. Restriction patterns were visualized on 2% agarose gel, and the same fingerprints were clustered together. Prior to sequencing, 5 μL of PCR products were enzymatically purified by addition of 10 U of Exonuclease I (Fermentas) and 1 U of Shrimp Alkaline Phosphatase (Fermentas). Representatives of every fingerprint pattern were sequenced. Sequences were analysed in the same way as clone sequences.

Metabolic tests with Phenotype Microarray plates

Representatives of most common OTUs, selected by 99% similarity using CD-HIT Suite: Biological Sequence Clustering and Comparison were chosen for metabolic tests. Most common OTUs were defined as the ones that appeared in hailstones at least 10 times or in at least three hailstones. Eight Methylobacterium and two Bradyrhizobium strains were selected. Isolates were grown in R2 medium to a late exponential phase, after which the cells were washed three times in a phosphate buffer, consisting of 0.4 M Na2HPO4 and 0.19 M KH2PO4 (pH 6.8). The cells were resuspended in liquid Nitrate Mineral Salts Medium (NMS, Whittenbury et al., 1970), supplemented with Vitamin Mix, Thiamin and Vitamin B12 (Widdel & Bak, 1992). The cell suspensions were inoculated into PM1 MicroPlate Carbon Sources (Biolog, Inc. Hayward) at a volume of 100 μL per well and incubated at room temperature, while shaking at 200 r.p.m. The OD was measured on the day of inoculation and then after 7 and 14 days of incubation. OD values that were interpreted as positive results had to exceed the value of the double measurement error, which was estimated from the blank OD values obtained on the day of inoculation for each isolate separately. The measurement error was defined as the maximum difference between the average of 96 blank OD values and either the minimum or the maximum blank OD value. One Methylobacterium and one Bradyrhizobium strain were removed from analysis because of a potential contamination problem.

Ionic composition of the hailstone meltwater

Nitrate, nitrite, sulphate and chloride concentrations were determined by ion chromatography (Dionex ICS 2500 with AS 18 column) with KOH as mobile phase. Ammonium concentrations were measured by nesslerization (Bower & Holm-Hansen, 1980).

Nucleotide accession numbers

The 16S rRNA gene sequences of the clones and the isolates have been submitted to GenBank under accession numbers JQ896628JQ897350.

Results and discussion

Diversity of the total bacterial community

We constructed clone libraries from nine individual hailstones to assess species richness and evenness, which represent the most important measures of bacterial diversity. Owing to the combination of low bacterial density (unpublished results) and low volume (Table 1) of our samples, clone libraries had to be made using a semi-nested PCR approach. It was shown that the nested PCR is not significantly more biased than the traditional PCR and can be used to investigate microbial diversity of low bacterial density environments (Fan et al., 2009). The number of PCR cycles also does not significantly influence the community composition studies, as the largest PCR-associated bias is likely generated during the first few PCR cycles (Acinas et al., 2005).

We obtained 485 sequences, which could be clustered into 231 OTUs0.01 (OTUs based on 99% similarity = species level, Stackebrandt & Ebers, 2006), 177 OTUs0.05 (OTUs based on 95% similarity = genus level, Ludwig et al., 1998) and 67 OTU0.15 (OTUs based on 85% similarity = lineage level). Rare OTUs0.01 within the total pool of sequences accounted for 84% of all OTUs0.01, with singletons (OTUs that occur once in the pooled assembly of sequences) accounting for 159 OTUs0.01 (68.8%) and doubletons (OTUs that occur twice in the pooled assembly of sequences) for 35 OTUs0.01 (15.2%). A similar proportion of rare OTUs was reported for uncontaminated soil (Hill et al., 2003) and the atmosphere (Fahlgren et al., 2010). Most abundant bacterial taxons identified by clone libraries are presented in Supporting Information, Table S2. Sixteen main OTUs0.01, represented by at least five sequences, belonged to four different phyla, Firmicutes, Actinobacteria, Bacteroidetes and Proteobacteria. Their closest relatives were previously detected in bulk and rhizosphere soil or on leaf surfaces of plants. Plant, rhizosphere and soil origin fits very well with the analysis of dissolved organic compounds in hailstones (unpublished results). The agricultural contribution has also been indicated by the ion composition (Supporting Information, Data S1, Table S1). The total bacterial community was not represented well by individual hailstones, which is illustrated in Fig. 1. Only very few genera of the total community were present in more than two hailstones (characteristic genera). In addition, 99.6% of all OTUs0.01 were uniques (OTUs that occur in one sample) or duplicates (OTUs that occur in two samples). In fact, only one OTU0.01, which belonged to the Actinobacteria genus Cellulomonas, appeared in three hailstones and no OTU0.01 was detected in more than three.

Figure 1.

Community composition showing all genera and common genera (present in at least three hailstones) for total community (TC) and cultivable community (CC).

The capacity of a bacterial community to resist environmental stress has been related to its species evenness (Wittebolle et al., 2008, 2009). We described the evenness using Pareto–Lorenz distribution curves (Marzorati et al., 2008; Wittebolle et al., 2008, 2009; Edwards et al., 2011), which are presented in Fig. 2a. The intercept of the respective Lorenz curve with the 20% x-axis line (Fig. 2a) can be used to distinguish between communities with high evenness (25% Lorenz curve, intercept at y = 25%), communities with medium evenness (45% Lorenz curve, y = 45%) and highly specialized communities (80% Lorenz curve, y = 80%, Wittebolle et al., 2008). Based on their intercepts, the Lorenz curves of the total storm cloud community suggest a community with medium evenness (Fig. 2a). Thus, approximately 55% of the species were dominant while the remaining 45% were present in low numbers. Gini coefficients, which can take values between 0 (complete equality) and 1 (complete inequality), were used to quantify species evenness (Wittebolle et al., 2009). The mean Gini coefficient for all hailstones was 0.413, and the Gini coefficient for pooled sequences was 0.448, which also points to medium evenness. Although also characterized by medium evenness, the cultivable community was more specialized than the total community, as is evident from its Lorenz curve (Fig. 2a) as well as from the Gini coefficient, which was 0.58. Community composition with medium evenness has been shown to be most balanced (Wittebolle et al., 2009) and can preserve its functionality by efficiently dealing with environmental stress. This is highly important for a bacterial community passing through the atmosphere, if we consider the rapid change that aerosolization poses for bacteria and the diverse environmental stress they encounter in the atmosphere. However, despite the fact that the medium evenness of the airborne community may be beneficial under harsh conditions, the actual activity and the resulting functionality of cloud-borne bacterial communities remain uncertain, especially considering the short residence times of bacteria in cloud droplets.

Figure 2.

(a) Lorenz curves showing total community evenness for total communities of individual hailstones (dashed lines) and the pooled hailstones (thick dashed line). The Lorenz curve showing community evenness for the cultivable community is also shown (thick solid line). The 20% x-axis line is indicated. (b) Chao 2 richness estimator of the total bacterial community as a function of number of replicate samples analysed, which shows estimated number of OTUs0.01 in two hailstones with three replicate subsamples (H24, thin black lines; H16, thin grey lines) and in the storm cloud with nine replicate hailstones (thick black lines). Mean Chao 2 richness estimator is plotted as solid lines and 95% confidence intervals as dashed lines. (c) Species accumulation curves are presented for clusters of different levels of similarity. From top to bottom: OTU 0.01, OTU 0.03, OTU 0.05, OTU 0.10, OTU 0.15. (d) Estimated richness of OTU 0.01 (black lines) and species accumulation curves (grey lines) as a function of number of samples analysed by cultivation-dependent methods. Mean values are plotted as solid lines and 95% confidence intervals as dashed lines.

Chao 2 was employed to determine the total bacterial richness of the cloud. We estimated that the storm cloud contained approximately 1800 OTUs0.01 (Fig. 2b), 500 OTUs0.05 and 80 OTU0.15, which points to a high species richness. Despite the low bacterial densities found in the storm cloud (unpublished results), which were several orders of magnitude lower than in soils or oceans, the diversity of storm cloud bacteria was comparable to that of soil and marine environments, hosting highly diverse communities (Torsvik et al., 1996; Venter et al., 2004; Schloss & Handelsman, 2005). The observed richness of the storm cloud (231 OTU0.01 detected by clone libraries) represented 12.5% of the estimated total richness. Despite the high proportion of richness that was captured in the clone libraries, the species accumulation curves for OTU0.01 and OTU0.05 are still linearly increasing, even after analysing nine hailstones (Fig. 2c). A strikingly high species richness, which was characteristic for the storm cloud, additionally supports the high functional stability of the bacterial community that was already indicated by the medium species evenness.

For the estimation of total richness in the cloud, the analysis of several hailstones was pivotal. In Fig. 2b, the estimated richness assessed by Chao 2 is presented as a function of the number of independent samples (hailstones) analysed. Examining a single hailstone, the species richness would be underestimated by 75%. Examining more than three hailstones facilitates a more reliable richness estimate (Fig. 2b). Two hailstones were also analysed in replicates, to assess bacterial diversity in a single hailstone and to estimate how comparable it is to the total diversity of the cloud. Even though the observed diversity (the number of OTUs0.01 detected by clone libraries) was very similar for both hailstones (56 OTUs0.01, H16; 51 OTUs0.01, H24), the total bacterial richness, estimated with Chao 2, differed strikingly. Hailstone H16 contained 300 OTUs0.01 (Fig. 2b), which is 15% of total estimated cloud richness, whereas hailstone H24 contained 1250 OTUs0.01, which is almost 70% of the estimated cloud richness. In addition, the estimated richness of H16 was described well with three subsamples, while the estimated richness of H24 was still increasing and was thus still underestimated (Fig. 2b). Hailstone H24 very well represented the total diversity of the cloud, which was further implied by the shape of the curve of estimated richness as a function of number of (sub)samples. The higher representation of H24 over H16 could partially be explained by the difference in the hailstones sizes, as the volume and the total number of bacterial cells in the hailstone H24 (42.5 mL, 64 000 cells) was more than two times higher than the volume and the total number of bacterial cells in H16 (20.0 mL, 30 000 cells). The Bray–Curtis coefficient was used to evaluate the compositional similarity between individual hailstones and compare it with subsamples of hailstone H16 and H24. Generally, (with an exception of subsample H16.2) the subsamples of hailstones shared a higher similarity than observed between individual hailstones (data not presented).

Cultivable bacterial community

The cultivable community represented a high proportion (up to 10.5%, unpublished results) of all cells. We isolated 424 bacterial cultures from 9 of 12 analysed hailstones, while three hailstones yielded no CFU. The isolates were clustered into 85 OTUs0.01, 61 OTUs0.03 and 44 OTUs0.05. If we consider all isolates, only 33 OTUs0.01 (38.8%) were singletons and 13 OTUs0.01 (15.9%) were doubletons. The cultivable bacterial community was represented well by some of the hailstones (Fig. 1). The majority of genera were present in more than two hailstones (characteristic genera), and there were only 74.1% of OTUs0.01 that appeared in only one or two hailstones (uniques and duplicates). This is a very different picture from the one we obtained by the cultivation-independent study. Despite the fact that nine hailstones used for the cultivation-independent study were not identical to the nine hailstones analysed in the cultivation-dependent study, the large number of replicates should diminish any influence of using distinct samples.

We estimated, using Chao 2, that the total richness of cultivable bacteria in the storm cloud is approximately 120 OTUs0.01 (Fig. 2d), 90 OTUs0.03 or 60 OTUs0.05, which represents approximately 7–12% of the estimated richness of the total bacterial community. Species accumulation curves reached a plateau, after the analysis of nine hailstones. In addition, the richness we actually observed by analysing 12 hailstones (three hailstones with no CFU were included in this analysis) is approaching the estimated richness (Fig. 2d), as we managed to cover between 68.5% and 77.0% of estimated cultivable bacterial OTUs in the cloud.

Characteristic isolates, which were found in three or more hailstones, belonged to 21 different OTUs0.03 representing five genera, Methylobacterium, Bradyrhizobium, Bacillus, Paenibacillus and Afipia. None of these genera were detected by our clone libraries, but all five were previously identified in air samples using cultivation-independent methods (Maron et al., 2005). The largest OTU0.03 with 17.2% of all isolates found in seven hailstones as well as five other characteristic OTUs0.03 belonged to the epiphytic genus Methylobacterium, which encompasses around a third of all isolates. Methylobacterium spp. are characterized by their pink pigmentation and their facultative methylotrophic lifestyle, traits that are common in epiphytic bacteria. Like the atmosphere, the phyllosphere is an extreme environment, with high levels of UV radiation, low water potential and sudden temperature shifts (Lindow & Brandl, 2003). Thus, adaptations to diverse stressful factors on plant leaf surfaces could help Methylobacterium to remain viable or even metabolically active in the atmosphere. In fact, Methylobacterium was found to possess a general stress response system, which enhances resistance to heat shock, desiccation, UV radiation and oxidative stresses (Knief et al., 2010). On plant surfaces, Methylobacterium strains are competitive against other plant colonizing bacteria because of their capability to utilize single carbon compounds, for example, methanol, formaldehyde and formic acid (Corpe & Rheem, 1989). Being capable of utilizing volatile methanol (Corpe & Rheem, 1989), cloud-borne Methylobacterium cells may grow on atmospheric methanol, the second most abundant organic compound in the atmosphere (Fukui & Doskey, 1998; Galbally & Kristine, 2002). Apart from methanol, Methylobacterium can utilize formaldehyde and formic acid that are both abundant in the atmosphere and in cloud water. Their ability to metabolize several single carbon compounds together with their stress tolerance may enable Methylobacterium to survive and even grow in cloud water.

Different strains of Methylobacterium are negative for ice nucleation and even have some antifreeze properties (Romanovskaya et al., 2001). Acting as antifreeze, large numbers of Methylobacterium in clouds could be important for patterns of precipitation, as the formation of ice crystals is often a prerequisite for precipitation in mixed phase clouds. In addition, Methylobacterium have been shown to have an antagonistic effect against some known plant-associated ice nucleating species (e.g. Xanthomonas campestris and Pseudomonas syringae). Interestingly, cultivable γ-Proteobacteria, to which the known ice nucleating plant pathogens belong, were not detected in our hailstones, although they have been found universally in atmosphere, cloud and precipitation samples (Fuzzi et al., 1997; Maron et al., 2005; Amato et al., 2007a; Ahern et al., 2007; Bowers et al., 2009). However, formation of hail does not depend on the presence of ice nuclei, as storm clouds reach high altitudes with temperatures below −40 °C (Ahrens, 2009).

Bradyrhizobium, represented by three characteristic OTUs0.03 (7.8% of all isolates), was another genus of α-Proteobacteria that was characteristic for the storm cloud hailstones. Like Methylobacterium, some Bradyrhizobium strains can grow on methanol and metabolize other single carbon compounds (Sudtachat et al., 2009). They may therefore grow on atmospheric methanol and consequently alter atmospheric chemistry. Bacterial strains belonging to the genus Bradyrhizobium fix nitrogen, and several Bradyrhizobium strains can carry out photosynthesis (Molouba et al., 1999). Their diverse metabolic capacities could be an advantage in the atmospheric environment, where at least some bacteria may be limited with bioavailable organic compounds. The residence time could, however, be an issue in the case of Bradyrhizobium, as their rRNA operon number (1.3 on average, Klappenbach et al., 2000; Lee et al., 2009) implies an ecological strategy that is characterized by oligotrophy and consequently slow growth. In contrast, Methylobacterium strains typically carry several copies of the 16S rRNA gene (5.5 on average), supporting a fast growth response that is characteristic for copiotrophic bacteria (Shrestha et al., 2007). Although we have not analysed the rRNA operon number, these values are frequently conserved within genera (Rastogi et al., 2009).

Bacillus and Paenibacillus represented almost 16.5% of all isolates, and Bacillus also formed the second most abundant OTU0.03, representing 9.2% of all isolates found in eight hailstones. Both genera have the ability to form endospores, which might provide them with means of withstanding environmental stress faced in the atmosphere. Several Bacillus isolates were closely related to known pathogenic strains, such as Bacillus anthracis or strains of Bacillus cereus and Bacillus thuringiensis. The last common genus, Afipia (third largest OTU0.03), could also be a potential pathogen, as it exhibits amoebae resistance, which is considered a possible virulence trait (Thomas et al., 2007).

Metabolic tests with Phenotype Microarray plates

Phenotype Microarray plates were used to assess the metabolic potential and niche specialization of selected Methylobacterium and Bradyrhizobium strains. We chose to test this group of plant-associated, nonspore-forming genera, which are more likely to remain metabolically active in the atmosphere, because of their preadaptations to atmospheric stress and their ability to utilize cloud-borne organics. Airborne Bacillus and Paenibacillus were likely present as endospores, which facilitates their survival under unfavourable conditions but excludes any metabolic activity. We tested seven Methylobacterium isolates and one Bradyrhizobium isolate for utilization of 95 organic compounds as single carbon sources. They were able to utilize between 2 and 9 carboxylic acids (Table 2), which on average represent 36% of total dissolved organic carbon in cloud water (Marinoni et al., 2004). Four Methylobacterium strains as well as the Bradyrhizobium strain were also able to utilize formic acid, confirming their potential to grow on major single carbon constituents of cloud water. The ability of cloud-borne bacteria to metabolize atmospheric organic compounds has previously been indicated by studies, showing that members of natural microbial communities could grow on the bulk of organic compounds found in cloud water (Sattler et al., 2001; Amato et al., 2007b; Hill et al., 2007). Bacterial isolates from dry air, precipitation and cloud water were also able to utilize common atmospheric organics (Ariya et al., 2002; Amato et al., 2005, 2007b, Vaïtilingom et al., 2010, 2011) at rates comparable to photooxidation, which means that microbial metabolism may compete with photooxidation in clouds. In addition, we demonstrated that the Bradyrhizobium strain was more specialized and could only utilize a small number of compounds, while the Methylobacterium strains were able to utilize between 17% and 49% of organic compounds of different types (Table 3). This observation provides supporting evidence for the generalistic (opportunistic) ecological strategy of cloud-borne Methylobacterium strains.

Table 2. Utilization of major carboxylic acids in cloud water (Marinoni et al., 2004) by 7 Methylobacterium and one Bradyrhizobium strain: 1 – can utilize, 0 – cannot utilize
Carboxylic acids% of total carboxylic acid conc.a 7 Methylobacterium strains Bradyrhizobium strain
  1. a

    The contribution of each carboxylic acid to the total carboxylic acid concentration in cloud water, as reported by Marinoni et al. (2004).

L-Lactic Acid711011110
Acetic Acid2701011010
Glycolic Acid500000010
Propionic Acid301000010
Formic Acid2410011011
Glyoxylic Acid500101010
α-Keto-Glutaric Acid411111111
Succinic Acid711010110
M-Tartaric Acid100000010
Table 3. Utilization of different compound types as single carbon sources by 7 Methylobacterium and one Bradyrhizobium strain
Compound type7 Methylobacterium strains Bradyrhizobium strain
Amines (no of compounds)00100000
Amino acid derivates (no of comp.)211245380
Carbohydrates (no of comp.)3181077342
Carboxylic acids (no of comp.)1013615159182
Nucleotides (no of comp.)05000000
Phenolic compound (no of comp.)00000000
Polymer (no of comp.)10010110

In summary, we hypothesize that the highly diverse bacterial community remains functional under stressful conditions and that it contains bacterial groups, which remain viable and are potentially active in the atmosphere. We propose that they are important for the patterns of bacterial distribution as well as for atmospheric chemistry. We also suggest that although they account for a small fraction of the highly diverse cloud community, epiphytic bacteria encompass species with particular importance for altering atmospheric chemistry.


T.S.T. was supported by a PhD fellowship granted by the Danish Agency for Science Technology and Innovation (Forsknings- og Innovationsstyrelsen). We thank Tina Thane for excellent technical assistance and Marijan Govedič for sample collection as well as the valuable discussions on diversity indices. We appreciate the helpful advice of Mark A. Lever and Kasper U. Kjeldsen regarding the molecular work with low bacterial density environments.