Phenol (>99 %) and hydrogen peroxide (30 %) were obtained from Fluka, and sodium chloride (>99 %), dichloromethane (>99.8 %), and sulfuric acid (>95 %–97 %) were from Sigma Aldrich; acetonitrile (>99.9 %) was from VWR Chemicals, NaOH (99 %) from Merck, and MgSO4 (>98 %) from Carlo Erba Reagents (SDS).

Cloud sampling. Cloud waters have been sampled at the PUY station (summit of the Puy de Dôme, 1465 m a.s.l.; 45∘46′ N, 2∘57′ E; France), which is part of the atmospheric survey networks EMEP (the European Monitoring and Evaluation Programme), GAW (Global Atmosphere Watch), and ACTRIS (Aerosols, Clouds, and Trace gases Research Infrastructure). The sampling site is fully described in Deguillaume et al. (2014). The global meteorological context was examined through 120 h back trajectories of the air masses sampled using the HYSPLIT model (HYbrid Single-Particle Lagrangian Inte-grated Trajectory). Two cloud water samples collected in 2016 (21 and 26 October) were analyzed in this work for phenol quantification by GC-MS. Three other samples were previously collected and analyzed in 2013 (5 November), 2014 (27 June), and 2016 (16 February; Lebedev et al., 2018). Samples were collected using a sterilized cloud droplet impactor and immediately filtered through a Minisart^® PES filter (0.22 µm porosity; Sartorius, Germany) under sterilized conditions; these have been stored at -25 ∘C.

GC-MS analysis. Sample preparation was carried out according to US EPA 8270 method. Prior to utilization, all the glassware was cleaned with a piranha reagent composed of 6 mL of sulfuric acid mixed with 2 mL of hydrogen peroxide. The reagents were kept in the glassware for one night and after all the glasses were washed two times with ultrapure water and two times with dichloromethane. With clean dishes, cloud waters kept frozen were melted at room temperature and the pH adjusted to 2 and 11. Organic compounds were extracted three times with dichloromethane (keeping the ratio 10 mL of water for 1 mL of CH2Cl2). All the dichloromethane fractions were then dried with MgSO4 and evaporated to 1 mL using a rotary evaporator under reduced pressure; the temperature of the water bath was 20 ∘C. Samples were kept at 4 ∘C until analysis

All analyses related to cloud samples collected on 21 and 26 October 2016 were performed at Saint Joseph, MI, at LECO Corporation (USA). Accurate GC-MS measurements were performed with the high-resolution time-of-flight mass spectrometer Pegasus^® GC-HRT in GC mode (software ChromaTOF-HRT). The obtained EI mass spectra were used for phenol identification by utilizing high mass accuracy data and retention time (Lebedev et al., 2013). Phenol concentrations were measured using naphthalene D8 as an internal standard. The response factor (0.7) was calculated using a standard solution of phenol. Phenol concentrations measured in cloud water samples collected on 5 November 2013, 27 June 2014, and 16 February 2016 are extracted from Lebedev et al. (2018).

Transcriptomic analysis. Cloud water samples were collected on 17 November 2014 for three consecutive periods of 5 h. The cloud air mass origin remained stable over the duration of sampling as attested by air mass backward trajectories (Fig. S1). The cloud droplets collected by impaction were immediately transferred by gravity into sterile collection bottles (Nalgene, Rochester, USA) through sterile (autoclaved) silicone tubing. Before cloud sampling operations started, control samples were made by pouring 200 mL of sterile water into the collection device and through the tubing and by processing it in parallel with the cloud water samples, including sequencing and data treatment. These controls were clearly distinct from samples: based on their contribution to identified ribosome sequences, these contained mostly Enterobacteriaceae (66 %), Dikarya (9.2 %), Streptococcaceae (5.4 %), Vibrionaceae (2.8 %), and Micrococcaceae (1.2 %), i.e., not the taxa of interest here. Conservatively, the sequences present in controls were further removed from sample files (BWA-MEM; Li, 2013). Immediately after collection, water samples were filtered (MO BIO 14880-50-WF) within a UV-sterilized laminar flow hood installed at the sampling site. The filters were then put into ∼5 mL of RNA Later solution (Sigma, Steinheim, Germany) and stored at -80 ∘C until further processing. Briefly, total RNA was extracted from filter halves using a MO BIO PowerWater RNA Kit, and bacterial ribosomal RNA was depleted using a MICROBExpress Bacterial mRNA Enrichment Kit (Life Technologies). Metatranscriptomes of the messenger RNAs were then obtained by multiple displacement amplification using a REPLI-g WGA & WTA kit.; shotgun libraries were sequenced on Illumina MiSeq paired-end 2×300 bp. Sequencing reads were quality checked (FastQC; Andrews, 2010) and trimmed (PRINSEQ-Lite; Schmieder and Edwards, 2011) before assembling the mate pairs using PANDA-SEQ (Masella et al., 2012). Annotations were made against the UNIPROTKB database (Leinonen et al., 2006), including protein sequences for bacteria, archaea, and fungi using BLASTX software (best hits with e value <10-4). All steps were performed using custom scripts. The sequence files have been deposited to the European Nucleotide Archive (ENA) under the study accession number PRJEB25802.

Bioinformatics treatment. Known enzymes involved in phenol degradation were found in the KEGG database (see Fig. 1). We only focused on aerobic metabolism as the cloud environment is highly oxidative. Four nucleotide sequence databases were created from NCBI corresponding to phenol hydroxylases (69 sequences), phenol monooxygenases (29 sequences), catechol (regrouping catechol 1,2-dioxygenases and catechol 2,3-dioxygenases; 145 sequences), and a fourth database including genes coding for putative phenol degradation enzymes (38 sequences). The sequences from the cloud metatranscriptomes corresponding to the different created databases were then extracted using bowtie2 (very sensitive option; Langmead and Salzberg, 2012). The affiliation of the extracted sequences was determined using BLASTN on a local server (e value = 0.00001; Camacho et al., 2009).

Bacterial strains. Bacterial strains were isolated from cloud waters sampled at the PUY station and identified as previously described in Vaïtilingom et al. (2012). From our lab strain collection, we choose all the potential bacteria that could biodegrade phenol. From the 145 strains tested, 33 of the strains were isolated for this work; the others were published earlier (see Table S1 in the Supplement).

119 Pseudomonas, 24 Rhodococcus strains, and 2 strains form the Moraxella family were selected for the screening of phenol degradation (see Table S1). More precisely, Pseudomonas strains included 4 P. fluorescens, 10 P. graminis, 1 P. grimontii, 2 P. poae, 1 P. reactans, 1 P. reinekii, 3 P. rhizosphaerae, 35 P. syringae, 2 P. trivialis, 2 P. veronii, 1 P. viridiflava, and 57 Pseudomonas spp. Rhodococcus strains included 1 R. erythropolis, 1 R. enclensis, and 22 Rhodococcus spp. Moraxella family strains were one Moraxella sp. and one Psychrobacter sp.

Pseudomonas and Rhodococcus strains represent 20.4 % and 4.10 % of the 584 strains of our cloud bacterial collection. From our experience, at the genus level, Pseudomonas and Rhodococcus are among the most frequent bacteria in clouds: Pseudomonas strains in particular have been frequently isolated by culture (Vaïtilingom et al., 2012; Joly et al., 2013), and both targeted and untargeted molecular analyses (and metagenomes, respectively) demonstrated high occurrence in the bacterial communities. These represented 0.1 % to >2 % of the prokaryotes ribosome sequences in amplicon sequencing investigations (Amato et al., 2017a). Based on the biomass in clouds (∼104 bacteria cells mL-1; Vaïtilingom et al., 2012) and assuming even ribosome amplification between bacterial groups, we can infer the presence of ∼103 Pseudomonas mL-1 and ∼102 Rhodococcus mL-1 of cloud water.

Incubations. All the strains were grown in 25 mL of R2A medium for 48 h at 17 ∘C and 130 rpm (Reasoner and Geldreich, 1985). Then cultures were centrifuged at 4000 rpm for 15 min. Bacteria pellets were rinsed first with 5 mL of NaCl 0.8 % and after with Volvic^® mineral water previously sterilized by filtration under sterile conditions using a 0.22 µm PES filter. Cells were resuspended in 5 mL of 0.1 mM phenol solution, prepared in Volvic^® mineral water, and incubated at 17 ∘C and 130 rpm agitation during 5 days in the dark. To identify the concentration, the optical density (OD) for each strain was taken during the experiment. Strain concentrations were around 109 cells mL-1. The ratio number of bacterial cells / phenol concentration was kept to that measured in cloud waters, and all the concentrations were multiplied by a factor of 104. Indeed, the mean bacteria concentration is around 105 cells mL-1 of cloud water, while that of phenol can reach 0.008 µM (0.74 µg L-1; see the Results section) in clouds collected at the PUY station. We showed in the past that when the cell / substrate ratios are kept constant the rates of biodegradation are constant (Vaïtilingom et al., 2010). The temperature (17 ∘C) corresponds to the average temperature at the PUY station in summer under cloud conditions. It is well known that under culture conditions in a laboratory, a lag time can be observed before bacteria starts to biodegrade phenol that corresponds to the induction period of the gene expression (Al-Khalid and El-Naas, 2012).

Before sampling, the evaporation of water has been compensated for by adding Volvic^® mineral water. A control experiment was performed by incubating phenol without bacteria; the phenol concentration remained stable with time (0.1 mM of phenol was obtained at the end of the experiment). For phenol quantification over time in the incubation experiments, 600 µL samples were centrifuged at 12 500 rpm for 3 min and the supernatants were kept frozen until HPLC analysis.

Phenol HPLC analysis. Before analysis, all the samples were filtered on an H-PTFE filter (pore size at 0.2 µm and diameter of 13 mm from Macherey-Nagel, Germany). Phenol detection was done on an HPLC VWR Hitachi Chromaster apparatus fitted with a DAD detector and driven by Chromaster software. Isocratic mode was used with a reverse-phase end-capped column (LiChrospher^® RP-18, 150 mm × 4.6 mm, 5 µm, 100 Å). The mobile phase was composed of acetonitrile and filtered water (Durapore^® membrane filters, 0.45 µm HVLP type, Ireland) in a 25 / 75 ratio with a flow rate of 1.2 mL min-1 (adapted from Zhai, 2012). The sample injection volume was 50 µL, spectra were recorded at 272 nm, and the run time was 10 min.

Phenol degradation. The percentage of phenol degradation was calculated by using the following equation. Phenoldegradation(%)=100-[Phenol]final×100[Phenol]initial The limit of phenol quantification was 3.8 µM. Strains are not considered active below 5 % of phenol degradation, corresponding to 5 µM.

Comparison of strain phenol degradation abilities was done using a nonparametric Kruskal–Wallis test (p value < 0.05) with Past software.

The objective of this paper was to explore the ability of microorganisms isolated or present in cloud waters collected at the PUY station to degrade phenol. From the 145 tested strains, 33 were isolated in this work; the others were already published (Table S1). We first checked for its presence in cloud waters sampled at the PUY station by performing GC-MS analysis. Figure S2 presents the back trajectories of the air masses corresponding to the five cloud events at the PUY station. The air mass origins of the five cloud samples determined from these back trajectories were classified as described in Deguillaume et al. (2014) and are reported in Table 1.

The GC-MS analysis performed on cloud samples allowed for the reliable identification and quantification of phenol in all samples (Table 1); the measured phenol concentrations ranged from 0.15 to 0.74 µg L-1. Figure S3a represents, as a typical example, the total ion current chromatogram of the cloud sample collected on 16 February 2016; the corresponding mass chromatogram based on the ion 94 current (characteristic for phenol) is represented in Fig. S3b. Quantification was done using similar mass chromatograms of all samples and the identification was proven by the correct retention time and accurate mass measurements (calculated: 94.0413; experimental: 94.0414).

The presence of transcripts involved in the biodegradation of phenol (Fig. 1) was investigated from prokaryotic messenger RNA-enriched metatranscriptomes obtained from three cloud water samples. Sequence data were searched for the presence of transcripts of genes involved in phenol biodegradation among the 281 sequences included in our database (more details about the affiliation of the sequences are given in Table S2).

Gene transcripts were detected for all the enzymes, except the catechol 2,3-dioxygenase, showing a possible implication of the microorganisms in the degradation of phenol in cloud (Fig. 2). However, the number of hits and the relative abundance of the transcripts coding for the different enzymes varied according to the cloud samples. A total of 257 hits (sequence homology) could be counted for cloud 2, with only 70 in cloud 1 and 130 in cloud 3. Transcripts corresponding to the enzyme involved in the first step of oxidation of phenol leading to catechols (phenol hydroxylases and phenol monooxygenases) were the most abundant in clouds 2 and 3, while those corresponding to the cleavage of the catechol ring (catechol 1,2-dioxygenase) were dominant in cloud 1. For all the samples the transcripts corresponding to putative phenol degradation enzyme pathways (i.e., none explicitly described enzymes) remained low. However, the slight differences observed between the three cloud samples are not significant when analyzed by a nonparametric Kruskal–Wallis test.

Figure 3 presents the relative abundance of the putative taxonomic affiliation of microorganisms involved in phenol biodegradation based on the information associated with sequences in the databases. All the sequences were affiliated with Gammaproteobacteria from only two genera, namely Pseudomonas and Acinetobacter, corresponding to only four species (P. fluorescens, A. gyllenbergii, A. oleivorans, and A. pitii) matched with cloud transcripts, among a total of 50 (Table S2). This very low diversity was unexpected considering that sequences from 50 bacterial genera, including 109 species, were used for our search in databases. In addition, the relative abundance of sequences affiliated with a bacterial species varied a lot with the considered enzymes and clouds (Fig. 3).

Gammaproteobacteria were found to contribute up to 21 % of the ribosome sequences identified in bacteria in targeted sequencing investigations. Pseudomonas in particular was highlighted as one of the most represented genera (contributing alone up to 2 % of the ribosome sequences) and the most active genus based on its representation in transcriptomes and consecutively high ribosomal cDNA : DNA ratio (Amato et al., 2017a; Fig. S4). Acinetobacter and Rhodococcus were much less represented (<0.1 % of the ribosome sequences) but also accounted for groups of interest regarding potential metabolic activity.

Rhodococcus were previously isolated from clouds at the PUY station (Vaïtilingom et al., 2012), but genes for phenol degradation affiliated with this genus were not detected here.

From our strain collection of 826 culturable microorganisms isolated from clouds collected at the PUY station between March 2003 and June 2016, we selected strains belonging to genera of interest concerning their potential ability for phenol biodegradation. We choose to test specifically Pseudomonas and Acinetobacter strains as they were detected in our metatranscriptomic analysis. As no Acinetobacter was available in our bacterial collection we choose closely related genera, namely two strains of Moraxella and Psychrobacter. In addition, Rhodococcus is well known to biodegrade phenol in the literature (as well as Pseudomonas and Acinetobacter). Pseudomonas and Rhodococcus are also the most frequently found genera in culturable bacteria from clouds (Renard et al., 2016; Vaïtilingom et al., 2012). Altogether 145 bacterial strains were tested (Table S1). The percentage of phenol degradation measured by HPLC after 5 days of incubation at 17 ∘C is reported in Fig. 4 and in Table S1. As the objective of the work was to perform a large screening with different types of cells, an incubation duration of 5 days was chosen to be sure that the induction period necessary for laboratory experiments was long enough to be able to detect biodegradation ability for all the tested cells. This time is quite long for a cloud but the objective here is not to evaluate a rate of biodegradation but to investigate the potential of biodegradation of microorganisms present in cloud waters.

We found that 93.1 % of the 145 tested strains were able to degrade phenol after 5 days of incubation. Globally, in our experimental conditions, all the families tested were very good phenol degraders (see Fig. 4a). No significant difference was found in the capacity of phenol degradation between Pseudomonas, Rhodococcus, and Moraxellaceae strains. A focus on the Pseudomonas strains according to their species is presented in Fig. 4b. The mean capacity of phenol degradation varied between 31 % and 67 % (for Pseudomonas rhizosphaerae and Pseudomonas graminis, respectively); however, no significant difference was observed between the species according to the Kruskal–Wallis test. Considering specifically Pseudomonas syringae strains, which are the most abundant species among cultural strains present in cloud waters (Renard et al., 2016), only 2 out of 35 were not capable of degrading phenol (strains PDD-32b-31and PDD-69b-20; see Table S1).