A set of atmospheric metagenomes (MG) and metatranscriptomes (MT) reported previously (ENA project PRJEB54740; Péguilhan et al., 2025) was reprocessed and reanalyzed specifically for genes and transcripts related to the processing of nitrogen compounds at the sequence level. Briefly, six aerosol (clear atmosphere) and nine cloud samples were originally collected in 2019–2020, from the instrumented atmospheric station of puy de Dôme Mountain summit (1465 m a.s.l., France) (Baray et al., 2020). For both atmospheric conditions, multiple high-flow rate impingers were deployed in parallel, each running at an air flow rate of 2 m³ h⁻¹ (Šantl-Temkiv et al., 2017), and the samples were accumulated for 2 to 6 h directly into a nucleic acid preservation buffer, at 0.5X concentration for clear atmosphere and 1X for clouds in order to compensate for concentration/dilution due to evaporation/water accumulation, so as to reduce potential methodological biases. Clear-atmosphere samples were collected at relative humidity (RH) ranging from 41 % to 78 % (55 % on average) while clouds were characterized by RH = 100 % and liquid water content >0 g m⁻³.

The DNA and RNA extracts obtained from each sample were sequenced by Illumina HiSeq (paired end reads of 150 bp). In our study, the reads of MG and MT were trimmed (quality > 30, length > 145 bp), and 20 million randomly selected of them were aligned against the protein sequences of the NCycDB database, using MMseqs2 (e-value threshold of 10⁻⁷). NCycDB was designed for metagenomic profiling of N-cycling genes (Tu et al., 2019). This contains 68 gene (sub)families, grouped into 8 main functions: nitrification, denitrification, assimilatory nitrate reduction, dissimilatory nitrate reduction, nitrogen fixation, annamox, organic degradation and synthesis, and “other”. For each sample, the number of reads similar to one of the 68 gene families (Tu et al., 2019) was normalized by the total number of reads in the dataset and by the average size (base-pairs) of orthologs inside each family. Values are presented in parts per million base pairs (ppmbp). In order to obtain taxonomic affiliations, the matching sequences (best hits) were compared with the generalist UniRef100 protein database (January 2025) (Suzek et al., 2015).

The ratio between the relative abundance of transcripts in MTs with respect to that of their corresponding genes in MGs (RNA:DNA ratio) is often considered a proxy for microbial activity (e.g., Baldrian et al., 2012). The proportion of reads identified as being involved in each N-cycling function (i.e., nitrification, denitrification etc) in both MG and MT were summed up, and the corresponding “RNA:DNA ratio” of a given function, i.e., the number of related transcripts in an MT with respect to that of genes in the corresponding MG, was calculated for each sample.

The presence of the biomarker gene for nitrogen fixation, nifH, was tested in 34 bacterial strains, previously isolated from clouds (Amato et al., 2007; Renard et al., 2016; Vaïtilingom et al., 2012; Vinatier et al., 2016) (Supplement Table S1). These were selected to include known nitrogen-fixing taxa such as Pseudomonadales, Sphingomonadales, Burkholderiales, Rhizobiales, Mycobacteriales, Rhodospirillales, Hyphomicrobiales, Rhodobacterales. Bacteria cells from cultures stored at -80 °C in 10 % glycerol were re-cultured on R2A medium at 17 °C. DNA was then extracted from colonies (QIAamp DNA Mini Kit; QIAGEN; ref 51304) following the manufacturer's protocol. A polymerase chain reaction targeting the gene nifH was performed from DNA extracts (∼ 20 ng), using the primers polF: 5'-TGCGAYCCSAARGCBGACTC-3' and polR: 5'- ATSGCCATCATYTCRCCGGA-3' from Poly et al. (2001). Amplification was carried out at final concentrations of 200 µM of each dNTP (Qiagen), 0.2 µM of each primer and 0.04 U of Platinum II Taq Hot-Start DNA Polymerase (Invitrogen; ref 14966005), in Platinum II PCR buffer (1X) and Platinum GC enhancer (1X) provided with the Taq, in a total final volume of 25 µL. The PCR was performed with the following thermocycling parameters: 7 min of initial DNA polymerase activation and DNA denaturation at 95 °C, followed by 30 cycles of denaturation (95 °C, 1 min), hybridization (55 °C, 1 min), elongation (72 °C, 2 min), and a final elongation step of 7 min at 72 °C. The length of the amplicons (∼ 450 bp) was controlled by electrophoresis on 2 % agarose gel.

In atmospheric MG, organic and inorganic nitrogen processing genes are equivalently represented (p>0.05, Student's test), contributing a total of 4.08 ± 2.33 parts per million base pairs (ppmbp). Organic nitrogen processes are dominated by glutamate and urea metabolisms (2.01 ± 0.53 parts per million) (Fig. 1A; Supplement Fig. S1 and Supplement Table S2). Regarding inorganic processes, together, denitrification, assimilatory and dissimilatory nitrate reduction, and biological nitrogen fixation contribute ∼ 99 % of the N-related reads, with respective contributions of 46 % ± 15 % (0.96 ± 0.36 ppmbp), 25 % ± 3 % (0.51 ± 0.09 ppmbp), 19 % ± 10 % (0.40 ± 0.24 ppmbp) and 9 % ± 4 % (0.18 ± 0.11 ppmbp). Nitrification and anammox contribute < 1 % (Fig. 1A; Supplement Fig. S1 and Supplement Table S2). Within denitrification, nitrite reduction prevails (0.71 ± 0.25 ppmbp), with two genes (nirK and nirS) contributing 75 % ± 26 % of the denitrification genes, and 35 % ± 12 % of the genes involved in inorganic nitrogen transformations. Nitrate reduction genes (napABC, narGHIJVWYZ) evenly contribute between 0 % and 40 % of the denitrification genes (Supplement Fig. S2). Within biological nitrogen fixation, the two most represented biomarker genes are nifH and nifW, whereas nifD and nifK are underrepresented (Supplement Fig. S2).

Overall, nitrogen processing genes are more represented in MG during clear atmospheric conditions (4.79 ± 1.66 ppmbp) than in clouds (3.36 ± 0.65 ppmbp) (p<0.05, Student's test). Inorganic nitrogen processing genes contribute 2.50 ± 0.90 and 1.63 ± 0.47 ppmbp in clear atmosphere and in clouds, respectively, of which biological nitrogen fixation genes are 0.26 ± 0.11 ppmbp (i.e., 10 ± 4 % of the inorganic N-related reads) and 0.10 ± 0.05 ppmbp (i.e., 6 ± 3 %) (p<0.05, Student's test), respectively (Fig. 1A; Supplement Fig. S1 and Supplement Table S2).

Functional expression was evaluated considering the relative representation of transcripts (mRNA) in MT (Fig. 1B) with respect to their corresponding genes in MG (RNA:DNA ratio, with higher values indicating higher expression levels). While ribosomal RNAs can persist up to several days at low temperatures (Schostag et al., 2020), the average half-life of an mRNA in a bacteria cell (Mycobacterium tuberculosis) at 37 °C is between 2 and 5 min, which is much shorter than the average duration of atmospheric transport (∼ 3–4 d; Burrows et al., 2009). Some studies have shown that under “stressful” conditions or during dormancy/inactive states, such as caused by a shift in temperature, the half-life may increase by a factor of 2–3, but it still remains on the order of only a few tens of minutes (Rustad et al., 2013). Most of the transcripts identified in cloud and aerosol samples were therefore likely produced by the cells while airborne.

On average, nitrogen processing genes are more represented in MT than they are in the corresponding MG with an RNA:DNA ratio > 1 (p<0.05, One-sample Wilcoxon test). Nitrate reduction, nitrate assimilation, biological nitrogen fixation, as well as glutamate and urea metabolisms all have an RNA:DNA ratio > 1 (p<0.05, One-sample Wilcoxon test), and this is higher in clear atmospheric conditions than in clouds (p<0.05, Mann-Whitney test) (Fig. 1C).

Functions with RNA:DNA ratios that are significantly greater than unity (p<0.05, One-sample Wilcoxon test) in clear atmosphere include nitrate reduction (3.23 ± 0.92), nitrite reduction (3.04 ± 0.81), nitrate assimilation (2.90 ± 0.94), dissimilatory nitrate reduction (1.73 ± 1.01), biological nitrogen fixation (2.22 ± 1.21), and genes associated with organic nitrogen metabolism, i.e., glutamate (2.90 ± 0.92) and urea metabolisms (3.15 ± 0.80). In clouds, functions with a ratio significantly > 1 include nitrate reduction (2.49 ± 1.62), nitrite reduction (4.71 ± 1.92), nitrate assimilation (2.01 ± 0.91), and glutamate metabolism genes (1.98 ± 0.64).

Functions with similar representation in MT as in MG (RNA:DNA ratio ∼ 1) (p>0.05, One-sample Wilcoxon test) include, in clear atmosphere conditions, nitric oxide reduction (0.76 ± 0.58) and nitrous oxide reduction (1.77 ± 0.83). In cloud conditions, the same applies to nitric oxide reduction (2.25 ± 2.77), nitrous oxide reduction (1.33 ± 0.76), dissimilatory nitrate reduction (2.10 ± 2.89), biological nitrogen fixation (1.40 ± 1.19), and urea metabolism (0.91 ± 0.78) (Fig. 1C).

Under clear-sky conditions, the atmospheric samples used to generate the data were collected at relative humidity (RH) values ranging from 41 % to 78 %, with a mean of 55 %; no relationship between the expression of biological functions and RH could be detected (see Péguilhan et al., 2025 for further details). Nevertheless, RH is known to impact the viability of model airborne bacteria, with often higher survival at extremely low or high RH levels (Cox and Goldberg, 1972; Wright et al., 1969), and to influence their gene expression patterns (Barnes and Wu, 2022). Larger datasets will be necessary to examine such relationships in the natural environment.

The bacteria associated with atmospheric nitrogen processing genes in MG and MT are affiliated with 61 distinct phyla and 331 orders (Fig. 2). Within each nitrogen function, the contributing taxa are evenly distributed at the phylum level, with no difference between clouds and clear atmospheric conditions, except for the rare taxa related to nitrification. The most represented phyla for most functions are Pseudomonadota (contributing from 5 % to 32 % of the N-associated functional sequences), Acidobacteriota, Actinomycetota, Bacteriodota, Bacillota and Planctomycetota (Fig. 2A). Some functions involve specific phyla, such as Nitrospirota and Nitrososphaerota for nitrification, and Cyanobacteriota for N₂ fixation.

At the order level, the dominant orders include Acetobacterales, Burkholderiales, Hyphomicrobiales, Sphingomonadales, and Rhodospirillales (Fig. 2B). Nitrification processes are contributed by orders such as Nitrososphaerales, Nitrosomonadales, Nitrospirales, Isosphaerales, and Cytophagales, which account for 100 % of the orders contributing transcripts in clouds. Certain bacterial orders are associated only with specific functions, such as Methylococcales for N₂ fixation and Pseudomonadales for urea metabolism.

In terms of diversity, similar taxa contribute to MG and MT for a given process, with differences in the relative abundance of taxa's representation. The major phyla in MG are also those dominating in MT, notably Planctomycetota and Acidobacteriota, which, for the denitrification process, account for <5 % and 7 % of the reads in MG and 36 % and 31 % in MT, respectively, under cloud conditions. Regarding nitrification, Acidobacteriota is the only phylum represented in MT, while this contributes only 10 % of N-related sequences in MG.

Thirty-four (34) strains of Alphaproteobacteria and Gammaproteobacteria isolated from clouds in earlier work were tested by PCR for carrying the N₂ fixation biomarker gene (nifH). These strains were selected to belong to taxa that include nitrogen-fixers, and they are among the most frequent viable bacteria in clouds (Vaïtilingom et al., 2012). In total, five strains (∼ 15 %) were positive: four Sphingomonadales and one Rhizobiales (Supplement Table S1).

Among the 30 rain samples assessed in this study, ammonium and nitrate concentrations ranged from 3.3 ± 0.1 to 79.8 ± 0.1 µmol L⁻¹, and from 1.0 ± 0.1 to 34.8 ± 0.1 µmol L⁻¹, respectively (Supplement Table S3). This was linked with the geographical origin of the air masses, as attested by backward trajectories, with continental sources as major contributors (Supplement Fig. S3) (p<0.05, Pearson test, on the correlation between nitrate + ammonium concentration and the origin of air masses).

Bacterial concentration in the samples ranged from 1.94 ± 0.88 × 10² to 6.73 ± 1.31 × 10⁴ cells mL⁻¹ (Supplement Table S3). This was not correlated with the origin of the air masses (p>0.05, Pearson test, Supplement Table S4).

The majority of bacteria in rain samples are affiliated with Pseudomonadota (between 80 % and 90 % on average, Fig. 3). The major orders are Burkholderiales, Cytophagales, Pseudomonadales, Sphingomonadales, Sphingobacteriales and Rhizobiales (Fig. 3). There are clear differences in the distribution of these orders, in particular Sphingomonadales and Burkholderiales, depending on the sample (Fig. 3). The amoA gene, related to ammonium utilization in bacteria (ammonia monoxygenase), was below the detection limit of qPCR (less than 10^0.25 copies mL⁻¹) in all rainwater samples, except 20230506_RAIN where 14 copies mL⁻¹ could be quantified.

The bacteria biomass and diversity in the nine rainwater samples investigated changed during the five days of incubation (Fig. 4 and Table 3). This is a plausible duration for bacteria's residence in the atmosphere, estimated around 3 to 4 d (Burrows et al., 2009), but the actual time spent by cells within atmospheric droplets is expected to be much shorter. The data were interpolated over the 5 d periods in order to determine rates, assuming, as a first-order approximation, that these remained constant throughout the incubation time.

Depending on samples, at the end of incubation, the cell number concentration ranged from 1/35 to 400 times that initially present. When increasing, this corresponded to generation times of 13.45 to 56.26 h gen⁻¹.

The dominant orders initially were Burkholderiales, Cytophagales, Pseudomonadales, Sphingomonadales, Sphingobacteriales and Rhizobiales (Fig. 3). In some of the samples, their relative abundance was modified after incubation. For example, Burkholderiales' representation increased from 7 % to 30 % in the 20230509_RAIN sample, and from 26 % to 65 % in the 20230313_RAIN sample. Similar trends were observed for Rhizobiales in 20231019_RAIN, and for Pseudomonadales in 20230507_RAIN, 20230313_RAIN, and 20230429_RAIN, in these cases at the expense of the initially dominant order Burkholderiales.

The amoA gene could be quantified in the samples only upon incubation for five days (Table 3), with values ranging from 20 to 40 gene copies mL⁻¹ of rainwater.

On average, NH4+ concentration decreased over the nine days of incubation (Wilcoxon test; p<0.05) from 28.43 ± 18.09 to 21.38 ± 18.02 µmol L⁻¹, i.e., between 0.26 ± 0.18 and 1.00 ± 0.50 of its initial concentration, depending on samples (Wilcoxon test; p<0.05) (Table 3). The inferred corresponding NH4+ bioassimilation rates ranged between 2.34 × 10⁻⁷ and 2.30 × 10⁻¹⁰ µmol h⁻¹ cell⁻¹ (4.61 × 10⁻⁸ µmol h⁻¹ cell⁻¹ on average) i.e., this ranged over three orders of magnitude. On the contrary, NO3- concentration did not significantly change, with concentration of 7.99 ± 4.66 µmol L⁻¹ in fresh samples and 7.87 ± 4.83 µmol L⁻¹ after incubation, i.e., 0.64 ± 0.41 to 1.25 ± 0.23 its initial values (Fig. 4 and Table 3).

PCA (Fig. 5) illustrates the variability of rainwater sample composition, and its evolution during incubations. The two first components explain 52 % of the variance and allow discriminating in particular marine from continental air masses. Samples from air masses originating from marine areas (Atlantic Ocean) were enriched in Na⁺ and Cl⁻ ions, whereas samples from continental air masses contained higher levels of NH4+, NO3-, and SO42- (p<0.05, Spearman's rank correlation). Continental air masses were also characterized by higher ambient temperatures at the sampling site, smaller water volumes, higher pH, and higher cell concentrations with respect to marine air masses (p<0.05). This is consistent with previous observations at this site (Péguilhan et al., 2021). Certain bacterial taxa could be also associated with air mass origin: the relative abundance of Sphingomonadales was significantly higher in samples from marine air masses, whereas Burkholderiales dominated in samples from continental air masses (p<0.05). Finally, the bioassimilation rates of ammonium and nitrate were positively correlated with the relative abundance of Sphingomonadales and negatively correlated with that of Burkholderiales (p<0.05), but they were independent from the initial concentrations of these ions, bacteria and amoA gene copies (p>0.05).

The biological and chemical composition of the atmosphere mirrors to some extent that of emitting surfaces. In atmospheric metagenomes and metatranscriptomes, we find approximately equal proportions (∼ 50:50 %) of genes associated with inorganic and organic nitrogen transformation. This is similar to findings in other environments such as oceans, rivers, sediments and plants (Deng et al., 2024; Nie et al., 2021; Song et al., 2022; Tu et al., 2017; Wang et al., 2021). The predominant metabolic functions identified are glutamate metabolism, denitrification and nitrate assimilation/dissimilation, as in plants, rivers and oceans (Deng et al., 2024; Nie et al., 2021; Song et al., 2022). In these environments, these processes can account for up to 80 % of the genes associated with inorganic nitrogen utilization, versus ∼ 90 % in the atmosphere in our study.

Active nitrogen-related functions are maintained by bacteria in the atmosphere, which might be critical for their physiology despite much shorter residence times (a few days) than in any other environments (Burrows et al., 2009). Surprisingly, higher gene expression levels (RNA:DNA ratio) of these functions were observed in clear air than in clouds, where the presence of condensed water is rather expected to promote biological activity. Liquid water can be retained by efflorescent aerosol particless at RH < 100 % (Cruz and Pandis, 2000), which could be sufficient to sustain biological activity in clear atmosphere. In addition, it is possible that multiple dry-wet cycles occurred, in particular, before collecting the clear-air samples, which could have contributed to enhance biological activity as compared with cloud water, as observed in soil (Xiang et al., 2008).

Bacterial diversity associated with nitrogen-related processes is composed of a core of taxa, which are recurrently present and active regardless of atmospheric conditions (clear atmosphere or cloud), and also predominant in surface ecosystems (freshwater, ocean, phyllosphere) (Deng et al., 2024; Nie et al., 2021; Song et al., 2022). This is made up of six phyla: Pseudomonadota, Acidobacteriota, Actinomycetota, Bacteroidota, Bacillota, and Planctomycetota, which contrasts with other ecosystems (plants, river etc.) where nitrogen-associated processes are typically dominated by a single bacterial phylum or order (>50 %–75 % relative abundance) (Deng et al., 2024; Song et al., 2022; Tu et al., 2017). Functions carried out by more diverse assemblages of microorganisms are less sensitive to environmental changes and therefore more stable over time (Maron et al., 2018).

In contrast, certain functions and their associated taxa have very low representation in the atmosphere, such as nitrification and anammox, usually carried out by Nitrososphaerales, Nitrosomonadales and/or Nitrospirales. These occur mostly in sediments from the ocean's mesopelagic zones (Deng et al., 2024; Song et al., 2022), i.e., in areas that are not in contact with the atmosphere.

Atmospheric bacteria assimilate and transform more nitrogen than they contribute through their biomass, suggesting that they may act as sinks of atmospheric nitrogen. However, the accuracy of these transformation estimates at global scale could be improved. For instance, the measured bioassimilation rates were obtained from rainwater incubations at 17 °C, which do not fully represent atmospheric aerosol conditions, as the liquid bulk phase does not account for the particulate nature of aerosols, and the temperature is relatively high compared to actual atmospheric conditions. Atmospheric regions such as the cloud transition zone (i.e., the interface between clouds and the free atmosphere) exhibit humidity levels close to saturation and a broader atmospheric coverage than previously estimated (Calbó et al., 2024; Ruiz de Morales et al., 2024). This region may be particularly favorable for bacterial activity, as it is expected to experience longer atmospheric residence times and higher temperatures than those prevailing within clouds.

Water availability is another important factor to consider. While ammonium and nitrate assimilation was extrapolated to cloud environments based on rainwater data, it is important to note that even under non-cloud conditions, aerosol particles are not entirely devoid of liquid water (Ervens et al., 2025; Pandis and Seinfeld, 1989), making nitrogen assimilation potentially feasible in clear atmosphere conditions as well.

Moreover, while these estimates are scaled to the global level, atmospheric conditions and stressors, nitrogen speciation and concentrations and microbial diversity and abundance are subject to strong spatial and temporal variability. Hence, nitrogen-related biological processes are expected to be highly variable in space and time as well. Our estimates, which are based on samples collected on a single location, for limited periods of time, and derived from laboratory incubations under defined fixed conditions are therefore associated with high uncertainties. Addressing variations and heterogeneity of these processes will require more investigations. One could envision enhanced microbial nitrogen cycling activity following major nitrogen release events (e.g., fertilizer application in agriculture). Conversely, such nitrogen “overloads” might induce nitrosative stress, potentially reaching toxic levels for airborne microorganisms. This is supported by chamber studies showing that high NO and N₂O concentrations can reduce bacterial culturability (Vernocchi et al., 2023).