Introduction

To optimize the scenarios that predict the risks associated with the emergence and spread of disease, epidemiologists need to understand the complete process of pathogen dissemination and survival. But more fundamentally, describing the dissemination pathways of plant pathogens and microorganisms reveals the range of environmental contexts that structure populations. For aerial plant pathogens, knowledge about the favorable or limiting conditions controlling their dissemination through the atmosphere from local to global scales determines the success of predictions and thereby of the management of the plant diseases they cause.

Mechanisms of aerial dissemination have been well documented for fungi and bacteria from several meters near the Earth’s surface up to several kilometers into the troposphere (see reviews by Gregory, 1961 and Burrows et al., 2009). Atmospheric dissemination involves three processes: (i) emission from a source, (ii) transport and survival in the atmosphere via the circulation of air masses and (iii) deposition back to the Earth’s surface through impaction, turbulent airflows or precipitation. For obligate biotrophs such as rusts that grow on a limited number of well-known host plants, the inoculum source and time of emission often can be identified, leading to rather clear descriptions of their atmospheric movements across continents and oceans and to the elucidation of a link between source and sink populations (Purdy et al., 1985; Nagarajan and Singh, 1990; Isard et al., 2005; Krupa et al., 2006; Pan et al., 2006; Tao et al., 2009; Wang et al., 2010). But for non-obligate pathogens with wide host ranges and saprophytic phases of growth, it is difficult to link the populations deposited by precipitation or dry sediments to a specific source, thereby complicating inference of their dissemination trajectory (Morris et al., 2014b).

Consequently, constructing dissemination trajectories for such organisms involves compiling corroborative evidence from the time frame and geographic contexts of possible backward trajectories that are calculated from the endpoint of deposition (Prospero et al., 2005). Tools are available for modeling the trajectories of air masses and they have been applied to evaluating the long distance dissemination of various fungal pathogens and insect vectors of plant disease (Davis, 1987; Prospero et al., 2005; Pan et al., 2006; Pfender et al., 2006; Zhu et al., 2006; Aylor et al., 2011). For microorganisms with a multitude of possible emission sources, such tools have been recently applied to describe the global dispersion of airborne bacteria and archaea with winds (Smith et al., 2013). Pseudomonas syringae and Botrytis cinerea are prime examples of such microorganisms for which dissemination history cannot be observed directly. These plant pathogens generate losses on a wide range of crops, for example, on kiwifruit, horse chestnut or tomato for P. syringae (Webber et al., 2008; Kunkeaw et al., 2010; Mazzaglia et al., 2012), or on tomato, raspberry and wine grapes for B. cinerea (Williamson et al., 2007). Both of these plant pathogens are aerially disseminated, have a saprophytic lifestyle and include many lineages characterized by wide host range (Holz et al., 2004; Morris et al., 2008). Both have been isolated from clouds as well (Amato et al., 2007). But they have fundamentally different traits that could have an impact on their deposition. First, P. syringae has the ability to nucleate ice formation (Lindow, 1983; Wolber et al., 1986; Möhler et al., 2008), thereby inciting conditions propitious for the formation of precipitation leading to deposition of the bacterium from the troposphere (Morris et al., 2008; Morris et al., 2010). To date, B. cinerea has been reported to not be ice nucleation (IN) active (Delort et al., 2007) and therefore cannot incite its own deposition in the same way. Second, bacterial cells are at least 10-fold smaller than most fungal spores: P. syringae is about 1 μm in length (Monier and Lindow, 2003) whereas conidia of B. cinerea are >10 μm in diameter (Pezet and Pont, 1990). These two properties, particle size and nucleation properties, are determining factors in the processes that lead to removal of microorganisms from the atmosphere with precipitation (Rodhe and Grandell, 1972; Flossmann et al., 1985; Rogers and Yau, 1989; Pruppacher and Klett, 1997). Small particles such as the cells of P. syringae cells are likely to be uplifted into clouds and eventually incorporated into cloud droplets. Their precipitation can be assured if crystallization of droplets occurs—a process that can be enhanced via the IN activity of these bacteria. Larger particles such as spores of B. cinerea will have shorter upward trajectories on average and many of the spores are likely to be washed out from the atmosphere by impaction with drops below the cloud. For these and all other non-obligate biotrophic plant pathogens, the role of precipitation as a dissemination vector, the capacity of these organisms to survive during the trajectory and the dissemination history of air masses leading to precipitation have never been explored.

In this study, we characterized the deposition of P. syringae and B. cinerea in snowfall and rainfall sampled between 2005 and 2010 in southern France. For 90 such samples, the abundance, IN activity and population structure of P. syringae were determined and pH and electrical conductivity were measured. Then, we inferred the dissemination trajectories of P. syringae by comparing these observations with backward trajectories and the associated climatic parameters computed under the HYSPLIT–4 model (Draxler and Hess, 1998). Finally, the conditions associated with the occurrence of P. syringae in precipitation were compared with those linked to the presence of the fungus B. cinerea (evaluated in a subset of the samples) to identify the conditions of deposition specific to each plant pathogen. This multidisciplinary approach, associating meteorological computations and microbial ecology, provides clues about the processes and conditions of their dissemination trajectories.

Materials and methods

Sampling of snowfall and rainfall

The field study was conducted from December 2005 to November 2011 at 14 sites mostly in southern France (Table 1). Samples were collected from a total of 25 fresh rainfalls and 65 fresh snowfalls by procedures that avoided contamination from local sources. Events are defined to be independent if they occurred at separate sampling sites or if there was at least a 24-h time lapse between the end of a precipitation event and the beginning of the next one. Details about samples are given in Supplementary Table S1. They were collected either in sterile plastic tubs (for rain) or on clean plastic tarps (1 m2) (for snow) that had been placed and elevated from the ground in situations where the snowfall event was anticipated. Rain collector tubs were 1 m tall and elevated to protect the rain samples from contamination by rain-splashed soil. Samples were recovered from the samplers no later than 3 h after the end of the snowfall event or when at least 200–500 ml of rainwater accumulated. Samples were transported in a cooler to the laboratory, kept at 4 °C and processed within 24 h. Electrical conductivity (EC) and pH were measured directly in an aliquot of the unprocessed sample with an electrochemical analyzer (Consort C561, UK, reference temperature at 25 °C). EC and pH values had an accuracy of ±1 μS cm−1 and ±0.01 pH unit, respectively.

Table 1 Sampling sites and number of samples collected
Full size table

Quantification of microorganisms

Snow samples were thawed overnight at 21 °C. Snowmelt and rainwater were processed within 24 h as described previously (Morris et al., 2008). Samples were concentrated 200 times by filtration across sterile nitrous cellulose filters (pore diameter 0.22 μm) before dilution plating. P. syringae and total culturable bacteria were enumerated in all samples and B. cinerea in 29 samples as described below (Table 1).

The concentrated samples were dilution-plated on KBC (King’s medium B supplemented with boric acid, cephalexin and cycloheximide) to enumerate P. syringae (Mohan and Schaad, 1987) and on 10% TSA (tryptic soy agar) for total mesophilic bacteria. Plates were incubated for up to 5 days at 22–25 °C. Up to 30 strains of P. syringae chosen randomly from a single dilution were purified from each sample. Putative strains of P. syringae were tested for production of fluorescent pigment on King’s medium B (King et al., 1954), for the absence of arginine dihydrolase and of cytochrome C oxidase as described previously (Morris et al., 2008). A set of 513 strains of P. syringae were stored at −80 °C in nutrient broth with 40% glycerol for further characterization.

For B. cinerea, aliquots were plated on selective BSTM (Botrytis spore trap medium) (Edwards and Seddon, 2001). The number of developing colonies of B. cinerea was recorded daily during 14 days of incubation at 20 °C and colonies were transferred to PDA medium as described by Leyronas and Nicot (2013). Strains were maintained in stock cultures at −20 °C in a glycerol-phosphate buffer 0.06 M (50/50, vol/vol for further characterization). The detection thresholds were 5 colony-forming units (CFU) per liter for the targeted plant pathogens, and 55 CFU per liter for total bacteria. This difference in threshold was due to the fact that 10 times more replicates of each dilution were plated on KBC and BSTM than on TSA to assure detection of the pathogens. A set of 83 B. cinerea strains was stored for further characterization.

Characterization of trajectories and climatic conditions of air masses

The trajectories and the associated climatic conditions of air masses arriving at the sampling location on the day of the precipitation event were determined with the HYSPLIT model version 4 (HYbrid Single-Particle Lagrangian Integrated Trajectory, http://ready.arl.noaa.gov/ready/HYSPLIT4.html). This model, developed by NOAA Air Resources Laboratory and Australia’s Bureau of Meteorology, uses previously gridded meteorological data to calculate forward and backward trajectories and the associated climatic conditions and provides simulations for atmospheric particle dispersion and deposition (Draxler and Hess, 1998; Draxler and Rolph, 2011). The backward trajectories of air masses for each precipitation event were determined for periods of 24 h, 48 h and 120 h preceding the event for endpoint elevations of 50, 500 and 1000 m above ground level. These altitudes represent a vertical section of the planetary boundary layer defined by Gregory (1961). Hourly climatic parameters were also obtained from the database including precipitation (mm), sun flux (W m−2), relative humidity (%), pressure (hPa), altitude (m) and temperature (°C). The meteorological data GDAS (Global Data Assimilation System) used by this model were produced by the National Center for Environmental Prediction. The total distance traveled by air masses (km) was determined under POSTGRESQL/POSTGIS (http://www.postgresql.org) and the map projections were made with Quantum GIS version 1.7. 0 (Development Team, 2011).

Statistical analyses

Statistical analyses were performed with data from all samples (n=90) for P. syringae and from 29 samples for comparisons of the two pathogens using the R software version 2.9.1 (The R Development Core Team, 2011). To compare abundances of the pathogens between samples when they were detected, values for population density were log10 transformed before calculations.

Phenotypic and genotypic characterization of microorganisms

All P. syringae and B. cinerea strains were characterized for their IN activity in distilled water as described previously (Morris et al., 2008). IN activity was evaluated by determining the freezing temperature, between −2 and −8 °C of three 30 μl drops each containing 106 cells or spores and was scored positive if at least two drops froze. The pathogenicity of bacterial strains was determined by testing for the presence of the Type III Secretion System, a key factor in virulence (Cunnac et al., 2009). This involved testing for the capacity to induce a hypersensitive reaction in tobacco leaves (Nicotiana tabacum cv. Xanthi) as described previously (Morris et al., 2008). Utilization of D(–)tartrate was also characterized because this trait is rare among P. syringae (Sands et al., 1970) and therefore could provide a means to discriminate among certain biotypes. The pathogenicity of the fungus was determined by testing its ability to induce symptoms on tomato, one of its known hosts. The aggressiveness of B. cinerea strains on tomato plants (cv. Monalbo) was determined as described in Ajouz et al. (2010) by inoculating stems with a calibrated spore suspension (106 spores ml−1). Lesion length was monitored on three plants for 7 days and the whole experiment was repeated three times. The area under the disease progress curve was computed as described by Decognet et al. (2009). To facilitate the comparison of aggressiveness between strains, a relative aggressiveness index (in percent) was computed as the ratio between the average area under the disease progress curve of the tested strain and that of the reference strain BC1.

Morris et al. (2010) showed that phylogeny reconstruction on the basis of the gene for citrate synthase (cts) allows identification of groups within the P. syringae species complex. For 209 randomly chosen strains (about 11 per sample), the cts gene was sequenced and then strains were classified as described previously (Morris et al., 2010; Monteil et al., 2012) to characterize the genetic structure of the P. syringae population.

Inference of population structures by landscape clustering

We determined the spatial population structure of P. syringae to test the hypothesis that genetic diversity of P. syringae populations in precipitation is different according to the sampling zone. We used a clustering model: GENELAND version 4.0.2 under R software (Guillot et al., 2005a,2005b, 2012) to (i) estimate the number of panmictic clusters (k), (ii) assign to each genetic cluster a membership to a sampling zone and (iii) assign individual strains to a cluster. The clustering approach can combine data on genotypic polymorphism with geographic coordinates projected on a rasterized map. The optimal number of clusters k and the posterior probabilities of cluster membership for any unit of the sampling map were identified by a Markov chain Monte Carlo procedure under the assumption that there is no admixture between populations and that populations are at Hardy–Weinberg equilibrium with linkage equilibrium between loci (Guillot et al., 2012). Here, we assumed that allele frequencies of the cts gene were independent between samples. We thus used the uncorrelated model (UFM), applied the default settings and performed 500 000 iterations with an initial k=10 and a burning-in period of 50 000 iterations. A second iteration of the calculation was then performed with a fixed number K. Strains were assigned to a cluster according a threshold of membership coefficient of q>0.9.

Results

The occurrence of P. syringae in precipitation is correlated with chemical properties of precipitation and the associated air mass trajectories

P. syringae was detected in 25 of the 90 precipitation samples analyzed (snow and rain combined) with population sizes varying between 102 and 106 CFU l−1 when detected (Figure 1 and Supplementary Table S1). Each air mass trajectory was classified according to the three geographical sectors defined by Celle-Jeanton et al. (2009) for precipitations in southeastern France: west (region 1), north and east (region 2) and Mediterranean (region 3) (Figure 2a). Among our samples, air masses came more frequently from region 2 compared with those from the other regions (Figure 2b). However, they carried P. syringae significantly less frequently (12%) than air masses from the other regions (37% and 46%) (pairwise Fisher’s exact test, P<0.05). Concerning the chemistry of the precipitations, those carrying P. syringae had significantly higher conductivities (13±2 vs 8±1 μS cm−1) and were more alkaline (pH 6.22±0.08 vs 5.76±0.23) than those in which P. syringae was not detected (Mann–Whitney U, P<0.05). Overall, precipitation from region 3, which carries P. syringae more often than precipitation from other regions, had significantly higher conductivity than precipitation from the other two regions (Multiple MWU, P<0.05).

Figure 1
figure 1

Population densities of total culturable bacteria, P. syringae and B. cinerea in rainfall and snowfall. Black, gray and white bars represent total culturable bacteria, P. syringae and B. cinerea, respectively. BDT means ‘under Bacterial Detection Threshold’.

Full size image
Figure 2
figure 2

Occurrence of P. syringae in precipitation according to the origin of air masses. (a) Backward trajectories of air masses associated with the 90 precipitation samples are plotted for the 48 h preceding the sampling for the 500 m above the ground level. Each circle corresponds to a spatial coordinate for each hour. Each air mass trajectory was classified according to the three geographical sectors defined by Celle-Jeanton et al. (2009) for precipitations in southeastern France: west (●), north and east (○) and Mediterranean (). (b) Precipitation frequencies coming from these three sectors are represented by hatched bars. Gray bars represent frequencies of samples carrying P. syringae for each origin. Values associated with the same letter are not significantly different (multiple Fisher’s exact test, P<0.05).

Full size image

Aggregate means of climatic parameters of air masses preceding the precipitation events were compared for samples carrying and not carrying P. syringae for the three periods and elevations mentioned above. An example of typical results obtained is shown in Figure 3 for the 48-h period. All data are provided in Table 2. On the whole, results for comparisons for the three periods and three elevations were consistent, with a better discrimination for the shorter periods and lower elevations. The cumulative solar radiation, temperature, pressure and distance traveled by the air masses were significantly greater when P. syringae was detected in precipitation than when it was absent (MWU, P<0.05). Mean altitude and relative humidity of the air masses during their trajectories did not have a consistently significant effect on the presence of P. syringae in precipitation, but when their effects were significant, the values for these parameters were smaller when the bacterium was detected compared with when it was absent. Sun flux, temperature, altitude and pressure were intercorrelated (Spearman, P<0.001) as expected. No significant effects of cumulative rainfall during the trajectory on the occurrence of P. syringae were observed.

Figure 3
figure 3

Physical properties of air masses carrying P. syringae during 48 h preceding the precipitation event. Aggregate mean of properties of air masses (cumulated sun flux (a), mean temperature (b), mean pressure (c), mean altitude (d), mean humidity (e), cumulated rainfall (f) and distance traveled (g)) are compared between precipitations that carried P. syringae (gray boxes) and those that did not carry P. syringae (white boxes). Comparisons were performed for each elevation (50 m, 500 m and 1000 m). The mean with its standard error are presented next to each box plot. Values associated with the same letter are not significantly different (n=90, MWU test, P<0.05).

Full size image
Table 2 Physical properties of air masses carrying P. syringae during the 24 h, 48 h and 120 h preceding the precipitation event
Full size table

Air mass trajectory and climatic properties associated with the occurrence of B. cinerea in precipitation are different from those associated with P. syringae

For the subset of samples analyzed for both pathogens, B. cinerea was detected in precipitation at twice the rate of P. syringae (62.1% vs 34.5%, Fisher’s exact test, P=0.06) (Table 3). In contrast with P. syringae, B. cinerea was present in precipitation that was more acidic than samples in which the fungus was not detected (5.80±0.11 vs 6.31±0.17, MWU test, P=0.13) and its presence was not linked to temperature and pressure (MWU test, P>0.25). However, for the 50-m elevation at 24 h and 48 h before the sampled precipitation event, the presence of B. cinerea was linked to higher rainfall rates and higher maximum relative humidity experienced along the air mass trajectory before the sample was collected (Supplementary Table S2). No significant difference was observed among the different origins of air masses in terms of the frequency of precipitations carrying B. cinerea (43, 67 and 75% of air masses from regions 1, 2 and 3, respectively). However, the frequency of the pathogens in precipitation from region 2 was five times higher for B. cinerea than for P. syringae (Fisher’s exact test, P<0.05).

Table 3 Frequency of precipitations carrying P. syringae and B. cinerea
Full size table

Populations of P. syringae, but not of B. cinerea, are disseminated differently by snow and rain and display distinct phenotypic patterns in these substrates

P. syringae was carried more frequently, but at lower concentrations, by rain than by snowfall. There was a significant effect of the type of precipitation on the occurrence of P. syringae, with 68% of the rain events carrying the bacterium versus 12% of snowfalls (Table 3). However the population densities of the bacterium in snowfall were on average 10-fold greater than in rain. For total bacteria the opposite trend was observed with snowfall carrying 10% of the population densities observed in rainfall (Table 4). For the subset of samples used to compare the two pathogens, this result was slightly less marked for P. syringae, and for B. cinerea no differences in frequency and population size were observed between rain and snowfall (Tables 3 and 4). Overall, the abundances of B. cinerea spores in precipitation were 100-fold smaller than those of P. syringae CFU (MWU, P<0.001).

Table 4 Population characteristics of P. syringae and B. cinerea in precipitation
Full size table

All B. cinerea strains were able to infect tomato petiole stubs and to generate symptoms on the stem of tomato plants. The aggressiveness index for the 83 strains tested ranged from 11% to 144% (compared with the reference strain BC1). No differences between strains from rain and snow were detected. Finally, none of the strains of B. cinerea tested were IN active at temperatures warmer than −9 °C (Table 4).

In contrast, the phenotypic structure of P. syringae populations was highly influenced by the type of precipitation. All strains of P. syringae in snow were IN active and induced hypersensitive reaction in tobacco whereas rain contained some strains incapable of IN activity or inducing hypersensitive reaction. However, for the IN active strains from these two types of precipitation, there was no significant difference in their freezing onset temperature. Strains from rain more frequently were capable of metabolizing D(−)tartrate compared with those from snow (Table 4). In addition to its strong relationship with the frequency of strains inducing hypersensitive reaction on tobacco, the frequency of IN active strains was also significantly correlated with minimum pressure experienced by the air mass and the distance it traveled (Table 5), but not with other climatic parameters (data not shown).

Table 5 Frequency of ice nucleation active strains in precipitation in relation to the frequency of strains inducing hypersensitivity (HR) on tobacco and air mass properties
Full size table

Population structure of P. syringae is associated with the location and type of precipitation

There were sufficient numbers of strains of P. syringae from rain and snowfall to assess the effect of precipitation type on genetic population structure. However, this was not the case for B. cinerea. Therefore analysis of the influence of the type of precipitation on genetic population structure was conducted only for P. syringae.

The cts sequences (427 base pairs, 80 SNPs) of the 209 strains of P. syringae selected randomly from snow and rain represented 54 unique haplotypes. Half of the haplotypes were represented by only one strain in the whole sample, and the three dominant haplotypes were represented by 28, 22 and 19 strains. GENELAND estimated that there were three clusters (k) whether spatial coordinates were included or not. These latter results were consistent with those obtained using the STRUCTURE 2.3 software (Pritchard et al., 2000; Falush et al., 2003) (data not shown). Assuming admixture between populations, 89% of the haplotypes clearly belonged to only one cluster (q>0.95), indicating that the three clusters were highly divergent. As shown in Figures 4a–d, each cluster corresponded to a distinct spatial domain covering several sampling sites that are geographically close. Half of the strains belonged to cluster 3 that was specific to rain (Figure 4e). Cluster 1 was represented by strains of group 2 defined by Sarkar and Guttman (2004) for which at least one strain was present in each sample, with a higher proportion in snow (Figure 5).

Figure 4
figure 4

Population structure of P. syringae over the study area. The computation was performed with cts sequence polymorphism and spatial distribution of individuals, considering the uncorrelated allele frequency model. (a–c) Maps of posterior probability for each pixel of the study area to belong to the panmictic clusters identified: cluster 1, cluster 2 and cluster 3, respectively. A gray-to-white graded scale represents probabilities from 0 to 1 and sampling sites are represented by black dots. (d) Map of cluster membership to the study area of the three clusters. Clusters are represented by a different shade of grey and sampling site names are associated with black dots (CV, Col de Vars; CL, La Clusaz; LA, Col du Lautaret; LY, Montfavet St Paul; MO, Montfavet St Maurice; SP, Savines le Lac; ST, St Saturnin-les-Avignon; SZ, Super Sauze; VI, Villars de Lans). (e) The estimated membership value of each strain to each k cluster according to the type of precipitation: snow (hatched bars) or rain (gray bars).

Full size image
Figure 5
figure 5

Frequency of strains belonging to phylogroup 2 in snow and rain. The mean and its standard error are indicated next to each box plot. Values associated with the same letter are not significantly different (n=19, MWU test, P<0.05).

Full size image

Discussion

Our work provides new insight into the role of precipitation in the dissemination of airborne plant pathogens and the unique opportunity to estimate the fluxes of plant pathogens due to precipitation over crops. On the basis of the hydrological fluxes of snowfall and rainfall estimated by Oki and Kanae (2006), and our estimates of the frequency of precipitation carrying P. syringae and the mean concentration, we can estimate that up to 1020 cells of this bacterium could be deposited by rain and 1019 cells by snowfall across the world every year. Considering a hypothetical but realistic situation where (i) a single rain event of 20 mm would bring 107 cells of P. syringae per hectare of field, (ii) a leaf area index of a given crop of 1, and (iii) that every cell fell on a leaf, such precipitation would bring only 1 CFU per 100 cm2 of leaf (i.e., one cell for about 5–10 leaves). Compared with the populations brought by irrigation water, this is much less important. Irrigation retention basins can contain 100 CFU l−1 of P. syringae pv. aptata (Riffaud and Morris, 2002). Therefore, a single irrigation event (2 × 105 l ha−1) would transfer a total of 1.36 × 106–4.06 × 106 CFU per hectare of crop for an equivalent of 5 to 15 cells per leaf. The same calculation for B. cinerea leads to estimates of fluxes of the same order of magnitude.

The role of precipitation in the long distance dissemination of plant pathogens is still poorly understood. Here, we showed that the presence of P. syringae in precipitation is statistically linked to biological, chemical and climatic parameters of the precipitation water and the air mass that carried it. In particular, the probability of P. syringae deposition with precipitation can be inferred by determining air mass characteristics (i.e., temperature, sun flux, origin, distance traveled) and precipitation pH/conductivity. It is unlikely that such trends would be apparent and particularly the trends linked to air mass origin if the deposition of P. syringae with precipitation were due primarily to the incorporation of local and near surface airborne cells by impaction with precipitation droplets (scrubbing). The comparison with B. cinerea, whose presence in precipitation is indifferent to air mass origin, is important evidence that a major part of the cells of P. syringae in precipitation originates from the air mass at cloud height whereas the presence of B. cinerea is more likely due to below-cloud scrubbing. Overall, differences in frequency of occurrence of the two pathogens in precipitation support that they have different sensitivities to the climatic conditions regulating their incorporation into precipitation: (i) P. syringae is favored by warm conditions whereas B. cinerea is favored by high humidity and colder temperatures and (ii), presence of P. syringae in precipitation is promoted by alkaline precipitation whereas that of B. cinerea is promoted by acidic precipitation. Furthermore, snowfall and rainfall are not equal in their capacity to deposit P. syringae, but they are both equally efficient in depositing B. cinerea.

The relationships between pathogen deposition and environmental conditions raise questions about their causes. Do they indicate differences in (i) atmospheric survival rate relative to chemical and physical conditions, (ii) the strength of emission sources and/or (iii) the atmospheric processes regulating their transport? Distinguishing the effects of all sources of variation is challenging because the occurrence of a microorganism in precipitation is likely multifactorial. Scrubbing of the atmosphere below cloud level is one factor that could contribute to the presence of microorganisms in precipitation. Washout and impaction by falling precipitation below the cloud is the most efficient scavenging process for non-IN active particles larger than 10 μm such as spores of B. cinerea whereas it is negligible for particles smaller than 4 μm (Gregory, 1961; Rodhe and Grandell, 1972; Pezet and Pont, 1990; Mircea et al., 2000; Burrows et al., 2009). Therefore, below-cloud scavenging of P. syringae by precipitation is likely to be low—except for cells on debris or on spores—whereas it could be important for B. cinerea.

On the basis of the reasoning above, it is likely that a large part of the cells of P. syringae in precipitation collected at ground level are in the precipitating air masses at cloud height. The presence of P. syringae in clouds is well established (Sands et al., 1982; Amato et al., 2007; Joly et al., 2013). Therefore, it is legitimate to consider how our results make sense in terms of atmospheric and cloud physics. For example, the positive correlations between the occurrence of P. syringae in precipitation and temperature, sun flux and air pressure might simply suggest that low temperatures could be detrimental to the survival of P. syringae in the atmosphere. However, these observations are likely to be associated with the natural seasonal solar and temperature cycles that regulate bacterial emission into the planetary boundary layer (Lighthart, 1999). Solar radiation heats surfaces and creates thermal convective updrafts lifting epiphytic and soil bacteria into the atmosphere following solstices and diurnal patterns (Lindemann et al., 1982; Lighthart, 1997, 1999; Tong and Lighthart, 2000). This process is more efficient for formation of bacterial than for fungal aerosols. Fungi may release spores in the atmosphere independent of solar radiation, for example, by puff or tap mechanisms. Interestingly, we found several parameters of humidity to be related to the presence of B. cinerea in precipitation that are also associated with its emission into the atmosphere (Leyronas and Nicot, 2013). On the other hand, our results (Table 5) support the hypothesis that warmer temperature may select ice nucleation active P. syringae in the atmosphere and promote its deposition. Indeed precipitation would be more efficiently catalyzed by the pathogen at warm temperatures when nonbiological ice nucleators are not efficient (Morris et al., 2014a).

The multiple possible sources of non-obligate biotrophs such as P. syringae and B. cinerea and their long-range transport ability through the atmosphere make it almost impossible to track dissemination history. Each cell or spore in a precipitation event might have different dissemination histories both in terms of trajectory and exposure to selection pressures. Here, we addressed that question through the sampling of precipitation at different locations in natural habitats upstream of crop areas and downstream within a catchment basin. Sampling effort was highly dependent on weather forecasts and on access to locations where precipitation occurred thereby adding a possible bias. Future studies should target the role of landscape structure and attempt to distinguish it from that of the underlying atmospheric processes on microbial deposition with precipitation. Tackling the tracing of dissemination history will require appropriate sampling strategies: this will require the sampling of not only precipitations, but of the suspected local emission sources as well, for example, canopies, grasslands, crops and watersheds. Equally important, such a challenge requires identifying the right molecular indicators. Appropriate indicators will be source specific and they must persist in microbial cells at the same time scale as the duration of the transport process.

Precipitation chemistry might give insight into processes involved in transport with rain and snowfall. P. syringae is favored by moderately alkaline conditions whereas B. cinerea prefers acid conditions (Manteau et al., 2003). Recently, Attard et al. (2012) showed that acidic pH had a negative impact on the IN activity of P. syringae and thus possibly on its deposition. Thus, we can assume that acid pH in clouds decreases P. syringae deposition. However alkaline pH and relatively higher electrical conductivities are indicators of minerals in the atmosphere and notably of dusts (aluminum and calcite) (Delmas et al., 1996; Lee et al., 2000; Ozsoy and Saydam, 2000). The presence of dusts associated with microorganisms has been reported in clouds, air and precipitation, where dust has a protective role against the extreme conditions in the atmosphere (e.g., ultraviolet, desiccation) (Griffin et al., 2002; Harrison et al., 2005; Prospero et al., 2005; Kellogg and Griffin, 2006; Maki et al., 2008; Hara and Zhang, 2012). Therefore, our observations could be the result of the presence of dust acting as a shelter for P. syringae during atmospheric transport. Interestingly, air masses with the most alkaline and conductive precipitation originate from the Iberian Peninsula where southerly winds are known to transport a large amount of solid particles, essentially calcite dusts coming from the Sahara (Loye-Pilot and Morelli, 1988; Delmas et al., 1996; Kellogg and Griffin, 2006). Winds can increase upward fluxes of soil dusts in the atmosphere. Among the parameters we used to characterize air mass movement, the total distance traveled in a given time period by an air mass is an indication of its average speed. Although distance traveled did not have a significant effect on the presence of the two targeted pathogens, distance traveled was consistently greater when they were present.

Population genetic structure might also be useful in elucidating trajectory. Here, the clustering of microbial diversity in precipitation over space suggested that P. syringae populations in precipitation would be influenced more by local sources than by distal sources. Indeed, if local sources were not dominant, we could expect more numerous sources and consequently a more complex and mixed population structure. But we cannot exclude that the structure of P. syringae populations along the landscape is the result of a bottleneck exerted by atmospheric processes. A set of genotypes and IN activity is closely linked with each type of precipitation. Thus selection of IN active populations could be at the origin of the differences between the genotypic patterns we observed. Our experimental data for P. syringae and B. cinerea and their biology are congruent with the predictions made by atmospheric physics concerning the removal of particles by precipitation in relation to particle size and nucleation/condensation properties and cloud physics (Rodhe and Grandell, 1972; Flossmann et al., 1985; Rogers and Yau, 1989; Pruppacher and Klett, 1997). At mid-latitudes, snow is initiated by an ice crystal-process; snowflakes arise from crystals that are initiated by ice nuclei and then aggregate (Rogers and Yau, 1989). A particle smaller than 1 μm such as P. syringae (Monier and Lindow, 2003) can be incorporated in a snowflake if it is IN active (Rodhe and Grandell, 1972). For rainfall at mid-latitudes, however, two different processes can lead to rain droplet formation. Rain can begin as ice via the same process described above if temperatures in clouds are sufficiently cold. However, under warmer conditions and when there is sufficient turbulence, rainfall can be initiated by a coalescence process that involves condensation nuclei, growth of ice particle by collision with supercooled droplets and melting in the lower zone of the cloud (Rogers and Yau, 1989). In this case, particles can be incorporated into rainfall within the cloud independent of their IN activity (Rodhe and Grandell, 1972). These phenomena are coherent with our observation that 100% of P. syringae strains in snow are IN active whereas this is not the case for those in rainfall and might also contribute to enhancing the abundance of IN active bacteria in snow compared with that in rain. B. cinerea is not IN active and therefore would not benefit from the role of ice crystal formation in rain or snow. Interestingly, there is no difference between rain and snow in the concentrations of B. cinerea. However, other phenomena related to the physics of rainfall might be important for this fungus.

It is important to understand how pathogens, and microbes in general, disseminate and to identify the processes structuring their populations. Overall, our estimation of atmospheric fluxes and our results about diversity give new insights into the role of precipitation in the dynamics of airborne microbial populations. They suggest that precipitation contributes to metapopulation mixing over short- and long-range distances, and to triggering microbial growth through rain splashing (Fitt et al., 1989; Hirano et al., 1996), rather than contributing directly to the size of microbial populations on crops. For IN active microorganisms, this dissemination can be part of a feedback process between land cover and the atmosphere (Morris et al., 2014a). The link between landscape structure, precipitation and atmosphere, and the time scale interactions that occur, remain to be investigated to better understand microbial population dynamics over time and space and to identify the selection pressures that contribute to their evolution.