Low airborne tenacity and spread of ESBL-/AmpC-producing Escherichia coli from fertilized soil by wind erosion

organic fertilizers were previously detected on soil surfaces of arable land and might be emitted by wind erosion. To investigate this potential environmental transmission path, we exposed ESBL-/AmpC-positive chicken litter, incorporated in agricultural soils, to different wind velocities in a wind tunnel and took air samples for microbiological analysis. No data exist concerning the airborne tenacity of ESBL-/AmpC-producing E . coli . Therefore, we explored the tenacity of two ESBL/AmpC E . coli strains and E . coli K12 in aerosol chamber experiments at different environmental conditions. In the wind tunnel, ESBL/AmpC-producing E . coli were detected in none of the air samples ( n = 66). Non-resistant E . coli were qualitatively detected in 40.7% of air samples taken at wind velocities increased emission of total viable bacteria with increasing wind velocity was observed. In the aerosol chamber trials, recovery rates of airborne E . coli ranged from 0.003% to 2.8%, indicating a low airborne tenacity. Concluding, an emission of ESBL/ AmpC E . coli by wind erosion in relevant concentrations appears unlikely because of the low concentration in chicken litter compared with non-resistant E . coli and their low airborne tenacity, proven in the aerosol chamber trials.


Introduction
Enterobacteriaceae including Escherichia coli (E. coli) can have the capability to produce ESBL-(extendedspectrum beta-lactamase) and AmpC-enzymes. These enzymes hydrolyze the beta-lactam ring of most betalactam antibiotics, including extended-spectrum cephalosporins belonging to the third, fourth and fifth generation. Genes encoding for the production of ESBL-and AmpCenzymes are found in pathogenic and commensal Enterobacteriaceae (Day et al., 2016). Among resistant bacteria, ESBL-/AmpC-producing Enterobacteriaceae are of outstanding importance and categorized as critical priority concerning the development of new drugs by the World Health Organization (WHO, 2017).
The presence of antimicrobial resistance in the environment represents an increasing 'One Health' problem. There is evidence of a transmission of ESBL-producing Escherichia coli (E. coli) and their mobile genetic elements encoding for antimicrobial resistance between humans, animals and the environment (Smet et al., 2010;Leverstein-van Hall et al., 2011). ESBL-producing E. coli are frequently detected in broiler barns, with a prevalence of up to 100% (Dierikx et al., 2010;Laube et al., 2013;Blaak et al., 2015;Hering et al., 2016;Daehre et al., 2017).
Besides the potential transfer of ESBL-producing E. coli from chickens to humans via direct contact (Dierikx et al., 2013;Huijbers et al., 2014) or contaminated meat (Leverstein-van Hall et al., 2011;Pietsch et al., 2018), ESBL-producing E. coli can spread in the farm environment and, thus, may act as a source for antimicrobial resistance. Apart from the spread through vectors such as flies (Blaak et al., 2014;von Salviati et al., 2015), another possibility of an environmental dissemination of ESBL-E. coli is the airborne route, which has been suggested from municipal sewage (Korzeniewska and Harnisz, 2013) and animal husbandries (Laube et al., 2014;Gao et al., 2015).
Applying ESBL-positive organic fertilizers to arable land presents a further important mode of transmission to the environment. For instance, ESBL-producing E. coli were detected in fertilized soil in studies carried out in Germany  and China (Zheng et al., 2017). In a French study ESBL-producing E. coli were even detected up to 1 year after the soil was fertilized with cow manure (Hartmann et al., 2012). Fields amended with ESBL-positive organic fertilizers might be affected by wind erosion, leading to bioaerosol formation (Jones and Harrison, 2004) and a possible airborne spread of these resistant bacteria. The microorganisms gain a lot of advantages from being transported together with soil particles, which ensure their survival during transport (Acosta-Martínez et al., 2015). The aim of this study was, therefore, to assess the relevance of airborne transmission of ESBL-producing E. coli by wind erosion after organic fertilization of fields. We performed three wind erosion trials with ESBL-positive chicken litter in a wind tunnel. In this context, the tenacity of resistant bacteria including ESBL-producing E. coli in the aerosolized state is of high interest and the current data situation is poor (Environment Agency, 2019). Therefore, we carried out a series of aerosol chamber experiments in order to investigate the airborne tenacity of two poultry-associated ESBL-/AmpC-producing E. coli strains and the nonresistant laboratory strain E. coli K12 under different conditions. This enabled a more in-depth interpretation of the results of the wind tunnel trial.

Results from the wind tunnel trials
Concentrations of ESBL-/AmpC-producing and nonresistant E. coli in the litter and soil-litter mixtures. The results of the quantitative and qualitative analysis of ESBL-producing and non-resistant E. coli in the chicken litter and soil-litter mixtures for all trials are displayed in Table 1.
The molecular characteristics of the E. coli isolate taken in the three wind tunnel trials are summarized in Table S1 (supplementary material).
Wind velocity, relative humidity (RH) and temperature measured in the wind tunnel trials and detection of ESBL-/AmpC-producing and non-resistant E. coli in the air samples. The wind velocity measured for the different levels of wind tunnel engine power and in different distances to the soil-litter mixture is depicted in Table 2. The wind velocity increased with increasing wind tunnel engine power and increasing distance to the soil-litter mixture because friction decreases with increasing distance to soil. Marginal deviations were observed in the wind velocity measurements between the three trials caused by changes in atmospheric conditions outside the tunnel, like air temperature and atmospheric pressure. For that reason, the arithmetic mean of the wind velocity of all three trials is depicted. A distinct variation of the environmental conditions (relative humidity (RH) and temperature) was observed between the three wind tunnel trials. The highest mean temperature was measured in the first trial at 25.7 C. The temperature was the lowest for the second trial with 16.1 C. The RH was highest for the first trial with 56% and lowest for the third wind tunnel trial with 29.7% (Table 2).
Sixty-six air samples were analysed throughout the three trials at different wind velocities (11 AGI-30 and 11 Coriolis μ air samples per trial). Neither ESBL/AmpCproducing nor non-resistant E. coli were detected in the AGI-30 samples. In the Coriolis μ samples, only nonresistant E. coli were detected, albeit in levels below the detection levels of the quantitative assay (Table 2). Statistical analysis using logistic regression revealed that an increased wind velocity significantly correlated with a qualitative detection of E. coli in the air samples (p = 0.022).
Total viable bacterial count, PM 10 and PM 2.5 emitted per m 2 of soil-litter mixture, soil aggregate size and collection efficiency of the AGI-30 and Coriolis μ air sampler. The geometric mean of total viable bacterial count emitted per m 2 of soil-litter mix for each wind tunnel trial is depicted in Fig. 1. Figure 1A shows the data for the AGI-30, Fig. 1B shows the data for the Coriolis μ air sampler. A similar emission was observed when comparing the three trials concerning the total viable count emitted at each wind velocity. However, with increasing wind velocity, a significant increase concerning the emission of total viable bacteria was observed in some trials for both air samplers. As all three trials showed a similar emission of total viable bacteria per m 2 , a pooled analysis was performed for both air sampling devices. The data are shown in the supplementary material in Fig. S1 for the AGI-30 (a) and the Coriolis μ air sampler (b). Here it becomes evident that there was a significantly increased emission between 5.6 and 7.3 m s −1 and also between 7.3 and 9.8 m s −1 but not between 9.8 and 10.6 m s −1 . Particle emission in Table 1. Quantitative and qualitative detection of ESBL-/AmpC-producing and non-resistant E. coli in the chicken litter directly before and after mixing with soil for the three wind tunnel trials. Trial 1  Trial 2  Trial 3 Litter (directly before mixing) Below detection limit (+) Barn A: 3.2 × 10 5 cfu g −1 Barn B: 4.2 × 10 5 cfu g −1 1.2 × 10 4 cfu g −1 Soil-litter mixture Below detection limit (+/−) 2.9 × 10 3 cfu g −1 Below detection limit (+)

ESBL-/AmpC-producing E. coli
Non-resistant E. coli Trial 1 Trial 2 Trial 3 Litter (directly before mixing) Not determined (+) Barn A: 2.0 × 10 7 cfu g −1 Barn B: 9.5 × 10 6 cfu g −1 8.0 × 10 5 cfu g −1 Soil-litter mixture Not determined (+) 2.4 × 10 4 cfu g −1 8.2 × 10 2 cfu g −1 The result of the qualitative analysis is displayed in brackets when samples were not quantifiable.  μg m −2 of soil-litter mixture is additionally depicted in Fig. 1 for PM 10 (c) and PM 2.5 (d). At wind velocities of 7.3, 9.8 and 10.6 m s −1 , significantly increased PM emissions were observed for the first trial. Soil-litter mixture aggregates were the smallest in the first trial with only 33.8% of the particle mass having a diameter of >500 μm. For the second and third wind tunnel trial, aggregates were coarser with 45.2% and 49.4% of the particle mass having a diameter of >500 μm respectively. Statistical modelling was used to investigate the influence of the air sampling device on the total viable bacterial count detected per m 3 of air in the wind tunnel trials. The three wind tunnel trials were not stratified for the analysis. The data, including the upper and lower 95% confidence intervals, are visualized in Fig. 2. Overall and between the different wind velocities, no significant differences were observed between the collection efficiency of the air sampling devices and confidence intervals show low maximum deviations.
The airborne concentrations of the three E. coli strains detected in the air samples at the different AGI-30 impinger heights are depicted in Fig. 3.
In the trials without organic soiling, the geometric mean concentration was lowest at 30% RH for all strains. The highest mean concentration in the trial without organic soiling was detected at 70% RH for E. coli K12 and E. coli R56 and at 50% RH for E. coli G-148-1. However, corresponding recovery rates were highest at 70% RH for all three strains.
In the trials with organic soiling, the highest airborne concentrations were detected at 30% RH for all three E. coli strains. The lowest concentrations were detected at 70% RH for all strains. The airborne concentrations and the recovery rates for all E. coli strains under different conditions are depicted in Table 3.
Differences in the recovery rates of the three E. coli strains For all experimental replicates under different conditions, the recovery rate of E. coli K12 from the aerosol was 0.16%. For the ESBL-producing E. coli R56, the recovery rate was 0.76% and for the AmpC-producing E. coli strain G-148-1, it was 0.79%.
The results of the repeated measures negative binomial model revealed that the recovery rate or the tenacity in the airborne status of the AmpC-producing E. coli strain G-148-1 was 4.1-fold higher compared with the non-resistant strain E. coli K12 (p-value = 0.016), taking all experimental replications under different conditions into consideration. For the ESBL-producing strain E. coli R56, the recovery rate was 3.5-fold higher compared with E. coli K12 (p = 0.041). The recovery rate of E. coli G-148-1 was 1.2-fold higher compared with E. coli R56 (p = 0.64).
Influence of RH and organic soiling on the recovery rate of airborne E. coli The influence of organic soiling on the recovery rate of E. coli was analysed in the repeated measures negative binomial model for the different RH. Since these factors had a similar effect on all strains, we did not stratify the three strains for the analysis. The calculated cfu of E. coli per m 3 , including the upper and lower 95% confidence intervals, are visualized in Fig. 4. At an RH of 30%, the recovery rate of E. coli from the aerosol is 41.3-fold higher when adding 10 g L −1 yeast extract and bovine serum albumin (BSA) (p < 0.001). At 50% RH, the recovery rate was only marginally (1.1-fold) higher with organic soiling (p = 0.91). At 70% RH, the addition of organic substances had an adverse effect. The recovery rate of aerosolized E. coli significantly decreased by 13.3-fold compared with the experiments where no organic substances were added (p = 0.003).

Discussion
This study aimed to assess the relevance of an airborne spread of ESBL/AmpC-producing E. coli from soil fertilized with chicken litter to the environment by wind erosion. The high total viable bacterial counts detected in the air samples indicate notable bio-aerosol formation, especially at high wind velocities. Wind erosion occurs when wind velocities exceed 6 m s −1 over dry soil . This is consistent with the significantly increased average total viable bacterial count (7.1 × 10 7 cfu m −2 ) detected at a wind velocity of 7.3 m s −1 in 5 cm distance to the soil-litter mix compared with the bacterial count detected at 5.6 m s −1 (2.0 × 10 6 cfu m −2 ). Additionally, only for wind velocities exceeding 7 m s −1 , non-resistant E. coli were detected in the Coriolis μ air samples. However, we did not detect ESBL/AmpC-producing E. coli in all air samples taken in the wind tunnel trials. Within each wind velocity, no significant difference was observed in the total viable count emitted for the three wind tunnel trials, which indicates a good predictability of total viable bacteria emitted from fertilized loamy sand soil for certain wind velocities. Although the emitted total viable count was similar for all trials, a significantly increased PM 2.5 and PM 10 emission was observed for the first wind tunnel trial. This increased fine dust emission was likely linked to the finer soil structure with smaller aggregates in the first wind tunnel trial, compared with the subsequent trials. The similar amount of total viable bacteria emitted in all three trials might be explained by assuming that larger particles, which were present in the second and third trials, might carry more total viable bacteria (Clauß, 2015). A further explanation for the similar bacterial emission for the three trials may be that the largest fraction of bacteria collected in the air samples was derived from the chicken litter and not from the soil, as recently suggested by Thiel et al. (2020), especially considering that the same amount of litter was used for all three trials. Based on our results on bacterial and particle emission, it can be assumed that fine dust and total viable count emission correlate. However, predicting wind-driven bacterial emissions from fertilized soil on the basis of particle emission is difficult, because it is influenced by many factors like soil type, soil structure, bacterial concentrations and composition of fertilizer and environmental conditions. Data on airborne ESBL/AmpC-producing E. coli in the environment are scarce and few studies investigating this issue were conducted to date. Laube et al. (2014) detected ESBL-/AmpC-producing E. coli in 7.5% (3/40) of air samples taken from the exhaust air of chicken barns, but a quantification was not possible for any of the samples. While Blaak et al. (2015) detected ESBL-producing E. coli in 7.7% of air samples taken inside broiler barns, all air samples taken in the vicinity of the barns were negative for ESBL-producing E. coli, which might indicate a low environmental tenacity of poultry-associated ESBL-E. coli in the airborne state. In a study by Korzeniewska and Harnisz (2013), 23.8% of air samples collected at a wastewater treatment plant tested positive for phenotypic ESBL-producing E. coli. In that study, a significantly positive correlation between the wind velocity and the number of ESBL-positive air samples was shown.
In our study, aerosol chamber trials were performed to gain additional knowledge on the tenacity of ESBL-/ AmpC-producing E. coli compared with non-resistant E. coli in the airborne state, which has never been investigated systematically to date. However, experimental studies on the airborne tenacity of non-resistant E. coli date back several decades (Poon,1966;Benbough, 1967;Cox, 1968;Wathes et al., 1986).
The airborne survival of bacteria under experimental conditions is influenced by various factors, including the bacterial strain, the composition of the culture and suspension fluid, the growth conditions, processing conditions of the bacterial cultures and the atmosphere into which the bacteria are released (Wathes et al., 1986). There are further important environmental factors, including the RH, oxygen concentration, temperature, ozone concentration, UV-radiation and air pollutants, which influence the tenacity of microorganisms in the aerosolized state (Zhao et al., 2014). The existence of overlapping lethal mechanisms for aerosolized bacteria renders the exact diagnosis of which mechanisms and conditions cause bacterial death difficult (Benbough, 1967).
Several studies have investigated the influence of temperature on the survival of E. coli in the aerosolized state.  Hoeksma et al. (2014) investigated the survival of the E. coli strain DSM-1936 between 10 C and 30 C and was able to show that after the initial decay, the bacteria survived the longest at 30 C. In contrast, Wathes et al. (1986) reported approximately four times higher death rates at 30 C compared with 15 C for a nalidixic acid-resistant strain of E. coli (serotype 0149). A possible explanation might be that the influence of temperature on the survival of airborne E. coli is strain-specific. Zhao et al. (2014) stated that, in general, the decay of microorganisms is faster at higher ambient temperatures. In our aerosol chamber experiments, a uniform temperature of 24 C was used for all experiments. The temperature in the wind tunnel trials varied between 16.1 C in the first trial and 25.7 C in the third trial. We did not observe a correlation between the temperature and the number of air samples positive for non-resistant E. coli or the total viable count in the wind tunnel trials.
Concerning the RH, we showed a significantly higher recovery rate of different E. coli strains in the aerosol chamber experiments without organic soiling at an RH of 50% and 70% compared with 30% RH. This finding is in line with a series of studies that showed decreased survival rates of aerosolized E. coli at low RH values. Wathes et al. (1986) reported half-life times of 3 min for low RH (<50%) and 14 min in humid conditions for E. coli aerosolized from phosphate-buffered saline (PBS) at 30 C. According to Poon (1966), the death rate of aerosolized E. coli is directly proportional to the rate of water evaporation and the water evaporation increases with decreasing environmental RH. Hoeksma et al. (2014) observed a fast biological decay of E. coli during the first 30 s after the aerosolization and a much smaller decay in the following 30 min, which he explains by a sudden cooling down effect caused by the evaporation of water. It is hypothesized that after aerosolization in dry conditions, water in the newly formed bacterial droplet will evaporate in milliseconds, thus altering the temperature, solute content and other environmental conditions of the droplet, which might promote biological decay (Liu et al., 2017). Additionally, under dry conditions, the toxicity of oxygen by the formation of free oxygen radicals which damage flavin-linked bacterial enzymes is increased (Benbough, 1967) compared with humid conditions, under which oxygen-induced free radicals are present to a lesser extent (HECKLY et al., 1963).
At 30% RH, bacterial suspensions with organic soiling showed a significantly increased recovery rate compared with suspensions without organic soiling. The explanation might be that airborne microorganisms are protected from external influences by particles coagulated within the viable bacterial particles, thus protecting bacterial proteins and membrane phospholipids, which appear to be targets for humidity and temperature-induced bacterial inactivity (Zhao, 2011).
In the trials with organic soiling, the lowest concentrations of aerosolized E. coli were detected in the aerosol at 70% RH (4.5 × 10 4 cfu m −3 air). This is unexpected because a high RH and the addition of organic substances both seemed beneficial for the survival of airborne bacteria. According to Marthi et al. (1990), a high RH might lead to clumping of cells, potentially increasing the odds of cell survival. This clumping effect may be enhanced by organic soiling in the E. coli suspensions. Viable E. coli may therefore have quickly deposited in large aggregates on the floor of the aerosol chamber and were undetectable in the air samples. Quantification of the deposited E. coli fraction might have been of particular interest. A limitation of this study is that we did not generate valid information about the particle size in the aerosol chamber trials, which may have supported this hypothesis.
In the wind tunnel trials, the RH was similar for the first and second wind tunnel trials with 56% and 54.4%. In the third wind tunnel trial, the RH was considerably lower, with 29.7%. Non-resistant E. coli were detected in 50% (5/10) of the Coriolis μ air samples in the first trial and 60% (6/10) in the second trial but in none of the air samples taken in the third wind tunnel trial. The lower environmental RH in the third wind tunnel trial might have led to inactivation by desiccation of E. coli adhering to soil or litter particles. This hypothesis seems inconsistent with the results of the aerosol chamber trial, where the highest survival rates under the presence of organic soiling were found at 30% RH. This deviation might be explained by the fact that we aerosolized the E. coli strains in the aerosol chamber trials from bacterial suspensions. Zhao (2011) stated that the biological decay of microorganisms aerosolized from dry sources might differ from the decay in wet aerosolization. He, therefore, recommended using dry aerosolization for microorganisms released from dry sources like faeces or litter. However, Hoeksma et al. (2014) pointed out that E. coli rarely survives the procedure for preparing dry aerosols.
The recovery rate for poultry-associated ESBL-/AmpCproducing E. coli strains R56 and G-148-1 was significantly higher than the recovery rate of the laboratory strain E. coli K12 under all experimental conditions. This is in accordance with findings by Marshall et al. (1988), who observed a prolonged survival of aerosolized E. coli of wild-type strains compared with the laboratory strain E. coli K12. The reasons for the significant differences in the recovery rates between the E. coli strains are still unclear. Additional research is warranted. To estimate the influence of ESBL-/AmpC-plasmid carriage on the tenacity of E. coli during aerosolization, a comparison of the recovery rates of ESBL/AmpC E. coli strains and their plasmid-cured variants would be of interest, as Ranjan et al. (2018) have demonstrated that the carriage of certain ESBL-plasmids was beneficial regarding competition fitness in vitro. Little is known about the molecular mechanisms and bacterial stress response in aerosolized bacteria. In a recent study by Ng et al. (2018), comparative transcriptome analysis was used to gain knowledge on gene expression in E. coli following aerosolization. Results indicate that E. coli responds to environmental stimuli in the air very quickly by changing the transcriptional signature. During aerosolization, 11 stress-responsive genes and 13 stimulus-responsive genes were regulated. It might be hypothesized that wild-type strains can change their transcriptional signature faster compared with laboratory strains in reaction to environmental stimuli.
Estimation of the potential of an airborne environmental spread of ESBL-/AmpC-producing E. coli by wind erosion In the wind tunnel trials, there were considerable differences in the concentrations of ESBL-producing E. coli in the chicken litter between the trials, reflecting realistic variations in the colonization of broiler flocks with these bacteria . While in the first trial, ESBL-E. coli were only qualitatively detectable in the chicken litter, in the second wind tunnel trial, chicken litter with a high concentration (3.2 × 10 5 cfu g −1 litter in barn A and 4.2 × 10 5 cfu g −1 in barn B) of ESBL-producing E. coli was used for aerosolization. Blaak et al. (2014) reported a similar average concentration (5.3 × 10 5 cfu g −1 ) of ESBL-producing E. coli in fresh chicken litter. The chicken litter used in the wind tunnel trials naturally contained ESBL/AmpC-producing E. coli, which were already adapted to the existing environmental conditions. This is beneficial because bacterial stress is minimized (Wesche et al., 2009) and may lead to a survival advantage over artificially added bacteria. The concentration of ESBL/AmpC-producing E. coli in the chicken litter and the mixing ratio of the soil-litter mixture reflected practical conditions, which ensures a good transferability of the results to field conditions. The two air samplers used in the wind tunnel trial have different properties. The airspeed at the inlet of the AGI-30 is 4 m s −1 , for the Coriolis μ it is 25 m s −1 . Therefore, air sampling was always super-isokinetic with the Coriolis μ, while it was only super-isokinetic for the AGI-30 at the lowest wind speed, where a mean of 3.3 m s −1 was measured in the suspensions chamber and sub-isokinetic for higher wind speeds (Table 2). Sub-isokinetic sampling is not ideal and has higher measurement errors compared with isokinetic or super-isokinetic sampling (Friedlander, 1977). Additionally, both air samplers have different particle cut-offs. The AGI-30 has a cut-off of 0.3 μm (Yao and Mainelis, 2006). Particles with a diameter above 15 μm are also not sampled in the AGI-30 collection fluid, because they are collected at the tube wall by inertial force (Lindsley et al., 2017). The Coriolis μ has a cut-off size of 0.5 μm at an operation flow of 300 L min −1 , which means that particles of 0.5 μm are sampled at 50% efficiency and larger particles are sampled at higher efficiency (Mbareche et al., 2018). Because of their different cut-offs, it was unexpected that both air sampling systems showed similar collection efficiencies. However, Clauß et al. (2013) stated that most airborne microorganisms are bound to particles with a size of 5-10 μm. For this size fraction, both air samplers have a high collection efficiency, which might explain the similar results concerning the total viable count detected per m 3 of air. Additionally, the detection limits of both air samplers differ. In our experimental setup, the quantitative detection limit was approximately 8.5 × 10 2 cfu m −3 for the AGI-30 air samples and 1.2 × 10 1 cfu m −3 for the Coriolis μ air samples. The qualitative detection limit was 8.0 × 10 1 cfu m −3 for the AGI-30 samples and only approximately 1 cfu m −3 for the Coriolis μ air samples. Despite the very low qualitative detection limit of the Coriolis μ, no ESBL-/AmpCproducing E. coli were detected in the air samples taken in the wind tunnel trials. The reasoning for this appears to be multifactorial. ESBL-producing E. coli are rapidly inactivated in chicken litter. In a study recently published by Siller et al. (2020), ESBL-producing E. coli concentrations decreased from an average of 3.4 × 10 5 cfu g −1 in fresh chicken litter below the detection limit after storage periods of 36 h in the summer and 72 h in the winter. To limit this inactivation in the litter used in the wind tunnel trials, the timespan between litter collection and aerosolization was minimized. A high concentration of ESBLproducing E. coli in the litter or fertilized soil seems to be essential in order to detect ESBL-/AmpC-producing E. coli in air samples. Chinivasagam et al. (2009) were able to show a direct link between levels of non-resistant E. coli in chicken litter and airborne E. coli concentrations in chicken barns. At concentrations of 10 8 cfu g −1 E. coli in the litter, airborne concentrations in the barns ranged from 10 2 to 10 5 cfu m −3 . In our study, despite the fact that ESBL/AmpC-positive litter was used in all trials, after mixing and diluting the litter with soil, quantitative detection of ESBL-producing E. coli in the soil-litter mixture was only possible in the second wind tunnel trial. Presumably, the vast majority of ESBL/AmpC E. coli present on the soil surface in the wind tunnel trials was subsequently inactivated during aerosolization. This hypothesis is strongly supported by the results of the aerosol chamber trials, which confirmed a low tenacity of aerosolized ESBL-/AmpC-producing E. coli with an average reduction of 2.25 LOG 10 .
The detection of high total viable bacterial counts and low amounts of E. coli in air samples taken in the wind tunnel might be explained by the fact that E. coli is a Gram-negative bacterium. Gram-negative bacteria account for a small proportion of airborne bacteria (Zucker et al., 2000). In poultry houses, Bakutis et al. (2004) reported a proportion of 2.6% of Gramnegative bacteria of the total bacterial count. Zhao et al. (2014) suspected that airborne Gram-negative bacteria are less frequently detected in air samples of livestock production systems, because they might be more vulnerable to environmental stress such as oxidation, radiation, and dehydration, presumably due to their thinner cell walls.
Additionally, it has to be considered that the ESBL/ AmpC-E. coli producing subpopulation represents only a small proportion of the total E. coli population. This could explain why ESBL/AmpC-producing E. coli have remained below the detection limit in the air samples while non-resistant E. coli were qualitatively detected. In the second wind tunnel trial, ESBL-E. coli represented 1.6% of the entire E. coli population in the chicken litter for barn A and 4.4% for barn B. In the third wind tunnel trial, 1.5% of E. coli were ESBL-producing. Considering all factors discussed, we conclude that an airborne spread of ESBL/AmpC-producing E. coli in the environment by wind erosion seems unlikely. However, because non-resistant E. coli were detected in the air samples, we suspect that ESBL-/AmpC-producing E. coli may have been present below the detection limit; therefore, a potential airborne spread of extremely low quantities of ESBL/AmpC-producing E. coli cannot be excluded. In two out of three wind tunnel trials, soil-litter mixtures with a low concentration of ESBL/AmpCproducing E. coli were used. If litter with a significantly higher concentration of ESBL-E. coli would be applied to arable land, the potential of emission might be increased. However, under practical conditions, it seems unlikely that ESBL-E. coli from chicken litter are applied to arable land in relevant concentrations because the largest proportion of ESBL-producing E. coli in chicken litter is quickly inactivated when transported from barns to arable land . Additionally, short-term storage (5 days) of chicken litter was proven to inactivate ESBL-E. coli effectively (Siller et al., 2020) and a recent labscale study by Thomas et al. (2019) confirmed the extinction of ESBL-E. coli in chicken litter within 2 h at temperatures exceeding 55 C. Additionally, the present study confirmed a low airborne tenacity for non-resistant and ESBL-/AmpC-producing E. coli. An airborne spread of ESBL/AmpC-producing E. coli from organic fertilizers from farm animals other than chickens seems even more improbable because ESBL/AmpC-producing E. coli are detected in broiler farms in the highest quantities and with a prevalence of up to 100% (Dierikx et al., 2010;Laube et al., 2013;Hering et al., 2016). Additionally, in the largest comparative study to date, by far higher concentrations of airborne bacteria were detected in poultry barns with 2.7 × 10 6 cfu m −3 air compared with 1.3 × 10 5 cfu m −3 in pig barns and 2 × 10 4 cfu m −3 in cattle buildings (Seedorf et al., 1998), which might be an indication for a pronounced potential of chicken litter, which typically has a high dry matter content, to form bioaerosols.
If emission of extremely low quantities of ESBL-/ AmpC-producing E. coli to the environment via wind erosion occurs, the risk of colonization in humans and animals remains unclear. Dungan (2010) stated that information on the infectivity of aerosolized enteric pathogens is scarce. However, it was recently shown by Robé et al. (2019) that an oral infection dose as low as 10 1 cfu ESBL-E. coli has led to persistent colonization of broiler chicks. This suggests that even very low quantities of these bacteria emitted to the environment might lead to health implications and thus highlight the crucial role of further research in the domain of airborne environmental antimicrobial resistance.

Experimental procedures
Wind tunnel trials Characteristics of the wind tunnel. Detailed technical descriptions of the wind tunnel located at the ZALF have been published by Funk (2000) and Funk et al. (2008Funk et al. ( , 2019. A modified sketch of the wind tunnel illustrating the most important technical features and the experimental setup is included in Fig. S2 in the supplementary material. The wind tunnel has a total length of 25 m and is of the Eiffel-type (push-type), meaning the air circulation is not a closed system. The airflow is generated by two axial ventilators, which are on top of each other at the beginning of the measuring section. The wind velocity is continuously adjustable up to 18 m s −1 . Rectifiers reduce turbulences and the profile former creates a logarithmic wind profile before the air enters the measuring section. The measuring section of the wind tunnel has a length of 7 m and a height and width of 0.7 m, equating a crosssectional area of 0.49 m 2 . It is accessible from one side and can be closed with acrylic glass plates. The floor space on which the soil-litter mixture was deposited measures 4.9 m 2 . Aerosolized particles are blown to the suspension chamber, where the air sampling devices were positioned. An exhaust air channel is located in the ceiling of the suspension chamber. The cross-sectional area of the suspension chamber is 12 times larger compared with the measuring section, resulting in a reduction of the average wind velocity to an eight of the original wind velocity.
Experimental design. To obtain ESBL-/AmpC-positive chicken litter for the wind tunnel trials, an initial screening was carried out to identify ESBL/AmpC-positive broiler barns. In total, 35 barns of two different broiler farms were tested by taking boot swab samples, pooled faeces and litter samples. All samples were qualitatively and quantitatively analysed for ESBL-/AmpC-producing E. coli (please see below for laboratory protocols). Litter from the barns with the highest ESBL/AmpC-producing E. coli concentration was chosen for further investigations in the wind tunnel trials. The concentration is shown in Table 1. In the first wind tunnel trial, the result of the qualitative analysis of the soil-litter mixture concerning AmpCproducing E. coli is uncertain, because only a few small colonies, which phenotypically resembled E. coli, were detected on the MC+ plates. These small colonies were transferred to a new set of plates and incubated overnight. The next day, no visible growth of E. coli was observed. In the second wind tunnel trial, litter from two barns was mixed because the concentration of ESBL-E. coli in the screening was similarly high.
For each trial, approximately 10 kg of chicken litter from the selected barns was collected in sterile plastic bags and stored in polystyrene boxes with ice packs until use. Trial 1 took place 1 day after litter collection; the chicken litter was stored overnight at 4 C. Trials 2 and 3 took place on the same day as litter collection. Two samples were stored for microbiological analyses: One directly after litter collection and one upon arrival in the wind tunnel.
The wind tunnel trials were carried out at the Leibniz Centre for Agricultural Landscape Research (ZALF). A concrete mixer was used to mix 3 kg of chicken litter with 250 dm 3 of loamy sandy soil taken from a trial plot of land. This ratio was chosen based on the German legislation, Paragraph 6 (4) of the Düngeverordnung (German Fertilizer Ordinance) which limits total nitrogen to 170 kg per hectare and data from the Leibniz-Institute of Vegetable and Ornamental Crops (2017), which estimates the average total nitrogen of fresh chicken litter from Brandenburg at 17.5 kg per metric ton. The grain size of the soil was measured by wet sieving (2000-63 μm) and the sedimentation method (<63 μm; DIN ISO 11277, 2002). This resulted in the following soil particle composition: Sand (2000-63 μm): 69%; Silt (63-2 μm): 25%; Clay (<2 μm) 6%. Derived from the particle composition, the soil can be classified as a 'loamy sand' and is therefore susceptible to wind erosion.
In the second wind tunnel trial, chicken litter from two barns was mixed at a ratio of 1:1, because these barns showed similarly high concentrations of ESBL-producing E. coli in the screenings. To ensure that the soil was negative for ESBL/AmpC-producing and non-resistant E. coli, soil samples for microbiological analysis were taken before mixing. After the mixing procedure, the soil-litter mixture was sampled again for microbiological analysis.
The wind tunnel was then filled with a 5 cm layer of soil-litter mixture and exposed to different wind velocities (5.6, 7.3, 9.8 and 10.6 m s −1 in 5 cm distance to the soil surface on average) for a time span of 10 min each. The wind velocity was monitored at a height of 5, 30 and 60 cm in the wind tunnel section using a hot wire anemometer (Lambrecht Thermal Anemometer 642, Lambrecht, Goettingen, Germany). The RH, temperature and PM 10 , as well as the PM 2.5 concentration [μg m −3 ], were measured using an Environmental Dust Monitor (EDM 164, GRIMM-Aerosol Technik, Ainring, Germany).
There was no exchange of the soil-litter mixture between wind velocities and the soil surface remained unaffected for each replicate of the wind tunnel trials. Therefore, total viable bacterial counts and particle concentrations were calculated by gradually adding up the total viable count and particles for each wind velocity, since a proportion of particles were already aerosolized at lower wind velocities and are therefore missing at higher wind velocity measurements, as described by Funk et al. (2019). Particle concentrations and total viable counts measured per m 3 of air were multiplied with the air volume (in m 3 ), which passed the wind tunnel for each wind velocity and divided by 4.9 (floor space in m 2 ) to calculate the emission of particles and total viable bacteria per m 2 of soil-litter mixture and therefore considering the dilution factor at increasing wind velocity. The experiments were carried out in triplicate for each trial. After one cycle, the soil was turned and mixed to create a fresh surface. Air samples were taken at the terminus of the wind tunnel using all-glass impingers 30 (AGI-30; Neubert Glas GbR, Geschwenda, Germany, VDI Norm 4252-3) and a Coriolis μ cyclone air sampler (Bertin Instruments, Montigny-le-Bretonneux, France). The impingers were filled with 30 ml of PBS (Oxoid, Wesel, Germany) and connected to a vacuum pump with a plastic tube. The airflow was monitored using a rotameter. It was approximately 12.5 L min −1 . Coriolis μ cones were filled with 15 ml PBS. The airflow for the Coriolis μ was set to 300 L min −1 . Sampling times for the AGI-30 impinger and Coriolis μ air samples were 10 min for each level of wind velocity. Impingers and Coriolis cones were stored at cool temperatures before and after air sampling. The air sampling devices were positioned 1.5 m above the ground, to imitate the height at which average humans respire. In total, 11 air samples were taken with the AGI-30 impingers and Coriolis μ per wind tunnel trial: Directly after the wind tunnel was filled with soil-litter mixture, a check plot sample was taken using impingement and cyclone sampling when the wind tunnel was turned off. At 5.6 m s −1 , only one air sample was taken for each trial. At 7.3, 9.8 and 10.6 m s −1 , three air samples were taken with each air sampling device per trial.
Laboratory analyses. All boot swabs, pooled faeces, litter, soil and air samples from the screenings of the barns and the wind tunnel trials were analysed qualitatively and quantitatively for ESBL/AmpC-producing and nonresistant E. coli as described by Siller et al. (2020). The air samples were additionally analysed for the total viable bacterial count.
Twenty grams of pooled faeces, litter and soil samples were mixed at a ratio of 1:10 with Luria/Miller-broth (LB; Roth, Karlsruhe, Germany) in stomacher bags. Boot swabs were put in stomacher bags and 200 ml LB was added. The samples were homogenized using a Stomacher 400 Circulator (Seward Limited, West Sussex, UK) at 200 rpm for 2 min. The quantitative analysis was performed by directly streaking 100 μl of the homogenized samples in triplicates on specific agar plates after serial dilution. For the quantification of E. coli, MacConkey agar No. 3 (Oxoid) was used (MC-). To quantify ESBL/AmpCproducing E. coli, 1 mg L −1 Cefotaxime (AppliChem, Darmstadt, Germany) was added to the MacConkey agar plates (MC+), as recommended by the EFSA (EFSA, 2011). Under optimal conditions, the quantitative detection limit of this method is 3.3 × 10 1 cfu g −1 and the qualitative detection limit is 1 cfu/20 g.
For the air samples, 100 μl were directly streaked on MC+ and MC-plates in triplicates. Additionally, 100 μl of the air samples were plated out on blood base agar (Blood agar Base No. 2, Oxoid) after serial dilution to determine the total viable bacterial count.
For qualitative analysis, the homogenized samples were incubated in LB medium for 24 h at 37 C. Three millilitre of air sample fluids were incubated under the same conditions after a 1:10 dilution with LB medium in Erlenmeyer flasks. Subsequently, 10 μl of each sample was streaked on MC+ and MC-agar with an inoculation loop and incubated again for 24 h at 37 C. The quantitative detection limit for the AGI-30 air samples was 8.5 × 10 2 cfu m −3 , the qualitative detection limit was 8.0 × 10 1 cfu m −3 . For the Coriolis μ air samples, the quantitative detection limit was 1.2 × 10 1 cfu m −3 , the qualitative detection limit was approximately 1 cfu m −3 .
Species confirmation of colonies suspected to be E. coli was achieved using MALDI-TOF Mass Spectrometry (MALDI Microflex LT and Biotyper database, Bruker Daltonics, Bremen, Germany).
Real-time PCR and sequencing. To detect the most important ESBL and CIT-type AmpC resistance genes (bla SHV , bla TEM , bla CTX-M and bla CMY ), isolates from the pre-trial screenings and the wind tunnel trials were tested using real-time PCR, as previously described by Roschanski et al. (2014). To identify the present ESBL-/ AmpC-variants, a selection of isolates was sequenced using Sanger-sequencing for each trial. The fewest isolates were analysed for the first wind tunnel trial (n = 6) because no AmpC-producing E. coli colonies were isolated from the soil-litter mixture. Most isolates (n = 16) were analysed for the second wind tunnel trial because isolates from two barns were characterised in this trial. For the third trial, 10 isolates were analysed. For Sanger sequencing, DNA isolation and PCR were performed as published previously . Purified PCR products were sent to LGC Genomics (Berlin, Germany), who provided the sequences. The obtained sequences were analysed using DNASTAR Lasergene (Madison, Wisconsin) and compared with the reference sequences of GenBank (https://www.ncbi.nlm.nih.gov/genbank/).
Phylotyping. Phylotyping of the isolates acquired from the litter used in the wind tunnel trials was performed as published by Clermont et al. (2013) with a modified PCR according to Projahn et al. (2017). Isolates that could not be assigned to one phylogroup due to unspecific band patterns were assigned a combined phylogroup.

Aerosol chamber trials
Characteristics of the aerosol chamber. The aerosol chamber trials were carried out in the aerosol chamber of the Institute for Animal Hygiene and Environmental Health (Freie Universtität Berlin).
Detailed technical descriptions of the aerosol chamber, including a Figure, have been published previously (Rosen et al., 2018). In brief, the aerosol chamber generates bioaerosols under different standardized climatic conditions (temperature, airflow rate and RH). The volume of the aerosol chamber is 7 m 3 . A perfusion pump transports the bacterial suspensions with a rate of 9 ml h −1 to an ultrasonic nebulizer. There is an axial ventilator in the ceiling of the chamber to disperse the aerosol and a separate opening in the ceiling for fresh air. The airflow rate was 100 m 3 h −1 and the temperature was 24 C for all experiments. The RH was set at 30%, 50% or 70% depending on the experimental setup.
The aerosol was sampled using three AGI-30 impingers at different heights (0.3, 0.8 and 1.3 m), each filled with 30 ml PBS. The airflow was approximately 12.5 L min −1 and the sampling time was 30 min.
Experimental design. In this experimental series, one non-resistant and two poultry-associated ESBL-/AmpCproducing strains of E. coli were aerosolized under various conditions. The ESBL and AmpC-producing strains were aerosolized together in a mixed bacterial suspension. Air samples were taken during aerosolization. We investigated the effect of different RH (30%, 50% and 70%) and the influence of organic soiling (10 g L −1 yeast extract and BSA, in combination) on the bacterial tenacity in the airborne state. Each E. coli strain was aerosolized at the three different RHs and with and without the addition of organic soiling respectively. Each experimental combination was carried out in triplicates.
The recovery rates of airborne E. coli were calculated by dividing the concentration of E. coli per m 3 measured in the aerosol chamber in the experiments by the expected concentration per m 3 . This expected concentration was calculated for each experiment individually by multiplying the concentration of the E. coli suspensions (per ml) with the forward speed of the perfusion pump (9 ml h −1 ) divided by the air exchange rate of the aerosol chamber (in m 3 h −1 ).
Preparation of the bacterial suspension and organic soiling solution. We used the well-described, nonresistant commensal laboratory strain E. coli K12 (DSM 423; https://www.dsmz.de/collection/catalogue/details/ culture/dsm-423) and two resistant, poultry-associated strains that were isolated from healthy chickens in a previous project in 2011 (Hering et al., 2016). The E. coli strain G-148-1 belongs to the multilocus sequence type (MLST) 10 and the phylogroup A. It harbours the CITtype AmpC resistance gene bla CMY-2 and the mcr-1 gene, which mediates colistin resistance. The second strain, first named R56 by Falgenhauer et al. (2016), belongs to the MLST 410 and phylogroup B1. It harbours the ESBL resistance gene bla CTX-M-15 and is additionally resistant to enrofloxacin.
Three colonies of the specific E. coli strain were added to 10 ml LB and incubated overnight at 37 C and 200 rpm in a shaking incubator (Multitron, Infors HT, Germany). The following day, 5 ml of this suspension was added to 100 ml of LB and incubated for 8 h. Then, 100 μl of the suspension was plated to blood base agar and incubated overnight (8 h) to achieve the exponential growth phase. The bacteria were removed from the agar plates with a plate spreader by adding 3 ml PBS and homogenized on a vortex mixer for 3 min with glass beads. To achieve the targeted concentration of 10 9 cfu (colony forming units)/ml, a fraction of the bacterial suspension was diluted with PBS at a ratio of 1:10. A measured value of 0.5 McFarland standards in the diluted fraction verified the targeted concentration of 10 9 cfu ml −1 in the bacterial suspension. The optical density was measured at 600 nm to validate the McFarland measurement (reference range: 0.073-0.11).
The organic soiling solution containing 10 g L −1 yeast extract (Merck, Darmstadt, Germany) and 10 g L −1 BSA (Sigma, St Louis, USA) was prepared by adding yeast extract to distilled water at a ratio of 1:50 and autoclaving. BSA was dissolved in distilled water at the same ratio and sterilized by membrane filtration. Afterward, the solutions were mixed at a ratio of 1:1.
For the addition of organic soiling, bacterial suspensions were centrifuged at 3000 rpm for 10 min, the supernatant was discarded and the pellet was resuspended with the organic soiling solution.
Microbiological analysis of the bacterial suspensions and air samples. All bacterial suspensions and air samples were analysed quantitatively. Serial dilutions were prepared for all samples.
In the experiments with E. coli K12, 100 μl was streaked in triplicates on MC-agar and incubated for 24 h at 37 C. In the experiments with ESBL-/AmpCproducing E. coli, all samples were streaked out in triplicates on two types of MacConkey agar No. 3 plates. Both plate types were supplemented with 2 mg L −1 cefotaxime. One set of plates additionally contained 4 mg L −1 of enrofloxacin, allowing a phenotypical selection for the R56 strain. The other set of plates additionally contained 7 mg L −1 colistin, leading to a phenotypical selection for the G-148-1 strain.

Statistical analysis
All statistical analysis was performed using R version 3.62 (R Foundation Vienna). Since bacterial counts (cfu) were lognormal distributed, we used the geometric mean for averaging. For statistical analysis, we used a mixed count regression. Due to overdispersion, we choose a negative binomial distribution. For the wind tunnel experiments, we used two hierarchical random effects. One random effect was used for the three soil-litter mixtures, which were measured at multiple wind velocities and within this effect, we used a random effect for the cfu plating triplicates as a repeated measures adjustment. The wind velocity and impinger type were modelled as a fixed effect interaction to assess the difference in measured cfu between AGI-30 and Coriolis μ at different wind velocities. Additionally, a model with the type of air sampler as the only fixed effect was run to compare air sampler performance overall. The probability of a qualitative detection of E. coli in the air samples depending on the wind velocity was analysed using logistic regression.
For the aerosol chamber experiments, a random effect for each of the experiments was utilized to account for repeated measures due to three AGI-30 impinger measurements per experiment and the combined aerosolization of two strains (E. coli R56 and G-148-1). We checked for interactions using interaction plots and found an interaction between humidity and organic soiling. Additionally, the strain was included as a categorical variable in the model. Post hoc comparisons between all strains were adjusted for multiple comparisons using the Tukey method. For a better comparison between the three strains, the concentrations in the bacterial suspensions were normalized for this analysis. All mixed models were performed using the R package lme4 (version 1.1-21). Estimated marginal means and multiple comparison post hoc tests were performed using the emmeans R package (version 1.4.5). Results are reported with 95% confidence intervals. A significance threshold of 0.05 was used. Figures were created using Graphpad Prism 8 (Gra-phPad Software, San Diego, CA).