The variations in the abundance of cultivable bacteria concentration in the air samples of laboratories and surroundings
Cultivable airborne bacteria concentration of all samples varied from 1 CFU/m3, sample RL2 to 41 CFU/m3, sample CO3 (Figure. 2), significantly lower than the cultivable airborne bacteria concentrations of library, hospital, other indoor environmental air samples and outdoor air samples such as wastewater treatment plant measured with tryptic soy agar (TSA) and blood agar (BA) medium in previous studies [15, 54-56].
Corridor samples generally had higher CFUs than respective laboratory samples, except SC1. Larger volume of human flow in the corridor may be the main reason, since studies have suggested airborne bacteria emission rate of human breath could be up to 4.85×105 CFU/h/person in an air-conditioned room, and the concentration of airborne bacterial genomes in an occupied classroom was 12-2700 times of that in a vacant room . This also applied to CO3, the sampling site with the highest CFU, as the Polybahn station had the largest stream of human flow among all outdoor sampling sites. A potential reason for SC1 to have lower CFUs than SL1 was that SL1 was a very big laboratory with several rooms linked by an internal corridor, thus people used the internal corridor more often than the external corridor.
Cloning experiments were performed in SL2 one day before the sampling, which explained why its cultivable bacteria concentration was the highest among the laboratories and as high as SC2. This suggests laboratories could be the source of airborne bacteria. Even for laboratories like SL6, in which cloning experiments had not been performed for a year, there was still a considerable amount of cultivable airborne bacteria. However, several biology laboratories had less airborne bacteria than the material lab.
The variation in the abundance of 16S rRNA in the air samples of laboratories and surroundings
The abundance of 16S rRNA in the air samples of city laboratories and surroundings did not vary too much (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1). CL2 was the lowest one with 7.32×104 copies/m3, while CO1 was the highest with 1.02×105 copies/m3 (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1), which was still lower than most samples from suburbs and all samples from rural area.
The outdoor samples from the suburb were lower in the abundance of 16S rRNA in air than most other suburb samples. The lowest one was SOB, 9.92×104 copies/m3 (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1). Among all the indoor samples, the material laboratory and the physics department corridor SC32 were right in the middle, 1.43×105 copies/m3 and 1.38×105 copies/m3 (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1). Laboratories with frequent cloning experiments such as SL2 and SL3 had the highest abundance of 16S rRNA among all the laboratory samples, and they clearly had an impact on their nearby corridors. Other labs, SL1, SL4, SL5, and SL6 had low abundance of airborne 16S rRNA. For SL5 and SL6, the reason was that they were on the ground floor with open doors to the outside, while SL1 and SL4 had lower concentrations, because more experiments on eukaryocyte instead of bacteria were performed there. This led to decreasing 16S rRNA concentrations from SL2 to SC2, then to SC1 and SL1 which physically comprised one entire floor of the building (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1).
The rural outside samples, RO1 and RO2 had the highest abundance of 16S rRNA among all samples, 2.35×105 copies/m3 and 3.82×105 copies/m3, respectively (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1). Earlier studies suggested in summer and autumn rural areas were heavier in atmospheric bacterial loadings than urban and industrial areas. This made the corridors a valley bottom of atmospheric bacterial loadings. RC12 with lots of open windows was clearly impacted by outside, while RC11 was the lowest, 1.18×105 copies/m3 (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1).
Overall, all the indoor samples were higher in atmospheric bacterial loadings than family residences[58, 59], vacant classrooms, but comparable to occupied classrooms [57, 60]. The outdoor samples were also higher than urban air of Seoul, Colorado, Ji’nan, and Nanjing investigated in several previous studies [36, 59, 61-63], but quite comparable to Beijing, Milan and Berkeley urban air [64-66]. The atmospheric bacterial loading represented by 16S rRNA measured by qPCR showed a totally different pattern from the one exhibited by cultivable airborne bacteria concentration. There was no correlation between them. Different from cultivable airborne bacteria concentration, human flow was not the major contributor for the atmospheric bacteria loading. CO3, the Polybahn station was likely relatively low in uncultivable airborne bacteria and dead bacteria, while rural outside samples were high in these.
The variation in the abundance of ARGs in the air samples of laboratories and surroundings
The only target ARG not detected in any sample was sul1. Other 15 target ARGs were found in almost all the samples, except that there was no floR in RC11 (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1). Based on the absolute concentrations, all target genes can be categorized into 3 groups: the abundant ones with log of copies/m3 more than 10 in most sites included blaTEM, floR, sul2, aadA1; the rare ones with log of copies/m3 less than 6 in most sites included aac6II, ermA, qnrS, blaOXA10 and Staphylococcus spp; while the rest target genes fell into the medium group. Clearly, ARGs resistant of the same kind of antibiotics can behave differently. For example, as ARGs against sulfonamide, sul2 was in the abundant group, while sul1 was not detected; as ARGs against aminoglycoside, blaTEM was in the abundant group, while blaOXA10 was in the rare group. The situation was similar for aadA1 and aac(6’)II. In contrast, all target genes against tetracycline and vancomycin were in the medium group. These patterns coincided with the study by Li et al.  that sul1 could not be detected in Zurich air, but sul2 could and blaTEM was the most abundant ARGs. However, ARGs such as ermA, tetW, which were not detected by Li et al.  were found in our study and the relative abundances for ARGs and MGEs showed different characteristics (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1). Compared to Li et al. , in our study the relative abundances of sulII and intI1 were much higher; aac(6’)II and blaTEM varied in a much larger ranges.
Both the absolute and relative abundances of aac(6’)II were high in CL2, CC1, SO2, and extremely high in CO2, 2.19×106 copies/m3 or 26.11 copies/m3/16S rRNA copies/m3 (Figure. 3 & 4 & Supplementary Table. 3-6, Additional File.1), while in some labs, corridors and all the rural sample sites, its relative abundances were lower than the ones reported in Li et al.. aac(6’)II was the only ARG that was extremely high in CO2, which suggested the hospital should be the main source of this specific ARG.
The relative abundance of blaTEM was high in some laboratories and all the suburb and rural outside sites and extremely high in CL2 and CC1, 53.13 and 40.79 copies/m3/16S rRNA copies/m3 (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1), respectively, while the relative abundances of blaTEM at all the urban outdoor sites were lower than the level reported for Zurich air by Li et al. . Since their samples were from cabin air filters of cars , it is not surprising that their results were between our urban results and suburb results for a city small in area as Zurich. The absolute abundance of blaTEM we detected were quite high compared to the ones in composting plants in Beijing , but similar to the ones in other districts including railway stations areas, educational districts, medical districts, residential areas, and commercial districts in Beijing, Tianjin and Shijiazhuang .
Other target genes with only high relative abundances in CL2 and CC1 were floR, vanB, qnrS, qnrA and tetG (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1).qnrS and qnrA which are both FCA type ARGs were more abundant both absolutely and relatively in CC1 than CL2 (Figure. 3 & 4 & Supplementary Table. 3-6, Additional File.1). The possible explanation is that some other animal laboratories next to CL2 were the source of these two ARGs. Though the absolute abundances of qnrS in our samples were higher than the ones in Nanjing, China, (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1), the relative ones in our samples, apart from those in CL2 and CC1, were actually lower (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1) , and fell into the range of the ones in the Eastern Mediterranean . For tetG, all abundances in our samples were higher than the ones in composting plants in Beijing , hospitals and farms in Ningbo. Most ones were higher than the ones in districts including railway stations areas, educational districts, medical districts, residential areas, and commercial districts in cities in Northern China, but they were comparable to the ones in Chinese wet markets in Shenzhen. The air from live poultry market had about 7.50 log(copies/m3 ) tetG , which was even higher than the values in CL2 and CC1, 3.05×106 and 3.65×106 copies/m3, respectively (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1). The results suggested that animal laboratories like CL2, could be a source of these ARGs because of their frequent contact with experimental animals in their research.
Both tetW and sulII were widely detected in previous studies. The absolute abundances of tetW in our samples were higher than the ones in districts including railway stations areas, educational districts, medical districts, residential areas, and commercial districts [36, 68], composting plants , clinics , and concentrated swine feeding operation , but comparable to concentrated poultry feeding operations . Its relative abundances in our samples were between 0.027 to 0.27 copies/m3/16S rRNA copies/m3 (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1), similar to the ones in Nanjing. The relative abundances of sulII in our samples were higher than the ones in Zurich air reported in Li et al.. The absolute ones were higher than the ones in composting plants, and comparable to the ones in districts including railway stations areas, educational districts, medical districts, residential areas, and commercial districts in Beijing, Tianjin, Shijiazhuang  and in Chinese wet markets in Shenzhen.
Other ARGs, such as ermA and acrA were also more abundant in our samples compared to the ones in hospitals and farms in Ningbo , even though both of them did not belong in the abundant group in our analysis (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1).
The variation in the abundance of MGEs and human pathogen bacteria (HPBs) in the air samples of laboratories and surroundings
Both MGEs were in the abundant target gene group (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1), but the relative abundance of TnpA was much higher than 1.00 copies/m3/16S rRNA copies/m3 at many sites, while intl1 was higher than 1.00 copies/m3/16S rRNA copies/m3 only in CL2, CC1, ML, SL3, RL1 and RC11 (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1). Nevertheless the intl1’s relative and absolute abundances in our samples were higher than those in previous studies [25, 62, 67, 69]. Especially, in the study of Li et al. , TnpA was not detected on air cabinet filters of automobiles in 14 cities including Zurich among all 19 cities distributed over the world. This contradiction could be caused by the property of TnpA that it can be more easily degraded than other target genes based on our experience. Our samples were freshly collected whereas the samples in Li et al.  were accumulated on cabin air filters on cars. Among our samples, TnpA was the highest in CO3 and CO1. It was also higher in SO1 and RO1 than in SO2 and RO2 (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1). These results lead to a hypothesis that TnpA may be contributed by large volume of human flow.
Staphylococcus. Spp belonged to the rare group in our study with the abundances varying from 41.85 to 521.99 copies/m3, while E.coli was in the medium group with the abundances varying from 3580.56 to 26893.71 copies/m3 (Figure. 3 & Supplementary Table. 3 & 4, Additional File.1). The range of Staphylococcus. Spp in our samples was similar to its range in composting plants, while E.coli’s abundance was close to concentrated poultry feeding operations, and higher than the composting plants.
The PCA analysis of target genes in the air samples of laboratories and surroundings
In PCA analysis (Figure. 5), Axis2 could be largely explained by the location factor from city to rural area, while Axis1 could be partially explained by the environment factor if CC1 and CL2 were excluded: biology laboratories and city outdoor sites were slightly on the right side with the suburb and rural outdoor sites slightly on the left. ermA, mphA2, aac(6)’-II and Staphylococcus. Spp had more contribution from the city samples. The first three were exclusively contributed by the hospital, since they were only extremely relatively abundant in CO2, the site near the hospital (Figure. 4 & Supplementary Table. 5 & 6, Additional File.1) while tetW, sulII, 16S rRNA and E.coli had more contribution from suburb and rural outdoor samples. CC1 and CL2 mainly contributed qnrS, qnrA, blaOXA10, tetG and intI1, while aadA1 and TnpA pointed to the opposite direction. The results show that animal laboratories were abundant in ARGs related to animal use as expected, while other laboratories were more likely to be the source of aadA1. aadA1 encodes protein which can inactivate aminoglycoside antibiotics. Due to the fact that aminoglycoside antibiotics can cause toxic side effects to inner ear and are contraindicated in patients with myasthenia gravis and mitochondrial disease, they are reluctantly used for medical purposes, but streptomycin, kanamycin are very commonly used in biology laboratories for experiments such as cloning. Furthermore, during the sampling, we were also informed that neomycin and ribostamycin were used in laboratory SL1 where immunology was studied. Since SL4 is the cell room for a structural biology laboratory and CL1 mainly studies fungi, a hypothesis is that biology laboratories working on eukaryocyte may release aadA1.
The co-occurrence network of ARGs, MGEs and HPBs independent of sites
Though previous studies did find the correlations between 16S rRNA and certain ARGs or MGEs in air, water, sediment and soil samples [8, 23, 36], there were no correlations found in this study between 16S rRNA and other target genes analyzed by Pearson Correlation Coefficient.
The correlations between other target genes based on their relative abundances clustered into two groups (Figure. 6): the city and hospital related group consisting of aac(6’)-II, ermA and mphA2, the rural and animal husbandry related group consisting of 8 ARGs, E.coli and intI1. Notably, aadA1 was the only ARG with negative correlations with other five ARGs which happened to point to the counter direction as aadA1 did in the PCA analysis. The other 7 ARGs in the group were strongly positively correlated with each other. The correlations between MGEs, intI1 and ARGs including qnrS, qnrA and blaOXA10 suggest these ARGs may have a higher risk to transfer horizontally. These correlations are consistent with previous studies. However, our study did not find TnpA and sulII correlate with other target genes like previous studies did in different environments[7, 23, 67], this suggests that more ARGs and the co-occurrence between ARGs and antibiotic residuals in aerosol of biology labs, pharmaceutical plants, fermentation industry should be taken into account in future studies.