Three elementary schools adjacent to major roads with an average daily traffic volume of over 100,000 vehicles in Seoul, Korea were selected.²¹ The schools were, on average, 226 meters (standard deviation = 49) apart from the main roads. Air samples were collected from nine classrooms (one per classroom and two to four classrooms per school) with natural ventilation made through opening interior windows between classroom and corridor during October through December 2019. We selected 1st, 2nd, and 3rd graders’ classrooms considering their higher classroom activity levels compared to the higher grades.

Indoor and outdoor air filter samples for total suspended particulate matter were simultaneously collected during school hours on weekdays and during the absence of students on holidays or weekends. Three indoor sampling pumps (Sensidyne, St. Petersburg, FL USA) were placed on the lockers at the back of the classroom at a height of 130 cm, similar to the average height of elementary school students in Korea,²² and two outdoor sampling pumps at the school playground (Figure S1). The sampling pumps connected to 37-mm closed cassettes loaded with polyvinyl chloride filters (pore size: 5 μm, SKC, Inc., Eighty Four, PA USA) were operated at 2.5 L/min indoors and 4 L/min outdoors from 8:00 am to 3:30 pm. To secure enough genomic DNA in samples for the 16s rRNA gene and ITS region sequencing, three indoor filters were combined for each classroom and two outdoor filters for the corresponding outdoor sample. All equipment, materials, and supplies were sterilized before use. After completion of sampling, the filters were placed in a sterile filter container and sealed with parafilm using a sterile technique. The sealed filters were then placed in a cooler with ice packs to maintain cool temperature and minimize potential changes in the microbial community during transportation to the laboratory where all filter samples were stored at −80 °C until the analysis. Environmental conditions of the classrooms were recorded on the day when the air samples were collected and they included the number of students, temperature, relative humidity, natural ventilation time, and air cleaner operation time. Outdoor wind speed, precipitation, temperature, and relative humidity on the sampling day were also recorded.

Genomic DNA (gDNA) in sampled filters was extracted using High Pure PCR Template Preparation Kit (Roche, Basel, Switzerland). The filter was put in a 2 mL screw-cap microcentrifuge tube along with 300 mg of 212–300 μm sterilized glass beads (Sigma, St. Louis, MO USA) and 650 μL of ultrapure water. It was pulverized twice for 30 s at 4,350 rpm using a homogenizer (Allsheng, Hangzhou, China).^23,24 After centrifugation at 20,000 g for 1 min, the supernatant was collected and centrifuged again at 20,000 g for 1 min. A 400 μL aliquot of the supernatant was collected and used as a pretreatment sample. Then, the sample was eluted from the column as recommended by the kit manufacturer and stored at −20 °C.

Each sample was prepared according to the Illumina 16S Metagenomic Sequencing Library protocols. The amount and purity of DNA were measured using the QuantiFluor dsDNA System (Promega, Madison, WI USA) and VICTOR Nivo (PerkinElmer, Waltham, MA USA). The samples that did not meet minimum cDNA concentration (10 ng/μL) were excluded from the 16s rRNA gene and ITS region sequencing (Table S1). ITS3-ITS4 region of fungal DNA²⁵ and 16S V3-V4 region of bacterial DNA²⁶ were used for gene amplification with the primer sequences shown in Table 1. The amplification step was performed to add multiplexing indices and Illumina sequencing adapters. The final products were normalized and pooled using the PicoGreen, and the size of libraries were verified using the TapeStation DNA screentape D1000 (Agilent, Santa Clara, CA USA). And then sequencing was performed using the Illumina MiSeq™ platform (Illumina, San Diego, CA USA).

Sequence analysis was performed using Quantitative Insights Into Microbial Ecology2 (QIIME2, v.2017.6.0), such as Caporaso et al. (2010).²⁷ At the Demultiplex phase, each sequence was divided into samples using DADA2 (ver.1.1.1) and quality checked to eliminate low quality.²⁸ The subsequent process also utilized DADA2 to filter out noisy sequences, correct errors in ambient sequences, and correct amplicon errors through the process of removing chimeric sequences and singletons. The clustering process used the q2-vsearch (ver.2019.7) tool and proceeded with an open reference OTU picking method. Sequences with similarity greater than 97% were clustered into Operational Taxonomic Units (OTUs), and OTUs were analyzed through SILVA and UNITE.^29,30

For the quantitative analysis of the total bacterial DNA, qPCR (quantitative polymerase chain reaction) was performed by using 518F and 800R (Table 1) among 16S rRNA region.³¹ The reaction mixture was prepared with 4 μL of 5x HOT FIREPol EvaGreen qPCR Supermix (Solis BioDyne, Tartu, Estonia), 14.6 μL of Nuclease-Free Water (QIAGEN, Hilden, Germany), and forward and reverse primer that were mixed with 0.2 μL and 1 μL of DNA template, respectively. Initial denaturation was carried out at 95 °C for 12 min, and 40 cycles of three-step process were carried out- denaturation at 95 °C for 15 s, annealing at 55 °C for 20 s, and elongation at 72 °C for 30 s.

For qPCR, a LineGene 9600 (BIOER, Hangzhou, China) was used. Using the qPCR absolute quantification method,³² qPCR was performed with the diluted Escherichia coli (E. coli) sample. PCR efficiency (R² = 0.993; PCR efficiency = 102.2%) was verified through a standard curve and specificity was confirmed through a melting curve. The amount of bacterial cDNA in sample was calculated according to the cycle threshold value (Ct value) of the diluted E. coli sample. To calculate the copy number in the bacterial sample, the Ct value of the diluted E. coli and the Ct value of the sample were compared. The OD600 value of the undiluted standard solution of E. coli sample used in the experiment was 0.7, and the corresponding DNA concentration measured after DNA extraction was 55.0 ng/μL. The E. coli standard sample solution was serially diluted to 10⁻⁵ times by a factor of 10.

Alpha and beta diversity were calculated and visualized using q2-diversity (ver.2019.7) in the QIIME2 plugin. Unidentified genera were excluded from the analysis. A rarefaction curve was created for each sample to estimate bacterial and fungal richness as a function of sequencing depth at genus level.³³ To explore the abundance and uniformity of bacterial and fungal taxa in the samples, a Shannon-Wiener index (HI), which is a diversity index considering both abundance and diversity, and Pielou’s evenness index (EI), which represents the evenness of a community, were calculated. Using principal coordinates analysis (PCoA), we evaluated similarity of the microbial community among schools and between different sample types (inside versus outside classroom) with Bray-Curtis dissimilarity index. Spearman correlation coefficients among the diversity indices and environmental factors were calculated by sample location (indoors and outdoors). For bacteria, the DNA copy numbers from the qPCR analysis were also included in the correlation analysis. All statistical analyses were conducted in R version 4.0 (CRAN) and statistical significance set at p < 0.05 and marginal significance at p < 0.1.

All nine classrooms in three schools had wood flooring and the number of students per occupied classroom ranged from 21 to 26 with an average of 23. Classroom average temperature was 22.6 ± 1.6 °C (standard deviation), and the average relative humidity was 42.2 ± 3.2 % (Table 2). The average time of natural ventilation made through opening interior windows between classroom and corridor was 47.8 ± 80.5 minutes/day while the average operating time of a classroom air purifier was 151.7 ± 142.5 minutes/day. Outdoor temperature during the sampling campaign varied widely from −0.5 (in December) to 17.0 °C (in October), but was significantly lower than classroom temperature (p < 0.01). The outdoor average relative humidity was 62.5 ± 11.2 % and significantly higher than indoors (p < 0.01). Average wind speed during outdoor sampling was 7.7 ± 2.5 m/s. It rained during the sampling period for classrooms 1, 3, and 4 in school ‘b’, and the average precipitation was 3.3 ± 6.7 mm.

Richness of bacterial and fungal samples is presented using the rarefaction curves (Figure 1). As rarefaction curve reached the asymptote (represented by a straight line with a zero slope), sufficient sequencing depth was confirmed. For bacteria, the indoor average OTUs was 240.3 ± 70.7, which was lower than outdoors (310.9 ± 81.4). For fungi, the average OTUs of classroom and outdoor samples were similar− 120.7 ± 27.1 and 122.2 ± 37.7, respectively. This indicated that bacteria in outdoor air samples were more diverse and more variable in diversity than classroom air while fungal diversity was similar indoors and outdoors (similar in average richness) but more variable in diversity in outdoor air samples. The number of bacterial OTUs was significantly (p < 0.01) higher than fungal OTUs, indicating that bacterial taxa in both indoor and outdoor air were more diverse than fungal taxa. This result was similar to the studies that found more diverse bacteria than fungi in soil, sugarcane (roots, shoots and leaves) and floor dust of homes.^34–36

Diversity and evenness (uniformity) of bacteria and fungi in the classroom and outdoor air is presented in Figure 2. Shannon-Wiener index (HI) of classroom airborne bacteria was lower (marginally significant, p = 0.09) than outdoors, but Pielou’s evenness indices (EI) indoors and outdoors were similar while HI for fungi was not different between indoors and outdoors even if significance level was set at 0.1 although fungal taxa were more evenly distributed in classroom air than outdoor air (p = 0.09). Kembel et al. (2012)³⁷ reported that diversity in the outdoor bacterial community was significantly greater than indoors, which was similar to our study finding. Comparison of bacterial and fungal diversities in our study indicated that both HI and EI for bacteria were significantly (p < 0.01) higher than fungi, which is consistent to the findings of Kettleson et al. (2015) report.³⁴

The top 30 bacterial and fungal genera in relative abundance are illustrated in the stacked bar graphs for the classroom and outdoor air samples (Figure 3). The average cumulative relative abundance of the top 15 genera accounted for 35.6 ± 9.0 % of total abundance for classroom bacterial samples, 22.3 ± 7.0 % for outdoor bacterial samples, 56.3 ± 7.8 % for classroom fungal samples, and 62.7 ± 8.0 % for outdoor fungal samples (Figure S2). For bacteria, Enhydrobacter, Streptococcus, and Micrococcus were the most abundant top three genera in classroom air while Methylobacterium, Pseudomonas, and Deinococcus were the most abundant top three genera in outdoor air (Figure 3A). The literature showed the main source of Enhydrobacter, Micrococcus, Staphylococcus, Streptococcus, Neisseria, and Haemophilus in indoor environments that were also found within the top 22 genera in our study was human occupants. These bacterial genera live in the human skin, mucous membrane, and the intestine.^38–43 Methylobacterium, a genus living in soil and water and existing widely in the environment,⁴⁴ was found in both indoor and outdoor environments as we observed in our study. Kocuria inhabits the human skin, mucous membrane, and larynx, and was also detected in indoor dust.^45,46 Acinetobacter is widely distributed in the environment such as soils and water and also found in hospital settings because it can survive exposure to various disinfectants.⁴⁷ For fungi, Cladosporium, Trametes, Aureobasidium, Sistotrema, and Alternaria were most abundant in both indoor and outdoor samples (Figure 3B). Cladosporium was the most commonly found genus inside and outside the classrooms.⁴⁸ Fungal spores of some species of Alternaria are allergens that cause hypersensitivity reactions, allergic asthma, keratitis, and skin infection.^49,50 Chronic exposure to Aureobasidium pullulans (a black and yeast-like fungus found in various environments such as soil, water, air, and limestone) from a contaminated humidifier or air conditioner may result in hypersensitivity pneumonitis (extrinsic allergic alveolitis).^51,52 Of the top 30 fungal genera in our study, 20 were parasitic or symbiotic on plants. Cladosporium, Trametes, Phlebia, Daedaleopsis, Clitocybe, and Irpex that were found to be relatively abundant in our study were mostly related to trees.^53–56 Our finding that more abundant bacterial genera indoors than outdoors were mostly related to humans is consistent to other study findings suggesting that students and teachers are likely to be a strong source of these airborne bacteria in classrooms; however, fungi in classroom air likely originate from outdoors.^9,57,58

To confirm the main source of microbes in classroom air, we examined the similarity of bacterial and fungal community compositions among schools, classrooms, and outdoor air samples using principal coordinates analysis. Our results showed that bacterial community compositions in classroom air in all three schools were generally similar (clustered together) but different from those in outdoor air that also varied widely by school (Figure 4A). On the other hand, fungal community compositions in the samples were generally clustered by school (different location) regardless of indoor or outdoor samples, indicating that fungal communities in indoor and outdoor air were similar within the same school but different among schools (Figure 4B). However, we did not observe different community compositions among classrooms with or without operating air cleaner. These results of community composition confirmed that human occupants were likely to be the major factor determining bacterial community composition in classroom air whereas outdoor air was likely the main source contributing to fungal community composition in classroom air as described in Burge’s report.⁵⁹

The copy number of bacterial DNA in the occupied classroom air was lower than that in outdoor air samples on the same day for the most samples, except for the classroom ‘1’ in school ‘a’ (Sa1) and classroom ‘1’ in school ‘c’ (Sc1) (Figure 5). For the classroom in the school ‘a’ (Sa1), the highest number of students (26) in the classroom might have influenced the concentration of airborne bacterial DNA in the classroom air. On the day of sampling at school ‘c’, outdoor temperature was below zero, and the sub-zero temperature might have influenced bacterial concentration in outdoor air. The copy number of bacterial DNA in the unoccupied classroom was similar to that in corresponding outdoor air. Our quantitative PCR results (amount of airborne bacterial DNA) were also consistent to the results of the semi-quantitative analyses of rarefaction curves and alpha diversity presented in the previous sections.

Evenness index for bacteria in classroom air samples was significantly and positively correlated (correlation coefficient = 0.81, p < 0.05) with classroom temperature (Figure S3A). The number of students was positively correlated with the number of bacterial DNA copies but not statistically significant. For outdoor bacterial air samples, temperature was negatively correlated with evenness index (−0.83, p < 0.01) but positively correlated with DNA copy numbers (0.74, p < 0.05) (Figure S3B). For fungi, the number of students indoors was positively correlated with evenness index (0.70, p < 0.05) (Figure S3C) whereas outdoor wind speed and evenness index of outdoor fungi had a positive correlation (0.77, p < 0.05) (Figure S3D). However, relative humidity, natural ventilation time, air cleaner operation time, and precipitation were not correlated with the microbial diversity indices both indoors and outdoors.

One of the limitations in our study was the small sample size, which might limit generalizability of the study findings. The small sample size also did not allow us to statistically evaluate the effects of environmental conditions on microbial community in classroom air. Future studies with a large number of schools and classrooms in different areas and seasons might provide more generalizable results. Nonetheless, our study findings provide useful information to better understand airborne bacterial and fungal microbiome including their major sources in occupied classrooms.