2.1 Experimental setup
Two lab-scale biofilters, BF1 and BF2, were constructed as acrylic cylinders with a 12 cm inner diameter and 25 cm height. Each biofilter was packed with porous perlite (0.54 void fraction) to form a 1.6-L filter bed with a height of 15.0 cm. An air compressor (Hailea ACO-318, Fuzhou, China) was used to feed air into the system. The gas flow rate was controlled by a flow meter (Zenxing LZD-4WB, Xianghu, China), leading into a sealed glass bottle containing liquid toluene to form mixed gas. A stainless-steel reactor equipped with a UV lamp (Cnlight ZW23D15W-Z436, Shenyang, China) was installed between the flow meter and the bottle containing liquid toluene, operating at 185 nm to produce gaseous ozone for BF1. For each biofilter, the packing media was initially mixed with 1.0 L of activated sludge collected from a municipal wastewater treatment plant (Xiaojiahe WWTP, Beijing, China). A nutrient solution containing NaNO3 (20 g/L), Na2HPO (1.6 g/L), and KH2PO4 (1.04 g/L) was sprayed directly on the filter beds of the two biofilters for sufficient humidity and nutrients. The leachate was discharged every day. The two biofilters were operated in parallel for 160 days in total. Both were operated in identical conditions without ozone for the first 44 days, and BF1 was fed with 200 mg/m3 gaseous ozone after day 45.
2.2 Performance analytical methods
Gaseous toluene concentrations were measured using a gas chromatograph (SHIMADZU, GC-14C, Kyoto, Japan) with a flame ionization detector (GC/FID). The temperatures of the column, injector, and detector were set at 100, 150, and 150°C, respectively. The CO2 level in the mixed gas sample was measured with an infrared detector (Testo, Testo 535 CO2, Lenzkirch, Germany), and ozone concentration was monitored by a UV absorbance detector (2B Technologies Inc, 106-M, Boulder, USA).
All concentrations were measured daily at 9:00, 13:30, and 18:00 (GMT +8) with six replicates at each point. The highest and lowest replicate points were discarded, and the remaining four data replicates were averaged to yield three resulting data points for each day. Removal efficiency and mineralization rate data used in correlation analysis with the microbiome at a particular time point were obtained by averaging the performance data of three days surrounding the date of microbial sampling, including the day of sampling and one day before and after.
2.3 Microbial sampling
Microbial samples were taken from both biofilters at days 38, 52, 66, 80, 94, and 160. Packing media were taken from depths of 1, 7, and 15 cm of the filter bed and the biofilms were detached and suspended in phosphate buffer saline (PBS) by sonication at 425 W, 21–25 kHz for 10 min (Ningbo Science Biotechnology SCIENTZ-IID, Ningbo, China). The sonicated suspension was centrifuged at 10000 ×g for 1 min and resuspended in 5 ml PBS. To exclude dead cells within the community, a fluorescent dye (propidium monoazide, PMA) was used to treat the microbial suspension by inactivating the DNA of cells with damaged cell membranes as well as exposed DNA (Guo and Zhang, 2014). PMA (Biotum, PMA™ dye, Fremont, USA) stock was prepared by dissolving 1 mg PMA in 100 μL of 20% dimethyl sulfoxide (DMSO) and stored at −20°C; 2.5 μL of 20 mmol/L PMA solution was added into a 500 μL microbial suspension. The mixture was incubated at room temperature for 5 min and occasionally mixed. The tubes were placed horizontally on ice and exposed to a 650 W halogen light at a 20 cm distance for 4 min. Then, DNA from PMA-treated aliquots were isolated with the FastDNA® SPIN Kit for Soil (MP Biomedicals, Solon, USA) following the manufacturer's instructions.
Samples from days 66, 80, 94, and 160 were divided into three identical samples after PMA treatment and before DNA extraction, therefore each sample was represented in triplicate to counter systematic biases from DNA isolation, PCR and sequencing procedures. Samples from days 38 and 52 were not sequenced as triplicates and therefore were not included in all analyses of this study to ensure consistent biological and statistical validity. They were used only for verification of factual differences caused by ozone and not systematic errors, as seen in Supplementary Material 1.
2.4 High-throughput sequencing and data analysis
The primer 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and reverse primer 806R (5′-GGACTACHVGGGTWTCTAAT-3′) were used with 12 bp barcode (Invitrogen, Carlsbad, CA, USA) (Caporaso et al., 2011). PCR reactions, containing 25 μl 2x Premix Taq (Takara Biotechnology, Dalian, China), 1 μl of each primer (10 mM), and 3 μl DNA (20 ng/μl) template in a volume of 50 µl, were amplified by thermocycling: 95°C for 3 min, followed by 30 cycles of 98°C for 20 sec, 55°C for 15 sec, and 72°C for 15 sec, with a final extension at 72°C for 10 min. The PCR instrument used was a BioRad S1000 (Bio-Rad Laboratory, Irvine, USA). The length and concentration of the PCR product were detected by 1% agarose gel electrophoresis. Samples with bright main strips between 288–310 bp were used for further experiments. PCR products were mixed in equidensity ratios according to the GeneTools analysis software (Version4.03.05.0, SynGene). Then, the mixture of PCR products was purified with an EZNA gel extraction kit (Omega, Norcross, USA). Sequencing of the 16S rDNA V4 region was carried out on an Illumina Miseq platform. Barcodes and adapter sequences were trimmed with Cutadapt (Martin, 2011), then truncated at 210 bp and denoised with DADA2 to formulate the ASV (amplicon sequence variants) table (Callahan et al., 2016). Taxonomy of 16S rRNA sequences were classified with the Silva 16s rRNA database (release 138) at 99% similarity with the Naïve-Bayes algorithm for all analyses requiring taxonomy (Quast et al., 2013; ; Glick, et al., 2004). The software Bugbase was used for functional prediction, which required the Greengenes database, for which Greengenes (version 13_5) was used (Desantis et al., 2006). Function predictions were done using Bugbase and PICRUSt2 (version 2.1.2b) (Ward et al., 2017; Douglas et al., 2020). Statistical analyses, including alpha and beta analysis, were conducted in the R package Phyloseq (version 1.3) (McMurdie et al., 2013). The R package ALDEx2 (version 1.2) was used to statistically identify pathways highly specific to ozone and control biofilters (Fernandes et al., 2014). Correlation analyses were performed and plotted using Pearson correlation incorporated in the R package ggcor (Huang et al., 2020). Co-occurrence network analyses were conducted with the molecular ecological network analyses with Spearman correlation, Bray-Curtis dissimilarity, and an RMT (Random matrix theory) threshold of 0.81 (Deng et al., 2012). Networks were plotted with Java software Gephi (Bestian et al., 2009). All R packages were conducted under R version 3.6. LEfSe analysis was performed on the only galaxy platform maintained by the author, with p-value threshold of the Wilcoxon test set to 0.05 and LDA log-score threshold set to 3.0 (Segata et al., 2011).
2.4 Metabolic activity analysis
The suspension from sonification containing microbiomes detached from the packing media as described in Section 2.3 was diluted in PBS to obtain the optical density at 600 nm wavelength (O.D.600) at 0.05. The ECO plate (Biolog, Inc, Hayward, USA), with 31 various sources of carbon substrates mixed with tetrazolium dye, was prepared for the determination. Then, 150 µL of the microbial dilutions was inoculated to each well of the ECO plate and incubated at 30°C. The plate was observed for the absorbance at 600 nm regularly over a period of 3 days by the microplate reader (Molecular devices, Spectramax M5, San Jose, USA). The absorbance over time from wells containing carbon sources from the same group (e.g. amino acids) was averaged and the absorbance from the control well was deducted to avoid systematic error and obtain the average metabolic rates of different groups of carbon sources.