Air samples were collected in the Pingshan market in Shenzhen, South China. Samplings were conducted by a high-volume TSP sampler (LS2031, IAT-Laoshan, Qingdao, China). Air was drawn at an average rate of 1.05 m³/min for 4 h (polluted air) and 23 h (clean air) resulting in approximately 252 and 1449 m³ of flow-through volume. Particulate matters were collected on 20.32 cm² × 25.4 cm² glass fiber filters (PALL, Port Washington, NY, USA) with 99.9% typical aerosol retention. The detail sampling information has been described in Supplementary Table S1. All filters were sterilized by heating at 500°C for 5 h before sampling.

The citric acid coated Fe₃O₄ nanoparticles were prepared by the co-precipitation method as previously described (Lin et al., 2014). The silica encapsulated Fe₃O₄ MNPs (Fe₃O₄@SiO₂) were prepared from the hydrolysis and condensation of tetraethylorthosilicate (TEOS; Xu et al., 2006). Subsequently, MNPs were characterized by transmission electron microscopy (TEM) and Fourier transform infrared (FTIR), as shown in Supplementary Figures S1 and S2.

To optimize separation efficiency during MNP assisted biomass separation, different buffering pH values (2.2, 7.5, and 10.0) or the introduction of silica encapsulates were evaluated. Briefly, after air sampling a 1/8 section of a TSP filter was placed in a clean beaker containing 40 ml buffer solution. After 30 s sonication for the biomass detachment, the glass filter changed into a pulp slurry. Then, 25 mg of MNPs was added and dispersed with another 30 s sonication. The beaker was gently shaken several times, to facilitate the attachment of MNPs onto microbial cells. Finally, the biomass was recovered through magnetic separation and was ready for the subsequent DNA extraction. In all experiments, the well-recognized conventional method described by Radosevich et al. (2002) was selected as the control method of our current study (Figure 1). The relatively small filter size selected here was aimed at a fair comparison with the conventional method control, in which 1/8 section of TSP filters was treated directly with DNA extraction kit without the separation of bacterial cells from glass fiber slurry. Total DNA extraction using the FastDNA Spin Kit for soil (MP, San Diego, CA, USA) was performed following the manufacturer’s instructions. DNA concentrations were determined by fluorescence spectrometer (Hitachi, Tokyo, Japan) using Hoechst 33258 staining dye. The zeta potential of MNPs and bacteria in different pH buffering were measured using Malvern Zetasizer (Malvern instruments Ltd., UK), Specifically, the pH buffer solutions were 0.2 M of Na₂HPO₄ (pH 7.5), 0.2 M NaH₂PO₄ supplemented with 0.1 M citric acid (pH 2.2), and 0.1 M of Na₂CO₃ fixed with 0.1 M NaHCO₃ (pH 10.0).

Bacterial DNA was amplified with primer pairs targeting the V4 region of the 16S rRNA gene (F515/R806; Albertsen et al., 2012). Barcodes and linker sequences that allow sample multiplexing during MiSeq sequencing were incorporated at the 5′ end of the forward primer (Zhou et al., 2011). The PCR amplification was conducted in an i-Cycler (Bio-Rad, Richmond, CA, USA). The PCR products from different samples were quantified using a Qubit 3.0 (Thermo Fisher Scientific, Waltham, MA, USA), mixed accordingly to achieve the equal concentration in the final mixture, and then sent to Macrogen Inc (Seoul, Korea) for the high-throughput sequencing on Illumina MiSeq platform. The next generation sequencing data were deposited in SRA database with the submission number of SUB1628949.

In total, 310,000 bacterial sequence reads (∼15,000 reads per sample) were recovered and subjected to MOTHUR software for all subsequent analyses (Schloss et al., 2009). The sequences were trimmed for the removal of primers, barcodes and sequences with ambiguous nucleotides, long homo-polymers or sequence length shorter than 200 bases. For the phylotype analysis, sequences were aligned, examined for chimeras, filtered, and finally classified taxonomically using the SILVA bacterial reference (Pruesse et al., 2007). For the OTU based analysis, aligned sequences were treated with the “dist.seqs” function to generate a distance matrix. Then, the “cluster” command was deployed to construct operational taxonomic units (OTUs) with cutoffs of 3 and 6% dissimilarity.

The target antibiotic resistance genes (ARGs) were quantified using QuantStudio 6 (ABI, Vernon, CA, US), the experimental conditions consisted of an initial denaturation for 5 min at 95°C, followed by 35 cycles of denaturation at 95°C for 15 s, annealing at 60°C for 30 s, and extension at 72°C for 30 s. The annealing temperature and primers are listed in Supplementary Table S2.

Development of a convenient method to separate bacterial cells from glass fiber pulp for the full capacity of TSP sampling is an urgent problem. Two major criteria for any potential separation methods are: (1) high efficiency, to eliminate the biomass loss; and (2) avoiding bias, to guarantee a similar microbial composition between original and separated biomass.

To realize the above criteria, citric acid coated Fe₃O₄ MNPs were utilized as the competing particle to scavenge biomass from glass fiber pulp. The design of our method in comparison with previous pretreatment methods is illustrated in Figure 1. It is believed that citric acid coated Fe₃O₄ MNPs have high surface area (Deng et al., 2008) and should be suitable for biomass adsorption. Different buffering pH and particle types were tested to optimize the biomass recovery, including bare Fe₃O₄ MNPs in neutral, acidic, and alkaline buffering solutions as well as the silica encapsulation of the MNPs. For a fair comparison, a 1/8 section of a filter membrane (∼60 cm²) was used in all extraction experiment corresponding to the maximum capacity of control method shown in Figure 1 (Radosevich et al., 2002). It should be noted that unlike the two newly developed methods shown in Figure 1, there were no separation of bacterial cells and filter fiber slurry was included during the whole workflow of the control method, thus no bias in the selection of bacteria could be guaranteed under this condition.

The ultimate DNA yields were used to quantify the separation efficiencies, as shown in Figure 2. Among the four treatments tested, Fe₃O₄ MNPs in acidic buffer resulted in the highest bacterial recovery. Under acid buffering conditions, the DNA recoveries of MNP methods were comparable to the conventional method control for both clean and heavily polluted air samples. Thus, much more DNA could be easily extracted by simply applying larger piece of the glass fiber filter during MNP assisted biomass separation.

To evaluate the possibility of any biased selection during MNP assisted biomass separation, the extracted DNA of different methods were compared with control method by their microbial community structures. Then, based on the OTUs distribution information, a co-linearity analysis was performed to quantify the similarity of microbial community structure between the different treatments, following Xia’s method (Xia et al., 2010). Briefly, the vector representing each of the five treatments was embedded in the N-dimensional space. N represents the total OTUs numbers present in all five samples. The angle between the vectors represents a measure of relatedness of different communities. The angle between acid buffered Fe₃O₄ MNP treatment and the conventional method control group was the smallest one [16.97° (cutoff = 0.03), 16.54° (cutoff = 0.06)] of the four treatments tested in the current study, as shown in Table 1 and also in Supplementary Figure S4. In addition to the heavily contaminated air sample near live poultry, the method was also used on a relatively clean ambient air samples outside the wet market. Similar small angels (Table 2) resulted between the acid buffered Fe₃O₄ MNP treatments and the controls, indicating good reproducibility of the present method. These angle values were also compared with some typical sample pairs, including the same DNA with different PCR amplifications (22.3°) or the same biomass with different DNA extractions (14.8°), as shown in Table 2. The quite small angel between Fe₃O₄ MNPs in acidic buffer treatment and control group indicated a high similarity of microbial communities and thus validated the unselective nature of the present method.

The bioaerosol samples collected in the current study harbor a diverse microbial community, which could be reflected by the rarefaction curve shown in Supplementary Figure S3. Facing such a complicated mixture of microorganisms, the MNP based methods in the acid buffer condition were successful and unselective in nature, which was comparable with duplicated PCR amplification of the same DNA sample.

For a deeper understanding of the variance of the separated microbial communities by different pH or particle coatings, a phylogenetic tree showing the exact deviations from the control method at class levels is shown in Figure 3. Among all four treatments, the acid buffered Fe₃O₄ MNPs treatment showed the least bias, in which 86% of separated populations showed acceptable recovery compared to the control method.

Buffering pH was a crucial factor for both efficiency and selectivity of biomass separation. At very low pH (pH 2.2), all types of bacteria cells almost completely adsorbed onto the Fe₃O₄ MNPs without any discrimination. As the pH was increased to 7.5, fewer bacteria (10∼20% of low pH) were found adherent to the MNPs and extracted by the magnetic force. The same results occurred for pH 10.0.

The two-step theory was well accepted for the understanding bacterial attachment on solid surface (Meinders et al., 1995). The first step involves the bacteria being transported close enough to allow initial attachment, with the forces including van der Waals forces, electrostatic forces, and hydrophobic interactions (Gilbert et al., 1991; Carpentier and Cerf, 1993). The next crucial step is the irreversible attachment process, by the production of exo-polysaccharides or specific ligands, which may complex with the surface (Dunne, 2002). In the transition from reversible to irreversible attachment, various short range forces are involved, including covalent and hydrogen bonding as well as hydrophobic interactions (Kumar and Anand, 1998). Under current experimental conditions, considering the very short Bio-Nano contact time, the reversible initial attachment no doubt dominate the efficiency and selectively of bacterial separation. Thus hydrogen bond, van der Waals and electrostatic forces were considered as the primary interactions between bacterial cells and MNPs according to DLVO theory (Hermansson, 1999).

Apparently, strong hydrogen bonding forms between protonated acid ligands from MNPs and carboxylic or amine groups from the bacteria accounting for the efficient adsorption of all bacteria. However, at higher pH conditions, the electrostatic repulsion between negatively charged carboxylate groups from MNPs and bacteria prevents sufficient biomass adsorption. This mechanism was well supported by surface charge characterization, as shown in Figure 4. Under acid buffering, the citric acid coated Fe₃O₄ MNP has a zeta potential of 3.1, which is close to zero. As we know the zeta potential measurement by the zetasizer from Malvern is usually a low-resolution method with a standard deviation of ± 5 mV. Therefore, the charge of the electrolyte may convert the zeta potential measurement to negative value; however, the overall result should be very close to zero as determined by the protonation of the carboxylate groups from citric acid at pH 2.2. This is why the hydrogen bonding and van der Waals interactions are critical under such conditions. Interestingly, the microbial cells enriched by Fe₃O₄@SiO₂ MNPs at neutral pH had nearly identical microbial composition with Fe₃O₄ MNPs at alkaline conditions (Table 1). In both case, significant negative surface charge was demonstrated by Zeta-potential values (Figure 4). Considering the low recovery efficiencies (Figure 2) of both tests, it is highly likely that only the rare positively charged microorganisms were captured, which again proved that surface charge was the deciding factor for microbial selectivity.

This is the first report that microbial biomass of low concentration in complex matrix could be easily be separated and enriched for downstream analysis, which has significant implications for trace microbial contaminants detection (e.g., airborne pathogen). Currently, qPCR is still the simplest and most widely used method for the quantification of target genes of interest. According to its definition, the method detection limit (MDL) for the qPCR detection of bioaerosols, is determined by the instrumental detection limit (IDL) of qPCR, the DNA extraction efficiency, and the sampling volume (Hospodsky et al., 2010; Tseng et al., 2015). Among these three factors, the IDL (typically 0.5∼1 copy/μL for a SYBR Green dye based detection) and extraction efficiency (approximately 15.5∼43.3% for most commercial kits; Mumy and Findlay, 2004) are the most difficult to improve further. Thus, most studies focus on novel sampling methods with a higher flow rate to lower the MDL of quantification in bioaerosols.

A high-volume TSP sampler is a desirable method because its sampling rate (∼1.0 m³/min) is 100 times higher than that of common organic membrane based filters (Li, 2011). While only up to 10% of the sampled biomass could be extracted by commercial kits, due to the swelling character of glass fibers. Such an actual sampling volume (150 m³ in 20 h) is acceptable for the detection of 16S rRNA genes, which present universally in all bacterial cells with multi copies(Lee et al., 2009).

However, the sampling volume has to be increased when the target gene of quantification is antibiotic resistant genes with concentrations significantly below the level of 16S rRNA genes. Actually, very few studies have been reported ARG contamination in bioaerosols.

Benefited from our novel methods, the actual sampling efficiency was significantly improved, indicated by the enhanced DNA yield shown in Table 3 (Ling et al., 2013; Luhung et al., 2015; McEachran et al., 2015). Approximately 300 and 5000 ng DNA was recovered from clean and polluted air environment, respectively, which was sufficient for the qRT-PCR based 16S rRNA and functional gene quantification.

In Ling’s study, the extracted DNA should be concentrated either by evaporation or ethanol precipitation before qPCR quantification based on their low absolute concentration [∼10 copies/m³ (tetX)] and small sampling volume (∼5.0 m³) (Ling et al., 2013). These concentrating procedures may induce additional errors to the overall measurement because losses are possible from inefficient precipitation and collection. Moreover, an increased PCR inhibition caused by concentrating contaminants in the extract is another important concern. Even in cases of a heavily polluted environment (Just et al., 2012) where the MDL is already satisfied, there are still technical demands for the improvement of the actual sampling rate by our novel method. A 10-folds increase in sampling rate indicates the sampling duration could be significantly shortened. Thus, a more precise time-course dynamic change of microbial compositions could be revealed (Qian et al., 2012).

Apart from the current application in biomass separation from glass fibers, the MNP based methods raised in the current study may have wider applications [e.g., the detection of sand adsorbed pathogens in beach health monitoring (Gu et al., 2003)]. Additionally, there are increasing studies using bimolecular functionalized NPs (e.g., Aptamers (Chen et al., 2008; Vikesland and Wigginton, 2010) or vancomycin (Gu et al., 2003; Neethirajan et al., 2014) functionalized NPs) for the capturing or sensing of pathogenic bacteria. During these detections, the non-specific bacterial adsorption on NPs was always underestimated or even neglected(Vikesland and Wigginton, 2010). According to our results, the buffering condition and surface coating are effective means for the tuning of non-specific adsorption, and have significant implications for the optimization of bacterial affinity materials.