Monitoring COVID-19 Transmission Risks by Quantitative Real-Time PCR Tracing of Droplets in Hospital and Living Environments

Several studies evaluated the presence of SARS-CoV-2 in the environment. Saliva and nasopharyngeal droplets can land on objects and surfaces, creating fomites.

surfaces by qPCR (Table 1). Fomites were found in indoor and outdoor areas exposed to human crowding or frequently touched by hands. Floors and walls were less contaminated than handles or buttons. Droplet DNA traces were detected in the majority of the sampled surfaces, and almost 10% of sampled points displayed multiple contamination from different biological fluids. Correlation between selected bacterial species and biological fluids in droplets and fomites was confirmed (P value , 0.01) ( Table 2), supporting the effectiveness of the approach. The combined action of different markers is synergic (Fig. 2), allowing a reliable identification of droplets and fomites. Indoor and outdoor samples showed the presence of traces from one or more human biofluids, although with different frequencies. qPCR can show not only the presence of biofluids and droplet traces but also the contamination level. Automated nucleic acid extraction and detection of anthropic contamination from environmental swabs could be implemented following the example adopted for the management of human swabs; therefore, detection of fomites by qPCR seems a feasible and promising approach even on a larger scale.
SARS-CoV-2 on surfaces in hospitals and public places is not widely disseminated, unless in proximity of an infected patient. SARS-CoV-2 was not detected in the majority of the indoor and outdoor sampling points, including all 15 sampling points within the air system of a COVID-19 hospital. SARS-CoV-2 RNA was detected only in one room where an infected patient was hospitalized and only in those samples collected near the patient (one on the bed rail and one on the surface of the call button). Furthermore, the stethoscope used for the patient was also positive. The lower (,4%) frequency of positive samples in comparison with other studies (20 to 30%) can be explained by the epidemiology rather than to differences in the sampling strategy or to a lower sensitivity of the method (36); other studies were conducted during the epidemic peak, whereas our study was carried out after the lockdown, when reopening of activities was carefully performed and preventive measures were strictly enforced (7-16, 18, 45-47). Environmental spread of SARS-CoV-2 was not relevant, with the only exception of the surfaces near a hospitalized infected patient.
NGS confirms microbial signature by qPCR. The microbial signature was confirmed by NGS, and all selected bacterial indicators were included within the microbiota identified by high-throughput sequencing and bioinformatic analysis. Both qPCR and NGS showed the prevalent contamination patterns F3-4 ( Fig. 3 and 4). DNA test can be easily performed within 1 day on any real-time apparatus, whereas NGS can be  performed within 1 week, adapting the laboratory protocol to the high-throughput sequencer. Therefore, each selected indicator amplified by qPCR was found by NGS, but only as a subcomponent between others (about 200 to 1,000 species for each sample), including unknown species (;5 to 10%). Mean values ranged from 0.24% (Bacteroides) to 5.78% (Corynebacterium). For example, Corynebacterium showed lower values (,1%) in environmental samples exposed to multiple sources of microbial pollutions, whereas the highest values (35 to 80%) were observed for environmental swabs contaminated by nasopharyngeal secretions. The correspondence between qPCR and NGS was reliable: two independent SARS-CoV-2-positive samples (YH1_01 and YH1_05) of a patient surroundings closely gathered, whereas the sample collected from the stethoscope (YH1_12) segregated at a different distance (Fig. 5). The anthropic contaminations on the right side of the bed (YH1_01) and on the call bottom (YH1_05) showed a similar biodiversity pattern (Shannon index, 2.602 and 2.893, respectively), suggesting a potential contamination performed by the patient's right hand. Conversely, sample YH1_12 displayed microbiota traces associated with the nasopharyngeal secretions and with a different biodiversity pattern (Shannon index, 2.161), suggesting a possible contact of the stethoscope. NGS found environmental bacteria of the microbiota of human biological fluids. Droplets and biofluids are frequently detected in fomites, representing a risk for indirect transmission of SARS-CoV-2. qPCR analysis of fomites allowed the identification of resident or ectopic microflora. NGS showed fomites on several surfaces exposed to anthropic contamination and inhabiting microorganisms or those from other environmental sources. Samples collected from indoor and outdoor surfaces independently grouped if compared with samples containing human biofluids (Fig. 6). Within the outdoor group, those with a higher anthropic contamination overlapped with indoor samples, far from those where the environmental component was overwhelming. Interestingly, only an outdoor sample segregated outside-and betweenboth groups: the external handle of the entrance of a public building (Z_04), characterized by multiple contaminations of anthropic origin overlaying the outside microflora.   Other indoor samples grouped together. These findings highlight the usefulness of microbiota-related data for tracing fomites. However, if fomites can be a risk, the transmission rate is low and depends on the presence of the specific viable pathogens (48). The environmental survival of SARS-CoV-2 depends on indoor and outdoor factors, including sanitation, time of release of the biological fluid, humidity, temperature, air circulation, and sunlight (49)(50)(51)(52)(53). Therefore, detection of fomites should be considered an indicator of transmission risk. Monitoring droplets and biofluids by qPCR can help prevent SARS-CoV-2 transmission suggesting continuous environmental surveillance and sanitization procedures.
Critical issues and limits of the study. The study aim was the assessment of a novel qPCR-based approach and not a detailed risk assessment. Therefore, the comparison between sampling points or buildings cannot be performed, not only because of the small sample size but also for the random collection from areas with a different incidence of infection. Several Italian regions were recruited to avoid a geographical bias; however, the epidemiological burden can affect the inference of the results. Therefore,  we suggested a strategy to assess contamination, which could point out a risk of indirect transmission of SARS-CoV-2. Interestingly, SARS-CoV-2-positive samples were collected only in Emilia-Romagna, Italy, where the incidence of infection was high during the epidemic peak (47). Nevertheless, our findings seem to confirm the key role played by hands. qPCR is faster than NGS (3 to 5 h versus 5 to 10 days) but not rapid compared with other molecular methods (33,54). Tracing droplets and biological fluids by qPCR or NGS has a high specificity (18,19,55,56). qPCR and NGS primers and probes for some bacterial genes were selected because of their feasibility and effectiveness; however, comparisons of sequences, indicators, or reaction conditions were not performed. Whole-genome sequencing would have been more informative, as well as the study of additional S16 rDNA regions (V1 and V2 together with V3, V4, V6, and V7). However, they are expensive (e.g., materials and reagents, bioinformatic analysis) and inappropriate for public health purposes on a larger scale. Finally, we used arbitrary thresholds to quantify droplets contamination based on C T values, recommending the highest sensitivity for droplet detection. A lower threshold could increase the specificity or can be helpful for monitoring transmission routes or sanitation.
Conclusion. Environmental monitoring of fomites can be performed by qPCR, with the identification of anthropic contaminations by detecting their microbiota component.
Droplets and biological fluids were observed in indoor environments exposed to humans and on surfaces frequently touched by hands. SARS-CoV-2 was not detectable, with the only exception of environmental areas near to a patient. Our results indirectly highlight the importance of handwashing and environmental sanitation, as well as the role of health education.
In addition to the environmental detection of SARS-CoV-2, the detection of fomites

MATERIALS AND METHODS
Sampling and experimental design. Surfaces at risk for the presence of biological fluids and the transmission of COVID-19 were sampled from different settings, both in indoor and outdoor areas. Environmental samples (n = 94) were collected during May-June, after the epidemic peak that occurred in Italy in March 2020. We sampled indoor surfaces from three COVID reference hospitals in three Italian regions (Parma, Emilia-Romagna; Sassari, Sardinia; Rome, Lazio), other buildings open to the public (one office, one fast food restaurant, and one church), surfaces in outdoor areas, and used handkerchiefs with nasopharyngeal secretions. All samples were tested for SARS-CoV-2 RNA, whereas anthropic contamination was assessed by searching for biological fluids from nose, mouth, and gut through detection of their microbiota traces by qPCR and/or NGS ( Table 3). As reported in Table 1, the sample reading code used letters and numbers as follows: A, anthropic contamination; Z, outdoor surface; Y, indoor surface; H, hospital; C, church; O, other sampling points in office or restaurant surfaces. All A, Z, and Y samples were collected within the Rome district, while YH1, YH2, and YH3 correspond to the hospitals in Rome, Parma, and Sassari, respectively.
Sampling collection. Surface sampling was carried out after their regular use and prolonged exposure to human presence (.4 h), following standard protocols. FLOGSwabs and CITOSSWAB were used and immediately soaked into a buffer solution of UTM-RM transport medium in a volume of 400 ml (Copan Diagnostics Inc., Murrieta, CA, USA). The nasopharyngeal secretions were collected on handkerchiefs with swabs (4N6FLOQSwabs; Copan Diagnostics Inc., Murrieta, CA, USA). All specimens were refrigerated at 4°C before being tested.
SARS-CoV-2 detection. All samples in UTM were heat inactivated at 56°C for 5 min to reduce the risk of accidental transmission of SARS-CoV-2 to laboratory personnel. Nucleic acids were purified and extracted using the eMag automated nucleic acid sample extraction system (bioMérieux, Marcy l'Etoile, France). Briefly, total nucleic acids were extracted from UTM using an input sample volume of 200 ml into 2,000 ml of easyMag lysis buffer using B protocol to a final eluted volume of purified nucleic acids of 50 ml. TaqPath one-step reverse transcriptase quantitative PCR (RT-qPCR) master mix (Life Technologies, Frederick, MD) and the 2019-nCoV CDC EUA kit (Integrated DNA Technologies, Coralville, IA) were used for target detection (57). Molecular detection of SARS-CoV-2 RNA was carried out by RT-qPCR, using primers and probes related to the E and N genes with a detection limit of 5.2 copies of RNA/reaction (55).  (58). The Detection kit for 2019 Novel Coronavirus (2019-nCoV) RNA (PCR-Fluorescence Probing) was designed for simultaneous identification of N and Orf1ab genes. Afterward, 5-ml portions of eluted RNA samples in a total volume of 25 ml were amplified on Bio-Rad CFX96 real-time system. In each round of extraction and amplification, positive and negative control samples (supplied by the manufacturer) were included. Laboratory controls included serial 10-fold dilutions of an internal positive sample that were analyzed in triplicate (R 2 = 0.99), further supporting the reproducibility and capability to detect SARS-CoV-2 RNA up to the limit of detection of the assay. The interpretation criteria were the following: (i) positive signals detected in E, N, and RdRP ORF1ab and N genes with cycle threshold (C T ) values of #40 were considered positive for SARS-CoV-2 RNA; (ii) positive signals in only one gene (E, N, or RdRP/Orf1ab) with C T values of #40 were considered inconclusive; and (iii) no fluorescent signals or over the 40th C T in RdRP/ORF1ab and N genes were considered not specific and reported as negative for SARS-CoV-2 RNA. The declared limit of detection (LoD) is 500 RNA copies/ml, following the manufacturer's information.
DNA extraction. An aliquot of COPAN UTM-RM transport medium (about 300 ml) was centrifuged at 16,000 Â g for 10 min, and the pellet was manually disaggregated with a pestle after adding glass beads   were added into the NAO Baskets and incubated at 37°C for 30 min. Then, 20 ml proteinase K and 400 ml buffer AL were added, and the sample was centrifuged at 10,000 Â g for 1 min, allowing the elution of the digestion solution. After incubation at 56°C for 10 min and addition of 400 ml ethanol, the washing step and DNA purification were performed in accordance with the manufacturer's instructions. DNA elution was completed in 60 ml elution solution [10 mM tris(hydroxymethyl)aminomethane-hydrochloride and 0.5 mM ethylenediaminetetraacetic acid at pH 9.0], as previously described (32,33). Analysis of mfDNA by multiplex real-time PCR and data interpretation. Amplifications were combined in four multiplex reactions: mix skin, for the identification of Staphylococcus aureus and Staphylococcus epidermidis; mix nasopharynx for Propionibacterium spp. and Corynebacterium spp.; mix oralpharinx for Streptococcus salivarius and Streptococcus mutans; mix feces for Enterococcus spp. and Bacteroides vulgatus (probes were labeled FAM/VIC/HEX, with the BHQ-1 quencher). Primers for different bacterial indicators and optimized reaction conditions were already established as previously described (4,30,59,60,61). Briefly, amplifications were performed in a volume of 25 ml, of which 12.5 ml JumpStart Taq ReadyMix for Quantitative PCR (Sigma-Aldrich, St. Louis, MO), containing 900 nM forward and reverse primers, and 250 nM each probe. For each mix, samples were tested in triplicate. The amplifications were performed using Bio-RadCFX96 (Bio-Rad, Hercules, CA) programmed for 10 min at 95°C and 40 cycles with 1 cycle consisting of 15 s at 95°C and 1 min at 60°C. For each sample, 11 ml template reaction was amplified. The PCR output was expressed as cycle threshold (C T ). Positive samples were those where $1 positive indicator (C T # 35) was found in at least two mixes. Conversely, a microbial indicator was considered negative when it was over the C T $ 39 threshold. In particular, for each biological fluid, at least two bacterial markers were used, and for each indicator, the following criteria were used: 111, positive, C T , 20; 11, positive, C T of 21-30; 1, positive with C T of 31 to 35; 1/2, low-confidence positive with C T of 36 to 38; 2, negative with C T . 39. Due to the use of recombinant and not native polymerase or Escherichia coli the criteria were modified as follows: 11, positive with C T , 20; 1, positive with C T of 21 to 29; 1/2, low-confidence positive with C T of 30 to 35; 2, negative with C T .36; the criteria were modified in order to avoid false-positive data due to the detection of E. coli DNA in the recombinant enzyme used for amplification (62).
16S rDNA amplicon sequencing analysis. Libraries for NGS were prepared according to the 16S Metagenomic Sequencing Library Preparation Guide (part 15044223 rev A; Illumina, San Diego, CA, USA). The PCR amplicons were obtained using Ba27F and Ba338R primers containing overhang adapters, as previously described (33,63). Tagged PCR products were generated using primer pairs with unique barcodes through two-step PCR. In this strategy, target primers containing overhang adapters were used in the first PCR to amplify the target gene, that product was then used in the second PCR using primers-containing barcodes. Each amplification reaction had a total volume of 25 ml, containing 12.5 ml of KAPA HiFi Hot Start Ready Mix (Roche, Pleasanton, CA, USA), 5 ml of each primer (1 mM), and 2 ml of template DNA. Reactions were carried out on a Techne TC-PLUS thermocycler (VWR International, LLC, Radnor, PA, USA). Following amplification, 5 ml of PCR product from each reaction was used for agarose gel (1%) electrophoresis to confirm amplification. The final concentration of cleaned DNA amplicon was determined using the Qubit PicoGreen dsDNA BR assay kit (Invitrogen, Grand Island, NY, USA) and validated on a Bioanalyzer DNA 1000 chip (Agilent, Santa Clara, CA, USA). Libraries were prepared using the MiSeq reagent kit preparation guide (Illumina, San Diego, CA, USA). Raw sequence data were processed using an in-house pipeline that was built on the Galaxy platform and incorporated various software tools to evaluate the quality of the raw sequence data (FASTA/Q Information tools, Mothur). All data sets were rigorously screened to remove low-quality reads (short reads . 200 nucleotides [nt], zero-ambiguous sequences). Demultiplexing was performed to remove PhiX sequences and sort sequences; moreover, to minimize sequencing errors and ensure sequence quality, the reads were trimmed based on the sequence quality score using Btrim (an average quality score of 30 from the ends, and remove reads that are less than 200 bp after end trimming) (64). OTUs (operational taxonomic units) were clustered at a 97% similarity level, final OTUs were generated based on the clustering results, and taxonomic annotations of individual OTUs were based on representative sequences using RDP's 16S Classifier 2.5. Observed OTUs were defined as observed species. A level of 97% sequence identity is often chosen as representative of a species and 95% for a genus (65). The sequence reads were analyzed also in the cloud environment BaseSpace through the 16S Metagenomics app (version 1.0.1; Illumina): the taxonomic database used was the Illumina-curated version (May 2013 release of the Greengenes Consortium Database) (65).
Statistical analysis. Relative abundances of community members were determined with rarefied data and summarized at each taxonomic level. The proportion of the microbiome at each taxonomic rank, such as phylum, order, class, family, and genus, was determined using the RDP classifier and the Greengenes Database. Alpha and beta diversity were calculated using EstimateS software at a level of 97% sequence similarity. Regarding alpha diversity, the Shannon index and equitability index at the species level were computed (66,67). Principal-component analysis (PCA) was performed using the METAGENassist platform and R (version 3.1.3, www.R-project.org) with packages "ggplot2," "ape," "psych," and "vegan" (68). Multivariate analysis, the PCA, and partial least square-discriminant analysis (PLS-DA) were performed in order to investigate the dissimilarity between groups. Feature selection was performed using PLS-DA and 10-fold cross validation to tune algorithm parameters and to check model validity. Dendrogram and clustering analysis were based on the Euclidean distance and Ward's method.
Data availability. The raw sequencing data were deposited in the NCBI Sequence Read Archive under accession number PRJNA685983. The analyses were carried out using an online Galaxy server available at http://huttenhower.sph.harvard.edu/galaxy.