Introduction Transmission of influenza virus between humans may occur by three routes: (1) direct or indirect contact between an infected and a susceptible person, usually resulting in contamination of a susceptible person's hands followed by hand to respiratory mucosa contact; (2) large droplet spray wherein droplets of respiratory fluid greater than approximately 100 µm in diameter are expelled with sufficient momentum to deliver a direct hit on the respiratory mucosa; and (3) aerosols generated by release of smaller, virus-containing droplets, as may occur during tidal breathing and coughing [1], [2], that rapidly evaporate into residual particles (droplet nuclei),which are inhaled and deposited in the respiratory tract [3]–[6]. There is significant evidence for each of these routes [7], [8], but their relative importance is not known [3]. As a result, the Institute of Medicine recommended that healthcare workers in contact with 2009-H1N1 patients use protection against all of the possible routes of infection, including use of fit-tested N95 respirators [3]. A year after the 2009 pandemic, there was no greater clarity on the importance of the various modes of transmission [9]. The U.S. Centers for Disease Control and Prevention recently funded an experimental study of person-to-person transmission to address this important knowledge gap [10]. However, an experimental study using intranasal inoculation to infect experimental donors [11] will need to show that the donors and naturally infected persons shed similar virus aerosols with regard to quantity, particle size distribution, and infectiousness, given that earlier experiments suggested that intranasal inoculation requires quantitatively larger doses and produces qualitatively milder illness than does inoculation via aerosol [12]. In an occupational hygiene context, personal protection is usually the last resort, after source mitigation and environmental controls are exhausted [13]. Thus, it is worthwhile considering whether surgical facemasks could be effective as a means of source control. The CDC recommends that persons with influenza wear surgical masks when in contact with susceptible individuals [14], [15]. However, there is only one report studying mask impact on containment of infectious large droplet spray during influenza infection [16], and no data on surgical mask impact on release of infectious viral aerosols. In the current study of patients infected with seasonal influenza, we describe the number of copies of viral RNA in two aerosol size fractions, report the culturability of virus in the fine-particle fraction, and the effect of surgical masks. Results We screened 89 volunteers: 33 (37%) tested positive for influenza using the rapid test (20 influenza A and 13 influenza B) and were asked to provide exhaled breath samples. Eight additional volunteers with negative rapid tests who reported a cough and who had a temperature of ≥37.8°C were also invited to participate. In total, 38 volunteers were confirmed to have influenza virus infection by PCR of nasopharyngeal specimens. Exhaled breath data with and without a surgical mask are complete for 37 of the 38 volunteers (21 influenza A, 16 influenza B); data for one volunteer has been excluded due to laboratory error in sample processing. One of the infected subjects reported receiving influenza vaccine for the current year. None of the subjects sneezed during the sample collection. Table 1 shows the sex, symptom and fever prevalence, and influenza virus type and Table 2 shows descriptive statistics for age and viral RNA copy number in swabs and exhaled aerosol fractions of the 37 volunteers with confirmed influenza infection. The viral copy numbers in each of the five specimens for all 37 cases are shown in Table S1. 10.1371/journal.ppat.1003205.t001 Table 1 Participant's sex, symptoms, temperature, and influenza virus type. N Percent Number with complete data 37 100 Male 30 81 On antiviral medicinea 0 0 Asthmatica 5 14 Flu shot this seasona 1 3 Flu shot previous seasonsa 12 32 Current smokera 9 24 Tachypneaa 13 35 Breathing difficultya 16 43 Lymphadenopathya 18 49 Feverisha 19 51 Temperatureb ≥37.8°C 10 27 Type A 21 57 a Self-reported. b At time of exhaled breath measurement. 10.1371/journal.ppat.1003205.t002 Table 2 Descriptive statistics. Percentiles Min 25th Median 75th Max Age 18 18 19 20 54 Days since onseta 0 1 2 3 5 Nasopharyngeal swab copy number 1.7×103 8.3×104 4.2×105 1.8×106 3.4×107 Coarse particle copy number with mask 0 0 0 0 7.7×101 Coarse particle copy number no mask 0 0 0 3.7×101 2.9×104 Fine particle copy number with mask 0 5 2.2×101 2.5×102 2.4×104 Fine particle copy number no mask 0 1.1×101 1.1×102 5.6×102 1.3×105 a At time of exhaled breath measurement. We detected influenza virus RNA in the coarse fraction (particles greater than 5 µm) collected from 11% (4 of 37 volunteers) while wearing surgical masks and from 43% (16 of 37) while not wearing a mask (relative risk for virus detection with mask = 0.25, 95% confidence interval (CI) 0.09 to 0.67; McNemar's test p = 0.003). The median number of coarse fraction viral copies (Figure 1) was below the limit of detection with and without facemasks; the 75th percentile dropped from 37 to below the limit of detection with use of surgical masks. Using Tobit analysis, we estimated that the geometric mean coarse fraction copy number without a facemask was 12 (95% confidence interval (CI), 4 to 37) and that the effect of facemasks was to produce a statistically significant 25 fold reduction in the copy number (95% CI 3.5 to 180, p = 0.002) to 50 µm) that we would not have detected. Furthermore, none of our subjects sneezed; an efficient method of generating droplets from the upper respiratory tract. This may imply that the smaller droplets we detected were generated in the lower respiratory tract and that the viral load at that location is not strongly correlated with the nasopharyngeal load. Alternatively, shedding into aerosol droplets may be driven by other host factors (e.g. asthma, symptom severity, and immune response), co-infection with other agents, virus factors affecting release from the epithelium, or the nature of the resident microbiome. If shedding into aerosol is determined in large part by the location of infection in the respiratory tract, this may have implications for experimental studies of transmission [11], [28]. Such studies will need to monitor aerosol shedding to determine whether nasal inoculation of donors results in aerosol shedding that mimics naturally acquired infection to validate the experimental design and aid the interpretation of results. Most of the viral aerosol generation we observed occurred during the first days of symptomatic illness (Table 3), consistent with studies of shedding monitored by nasal washes [29]. We studied each individual on only one occasion and, by design, have little data beyond day 3. Further longitudinal studies of viral aerosol generation are needed to confirm these findings. New study designs will be needed to examine aerosol generation before and on the day of symptom onset in community acquired infection. A limitation of our study is that we recruited patients with certain signs and symptoms or who were positive on a rapid test or had fever, and therefore our data could be biased towards patients with higher viral loads [21]. However, we still observed significant inter-individual variation and modeling suggests that cases with higher viral loads are disproportionately important in the spread of influenza [30], [31]. Additional studies are also needed to determine how aerosol generation correlates with symptoms (including milder disease), presence of other health conditions, age (we studied a narrow age distribution), and co-infection with other respiratory viruses so that recommendations for infection control can be critically evaluated. Methods Patient population We recruited volunteers with influenza-like illness from the Lowell, MA community, primarily among students and staff of the University of Massachusetts, beginning January 29 and ending March 12, 2009. The study protocol was approved by the Institutional Review Boards of the University of Massachusetts Lowell, Lowell General Hospital, and Saints Memorial Hospital, Lowell, MA. Oral informed consent was obtained by providing each subject with a detailed consent information form. Collection of a signed copy of the form was waived because it would have been the only personally identifiable information retained by this minimal risk study. Volunteers learned of the study through flyers and notices posted on campus and by referral from health care providers. We screened self-referred volunteers by telephone for influenza-like illness (ILI). Persons who reported onset of fever and cough within the preceding 72 hours or were referred by a health care provider were invited to the laboratory for testing. We collected a nasopharyngeal specimen using a flocked swab (501CS01, Copan Diagnostics, Murrieta, CA) and temperature was taken with a digital ear thermometer (Model 18-200-000, Mabis Healthcare, Waukegan, IL). All volunteers with a temperature ≥37.8°C and a cough and volunteers without fever who provided a nasopharyngeal specimen positive for influenza by point of care testing (QuikVue Influenza A/B, Quidel Corp., San Diego, CA) were invited to provide exhaled breath samples, answer a questionnaire, and provide a second nasopharyngeal specimen for analysis by PCR. Only subjects with influenza infection confirmed by PCR were included in the data analysis. Exhaled breath collection We collected exhaled breath with the subject seated in front of the inlet for a sampler designed for human exhaled breath collection, Figure 2, (G-II) described in detail by McDevitt et al. [27] Briefly, the G-II inlet was cone shaped so that the subject's face was situated inside the large end of an open cone with air drawn continuously around the subject and into the sampler. The cone allows the subject to breathe normally and unlike use of a mouthpiece, the subject could also wear a mask. The cone served as a capture type ventilation hood allowing collection of exhaled breath with minimal fugitive emissions even when the subject was wearing a mask with resultant redirection of flow. Intake air (130 L/min) flowed through a conventional slit impactor that collected particles larger than 5 µm on a Teflon surface (“coarse” particle fraction). To collect a “fine” particle fraction, water vapor was condensed on the remaining particles, which created droplets large enough to be captured by a 1.0-µm slit impactor. The 1.0-µm impactor was composed of a slit and a steel impaction surface sealed inside a large reservoir. Impacted droplets drained from the impaction surface into a buffer-containing liquid in the bottom of the reservoir. Concentrated buffer was pumped into the reservoir during collection to match the accumulation of water from collected droplets and maintain phosphate buffered saline with 0.1% bovine serum albumin throughout collection. The sampler was shown to be 85% efficient for particles greater than 50 nm in diameter and was comparable to the SKC BioSampler for detection and recovery of influenza A/PR/8/34 H1N1 by PCR and culture. Between subjects, the apparatus was disassembled and cleaned with a 0.5% hypochlorite solution. 10.1371/journal.ppat.1003205.g002 Figure 2 Exhaled breath collection system. Each volunteer sat as shown with face inside the inlet cone of the human exhaled breath air sampler inside a booth supplied with HEPA filtered, humidified air for 30 min while wearing an ear-loop surgical mask. Three times during the 30 min each subject was asked to cough 10 times. After investigators changed the collection media, the volunteer sat in the cone again, without wearing a surgical mask, for another 30 min with coughing as before. Exhaled particles were collected for 30 minutes while the subject wore an ear-loop surgical mask (Kimberly-Clark, Roswell, GA) and then for 30 minutes without a mask. Subjects were asked to cough 10 times at approximately 10-minute intervals for a total of 30 coughs during each 30 minute sample. One subject coughed frequently such that forced coughs were not required. No subjects were observed to sneeze. Sample analysis Immediately after collection, the Teflon impaction surface was removed and temporarily stored at −20°C. The impactors were scraped with a flocked swab wetted with Dulbecco's phosphate buffered saline with calcium and magnesium (Hyclone, Thermo Scientific, Waltham, MA) with 0.1% bovine serum albumin (DPBS++BSA). The swab was eluted in 600 µl of DPBS++BSA for 1 minute with vortexing. The resulting sample was stored at −80°C. The fine particle fraction collected in DPBS++BSA buffer (100 to 150 ml volume) was maintained at 4°C and concentrated by ultrafiltration using Amicon Ultra 15 filter units with a molecular weight cut off of 100 kD (Millipore, Bedford, MA) to a volume of approximately 400 µl. Following ultrafiltration, the filter was washed with 200 µl of DPBS++BSA, and the wash solution was combined with the retentate. Samples were stored at −80°C. RNA extraction in Trizol-chloroform, reverse transcription, and quantitative PCR were performed as previously described [1], [32]. Quantitative PCR was performed using an Applied Biosystems Prism 7300 detection system (Foster City, CA) for coarse fraction samples or a LightCycler 480 (Roche, Indianapolis, IN) for the fine particle fraction. Duplicate samples were analyzed using influenza A and B primers described by van Elden et al. [33] A standard curve was constructed in each assay with cDNA extracted from a stock of influenza A (A/Puerto Rico/8/1934, Advanced Biotechnologies Incorporated, Columbia, MD) with a concentration of 3.0×1011 virus particles per mL or a stock of influenza B (B/Lee/1940, Advanced Biotechnologies Incorporated, Columbia, MD) with a concentration of 8.6×1010 virus particles per mL as determined by electron microscopy. Results are expressed as the total number of virus particles by reference to the standard curve, rounded to the closest integer value. The limits of detection were 6 and 11 viral RNA copies per qPCR well for influenza A and B respectively. Fine particle samples from all subjects were cultured for infectious virus on MDCK cells. Confluent cells in 24-well plates (Corning, NY, USA)were inoculated with 0.1 ml of the concentrated sample diluted 1∶1 in OptiMEM® I medium (Invitrogen, Carlsbad, California). The plates were incubated at 37°C for 1 h with rocking every 15 min, and 0.8 ml of OptiMEM® I media with 1 µg/ml of TPCK-trypsin was added to each well and incubated for 72–96 h. The cells were checked daily for cytopathic effect (CPE) and if none was detected, two blind passages were performed using cell supernatant. At each passage, supernatants were tested for influenza virus by hemagglutination (HA) assay using 0.5% chicken red blood cells. Positive samples were confirmed by Flu DETECT (Synbiotics, CA, USA) strip test and by amplification of the hemagglutination (HA) gene by RT-PCR followed by sequencing. Statistical analysis We analyzed the effect of surgical masks as a) log relative risk for production of any virus aerosols assuming a binomial distribution using generalized estimating equations with exchangeable within-subject correlation to account for repeated measures, and b) the geometric mean counts of virus particles detected in exhaled breath by qPCR and fractional reduction in copy number using Tobit regression analysis on log copy number with a random effect to account for variability between individuals. Tobit analysis was also used to compare coarse and fine particle fractions. Tobit regression avoids bias that would arise from assigning samples below the limit of detection a specific value such as zero or the limit divided by the square root of 2. Surgical mask use was the dependent variable. We also computed McNemar's test for paired samples to examine mask effect and Spearman's correlation coefficient to examine the relationship between the load in the nasopharyngeal swab and aerosol fractions. Statistical analyses were performed using SAS (Procs GenMod, NLMixed, Lifereg, Freq, Corr, and Means, version 9.2, Cary, NC). Supporting Information Table S1 Copy number and influenza type in five assayed samples per subject. (DOCX) Click here for additional data file.
