Isolation of H5N6, H7N9 and H9N2 avian influenza A viruses from air sampled at live poultry markets in China, 2014 and 2015

Zoonotic infections by avian influenza viruses occur at the human–poultry interface, but the modes of transmission have not been fully investigated. We assessed the potential for airborne and fomite transmission at live poultry markets in Guangzhou city and in Hong Kong Special Administrative Region (SAR), China, during 2014 and 2015. Viral genome and infectious avian influenza A viruses of H5N6, H7N9, and H9N2 subtypes were detected predominantly from particles larger or equal to 1 μm in diameter in the air sampled with cyclone-based bioaerosol samplers at the live poultry markets in Guangzhou. Influenza A(H9N2) viruses were ubiquitously isolated every month during the study period from air and environmental swabs, and different lineages of H9N2 virus were isolated from markets where chickens and minor land-based poultry were sold. The use of de-feathering devices increased the quantity of virus-laden airborne particles while market closure reduced the amount of such particles. The results highlight the possibility of airborne transmission of avian influenza viruses among poultry or from poultry to humans within such settings. This may explain epidemiological observations in which some patients with H7N9 infection reported being in markets but no direct contact with live poultry or poultry stalls.


Introduction
Influenza A viruses infect a wide range of animal species and are transmitted via virus-laden particles through multiple non-exclusive modes. Interplay between multiple viral, host and environmental factors determine influenza viral transmission efficiency [1][2][3][4][5]. Virus-host compatibility establishes viral tropism and the quantity of virus-laden particles that may be released from infected hosts [1,2]. Gravity limits the distance that virus-laden particles can travel; large droplets settle rapidly and contribute to fomite transmission while droplet nuclei less than 5 μm in diameter may remain suspended in the air and mediate airborne transmission [3,4]. Humidity and temperature may impact on particle size and viability of the virus [5].
Zoonotic infections by avian influenza viruses occur at the human-avian interface [6] and live poultry markets play a critical role in maintaining, amplifying and disseminating avian influenza viruses between poultry species and from poultry to humans [7]. Exposure to live poultry has been reported by many patients with illness due to H5N1 and H7N9 infection, but sometimes such exposure has been indirect, for example visiting a vegetable stall within a large market where live poultry were sold [8]. Thus the modes of transmission are not well defined. The importance of contact or fomite transmission is supported by the detection of avian influenza viruses from various environmental swabs (e.g. counter surfaces, cages, water) at live poultry markets [9,10]. In addition, virus-laden particles that may mediate droplet or airborne transmission could be released from infected birds or as a result of aerosolgenerating procedures during poultry slaughtering at markets. Currently, however, there is no information on the quantity, particle size and viability of virus-laden particles at live poultry markets.
To systematically assess the potential modes of transmission of avian influenza viruses at the human-poultry interface, we conducted monthly air and environmental sampling during July 2014 and October 2015 at three types of live poultry markets in Guangzhou city, Guangdong Province, China, and at one wholesale market in Hong Kong Special Administrative Region (SAR), China. In Hong King SAR, a ban on keeping live poultry overnight at retail live poultry markets has been implemented since 2008 [7].

Methods
Samples were obtained from three different market types in Guangzhou: one wholesale market (two sites), one mixed animal market (two sites) and one retail market (one site). In Hong Kong SAR, we sampled at one wholesale poultry market.
Sampling in the Guangzhou wholesale market and mixed animal market was carried out from July 2014 to October 2015. In the retail market, sampling was conducted from January to October 2015; in the Hong Kong SAR market, sampling was carried out in October and November in 2014 and March, April, July, August, September and October in 2015.

Bioaerosol and environmental sampling at live poultry markets
Two types of cyclone-based bioaerosol samplers were used. The NIOSH bioaerosol sampler (BC251) collects particles based on their aerodynamic diameters into > 4, 1-4, and < 1 µm fractions at a flow rate of 3.5 L per minute [11]. The NIOSH samplers were set 1.2 m above ground and 0.5 m distance from poultry housing; samplers without connection to a vacuum pump were similarly placed as negative controls. After 30 min, a total of 0.105 m 3 air was sampled; 1 mL of minimum essential media with 4% bovine serum albumin was added to each of the collection tubes and polytetrafluorethylene filters and transported on ice packs to the laboratories at Guangdong Provincial Center for Disease Control and Prevention or at the University of Hong Kong.
The Coriolis μ air sampler (referred hereafter to as Coriolis) (Bertin Technologies) collects air at 100-300 L per minute. After 10 min sampling using 300 L per minute, a total of 3.0 m 3 air was sampled into a conical vial containing 5 mL MEM, which was concentrated using the 100 kDa Amicon Ultra-15 (Millipore) to a final volume of 1.5 mL. The sampler was placed 1 m above the ground and 0.5 m distance from poultry housing.
In parallel, environmental swabs were also collected from drinking water, fresh faecal droppings, or surfaces (cages, de-feathering machine and waste bins) at the markets. Temperature and humidity were recorded using a hygro-thermometer (Extech).

Detection and quantification of influenza viral RNA genome
Viral RNA for testing by quantitative real-time reverse transcription polymerase chain reaction (qRRT-PCR) was extracted from 400 µL of the specimen using the QIAGEN EZ Robot or the RNeasy Mini Kit (Qiagen) and eluted into 60 µL H 2 O. Influenza viral RNA was detected using AgPath-ID One-Step RT-PCR Reagents (Life Technologies) with specific primers and probes [12], using 5 µL of the eluted RNA. The number of influenza A virus M gene copies per m 3 air was calculated, where V w is the volume of medium added to the sampler, V r is the volume of specimen used for RNA extraction, U is the airflow rate (m 3 per minute) and t is the sampling time.

Formula 1
M gene copies per cubic metre air = copies per µL × 60 µL × VwVr ÷ (U × t) The minimum linear range of quantification (LoQ) was two copies M gene per µL, and the LoQs were determined as 2,857 and 150 copies/m 3 air for the NIOSH and Coriolis samplers, respectively. Influenza A virus M gene-positive samples were subtyped using H5-, H7or H9-specific primers and probes by qRRT-PCR [9].

Virus isolation in embryonated chicken eggs
All samples with threshold cycle (Ct) values ≤ 35 for influenza A virus M gene by qRRT-PCR were propagated in embryonated chicken eggs by injecting 0.2 mL of specimen into the allantoic cavity and incubated at 37 °C for 48-72 hours. Allantoic fluid that agglutinated chicken or turkey red blood cells were further characterised by qRRT-PCR; samples with increasing copy numbers for influenza viral H5, H7 or H9 gene (reduced Ct values relative to the original field samples) after egg propagation were considered positive by virus isolation.

Genome sequencing and phylogenetic analysis
Viral RNA from an isolated virus was extracted using the RNeasy Mini Kit (Qiagen), amplified by RT-PCR [13] and was subjected to dideoxynucleotide sequencing or next-generation sequencing using the Ion PGM System with PathAmp FluA Reagents (Life Technologies). The sequences were submitted to the Global Initiative on Sharing All Influenza Data (GISAID) [14] (EPI674320, EPI674374 to EPI674424, EPI676397 to EPI676400, EPI676490, EPI676491, EPI696727 and EPI696728). Phylogenetic analysis was performed with the H9 haemagglutinin (HA) coding sequence (1,093 nt, 115-1,207 nt from ATG) aligned with reference strains from GISAID (Table 1). Phylogenetic trees were constructed by maximum likelihood method with bootstrap analysis (n = 1,000) by MEGA (version 6.0).

Statistical analysis
Correlation analyses were done by determining Spearman's rank-correlation coefficients (r s ). Fisher's exact test was applied to assess if the subtypes detected were statistically significantly different. Statistical analyses were performed using Graphpad Prism 6.0.

Ethics statement
Permission from the vendors at the poultry markets was obtained before the bioaerosol and environmental sampling. All sampling was performed without Table 1 Origin of the haemagglutinin sequences of influenza A(H9N2) isolates used for the phylogenetic analysis Segment    directly handling the poultry, thus animal ethics were not applicable for our study.

Sampling at a wholesale market in Guangzhou
The wholesale market was organised into areas for holding live poultry, slaughtering and selling dressed poultry (poultry carcasses). Two sites were sampled. Site A1 was within the live poultry holding area of ca 5,500 m 2 , where 10,000-20,000 poultry (predominantly chickens) were kept at any one time. Chickens were kept on a litter-bedded floor and were often sold to other retail markets within three days. Site A2 was a stall for chicken slaughtering with a de-feathering machine. There was one routinely scheduled market rest day per month; additional rest days may be scheduled in response to reports of human zoonotic infections.
The quantity and subtypes of influenza virus-laden particles detected in the air using a NIOSH bioaerosol sampler, at particle sizes of > 4, 1-4 and < 1 μm diameter are shown. The horizontal dotted lines indicate the linear range of quantification for the influenza A virus M gene by quantitative real-time reverse transcription polymerase chain reaction (qRRT-PCR) assay. Samples in which virus was isolated after egg passage are indicated by an asterisk (*).
H9N2 viruses (five isolates) and mixture of H7N9/H9N2 viruses (one isolate) were further isolated from the air samples collected by the NIOSH sampler at the fraction of > 4 μm, with an isolation rate (number of isolates/ number of PCR-positive samples) of 6/14 (Tables 2 and  3). From the fraction of 1-4 μm, one H9N2 virus was isolated from 11 influenza A virus M gene-positive samples after egg propagation ( We analysed if viral load or environmental conditions might be associated with virus isolation from the air samplers; however, the M gene copy numbers, temperature, and relative humidity were not significantly different between months in which virus was isolated and those in which it was not, using the NIOSH sampler (p = 0.17, 0.07 and 0.72, respectively, Mann-Whitney  test) or the Coriolis sampler (p = 0.86, 0.49 and 0.32, respectively).
In December 2014 and October 2015, neither air sampler detected the influenza A virus M gene. In December 2014, sampling was coincidentally performed on the market rest day (when the market was closed); all chickens were removed from the market but the environment had not yet been disinfected. In October 2015, sampling was performed the day after market closure. These results suggest that market closure may effectively reduce the viral load at the markets for a short time period.

Site A2
We performed air sampling while the de-feathering machine at site A2 was in operation (five samples) or not in use (three samples). While the machine was in operation, influenza A virus M gene was detected by qRRT-PCR from particles > 4 µm in 5/5 samples at 4,157-28,929 copies/ m 3 air (2/5 below LoQ) and from particles 1-4 µm in 2/5 samples (2/2 below LoQ); no viral RNA was detected from particles < 1 µm (0/5 samples). H9 RNA was detected in 4/5 samples and mixed H5/H9 RNA was detected in 1/5 samples from particles > 4 µm; one H9N2 virus was isolated from the air sample.
In contrast, influenza A virus M gene was not detected in air sampled while the de-feathering machine was not in use (0/3 samples). At the same time, environmental swabs collected from the de-feathering machine were consistently positive for the M gene by qRRT-PCR, regardless of whether the machine was in use or not. Overall, the results suggest that infectious influenza A virus-laden particles can be generated during the defeathering process.

Sampling at a mixed animal market in Guangzhou
This mixed animal market sold live poultry, reptiles and mammals, although poultry were kept in a separate area. The predominant poultry species sold were aquatic birds (ducks and geese) and minor land-based poultries (pheasants, guinea fowls, chukar partridges, quails). Each vendor may have a few hundred birds of different species, which were kept in separate cages or pens of various sizes. There was no clear all-in/all-out policy or known routine market rest days.
NIOSH samplers were set up at two separate vendors' stalls (sites B1 and B2). At site B1, influenza A virus M gene was detected by qRRT-PCR from particles > 4 µm in 15/16 samples at 6,179-1,650,000 copies/m 3 air (2/15 below LoQ), from particles 1-4 µm in 9/16 samples at 3,450-210,714 copies/m 3 air (3/9 below LoQ) and from particles < 1 µm in 1/16 samples (1/1 below LoQ) (Figure 2). At site B2, influenza A virus M gene was detected from particles > 4 µm in 15/16 samples at HA: haemagglutinin; qRRT-PCR: quantitative real-time reverse transcription polymerase chain reaction. a A sample with copy numbers of influenza A virus H5, H7, or H9 genes (reduced threshold cycle (Ct) values by qRRT-PCR) higher than those of the original filed sample after egg propagation was considered positive by virus isolation. b The sampling site was located at the poultry holding area within the wholesale live poultry market (see site A1 in the text). c Sites B1 and B2 were two separate vendors' stalls within the mixed animal market.
3,590-204,286 copies/m 3 air (4/15 below LoQ), from particles 1-4 µm in 11/16 samples at 3,050-20,857 copies/m 3 air (6/11 below LoQ) and from particles < 1 µm in 3/16 sampling (3/3 below LoQ) (Figure 2). H9 and mixed H5/H9 RNA were detected from the M gene-positive samples by qRRT-PCR. H9N2 (n = 6) and mixed H9N2/H5N6 (n = 3) viruses were isolated from the fraction of particles > 4 µm, with isolation rates of 5/15 and 4/15 at sites B1 and B2, respectively (Tables 2  and 3). Higher M gene copy numbers (p = 0.01, Mann-Whitney test) and lower relative humidity (p = 0.04) were noted in the months when influenza virus was isolated in air sampled by the NIOSH sampler. Using the Coriolis sampler, influenza A virus M gene was detected from 14/14 samples at 201-29,888 copies/ m 3 air (1/14 below LoQ), which were subsequently confirmed as H9 or mixed H5/H9 subtypes. Six H9N2 viruses were isolated from 14 air samples collected by the Coriolis sampler (Table 2).  The quantity and subtypes of influenza virus-laden particles at particle sizes at > 4, 1-4 and < 1 μm detected in the air using a NIOSH bioaerosol sampler at two sampling sites (sites B1 and B2) within the market are shown. The horizontal dotted lines indicate the linear range of quantification for influenza A virus M gene by quantitative real-time reverse transcription polymerase chain reaction (qRRT-PCR) assay. Samples in which virus was isolated after one passage in embryonated eggs are indicated by an asterisk (*).

Sampling at a retail market in Guangzhou
This retail market had 10 stalls that sold live poultry. Sampling was performed in one stall of 4 m 2 , which held 30-50 birds daily (co-housed chickens, ducks, pigeons, geese and quails). There were no clear all-in/ all-out policy or known regular market rest days for disinfection. Sampling at the retail market was conducted from January to October 2015.

Sampling at a wholesale poultry market in Hong Kong SAR
This wholesale poultry market served as a temporary holding site for chickens imported from mainland China or raised locally. The chickens stayed for no longer than 48 hours until sold to retail markets, with a firstin/ first-out policy, segregation and strict biosecurity measures. Since 2013, chickens imported from mainland China and those raised locally have been housed separately at different locations.  (Table 2). Furthermore, none of the 39 environmental swabs were positive for the influenza A virus M gene (

Genetic analysis of H9N2 viruses isolated from the live poultry markets
The H9N2 virus was the most frequently isolated subtype from the markets in Guangzhou we sampled, with a total of 58 isolates of H9, H9/H7, or H9/H5 subtypes ( Table 2). We performed a phylogenetic analysis of the HA gene of 46 selected H9N2 viruses isolated from the wholesale market (10 air samples, 15 environmental swabs) and the mixed animal market (10 air samples, 11 environmental swabs) in Guangzhou city. The H9N2 viruses isolated from the air and environment from the same market were genetically related. Furthermore, the H9N2 viruses isolated from the wholesale and the mixed animal markets were separately clustered into two clades (Figure 3).

Discussion
Influenza viruses are transmitted via different but nonmutually exclusive modes [4]. Infections are mediated via virus-laden particles of various sizes that confer fomite, droplet or airborne transmission [19][20][21], but the modes of transmission for human zoonotic infections by avian influenza viruses at the human-poultry interface are not well defined. In our study, we determined the quantity, viability, subtype and size of influenza virus-laden particles in the air at three types of live poultry markets in Guangzhou city. Although our study is limited to a small number of markets in Guangzhou city and Hong Kong SAR and the results should be interpreted with caution, we show that viral RNA or viable avian influenza viruses of H5, H7 and H9 subtypes with human zoonotic infection potential are readily detectable in the air, suggesting the feasibility of airborne transmission of avian influenza viruses at the human-poultry interface. Furthermore, human activities, such as operation of de-feathering machines commonly used at live poultry markets in China, may facilitate generation of viable virus-laden particles in the air. In contrast, the negative air sampling results obtained at the wholesale market in Guangzhou on or after market closure day suggest that appropriate interventions may reduce the viral load effectively in the environment. While poultry markets are not common in Europe, the result is consistent with the detection of influenza viral RNA in the air at poultry farms sampled during avian influenza outbreaks in the Netherlands [22]. Our study provides experimental evidence showing that viable avian influenza viruses can be detected in the air where live poultry are kept, which is consistent with previous reports that detected viral RNA and infectious influenza viruses at swine barns or at live pig markets in the United States [23,24]. Although it is difficult to compare our results with those reported previously due to differences in the air samplers used, the concentrations of viral RNA we detected in the air at the live poultry markets were comparable with those detected at the swine barns in the United States in 2011 [24].
Our results suggest that poultry workers in the live poultry markets are constantly exposed to high viral loads in the air and the environment, but human symptomatic infections caused by avian influenza viruses in this population remain uncommon. Excluding the samples collected at the slaughtering area of the Guangzhou wholesale market (site A2) and from the live poultry market in Hong Kong, using the NIOSH bioaerosol sampler, viral RNA or viable virus was identified predominantly from particles > 4 µm (16 viable isolates of 58 samples collected), occasionally from 1-4 µm (1/58), and none from particles < 1 µm (0/58). Previous studies that analysed particle deposition suggest particles < 3 μm are more likely to deposit in the deep lungs [25] where avian influenza viruses with binding specificity for α2,3-linked sialic acids preferentially replicate [26]. In addition, seroepidemiological studies have reported a limited number of cases with low levels of neutralising antibody titres using hemagglutination inhibition assay or neutralisation assay [27][28][29]; however, the mechanism of cross-protection may be via non-neutralising antibodies or T-cell response. Further studies are needed to evaluate the percentage of subclinical infections and to assess the crossprotective adaptive immune response between poultry workers and the general population.
The H9N2 avian influenza virus ubiquitously present among land-based poultry in China and other countries [30] was the predominant subtype detected from the air and environmental samples in our study. Genetically diverse H9N2 viruses have been shown to possess human-like receptor binding specificity [31], transmission potential among ferrets [32] and have provided the internal genes for the H7N9 or H10N8 viruses that have caused fatal human infections since 2013 [33]. Unlike highly pathogenic viruses of H5 subtype that replicate systematically and cause high mortality, the low pathogenic H9N2 and H7N9 viruses generally do not cause apparent clinical signs in infected poultry [30,34]; this poses a challenge in identifying the infected birds for infection control and facilitates the spread of the H9N2 and H7N9 viruses in live poultry markets. H9N2 and H7N9 viruses are known to replicate more efficiently in the respiratory tract than the gastrointestinal tract of the land-based poultry [34,35], and the highly prevalent H9N2 virus has been the dominant subtype detected in the air at the poultry markets, as shown in the present study. Determining the viral loads and subtypes from oropharyngeal and cloacal swabs from different poultry species may help to understand the effect of viral respiratory tropism versus the quantity of virusladen particles released in the air. We also observed segregation of species-adapted H9N2 lineages at different markets; further studies should investigate if the segregation is due to repeated re-introduction of a species-adapted virus as a result of selling different species at different markets or if insufficient cleaning of the environment facilitated the persistence and segregation of the H9N2 virus.
Among the three different types live poultry markets in Guangzhou, we noted higher virus isolation rates from air samples collected at the wholesale market and the mixed poultry market than that of the retail market, suggesting the number of poultry sold on site may affect the quantity of viable virus detected in the air. Cleaning practices, such as the market rest day, may have an impact as well. In addition, we noted a higher detection rate and isolation rate from particle > 4 μm, regardless of the viral subtype, suggesting that there is no correlation between avian influenza A subtype and virus detection at specific particle sizes. Since the subtypes detected in the air correlate well with the subtypes detected from the environment (water, faecal droppings and surfaces), the prevalence of a subtype in poultry (e.g. H9N2) may be a major contributing factor to the subtype detected in the air; however, other factors including viral tropism in poultry should also be considered. Temperature and relative humidity can affect viral viability and the sizes of virus-laden particles in the air. However, we did not observe a strong impact of temperature and humidity on viral detection at specific particle sizes; a longer observation period and/or frequent sampling will be needed to address this question.
Taken together, our results indicate the possibility of airborne transmission for avian influenza A viruses and may explain some human cases who appear to have acquired H7N9 infection by visiting live poultry markets but without direct or indirect contact to poultry [8]. Furthermore, the observation that known zoonotic infections have been in people with transient contact with, or passing the vicinity of live poultry marketsrather than those working within them, who are clearly exposed to avian influenza viruses on almost a daily basis -suggests a role for host susceptibility as one of the key determinants of zoonotic infection.