Scavenging Ducks and Transmission of Highly Pathogenic Avian Influenza, Java, Indonesia

These ducks may be a source of infection for chickens and humans.

S ince 1997, when highly pathogenic avian infl uenza (HPAI) subtype H5N1 outbreaks occurred in poultry in Hong Kong, People's Republic of China (1)(2), the virus has caused epidemics in Asia, Europe, and Africa (3). In Indonesia, the fi rst HPAI (H5N1) virus infections in poultry were offi cially announced in early 2004 (4); human cases have been reported since mid-2005 (5). Although extensive HPAI control efforts helped reduce the frequency of outbreaks in poultry (6), by 2009, subtype H5N1 virus had been detected in 31 of Indonesia's 33 provinces (7).
In 2009, Indonesia had the highest incidence worldwide of human infections and deaths (8).
Waterfowl are the natural reservoir of avian infl uenza viruses (9), and experimental research indicates that ducks may play a role in the maintenance of HPAI (H5N1) viruses. Infected ducks may exhibit no clinical signs yet can excrete high concentrations of virus that are pathogenic to other poultry species (10)(11)(12)(13). Possible risk factors for HPAI spread in Indonesia include duck movements, contacts between ducks and other poultry and animal species, poor poultry husbandry, inadequate handling of sick and dead ducks by fl ock owners, and poor awareness of control strategies among poultry farmers (14). However, no analytical study assessing risk factors for HPAI infection has been conducted in Indonesia.
In 2005, Indonesia's duck population was ≈34.3 million, of which 40% were on the island of Java, mainly on smallholder farms, i.e., backyard and small commercial farms (14). As in many other Asian countries, domestic ducks on smallholder farms in Indonesia are allowed to scavenge freely during the day around houses, in the villages, or in rice paddies; duck owners supply little or no feed (15). To assess the hypothesis that ducks contribute to the maintenance and transmission of avian infl uenza (H5N1) viruses, we conducted a longitudinal investigation describing temporal patterns of antibodies against HPAI (H5) and virus prevalence in unvaccinated scavenging ducks and chickens that have contact with these ducks (in-contact chickens) in Java, Indonesia.

Study Design
Ducks and in-contact chickens on 96 smallholder duck farms in 4 districts of Central Java were monitored once ev-Scavenging Ducks and Transmission of Highly Pathogenic Avian Infl uenza, Java, Indonesia ery 2 months over 13 months. Four districts were selected (Sleman, Magelang, Bantul, and Kulon Progo) because of their high abundance of duck farms and proximity to the Disease Investigation Center (DIC) in Wates, where fi eld investigators were based and diagnostic work was conducted ( Figure 1). Sample size calculations were based on DIC surveillance data collected in Central Java in 2006; 13 (4.7%) of 278 cloacal swabs from ducks were positive for H5 viral RNA on real-time reverse transcription-PCR (RT-PCR). On the basis of an expected true bird-level virus prevalence of 5%, a precision of the estimate of ±1.5 % and a 95% confi dence interval (CI), a total of 811 ducks had to be sampled (16). We enrolled 96 duck farms in the study and sampled a total of 960 ducks (10 ducks per farm) and 480 in-contact chickens (5 chickens per farm) during each of 7 visits over 13 months (initial visit plus 6 bimonthly visits). We used a multistage sampling strategy with stratification by district and a 3-level sampling process that involved villages, duck farms, and birds. A sampling frame was prepared by agriculture extension offi cers who listed all villages in the selected districts, including the total number of duck farms within each village. We selected 4 villages within each district using probability proportional to size sampling. Field veterinarians then prepared a second sampling frame containing the names of all duck farmers within the 16 villages selected and the number of ducks kept by each farmer. From this sampling frame, 6 duck farms per village were selected by using simple random sampling. Farms with <10 ducks were excluded (we wanted to sample 10 ducks per farm) as were farms with >700 ducks (which we considered to be large commercial farms). Random numbers for village and duck farm selection procedures were produced in STATA version 10.0 (StataCorp, College Station, TX, USA).

Data Collection and Diagnostic Tests
Four veterinarians from the DIC were trained in the use of data collection tools and interviewing techniques. Field visits were conducted once every 2 months from March 2007 through March 2008; duck owners were interviewed and swab and blood samples from birds were obtained during each visit. On the fi rst visit, birds were selected for the study. The duck owner enclosed all ducks in a pen and selected the fi rst 10 ducks that could be caught. If available, 5 chickens kept on the same farm were also selected in the same manner. Wing tags or leg bands were attached to each selected duck and chicken. Blood samples were collected from the wing vein of each bird, and an oropharyngeal swab and a cloacal swab were collected from each bird and placed into a single tube containing virus transport media (Universal Viral Transport 3mL; Becton Dickinson, Franklin Lakes, NJ, USA). Duck owners confi rmed that none of the ducks and chickens sampled had been vaccinated against HPAI before the study and that none were vaccinated during the study.
Serum samples were tested for antibodies to avian infl uenza (H5) by using the hemagglutination inhibition (HI) test according to methods recommended by the World Organisation for Animal Health (OIE) (17). Antigen and control antiserum were supplied from Pusat Vetenerinaria Farma (Surabaya, Indonesia). The antigen was derived from an HPAI (H5N1) chicken isolate obtained in 2004 in Indonesia (A/chicken/Pare/East Java/2004). This antigen is commonly used for HI tests to detect antibodies to avian infl uenza (H5N1) at all veterinary diagnostic laboratories in Indonesia. A titer >2 4 against 4 hemagglutinating units of antigen was considered positive (17). In accordance with the Australian Animal Health Laboratory protocol (18)(19), RT-PCR was used to test the combined oropharyngeal and cloacal swabs of individual birds in pools of 5 for subtype H5 virus RNA. Sequencing was conducted on the H5 RT-PCR-positive samples to confi rm the HPAI multiple basic amino acid motif at the cleavage site of the hemagglutinin gene and to determine whether the neuraminidase gene of the isolate belonged to the N1 subtype.

Investigations of Bird Deaths
Duck farmers involved in the study were asked to immediately report sickness or deaths of birds to the DIC. Compensation was paid to duck farmers to encourage reporting. Upon notifi cation, veterinarians conducted an outbreak investigation at the reported farm by using a predesigned questionnaire. Clinical signs were recorded, and carcasses were collected for postmortem examination. Blood and swab samples from clinically normal birds from the same farm were obtained on the day of the investigation. Blood samples were tested for antibodies to avian infl uenza (H5) as already discussed; swab samples from carcasses (combining lung, heart, liver, spleen, pancreas, and intestinal tissues) and from live birds were processed by virus isolation in embryonated eggs. Two passages of virus isolation were conducted, and allantoic fl uid was tested for H5 antigen of avian infl uenza by using the HI test. An HPAI outbreak was defi ned as >1 bird dying within a few days of each other from HPAI (i.e., positive by subtype H5 virus isolation or RT-PCR).

Data Analyses
For both ducks and in-contact chickens, bird-level seroprevalence (proportion of study birds with antibodies to avian infl uenza [H5]) and fl ock-level seroprevalence (in which at least 1 study bird had antibodies) were calculated for each of the 7 sampling periods and pooled across the entire study period. Virus prevalence was calculated only at fl ock level (proportion of fl ock visits in which at least 1 pool of swab samples from the farm was positive for H5 RNA) for the entire study period. We accounted for the multistage sampling strategy in the data analyses by using survey commands in STATA version 10.0 (StataCorp); districts were treated as strata; villages were specifi ed as primary, and farms as secondary, sampling units. For bimonthly bird-level prevalences, and for bird-and fl ocklevel prevalences over the entire study period, sampling weights were the inverse of the product of the proportion of villages in the district that were sampled and the proportion of duck farms in the village that were sampled (20). The fi nite population correction factor for primary sampling units was the total number of villages in the district. Finite population correction accounted for reduction in variance associated with sampling without replacement (21). For bird-level seroprevalence calculations over the entire study period, we accounted for repeated measurements within the same birds by specifying the individual bird as the third level of sampling and incorporating the number of duck farms per village as the fi nite population correction factor for secondary sampling units. For the bimonthly fl ock-level seroprevalence, only primary sampling units with their fi nite population correction factor were specifi ed in the analyses. Sampling weights for bimonthly fl ock-level seroprevalence were the inverse of the proportion of villages in the sampled district.
We used logistic regression models accounting for 3 levels of clustering (birds within farms within villages) to compare the odds of birds having titers positive for avian infl uenza (H5) between ducks and in-contact chickens. For fl ock-level comparisons, logistic regression models accounting for 2 levels of clustering (farms within villages) were used to compare the odds of fl ocks having at least 1 bird with antibodies to avian infl uenza (H5) between duck and in-contact chicken fl ocks, and for duck fl ocks between sampling months. Logistic regression accounting for 2 levels of clustering was also used to evaluate whether the odds of duck (or chicken) fl ocks being seropositive were independent of the results of the other species at the same farm and sampling. All logistic regression models also accounted for sampling weights and incorporated fi nite population correction. Adjusted Wald tests were used to assess the overall effect of sampling month. After fi tting the logistic regression models taking the survey sampling design into account, we applied the F-adjusted mean residual goodness-of-fi t test (22).

Results
From March 2007 through March 2008, a total of 8,993 serum and swab samples were collected from 6,705 clinically healthy ducks and 2,288 chickens during 670 farm visits (at 2 farm visits, all birds had been sold). During ≈80% of farm visits, chickens were present. Flock sizes for ducks and chickens averaged 53.7 and 8.5, respectively. Of all combined oropharyngeal and cloacal swab sets from individual birds, 8,900 were analyzed in pools of 5 by RT-PCR, and all serum samples were tested for antibodies to subtype H5 virus. In addition, during outbreak investigations, 174 sets of swabs from dead birds and 136 from apparently healthy live birds were collected from the outbreak farms.

Prevalence of Antibodies to Avian Infl uenza (H5)
Bird-level seroprevalences of subtype H5 antibody titers >2 4 in clinically healthy birds for all bird samplings pooled over the entire study period were 2.6% (95% CI 1.8-3.5) for ducks and 0.5% (95% CI 0.0-0.9) for in-contact chickens. The odds of ducks being positive for avian infl uenza (H5) were 5.5× (95% CI 2.1-14.4) higher than for in-contact chickens. Flock-level seroprevalence of antibodies to avian infl uenza (H5) was 19.5% (95% CI 14.3-24.6) for ducks and 2.0% (95% CI 0.1-3.9) for in-contact chickens. The odds of duck fl ocks being seropositive were 12.4× (95% CI 3.9-40.1) higher than those for chicken fl ocks. At 21.4% of 501 farm visits, >1 study duck was seropositive for infl uenza (H5) when during the same farm visits, all in-contact study chickens on these farms were seronegative (Table). Conversely, at only 1.4% of farm visits was >1 study chicken seropositive for avian infl uenza (H5), while all study ducks on the farm were seronegative. At fl ock level, seropositivity of ducks was not associated with seropositivity of chickens on the same farm (OR = 3.9, 95% CI 0.4-43.0). The goodness-of-fi t statistics calculated after fi tting the survey design-adjusted logistic regression models provided no evidence of lack of fi t of any of the models (p>0.05).

HPAI (H5) Virus Prevalence in Clinically Healthy Birds
Birds on 25 (26%) of the 96 monitored farms tested positive for avian infl uenza (H5) virus RNA; on 20 farms, birds tested positive on 1 sampling occasion and, on 5 farms, on 2 different sampling occasions. On these 25 farms, 30 fl ocks (22 duck and 8 chicken fl ocks) tested positive for subtype H5 virus RNA. On 3 farms, both duck and chicken fl ocks tested positive for subtype H5 virus RNA at the same visit (6 fl ocks); otherwise, only 1 fl ock (either ducks or chickens) was positive for subtype H5 virus RNA at any 1 visit (19 duck and 5 chicken fl ocks). The fl ock prevalence of subtype H5 virus RNA (proportion of fl ock-visits during which at least 1 study bird was positive) in clinically healthy birds for all fl ock samplings pooled over the entire study period was 2.5% (95% CI 0.9-4.1) for ducks and 1.5% (95% CI 0.4-2.7) for chickens.

HPAI Outbreaks
Of 96 the study farms, 34 (35%) across all 4 districts had HPAI outbreaks during the study period (Figure 3 On 16 of the 34 farms that experienced HPAI outbreaks, combined swab samples were collected from 136 clinically normal birds (109 ducks, 27 chickens) at the same time as samples from dead birds on the same farm. For 11 (69%) of 16 outbreak farms, HPAI (H5) virus was isolated from 37 (27%) of 136 clinically normal birds (28 ducks, 9 chickens).
Carcasses of 180 marked and unmarked birds (59 ducks, 121 chickens) were obtained; HPAI (H5N1) virus was isolated from 65 (10 ducks, 55 chickens). Another 14 birds (3 ducks, 11 chickens) had most likely died from HPAI (H5) infection; we based this determination on 1) sudden death with or without clinical signs of HPAI (such as lethargy; swelling or discoloration of combs, wattles, and legs; nasal discharge; coughing and sneezing; diarrhea; and lack of coordination) and 2) isolation of HPAI (H5) virus from other dead birds within the fl ock at the same time. Therefore, a total of 79 (44%) of 180 birds most likely died from HPAI infection.

Discussion
Scavenging duck farming has been proposed as an important contributor to HPAI in poultry fl ocks in Southeast Asia, predominantly on the basis of fi ndings obtained through spatial analyses of national surveillance data of HPAI outbreaks (23)(24)(25). However, no fi eld studies have investigated infection patterns over time in duck farming systems. Results of the current study indicate that scavenging ducks are a source of infection for other poultry and, possibly, for humans.
One explanation for the higher seroprevalence in ducks than in chickens is that HPAI (H5N1) virus circulated more successfully among ducks than among in-contact chickens; hence, ducks were more likely to harbor and transmit the virus. This could because of a higher risk for death among infected chickens, resulting in fewer surviving chickens with H5 antibodies, or to differences in scavenging behavior between ducks and chickens. Virulence of HPAI (H5N1) virus for ducks varies from inconsequential to highly lethal (10,26), and some of the 2003-04 Asian-lineage subtype H5N1 viruses can be shed by domestic ducks for up to 17  days postinfection (11,12). When viruses harbored by ducks are transmitted to gallinaceous species, such as chickens, severe clinical signs and high death rates can occur (10). However, for high-incidence HPAI outbreaks in chickens, virus excretion from infected ducks must be combined with an effi cient reproductive number (R 0 ) to produce secondary cases in a susceptible chicken population. R 0 is infl uenced by the infectiousness of the agent causing the disease, the probability of transmission (determined by factors such as housing, mixing, and feeding practices), and the level of population immunity. During the 2004 HPAI (H5N1) epidemic in Thailand, R 0 estimates were lower for backyard chickens than for broilers and layers (27). Birds of the latter 2 groups are typically housed together; but backyard chickens usually have less contact with each other. On our study farms, ducks usually grazed together, behavior conducive to virus circulation between ducks; individual incontact chickens scavenged more independently. Another possible explanation for the difference in seroprevalences between poultry species is that duck fl ocks were exposed to HPAI more frequently than were chickens. Duck fl ocks may graze in the same rice fi elds where other potentially infected domestic or wild birds may have grazed. The higher fl ock-level seroprevalence in ducks than in chickens was probably not biased substantially on differences in the numbers of birds sampled (10 ducks, 5 chickens) at each farm visit. In fl ocks where no study birds were detected with avian infl uenza (H5) virus or antibody, the virus may have been in other ducks or chickens in the same fl ock, and therefore our fl ock-level prevalence estimates underestimated the true fl ock prevalences. Because we sampled more ducks than chickens in each study fl ock, the risk for nondetection of infection was higher for chickens. However, on the basis of sample size calculations for assessing freedom from disease, this bias is unlikely to explain the differences in fl ock prevalences of antibodies between ducks and chickens (28). For example, if the true se-roprevalence was 3% in populations of 100 ducks and 100 chickens, the probability of detecting at least 1 seropositive bird from 10 sampled ducks is 0.27 and, from 5 sampled chickens, 0.16. This equates to an OR of 1.9, which is substantially lower than the observed difference in fl ock-level prevalences in which the odds of duck fl ocks being seropositive were 12.4× higher than that of chicken fl ocks.
HI tests in which horse erythrocytes were used to detect avian infl uenza antibodies in human serum were more sensitive than HI tests in which chicken erythrocytes were used (29). Because OIE does not recommend the horse erythrocyte method for HI tests on poultry serum samples, it is rarely used in poultry diagnostics, although some evidence supports a higher sensitivity in these species (30). We compared HI tests based on horse and chicken erythrocytes by using serum samples from 60 ducks experimentally infected with 2 of the HPAI (H5N1) fi eld isolates from this study. These tests showed substantial agreement when results were categorized as positive (>2 4 ) or negative (κ = 0.74, 95% CI 0.57-0.90), although some samples tested with horse erthrocytes had higher titers than when tested with chicken erythrocytes. Thus, seroprevalence estimates may have been similar to those reported here had the HI tests been conducted by using horse erthrocytes. We suggest that OIE review this issue and, if warranted, modify the recommended diagnostic methods.
Virus shedding was reported in apparently clinically healthy birds on nearly 11/16 outbreak farms. Despite a high risk for death in chickens and some deaths in ducks, other birds carrying the virus appeared to be unaffected, which indicates that host-specifi c characteristics of susceptibility might have varied among birds. Alternatively, some of these virus-positive clinically normal birds could have been sampled early in infection and had not yet developed clinical signs. However, a small number of chickens in the longitudinal study had antibodies to avian infl uenza (H5), providing further evidence that some chickens survive infection. These birds could have been infected with low pathogenicity avian infl uenza (LPAI) viruses. To our knowledge, the prevalence of LPAI in poultry in Indonesia is unknown. However, infl uenza (H5N1) viruses isolated from dead and live birds in our study were confi rmed to be highly pathogenic. Further molecular characterization of these isolates is under way (31).
The frequency of HPAI outbreaks varied throughout the study period. Outbreaks increased in July 2007 (the beginning of the dry season), coinciding with an increased proportion of fl ocks with seropositive ducks. This increase suggests that HPAI (H5) virus was circulating among more duck fl ocks during this time and may be related to the practice of herding free-ranging ducks to scavenge on paddy fi elds postharvest as described for Thailand and Vietnam (24,25). Intermingling of ducks on paddy fi elds may al-Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 16  low extensive opportunities for virus release and exposure and contact with wild birds that also feed on leftover rice (25). However, the relationship between rice farming and HPAI outbreaks in Indonesia is likely to vary from those in Thailand and Vietnam because of different climatic conditions and rice farming calendars. Rice farming in Central Java is less seasonal, and rice paddies are smaller, often not separated by wide waterways, unlike in the Mekong Delta of Vietnam. Other factors, such as the long distance movement of duck fl ocks, may infl uence outbreak patterns in Indonesia. Further studies are needed on the management and movement of duck fl ocks, HPAI transmission pathways between different poultry species, and the association between rice harvest activities and increased HPAI outbreaks.