This study represents the largest single-year AI surveillance effort of wild birds in North America reported to date. Our program was implemented to monitor for possible introduction of HPAI H5N1 by migratory birds from Asia and 16,797 wild birds were sampled in Alaska. Although HPAI H5N1 was not found in any of the birds tested in 2006 and spring 2007, we detected low pathogenic AI viruses in 293 birds. Similarly, 189 low pathogenic viruses were detected by virus isolation. Because no outbreaks of HPAI H5N1 were detected during or subsequent to field sample collections, we infer that our surveillance program correctly concluded that this specific virus was not present in Alaska during our sampling. While our study was not specifically designed to compare low pathogenic avian influenza exposure rates, an examination of our findings does provide insight useful for design of future research.
AI prevalence varied among areas with the highest rate found in the interior of Alaska, and lower rates discovered along the West coast and Aleutian/Bering Sea Islands (Fig. 1). The lowest prevalence rates were found along the Arctic Coastal Plain and in Southeast Alaska. Similarly, a study of influenza A viruses in waterfowl in Alaska during the early 1990s found a higher prevalence in the interior versus coastal areas . Obviously, these bird conservation regions represent broadly differing habitat types that are occupied by different species of birds. Further, these regions may represent linkages with different wintering areas within species. For example, tundra swans (Cygnus columbianus) sampled in the Arctic spend the winter along the East coast of North America in the Atlantic Flyway; conversely, tundra swans sampled in Western Alaska spend the winter in the western US along the Pacific and Central Flyways. Our study could not determine if the geographic variation resulted from inherent habitat characteristics that influence virus persistence and transfer, or from variation in characteristics of species or populations.
Age and sex
Our results and those from other studies [1, 19] show that AI was more prevalent in juvenile waterfowl. This pattern implies that either the adult population transmitted the viruses to young birds or that the viruses were maintained in the environment and young birds were infected following hatch . The higher rate in young birds may be because they are immunologically naïve whereas adults are more resistant, particularly to viruses to which they may have previously been exposed .
Results of studies that examined differences in AI prevalence between sexes in birds have been inconsistent. In one case, female mallards had a greater prevalence of AI infections than males , more AI positives were found in males in an examination of several species , while a third study showed that males were more likely to test positive by H5 rRT-PCR, but not by matrix rRT-PCR, than females . Other studies detected no sex difference in AI prevalence rates [22, 23], however, we found differences in prevalence between sexes within age classes. The absolute difference in prevalence between sexes was greater in adults than in juveniles. For many species of waterfowl, males do not incubate eggs or rear offspring. Males of species that exhibit a sexual bias in AI prevalence rate might utilize a wider range of habitats and different foraging or roosting areas where they may encounter different groups of viruses than females. This hypothesis is supported by the greater difference between the sexes for adults compared to juveniles. At the time of our sampling, juvenile males and females would not be expected to have variation in exposure probabilities based on differing life histories.
We found considerable differences in AI prevalence among taxa but, overall, our rates are somewhat lower than found in surveillance studies conducted in Europe [22, 24]. In Alaska, dabbling ducks had the highest prevalence, which is generally consistent with results from previous sampling in North America and elsewhere [3, 19, 25]. Detection of AI in 4 eider species, whose populations from Asia and North America co-mingle annually in Alaska, suggests they may be important hosts for the intercontinental transport of these viruses. The prevalence in Charadriiformes (0.45%) in our study is similar to the 0.42% detected in northern Europe , but is lower than that observed in the Chesapeake Bay region of North America (14.2%, ). Although several seabird species (glaucous gull [Larus hyperboreus], Aleutian tern [Onychoprion aleuticus], common murre [Uria aalge], and thick-billed murre [U. lomvia]) tested rRT-PCR positive for AI viruses, samples from only 2 species of shorebirds (bar-tailed godwit [Limosa lapponica] and dunlin [Calidris alpina]) were positive, and AI viruses were not detected in passerines.
The prevalence of AI in waterfowl is typically low on wintering areas and declines further in spring, with prevalence rates in ducks of 0.4% in late winter and 0.25% as they return to nesting areas in Canada [1, 27]. In contrast, we found an unexpectedly high prevalence rate of AI (2.5% overall) in some species of spring migrants returning to Alaska. A similarly high (4.0%) prevalence was found in European ducks in the spring . These data suggest that bird-to-bird transmission in waterfowl might be sufficient to perpetuate influenza viruses from year to year. In addition, high-latitude wetlands may maintain viable viruses across years  and infect susceptible birds upon their return. For the 3 species where we had sufficient data (i.e., pintails, mallards, green-winged teal), there were consistent non-linear trends in AI prevalence with date of sampling. As noted previously, juveniles may be more susceptible than adults and this susceptibility may explain their seasonal increase in exposure rates. However, it is not clear why the exposure rate in adults increases simultaneously. Further studies of the causes and consequences of seasonal variation in AI prevalence are needed.
Differences between laboratory techniques
Contrary to our a priori expectations based on sampling, but similar to previous studies, we found higher prevalence from rRT-PCR compared to virus isolation. For example, in one study, molecular screening yielded 14.8% AI positives compared with 8.4% by virus isolation, and in another report, viruses were isolated from 60% of the molecular positives [20, 22]. This disparity has been attributed to factors such as amplification from nonviable viral particles, degradation of viruses prior to egg inoculation, and the inability of some AI viruses to grow to high enough titers to be detected in eggs [20, 28]. Further research is needed to identify the source of the disparity in results based on technical differences in methods.
Munster et al.  proposed that molecular testing is superior to the traditional technique of virus isolation for the detection of avian influenza in wild birds. We note that for large-scale surveillance programs where early detection of a possible pathogen is of paramount importance, molecular detection techniques can be readily scaled to accommodate large sample numbers. However, as noted by Munster et al. , even under ideal sample collection, specimen transport, and storage conditions, only 33.5% of their rRT-PCR test-positive samples resulted in virus isolation. In this study, viruses were isolated from 45% of the rRT-PCR positive samples. While some of the differences between rRT-PCR and traditional virus isolation may be due to low viral titer, inactivated or non-infectious viruses, many (32%, n = 61) of our virus isolates were from matrix rRT-PCR test-negative samples. Various species of wild birds have different diets that may lead to varying levels of non-specific inhibitors present in the cloacal swab samples. Future wild bird surveillance programs should consider the use of rRT-PCR methods with internal controls to monitor for the effects of non-specific inhibition . Until the issue of rRT-PCR-negative but virus-isolation-positive samples can be resolved, we suggest that surveillance programs in wild birds should comprise a combination of molecular and traditional virus isolation methods to provide a more comprehensive assessment of the avian influenza viruses in wild birds.
Comparison with other studies
The sources of variation noted above complicate exposure rate comparisons among studies. For example, our study found that AI prevalence, as indicated by positive matrix gene rRT-PCR test, was detected in 1.7% of the birds tested which contrasts with the 0.06% prevalence reported by Winker et al. , who tested 8,254 birds of 64 species in Alaska between 1998–2004. Conversely, Runstadler et al.  found an AI prevalence of 25.6% but only sampled 4 species of dabbling ducks, in August, at one location in interior Alaska. A recent study in Canada found AI positives in 37% of 17 species of ducks (90% of which were dabbling ducks) sampled primarily in August and September in several provinces . As our results indicate, differences in prevalence between studies may be due to factors such as species composition, sex and age distribution, timing and location of sampling, as well as differences in sampling and detection methods. Thus, direct comparison and interpretation of simple prevalence rates are likely uninformative.