Microevolution of DENV in Kamphaeng Phet over four dengue seasons
To investigate the genetic diversity and structure of DENV populations circulating in Kamphaeng Phet () from 2004 to 2007, we sequenced and analyzed the E genes of viruses collected during school-based surveillance and geographically-based community cluster studies. At least three serotypes circulated in our study population during each year of the study, with all four serotypes detected only in 2005 (). This pattern generally reflects the relative proportions of DENV isolated through passive surveillance across the region during the period of sampling, although passive surveillance data from the Kamphaeng Phet Provincial Hospital detected all four serotypes circulating each year (). Years in which a given serotype was not detected in our study population corresponded to instances in which that serotype was detected at less than 10% frequency during passive surveillance. This suggests that our sampling and sequencing protocol adequately captures the diversity of the viral population circulating in KPP in a given year.
KPPPH data of all reported and PCR confirmed dengue cases in Muang District from 2004–2007.
Only a single genotype was represented for each serotype in this data set (DENV-1, genotype I; DENV-2, Asian I genotype; DENV-3, genotype II; DENV-4, genotype I). These four genotypes have been the dominant circulating genotypes in Thailand and much of mainland Southeast Asia for several years, and were all present among previous Thai E gene sequences from Thailand dating back to 2001 and earlier. Although other genotypes were detected in Thailand prior to our study, only DENV-1 genotype I, DENV-2 Asian I genotype, and DENV-3 genotype II were isolated in a study performed in Kamphaeng Phet in 2001; no DENV-4 sequences were isolated during the 2001 study 
. Thus, it appears that these were the dominant circulating genotypes in Kamphaeng Phet, at least through 2007.
Phylogenetic analysis revealed considerable viral genetic diversity within genotypes in the Kamphaeng Phet region during this period, demonstrated by the presence of phylogenetically distinct lineages within each serotype. Among the DENV-1 isolates detected, five major lineages were present, suggestive of at least five separate virus introductions prior to or during the sampling period ( and ). One lineage was detected briefly in 2005, but was not observed in the following years. Three lineages circulated in 2006, the year in which DENV-1 incidence greatly increased in the region (). All of these lineages persisted into the 2007 season, during which one additional lineage was detected. With the exception of Lineage 4, these viral populations were most closely related to viruses isolated in Thailand in the previous decade. Lineage 4 may have been a recent introduction from Singapore (the most closely related geographical region on the phylogenetic tree) that invaded Na Bo Kham sub-district and persisted within that population for at least two years. In all of the lineages except Lineage 2, viruses commonly mixed among sub-districts within a single season.
Timing of DENV lineages circulating in five participating sub-districts of KPP, Thailand from 2004–2007.
ML tree of representative E gene sequences of DENV-1 genotype I.
Despite the small number of sequences obtained, the DENV-2 population also showed considerable genetic diversity. Four lineages were detected between 2004 and 2007, with little mixing detected among sub-districts within a given season ( and ). Interestingly, however, Lineage 2 was first detected in rural Na Bo Kham in 2005, and was not isolated again until in 2007 in the most cosmopolitan of our populations, Nong Pling. This suggests that in this environment viral populations may sometimes move from rural to more densely populated areas rather than the opposite, as has been reported for other locations 
ML tree of representative E gene sequences of the Asian I genotype of DENV-2.
Although DENV-3 exhibited the lowest incidence in the study population during this time period, one lineage was detected in 2004 and a different lineage was detected in 2005 ( and ). Both of these were derived from lineages that were circulating in Thailand in the previous decade, which suggests that rare lineages may persist in rural populations even when disease incidence is low.
ML tree of representative E gene sequences of DENV-3 genotype II.
DENV-4 was the dominant serotype detected in Kamphaeng Phet during the first two years of this study. Two lineages were present in 2004; one was represented by a single isolate detected in Na Bo Kham (Lineage 3), and the other was represented by three distinct, primarily sub-district-specific clusters, showing limited mixing in a given year (Lineage 1) ( and ). Lineage 1 appeared to persist as the dominant DENV-4 lineage in the population throughout the study, although multiple distinct populations were present through 2007. Two additional introductions were observed in the Na Bo Kham and Nakon Chum sub-districts in 2006 (Lineage 4 and Lineage 2, respectively). Based on the sequences in this study, these introductions did not appear to result in extended transmission in the region. Interestingly, while most lineages detected appeared to have entered following long-term transmission within Thailand, a single sequence isolated in Na Bo Kham in 2006 (Lineage 4) may have entered through the re-introduction of a Thai lineage following extended transmission elsewhere in Southeast Asia.
ML tree of representative E gene sequences of DENV-4 genotype I.
It is interesting that many of recent introduction events occurred in Na Bo Kham sub-district; this is the most rural of the populations studied and more distantly positioned and isolated relative to the other study populations. A previous study performed in Kamphaeng Phet during a single, high incidence dengue season detected the greatest genetic diversity among viruses in the most densely populated areas 
, while the current study indicates the opposite. Because most of these lineages did not appear to immediately move into the greater population of Kamphaeng Phet, it is possible that the DENV populations present in Na Bo Kham differ greatly from the rest of the area because there is less mixing with the other populations we sampled. Alternatively, Na Bo Kham may depend on a different population center through which viral variants are introduced. Differences among DENV from Na Bo Kham and the rest of the sub-districts studied indicate that geographically distinct patterns of human movement are important processes in the structure and dynamics of DENV populations 
Spatial and temporal clustering of DENV in Kamphaeng Phet
Among village clusters initiated by dengue cases detected in the school-based cohort, nearly all of the DENV viruses sequenced from both humans and mosquitoes within a cluster during the 14-day period following the initiation of sampling exhibited identical or nearly identical E gene sequences to the index case of their respective clusters. All but three of these sampling clusters involved viruses originating from the same lineage within a single serotype. The three exceptions are described here. First, in the case of DENV-1, one cluster of three samples included an index case and one village contact that possessed identical E genes in Lineage 2 and a second village contact that was infected with a virus in Lineage 1, which was found to be circulating simultaneously in multiple sub-districts during the 2006 and 2007 seasons. Second, a single DENV-1 virus was isolated from a cluster investigation in which the index case and all other village contacts were infected with DENV-4. This DENV-1 isolate fell within Lineage 1 and was the first virus collected within a large phylogenetic cluster of identical E gene sequences isolated in the same village for the following seven to 31 days, from elsewhere in Nong Pling sub-district on days 62 and 97 after the first isolation, and from Thep Na Korn and Kon Tee sub-districts in the following 97 and nine to 108 days, respectively, spanning 2006 to 2007. Third, among DENV-4 isolates, one cluster in Na Bo Kham in 2004 (Lineage 1) included a virus of a divergent lineage (Lineage 3). Multiple closely related viruses were isolated from both humans and mosquitoes, forming a well-supported clade with other viruses from the same sub-district in the same season. A single mosquito was found to be infected with a divergent virus, comprising the only isolate of DENV-4 Lineage 3 detected in this study. That this lineage was only detected in a mosquito and we did not detect evidence of further transmission within the study population, as well as the number of lineages represented by a single isolate in the data set, suggests that multiple DENV lineages commonly enter this population in a single season, some of which fade out without detection, or may persist at low levels with little to no detection in the human population.
Both the sequence data presented here and the epidemiological data collected from the cohort and cluster studies 
suggest that DENV transmission is indeed highly spatially and temporally focal rather than occurring via simultaneous circulation and mixing of multiple DENV lineages across the region 
. For confirmation, we performed a statistical analysis of the strength of phylogenetic clustering of viruses sampled closely in space and time. At all levels of spatial aggregation (sub-district, village, school, and sampling cluster) and for all three serotypes investigated (DENV-1, DENV-2, and DENV-4), we detected a significant relationship between phylogeny and space; i.e., there was more clustering by spatial variable than expected by chance alone (). Similarly, we found a strong phylogenetic clustering of viruses by year of sampling. Hence, we conclude that viral genetic diversity in this population tends to turnover on an annual basis, although lineages may occasionally persist over multiple seasons.
Phylogeny-trait association tests of phylogenetic structure of geographic, temporal, and clinical traits of DENV infections.
Among the levels of spatial aggregation analyzed, the home sub-district and village of subjects showed stronger phylogenetic clustering (sub-district association index (AI): 0.09 to 0.31, parsimony score (PS): 0.3 to 0.51; village AI: 0.28 to 0.5, PS: 0.55 to 0.61) for all three serotypes than did the school (AI: 0.45 to 0.62, PS: 0.63 to 0.65) and sampling clusters (AI: 0.34 to 0.67, PS: 0.68 to 0.77) (). This result is not unexpected because epidemiological data suggest that DENV transmission in this region primarily occurs at a person's home or the home of a friend or relative 
. Our virological data, therefore, support these earlier results and indicate that these trends continued over the four-year time period. No significant phylogenetic clustering of DENV was observed for age, sex, or clinical syndrome, except in the case of DENV-2, for which relatively weak phylogenetic structure was detected according to clinical syndrome (). This result was strongly influenced, however, by the existence of two clusters of unknown syndrome and non-hospitalized DF, as well as, by the small number of DENV-2 sequences isolated in this study. It does not appear to result from differences in virulence among viral populations. The use of coding scheme 1, in which subjects associated with no school and mosquitoes were coded as NA and mosquito, respectively, produced weaker associations in the school-based analysis than when these were coded according to the school of the respective index case (data not shown). Specifically, these sequences rarely showed clustering, but instead interrupted or reduced the size of school-based clusters.
By restricting our analysis to those sequences obtained from index cases and through school-based surveillance, we were able to further assess whether highly focal sampling in the village-based cluster design was a primary factor influencing the spatial and temporal structure detected. Results of these analyses were generally similar to those obtained using the full set of sequences (). Significant clustering was observed at the sub-district, village, school and 100-meter radius cluster levels and within years for both DENV-1 and DENV-4. The strength of village-level clustering was reduced, however, relative to that of sub-districts and schools (). Further, the diversity within this subsample was comparable to that observed in the full data set, with the loss of only a single lineage (DENV-4 Lineage 3) in subsampling. These results indicate that distinct viral populations may be present in areas separated by only a few kilometers, and suggest that school-based virological surveillance alone captures much of the genetic diversity of DENV within a given area. As such, school-based surveillance may be a practical and efficient indicator of DENV circulating in communities in endemic areas.
Phylogeny-trait association tests of phylogenetic structure of geographic and temporal traits of DENV-1 and DENV-4.
An important observation from our study is the strength of temporal DENV clustering by year. This suggests that seasonal bottlenecks commonly occur in this population and, hence, that only a subset of viral genetic diversity within a given year survives into the next. This may in turn result in regular reductions in population immunity and competition among viruses, thus allowing the introduction and dissemination of viruses from outside of the study area. While persistence of viral lineages and in situ
evolution occur over multiple seasons in some cases, these processes are relatively limited compared to that of viral migration (i.e. importation), which appears to occur via movement of infected people, play a dominant role in generating diversity within each serotype, and contribute to dynamics in patterns of DENV transmission. This finding is similar to that reported for Kamphaeng Phet in 2001 
, and further indicates that it is important to account for the entry and re-entry of DENV lineages from external populations when considering the potential response to vaccination programs. In the context of forthcoming vaccine trials, and given the potential for imperfect vaccination and complex immune interactions, these results imply that the risk of DENV transmission and severe disease within a community will not only be determined by the vaccination levels within that geographic area, but also by the vaccination status and movement of people in neighboring regions.