Newly emerging influenza pandemics provide the best opportunity to follow the dynamic changes of viral mutants in the initial phase as well as after public health measures have been instituted. We performed such studies in two high-density metropolitan areas in Taiwan, and made four observations that may enhance our understanding of influenza epidemiology. First, a significantly higher cumulative number of amino acid changes in HA and NA was found in the post-peak period of the epidemic. Second, a pH1N1 mutant with a unique change E374K in HA2 detected first at 34th week in Taipei and 6 weeks later in Kaohsiung, survived more successfully than other variants through transmission chains and became the major circulating strains in the post-peak period. Such well-adapted mutants were favored in certain areas of metropolitan Taipei, where pH1N1 cases had significantly higher spatial clusters during the peak of the second epidemic wave (weeks 40–47). Third, tempo-spatial increases in E374K for six weeks matched well with the second wave of class suspension (41–45th weeks). These increasingly dominant mutants persistently circulated for an additional three weeks before the first day of vaccination for healthcare workers and continued for two more weeks before implementing mass-vaccination for schoolchildren. Fourth, vaccination sharply reduced pH1N1 cases but at 10 months post-vaccination there was still inadequate antibody pressure to drive antigenic drifts in the circulating new dominant mutants. These findings imply that continuous selection of pandemic influenza viruses occurs most frequently at the time periods and places where numerous transmission events are happening and stress the importance of prevention measures to reduce the public health threat.
The higher frequency of cumulative amino acid changes of HA after the epidemic peak identified in this study can be explained by several factors. Due to lack of proofreading of viral RNA polymerase activity, HA has a very high rate of spontaneous mutation (estimated at 2×10−3
base substitutions/position per virus generation) 
. The quasispecies of pH1N1 virus within an individual host 
indicate that the intra-host selection of unique variants in viral populations should occur and subsequently be transmitted 
. This natural process leads to selective advantage variants that continue to circulate in the human population 
. Additionally, the recombination of influenza virus strains within an individual (e.g. intra-host) is also likely to affect host selection. This might drive the mutated virus variants with higher replication, better fitness and more efficient transmissibility to have a selection advantage and become the dominant variant 
The phenotypic variations in HA with public health significance involve five major dimensions: (1) RBS important for viral entry, (2) antigenic variations, (3) glycosylation sites related to viral virulence and immune escape, (4) clinical severity, and (5) increasing epidemiological significance. Most pH1N1 viruses with one or two amino acid residue changes in the antigenic sites of HA isolated in Taiwan occurred sporadically and were not fixed during the 2009–2010 epidemic. In the RBS of pH1N1 HA, the finding of a substitution at the 220-loop region (D222E, D222G), particularly D222G, reported to be associated with clinical severity 
, was not supported by our results or other reports 
. Other RBS mutant such as I216L, which led to efficient airborne viral transmission in ferrets 
, was also not found in our study.
Antigenic variation can involve changes in antigenic sites and glycosylation sites. N125D located at antigenic site Sa of HA, which was first detected on June 6, 2010 (week 23) (e.g. late epidemic phase) and circulated through October 2010. All nine Taiwanese N125D mutants had co-existing E374K mutations, consistent with the findings from Singapore, Australia and New Zealand 
. This N125D antigenic variant, similar to E374K, appeared after the epidemic peak but with lower frequency, indicating it had insufficient high immunological pressure to be fixed. Furthermore, HI serological titers showed no significant differences between these two strains with double mutants (HA: N125D and E374K) and the other two strains without double mutants, using pH1N1 vaccinated human serum and immune sheep serum. Moreover, the other substitutions located at Ca site (K142R, G170R, S203T and D222G) also showed no antigenic changes by HI reaction. These results imply that pH1N1 variants observed at the beginning of pandemic period had not developed the capability to escape from immune pressure, regardless of the cumulative number of amino acid changes identified at the antigenic sites. In addition, we did not detect any new glycosylation sites in the HA 
. Taken together, the lower numbers of cumulative amino acid changes, the less frequent multiple mutations at four antigenic sites and the lack of additional glycosylation sites identified in this study support the evidence of insufficient selection pressure. Immune selection pressure has been the main driving force for antigenic drift of human seasonal influenza viruses 
. This was not observed from the beginning pandemic period in 2009 till ten months after vaccination.
Our results in Taiwan did not find the three pH1N1 variants, Q293H, D131E, and S186P, that were associated with clinical severity. Such an inconsistent finding is likely due to mass application of antiviral and other public health prevention measures implemented in Taiwan to reduce viral transmissibility. This implies that clinically severe cases might have emerged under special epidemiological settings and searching for these answers will be helpful for informing future public health prevention decisions.
After analyzing all pH1N1 variants throughout the epidemic, this study identified two fixed mutation changes (S203T and E374K) in HA with epidemiological significance. However, these two mutations were not significantly co-varied (Fisher's exact p
0.08). The overall high percentage (97.13%, 169/174) of total Taiwanese pH1N1 viruses which had the S203T substitutions is consistent with other reports 
. These results indicate that such substitutions might have occurred before their introduction to Taiwan. Interestingly, this small change in a side chain near the monomer-monomer interface appears not to have had a dramatic effect on the structure of HA 
, nor on HI titers. In contrast to S203T, the E374K mutation occurred throughout the epidemic in Taiwan. Application of Tamiflu, that proved to reduce viral load 
, was implemented for all influenza A antigen-positive ILI patients from Aug 1, 2009 (week 30) and this might have delayed the peak of epidemic before the vaccination program 
. When the fall semester began in early September, the E374K mutant sharply increased in the first three weeks despite the nationwide class suspension policy initiated on September 1, 2009. After week 37, pH1N1 variants strikingly increased the number of amino acid substitutions to five or more and elevated the percentage of E374K variant replacing E374E. The class suspension did not stop the transmission of this E374K mutant, while the percentages of E374K mutant in the first wave (wks 35–40) and second wave (wks 41–45) of class suspension in Taipei City were 30.16% (19/63) and 100% (6/6) respectively. The increased transmission of pH1N1 might provide more chances to spread E374K mutant in the human population. The period of high percentage of E374K mutant matched well with the period before the start of mass vaccination of pH1N1 in elementary and high schools on November 16, 2009.
The tempo-spatial epidemiological conditions facilitating the increasing dominance of E374K mutants even before vaccination, supported by spatial epidemiology and multivariate analyses, were the higher spatial clusters of E374K mutants occurring in the post-peak of the epidemic period and significantly high spatial clustering in seven district regions with greater population densities. Because of the limited number of Taiwanese pH1N1 viruses (n
121) captured in this study, it may be difficult to fully characterize the impact of the tempo-spatial effects on E374K mutation. A larger sample size is needed to verify this conclusion.
For the newly emerged pH1N1, most of the persons born after 1957 were susceptible to this virus 
. A lesser level of herd immunity, that may be insufficient to result in high immune selection pressure, explains why less substitution of amino acids and less percentage of E374K mutants were isolated in the early pre-peak period of the epidemic. In fact, the overall vaccine coverage rates of pH1N1 in all Taiwan areas and Taipei by January 29, 2010 were about 24.3%, and 21.8%, respectively 
. The level of community-based herd immunity plus the extremely high vaccine coverage (74.7%, 271460/363403, January 29, 2010) in Taipei's 7–18 year old school children did reduce the size of the pH1N1 epidemic and transmission opportunities. Therefore, even at ten months after the nationwide vaccine campaign, there was insufficient selection pressure for antigenic drift of pH1N1 in Taipei. Taken together, all these results strongly support that E374K widely spread in areas with day-to-day schoolchildren gatherings even during the period of class suspension and anti-viral programs. It is very likely that unhygienic behavior, the high-density public transportation system, crowding, or all of the above might contribute to the tempo-spatial clustering of E374K mutants. Thus, the viral population size of the E374K mutant increased through epidemics, particularly at the pre-vaccination phase during post-peak periods. Certainly, without anti-viral agents, class suspension, and vaccination, more virus circulation and diversities of mutants may have emerged. On the other hand, the tempo-spatial epidemiological conditions at peaks might facilitate the selection of more virulent viral variants and thus confer increasing epidemic severity in later waves with more fatalities. This happened as documented in past influenza pandemics 
and other viral pandemics as well 
, once public health interventions had not been timely effective. These findings indicate that more attention s
hould be paid to non-pharmaceutical measures 
for future novel influenza viruses with pandemic potential.
Global comparison of the evolution of the E374K mutations from June to December of 2009 
found that lower numbers of cases with E374K mutation were reported in Canada (3.4%, 8/235) and Finland (2.38%, 3/126) in the temperate zone but more cases with this mutation were reported in Taiwan (37.24%, 73/196) and Singapore [September (14%), October (28%), November (55%)] in the subtropical and tropical zones with high population density, respectively 
. The roles of climate, population density, and human-to-human contact behaviors in facilitating the fixing of E374K mutants in those particular regions have not been clarified, indicating that timely international collaboration is needed among countries in different climate zones with various epidemiological characteristics once novel influenza virus emerges in future years.
The virological mechanism that might explain the fitness of the E374K mutant in the viral population is worth pursuing. Among several pH1N1 variants, E374K variants were not evident in the early epidemic period and their fitness would be random due to the lack of consistent co-substitutions 
. Single residue deletions in the HA fusion peptide can lead to loss of membrane fusion activity 
, however, the first 24 N-terminal residues of HA2 were conserved in all the 78 Taiwanese pH1N1 isolates. No significant differences in cross-MNt and HI serologic reactions were observed between the wild-type E374E and the E374K mutant, even using pH1N1-positive human and sheep serum samples with high titers. More results are needed with serum samples from patients experienced the 1918 influenza pandemic or with higher B-cell memory responses to clarify the possible mechanism of immune selection of this mutant, particularly because the fusion portion of HA2 offers partial protection for heterologous neutralization 
. Although E374K mutants had the similar growth capability in MDCK cells (Figure S3
), other biological characteristics that might be changed through affecting the stability between oligomers interaction 
or be less recognized or cleared by immune system facilitating persistent transmission in human population 
need further investigation.
In conclusion, this is the first study examining dynamic changes of pH1N1 viruses through integrating virological surveillance, tempo-spatial epidemiological characteristics, public health interventions, clinical severity and serological findings, hopefully providing fundamental information during the pandemic. This study had selection bias in the hospitals, patients' giving specimens, and fatal cases. A prospective large cohort study that can integrate with a high quality virological surveillance system through collecting clinical, epidemiological and intervention attributes, and meteorological variables plus deep sequencing analysis 
of virus population random directly from the ILI patients, will provide the best understanding on the micro- and macro-mechanisms of viral selection and fitness. Future efforts should identify the biological significance and explore the mechanisms involved in the fitness of pH1N1 viruses with amino acid residue substitutions in RBS sites, antigenic sites and the fusion regions of HA and their phenotypic characterization with human respiratory cell lines 
. Moreover, other segments of the viral genome, such as NS1 
and polymerase genes 
, might impact mutant fitness, and requires a more detailed investigation. Certainly, real-time comparative studies on the relationship between intervention strategies and molecular changes of pandemic influenza virus through international collaboration will be an important step for global control of influenza.