Divergent Evolution of Norovirus GII/4 by Genome Recombination from May 2006 to February 2009 in Japan^{▿ ‡}

Stool specimens.

The Norovirus Surveillance Group of Japan collected stool specimens from NoV-GII- or GII/4-positive individuals with acute gastroenteritis (n = 247). Most of the specimens were from NoV outbreaks around the collection sites. The group collected the specimens in spring, summer, autumn, and winter for 3 years: the 2006/2007 season (May 2006 to January 2007), 2007/2008 season (March 2007 to February 2008), and 2008/2009 season (May 2008 to February 2009). The collection sites were located at 20 different regional public health institutes in Japan (five samples from each institute per year). The genogroup of NoVs was evaluated by real-time reverse transcription-PCR (RT-PCR) (23). In some cases, the genotype of NoVs was evaluated by sequencing of the reverse transcription-PCR products of the ORF1 and ORF2 bordering region (29). Near-full-length genome sequences were obtained with 199 of the 247 specimens. Epidemiological information on 37 of the 199 samples from the 2006/2007 season was described previously (38). Information on the rest (n = 162) is described in Tables S1 and S2 in the supplemental material. Briefly, the 162 specimens were from outbreaks (n = 90), sporadic infection cases (n = 15), and undescribed cases (n = 57) during December 2006 to February 2009 in Japan. The major sites of the incidences were a nursing care center (n = 19), restaurant (n = 17), kindergarten (n = 15), hotel (n = 8), hospital (n = 7), sports event (n = 1), self-defense force (n = 1), family home (n = 1), elementary school (n = 1), and bank (n = 1), and one was undescribed (n = 91). The viral RNA copy numbers in the specimens ranged from 5.0 × 10⁴ to 1.9 × 10¹¹ copies/g stool (average, 6.1 × 10⁹ copies/g stool) as judged by the real-time quantitative reverse transcription-PCR assay (23). All stool specimens were stored at −80°C until use.

Viral genome sequencing.

NoV GII/4 genome sequencing was done as described previously (38). Briefly, two overlapping fragments (approximately 5.2 and 2.5 kb) were amplified by RT-PCR from stool specimens. The PCR products were purified and used as a template for sequencing in a 96-well scale using an ABI 3730 xl DNA analyzer (Applied Biosystems, Foster City, CA). The sequences of 5.2-kb and 2.5-kb segments from the same individual were used to reconstruct near-full-length genome sequences (about 7.5 kb) by alignment at an overlapping region using the Staden Package (http://staden.sourceforge.net). The 5.2-kb fragment covers the complete ORF1 and the 5′ end of ORF2. The 2.5-kb fragment covers the 3′ end of ORF1, complete ORF2 and ORF3, and 3′-end noncoding region of the genome. The primers used for reverse transcription and nested PCR for the 5.2-kb fragment were GII4-1F/GII4r5412 (outer primer pair) and GII4-2F/GII4r5295 (inner primer pair) (38). Those for the 2.5-kb fragment were COG-2F/Tx30SXN (outer primer pair) and G2SKF/Tx30SXN (inner primer pair) (38). The initial 22 nucleotides at the 5′ ends of the reconstructed genomes were from PCR primers. The final 45 nucleotides at the 3′ ends of the genome were excluded from analysis because of the low levels of sequence accuracy. We obtained 199 near-full-length genome sequences from 247 GII-positive specimens. The 199 sequences included 37 GII/4 sequences previously reported between May 2006 and January 2007 (38) and 162 sequences newly obtained between December 2006 and February 2009.

Molecular cloning and sequencing of genome segments.

The 5.2-kb, 1.0-kb, and 2.8-kb genome segments were amplified by RT-PCR products as described above and cloned into pPCR-XL-TOPO vectors (Invitrogen, Carlsbad, CA). Each of the segments covers a junction of putative recombination breakpoints around the 5′ end of ORF2: the 5.2-kb segment contains the near-full-length ORF1 and 5′-end portion of ORF2, the 2.8-kb segment contains the 3′-end portion of ORF1, complete ORF2, and complete ORF3, and the 1.0 kb segment contains the 3′-end portion of ORF1 and 5′-end portion of ORF2. The primers used for the nested PCR of the 5.2-kb segment were the same ones described above: GII4-1F/GIIr5412 (outer primer pair) and GII4-2F/GIIr5295 (inner primer pair) (38). The primers used for the nested PCR of the 2.8-kb fragment were GII4f4117 (5′-CTGACAAAATTTATGGTAAGATCAAGAAGAGG-3′)/Tx30 SXN (outer primer pair) and GII4f4762 (5′-GACCCAGCTGGTTGGTTTGGAAAA-3′)/GII4r7516 (5′-ATAGTTTAGCGGCCGCATTCTTATCACATTACACCCGTGACTCCCCTCG-3′) (inner primer pair). The primers used for the nested PCR of the 1.0-kb fragment were GII4f4117/GII4r5412 (outer primer pair) and GII4f4223 (5′-GGTATGAATATGAATGAGGATG-3′)/GII4r5295 (inner primer pair). The single clones of the 5.2-kb, 2.8-kb, and 1.0-kb genome segments of the GII/4 subtypes were randomly chosen and sequenced in 96-well plates using an ABI 3730 xl DNA analyzer as described above.

Phylogenetic analysis.

Phylogenetic trees were constructed using the neighbor-joining method and maximum-likelihood method. Briefly, the near-full-length genome sequences from this study were aligned with the available GII/4 genome sequences from past NoV epidemics occurring over the past 3 decades, using CLUSTAL W software included in the MEGA software package, version 4.0 (58) (http://evolgen.biol.metro-u.ac.jp/MEGA/) and the MAFFT multiple sequence alignment software program, version 6.0 (26) (http://align.bmr.kyushu-u.ac.jp/mafft/software/). The neighbor-joining trees were constructed with the nucleotide substitution values estimated with the maximum composite likelihood model (59) using MEGA. The maximum-likelihood trees were inferred on the basis of the general time reversible models (31) using the PHYML software program included in the RDP3 software package (35) (http://darwin.uvigo.es/rdp/rdp.html). The reliability of interior branches in the phylogenetic tree was assessed by the bootstrap method with 1,000 resamplings. The GII/4 genome reference sequences were from samples taken before 1990 (<1990) (6 sequences, CHDC591-1974, CHDC2490-1974, CHDC4871-1977, CHDC4108-1987, Lordsdale, MD145-12/US/1987, and CHDC3967-1988), before 2000 (<2000) (2 sequences, and Dresden174/US/1997), in 2002/2003 (6 sequences, Famington Hill, B2S16/2002/UK, B5S22/2002/UK, Langen1061/2002/DE, YURI32073/2002/JPN, and MD-2004/2004/US), and in 2004/2005 (4 sequences, Guangzhou/NVgz01/CHN/2006, Chiba/04-1050/2005/JP, Sakai/04-179/2005/JP, and Ehime/05-30/2005/JP). Accession numbers for the reference genome sequences are given elsewhere (7, 38).

We initially constructed the phylogenetic trees with 199 genome sequences from the Japanese variants from 2006 to 2009 and 6 representative sequences of GII genotypes whose complete genome sequences were available in GenBank in October 2009 (GII/1, GII/3, GII/4, GII/6, GII/10, and GII/12; accession no. U07611, AB067542, X86557, AB039776, AY237415, and AB039775, respectively). The trees showed that the 199 genome sequences reproducibly grouped with the GII/4 reference sequences outside other GII references. The GII/4 cluster was positioned most closely to the GII/12 reference (Saitama U1/JP). Therefore, we used GII/12 as an outgroup in the present study for a better grasp of the relationship of the phylogeny among the Japanese GII/4 variant subgroups and between GII/4 and GII/12 variants.

Bootscanning-plot analysis.

Bootscanning-plot analysis was performed as described previously (69). Briefly, each query sequence was aligned with three NoV reference sequences using CLUSTAL W software, version 1.4 (62). The bootstrap values were plotted for a window of 300 bp, moving in increments of 10 bp along the alignment using the software program Simplot (48) (version 3.5.1; http://sray.med.som.jhmi.edu/SCRoftware/simplot/). Thus far, 19 genotypes of the NoV GII variants have been reported on the basis of complete capsid sequences (65, 71). Among them, only 7 genotypes have been fully sequenced at the genome level (GII/1, GII/3, GII/4, GII/6, GII/8, GII/10, and GII/12; accession numbers U07611, AB067542, X86557, AB039776, AB067543, AY237415, and AB039775, respectively). To search for sequences that are phylogenetically relevant to the query sequences, we constructed phylogenetic trees of the complete ORF1, ORF2, and ORF3 sequences using all available representatives of the 19 genotypes in the GenBank database. We also used the automated exploratory analysis tool included in the RDP3 software package (35). The genome sequence set used for the analysis consisted of 7 query sequences (2004/05, 2006a, 2006b, 2007a, 2007b, 2008a, and 2008b), all available GII genotype representatives, and all available GII/4 variant subgroups which caused epidemics over the past 34 years (7, 31). Two putative parent sequences with the best confidence values and a single distantly related sequence were used for the bootscanning plots with MEGA. The confidence values of the recombination events were also assessed with tools included in the RDP3 software package, such as RDP, GENECONV, Maxchi, Chimera, 3seq, and Siscan. The query sequences used in this study were Sakai2/2006/JP for the 2004/05 subtype (accession no. AB447448) and representative genomes of the 2007a (Osaka1/2007/JP), 2007b (Iwate5/2007/JP), 2008a (Hokkaido5/2008/JP), and 2008b (Hokkaido4/2008/JP) subtypes obtained in this study. The reference sequences were Saitama_U1/JP (GII/12 genotype [25], accession no. AB039775), B2S16/2002/UK (2002/03 subtype [38], accession no. AY587989), Saitama_U3/JP (GII/6 genotype [25], accession no. AB039776), Sakai2/2006/JP (2004/05 subtype [38], accession no. AB447448), Aomori1/2006/JP (2006a subtype [38], accession no. AB447432), Aichi3/2006/JP (2006b subtype [38], accession no. AB447446), and Hokkaido5/2008/JP and Hokkaido4/2008/JP (2008a and 2008b subtypes, respectively, obtained in this study).

Informative-site analysis.

The informative-site analysis was performed as described previously (50). Briefly, each query sequence was aligned with two putative parental sequences and an outgroup sequence. The alignments were used to identify informative sites that support alternative tree topologies between downstream and upstream regions using the Simplot software program (48), version 3.5.1. This information allowed identification of genome regions that were assigned as chimeras of heterologous sequences of distinct evolutionary origins. The statistical significance of the resultant division by the informative sites was evaluated by the maximal χ² test using in-house programs. The programs were designed to execute the calculation algorithms described by Robertson et al. (50, 55).

Molecular modeling.

Three-dimensional (3-D) structural models of the capsid P-domain dimers were constructed by homology modeling as described previously (38). Briefly, the P-domain monomer models were first constructed using the crystal structure of the NoV capsid P domain of the GII/4 VA387 strain at a resolution of 2.00 Å (PDB code 2OBS [13]) as the template. The P domains of the GII/4 subtypes described in this study have sequence similarities of greater than 90% to that of VA387, high enough to construct models with a root mean square distance (RMSD) of ∼1 Å for the main chain between the predicted and actual structures (3). The P-domain monomer models were used to construct the P-domain dimer models by superimposing the chains A and B using the crystal structure of the NoV capsid dimer (PDB code 1IHM [47]).

Nucleotide sequence accession numbers.

The DDBJ database accession numbers for the nucleotide sequences of NoV genomes for the 2006/2007 season (n = 37) have been reported elsewhere (38). The DDBJ database accession numbers for the nucleotide sequences of NoV genomes for the 2007/2008 and 2008/2009 seasons (n = 162) are AB541201 to AB541362. The DDBJ database accession numbers for the nucleotide sequences of NoV genome segment clones (n = 11) are AB541190 to AB541200.

Phylogenetic classification of NoV GII/4 subtypes in Japan during 2006 and 2009.

First, we investigated the phylogeny of the NoV near-full-length genome sequences (about 7.5 kb). For this study, we used sequences obtained in this study from 19 sites in Japan between May 2006 and February 2009 (n = 199), various reported GII/4 reference sequences of past global or Japanese epidemics, and various reported outgroup sequences of other NoV genotypes. Figure 1 shows a maximum-likelihood tree constructed with the 199 Japanese genome sequences and the 19 GII/4 reference sequences from past NoV epidemics throughout the world during the 1970s and 1980s (7), <2000, in 2002/2003, and in 2004/2005 (38). The tree shows that the 199 Japanese sequences are divisible into 7 distinct lineage groups within a GII/4 cluster with a high bootstrap value (the 7 colored ovals in Fig. 1). The monophyly of the 7 GII/4 groups was reproducible independently of the algorithms to infer the phylogeny and reference sequences used. We tentatively named the 7 monophyletic subtypes of GII/4 variants 2004/05, 2006a, 2006b, 2007a, 2007b, 2008a, and 2008b.

The 2004/05 genome sequences were first obtained in Japan in the winter of 2004-2005 (accession no. AB220921 to AB220923 [42]). The geographic distribution of the 2004/05 sequences seemed to be restricted to East Asia (54). The 2006a and 2006b genome sequences were first obtained in Japan during the winter of 2006-2007 (accession no. AB447427 to AB447463 [38]). The 2006a and 2006b sequences were detected in many countries in Europe, North America, and East Asia during 2006-2007, wherein the 2006b subtype was generally more dominant than the 2006a subtype (54). The 2007a, 2007b, 2008a, and 2008b genome sequences were newly obtained in this study. Phylogenetic tree analyses showed that the nucleotide sequences of ORF2 of the 2008a subtype were genetically closely related to the ORF2 sequence obtained in the Netherlands in 2008 (accession no. AB445395), and together these sequences formed a single monophyletic group with a high bootstrap value (data not shown). These results suggest that at least 4 of the 7 GII/4 subtypes identified in Japan during 2006 and 2009, i.e., the 2004/05, 2006a, 2006b, and 2008a subtypes, caused NoV infections outside Japan.

We estimated the genetic divergence within and between the 7 monophyletic groups on the basis of the maximum composite likelihood model using MEGA software. The intragroup divergence was comparably high in the 2006b subtype among the 7 groups (see Table S3, diagonal lines, in the supplemental material), suggesting that the diversity of the 2006b genome is higher than that of the other subtypes. This is consistent with the epidemiological data that 2006b had predominated for 3 years in Japan whereas the others emerged only temporally. The intergroup divergence was comparably high between 2004/05 and the other groups and between 2007a and the other groups, and about 12 to 15% sequence divergence existed in the genomes (see Table S3, bottom left portion, in the supplemental material).

Temporal and geographical distribution of Nov GII/4 subtypes in Japan.

We then analyzed the temporal and geographical distribution of the 7 GII/4 subtypes in Japan. The 199 near-full-length genome sequences were divided into 3 groups according to the collection periods: the 2006/2007 (May 2006 to January 2007) (n = 39), 2007/2008 (March 2007 to February 2008) (n = 78), and 2008/2009 (May 2008 to February 2009) (n = 82) seasons. The frequencies of detection of particular NoV subtypes were obtained for each of the three seasons. We also used published subtyping data for the analysis of the previous winter season in Japan (November 2005 to March 2006) (n = 38) (38, 43).

The 2004/05 and 2006a sequences were detected at multiple collection sites and were prevalent in the 2005/2006 season (38, 43) (Fig. 2 A and B, 2004/05 and 2006a). However, they became minor in the 2006/2007 season and were hardly detected thereafter. The 2006b sequences were minor in the 2005/2006 season (38) (Fig. 2A, 2006b). However, they rapidly became dominant in the 2006/2007 season and continually predominated in most of the collection sites in Japan, representing 176 of the 199 genome sequences (88.4%) during the study period. This result is consistent with the data of partial capsid sequences obtained during December 2007 to January 2008 in Japan (28). The 2007a and 2007b sequences were detected only at single collection sites in the 2007/2008 season (Fig. 2A and B, 2007a and 2007b). The 2008a and 2008b sequences were detected most recently at multiple collection sites in the 2008/2009 season (Fig. 2A and B, 2008a and 2008b). These data indicate that the 2006b subtype displaced the 2004/05 subtype in the 2006/2007 season and continued to predominate for the next 2 years in Japan. During the period of the 2006b predominance, however, several GII/4 subtypes caused NoV outbreaks in Japan, and the frequencies and sites of non-2006b outbreaks increased slightly in the 2008/2009 season.

Phylogenetic evidence for NoV genome mosaicism.

Next, we investigated the possibility of genome mosaicism of the 7 GII/4 subtypes. For this purpose, we first compared the branching orders of the subtype clusters in the maximum-likelihood and neighbor-joining trees of the ORF1, ORF2, and ORF3 sequences using representative sequences of the 19 GII genotypes (GII/1 to GII/19) reported to date in the GenBank database. Figure 3A shows the maximum-likelihood trees, in which most of the non-GII/4 sequences were positioned far from the GII/4 cluster and were therefore excluded for a better grasp of the relationship of the phylogeny among the GII/4 variant subgroups. The exception was the ORF1 tree, in which the GII/12 sequence branched inside the GII/4 cluster. The comparisons of the three trees revealed that there was marked inconsistency in the branching orders of the GII/4 subgroups. The inconsistency was reproducible independently of the algorithms to infer the phylogeny and reference sequences used. First, the ORF1 sequences of the 2006b, 2007a, 2007b, and 2008b subtypes formed independent monophyletic clusters, whereas the ORF2 sequences of the 2006b, 2007a, and 2008b subtypes formed a single cluster and the ORF3 sequences of these four subtypes formed the same cluster (Fig. 3A, light blue circles). Second, the ORF1 sequences of the 2004/05 subtype were clustered near the ORF1 sequence of a GII/12 strain (Saitama_U1/JP [25]) and relatively distant from the reported GII/4 reference sequences, whereas the ORF2 and ORF3 sequences of the 2004/05 subtype were very distantly related to the Saitama_U1/JP sequence and closely related to the GII/4 reference sequences (Fig. 3A, yellow circles). Third, the branching orders of the 2008a sequences were also different in the ORF1, ORF2, and ORF3 trees (Fig. 3A, red circles). These results suggested that most subtypes identified in this study had mosaic genomes.

To further assess this possibility, we performed bootscanning-plot analyses as described previously (69). For each bootscanning plot, we used a query genome sequence of a given subtype, two to three reference sequences that were positioned relatively closely to the query sequence in the neighbor-joining trees, and a distantly related outgroup sequence. The analyses showed that the genomes of the 2004/05, 2007a, 2007b, 2008a, and 2008b subtypes were indeed composed of multiple segments from recently prevalent or as-yet-undefined genogroups, genotypes, and subtypes of NoVs in this and previous reports (2, 7, 25, 38, 53, 65, 71) (Fig. 3B; see also Fig. 3A). The 2004/05 genome (Sakai2/2006/JP) was comprised of the ORF1 related to GII/12 (Saitama_U1/JP) and the ORF2/3 related to GII/4 2002/03 (B2S16/2002/UK). The 2007a genome (Osaka1/2007/JP) was made up of the ORF1 related to GII/12, the ORF2 of as-yet-undefined classes of GII/4, and the ORF3 related to GII/4 2006b (Aichi3/2006/JP). The 2007b genome (Iwate5/2007/JP) was made up of the ORF1 related to GII/4 2006b and 2006a (Aomori1/2006/JP) and the ORF2 and ORF3 related to GII/4 2006b. The 2008a genome (Hokkaido4/2008/JP) was comprised of the ORF1 related to GII/4 2008b (Hokkaido5/2008/JP) and as-yet-undefined classes of GII/4 and the ORF2 and ORF3 of GII/4 2006a and as-yet-undefined classes of GII/4. The 2008b genome was made up of the ORF1 of 2008a and the ORF2 and ORF3 related to GII/4 2006b. We also investigated possible genome mosaicism for the 2006a and 2006b subtypes, but we could not identify putative ancestral sequences of ORF1, ORF2, and ORF3 that were genetically closely related to 2006a and 2006b when we used the available NoV sequences in the public database as references.

To define statistically the possible recombination breakpoints of the 2004/05, 2007a, 2007b, 2008a, and 2008b genomes, we performed informative-site analysis (50) using the same reference sequences used in the bootscanning-plot analysis. With this approach, we identified several patches of genome regions that were assigned with statistical significance as putative recombination breakpoints. Notably, a putative breakpoint located around the junction of ORF1 and ORF2 constantly gave the highest statistical significance, i.e., maximum χ² values, in the 2004/05, 2007a, 2007b, 2008a, and 2008b genomes (P ≤ 0.000001) (Fig. 3B, arrows). The results were in good agreement with the phylogenetic-tree and bootscanning-plot analyses. These data consistently suggest that the new GII/4 subtypes identified in Japan were mostly hybrid viruses composed of viral protein elements from distinct genetic lineages of NoVs.

We further assessed possible genome recombination events using other tools included in the RDP3 software package (30). The analysis again identified single recombination breakpoints with the best or second-best confidence values around the junction of ORF1 and ORF2 in the 2004/05, 2007a, 2007b, 2008a, and 2008b genomes (P < 0.001). The analysis also identified additional putative breakpoints around the junction of ORF2 and ORF3 of 2007a. However, we could not obtain evidence for genome mosaicism with 2006a and 2006b using a selected sequence data set of the NoV GII genotypes reported to date (GII/1 to GII/19) (25, 65, 71) and GII/4 subtypes (7, 38, 53). Because information on the entire genome sequences of NoV is very limited, it remains to be determined whether 2006a and 2006b also have mosaic genomes.

Isolation of NoV mosaic genome segments.

To clarify the presence of the mosaic viral genomes in nature, we cloned and sequenced the genomes of the 2007a, 2007b, 2008a, and 2008b subtypes. For this study, we cloned the genome segments, i.e., the 5.2-kb, 2.8-kb, and 1.0-kb genome segments, that presumably contain a junction of putative recombination breakpoints around the ORF1/ORF2 junction (Fig. 4). The subtype 2004/5 was not included in the cloning analysis because we did not have sufficient amounts of clinical specimens for the cloning. Likewise, the subtypes 2006a and 2006b were not included because the major parental sequences are not clear in the present study. We successfully obtained the molecular clones except for the 5.2-kb fragment of 2007a. We could amplify but failed to clone the 5.2-kb fragment of 2007a. Although the precise reason for the failure is unclear at present, it might be due to the decreased cloning efficiency of the larger insert by the TA-cloning method. Because the appropriate restriction enzyme sites for the cloning were absent in the 2007a 5.2-kb fragment, we did not include this fragment in further analyses. Nucleotide sequences of the segments were used for the bootscanning-plot analysis using the same sets of reference sequences described in Fig. 3B, and the statistically significant putative recombination breakpoints were assessed by informative-site analysis.

Figure 4 shows representative results of the bootscanning-plot and informative-site analyses with the 5.2-, 1.0-, and 2.8-kb segment clones. Importantly, all 11 clones from the 2007a, 2007b, 2008a, and 2008b stool specimens had the same putative recombination breakpoints, with the highest statistical significance around the ORF1/ORF2 junction region identified with direct sequencing analyses (Fig. 4A and B, arrows). In addition, the patterns of the bootscanning plots were almost identical over the viral genomes examined between the sequences of the uncloned and cloned genome segment except for the 5′-half region of the 2007b ORF1 (Fig. 3B and 4). Although the precise reason for the discrepancy is unclear at present, it might be due to the cloning of the minor population of the 2007b quasispecies in the stool specimens. The overall good agreement of the results by the two sequencing strategies strongly suggests that the genome mosaicisms we found by analysis of the direct sequencing data were intrinsic rather than an artifact of the analysis. Taken together, these data indicate that the NoV mosaic genomes were present in the human stool specimens and that the ORF1/ORF2 junction region is the common hot spot for generating the mosaic genomes in GII/4 subtypes in nature.

Amino acid signatures of the NoV GII/4 subtypes.

We then investigated sequential characteristics of the proteins of the 7 GII/4 subtypes by searching for unique amino acid signatures in viral proteins. The deduced amino acid sequences of ORF1, ORF2, and ORF3 of a given subtype were aligned with reference sequences of the past GII/4 subtypes (38) that were identified before detection of the query subtype. Amino acids specific to the query subtype were extracted and referred to as amino acid signatures of the new epidemic subtype. In the case of the 2006b subtype, we also analyzed the changes in the signatures in the capsid protein VP1 between 2006 and 2009, because information on the structure and function is more abundant for the capsid than for other viral proteins.

The amino acid signatures of the 7 GII/4 subtypes were distributed throughout the three ORFs (Fig. 5, asterisks). 2004/5 and 2007a had more substitutions in ORF1 than the others because their ORF1s seemed to have originated with the GII/12 relatives (Fig. 3A and 4). When they were compared with the two available complete ORF1 sequences of GII/12, they still had many amino acid substitutions (27 and 63 for 2004/05 and 2007a, respectively). 2007b and 2008b had fewer substitutions in ORF2s and ORF3s than the others because these regions seemed to have originated from the 2006b relatives.

As seen in the 2006b variants in the 2006/2007 season (38), the capsid protein signatures were preferentially distributed on the P2 domain in other GII/4 subtypes (Fig. 5, blue box). All 7 capsid signatures identified in the 2006b variants in the 2006/2007 season were highly conserved during the 2006/2007 season, although two of them (P357 and N412) were gradually lost in the 2006b variant population during 2007 and 2009. Instead, other amino acid substitutions were sporadically accumulated in the P2 domain of the later 2006b variants (data not shown). The 7 signatures in the P2 domain were also well retained in the 2007b and 2008b subtypes, whose genomes had capsid gene segments from the 2006b relatives (Fig. 3B). These data indicate that (i) all of the 7 GII/4 subtypes had unique amino acid substitutions in viral capsid and replication proteins at the time of their outbreaks in Japan, (ii) the dominant 2006b subtype retained the capsid signatures during its persistence between 2006 and 2009, and (iii) some GII/4 subtypes acquired unique mutation sets of the 2006b capsid P2 domain by putative genome recombination events.

3-D locations of the subtype-specific amino acids in the capsid P domain dimer.

To clarify 3-D locations of the capsid signatures, we constructed structure models of the VP1 P-domain dimer of the GII/4 subtypes by the homology modeling method as described previously (38). The 2007b and 2008b models were not included for the study because their capsid proteins had no signature or a single signature in the P2 domain due to putative genome recombination with 2006b (Fig. 3, 4, and 5). The thermodynamically and sterically optimized structural models of the P-domain dimer of the 2004/05, 2006a, 2006b, 2007a, and 2008a subtypes showed no major differences in the folding of the main chains (Fig. 6). This result suggests that the capsid amino acid substitutions primarily influenced physicochemical properties around the substitution sites by changing the size and chemical properties of the side chains. These models were then used to map the 3-D locations of the P2 domain mutations.

Importantly, the capsid P2 domain signatures were mostly mapped on the outer surface loops in all of the GII/4 subtypes examined (Fig. 6). These loops form an accessible protein surface with which host proteins, such as a cellular receptor(s) and antibodies, can directly interact. The P2 mutations are often positioned near the putative functional sites for virus entry into the cells: the fucose ring binding sites formed by the P-domain dimer (8, 13) (blue-dotted ovals) and an RGD motif (60) on the β2 sheet of the P domain (cyan chain). Notably, the P2 mutations were more abundant in the widely prevalent subtypes, 2004/05 and 2006b, than in the others (Fig. 2 and 6). The 2008a subtype, which was detected most recently and caused NoV epidemics at multiple sites in the 2008/2009 season, also had 5 unique substitutions in the P2 domain (Fig. 2 and 6). These data indicate that the GII/4 subtypes that were dominant between 2004 and 2009 in Japan had a greater number of unique amino acids preferentially positioned on their capsid surfaces at the time of their first outbreaks.