Characterization of small genomic regions of the hepatitis B virus should be performed with more caution

Here, 57 plasma samples were collected from 57 patients co-infected by HBV and HIV-1 in Guangdong Province from 2007 to 2011. The plasma samples were stored at − 80 °C until use. HBV DNA was extracted from plasma samples with the QIAamp DNA Mini Kit (Qiagen). The DNA sample served as a template for HBV DNA amplification by nest-PCR with PrimerStar PCR kit (Takara). Two fragments which could cover full-length of HBV genome were amplified, including long fragment (L fragment, from 1848 to 1807, 3174 bp) and short fragment (S fragment, from 1603 to 2327, 725 bp). All primers were listed in Table 1. The nest-PCR protocol of L fragment was performed as following: 95 °C 2 min, 95 °C 30s, 60 °C 30s, 72 °C 3 min 20s, 30 cycles, in the 1st round of nest-PCR, and then 95 °C 2 min, 95 °C 30s, 60 °C 30s, 72 °C 3 min 20s, 35 cycles, in the 2nd round of nest-PCR. The nest-PCR protocol of S fragment was performed as following: 95 °C 2 min, 95 °C 30s, 60 °C 30s, 72 °C 30s, 30 cycles, in the 1st round of nest-PCR, and then 95 °C 2 min, 95 °C 30s, 60 °C 30s, 72 °C 30s, 35 cycles, in the 2nd round of nest-PCR.

PCR products were purified and then sequenced directly. Among these 57 complete sequences, initial jumping profile hidden Markov model (jpHMM) analysis indicated one sequence from a 33-year-old woman in 2016, is a novel B/C/D intergenotype recombinant. It is the first 3-genotypes recombinant discovered in China. Multiple recombination and phylogenetic analyses were performed to confirm this. The HBV genomic DNA is 3215 base pairs in length. It was submitted to GenBank (GenBank accession number: KY417926).The other 56 nucleotide sequences reported in this study were deposited in GenBank under accession numbers MG571321–MG571376.

The usual performance of recombination detection on circular viruses is that the circular genomes are first manually linearized and then a linear model is used [6, 23, 32, 38, 41]. A major defect of this strategy is that dependencies between nucleotides at the 5′ and 3′ end of a sequence cannot be modeled [32]. It is known that when linear sequences of an alignment are analyzed as though they were circular and some recombination is detected, it becomes possible to detect a strong recombination hotspot spanning the beginning and end of the analyzed sequences [20, 21]. In contrast, when circular genomes are analyzed as though they were linear, recombination breakpoints occurring closely to the 5′ or 3′ ends of the linearized sequence may be overlooked [32].

The detailed recombination analysis is performed as previously described [14‐17, 43, 46]. Both the jpHMM and recombination detection program (RDP4) used in the current work take the circularity of the HBV genomic sequences into consideration and can facilitate accurate predictions of recombination breakpoints even close to the 5′or 3′end of the linearized sequence [21, 32]. As demonstrated in a previous study [16], jpHMM and RDP can easily produce highly accurate and impartial recombination data.

JpHMM was first used to perform recombination analysis. This tool is very intelligent and can produce a genome mosaic map with position numbers given either in the original sequence or relative to the HBV reference genome AM282986. The recombination prediction in jpHMM is based on a precalculated multiple sequence alignment of the major HBV reference genotypes, and the evaluation of its prediction accuracy showed that it to be more accurate than the competing methods used for phylogenetic breakpoint detection [32, 33, 46].

To confirm the data obtained by jpHMM, another recombination analysis tool, RDP4, was used for further analysis. RDP4 is a software package suitable for statistical identification and characterization of recombination events in nucleotide sequences [20, 21]. RDP4 is also very intelligent and utilizes a range of non-parametric recombination detection methods simultaneously: RDP, GENECONV [28], BOOTSCAN [22, 31], MAXCHI [29, 39], CHIMAERA [29], SISCAN [10], 3SEQ [5], and LARD [13], which can greatly increase sensitivity. Here, the sequences were set to linear. The highest acceptable P-value was set to 0.05. The other parameters are default RDP4 settings. To ensure reliability, the HBV sequence were considered recombinant when the recombination signal was supported by at least 4 methods with P-values of ≤0.05 after Bonferroni correction for multiple comparisons implemented in RDP4 [17, 20, 36]. The breakpoint positions inferred were manually checked using recombination signal analysis implemented in RDP4. Details regarding the methods and algorithms of the recombination analysis tools used here are given in the comprehensive list of recombination analysis software maintained by the Robertson Lab (http://​www.​bioinf.​manchester.​ac.​uk/​robertson/​recombination/​programs.​shtml).

To further confirm the results of the recombination, maximum likelihood phylogenetic trees (ML) were constructed based on the inferred breakpoint locations using the PhyML 3.0 implemented in RDP4 [11]. Automatic model selection with PhyML3.0 was used to find the best-fitting model of nucleotide substitution. Tree topologies were searched using NNI and SPR procedure. The confidence of each node in phylogenetic trees was determined using the bootstrap test with 1000 bootstrap replicates. The final ML trees were visualized using Mega 6 [42].

The clear and unified definition of genotype representatives is important to both recombination analysis by RDP4 and subsequent confirmation by phylogenetic analysis. All reference sequences are selected based on previous works and HBVdb [6, 12]. These sequences were all further verified via jpHMM to ensure the consistency. Inconsistent values were all removed from the references alignment.

Here, 7.02% of the sequences (4 out of 57) were identified as genotypes C by jpHMM. 91.23% of the sequences (52 out of 57) were identified as BC intergenotypic recombinants, i.e. genotype Ba. The 52 BC recombinants all had a fragment of genotype C over the precore region plus the core gene. This type of HBV has been previously discovered and is very common in southern China. Specifically, one sequence is typically identified as a 3-intergenotype recombinant comprising the B, C, and D genotypes. Although there are multiple genotypes co-circulating in China, such as B, C and D, a 3-genotype recombinant has never been found.

The jpHMM analysis of the strain obtained in the current work (GenBank accession number: KY417926) indicated clearly that it is a recombinant related to B, C, and D genotypes. The breakpoints clearly located at 1899, 2296, and 2526, respectively in the original sequence. The genotype assignments are supported by significant posterior probabilities (Fig. 1a, Table 2). Recombination of HBV between 2 genotypes is very common. However, recombinants involving more than 2 genotypes are rare [2]. To date, “genotype” I has been shown to be related to 3 genotypes: A/C/G [44].

In order to confirm the results of jpHMM, the 3 sub-regions delimited by the jpHMM-derived breakpoints in the strain were used to construct an ML tree with genotype references, respectively. The computed GTR + I + G4 model by PhyML3.0 was suitable for all trees. The tree of the B genotype region (spanning 1–1898, 2526–3215) clearly showed the clustering of the fragment with the B references (Fig. 1b). The tree of the region spanning 1899 to 2295 clearly showed the clustering of the fragment with the C references (Fig. 1c). Both these results are consistent with the jpHMM results. However, an apparent conflict became visible in the determination of the genotype assignment of the region spanning 2296 to 2525 (230 bases). Despite the low bootstrap values, the tree clearly showed that the fragment, unexpectedly, did not cluster with D references but rather with the B references (Fig. 1d).The topology of the tree was further confirmed by another round of phylogenetic analysis using the MrBayes tool implemented in RDP4 based on Bayesian inference [30].As shown in Additional file 1: Figure S1, the Bayesian tree showed the same results with respect to the assortment of D fragment.

To investigate this discrepancy, another recombination analysis tool, RDP4, was used to further explore the mosaic structure of the strain. The unique tool simultaneously utilizes a range of non-parametric recombination detection methods and thus has increased sensitivity and reliability. Significantly different from the result from jpHMM, RDP4 analysis clearly indicates that the strain is a B\C recombinant, excluding any other genotypes, with the breakpoints located at 1821 and 2199 in the original sequence (Fig. 2a, Table 3). This recombination event is supported by all 8 recombination detection methods (P-values are listed in Additional file 1: Table S1). Subsequent confirmations of genotype assignment by phylogeny are consistent with the RDP4 results. The major parent clustered with B references and 1000 replicates supported (Fig. 2b). The minor parent clustered with C references and 977 replicates supported (Fig. 2c). Thus, 2 major recombination tools have been used, variously displaying the recombination pattern of the HBV strain. Given that 2 of the 3 methods (jpHMM, RDP4, and the subsequent phylogenies) indicated a similar recombination event, the strain identified in the current work is assigned to a B/C recombinant, i.e. Ba, with C fragment spanning 1899 to 2295 (jpHMM) or 1821 to 2199 (RDP4). The breakpoint positions relative to reference genome AM282986 numbering are given in Additional file 1: Table S2–S3. All the other 56 stains were also further confirmed by RDP4 and the results are accordance with those from jpHMM.