Is Hepatitis Delta infections important in Brazil?

A total of 50 samples from patients in medical care in different regional clinics at the Amazon region of Brazil was collected between 2009 and 2010. The samples were originated from the states of Acre (Cruzeiro do Sul and Rio Branco Cities), Amazonas (Manaus City) and Rondônia (Porto Velho City), all located in the North region of Brazil. The samples were collected and transported in dried-blood spots (DBS 903- Whatman). The mean age of the patients was 32 years (maximum of 69 and minimum of 18), and 52 % were male.

The samples were processed only after the CEP/UNIFESP (# 72971) was approved, and analyzed anonymously to ensure the protection of the participants. Informed consent was not required because the samples had been previously collected and stored. Patients co-infected with HIV and/or HCV were excluded.

The sample RNA was purified by the method developed by Boom [24], and the cDNA were synthesized using SuperScript III reverse transcriptase (Invitrogen, Carlsbad, CA). HDV full genome was sequenced using the shotgun technique, which amplified five overlapping fragments named A - E. The PCR methodology used was based on the study performed by Nakano [25]. The primers used amplified all HDV genotypes, and the sensitivity of amplification was up to 10 ³ copies/mL. For the D fragment a rescue strategy was necessary, where a nested PCR was implemented, combining the primers 1267_F with F2 [26] in the 1st round and A3 with 138_R in the 2nd round [26, 27].

The processing of all samples included a negative control to exclude the possibility of cross contamination and a positive control to ensure the reliability of the reaction. Moreover, both strands of the PCR were sequenced and thereafter a consensus sequence was created, which was used in the analysis to guarantee the best quality of the results obtained.

Sequencing was performed using the ABI Prism 3130 Genetic Analyzer (Applied Biosystems, USA). The sequences were edited using the Sequencher v.4.2 Program (Gene Code, Ann Arbor, USA) and analyzed using Blastn http://blast.ncbi.nlm.nih.gov/Blast.cgi ), to verify their similarity with the database sequences.

The HDV references sequences were obtained from the GenBank database (http://​www.​ncbi.​nlm.​nih.​gov/​genbank), described in Table 1. All the sequences were aligned using ClustalX software version 2.1 [28]. After the first alignment, the sequences were manually edited using the SE Al program version 2.0 (http://​tree.​bio.​ed.​ac.​uk/​software/​seal/​).

Maximum likelihood trees and bootstrap values were obtained using PhyML software [29]. The HKY model and gamma distribution (Γ) were selected and, based on the likelihood ratio test (LRT), implemented in the jModeltest software [30, 31]. The resulting trees were created and edited using FigTree (http://​tree.​bio.​ed.​ac.​uk/​software/​figtree/​).

The RDP v.4 software was used to determine the presence of recombination in the sequences [32]. 3Seq performs an exact nonparametric test to detect recombination in triplets of sequences [33]. The maximum χ ² (MaxChi) is a method implemented by Maynard Smith [34], and it uses variable/invariable sites to detect recombination in pairs of sequences. The maximum match χ ² (Chimaera) is a modification of Smith’s method that uses variable sites to calculate the maximum χ ² match statistics [35]. Geneconv detects recombination by evaluating conserved substitutions in fragments between two sequences [36]. Although evolutionary methods are not explicitly implemented in Geneconv, it is robust and has low levels of false positive [37].

Initially, default parameters were used; then, these were optimized to avoid detection of false positive recombination. In addition, window sizes of 20 to 50 as well as Bonferroni correction were utilized.

A Bayesian Markov chain Monte Carlo (BMCMC) coalescent framework was used to estimate the ancestral genealogy, phylogeographic and time to the most common ancestor (tMRCA). After Markov chain runs enough so that the parameter space have been sufficiently visited, the mean and the 95 % HPD intervals of each parameter was calculated. The HKY model plus a gamma correction (Γ) was applied to all analyses; the evolutionary and demographic parameters were iteratively adjusted. We used the Bayesian discrete methods to estimate the node probability in regard to the geographical location of samples [38].

In order to explore the correlation between sample location and tree topology, the Bayesian discrete model [38] was used because it allows the reconstruction of ancestral location states while accounting for the uncertainties of maximum posteriori (MP) tree topology and mapping states at nodes.

The Bayesian skyline plot method (BSP) that provides an unbiased description of genetic diversity over time was used (expressed as the product of the effective sample size Ne and generation time τ). BSP contemplates a wide varied of demographic scenarios without formally assuming any determined model [39]. Since absolute time was used (years) to scale branch lengths and a specific generation time was not assumed, our estimates of Ne.τ will reflect only the relative genetic diversity of the infections over time.

The Markov Chain Monte Carlo (MCMC) processes were run for 3x10⁸ generations with the initial 10 % of each run discarded as burn in. The convergence of chains was evaluated using the TRACER software, version 1.5 (available at: http://​beast.​bio.​ed.​ac.​uk/​), runs were accepted when all parameters presented the effective sample size number (ESS) greater than 200. Two independent chains were run for each dataset and combined with LogCombiner software [39, 40].

The MP tree was inferred using the BSP method and the maximum clade credibility parameter were annotated using the TreeAnnotator software [39, 40]. All these analyses were performed with the BMCMC approaches implemented in the BEAST package version 1.8.0 [40].

The codon based maximum likelihood method was used to estimate the selective pressures on the PTV sequences. This approach estimates the likelihood of distinct models of codon evolution and computes the ratio (dN/dS = ω) of the number of non-synonymous (dN) and synonymous (dS) substitution rates between sites considering the phylogenetic relationships of the sequences [41, 42]. To estimate the overall mean ratio, the one ratio model (M₀) was used. This model assumes a single ω for all sites in the alignment and is the simplest model. The hypothesis of positive evolution is based on the likelihood ratio test (LRT) comparing pairs of nested models, with the corresponding statistics approximated by a chi square (χ ²) distribution. To detect positive selection, we compared the null hypothesis of model M1, which allows for different proportions of conserved (ω = 0) and neutral codons (ω = 1), against the alternative hypothesis of model M2, which has an additional class of codons with ω ratio > 1, thus assuming positive selection. These models were implemented in the CODEML program of the PAML package, version 4.8, available at http://​abacus.​gene.​ucl.​ac.​uk/​software/​paml.​html.

Out of a total of 50 samples, 46 demonstrated the necessary quality for further analysis. The demographic information and the GenBank ID of each sample used in the analysis are detailed in Table 2 (present in the end of the manuscript). A maximum likelihood tree was initially constructed using the full length genomes of 46 samples and reference genomes of the main genotypes. Figure 1 shows that all sequences generated in the current study belong to genotype 3 and clustered within a single sequence with a bootstrap proportion of 97. Although the samples belong to genotype 3, they formed two groups: one composed of isolates from Yucpa Indians in Venezuela [43] and another composed of Brazilian isolates plus one isolate from the Yavari River in the Amazon basin close to the Peru Brazil border [19]. All sequences used to infer the tree mentioned above lacked a recombination signal.

42 samples were used for this analysis, four of the 46 original sequences were excluded because they had significant deletions along the genome. The phylogeography of HDV-3 isolates (Fig. 2), show that Brazilian samples clustered into four clades (named A, B, C, D), one sample from Peru isolated in 1986 is located at the base of these clades. The MP tree presents two large clusters, A and B, containing samples from different geographic locations (Rio Branco, Porto Velho, Cruzeiro do Sul and Manaus) and two small clusters, C and D, with sequences from Rio Branco and Porto Velho, respectively.

The topology also indicates two sets of nearly identical isolates: one from Rio Branco and another from Porto Velho (filled triangles in Fig. 2). The branching pattern of intermingled samples from Rio Branco and Porto Velho suggests a bidirectional flow of infections between these two locations.

The pattern of geographical dispersal of HDV-3 is depicted by the highest location probability on the nodes of the MP tree (Fig. 2; the location is indicated by colored branches and the location probability by numbers at the nodes). Although the multi-location clades A and B indicate that Rio Branco contributes significantly to the geographical spread of HDV-3, this may occur because samples from this location were overrepresented. When random sets of sequences containing equivalent numbers of samples from Rio Branco and Porto Velho were evaluated, the location probability to support Rio Branco at internal nodes decreased to values below 0.7 (data not shown).

Figure 3 shows the reconstructed dynamics of the genetic diversity (Ne.τ, y axis) plotted against time in years (x axis). The plot shows the mean (thick solid line) and the 95 % highest density interval (thin solid lines). The shape of the plot demonstrates a period of exponential growth that began in the late 1970s and lasted until 1995. The estimated tMRCA values for the HDV-3 and for the Amazon clade were 1934 (1927 until 1940) and 1952 (1948 until 1955), respectively.

A systematic codon based analysis was performed on HDV isolates using the 215 codons from the L-HDAg region. In Table 3, the estimates of the codon based analysis are summarized. To detect positive selection models M7 and M8 were compared using likelihood ratio test assuming two degrees of freedom between models. Initially an intra-subtype analysis was performed using a dataset with 47 isolates of HDV-3, 46 from this study and one reference sequence from Peru (L22063), results of Model M8 indicated 24 codons in the L-HDAg under positive selection with mean ω values of 5.04.

When this same analysis was performed with other HDV genotypes (78 sequences) the Model M8 indicated only nine codons under positive selection, four of which were similar with the result demonstrated with the HDV-3 genotype (8 K, 9 K, 37 T and 120S).