Evidence of frequent recombination among human adenoviruses

Adenoviruses are ubiquitous, non-enveloped, double-stranded DNA viruses. In humans, they are a frequent cause of minor upper respiratory-tract infections, being responsible for about 5 % of acute respiratory disease in infants (Horwitz, 2001), and a rare cause of life-threatening infections in immunosuppressed patients. Human adenoviruses belong to the genus Mastadenovirus and are represented by six species (formerly groups), Human adenovirus A–F (AdA–AdF), which are classified further into 51 serotypes (Shenk, 2001). Serotyping of adenoviruses is traditionally performed by neutralization test. As the neutralization epitopes are located mainly within the hexon protein, sequencing of the hexon gene gives results almost indistinguishable from those of traditional serological assays (Madisch et al., 2005).

Recombination is a well-known feature of adenovirus genetics. Experimental recombination was pursued extensively in the 1970s (Grodzicker et al., 1974; Mautner et al., 1975; Williams et al., 1975). In the absence of extreme selection in vitro, recombination occurs only between strains of the same species, predominantly in regions of homology (Boursnell & Mautner, 1981; Mautner & Boursnell, 1983). Several adenovirus prototype strains were found to be interspecies recombinants (Ebner et al., 2005), but such events are seemingly rather uncommon. Illegitimate recombination (insertion and deletion) was suggested to be one of the major means of adenovirus evolution in AdD (Crawford-Miksza & Schnurr, 1996). Natural recombination and reassortment are key evolutionary mechanisms in most RNA viruses. In ubiquitous respiratory viruses, such as influenza viruses (Hay et al., 2001) or enteroviruses (Lukashev, 2005), frequent co-infection facilitates recombination so commonly that genome fragments have independent evolutionary histories even on a timescale of years. Both the possibility of recombination and the considerable chance of co-infection in adenoviruses are well known, but the true extent of natural recombination among adenoviruses has remained obscure.

The wealth of early studies on adenovirus replication, relative ease of genome manipulation and apparent genetic stability made adenoviruses one of the most attractive early vectors for gene therapy and cancer therapy (Young et al., 2006). Multiple clinical studies of adenovirus vectors are currently ongoing (Breyer et al., 2001; Palmer et al., 2002) and a plethora of improved vectors are being developed. Most gene-therapy vectors are based on E1-deleted forms of Ad5. Production of such replication-incompetent viruses is carried out in cell lines with integrated E1 genes. Recombination of the adenovirus vector with the integrated E1 genes in the host cell, which restores the replication competence of the virus, is a common and well-characterized event that can only be avoided by rigorous exclusion of any homology between the viral and cellular sequences (Fallaux et al., 1998; Hehir et al., 1996; Lochmuller et al., 1994). The main limitation to adenoviral therapy for cancer has been the low activity of the vectors. Hence, current interest centres on making the vectors more active, for example by expressing prodrug-activating enzymes or by modifying the capsid proteins to alter the tropism. Regulatory bodies ask for evidence that the genome of recombinant viruses is stable, but this question is rarely framed in the context of a deep understanding of the stability of natural adenoviral strains. We have used a range of phylogenetic techniques to analyse the stability of the prototype AdB, AdC and AdD strains, as well as 16 sporadic AdC field isolates. We provide evidence that recombination between strains of the same species is not only a very common event, but also one of the major driving forces in adenovirus evolution.

AdC strains used in this work (Table 1) were sporadic isolates from stool samples and, in one case, from a sewage specimen collected in accordance with the WHO Polio Eradication Initiative in Russia and New Independent States. Two cell-culture lines were used for isolation – RD (human rhabdomyosarcoma) and (mainly) HEp2 (a HeLa-derived cell line). Most strains underwent two to four passages before being used for this study. Isolates were identified provisionally as adenoviruses by cytopathic effect morphology. Viral DNA was extracted from infected cells by using a standard phenol/chloroform protocol. Human AdC strains were identified initially by sequencing of a fragment of the polymerase (Pol) gene, using oligonucleotides 7280F and 7900R (see Supplementary Table S1, available in JGV Online). Five genome regions were then amplified by PCR using the oligonucleotides shown in Supplementary Table S1. For a fragment of the E3 genome region, only the size of PCR products was analysed. The resulting DNA sequences (GenBank accession numbers EU192298[GenBank] –EU192361[GenBank] ) for other genome regions were aligned with the corresponding fragments of the complete sequences of human adenovirus type 1 (GenBank accession no. AF534906[GenBank] ; Lauer et al., 2004), type 2 (GenBank accession no. J01917[GenBank] ; Roberts et al., 1986) and type 5 (GenBank accession no. M73260[GenBank] ; Chroboczek et al., 1992) prototype strains. The complete sequence of Ad5 isolate NHRC Ad5FS 7151 (GenBank accession no. AY601635[GenBank] ; Lin et al., 2006) was not used because it was very similar to the prototype Ad5 ATCC strain in all genome regions. Other complete sequences used were Ad3 GB (GenBank accession no. AY599834[GenBank] ), Ad3 NHRC 1276 (GenBank accession no. AY599836[GenBank] ), Ad7 Gomen (GenBank accession no. AY594255[GenBank] ), Ad7 NHRC 1315 (GenBank accession no. AY601634[GenBank] ), Ad9 (GenBank accession no. AJ854486[GenBank] ), Ad11 (GenBank accession no. NC_004001), Ad14 (GenBank accession no. AY803294[GenBank] ), Ad16 (GenBank accession no. AY601636[GenBank] ), Ad17 (GenBank accession no. NC_002067[GenBank] ), Ad34 (GenBank accession no. AY737797[GenBank] ), Ad35 (GenBank accession no. AC_000019), Ad46 (GenBank accession no. AY875648[GenBank] ) and Ad49 (GenBank accession no. DQ393829[GenBank] ). Partial E1A and Pol sequences of the prototype Ad6 strain Tonsil 99 (obtained from the ATCC; catalogue no. VR-1083) were identified in this work, whilst hexon and fiber sequences of Ad6 were acquired from GenBank [accession numbers DQ149613[GenBank] (Ebner et al., 2005) and AB125751[GenBank] (Adhikary et al., 2004), respectively]. A large part of the complete sequence (nt 18838–33452) was identified for strain #16700 and for the Ad6 ATCC strain. Nucleotide positions throughout the work are given in accordance with the Ad2 complete sequence (GenBank accession no. J01917[GenBank] ). The sequences of the oligonucleotides used for this purpose are available upon request.

Table 1. Adenovirus strains used in this work

To facilitate analysis of the #16700 strain, a full-length bacterial artificial chromosome clone was produced by gap repair in yeast, as described by Gagnebin et al. (1999). The homology arms in the recipient vector (pMB19; Gagnebin et al., 1999) contain short Ad5 sequences (<300 nt).

The sequences were aligned by using CLUSTAL_X 1.81 software (Thompson et al., 1997). Fiber region nucleotide sequences that had a significant number of insertions and deletions were translated by using BioEdit 7.0.5.2 software (Hall, 1999) and the protein sequence was aligned by using CLUSTAL_W (Thompson et al., 1994). Phylogenetic trees were created with CLUSTAL_X (neighbour-joining algorithm, Kimura substitution model), using the exclude positions with gaps and correct for multiple substitutions options. Larger alignments were also analysed by using SimPlot software v. 2.5 (). Similarity plots were created with window size of 500 nt, which provided a good balance between lowering the noise and showing a good number of putative recombination events. Bootscan analysis (Salminen et al., 1995) was run with a neighbour-joining tree algorithm, Kimura distance model and 1000 pseudoreplicates. The trees were drawn with the NGRAPH module of CLUSTAL_X, and the in-tree comments were added in CorelDraw 12. Mean similarity plots were created by calculating mean nucleic sequence distance within a sliding 500 nt window, using the DNADIST module of the PHYLIP software package (Felsenstein, 1989) operated via Perl script. The graphs were plotted in Microsoft Excel.

Virus titration was performed by plaque assay in HER911 cells (Fallaux et al., 1996), kindly provided by Professor P. Beard (Swiss Institute for Experimental Cancer Research, Lausanne, Switzerland). Serotyping by plaque neutralization test was carried out with rabbit type-specific antisera, kindly provided by Professor W. Russell (University of St Andrews, UK). Virus (10–50 p.f.u.) was incubated for 1 h at 37 °C with serial twofold dilutions of serum in Dulbecco's modified Eagle's medium (Gibco); medium without any supplements was used as a control. Each virus tested was challenged separately with four type-specific anti-AdC antisera. Neutralizing titre was identified as the serum dilution that neutralized approximately 50 % of the p.f.u.

We analysed the phylogenetic relationships of four AdC prototype strains and 16 AdC field isolates in four genome regions. The fragment of the hexon gene studied (nt 19198–19930) includes the hypervariable loops, which carry the main neutralizing epitopes that define the serotype (Madisch et al., 2005). The phylogenetic grouping in this genome region was therefore used to identify the serotype of the strains studied here (Table 1). All serotype groups were supported with 100 % bootstrap values (Fig. 1a), and the nucleotide sequence difference between serotypes was 31–47 % (Kimura model). Strains of the same serotype differed by <5.5 %, and some isolates of the same serotype were almost identical to each other and to the corresponding prototype strain isolated in the 1950s. For example, among eight Ad2 strains studied here, there were 17 positions with nucleotide variations, yet only one of them resulted in a protein sequence alteration. Therefore, fixed mutations were predominantly neutral, and the protein sequence within a serotype was conserved even in the hypervariable regions.

(22K): Fig. 1. Phylogenetic relationships of AdC field isolates and ATCC strains in four genome regions: (a) hexon (nt 19198–19930); (b) fiber (nt 31090–31875); (c) polymerase (nt 7340–8466); and (d) E1A (nt 145–864). Neighbour-joining trees using the Kimura distance model are shown. Numbers at tree nodes indicate percentages of bootstrap pseudoreplicates that supported the group below. Bars indicate branch length as nucleotide sequence distance per nucleotide position and as a discrete number of nucleotide changes (c and d only).

Strain #16700 differed from every other adenovirus strain studied by 29–41 %, i.e. by about as much as distinct adenovirus serotypes differ from each other, and it did not group reliably with any of the AdC serotypes. We therefore suggest that this strain represents a novel serotype of AdC. Strain #16700 was challenged in a neutralization test with antisera to all four known serotypes of AdC (Table 2). All antisera were able to neutralize the adenovirus strain of the corresponding serotype specifically. Strain #16700 showed some cross-reactivity with Ad6 antiserum, but the titre was four to eight times lower than the titre against authentic Ad6. This result could be explained by the slightly higher similarity of this strain to Ad6 than to any other AdC serotype in part of the hexon gene (Fig. 2a), but it seems more likely to be due to antibody binding to minor neutralization epitopes in the fiber, which is similar to the Ad6 fiber (see below).

Table 2. Neutralization test for strain #16700

(62K): Fig. 2. (a, b) Similarity plot (a) and bootscan graph (b) of the partial genome sequence (nt 18838–33452) of strain #16700 compared with AdC ATCC strains. Window, 500 nt; step, 30 nt. The Kimura model with Jukes–Cantor correction and neighbour-joining tree algorithm were used. (c, e) Mean nucleotide sequence difference plots (Kimura model) for AdC (c), AdB1, AdB2 and AdD (e). Window, 500 nt; step, 20 nt. (d, f, g, h) Bootscan graphs of Ad2 to AdC strains and Ad7 Gomen (d), Ad17 to other AdD (f), Ad7 Gomen to other AdB1 strains (g) and Ad14 to other AdB2 viruses (h). Window, 1000 nt; step, 100 nt. The Kimura model with Jukes–Cantor correction and the neighbour-joining tree algorithm were used. The dotted line shows a 70 % bootstrap value cutoff, which is generally considered reliable.

In the fiber region (nt 31090–31875), the phylogenetic grouping corresponded almost perfectly to that for the hexon region (Fig. 1b). Apart from strain #16700, no viruses had mismatched hexons and fibers. As in the hexon region, there was a huge sequence difference (21–38 %) between strains of different serotypes, but <0.9 % difference between strains of the same serotype, which mainly resulted from synonymous substitutions. Ad2 strains, for example, could differ in up to 10 nucleotide positions, yet only one of them resulted in amino acid substitution. The fiber of strain #16700 was indistinguishable from Ad6 fibers. This was the only apparent example of recombination between the fiber and hexon regions among the 16 strains studied.

We also analysed a fragment of the viral polymerase gene (nt 7340–8466). This genome region was much more conserved, with <1.7 % variation in nucleotide sequence between the AdC strains studied. Importantly, of 49 nucleotide positions where substitutions were observed, only 10 resulted in amino acid substitutions. Therefore, a majority of these substitutions probably resulted from random fixation of neutral mutations (genetic drift) rather than adaptive evolution. Hence, the phylogenetic signal in this region should reflect ancestry rather than convergence. The phylogenetic grouping of the polymerase region was not concordant with those for the hexon and fiber genome regions (Fig. 1c). Strains of different serotypes were shuffled randomly, with many reliably observed groups uniting strains of different serotypes. In several cases, field isolates of different serotypes were identical in the polymerase region and distinct from both prototype ATCC strains, e.g. strains Ad2 17676 and Ad1 22415, or strains Ad6 14555, Ad2 20608 and Ad5 24235. This phylogenetic pattern is best explained by intertypic recombination in natural AdC populations.

In order to verify the results for the polymerase region, we analysed a fragment of the E1A region (nt 145–864) that included both the E1A promoter and a part of the E1A coding sequence. Substitutions were found in 28 sites, including 13 sites with substitutions in the E1A coding sequence, of which nine resulted in amino acid substitutions. Due to the high degree of conservation, phylogenetic analysis of this region (Fig. 1d) did not produce high bootstrap values. Nevertheless, we could confirm our observations for the polymerase region. The phylogenetic grouping did not correspond to the serotype (i.e. grouping in hexon region), and several reliably observed groups included strains of different serotypes. Similar to the polymerase region, isolates of different serotypes could be identical, and distinct from their corresponding prototype strains. In some cases, the phylogenetic relationships were similar in the E1A and polymerase regions, whilst in many other cases they were disturbed.

The conflicting phylogenetic relationships of the adenovirus field isolates in different genome regions strongly suggest recombination in their phylogenetic history. As strain #16700 seemed the clearest case of recombination and defines a putative novel serotype, we sequenced a large fraction of the genome (nt 18838–33452). As the complete Ad6 sequence is not available in GenBank, we also sequenced the corresponding part of the Ad6 genome. Similarity plots and bootscan graphs (Salminen et al., 1995) were used to analyse the relationship of strain #16700 to the four prototypic AdC strains. Both approaches showed that strain #16700 was a mosaic recombinant relative to the prototypic strains (Fig. 2a, b). Discrete parts of the strain #16700 genome were more similar to (Fig. 2a) and grouped reliably with (Fig. 2b) different AdC strains. The exception was the hypervariable part of the hexon gene in strain #16700, which was not similar to that of any other AdC strain (Fig. 2a, arrow). Importantly, many of the crosses were obvious not only on the bootscan graph, but also on the similarity plot. Recombination crossover regions could not be linked to distinct genes or transcription units.

Study of phylogenetic relationships of AdC field isolates suggested rather frequent recombination. We examined complete genome sequences of other adenovirus species to seek support for our findings. These viruses are very similar (>97 % on average) over most of the genome, but very different in the hexon, fiber and E3 regions (Fig. 2c). Interestingly, there was a very sharp border between conserved and variable genome regions. There are only three complete sequences of ATCC AdC strains available; therefore, we added the Ad7 Gomen (AdB) sequence to the alignment to facilitate bootscan analysis of Ad2 versus the other AdC serotypes (Fig. 2d). The multiple flips in the grouping pattern are probably explained by frequent recombination within species, especially as the large window size used (1000 nt) minimized the risk of putative adaptive changes affecting the grouping pattern. It should be noted here that regions of Ad2 that grouped reliably with Ad7 indicate only that, in these genome regions, Ad1 and Ad5 grouped together very reliably, forcing Ad2 to group with Ad7.

To test whether the observation of very incongruent phylogenetic relationships within species is unique to AdC, we analysed the available complete AdB and AdD sequences for signs of recombination. AdB is divided into subspecies B1 and B2 (Wadell et al., 1980), which show partially overlapping tropism at the level of the cellular receptor (Segerman et al., 2003). Similarity plots for AdB strains (data not shown) suggested that the subspecies were distinct over the entire genome; we therefore analysed them separately. The results were very similar to what was observed for AdC. The complete sequences of viruses of the same (sub)species were very similar over most of the genome, with the exception of the hexon, fiber and E3 regions in AdD, plus, to a lesser extent, the whole late region in AdB1 and the hexon, fiber and penton base regions in AdB2 (Fig. 2e). Again, as observed in AdC, there was a sharp transition between similar and variable genome regions. Bootscan graphs for AdD, AdB1 and AdB2 groups (Fig. 2f–h) bore multiple points of abrupt change of the most likely grouping over the genome. Analysis of complete AdB1, AdB2 and AdD genomes using the informative-sites test implemented in SimPlot and several algorithms implemented in the RDP v. 2.0 package (Martin et al., 2005), such as RDP, MaxChi and CHIMAERA, also provided clear evidence of multiple recombination events distributed over the genome. We do not show the data here, as they could simply confirm the bootscan results shown in Fig. 2, and apparent recombination breakpoints often reflect not true recombination spots, but regions with a sufficiently high phylogenetic signal to detect a significant phylogeny conflict.

Recombination is a hallmark of adenovirus genetics, but a dearth of sequence information limited the scope of previous studies. In this work, we investigated the phylogenetic relationships of 16 AdC field isolates in four genome regions. Multiple phylogenetic conflicts, the gold standard of recombination studies, indicate that all 16 field strains are recombinant relative to the ATCC prototype strains and to each other. The fact that most of the phylogenetic signal in the hexon, fiber and polymerase genes comes from silent mutations indicates that our results represent true phylogenetic relationships of the strains studied. Analysis of strain #16700, probably a representative of a novel serotype, suggests multiple recombination events relative to other prototype AdC strains. Further analysis of the available complete sequences of AdC, AdB1, AdB2 and AdD strains conducted with bootscanning, a common tool used to detect recombination events in viruses (Lukashev et al., 2005; Oberste et al., 2004; Salminen et al., 1995), also provided evidence of multiple intertypic recombination events in their phylogenetic history (Fig. 2d, f, g, h). Taken together, this shows that natural recombination is a very common event among circulating adenoviruses. Such recombination requires frequent co-infection, which is plausible in cases of long-term adenovirus persistence in the tonsils (Garnett et al., 2002) or persistence of adenoviruses after acute infection (Adrian et al., 1988).

According to mean sequence similarity plots (Fig. 2c, e), prototype adenovirus strains are very similar to each other (about 97 % nucleotide sequence similarity) over most of the genome, but differ by 20–60 % in hexon, fiber and some other genome regions. These observations were made only for the few complete sequences available, but they agree well with those of Bailey & Mautner (1994) and with our results for 16 AdC field isolates. The rate of genetic drift in adenoviruses was previously reported to be low, yet appreciable (of the order of 1 % nucleotide sequence in 200 years; Crawford-Miksza & Schnurr, 1996). The number of sites with possible silent mutations in the Ad2 hexon region was similar to that in the polymerase gene; we could therefore expect a comparable fixation rate of spontaneous mutations in these genome regions. In the absence of recombination, strains of the same serotype would have drifted away from other serotypes and made a distinct phylogenetic group all over the genome. As we observed, viruses of different serotypes can be identical or very similar in the polymerase and E1A genome regions (Fig. 1c, d), and the general DNA sequence variability within the species over most of the genome is rather low (Fig. 2c, e). Our results suggest that frequency of recombination within a species is so high that it maintains the common species consensus over most of the adenovirus genome. More variable genome regions escape this constraint, presumably due to incompatibility of more divergent proteins. If this model is correct, it implies that the rate of natural intertypic recombination in adenoviruses is at least equal to or higher than the spontaneous mutation fixation rate. Unfortunately, due to the generally slow genetic drift in adenoviruses, it is currently not possible to quantify the frequency of intertypic recombination, which could have taken place over decades or millennia. The fact that adenoviruses from different countries are equally involved in recombination and maintenance of the species consensus suggests that adenovirus species exist as a global, worldwide reservoir of genetic information, with free exchange of genes between viruses of different serotypes within the same species. Speaking in more common terms, sex in adenoviruses is not only a common event, but the mechanism of species maintenance.

The phylogenetic and serological properties of strain #16700 suggest that it may be a novel serotype of AdC. Analysis of a partial genomic sequence provided evidence of multiple recombination events in its phylogenetic history. Importantly, some of the recombination events mapped to the hexon region, albeit outside the hypervariable loop region. It has been suggested that adenovirus serotypes emerge as a result of illegitimate recombination and polymerase slippage in polypurine regions (Crawford-Miksza & Schnurr, 1996). Compared with other AdC serotypes, strain #16700 shows many amino acid substitutions, but few insertions or deletions in the hexon. We would thus give greater weight to legitimate recombination and point mutation than to polymerase slippage in the evolution of adenovirus serotypes.

We observed almost no isolates with mismatched hexon and fiber genes, which implies that the hexon and fiber proteins from different serotypes are incompatible, at least in nature. A wealth of experimental evidence suggests that viruses with fibers from different serotypes and even different species are viable (Kangasniemi et al., 2006; Ni et al., 2005; Stoff-Khalili et al., 2005). Whether these viruses will be stable in the long term is an open question but, based on our results, it would be prudent to maintain the concordance of hexons and fibers in gene-therapy vectors. It remains unclear how such a perfect match of fibers and hexons is maintained in AdC. There is no direct interaction between these proteins in the viral capsid, as they are linked via the penton base protein. Mean similarity graphs for AdB and AdD (Fig. 2e) revealed elevated variability in the penton base gene, which was not observed in AdC. In these species, penton base could be serotype-specific and thus require a perfect interaction with the corresponding hexon and fiber. The exact reason that fibers and hexons are so well-matched in AdC remains obscure.

Frequent natural recombination within serotypes has important practical implications for gene therapy and cancer therapy. Thanks to the excellent safety record of adenovirus therapy, many groups are trying to increase the activity of oncolytic vectors by mutating the promoters, expressing transgenes or modifying the capsid. Based on our work, we suspect that there is a significant risk of recombination when such a vector meets a genetically similar wild-type strain in a patient.

We thank Professor W. Russell for providing antibodies and for critical reading of the manuscript, and the University of St Andrews for financial support. We also thank the EU fp6 Theradpox STREP for financial support. We are also grateful to both anonymous reviewers for very helpful comments.