Productive Homologous and Non-homologous Recombination of Hepatitis C Virus in Cell Culture

Genetic recombination is an important mechanism for increasing diversity of RNA viruses, and constitutes a viral escape mechanism to host immune responses and to treatment with antiviral compounds. Although rare, epidemiologically important hepatitis C virus (HCV) recombinants have been reported. In addition, recombination is an important regulatory mechanism of cytopathogenicity for the related pestiviruses. Here we describe recombination of HCV RNA in cell culture leading to production of infectious virus. Initially, hepatoma cells were co-transfected with a replicating JFH1ΔE1E2 genome (genotype 2a) lacking functional envelope genes and strain J6 (2a), which has functional envelope genes but does not replicate in culture. After an initial decrease in the number of HCV positive cells, infection spread after 13–36 days. Sequencing of recovered viruses revealed non-homologous recombinants with J6 sequence from the 5′ end to the NS2–NS3 region followed by JFH1 sequence from Core to the 3′ end. These recombinants carried duplicated sequence of up to 2400 nucleotides. HCV replication was not required for recombination, as recombinants were observed in most experiments even when two replication incompetent genomes were co-transfected. Reverse genetic studies verified the viability of representative recombinants. After serial passage, subsequent recombination events reducing or eliminating the duplicated region were observed for some but not all recombinants. Furthermore, we found that inter-genotypic recombination could occur, but at a lower frequency than intra-genotypic recombination. Productive recombination of attenuated HCV genomes depended on expression of all HCV proteins and tolerated duplicated sequence. In general, no strong site specificity was observed. Non-homologous recombination was observed in most cases, while few homologous events were identified. A better understanding of HCV recombination could help identification of natural recombinants and thereby lead to improved therapy. Our findings suggest mechanisms for occurrence of recombinants observed in patients.


Introduction
RNA viruses are rapidly adapting to their environment. The error-prone viral polymerases and the lack of proofreading mechanisms for most RNA viruses lead to high mutation rates. Genetic recombination between viral genomes is an additional mechanism increasing genetic diversity, which has proven to be epidemiologically relevant and allows RNA viruses to adapt to their surroundings [1]. Recombination could allow escape from natural or therapeutically induced immunity [2], or during antiviral treatment constitute an escape mechanism to antiviral compounds with an otherwise high barrier to resistance [3]. In addition, viral recombination has been associated with increased pathogenicity [4], and has caused the emergence of new human pathogens, such as Western equine encephalitis virus [5]. The use of live attenuated viral vaccines has led to re-emergence of disease due to recombination of vaccine strains with related viruses [6,7]; this remains a problem in poliovirus eradication. Thus, understanding the nature of viral recombination has general evolutionary implications, and might affect treatment and vaccination for important human pathogens.
Significant differences have been reported in recombination frequencies for different virus families, with high frequencies among Picornaviridae and lower frequencies among Flaviviridae and Alphaviridae [8]. Although hepatitis C virus (HCV) belongs to the Flaviviridae family, several epidemiologically important recombinant strains have been reported [9][10][11]. HCV constitutes a major public health burden with 130-170 million people chronically infected. Infection leads to increased risk of hepatitis, liver cirrhosis and hepatocellular carcinoma. The single positive-stranded HCV RNA genome of around 9600 nucleotides encodes one long open reading frame (ORF) flanked by 59 and 39 untranslated regions (UTRs). The HCV polyprotein is co-and post-translationally processed into structural (Core, E1 and E2), and nonstructural proteins (p7, NS2, NS3, NS4A, NS4B, NS5A and NS5B). Significant diversity is found among HCV isolates, which are grouped into seven major genotypes and many subtypes [12]. Genotypes, subtypes and isolates/strains differ at around 30%, 20% and 2-10%, respectively, at the nucleotide and amino acid levels.
The epidemiologically most important HCV recombinant is the homologous recombinant of genotype 2k/1b that was first identified in St. Petersburg [13]. Since then, a number of naturally occurring inter-and intra-genotypic recombinants have been reported [9][10][11]; most inter-genotypic recombinants have junction in or close to the NS2 gene. Further, naturally occurring subgenomic deletion mutants have been described to persist in around 20% of patients [14,15]. The prevalence of recombinants might be underestimated due to lack of routine screening; in addition, recombination events between isolates of the same subtype could be difficult to identify and distinguish from new isolates [16]. While mechanisms and kinetics remain problematic to study in patients, in vitro systems could provide a better understanding of HCV recombination, leading to improvements in detection of natural recombinants.
Treatment with interferon-a and ribavirin leads to sustained viral response for only around half of HCV infected patients, and many cannot be treated due to side effects or contraindications. The recent approval of novel directly acting antiviral compounds is expected to increase successful treatment rates. Great HCV genotype-specific differences exist in the outcome of antiviral therapy, and in the recommended treatment regimens [17,18]. Thus, genotyping from a single gene region could mislead decisions on treatment regimens for recombinant viral strains. In addition, RNA recombination could function as an escape mechanism to therapy with novel directly acting antiviral compounds.
Two possible mechanisms of RNA recombination are generally considered for RNA viruses: replicative and non-replicative. In the replicative copy-choice mechanism, the viral polymerase changes template during RNA synthesis whereas in the non-replicative mechanism, RNA breakage and rejoining occur. Both mechanisms can in principle lead to homologous and non-homologous recombinants. The copy-choice mechanism is the best characterized [1,19,20], and was first described for poliovirus [21]. Productive non-replicative recombination was so far only demonstrated in few studies on poliovirus [22] and bovine viral diarrhea virus (BVDV) [23], which belongs to the HCV-related pestiviruses.
Few experimental studies have investigated recombination of HCV, and our understanding of its mechanisms is still limited. One study examined HCV recombination in co-infected chimpanzees and identified homologous recombinants between genotypes 1a and 1b [24]. In another recent study, recombination frequency was investigated using the bicistronic selectable HCV replicon system [25]. In the present study, we aimed at investigating the nature of HCV recombination in infectious cell culture systems.

Co-transfection of HCV genomes lacking viability in vitro led to productive non-homologous recombination
To study HCV recombination, an assay was established using the Huh7.5 hepatoma cell line. Since recombination of HCV is thought to be a relatively rare event, HCV genomes lacking viability in vitro were co-transfected to facilitate the identification of viable recombinants. RNA transcripts of the JFH1DE1E2 genome were transfected alone or in combination with either the J6CF or J6/JFH1-GND genome (all genotype 2a, Figure 1). JFH1DE1E2 carries a partial deletion of the envelope genes, which allows replication but not viral particle production. The consensus fulllength clone of the J6 isolate, J6CF, does not replicate in Huh7.5 cells [26] but has a functional 59UTR-NS2 region in vitro [27], while the replication-deficient J6/JFH1-GND, carries an NS5B polymerase mutation in the viable J6/JFH1 background [28].
In all experiments, around 30% of cells were positive for HCV Core one day after transfection (Figure 2A); this percentage rapidly decreased due to lack of spread of infection and growth advantages of untransfected cells, as previously shown [29]. HCV RNA levels in the supernatant were comparable for all cultures during the first 8 days ( Figure 2B) and no infectious particles were released from any of the cultures on day 3 and 6 ( Figure 2C). An increase in percentage of HCV positive cells and HCV RNA levels was observed for the culture co-transfected with JFH1DE1E2 and J6CF from day 10 and infection spread to the almost entire culture on day 13. Similarly, infection spread to the majority of cells around day 36 in the culture co-transfected with JFH1DE1E2 and J6/JFH1-GND. After spread of infection in culture, infectivity titers of around 10 4 focus-forming units (FFU)/mL or 10 3 FFU/ mL, respectively, were observed in supernatant from the two cultures ( Figure 2C). After passage of supernatant from the J6CF co-transfected culture to naïve cells, HCV RNA titers above 10 7 IU/mL and infectivity titers around 10 4 FFU/mL were produced. Two additional co-transfections of JFH1DE1E2 and J6CF led to similar results, with spread of infection to the majority of the culture after 8 and 25 days, respectively.
To determine the nature of the infectious HCV genomes from the original co-transfection of JFH1DE1E2 with J6CF after passage to naïve cells, we performed direct sequencing of 12 overlapping PCR amplicons covering the entire ORF. While amplicons 1-2 (59UTR-E2) had J6 sequence, amplicons 3-12 (E2-39UTR) had JFH1 sequence; amplicons 2 and 3 contained overlapping sequence in E2 from both strains, which indicated the presence of a duplicated region. This was further analyzed for all

Author Summary
Genetic recombination is the alternative joining of nucleic acids leading to novel combinations of genetic information. While DNA recombination in cells is of importance for evolution and adaptive immunity, RNA recombination often has only transient effects. However, RNA viruses are rapidly evolving and recombination can be an important evolutionary step in addition to mutations introduced by the viral polymerase. Recombination can allow escape from the host immune system and from antiviral treatment, and recombination of live attenuated viral vaccines has led to re-emergence of disease. Hepatitis C virus (HCV) is an important human pathogen that chronically infects more than 130 million worldwide and leads to serious liver disease. For HCV, naturally occurring recombinants are rare but clinically important. HCV recombination constitutes a challenge to antiviral treatment and can potentially provide an escape mechanism for the virus. In this study, we established an assay for HCV RNA recombination and characterized the emerging homologous and non-homologous recombinant viruses. Interestingly, recombination did not depend on viral replication, occurred most efficiently between isolates of the same genotype and did not occur with strong sitespecificity. Better diagnosis of clinically important recombinants and an increased knowledge on viral recombination could strengthen antiviral and vaccine development.
three cultures co-transfected with JFH1DE1E2 and J6CF by cloning of longer PCR amplicons and amplicons generated by inverted primer sets. The resulting sequences revealed nonhomologous recombinant genomes with different genomic structures. The first recombinant had J6 sequence from the 59UTR to nucleotide (nt) 2986 (NS2), recombined with JFH1DE1E2 from nt 872 (Core) to the 39UTR (Rec#1; including the envelope deletion from nt 991 to 2040) ( Figure 3). This recombination produced an in-frame non-homologous recombinant HCV ORF containing 1065 duplicated nts (355 amino acids) with a total predicted genome length of 10743 nts, compared to 9678 for JFH1 and 9711 for J6CF. A second recombinant had J6 sequence from the 59UTR to nt 2870 (NS2), recombined with JFH1DE1E2 at nt 561 (Core) (Rec#2) (Figure 3). The third recombinant had breakpoint further downstream with J6 sequence from the 59UTR to nt 4254 (NS3) joined to JFH1DE1E2 from nt 796 (Core) (Rec#3). The resulting genome had a predicted length of more than 12 kb, over 2400 nucleotides longer than natural HCV isolates. While this is longer than typical infectious HCV reporter constructs expressing fluorescent or luminescent markers [30], much longer BVDV recombinants (up to around 20 kb) were identified in similar cell culture recombination experiments [23].
It was previously demonstrated that the NS3 helicase contributes to the unique replication abilities of the JFH1 isolate [31]. Since this might have restricted the region of recombination in cotransfections of JFH1DE1E2 and J6CF, we investigated whether a different type of recombination event had occurred in the culture co-transfected with JFH1DE1E2 and J6/JFH1-GND, where both genomes carried an NS3 protein of JFH1 origin. After passage of viral supernatant to naïve cells, sequencing of the entire ORF from recovered viruses again showed J6 sequence for amplicons 1-2 and JFH1 sequence for amplicons 3-12. In further analysis, PCR amplicon clones covering the junction revealed a recombinant genome with J6/JFH1-GND sequence from the 59UTR to nt 2971 (NS2), followed by JFH1DE1E2 from nt 860 (Core) to 39UTR (Rec#4) (Figure 3), similar in structure to those already identified.

Recombination does not depend on a functional HCV polymerase
In the initial recombination assay, a replicating genome (JFH1DE1E2) was co-transfected with a non-replicating genome   (J6CF or J6/JFH1-GND). To determine whether putative lowlevel replication of J6CF or replication of J6CF in trans by the JFH1 replicase played a role in recombination, we co-transfected JFH1DE1E2 with J6D39. J6D39 was produced by linearization of the DNA in the beginning of NS5B and would therefore not express the polymerase or carry a 39UTR ( Figure 1). This experiment led to results similar to co-transfections of JFH1DE1E2 with J6CF, with spread of infection to the majority of the culture after 13 days. After passage to naïve cells, sequencing of the replicating genome demonstrated a junction from NS2 of J6D39 to Core of JFH1DE1E2 (Rec#5, Figure 3). Thus, a functional J6 polymerase and a complete 39UTR was not a requirement for recombination, which apparently did not depend on replication of both genomes.
To determine whether at least one functional HCV polymerase would be required for recombination, we co-transfected two nonreplicating genomes. Four replicate co-transfections were performed using J6CF, which is unable to replicate in vitro, and JFH1D59, which lacks the entire 59UTR and therefore cannot undergo translation or replication (Figure 1), such that no viral replication could occur in the transfected cells. In addition, JFH1D59 was co-transfected with J6D39 (one replicate) or with transcripts from the pJ6 1-7666 plasmid (four replicates), which was constructed to only contain J6 59UTR-NS5A sequence, thus ensuring that no polymerase protein was produced ( Figure 1). In these experiments, no or very few HCV positive cells were observed by immunostaining one day after transfection. However, infection emerged in few cells in all cultures by day 4 and spread to the majority of all nine cultures in 10-32 days. After passage to naïve cells, replicating genomes were characterized by sequencing. Three of the four recombinants from the cultures co-transfected with complete J6CF genomes had structures similar to those identified in the JFH1DE1E2 co-transfections; one had junction from p7 to E2 (Rec#6), another from NS2 to Core (Rec#7), and the third from NS3 to E1 (Rec#8). Interestingly, the last recombination event was homologous with breakpoint between nt 2710-2717 in p7 (Rec#9) ( Figure 3). In the culture cotransfected with J6D39, we identified a heterologous recombinant with a short duplication of just 33 nts and junction from nt 2811 (NS2) to nt 2779 (p7) (Rec#10) (Figure 3). Heterologous recombinants were also observed in all four cultures after cotransfection with J6 1-7666 , with junctions from NS2 to Core (Rec#11), from NS2 to E2 (Rec#12) or from NS2 to p7 (Rec#13 and Rec#14) ( Figure 3).
To validate that no translation was occurring from JFH1D59 leading to the presence of HCV polymerase, we generated a JFH1D59-RLucD40 reporter construct with renilla luciferase inserted into NS5A [30], and measured low-level translation from transfected input RNA in luciferase assays. In measurements from 4-48 hours post transfection luciferase signals were observed for the positive control, J6/JFH1-RLucD40, and 4-8 hours after transfection for J6/JFH1-GND-RLucD40, for which translation but not replication could occur. In contrast, signals for JFH1D59-RLucD40 were comparable to the background signal for all time points ( Figure 4). Thus we concluded that a functional HCV polymerase was not required for recombination to occur in cell culture.

Sequential recombination events observed after serial passage in culture
To determine whether sequential recombination events could occur on the same genome, we performed long term passaging of Homologous (homol.) recombination events are indicated. The predicted total genome length is given, assuming that the recombination breakpoint was the only recombination event present. Schematic drawings of the genome structure of individual recombinants are shown. A Junction identified by direct sequencing of PCR products. B Junction identified by sequencing of cloned fragments. C One of seven clones contained an in-frame deletion of JFH1DE1E2 nt 926-957. D The same junction was subsequently also found for co-transfections of J6/JFH1D59, J6/JFH1D(59-p7) and J6/JFH1D(59-NS4A). doi:10.1371/journal.ppat.1003228.g003 the J6/JFH1DE1E2(Rec#1) and J6/JFH1(Rec#10) recombinants by serial inoculation of naïve cells with supernatant from fully infected cultures. Interestingly, after three passages to naïve cells a novel recombinant was detected in the J6/JFH1DE1E2(Rec#1) culture. The genetic structure of the new genome showed that an additional non-homologous recombination event had taken place and removed most of the duplicated region, resulting in a junction from nt 2823 (NS2) of J6 to nt 2638 (p7) of JFH1 (Rec#1.1, Figure 3). This second-generation recombinant was detectable from passage 3 and dominated the virus population from passage 6 ( Figure 6A and B). The peak supernatant infectivity titer increased in passage eight, where the shorter Rec#1.1 genome dominated ( Figure 6A and C). In contrast, no changes occurred to the comparably short duplicated junction region of J6/JFH1(Rec#10) during 8 serial passages. Infectivity titers of almost 10 5 FFU/mL were observed in most passages for this apparently genomically stable recombinant ( Figure 6C). Thus, sequential recombination events could take place in culture to eliminate long duplicated and presumably non-functional genome regions, apparently leading to increase of viral fitness.
Evaluation of recombination frequency between isolates of the same genotype All 14 co-transfection experiments with J6 and JFH1-based genomes performed so far led to emergence of viable recombinants. To get a more quantitative understanding of recombination frequencies we re-plated cells co-transfected with JFH1D59 and J6 1-7666 into 96-well format before virus production was expected to occur. This would allow an estimation of recombination frequency between the genotype 2a isolates J6 and JFH1 over the Core-NS5A region. Through 22 days of follow up, 8 controls transfected with J6/JFH1 were positive, while recombination occurred in 4/72 (5.6%) co-transfected wells (Figure 7). Taking into account that 7000 cells were plated per well and that the transfection efficiency was 50% (assuming that co-transfection had the same efficiency as observed when evaluating NS5A positive cells one day post transfection of J6/JFH1) this equals to one productive recombination event for every 63,000 co-transfected cells, or recombination in 0.0016% of the cells.

Infrequent recombination between isolates of different genotypes
Intergenotypic recombinants were previously identified in vivo [9][10][11], and synthetic intergenotypic recombinants could establish infection in cell culture [32,33]. Thus, we next investigated whether recombination in vitro could also occur between isolates of different genotypes. Since efficient replication in the infectious cell culture system at the outset of this study relied on the JFH1 isolate, we co-transfected JFH1DE1E2 with consensus clones of genotype 1a (H77C and HC-TN), 1b (J4L6S), 3a (S52) or 4a (ED43) or with 39 truncated versions (truncation in NS5B) of the same genomes. Similarly to J6CF, these clones are infectious in chimpanzees but cannot replicate in Huh7.5 cells [26]. We previously demonstrated that most synthetic JFH1 recombinants with genotype-specific Core-NS2 relied on adaptive mutations for efficient production of intracellular infectious particles [32,34]. Since many recombination events identified in this study occurred in the NS2 region, we speculated that recombination between genomes carrying previously identified adaptive mutations might enhance the production of functional intergenotypic recombinants in our assay. We thus co-transfected JFH1DE1E2 with J4L6S F886L or ED43 T827A,T977S that carried mutations previously shown to confer adaptation to the Core-NS2    recombinants, J4/JFH1 and ED43/JFH1 [32,35]. While no recombination occurred in triplicate co-transfections with ED43-T827A,T977S , co-transfection with J4L6S F886L resulted in spread of infection to the majority of cells 93 days post-transfection ( Figure 8). After passage to naïve cells, sequencing identified intergenotypic non-homologous recombination. The replicating genome contained J4L6S F886L sequence from the 59UTR to NS3 and JFH1 from NS2 to the 39UTR, and carried the introduced mutation F886L (Rec#16, Figure 3). Thus, introduction of mutations conferring adaptation to synthetic intergenotypic JFH1-based Core-NS2 recombinants had only limited effect, on recombination frequency.
While all intragenotypic co-transfections performed with high input RNA led to emergence of viable recombinants, only two intergenotypic recombination events were identified from a total of 18 co-transfection experiments. Considering all co-transfection experiments with JFH1DE1E2 and different clones of other genotypes, an estimated generalized recombination frequency would be one productive recombination event per million cotransfected cells, or recombination in 0.0001% of the cells, taking into account two productive recombination events, the starting number of 400,000 cells in each of 18 experiments and an estimated average transfection efficiency of 30%.

Recombination sites were not restricted to specific regions of the HCV genome
The recombination events identified so far all had breakpoints in the p7-NS3 region of the 59 fragment and the Core-NS2 region of the 39 fragment. Due to the lack of functional envelope genes in the JFH1DE1E2 construct, many recombination breakpoints were however restricted from occurring further upstream. Likewise, due to the importance of the NS3 helicase for the unique replication abilities of the JFH1 isolate [31], breakpoints could be restricted from occurring further downstream of non-JFH1 genomes. To investigate whether recombination events could occur in other regions, we co-transfected JFH1D59 with versions of J6CF truncated at nt 708, 1344, 2407, 2564, 2972 or 3479 (Figure 1). While no spread of infection was identified in two cultures (J6 1-708 and J6 1-2564 ), the majority of cells in the other cultures became infected after 13-22 days. Recombined genomes were identified after passage to naïve cells. Another case of homologous recombination was identified in the culture co-transfected with J6 1-1344 , occurring in the nt 858-883 region (Core) (Rec#17). The three other recombination events were non-homologous with junctions from E1 to Core (Rec#18; J6 1-2407 ), a mixed population of 2878 (NS2)/2261 (E2) and 2901 (NS2)/2521 (E2) (Rec#19a/b; J6 1-2972 ), and from NS2 to E2 (Rec#20; J6 1-3479 ) ( Figure 3). Thus, recombination of J6 and JFH1 occurred outside the NS2 region, even in the most upstream gene, Core.
Next, we wanted to determine whether recombination could occur downstream of NS3. Since JFH1 exhibits efficient function of the NS3-NS5B region in Huh7.5 cells, we transfected 59 truncated transcripts of J6/JFH1 together with J6/JFH1/39X, which carried the 59UTR-NS2 from J6CF, NS3-39UTR(polyU) from JFH1 and an irrelevant human mRNA sequence replacing the 39X region (Figure 1 (Figure 3); this recombination took place in a region where 11 of 12 consecutive bases were conserved. Depending on the primers used, wild-type NS5B sequence could also be amplified from these cultures. Independent confirmation of the junction site by RT-PCR excluded cross contamination between the samples with identical breakpoint.
Since four identical recombinants were observed, we also cloned this recombinant type, J6/JFH1(Rec#21), and analyzed it in reverse genetic studies. Surprisingly, the input recombinant with the duplicated region could only be detected one day after transfection, while wild-type virus was detected thereafter. A silent mutation introduced in NS4B was amplified together with the duplicated region to exclude contamination. Thus, Rec#21 apparently resulted from one recombination event leading to a transient state, which was rapidly followed by a second recombination event leading to wild-type J6/JFH1 sequence. The presence of wild-type NS5B sequence also in the original cultures was in accordance with Rec#21 representing a transient state.
Thus, efficient recombination was demonstrated also in the 39end of the HCV genome. In co-transfections unbiased by the selection of HCV isolates [both J6/JFH1D59 and J6/JFH1/39X carried the complete J6/JFH1 ORF], a longer stretch of conserved nucleotides seemed to be preferred over the NS2 region for the recombination breakpoint.

Discussion
In this study, efficient HCV RNA recombination leading to robust virus production was demonstrated in cell culture. Most recombination events were non-homologous with large in-frame insertions of up to 2400 nucleotides, while fewer homologous events were identified. Almost all recombinants identified from replication defective genomes were of different nature, and we thus found no strong site specificity. Further, recombination occurred most efficiently between isolates of the same genotype. Most identified recombinants maintained at least one complete copy of each HCV protein and many recombinants carried two copies of one or more genes. It remains to be determined whether such duplications could produce two different functional protein copies, e.g. leading to viral particles carrying envelope proteins of different isolates or give any advantage to the virus. Only one recombinant type did not have at least one intact copy of all HCV genes (Rec#21). Though this recombinant type had an internal junction in NS5B, it carried an intact globular finger-palm-thumb structure followed by duplicated sequence and finally the C-terminal membrane anchor [36].
Interestingly, HCV RNA recombination did not depend on HCV replication as co-transfection of two replication incompetent genomes led to productive recombination (Rec#6-14 and #17-21). Further, the frequencies of recombination and the time until spread of infection in culture did not seem to differ between cotransfections with and without replication competent genomes. A non-replicative mechanism for HCV recombination is in agreement with findings in cell culture for the related BVDV [23] and for poliovirus [22]. This type of recombination was shown primarily to take place at single-stranded RNA structures [37], and it is hypothesized to occur through endoribonucleolytic cleavage and subsequent ligation of 39-phosphate and 59-hydroxyl partners. It remains to be determined by which mechanism(s) HCV recombination occurs in patients. The replicative copychoice mechanism has previously been favored, since it is straightforward to envision how this strategy could produce the homologous recombinants observed in vivo. Accordingly, a model that could explain the generation of the 2k/1b recombinant from St. Petersburg by template switching was previously suggested [38]. Here we demonstrated that homologous recombinants could be produced through a non-replicative mechanism (Rec#9 and Rec#17), which could represent an alternative or parallel pathway to replicative recombination in vivo.
After long-term passage in culture of the non-homologous recombinant J6/JFH1DE1E2(Rec#1) a more fit variant emerged, replacing the original replicating genome and leading to higher viral titers. This new recombinant resulted from a second recombination event and carried a duplication of only 186 nts compared to the original 1065 nts. Since the original recombinant was cloned and the second event occurred after a new transfection and subsequent cell-free passages, recombination must have occurred from the same genome or among genomes with identical structures and sequentially led to a more fit variant with a smaller insertion. Similar deletions of heterologous sequences have been observed in cytopathogenic BVDV genomes with heterologous sequences [39], and HCV genomes with inserted reporter genes in cell culture and in vivo [30], reflecting the virus ability to evolve and increase its fitness. Non-homologous recombinants have not been observed in patients [9][10][11], potentially due to strong fitness selection for homologous recombinants. However, non-homologous recombinants could represent precursors to more fit homologous recombinants through sequential recombination events, as we observed in reverse genetic experiments with J6/ JFH1(Rec#21).
Co-transfections with two genomes of the same genotype led to productive recombination events in 22 of 25 experiments (86%) or 0.0016% of cells, whereas only 2 out of 18 (11%) or 0.0001% of cells in intergenotypic experiments led to productive recombination. Except for two cases of homologous recombination, all identified events were non-homologous. Reiter et al. previously described homologous recombination in the HCV replicon system [25]. However, since duplicated regions generated by nonhomologous recombination between fragments of the same isolate could be obscured in direct sequencing from PCR products, nonhomologous recombinants could possibly also have been occurring in that study. The recombination frequency in the replicon-based study ranged from one event per 3,000 to 30,000 cells, depending on the length of the genomic region available for recombination, or 0.003 to 0.03% of cells replicating wild-type replicons in parallel experiments [25]. This was slightly higher than frequencies observed in the present study, however in the replicon system, selection could allow less fit recombinants to survive and some recombination events might be compatible with replication but not with the complete viral life-cycle. In a study of cells infected with a non-cytopathogenic BVDV strain, which were subsequently transfected with a defective cytopathogenic genome, recombination events were observed in 33-58% of cultures when electroporated cells were plated in 24-well format, or roughly equivalent to one event per 0.001% of cells (assuming around 10 5 cells per culture) [23]. This was in the range of what was observed in the present study on HCV. A notable difference, however, is that for BVDV this occurred for a viable genome, while observation of similar recombination frequencies for HCV depended on two nonviable genomes. A direct comparison of frequencies is complicated, since recombination is thought to be affected by the length of the genomic region available for recombination [25], replication capacity and constraints on genome organization of productive recombinants. Studies with poliovirus and BVDV previously showed that the frequency of homologous recombination decreased with decreasing sequence homology between the RNA molecules [21,39], and that non-homologous recombination was the most frequent for recombination between different BVDV strains.
In this study, productive recombination more often took place between isolates of the same HCV genotype. The identification of several recombination events at a conserved nucleotide sequence in NS5B supports the importance of similar sequences for recombination to occur. Another explanation could be the higher functional compatibility between proteins of the same genotype expressed by the recombined RNA. The lack of sequence conservation at a number of recombination sites (Figure 3) indicates that sequence similarity is not a prerequisite for nonhomologous recombination to occur. On the other hand, the high frequency of ambiguous nucleotides in recombination sites in this study (residues around the recombination site that are identical in the two parental sequences; Figure 3), indicates a role for primary sequences in dictating junction sites. Random joining would leave one ambiguous nucleotide in one of four recombination events, two ambiguous nucleotides in one of 16 events etc. Thus, the frequency of ambiguous nucleotides in cross over sites in this study is higher than expected. The low frequency of intergenotypic recombination events identified in this study is in some contradiction to the ratio of inter-and intragenotypic recombinants identified in patients [10]. However, since intragenotypic recombinants by nature are harder to define, their existence could be underrepresented in the literature.
The recombination frequency calculations from the replicon study indicated that no recombination hotspots are present in the HCV genome [25]. This is in agreement with our findings that productive HCV recombination in the infectious cell culture system is not restricted to certain regions of the genome. However, several cases of recombination between two nearly identical 12 nt stretches in NS5B indicated some preference for conserved sequences. Interestingly, the experimental setup in the replicon study did not allow recombination to occur at this potential hotspot [25]. Recombination site specificity remains to be fully investigated in the absence of constraints using identical HCV isolates covering the entire genome. It could be speculated that some restrictions on recombination sites could apply at least to non-homologous recombination. Interestingly, all recombination sites identified in this study fall in regions where recombination of natural strains was also described [10].
All recombination events identified in the present study led to joining of viral RNA fragments. While insertion of cellular sequences has been reported for several other viruses [4,[40][41][42], and is an important regulatory process for cytopathogenicity of the related BVDV [40,43], this has not been reported for HCV in vivo. However, by cell culture transfection of deletion mutants of stem loop I of the HCV 59UTR, we previously recovered viable genomes that acquired RNA stem loop structures derived from viral or host sequences compensating for the deletion [27]. Now knowing that replication independent recombination is possible for HCV, these variants could have arisen by such a mechanism.
Non-homologous recombination could initiate important evolutionary steps in generation of novel types of viral genomes or cause diversity in genome regions tolerating insertions and deletions. Such productive non-homologous recombination events might potentially be followed by another recombination event to get rid of duplicate fitness-lowering sequences. The importance of RNA recombination for the evolution of RNA viruses is well documented [19], and many recently emerged human diseases are caused by viruses that display active recombination or reassortment [1,5]. The presence of reverse transcriptase could even fix such sequences in the cellular genome [44]. Thus, RNA recombination could have played an important role in cellular and viral genetic evolution.
The prevalence of HCV recombinants in patients is relatively low, which could in part be caused by the super-infection exclusion principle [45,46], which would reduce the chance of having two different HCV strains replicating in the same cell. In vivo, the amount of replicating RNA is further expected to be much lower than the amounts of RNA present after co-transfection in vitro. Thus, the recombination frequency reported here could well be overrepresented compared to the in vivo setting. Further, fitness of novel recombinants in vivo should be high for the recombinant to eventually dominate over the parental strains. In a treatment setting this might however be accomplished, e.g. if parental genomes each carried resistance to one of two antiviral compounds in a combination therapy, with recombination leading to a doubleresistant recombinant genome. Subgenomic deletion mutants [14,15] are naturally occurring in patients and are similar in structure to the JFH1DE1E2 construct used in this study. These could therefore constitute a reservoir of independent genomes that could potentially recombine with the wild-type to generate treatment-resistant or otherwise high-fitness genomes. With an increased knowledge on HCV recombination, better diagnosis of clinically important recombinants could become available, thereby facilitating selection of optimal therapeutic regimens for the patients. Our findings shed new light on how HCV recombination could occur in patients. Further, viral recombination might be an important escape mechanism to specific antiviral therapy in general, which could be important to consider in design of treatment regimens for certain viruses.

Culturing, transfection, infection and evaluation of cell cultures
Culturing of Huh7.5 hepatoma cells was done as previously described [51]. One day before transfection or infection, 4610 5 cells per well were plated in six-well plates. Before RNA transcription, plasmids were linearized with XbaI to generate the HCV 39-end. To produce D39 transcripts (Figure 1), linearization was done using EcoRV (nt 7764) for pJ6CF, NotI (nt 9221) for pH77C and pHC-TN, AflII (nt 9399) for pJ4L6S, NotI (nt 8549) for pS52 and KpnI (nt 9014) for pED43. Shorter truncated versions of pJ6CF were generated using ClaI (nt 709), BsiWI (nt 1345), SalI (nt 2408), BsaBI (nt 2565), NdeI (nt 2973) or AleI (nt 3480). In addition, digestion with XbaI was performed to avoid influence of minus-strand synthesis from a reverse T7 promoter. In vitro transcription of RNA was performed as previously described [35]. To exclude that recombination during in vitro transcription led to emergence of recombinant viral genomes, each transcript was synthesized separately. For transfections, 1.25 mg RNA of each construct (a total of 2.5 mg in co-transfections, as estimated by gel-electrophoresis) were incubated with 5 mL Lipofectamine2000 (Invitrogen) in 500 mL Opti-MEM (Invitrogen) for 20 min at room temperature. Cells were incubated with transfection complexes for 16-24 hours in growth medium or in pure Opti-MEM. For infection experiments, cells were inoculated with virus-containing supernatant for 16-24 hours. Cell cultures were split every 2-3 days and monitored by immunostaining using mouse anti-HCV-Core-protein monoclonal antibody (B2, Anogen) or anti-NS5A 9E10 [28] as previously described [35,51]. Supernatants collected during experiments were sterile-filtered and stored at 280uC. HCV RNA titers were determined as previously described [51]. Infectivity titers were determined by adding 100 mL of triplicate sample dilutions (diluted 1:2 or more) to 6610 3 Huh7.5 cells/well plated 24 hours before infection on poly-D-lysine-coated 96-well plates (Nunc). Cells were fixed 48 hours post-infection and immunostained for HCV following a previously established protocol [51] using anti-NS5A 9E10 as primary antibody [28]. FFU were defined by clusters of infected cells separated by at least two uninfected cells. The number of FFU was determined by manual counting or by using an automated counter (ImmunoSpot Series 5 UV Analyzer, CTL Europe GmbH) with customized software, as previously described [26,52].
To analyze recombination frequency, cells were split 18 hours after transfection and 7000 cells were plated per well in 96-well format. Starting day 5, cells were split every 2-3 days using split ratios adjusted to ensure .50% confluency in all wells throughout the experiment. At each split a replica 96-well plate was plated and incubated for 2-3 days before staining as described above for infectivity titration. In this experiment, single infected cells were counted using the ImmunoSpot Series 5 UV Analyzer.

Luciferase translation assay
For luciferase assays, RNA was transfected into 10 5 Huh7.5 cells/well of 24-well plates. At indicated time points, cells were lysed for 15 min according to the Renilla Luciferase Assay System (Promega) protocol, and luciferase signals were measured in 5 replicates using optical bottom 96 well plates on a FluoStarOptima (BMG) plate reader.
Sequence determination of culture-derived HCV HCV RNA was extracted from culture supernatant using High pure viral nucleic acid kit (Roche). For direct sequencing of the complete HCV ORF, reverse transcription, 1 st round PCR covering the entire ORF and 12 overlapping 2 nd round PCR amplifications were performed as previously described for J6/ JFH1 [51]. For J4L6S and S52 intergenotypic recombination events, primers designed for the corresponding JFH1-based recombinants were used [34,51]. Non-homologous recombination events that resulted in duplicated primer binding sites for the 2 nd round PCR could not be identified using the direct sequencing approach. In these cases, additional 2 nd round PCRs were set up with forward primers downstream of reverse primers (inverted primer sets), to specifically amplify the region containing a nonhomologous recombination breakpoint with duplicated sequence. For supernatants originating from co-transfection of different HCV isolates, the initial 12 amplicon ORF direct sequencing was used to determine in which region such primers should be designed. For supernatants originating from co-transfection of RNA from the same HCV isolate, a scanning approach was used, in which inverted primer pairs placed for each ,500 nts were tested positive or negative by PCR. Amplified PCR bands were sequenced to identify breakpoints of non-homologous recombination. In selected cases, and in cases where the recombination site could not be uniquely identified by direct sequencing, PCR products were TOPO-cloned (Invitrogen) and sequenced.
The occurrence of sequential recombination events for Rec#1 over time was monitored by PCR using primers JF1848 (CTGTGTGTGGCCCAGTGTAC) and 2763R_J6 (AGCGT-GAGCCCTGACGAAGTACGG) on cDNA. The amplified product sizes varied according to the recombinant and thereby allowed differentiation. Sequence analysis was performed with Sequencher (Gene Codes Corp.).

Control experiments
As a control for correct identification of recombinant junctions, independent RNA extraction and RT-PCR were done on supernatant of selected recombinants (Rec#1, Rec#17 and all four cultures leading to Rec#21). This confirmed the identified breakpoints. It was further confirmed that recombinant-specific PCR products could be amplified from supernatant-derived cDNA and from cloned recombinant plasmids but not from pJ6/JFH1 using inverted primer sets JF2845 (CACCCCCGGGTATAA-GACC)/2111R_JFH1 (TGTACGTCCACGATGTTCTGGTG) (Rec#1) or JF1848 and the junction-specific reverse primer Rec10_R (CGTGCACAGGTGCGTCATAGGCTCC-TATCTGGCCATGCACAG) (Rec#10). To exclude in vitro introduced recombination during T7-driven transcription or during reverse transcription after RNA extraction from supernatant, RNA produced by T7 transcription was subjected to 3 sequential rounds of DNAseI (Fermentas) digestion using the RNeasy kit (Qiagen), mixed to yield combinations of 59 and 39 partners that previously led to successful recombination in cell culture, diluted to 50 pg (equivalent to around 10 7 copies) and subjected to RT-PCR. PCR amplification using inverted primer sets JF2845/2111R_JFH1 on J6CF and JFH1DE1E2 RNA (Rec#1), JF1848 and the junction-specific reverse primer Rec10_R on J6CF and JFH1DE1E2 RNA (Rec#10), or inverted primer sets JF8806 (CAGATACTACCTGACCAGAGAC)/ JR8688 (TCCGTGAAGGCTCTCAGGTTC) on J6/JFH1/39X and J6/JFH1(D59-NS5A) RNA (Rec#21), did not lead to the specific amplicons that were observed for the cloned plasmids of the respective recombinants. Further, the viable phenotypes of all cloned recombinants and the fact that all recombination events identified led to in-frame recombinant ORFs supported that RT-PCR-induced artifacts were not misleading our conclusions.