Multiple invasions of an infectious retrovirus in cat genomes

Endogenous retroviruses (ERVs) are remnants of ancient retroviral infections of host germ-line cells. While most ERVs are defective, some are active and express viral proteins. The RD-114 virus is a replication-competent feline ERV, and several feline cell lines produce infectious RD-114 viral particles. All domestic cats are considered to have an ERV locus encoding a replication-competent RD-114 virus in their genomes; however, the locus has not been identified. In this study, we investigated RD-114 virus-related proviral loci in genomes of domestic cats, and found that none were capable of producing infectious viruses. We also found that all domestic cats have an RD-114 virus-related sequence on chromosome C2, termed RDRS C2a, but populations of the other RDRSs are different depending on the regions where cats live or breed. Our results indicate that RDRS C2a, the oldest RD-114-related provirus, entered the host genome before an ancestor of domestic cats started diverging and the other new RDRSs might have integrated into migrating cats in Europe. We also show that infectious RD-114 virus can be resurrected by the recombination between two non-infectious RDRSs. From these data, we conclude that cats do not harbor infectious RD-114 viral loci in their genomes and RD-114-related viruses invaded cat genomes multiple times.

Endogenous retroviruses (ERVs) are remnants of ancient retroviral infections of host germ-line cells. While most ERVs are defective, some are active and express viral proteins. The RD-114 virus is a replication-competent feline ERV, and several feline cell lines produce infectious RD-114 viral particles. All domestic cats are considered to have an ERV locus encoding a replication-competent RD-114 virus in their genomes; however, the locus has not been identified. In this study, we investigated RD-114 virus-related proviral loci in genomes of domestic cats, and found that none were capable of producing infectious viruses. We also found that all domestic cats have an RD-114 virus-related sequence on chromosome C2, termed RDRS C2a, but populations of the other RDRSs are different depending on the regions where cats live or breed. Our results indicate that RDRS C2a, the oldest RD-114-related provirus, entered the host genome before an ancestor of domestic cats started diverging and the other new RDRSs might have integrated into migrating cats in Europe. We also show that infectious RD-114 virus can be resurrected by the recombination between two non-infectious RDRSs. From these data, we conclude that cats do not harbor infectious RD-114 viral loci in their genomes and RD-114-related viruses invaded cat genomes multiple times. E ndogenous retroviruses (ERVs) are retroviruses which have been integrated into the host genome of germline cells, comprising approximately 10% of the host genome in mammals 1 . ERVs behave as host genes and are transmitted from parents to offspring in a Mendelian manner. Although most ERVs are inactivated through the accumulation of mutations and deletions, leading to an introduction of stop codons within coding sequences, some ERVs are active and have the potential to produce infectious viral particles.
The RD-114 virus is a replication-competent feline ERV 2 and several feline cell lines produce infectious RD-114 viral particles 3 . The RD-114 virus was first isolated from human rhabdomyosarcoma (RD) cells during passages through fetal cat brains and was mistakenly regarded as the first human retrovirus 4 . Subsequently, it was revealed that cellular RNA from cats contain sequences homologous to RD-114 viral RNA 5,6 , leading to the discovery of RD-114 virus-related proviral sequences in the cat genome [7][8][9][10] . In addition, several groups reported that infectious RD-114 viral particles were produced, either spontaneously or after chemical induction, from cells originating from domestic cats [11][12][13] . From these data, it was concluded that the RD-114 virus was an active ERV of cats, and not an exogenous human retrovirus.
Several research groups have searched for the active RD-114 proviral loci 2,14,15 . Spodick et al. identified thirteen RD-114 virus-related clones, consisting of conserved gag-pol and unrelated env regions from six domestic cat genomes 14 . At the same time, another group identified nine endogenous RD-114 virus-related sequence clones representing at least eight distinct loci from cat DNA libraries 2 . Although the gag-pol region of the clones was conserved, the env region displayed considerable divergence from the replication-competent RD-114 virus clone derived from human RD cells, both in size and sequence homology, coinciding with previous results. In their followup study, feline chromosomes which contain RD-114 virus-related segments were identified with variation of distinct viral segments among individual cats 15 . These two groups identified the restriction endonucleasedigested genome fragments corresponding to fragments of the infectious RD-114 virus clone in cat genomic DNA and concluded that the active infectious RD-114 viral locus may be a single copy. The former group detected a 4.8kb PstI-digested pol-env fragment in all cats genomic DNAs examined 14 , and the latter group detected a 1.0-kb BamHI-digested gag fragment 2,15 . Although the latter fragment was found to be located on chromosome B3 15 , the locus of the former fragment was not identified. Further it has not been confirmed whether both of them were present on the same single locus as the complete full-length provirus. After these studies, a new endogenous type C retrovirus named Felis catus endogenous retrovirus (FcEV) which consists of the RD-114 virus-related gag-pol gene and unrelated env gene was discovered in all domestic cats genomes examined 16 . They reported that there were fifteen to twenty FcEVs in domestic cat genomes and suggested that all or most RD-114 virus-related sequences (RDRSs) possessing unrelated env genes are more likely to be FcEV integrations 16,17 . Therefore, as mentioned above, the RD-114 proviral locus encoding an infectious RD-114 virus has not been identified.
In this study, we searched for loci encoding infectious RD-114 virus. Contrary to our expectations, we could not find any RDRSs capable of producing infectious RD-114 virus, but discovered that a RD-114 virus-related virus (RDRV) nearly identical to the RD-114 virus can be produced by recombination between two defective RD-114 virus-related proviruses. We also noticed that most RDRSs have not been fixed in domestic cat genomes, suggesting multiple invasions of cat genomes by the ancient RDRV. Our study sheds light on host-virus interactions and provides novel genetic markers of domestic cats available for tracing footsteps of cats.

Results
Characterization of RDRS loci in the genomes of domestic cats. Firstly, we searched for RD-114 proviral loci in the genome database of domestic cats (Felis_catus_6.2) using the SSEARCH program 18 with the complete genome sequence of RD-114 strain CRT1 (GenBank accession number: AB559882.1) 19 as a query; however, we could not identify any proviruses which were identical to the sequences of three previously reported infectious RD-114 virus clones 20 . Instead, we found one defective RDRS locus on feline chromosome C2 and designated it as RDRS C2a (proviruses were named after the chromosome numbers on which they are located). Quite recently, by using the sequences of RD-114 virus infectious clones as queries, Tamazian et al. reported that there are twelve RD-114 virus-like elements with more than 50% sequence identity, which cover at least 50% of query sequences in domestic cat genomes 21 . However, their sequences are not available for further analyses. Next, we investigated whether RDRS(s) capable of inducing RD-114 virus are present in all cat genomes by Southern blot analysis. By using HindIII and a gag probe, we detected a 2.6-kb fragment corresponding to the HindIII fragment of RD-114 virus molecular clones in RD-114 virus producer Crandell-Rees feline kidney (CRFK) cells and RD-114 virus non-producer cell lines such as G355-5 (a feline fetal astrocyte cell line), Felis catus whole fetus-4 (fcwf-4) (a feline macrophage-like cell line) and FeT-J (an interleukin-2-independent feline T-cell line) cells (Fig. 1B). By using HindIII and an env probe, we detected a 3.2-kb fragment corresponding to the HindIII fragment of RD-114 virus molecular clones only in CRFK cells but not RD-114 virus non-producer cell lines (Fig. 1D). In accordance with the previous studies 2, 14,15 , there are polymorphisms present in RD-114 virus-related sequences with a conserved env region. By southern blotting hybridization using SacII, EcoRI or SphI and probes for gag, pol and env genes, we revealed that RD-114 virus-related env copy numbers were less than gag-pol copy numbers (Fig. 1B, C, D). Inverse nested PCR targeting RD-114 virus env identified three proviruses (designated as RDRS A2, C2a, and C1, respectively) in CRFK cells ( Fig. 2 and Fig.  S1). We also identified three proviruses (RDRS C2a, E3 and D4) in primary feline peripheral blood mononuclear cells (PBMCs) and two proviruses (RDRS C2a and C2b) in FER cells (feline embryonic fibroblasts that produce RD-114-related virus) ( Fig. 2 and Fig. S1). Again, these proviruses were not identical to the sequences of infectious RD-114 virus clones. Four of these proviruses (RDRS A2, C1, D4 and C2b) have the entire open reading frames (ORFs) for env and two of them (RDRS C1 and C2b) have intact ORFs for all viral genes (gag, pol and env) (Fig. 2). We found that the 3.2-kb HindIII fragment containing env could be generated by digestion of four RDRSs (RDRS A2, C1, D4 and C2b) which have the intact ORF for env gene. RDRS C2a of CRFK, G355-5 cells and feline PBMCs (n52) have stop codons within all viral genes in the same fashion. Phylogenetic analysis of RDRSs and RD-114 virus clones indicated that RDRS C2a belongs to a separated group from the other RDRSs and infectious RD-114 virus clones (Fig. 3). In feline cell lines, although all lines have RDRS C2a, RD-114 viral producer cell lines do not have any other loci in common (Table 1). These data suggest that a single RDRS is not responsible for the production of RD-114 viral  Timelines for acquisitions of RDRSs in feline species. We next estimated the integration time of RDRSs. The divergence accumulated between 59 and 39 LTRs over time was previously used as a molecular clock to investigate the integration time of ERV 22 . Thus, we tried to estimate the integration time of the RDRSs by assessing the sequence differences between 59 and 39 LTRs. The proposed rate of 1.2 3 10 -8 substitutions/site/year in cats was adopted for calculating proviral insertion time 23 . Four of the six proviruses (RDRS A2, E3, D4 and C2b) have identical 59 and 39 LTRs, which are suggestive of relatively recent integrations into the host genome, less than 0.2 million years ago (MYA), possibly by RDRV infection (Fig. 2C). On the other hand, RDRS C1 and C2a have one and eight nucleotide differences, resulting in an estimated time of integration of RDRS C1 and C2a around 0.2 and 1.6 MYA, respectively (Fig. 2C). These periods are after domestic cats separated from the Sand cat lineage 24 . The time of ERV integration can be also estimated by determining whether phylogenetically closely related animal species share these proviruses in the same genome location. The oldest RDRV provirus C2a was not present in the genome of the Leopard cat (Prionailurus bengalensis) (n54) which inhabits Vietnam. Instead, we found a 54bp sequence within the same locus. Intriguingly, the 54-bp sequence is part of a SINE element (Repbase ID: SINEC_Fc2), predicted by the Repeatmasker program (http://www.repeatmasker.org/) (Fig. 5). The SINE-like sequence is defined by a tRNA-related region, harboring RNA polymerase III-specific internal promoter with its conserved regulatory elements, A-box and B-box, followed by a microsatellite region (TC)n and by an A/T-rich tail with the polyadenylation signal AATAAA 25,26 . The 54-bp sequence in the leopard cat genome contains the A-box, but the domestic cat's genome lacks this part of the SINE element. In other Felidae, including the Serval cat (Leptailurus serval), Snow leopard (Panthera uncia), and Tiger (Panthera tigris) (whole genome shotgun sequences, accession number: NW_006711997.1), the full length SINE elements were maintained in their genomes at the same location as the leopard cat, suggesting that SINEC_Fc2 integrated into the genome of the common ancestor of Felidae species and then RDRS C2a integration occurred, leading to the disruption of a part of the SINE sequence (Fig. 5). Anai et al. also reported the absence of the RD-114 viral env gene in the Tsushima cat (Prionailurus bengalensis euptilurus) 27,28 . The domestic cat is estimated to have separated from the leopard cat lineage approximately 6.2 MYA 28 . We hypothesize that the oldest RDRV C2a entered the ancestor of domestic cats from 6.2 to 1.6 MYA before the diversification of cat breeds.
Integration of RDRS coding for infectious-type env. Because RDRS sequences vary depending on where cats live or breed, it is possible that unidentified RDRS loci are present in domestic cat genomes of breeds investigated above. We developed a PCR method to differentiate infectious-type env on any loci from the defective C2a-type env (Fig. 4B). RDRS C2a has a 15-bp insertion in the env coding region, whereas the other RDRSs do not. We designed a reverse primer to step over the region of the 15-bp insertion present in RDRS C2a (Fig. 4B). As a result, we revealed that 40 and 56% of domestic cats in Europe and North America have infectious-type env, respectively, whereas almost all domestic cats in Asia do not (Fig. 4A). Asian domestic cats containing infectious-type env were mongrel cats and their origins were not identified. Interestingly, none of the Asian cats have any RDRSs harboring infectious-type env identified in this study. Similarly, most Scottish fold cats have RDRS harboring infectious-type env; however, they are unknown RDRSs. These data may indicate that unidentified RDRS(s) with infectious-type env are present in Scottish fold cats. Domestic cats (Felis silvestris catus) originate from the African wildcat (Felis silvestris lybica), a wild cat living in the Middle East, and then spread over the world to Europe or a different route to Asia 24 (Fig. 4C). Our results suggested that the newer RDRVs (i.e. RDRSs possessing infectious-type env) might have infected domestic cats after cats started moving from the Middle East. There is a possibility that some additional loci with the infectious-type env were acquired in cats through introgression of newer RDRSs into their genomes by crossbreeding between different feline species or cat breeds 21,[29][30][31] . Although the original cat breeds of CRFK, 3201 (feline thymic lymphoma) and FER cells possessing the newer RDRSs are unclear, these cell lines were established in the United States or the United Kingdom 32-34 , suggesting that they originated from cats living in these countries.
Infectivity of RDRSs. Among the RDRS loci, RDRS C1 and C2b have intact ORFs for gag, pol and env genes, and RDRS C1 was most similar to the sequences of infectious RD-114 virus clones 20 . RDRS C1 has five differences, when compared with an infectious RD-114 virus clone, termed pSc3c; LTR has one additional direct repeat (termedDDR-A), one point mutation at primer binding site (the ''a'' to ''g'' change at position 491 of pSc3c), and three point mutations in pol gene (''a'' to ''g'' at 3748, ''t'' to ''c'' at 5394 and ''a'' to ''g'' at 5719) resulting in amino acid substitutions (Fig. 6A). To examine if RDRS C1 encodes infectious RD-114-like virus, we transfected human embryonic kidney (HEK) 293T cells with a plasmid encoding RDRS C1. We then collected culture supernatants at 48 hours post transfection, and inoculated them onto HEK293T cells transduced with the LacZ marker gene [HEK293T(LacZ) cells]. Virus titers produced in the   culture supernatants were measured by the LacZ marker rescue assay as described previously 35 . As a result, we found that RDRS C1 was not infectious. We further evaluated the effects of the three point mutations in the pol region on the infectivity of RDRS C1. We constructed single mutants, which were made by site-directed mutagenesis 36 . Two mutants were not infectious, whereas one mutant (5394t mutant) with a mutation introduced in the integrase core domain (Pfam accession number: PF00665), predicted by Pfam, (http://pfam.sanger.ac.uk/) had infectivity as an RD-114 virus clone ( Fig. 6B and C). From these results, we conclude that the ''t'' at 5394 is crucial for infectivity for RDRS C1 infectivity. The same mutation in RDRS C1 has also been observed in the sequences of two unrelated cats.
In addition to CRFK cells, FER cells produce infectious RD-114related viral particles 3 . Because FER cells have RDRS C2b which has intact ORFs for gag, pol and env genes (Fig. 2), we suspected that RDRS C2b may encode an infectious RD-114-like virus. RDRS C2b has a 9-bp insertion at the pol coding region, resulting in the insertion of three amino acids in comparison with pSc3c (Fig. S2). We examined the infectivity of RDRS C2b in the same way as described above. Consistent with results of the other RDRSs identified in this study, we found that RDRS C2b was not infectious (Fig. S2).
Recombination of RDRSs. Because we could not find any loci which encode infectious RD-114 virus, we hypothesized that recombination between RDRSs resulted in the production of infectious RD-114 viral particles. We attempted to reproduce the recombination event by transfection of two RDRS plasmids. Because CRFK cells, from which infectious RD-114 virus is produced, have RDRS A2, C2a and C1 loci, we suspected that an infectious RDRV might be generated by recombination between RDRS A2 and C1. We transfected two plasmids encoding RDRS A2 and C1, and succeeded in recovering replication-competent viral particles (Fig. 7A). By comparing sequences of RDRS A2 and C1, and the recombinant RDRV obtained by transfection of RDRS A2 and C1 (hereinafter referred to as RDRV AC), it was suggested that three changes had occurred (two in the middle of pol and one in pol-env region) (Fig. 7B). Of the two single nucleotide differences between RDRV AC and pSc3c at positions 3224 and 4357, only the ''c'' to ''a'' at 3224 results in an amino acid substitution. The amino acid difference was not present in the predicted functional domains of Pol. The length of LTR of RDRV AC was 27 bp shorter than that of the recombination origin RDRS C1 at the repeat region (Fig. 7C). The 27-bp sequence was repeated in the U3 region of LTR in pSc3c, RDRS A2 and C1, and termed direct repeat A1 (DR-A1). Altogether, infectious RD-114 virus-like particles can be generated by recombination between defective RDRS A2 and C1.

Discussion
The RD-114 virus was originally isolated from human RD cells after passages in fetal cat brains 4 . Because the parent cell line (RD cells) does not contain the RD-114 virus genome, there has been speculation about the origin of the RD-114 virus. In previous studies, restriction endonuclease map analysis and Southern blot hybridization of cat cellular DNA revealed that there are approximately twenty RDRSs in domestic cat genomes but most have a deleted-env region 2,14,15 . The 4.8 kb PstI-digested fragment corresponding to a complete env region was detected in all cat DNAs examined as a single copy in the previous study; however, they did not isolate a clone containing a complete RD-114 provirus 14 . We confirmed that the PstI-digested env fragments derived from RDRS C2a is similar to infectious RD-114 virus clones, pSc3c and pCRT1, in size (4.7 kb) with high sequence homology (95.7%, data not shown). These data suggest that the single locus in all cat genomes which Spodick et al. identified was in fact RDRS C2a and not the complete infectious RD-114 provirus. Another group identified eight RDRSs (Ren 18c, 20a, 8a, 7c, 10a and 6c) in domestic cat genomes 2 and found that the 1.0kb BamHI gag fragment specific to the RD-114 virus clones was present in cat genomes as a single copy on chromosome B3; however, domestic cats are polymorphic with respect to the presence or absence of this RD-114 virus-related segment 2,15 . Although it has been proposed that all domestic cats harbor infectious RD-114 provirus in their genomes, the proviral locus has not been identified even after extensive mining of feline genome databases.
In this study, we identified six RDRSs in domestic cats genomes. These proviruses are not identical to the sequences of infectious molecular clones of RD-114 virus reported previously 20 . RDRS C1, whose sequence is most similar to those of RD-114 infectious clones, could not produce infectious viral particles. The defect was ascribed to a single nucleotide mutation resulting in an amino acid substitution in the integrase core domain which plays a key role in catalysis 37 . The sequences of RDRS LTRs are rich in diversity, suggesting  Table 2) and there were no differences between 59 and 39 LTRs in RDRSs A2, C2b and E3. These data suggest that LTR-LTR recombination did not occur and RDRS invasions into cat genomes occurred relatively recently. Except for RDRS C2a, RDRSs have diverse TSDs, indicating that multiplication of RDRSs occurred through viral re-infection but not genome rearrangements. The mechanism of multiple invasions of RDRVs is unknown at present. Generally, it is considered that integration of ERVs into host genomes confers resistance to the infection of exogenous counterparts 38 . However, it is possible that the expression of newly integrated RDRS is tightly regulated in cats, especially in germline cells, to allow multiple integrations of RDRV in cats.
Remarkably, number of RDRS proviral copies vary considerably with type of cell lines and cat breeds. All cats have RDRS C2a in common, but all proviral genes (gag, pol and env) are disrupted in RDRS C2a. The other RDRSs have not yet been fixed in domestic cat genomes. Furthermore, we found that most Asian cats are free from new RDRSs harboring infectious-type env.
Because domestic cats originated from the African wildcat living in the Middle East (Felis silvestris lybica) 24 , we propose the following scenario (Fig. 4C)  British Shorthair (BSH) was established in England from European Shorthair which first appeared in Sweden, and later American Shorthair (ASH) and American Curl (a variant of ASH) originated from BSH 40 . Some of these cat breeds have RDRS E3 but others do not. These data indicate that RDRS E3 would be a useful genetic marker to trace the footsteps of these breeds from Europe to America. In future studies, we will investigate the presence of RDRSs in domestic cats living in various regions as well as closely related wild cats to further map cat migration.
Several feline cell lines including CRFK and FER cells produce RD-114 or RD-114-related viruses, but there are no proviruses identical to previously reported RD-114 viral sequences. Several novel retroviruses have been reported to be generated by the recombination of ERVs [41][42][43] . Therefore, we hypothesized that RD-114 virus-like particles were generated by the recombination between two defective RDRS proviruses. By co-transfection of two defective RDRS A2 and C1, we demonstrated that infectious RD-114-like viral particles were generated through recombination. The resultant recombinant, RDRV AC, and infectious clones of RD-114 virus were nearly identical in coding sequences; however, the LTR of RDRV AC was not identical to those of the infectious clones, pCRT1 and pSc3c. The changes in LTR length frequently occurred in repeat sequences during cell passages in vitro as observed in other gammaretroviruses such as porcine endogenous retrovirus 44,45 . FER cells also produce RD-114-related viral particles 3 , despite the lack of RDRS A2 and C1 in their genomes (Table 1). Similar to the case of CRFK cells, we speculated that an RD-114-related virus could have been generated by recombination between RDRS C2b which has the infectious-type env and some other unidentified RDRSs or FcEV. Retroviral recombination events occur by template switching during reverse transcription 46,47 . Xenotropic murine leukemia virus-related virus (XMRV) is one such virus which has arisen through the recombination of two defective murine ERVs during passages of human prostate cancer xenografts in immunocompromised nude mice 48 . The defective viruses recombined in human cells but not in mouse cells because the XMRV is xenotropic. Similarly, the first RD-114 virus may have arisen by recombination in human RD xenografts in cat brains 4,8 . RD-114-related viruses have also been induced by chemicals in feline cells without xenotransplantation [11][12][13] , where a RD-114-related virus may have arisen in feline cells without contact with cells from other species. In these cases, defective viral particles co-packaged with RDRS genome infected feline cells and recombined during reverse transcription. Resultant replication-competent viruses may have expanded in the feline cells, because feline cells are permissive to RD-114 virus 3,49 . Additionally, we should consider a possibility that RDRS C1 produces infectious viral particles by spontaneous mutation like Emv2 in C57BL mice [50][51][52] . Emv2 has a defect in the pol gene, compromising the ability of the Emv2 to produce infectious virus particles 53 . However, Emv2 can produce an infectious endogenous ecotropic murine leukemia virus (E-MLV) through the acquisition of a spontaneous mutation in the pol gene in vivo in aging C57BL mice 50 or by backcrossing between C57BL and E-MLV-negative mice 51 . Therefore, we cannot exclude the possibility that RDRS C1 produces infectious viral particles through spontaneous mutations in vivo.
In the previous study, we revealed ubiquitous expression of the RD-114 viral receptor, termed ASCT2, in domestic cat tissues 49 ,  suggesting that RD-114 viral recombination events may occur in cats, and the infectious RD-114 virus may be generated in some groups of cats having RDRSs harboring infectious-type env.
Recently, it was reported that in immune-compromised mice, ERVs in the host genome were resurrected and induced lymphoma 54,55 . In these cases, receptors for infectious ERVs are expressed in tissues of mice and infectious ERVs re-infected the host cells and induced lymphoma possibly via insertion of the ERVs in the vicinity of cellular proto-oncogenes 54,55 . In this study, most RDRSs are not yet fixed in cat genomes, but some of them have the potential to generate a recombinant virus which could re-infect the host cells. This recombinant virus may re-infect cat cells which do not express RDRS Env, leading to new integrations. Some of the new integrations may induce diseases in cats, if the integration occurred in the vicinity of proto-oncogenes. While it is still unknown whether RD-114 virus causes any diseases in cats, our report that RD-114 virus can be generated by recombination will be useful for further investigation of the pathogenicity of this virus. Blood and tissue samples. Domestic cat blood samples for isolation of PBMCs were provided from Fujimura Animal Hospital (Osaka, Japan), Kyoritsu Seiyaku Corporation (Tokyo, Japan) and Veterinary Clinics in Japan. The Snow leopard and Serval cat tissue samples were provided from Sapporo Maruyama Zoo (Hokkaido, Japan). These animals died from feline parvovirus infection 57 .
Genomic DNA isolation and genomic PCR. Genomic DNA was extracted from cell lines and PBMCs using a QIAamp DNA Blood Mini Kit (QIAGEN, Valencia, CA). Genomic PCR was performed by using PrimeSTAR GXL polymerase (TaKaRa, Shiga, Japan) or Ex Taq polymerase (TaKaRa). The former was used for cloning RDRSs and the recombinant RDRV, and the later was used for screening for the presence of RDRSs. Amplicons were analyzed by electrophoresis in 0.8% or 2% agarose gels. Primers used in this study are listed in Table S1.
Southern blotting. A total of 5 mg (or 3 mg) genomic DNA was isolated from PBMCs or cell lines, digested with HindIII for detection of gag and env genes, SacII for detection of gag and pol genes, EcoRI for detection of env genes, or SphI for detection of gag and env genes. Digested fragments were separated by a 1.0% agarose gel and transferred to positive charged-nylon membrane (Roche Diagnostics, Indianapolis, IN). Hybridization was performed using digoxigenin (DIG)-labeled PCR probes and DIG easy Hyb (Roche) at appropriate temperatures (50uC for gag and pol, 65uC for env detection) for 16 hours. After hybridization, the membrane was washed with a low stringency buffer (0.1% SDS and 2 3 SSC) at room temperature followed by a high stringency buffer (0.1% SDS and 0.1 3 SSC) at 68uC. The hybridized bands were visualized using CDP-Star reagent (Roche) following the manufacturer's instructions. Probes were prepared from an infectious molecular clone of RD-114 virus, termed pCRT1 19 , using DIG-labeled PCR probe synthesis kit (Roche) according to the manufacturer's instructions with primers used listed in Table S1.
Inverse PCR. Genomic DNA was digested with EcoRI or SphI and 100 ng of digested genomes were ligated by T4 ligase (TOYOBO, Osaka, Japan) at 16uC overnight. Ligated DNA was amplified with specific primers listed in Table S1 using PrimeSTAR GXL polymerase (TaKaRa). Second PCR products were cloned into a pCR4 Blunt TOPO vector (Invitrogen) and sequenced. Loci of RDRSs were determined by comparing flanking sequences with the genome database of domestic cats (Felis_catus_6.2).
Cloning of RDRSs. Full-length RDRSs were amplified using specific primers (listed in Table S1) on flanking regions and amplicons were cloned into pCR BluntII TOPO vector (Invitrogen).
Infection assay. HEK293T cells were transfected with 4 mg of RD-114 virus molecular clones or plasmids containing RDRSs using Lipofectamine2000 (Invitrogen). The pcDNA3.1 vector was used for mock transfection. Infectivities of RDRSs were examined by the LacZ marker rescue assay 28 . Culture supernatants were collected 48 hours post-transfection and filtrated through 0.45 mm-pore-size filters and then inoculated onto HEK293T(LacZ) cells with 8 mg/ml polybrene. After inoculation, supernatants were collected every four days. TE671 cells were seeded in 96-well plates at 1 3 10 4 cells per well one day before infection and diluted supernatants of inoculated HEK293T(LacZ) cells were inoculated onto TE671 cells in the presence of 8 mg/ml polybrene. Two days after inoculation, the inoculated cells were stained with X-Gal, and virus titers, expressed as focus forming units (f.f.u.)/ml, were determined as described previously 35,47 . Assays were conducted in triplicate for each individual sample.
Preparation of RDRS C1 mutants. RDRS C1 mutants were made by DpnI method 36 . PCR was performed using PrimeSTAR GXL Polymerase (TaKaRa) with primers listed in Table S1. RDRS C1 plasmid DNA was used as template. Amplicons were digested with DpnI to eliminate the parental plasmid DNA. Digested DNA fragments were purified and transformed into Escherichia coli DH5 alpha.
RDRV AC isolation. HEK293T cells were transfected with 4 mg of pCRT1 or plasmids containing RDRSs using Lipofectamine2000 (Invitrogen). pcDNA3.1 vector (Invitrogen) was used for mock transfection. Culture supernatants were collected 48 hours post-transfection and filtrated through 0.45 mm-pore-size filters and then inoculated into HEK293T cells with 8 mg/ml polybrene (Sigma-Aldrich). After inoculation, supernatants were collected and filtrated every four days. TE671 cells were seeded in 6-well plates at 1 3 10 6 cells per well one day before infection and diluted supernatants of the inoculated HEK293T cells (36 d.p.i.) were inoculated onto TE671 cells in the presence of 8 mg/ml polybrene. Six hours after inoculation, genomic DNAs were extracted and proviruses were amplified by PCR. PCR was performed with PrimeSTAR GXL polymerase (TaKaRa) and primers listed in Table  S1, according to the manufacturer's instructions. The forward primer was designed in the R region of 59LTR and reverse primer was designed in the U5 region of 39LTR (positions of primers were indicated in Fig. 7).
Sequence Analyses. Sequencing was performed by a commercial DNA sequencing service (FASMAC, Kanagawa, Japan). Alignment of nucleotide sequences was computed using L-INS-i program in MAFFT version 7 58 , and their phylogeny was inferred by RAxML version 7.2.6 that is based on the maximum likelihood method 59 .