An Array of Novel Murine Spleen Focus-Forming Viruses That Activate the Erythropoietin Receptor

Abstract

The Friend spleen focus-forming virus (SFFV) env gene encodes a 409-amino-acid glycoprotein with an apparent M_r of 55,000 (gp55) that binds to erythropoietin receptors (EpoR) to stimulate erythroblastosis. We reported previously the in vivo selection during serial passages in mice of several evolutionary intermediates that culminated in the formation of a novel SFFV (M. E. Hoatlin, E. Gomez-Lucia, F. Lilly, J. H. Beckstead, and D. Kabat, J. Virol. 72:3602–3609, 1998). A mouse injected with a retroviral vector in the presence of a nonpathogenic helper virus developed long-latency erythroblastosis, and subsequent viral passages resulted in more pathogenic isolates. The viruses taken from these mice converted an erythropoietin-dependent cell line (BaF3/EpoR) into factor-independent derivatives. Western blot analysis of cell extracts with an antiserum that broadly reacts with murine retroviral envelope glycoproteins suggested that the spleen from the initial mouse with mild erythoblastosis contained an array of viral components that were capable of activating EpoR. DNA sequence analysis of the viral genomes cloned from different factor-independent cell clones revealed env genes with open reading frames encoding 644, 449, and 187 amino acids. All three env genes contained 3′ regions identical to that of SFFV, including a 6-bp duplication and a single-base insertion that have been shown previously to be critical for pathogenesis. However, the three env gene sequences did not contain any polytropic sequences and were divergent in their 5′ regions, suggesting that they had originated by recombination and partial deletions of endogenously inherited MuLV env sequences. These results suggest that the requirements for EpoR activation by SFFV-related viruses are dependent on sequences at the 3′ end of the env gene and not on the polytropic regions or on the 585-base deletions that are common among the classical strains of SFFV. Moreover, sequence analysis of the different recombinants and deletion mutants revealed that short direct and indirect repeat sequences frequently flanked the deletions that had occurred, suggesting a reverse transcriptase template jumping mechanism for this rapid retroviral diversification.

Friend virus is a complex of a replication-competent murine leukemia virus (F-MuLV) and a replication-defective spleen focus-forming virus (SFFV) (32, 40). The SFFV component encodes a gp55 membrane glycoprotein that binds to erythropoietin receptors (EpoRs) to cause erythroblast proliferation and splenomegaly in susceptible mice (13, 24, 42). The gp55 is inefficiently processed to cell surfaces as a disulfide-bonded dimer, and it appears that these cell surface dimers are necessary for activation of EpoR (4, 8, 11, 25). The gp55 is a modified recombinant glycoprotein with domains that are closely related to the envelope glycoproteins of endogenously inherited polytropic MuLVs and ecotropic MuLVs (20). Specifically, gp55 contains an amino-terminal polytropic domain, a proline-rich linker, and a carboxyl-terminal region that is related to the Env glycoproteins encoded by ecotropic host range MuLVs (1, 6, 11, 18, 36, 56). The env genes of pathogenic SFFVs have common features including a 585-base deletion and a single-base insertion in the ecotropic region that causes a translational frameshift and premature termination of the encoded protein. These features occur in all previously described SFFV isolates and are thought to be important for pathogenesis (2, 55). Identical features occur in the independently isolated Rausher SFFV (3).

Both the pathogenic activities and infectivities of replication-competent murine retroviruses rely on the virion envelope glycoproteins, which are synthesized as gPr90 precursors that are cleaved by partial proteolysis to form surface (SU) gp70 and transmembrane (TM) p15E subunits (26, 32, 50). SU is involved in receptor binding, and thus in specifying host-range and interference properties (10), whereas TM is partially embedded in the cell membrane and contributes to membrane fusion and to immunosuppression (5, 54, 55). Although gp55 is encoded by the env gene of SFFV, it lacks the gp70/p15E cleavage site and a cytoplasmic tail and is not incorporated into virions (18).

Like the genomes of many retroviruses, the genome of SFFV is highly variable (19). Thus, strains of Friend SFFV passaged in different laboratories are distinct (1, 6, 32, 40, 50, 57). The genetic variability of retroviruses depends on the mutation rate per replication cycle, the number of replication cycles, and the selective advantage or disadvantage of the particular mutation (17, 58). The mutation rate in retroviruses is higher than in some other viral systems due to an error-prone reverse transcriptase (19, 48). Retroviral reverse transcriptase is poorly processive owing to its need to switch templates at least twice during normal replication (34). The low degree of processivity is supported by numerous experiments in vitro using purified enzyme (33). Moreover, the fact that the retroviral RNA is packaged as a dimer increases the possibilities for recombination (19). The major types of retroviral genetic variations include base pair substitutions, frameshifts, deletions (with or without insertions), and homologous and nonhomologous recombinations (51).

This report describes the evolution of a group of MuLVs which arose after the inoculation of 4- to 8-week-old NIH/Swiss mice with a retroviral vector lacking an expressed env gene plus a nonpathogenic isolate of MuLV (30). Initially, a mouse developed erythroblastosis after a long latency of several months. Subsequent in vivo passages resulted in a shortened latency of disease and culminated in formation of a novel SFFV that is closely similar to previous SFFV isolates (15a). The evolutionary intermediates were isolated based on their abilities to convert an erythropoietin (Epo)-dependent cell line, BaF3/EpoR (BER), to factor-independent derivatives. Sequence analysis of the evolutionary intermediates revealed several rearrangements, including recombinations, deletions, and base pair substitutions. Certain deletions and recombinations appear to occur frequently because of flanking direct and indirect repeat sequences that occur in the viral genomes. We found that the 585-base deletion common to all previously isolated SFFVs is not necessary for pathogenesis but appears to be a hotspot for deletions due to a flanking direct repeat sequence. Since the viruses that we analyzed were selected based on their abilities to activate EpoR, their common features implicate specific glycoprotein sequences in activation of this receptor.

MATERIALS AND METHODS

Viruses and cells.

EpoR-encoding virions were used to infect the interleukin-3 (IL-3)-dependent hematopoietic cell line BaF3 (29) to produce the BER cells used in this study as previously described (14). BaF3 cells were grown in RPMI 1640 medium supplemented with 10% fetal bovine serum and 5 × 10⁻⁵ M β-mercaptoethanol with 10% WEHI-3 conditioned medium as a source of IL-3. BER cells were maintained in the same medium with Epo (Boehringer Mannheim, Indianapolis, Ind.) at 0.5 U/ml instead of IL-3. Preparation of passaged virus from spleens was described previously (16).

Pathogenic assays.

Female NIH/Swiss mice (4 to 8 weeks old) were inoculated with virus encoding wild-type EpoR mixed with ecotropic helper virus B4 (30) as described previously (16). Passaged virus was prepared from the enlarged spleens of the diseased animals and was used immediately or frozen (−80°C). The cell-free spleen homogenate was inoculated into other mice as first passage or from these into other mice as second passage. Subsequent passages were performed similarly. For factor-independent growth assays, BaF3 cells or BER cells were exposed to the spleen filtrates for 2 h at 37°C in the presence of Polybrene (8 mg/ml). The cells were pelleted by centrifugation and resuspended in medium containing growth factor for 48 h. The cells were then sedimented by centrifugation, washed twice with phosphate-buffered saline, and resuspended in complete medium without growth factors to allow for selection of factor-independent cells. Some of the factor-independent BER cells infected with the virus were cloned by limiting dilution (32) to obtain approximately 30 clones/96-well plate.

Western blots.

For Western blotting, cell lysates were immunoprecipitated with an anti-F-MuLV gp70 antiserum that has broad reactivity with MuLV Env glycoproteins (11, 43, 44) and electrophoresed on polyacrylamide gels under reducing conditions in the presence of 1% sodium dodecyl sulfate. The proteins were then transferred to nitrocellulose membranes, immunoblotted with the same antibody, and detected with [¹²⁵I]protein A as described previously (11, 23) or by using the Renaissance system (NEN, Boston, Mass.).

Genomic studies.

Total RNA was extracted from cells or spleens by standard methods. The DNA sequence of the viral genes was obtained by reverse transcription-PCR (RT-PCR) as described in the accompanying report (15a). Briefly, cDNA was made from total cellular RNA by standard methods, and PCR was performed with either Elongase (Gibco BRL) or PCR Supermix (Gibco BRL), using combinations of the following primers. Sequences for the two forward primers were based on homology to known MuLVs: U5 (5′-TCAGCGGGGGTCTTTCATTTG-3′), located in the 5′ long terminal repeat (LTR) (38); and SF1 (5′-CGCAACCCTGGGAGACGTCC-3′), which overlaps the retroviral packaging signal (10). The reverse primers were U3 (5′-ACAGGTGGGGTCTTTCATTCC-3′) (38) and PV3 (5′-CGTTACAGCGGCATCAGGCTAAGC-3′), both located within the 3′ LTR.

PCR products were TA cloned into the pCR 2.1 vector (Invitrogen), sequenced, and compared to entries in sequence databases with the BLAST algorithm. Multiple sequence alignments were done with the Clustal W tool from the MacVector sequence analysis program (Oxford Molecular Group, Ltd.).

Nucleotide sequence accession numbers.

The accession numbers for the EE449, EE187, EE644, and DE410 sequences are AF030174, AF030175, AF030173, and AF030182, respectively. The GenBank accession numbers for sequences used to determine the recombination sites are M93134, M10100, Z11128, V01552, J02193, K00021, K02375, M93134, K02725, and Z22761.

RESULTS

An assortment of Env glycoproteins in factor-independent BER cells.

In the accompanying report (15a), we described the genesis of a new erythroleukemia virus that formed after mice were injected with a nonpathogenic helper virus plus a pSFF-based retroviral vector that lacked a functional env gene (3). Initially, a mouse developed erythroblastosis after a long latency, but subsequent passages in mice yielded more rapid erythroblastosis and polycythemia that culminated in the formation of a new SFFV. At intermediate stages in the passaging, viruses that enabled BER cells to survive and proliferate in the absence of growth factors were isolated. These viruses had no effect on the growth factor dependency of the parental BaF3 cells, suggesting that the mitogenic effect was mediated by EpoR. Interestingly, these virus isolates encoded SFFV-related Env glycoproteins. A flow chart of the relationships between the infected mice, BER cell clones, and virus nomenclature is shown in Fig. 1.

FIG. 1.

Relationships among animals, cell lines, and viruses used in this study. The in vivo and in vitro passage histories of the viruses described in this report are indicated with the terminology for animals, BER cell lines, and viruses (boxed). The origin and terminology for the cells and the corresponding viruses are described in the text. The viruses are named according to the characteristics of the envelope glycoproteins (EE, ecotropic amino-terminal domain and ecotropic carboxyl-terminal domain; DE, dualtropic [or polytropic] amino-terminal domain and ecotropic carboxyl-terminal domain). The sequence from the 1129 spleen is a novel SFFV (DE410) and is described in the accompanying report (15a). d, days.

The population of factor-independent BER cells derived from mouse 1218 (BER/1218) was used to isolate single-cell clones by limiting dilution. An immunoblot of Env glycoproteins in 11 of the resulting cell clones is shown in Fig. 2 (lanes 3 to 13). In comparison with BER cells infected with wild-type SFFV (lane 1) or with the uncloned population of BER cells infected with the 1218 virus (lane 2), all of the cell clones contained a slowly migrating component with an apparent M_r of ∼90,000 which occurs endogenously in the BaF3 cell line (14) in addition to novel components. The results suggest that the early-passaged 1218 virus that was used to infect the BER cells contained at least three different viruses that were able to activate EpoR. The most abundant of these viruses encoded an M_r-60,000 glycoprotein that was present in the majority of the factor-independent cell clones, whereas a gp55-encoding virus was evidently present in one cell clone (lane 6). One of the factor-independent cell clones appeared to lack both gp60 and gp55 glycoproteins and to contain additional components with sizes expected for the gPr90 and gp70 glycoproteins, consistent with what would be encoded by a replication-competent helper virus (lane 11). In addition, several of the lanes (including lane 11) contained traces of smaller glycoproteins that we initially presumed were breakdown products of the components shown in Fig. 2.

FIG. 2.

Env glycoproteins in growth factor-independent clones of BER cells infected with passaged virus. Lysates from BER cells infected with wild-type SFFV (lane 1), BER cells infected with passaged virus 1218 (from which 11 clones analyzed in this blot were obtained by limiting dilution) (lane 2), clone 5S (lane 3), clone 4S (lane 4), clone 3S (lane 5), clone 2S (lane 6), clone 1S (lane 7), clone 6F (lane 8), clone 5F (lane 9), clone 4F (lane 10), clone 3F (lane 11), clone 2F (lane 12), and clone 1F (lane 13) were immunoprecipitated with anti-Env antibody, run on electrophoresis gels under reducing conditions, and transferred to nitrocellulose membranes. The membranes were again incubated with anti-Env antibody and then with [¹²⁵I]protein A. The higher-M_r band at 70,000 to 90,000 occurs endogenously in the BaF3 cells (14).

env sequences of the novel viruses.

To determine the viral sequences present in infected BER factor-independent cells, RT-PCR was performed on total RNA isolated from BER/1218 clones 2F and 3F (Fig. 2, lanes 11 and 12) and from BER cells that were infected with a later passage of the 1218 virus (termed BER/1129) as well as with control factor-dependent cells. Amplified products that were reproducibly observed and unique to the factor-independent cells were cloned and sequenced (see Materials and Methods). The DNA sequences obtained are shown in Fig. 3 in comparison to an ecotropic env gene with the closest homology to the viral sequences. The 3′ regions of the env genes analyzed in this work were all identical to each other and with those remaining in the pSFF vector (pSFF GenBank accession no. Z22761), suggesting that the pSFF vector was a common source in the recombinations that we observed. Although these 3′ env sequences remain in pSFF, there is a large deletion in 5′ sequences that eliminates the initiation codon and the pSFF vector consequently does not encode any Env glycoprotein. All of the novel sequences contained several characteristic hallmarks of an SFFV, such as (i) a single-base insertion that causes a frameshift mutation and leads to a premature termination of the encoded protein and (ii) a six-base duplication. EE449 and EE187 also contained the 585-base deletion (the positions of these features are indicated in Fig. 3). However, as shown in Fig. 3, the 5′ regions of the EE644, EE449, and EE187 sequences were closely related to the ecotropic MuLV reference sequence shown on line 1. Based on extensive sequence comparisons with many related viral env sequences (see the legend to Fig. 3 for sequences used in the comparison), we propose that the novel viruses formed by recombinations of the pSFF vector with ecotropic env sequences that were present in the mice. The likely recombination sites are boxed in Fig. 3. The recombination sites are difficult to identify because there is homology among ecotropic and polytropic sequences in several regions. However, because we found that the 3′ sequences of the novel viruses were nearly identical to the 3′ sequence of the pSFF vector, a divergence in the nucleotide substitutions indicated that a recombination had occurred. The figure also shows the partial sequence of the 585 bases that are deleted in SFFVs and in EE449 and EE187. This 585-base sequence occurs in EE644 and is flanked by a short direct repeat (Fig. 3 and 5).

FIG. 3.

Nucleotide sequences of the novel env genes. The sequences are compared to an ecotropic env sequence (GenBank accession no. Z11128) shown on line 1. Nucleotides for EE644 (line 2), EE449 (line 3), and EE187 (line 4) are shown where the sequences differ from the reference sequence. Sequences flanking deletions are shown, and repeated bases are underlined. Lowercase letters represent sequences proposed to have participated in the deletion and are not present in the final viral sequence. Proposed recombination sites for the viral sequences are shown on lines 2, 3, and 4 and are boxed and labeled. Sequences of EE187, EE449, and EE644 are identical to the pSFF vector sequence after the recombination site proposed for EE644. The 786-nucleotide in-frame deletion in EE187 is indicated by arrows. The 585 bases deleted in EE449 and EE187 (lines 2 and 3, respectively) but present in the reference sequence and EE644 are bracketed (lines 1 and 2, respectively). The sites of the 6-base duplication (▿⁶) and single-base insertion (▿¹) are indicated by inverted triangles. A series of dots indicates where additional sequence exists but is not shown.

FIG. 5.

Deletions observed in viral sequences are flanked by direct (A) or indirect (B) repeat sequences. DNA sequences were determined by dideoxy sequencing from the cloned RT-PCR products from BER factor-independent cell lines or enlarged spleens as indicated and as described in the text. Details of two different types of deletions are shown at the deletion junction with flanking sequence. The repeated sequences are boxed. nts, nucleotides.

Figure 4 shows the amino acid sequences of the Env glycoproteins in the viral isolates in comparison to each other and to the novel SFFV DE410, the gp55 sequence that ultimately formed in the mice. The DE410 SFFV sequence is shown in Fig. 4 as a reference because it contains the classical features of SFFV gp55s, including a polytropic-related sequence in the amino terminus and an ecotropic-related sequence in the carboxyl-terminal region. As shown, the DE410 amino acid sequence is very similar to the ecotropic envelope sequences at the carboxyl-terminal region, and it deviates dramatically from EE644, EE449, and EE187 sequences in the amino-terminal region. Thus, the latter Env glycoproteins have amino-terminal regions very different from those of gp55s.

FIG. 4.

Retrovirus-related env sequences present in factor-independent BER cells. The envelope glycoproteins are encoded by the sequences obtained from the BER factor-independent cells (EE449, EE187, and EE644) and an enlarged spleen from an animal injected with a highly passaged virus preparation (DE410). Five amino-terminal residues common to the novel env sequences are underlined. Amino acid identities are shaded, nonconservative substitutions are shown without shading, and conservative substitutions are boxed. Deletions are indicated by dashes. SFFV characteristics and viral protein landmarks are numbered as follows: 1, the 786-nucleotide (262-amino-acid) in-frame deletion in EE187; 2, the proline-rich region; 3, the site of the 585-base in-frame deletion that eliminates the gp70/p15E cleavage site in SFFVs and in EE449 and EE187; 4, the gp70/p15E cleavage site in MuLVs which is eliminated in SFFVs; 5, insertion of two leucines caused by a 6-base duplication; 6, site of the single-base insertion causing early termination. Proposed recombination sites are indicated by asterisks and labeled.

The EE449 sequence indicated an open reading frame encoding 449 residues with a predicted M_r of 49,000. Because several potential N-glycosylation sites are present in the deduced amino acid sequence, the calculated M_r is compatible with the observed M_r of 60,000 for the glycoprotein in the BER/1218/2F cells. In comparison, most strains of SFFV encode glycoproteins of 409 amino acids and an observed M_r of 55,000.

The EE187 sequence contained an open reading frame predicted to encode 187 amino acids. Like the Env encoded by EE449, the sequence of EE187 has overall homology to ecotropic Envs and lacks the polytropic amino-terminal region found in DE410 and other SFFVs. An M_r of 20,500 is predicted for the unglycosylated EE187 Env glycoprotein. It is difficult to determine with confidence whether this protein is present on Western blots of BER/1218/3F lysates (Fig. 2, lane 11). Many weak bands are present in this size range and could possibly be proteolytic products of larger Env glycoproteins. Moreover, since EE187 has two large deletions, the polyclonal antibody used in the immunoblot in Fig. 2 may not have recognized the EE187 glycoprotein. Since the EE187 sequence was reproducibly isolated, we conclude that it was present in the BER/1218/2F cells.

The EE644 virus was isolated in a clone of BER cells (BER/1129) that became factor independent after infection with a virus from a spleen extract. This same mouse spleen also contained the DE410 virus as shown in Fig. 1. The EE644 sequence encodes a predicted Env protein of 644 amino acids (Fig. 4), starting with the characteristic MACSTL… of ecotropic viruses. Since the deduced sequence includes the gp70/p15E cleavage site, the EE644 Env glycoprotein would be expected to encode a gp70 glycoprotein similar to that of the helper virus used in these experiments. This prediction was confirmed by Western blotting (data not shown). However, the p15E region of the EE644 glycoprotein is truncated in a manner identical to the carboxyl-terminal region of gp55.

Total RNA extracted from the spleen of a mouse injected with a highly passaged virus was also sequenced. The sequence DE410, discussed in the accompanying report (15a), is predicted to encode a 410-amino-acid glycoprotein for a new SFFV. As reported previously, sequence landmarks in the DE410 glycoprotein are consistent with a recombination between ecotropic sequences and endogenous polytropic sequences (this recombination site is indicated in Fig. 3), which resulted in a sequence typical for SFFVs.

Nucleotide sequences flanking the observed deletions.

During the course of this investigation, nucleotide sequences were cloned not only from the env gene regions but also from the gag-pol regions of the retroviral sequences. In many cases, the clones contained deletions in comparison to reference sequences in the databases or in comparison with other clones that we isolated. Interestingly, these deletions (including the 585-base deletion that is common to SFFVs) were frequently flanked either by short direct repeats or by inverted repeats in the reference sequences. The sequences that we isolated were all deposited in GenBank (accession numbers are listed in Materials and Methods). Figure 5 shows a compilation of the types of deletions and flanking sequences that we observed.

DISCUSSION

Requirements for EpoR stimulation and pathogenicity by retroviral envs related to SFFV.

MuLVs produce a variety of progressive hematopoietic disorders, including erythroleukemias (7, 18, 41, 50). The pathogenic activity of the replication-defective SFFVs is mediated by the binding of the SFFV-encoded gp55 glycoprotein to EpoR in the erythroblasts (24), and this triggers cell proliferation that will eventually lead to erythroleukemia (9). The gp55 glycoprotein is processed to cell surfaces as a disulfide-bonded dimer, and this dimeric structure on cell surfaces is required for activation of EpoR (4, 8, 11, 12, 25). Several of the characteristics that distinguish gp55s from the envelope glycoproteins of MuLVs have been previously implicated in the induction of erythroblastosis: a polytropic amino-terminal region, a 585-base deletion, a 6-base duplication, and a single-base insertion, as described above. These modifications have been proposed to be important based on their occurrence in the sequences of different strains of SFFV and on experiments designed to test each modification separately for its contribution to pathogenicity (2, 5, 27, 47, 54, 55, 59), although the role of polytropic amino-terminal sequences in activation of EpoR and in SFFV pathogenesis has been uncertain (46, 54). The common occurrence of the identical 585-base (195-amino-acid) deletion in natural SFFVs, including the independently isolated Friend and Rausher SFFVs, has been confusing because expanded deletions in this region do not reduce pathogenesis. Indeed, these deletions substantially enhance pathogenicity in mice homozygous for the Fv-2^r resistance allele (15, 28). The surprising result that we report here is that although the BER cells infected with virus from mouse 1129 or 1218 became factor independent, and this correlated with the detection of Env glycoproteins on Western blots, these Env glycoproteins did not include all four of the SFFV characteristic features that were common to previously analyzed gp55s. For example, EE644 contains the 6-base duplication and the single-base insertion but has an ecotropic amino-terminal region and no 585-base deletion. Similarly, EE449 and EE187 have the 6-base duplication, the single-base insertion, and the 585-base deletion but also contain an ecotropic amino-terminal region and lack any polytropic sequences. By characterizing and comparing env sequences able to activate EpoR, a profile of minimal required sequences for receptor activation may emerge. For example, in addition to the striking carboxyl-terminal similarities observed, the novel sequences contain a five-amino-acid identity in the amino-terminal region (Fig. 4, positions 40 to 45).

A strength of this work was our ability to detect viral intermediates in a process of SFFV evolution in vivo and to clonally isolate these intermediates based on their abilities to activate EpoR in BER cells. This enabled us to isolate and to study viruses that were only weakly pathogenic in the mice and that presumably contributed to the observed erythroblastosis and to the process of SFFV evolution. The fact that all of our virus isolates encoded in-frame Env glycoproteins that were absent from the Epo-dependent BaF3/EpoR cells but present in the factor-independent derivatives strongly supports our conclusion that they were able to activate EpoR and that they contributed to the erythroid cell proliferation in these mice. The parental BaF3/EpoR cells are strictly dependent on either IL-3 or Epo for survival and growth. It should be emphasized that the individual Env-encoding viruses that we have identified were obtained from mice that had erythroproliferative disease and contained a swarm of distinct viral components. It is difficult in this circumstance to precisely trace the evolutionary lineages of the individual viruses or to unambiguously know their pathogenic characteristics. Studies to reconstitute the molecularly cloned env sequences into proviral structures and test their individual roles in pathogenesis are in progress.

Recombination with nonpathogenic input vector sequences resulted in novel viruses.

Of the three types of recombinations previously described by Zhang and Temin (58), we have observed only the general type, in which the sequences used as a vector recombined with endogenous polytropic and ecotropic sequences upstream of the 3′ LTR. Little sequence identity (<7 nucleotides) is needed for such recombination events to occur (49), and since ecotropic sequences remaining in the vector are similar to the ecotropic helper virus sequences, homologous recombinations could occur. It is believed that reverse transcriptase jumps between templates at least twice during normal replication (33) and that such template switches may also occur during recombinations and deletions. Reverse transcriptases lack editing functions and have been shown to be error prone in vitro and therefore are presumed to contribute to the high mutation rate in vivo (19).

We conclude that a recombination took place in the first three mice that became sick. Virus was prepared and was further passed to other mice, and from one of these mice (1218) it was passed to BER cells (BER/1218) or to other mice (e.g., 1129). The sequences isolated from the BER/1218 cells (EE449 and EE187) confirmed that a recombination had taken place within a short sequence 36 bp downstream of the 3′ end of the proline-rich region between the input vector and ecotropic sequences, most probably with the ecotropic B4 helper virus used in these experiments. Further passages also led to recombination near this site, resulting in EE644 and DE410. Thus, all of these recombinations probably occurred within a 160-bp area just downstream of the proline-rich region in SFFV (Fig. 3).

Purcell and others have reported a complex and related group of viral elements that apparently formed by recombination in a murine retroviral packaging cell line and were transmitted to rhesus macaques during subsequent gene transfer experiments (38). Some of the sequences isolated were highly related to polytropic virus env sequences, suggesting that complex and interrelated retroviral recombinants may form from input sequences and endogenous sequences. As the Purcell study and this report show, the use of RT-PCR and universal primers may allow the true complexity, frequency, and biological consequences of these events to be identified.

The deletions observed suggest a general mechanism for MuLV diversification.

As previously proposed, deletions involving two direct repeats presumably occur when reverse transcriptase copies the first of the direct repeats and the growing DNA strand misaligns with the homologous direct repeat found downstream in the template (31, 34, 35, 37, 52). This mechanism results in the deletion of intervening sequences together with one of the repeats. In addition to the deletions shown in Fig. 5, we observed three other deletions of this category in the sequence of DE410 upstream of env (data not shown). We also observed one deletion involving nonhomologous sequences or sequences with very little similarity (data not shown). These latter deletions confirm that reverse transcriptase is able to successfully transfer between templates that have little or no sequence similarity. Deletions of sequences flanked by divergent sequences with only 1 bp in common have been observed in other retroviruses (33, 37, 45, 52, 53). Repeat-mediated deletions are not restricted to retroviral replication since similar deletions have been observed in human genetic diseases (21, 22, 39).

In addition to the 585-base deletion that is common in SFFV envs, another large deletion, which eliminated a large portion of the amino-terminal region and most of the proline-rich region of the viral sequence, was observed in EE187. This deletion is unusual because it is flanked by inverted repeats, which to our knowledge have not been reported earlier. Inverted repeat-mediated deletions are not obvious unless antecedent sequences are examined, since both flanking repeats are absent from the final sequence. Common to all deletions observed in the viral sequences that we report here is that the number of bases deleted is always a multiple of 3, emphasizing the functional importance of keeping the sequence in frame for retroviral diversification to have biological consequences. Unless a virus mutant or recombinant can stimulate mitosis, the virus sequence would be expected to be eliminated by dilution, because nonproliferating cells produce relatively few retrovirions. These results support our conclusion that the envelope glycoproteins of the viruses that we cloned were under selective pressure in vivo and had pathogenic effects.

Besides recombinations and deletions, base substitutions were also observed in the viruses. The same substitutions were not in the parental sequences but have been previously observed in similar viruses. It is uncertain whether these substitutions contributed to pathogenesis.

One interesting implication of our results is that the 585-base deletion observed in the env genes of all previously isolated SFFVs may be a consequence of an inherent instability of this sequence and not necessarily due to selective pressure to produce EpoR stimulation. This 585-base deletion is flanked in the progenitor viruses by a direct repeat sequence (Fig. 3). These observations suggest that certain genes prone to repeat-mediated deletions (e.g., Fanconi anemia group A and BRCA1) (22, 39) may be unstable when expressed in retroviral vectors for gene therapy applications.

Molecular evolution of an SFFV.

We have described an array of viruses capable of activating EpoR that emerged as a result of in vivo selection. Between 60 and 80 days postinfection, three mice displayed massive splenomegaly and polycythemia. The spleen homogenates prepared from the enlarged spleen of one mouse caused fulminant disease in all subsequently infected mice. The viruses present in these mice enabled Epo-dependent BER cells to become Epo independent. The viruses characterized from the mice or from the factor-independent BER cells contained recombinations as well as deletions and base pair substitutions. Based on the relationships of the sequences, and their temporal appearance, we propose that the EE449 and EE187 viruses may have formed during the initial stages of infection and that these and/or related variants caused a slowly developing erythroblastosis. The gp60 glycoprotein was prominent in the initial spleens, implying that the EE449 virus was abundant and that it was highly amplified in that mouse. During subsequent viral passages, further viral evolution that culminated in production of the novel DE410 isolate of SFFV occurred. Our results demonstrate that the intermediate stages of this retroviral evolution can be trapped in indicator cell lines (e.g., BER) that have requirements for specific growth factors such as Epo. Although such weakly pathogenic retroviruses have not been previously studied and cannot be stably passaged in vivo, they are clearly capable of causing serious indolent disease in individual animals.