Latent Cellular Oncogenes: The Paradox Dissolves

In order to understand cancer, it is necessary to identify cancer genes. The search for such genes and for mechanisms that generate such genes must take into consideration that at the cellular level cancer is a very rare event. The kind of cellular transformation that leads to monoclonal tumours in vivo occurs only in about 1 out of 2x 10¹⁷ mitoses in humans or animals. The basis for this estimate is that most animal and human tumours are monoclonal (Wolman, 1983; Rowley, 1984; Trent, 1984; Duesberg, 1987) and that humans and corresponding animals represent about 10¹⁶ mitoses (assuming 10¹⁴ cells, that go through an average 10² mitoses) and that about 1 in 5 persons die from tumours (Silverberg & Lubera, 1986).

As yet the only proven cancer genes are the transforming one genes of retroviruses. These are autonomous transforming genes that are sufficient for carcinogenesis (Duesberg, 1983, 1985). Thus, these viruses are by far the most direct and efficient natural carcinogens. They transform susceptible cells in culture with the same kinetics as they infect them, and they cause tumours in animals with single-hit kinetics (Duesberg, 1983, 1985). Therefore these viruses are never associated with healthy animals.

However, tumours with retroviruses that contain one genes are very rare in nature, as only less than 50 cases are recorded from which such viruses were isolated (Duesberg, 1987, 1983, 1985; ‘RNA Tumor Viruses’ 1985). Moreover these viruses never have been reported to cause epidemics of cancer. The probable reasons are that viral one genes arise naturally only with great difficulty via two illegitimate recombinations, and that once arisen they are very unstable because they are not essential for virus replication (Duesberg, 1983, 1985). Nonessential genes are readily lost due to spontaneous deletion or mutation. Indeed, one genes were originally discovered by analysis of spontaneous deletions of the sre gene, the one gene of Rous sarcoma virus (RSV) (Duesberg & Vogt, 1970; Martin & Duesberg, 1972). Subsequently, about 20 other viral one genes were identified in retroviruses (Duesberg, 1983, 1985; Weis et al. 1985, 1982). All of these viral one genes were originally defined by ‘transformation-specific’ sequences that are different from the known sequences of essential virus genes (Duesberg, 1979).

Since one genes are unstable, they must also be recent additions to retroviruses. Indeed, the cellular genes from which the transformation-specific sequences on oncogenic retroviruses were transduced have been identified in normal cells. This was initially done by liquid hybridization of transformation-specific viral sequences with cellular DNA (Scolnick et al. 1973; Scolnick & Parks, 1974; Tsuchida et al. 1974; Frankel & Fischinger, 1976; Stehelin et al. 1976), and later by comparing cloned viral one and corresponding cellular genes (Watson et al. 1983). The cellular genes from which viral one sequences are derived have since been termed proto-one genes (Duesberg, 1983).

The cellular origin of the transformation-specific sequences of retroviral one genes is frequently presented as a particular surprise (‘RNA Tumor Viruses’ 1985, 1982). However, cells are the only known source of genetic material from which viruses could transduce genetic information and viral transduction is canonical knowledge ever since phage lambda was first shown to transduce beta-galactosidase in the 1950s. Indeed virsues are themselves derivatives of cellular genes that have evolved away from their progenitor genes as they acquired their capacity of self-replication.

On the basis of the sequence homology between viral one genes and proto-one genes, viral one genes have been postulated to be transduced cellular cancer genes, and proto-owc genes have been postulated to be latent cancer genes (Bishop et al. 1979, KaressetrzZ. 1979; Wanget al. 1979; Bishop, 1981; Klein, 1981; Bishop et al. 1982; Bishop, 1983; Varmus & Bishop, 1986, Weiss, 1986). According to this view, termed the oncogene concept (Weiss, 1986), proto-ozzc genes are not only converted to transforming genes from without by transducing viruses, but also from within the cell by increased dosage or increased function (Bishop et al. 1979; Karessei al. 1979; Wang et al. 1979; Bishop, 1981; Klein, 1981; Bishop et al. 1982; Bishop, 1983; Varmus & Bishop, 1986; Weiss, 1986). Activation of latent oncogenes from within the cell is postulated to follow one of four prominent pathways: (i) point mutation (Tabin et al. 1982; Reddy et al. 1982; Taparowsky et al. 1982); (ii) chromosomal translocation that brings the latent oncogene under the control of a heterologous enhancer or promoter (Klein, 1981; Leder et al. 1983); (iii) gene amplification (Varmus & Bishop, 1986; Weiss, 1986); and (iv) activation from a retroviral promoter integrated adjacent to the latent oncogene (Bishop, 1981; Klein, 1981; Bishop et al. 1982; Bishop, 1983; ‘RNA Tumor Viruses’ 1985; Varmus & Bishop, 1986; Weiss, 1986). Thus, this view predicts the latent cancer genes preexist in normal cells. However the existence of latent cancer genes is a paradox, because such genes would obviously be most undesirable for eukaryotes.

Indeed, the oncogene concept was a revision of Huebner’s oncogene hypothesis, which postulated activation of latent oncogenic viruses instead of latent cellular oncogenes as causes of cancer (Huebner & Todaro, 1969). Nevertheless Huebner’s hypothesis remained unconfirmed because most human and animal tumours are virus-negative (‘RNA Tumor Viruses’ 1982, 1985). Moreover, the retroviruses and DNA viruses that have been isolated from tumours are not directly oncogenic (Duesberg, 1987), except for the less than 50 isolates of animal retroviruses which contain one genes (Duesberg, 1985; ‘RNA Tumor Viruses’ 1982, 1985).

The revised oncogene hypothesis was at first sight highly attractive because it derived credibility from the proven oncogenic function of retroviral one genes, the viral derivatives of proto-one genes, and because it promised direct access to the long sought cellular cancer genes in virus-free tumours with previously defined viral one genes as hybridization probes. Predictably the hypothesis has focused the search for cellular cancer genes from the 10⁶ genes of eukaryotic cells to the 20 known proto-one genes (Duesberg, 1985;‘RNA Tumor Viruses’ 1985; Klein, 1981; Bishop etal. 1982; Bishop, 1983; Varmus & Bishop, 1986; Weiss, 1986).

The hypothesis makes four testable predictions, namely, (i) that viral one genes and proto-one genes are isogenic; (ii) that expression of proto-one genes would cause cancer; (iii) that proto-owc genes from tumours would transform diploid cells as do proviral DNAs of viral one genes; and above all (iv) that diploid tumours exist, that differ from normal cells only in activated proto-one genes. Despite record efforts in the last six years, none of these predictions is confirmed. Instead: (i) Genetic and biochemical analyses, that have defined essential retroviral genes, viral one genes and proto-one genes during the last sixteen years, show that viral one genes and proto-one genes are not isogenic (Duesberg, 1983, 1985) (see below), (ii) Further, it turned out that most proto-owc genes are frequently expressed in normal cells (Duesberg, 1985). (iii) Contrary to expectation, none of the 20 known proto-onc genes isolated from tumours functions as a transforming gene when introduced into diploid cells. The apparent exceptions of proto-ras and proto-myc are discussed below. By comparison, proviral DNAs of retroviral one genes transform normal cells exactly like the corresponding viruses (‘RNA Tumor Viruses’ 1982, 1985). (iv) As yet, no diploid tumours with activated proto-one genes have been found, except for those caused by viruses with one genes (Pitot, 1978; Klein etal. 1980). Instead of activated oncogenes Duesberg, 1983), clonal chromosome abnormalities are a consistent feature of virusnegative tumours (Wolman, 1983; Rowley, 1984; Trent, 1984; Levan, 1956) and also of all those tumours that are infected by retroviruses without one genes (Duesberg, 1987).

Proto-ras is the cellular precursor of Harvey, Balb and Rasheed murine sarcoma viruses (‘RNA Tumor Viruses’ 1982, 1985). Both proto-ras and the viruses encode a colinear protein, termed p21, of 189 amino acids (Fig. 1). In 1982 it was discovered that proto-ras extracted from a human bladder carcinoma cell line, but not from normal cells, would transform the morphology of a few aneuploid murine cell lines, in particular the 3T3 mouse cell line (Tabin et al. 1982; Reddy et al. 1982; Taparowsky et al. 1982). Subsequently proto-ras DNAs from some other cell lines and rarely from primary tumours (Duesberg, 1985; Feinberg et al. 1983; Fujita et al. 1985; Milici et al. 1986) were also found to transform 3T3 cells. Since proto-ras behaves like a dominant and autonomous transforming gene in this morphological assay, it was claimed to be a cellular cancer gene (Tabin et al. 1982; Reddy et al. 1982; Taparowsky et al. 1982). The 3T3 cell transforming function of the proto-ras gene from the bladder carcinoma was reduced to a single point mutation that changed the 12th ras codon of p21 from the normal gly to val (Tabin et al. 1982; Reddy et al. 1982; Taparowsky et al. 1982). In the meantime, over 50 different point mutations in five different ras codons have been identified that all activate 3T3 cell transforming function (Cichutek & Duesberg, 1986; Lowy & Willumsen, 1986). Since the viral ras genes and proto-ras genes encode the same p21 proteins, whereas most other viral one genes encode proteins that are different from those encoded by proto-one genes (Fig. 1; Duesberg, 1983, 1985), this system has been considered a direct support for the hypothesis that viral one genes and proto-one genes are indeed isogenic and hence can become functionally equivalent by point mutations (Bishop, 1983; Varmus & Bishop 1986; Weiss, 1986; Tabin et al. 1982; Reddy et al. 1982; Taparowsky et al. 1982; Lowy & Williumsen, 1986).

However the following arguments cast doubt on the claims that point mutations are indeed necessary or sufficient to convert proto-ras-to a dominant cancer gene:(1) The claim that point mutations convert proto-ras to dominant cancer genes is significantly weakened by the very rare occurrence of ras mutations in the kind of spontaneous tumours in which they are occasionally found (Duesberg, 1985; Feinberg et al. 1983; Fujita et al. 1985; Milici et al. 1986). In fact, the gly to val mutation that was originally found in the human bladder carcinoma cell line (Tabin et al. 1982; Reddy et al. 1982; Taparowsky et al. 1982), has never been found in a primary tumour. Further, the origin of this mutation may not have coincided with the origin of the tumour, since the mutation was not found in the original tumour but only in a cell line, 10 years after this line was derived from the original bladder carcinoma (Hastings & Franks, 1981).

On the basis of a numerical argument it is also unlikely that point mutations are sufficient to convert proto-ras genes to dominant cancer genes. The frequency of point mutations of eukaryotes is 1 in 10⁸ nucleotides per mitosis (Wabl et al. 1984). Thus 50 out of 10⁸ mitoses are expected to generate mutant Harvey ras genes with dominant transforming function. By contrast, spontaneous transformation that leads to clonal tumours only occurs in less than 1 out of about 2×10¹⁷ mitoses and only a small minority of these contains mutant ras genes.

It may be argued however that indeed 50 out of 10⁸ mitoses generate tumour cells with activated proto-ras and that the immune system eliminates these cells. However this is unlikely, since a point mutation is not an easy target for immunity. Further, animals or humans that are tolerant to ras point mutations would be expected to develop tumours at a very early age, if point mutated proto-ras genes were dominant cancer genes, as the 3T3 assay suggests. Instead spontaneous human tumours with activated proto-ras are very rare and all were observed in adults (Rowley, 1984; Feinberg et al. 1983; Fujita et al. 1985; Milici et al. 1986).

Moreover the argument that cellular oncogenes exist that can be activated by point mutation and then controlled by immunity is hard to reconcile with the existence of athymic or nude mice which do not develop more spontaneous tumours than other laboratory mice (Sharkey & Fogh, 1984). Finally, one would predict that in the absence of immunity, as in cell culture, 50 out of 10⁸ normal cells should spontaneously transform due to point mutation of Harvey proto-ra.v and probably the same number due to mutation of Kirsten proto-ra.v (‘RNA Tumor Viruses’ 1985). Yet spontaneous transformation of cells in culture is clearly a much less frequent event.

In an effort to directly test the hypothesis that ras genes are activated to dominant cancer genes by point mutation, we have analysed whether the transforming function of ras genes does indeed depend on point mutations. Using site-directed mutagenesis, we have found that point mutations are not necessary for transforming function of viral ras genes and proto-m.v genes that had been truncated to be structurally equivalent to viral ras genes (Cichutek & Duesberg, 1986).

(2) Contrary to expectation, the same proto-ras DNAs from human tumours that transform aneuploid 3T3 cells do not transform diploid human (Sager et al. 1983) or diploid rodent cells (Land et al. 1983; Newbold & Overell, 1983), the starting material of natural tumours. Thus transformation of 3T3 cells does not appear to be a genuine assay for transforming genes of diploid cells. Instead of initiating transformation, mutated proto-ras may just alter the morphology and enhance tumorigenicity of aneuploid 3T3 cells. It is consistent with this view that untreated 3T3 cells are tumorigenic in nude mice (Boone, 1975; Littlefield, 1982; Greiget al. 1985). Thus, proto-ras genes with point mutations are not sufficient for transformation. They only appear as dominant cancer genes in certain aneuploid cells like the 3T3 cells.

(3) Assuming that mutated proto-ras genes are cancer genes, like viral one genes, one would expect diploid tumours that differ from normal cells only in ras point mutation. Contrary to expectation, chromosome abnormalities are consistently found in those tumours in which proto-ras mutations are occasionally found (Wolman, 1983; Trent, 1984). The human bladder carcinoma cell line, in which the first proto-ras mutation was identified, is a convincing example. This cell line contains over 80 chromosomes instead of 46 and it includes rearranged marker chromosomes (Hastings & Franks, 1981). In view of such fundamental chromosome alterations, a point mutation seems to be a rather minor event.

Thus, proto-ras genes with point mutations are neither sufficient nor proven to be necessary for carcinogenesis and are not autonomous cancer genes like viral ras genes. In addition there is no kinetic evidence that the origin of the mutation coincides with the origin of the tumours in which it is found. It is consistent with this view that proto-ras mutations, that register in the 3T3 cell transformation assay, have been observed to occur in vivo in benign hyperplasias, as for example in benign murine hepatomas (Reynolds et al. 1986) or benign purely diploid mouse skin papillomas that differentiate into normal skin cells (Balmain et al. 1984, 1983; Balmain, 1985; Klein & Klein, 1984; Burns et al. 1978). Ras mutations have also been observed to arise after carcinogenesis in aneuploid cancer cells (Albino et al. 1984; Tainsky et al. 1984; Vousden et al. 1984), rather than to coincide with the origin of cancer. By contrast, viral ras genes are sufficient for transformation and thus initiate transformation of diploid cells in vitro and in vivo with single hit kinetics and concurrent with infection (Duesberg, 1985; Aaronson & Weaver, 1971; Hoelzer-Pierce & Aaronson, 1982).

This then raises the question why viral ras genes are obligatory carcinogens under conditions where proto-ras genes with point mutations are not. Unexpectedly a sequence comparison between proto-ras genes and the known viral ras genes has recently revealed a proto-ras-specific exon, that was not transduced by any of the known retroviruses with ras genes (Cichutek & Duesberg, 1986). It is as yet unclear whether the untransduced exon has a regulatory or a coding function (Cichutek & Duesberg, 1986). It follows however that proto-ra.v and viral ras genes are not isogenic (Fig. 1). Since four different viral ras genes have been shown to lack the same proto-ras exon, and since point mutations are not necessary for transforming function, we have proposed that proto-ras genes derive transforming function for diploid cells by truncation of an upstream exon and recombination with a retroviral promoter (Cichutek & Duesberg, 1986) (see below).

Proto-myc is the cellular precursor of four avian carcinoma viruses termed MC29, MHZ, CMII and OKIO with directly oncogenic myc genes (Duesberg, 1985). The transforming host range of viral myc genes appears to be limited to avian cells, as murine cells are not transformed by cloned proviral DNAs (Land et al. 1983; Rapp et al. 1985). Nevertheless, it is thought that proto-myc, brought under the control of heterologous cellular enhancers or promoters by chromosome translocation, is the cause of human Burkitt’s lymphoma or mouse plasmacytoma (Leder et al. 1983; Klein & Klein, 1984; Adams et al. 1985).

The following arguments cast doubt on whether such activated proto-myc genes are indeed necessary or sufficient for carcinogenesis:

(1) The human proto-myc gene is located on chromosome 8. This chromosome is typically rearranged in B-cell lines derived from Burkitt’s lymphomas (Duesberg, 1985; Leder et al. 1983; Klein, 1985). However, although chromosome 8 is subjected to translocations, proto-myc is frequently not translocated and when translocated it is frequently not rearranged (Duesberg, 1985; Leder et al. 1983; Klein & Klein, 1984). Moreover no rearrangements of chromosome 8 were observed in about 50% of primary Burkitt’s lymphomas; instead other chromosome abnormalities were recorded (Biggar et al. 1981). Thus, proto-myc translocation is not necessary for lymphomagenesis.

(2) Expression of proto-myc is not consistently enhanced in lymphomas (Duesberg, 1985).

(3) As yet no proto-myc gene isolated from a tumour has been demonstrated to transform any cells (Duesberg, 1985). In an effort to assay transforming function in vivo, a proto-myc gene, that was artificially linked to heterologous enhancers, was introduced into the germ line of mice (Adams et al. 1985). Several of these transgenic mice developed lymphomas after 1 to 5 months, implying that activated proto-myc had transformed diploid cells. However the lymphomas of the transgenic mice were all monoclonal (Adams et al. 1985). Thus, if the activated proto-myc gene was indeed responsible for the lymphomas, it would be an extremely inefficient carcinogen, because only 1 of about 10⁸ ‘control’ B-cells of the same mouse with the same transgenic myc gene was transformed. Moreover, there is no deletion or mutation analysis to show that the activated proto-myc played a direct role in the tumours of the transgenic mice (Adams et al. 1985). By contrast, viral myc genes transform all susceptible cells directly and inevitably (Duesberg, 1985).

(4) If translocated proto-myc were the cause of Burkitt’s lymphomas, one would expect all tumours to be diploid and to carry only 2 abnormal chromosomes, namely, number 8 and the chromosome that was subject to reciprocal translocation with number 8. However, as mentioned above, primary Burkitt’s lymphomas exist with two normal chromosomes 8 but with other chromosome abnormalities (Biggar et al. 1981). Thus, translocated proto-myc genes are not sufficient nor proven to be necessary for carcinogenesis.

It was estimated above that the probability of spontaneous transformation that leads to monoclonal tumours in humans is 2×10^−1/ per mitosis. By contrast, one would expect activation of a preexisting, latent proto-orcc gene to be a much more frequent event. For a given proto-onc gene, the probability of activation per mitosis would be the sum of the probabilities associated with each of the four putative pathways (Varmus & Bishop, 1986; Weiss, 1986) of proto-onc activation: (1) The probability of a point mutation per nucleotide per mitosis is about 1 in 10⁸ (Wabl et al. 1984) (and thus about 50 in 10⁸ for Harvey proto-ras). The probability that any one of the 20 known proto-onc genes is activated would be 2×10⁻⁷.

(2) The probability of a given proto-onc gene to be activated by amplification is about 1 in 10⁸, considering that about 1 in 10³ to 10³ mitoses leads to gene amplification in vitro and possibly in vivo and that about 10³ out of the 10⁶ kilobases (kb) of eukaryotic DNA are amplified (Stark, 1986; Schimke et al. 1986). The probability that any one of the 20 known proto-onc genes would be activated by amplification would then be 2×10⁻⁷.

(3) The probability of oncogene activation by chromosome translocation depends largely on what distances between a proto-onc gene and a heterologous enhancer, and on which enhancers are considered sufficient for activation. Since distances of over 50 kb of DNA have been considered sufficient for activation of proto-myc (‘RNA Tumor Viruses’ 1985; Klein & Klein, 1984) and proto-abl (‘RNA Tumor Viruses’ 1985; Heisterkamp et al. 1985), the proto-owc gene of murine Abelson leukemia virus (‘RNA Tumor Viruses’ 1985), and since an enhancer is likely to be found in every 50 kb of cell DNA, nearly every translocation within a 50-kb radius of a proto-onc gene should be activating. Thus the probability that a given proto-onc gene is activated per translocation would be 5×10⁻³ (50 kb out of 10⁶kb). The probability that one of the 20 known proto-onc genes is activated would then be 10⁻³ per translocation.

Translocation frequencies per mitosis are not readily available. However in hamster cells they are estimated to occur with a probability of 10⁻⁶ per mitosis (Kraemer et al. 1986; Ray et al. 1986). In mice and humans even higher frequencies of 10⁻² to 3 × 10 ¹ have been reported in somatic cells studied in vitro (Terzi & Hawkins, 1975; Harnden et al. 1976; Martin et al. 1985). The probability of a translocation per meiotic cell division in humans has been determined to be 10⁻³ to 10⁻⁴ based on chromosome abnormalities in live births (Hook, 1985). Assuming one translocation in 10⁴ mitoses, the probability that 1 out of the 20 known proto-onc genes is activated per mitosis by translocation would then be about 10⁻⁷.

(4) The probability that a proto-onc gene would be activated from without by the promoter or enhancer of a retrovirus integrated nearby is even higher than those associated with the intrinsic mechanisms. Since retrovirus integration within 1 to 10 kb of a putative latent cancer gene is considered sufficient for activation (‘RNA Tumor Viruses’ 1985; Bishop, 1981; Klein, 1981; Bishop et al. 1982; Bishop, 1983; Varmus & Bishop, 1986), and since retrovirus integration is not site-specific (‘RNA Tumor Viruses’ 1982, 1985) and since eukaryotes contain about 10⁶kb of DNA, a given proto-onc gene would be activated in at least 1 out of 10⁶ infected cells (Duesberg, 1987, 1985). The probability that any one of the 20 known proto-onc genes would be activated would be 2× 10⁻³ per infected cell.

The sum of these probabilities should reflect the spontaneous transformation frequency of cells per mitosis in vivo and in vitro. It would be between 10⁻⁵ and 10⁷. However, it should be at least 5×10⁻⁷ due to proto-ras mutations alone. The actual number may be 10 times lower or about 10⁻⁸, depending on whether all or only some of these four putative mechanisms could activate a proto-one gene and depending on whether a given cell is susceptible to transformation by a given one gene or to a given retrovirus. Instead, spontaneous transformation per mitosis that leads to monoclonal tumours is only about 2×10 ¹⁷in vivo. Thus the expected probability of spontaneous transformation due to activation of preexisting oncogenes differs at least by a factor of 10⁹ from that observed in diploid cells in vivo.

Again it may be argued that spontaneous malignant transformation does indeed occur at this rate but that immunity eliminates nearly all transformants. However in this case athymic or nude mice should not exist (Sharkey & Fogh 1984) and diploid cells in culture would transform at the above rate.

In view of the consistent difficulties in demonstrating oncogenic function of proto-onc genes, a further revision of the oncogene concept has recently been favoured. It proposes that ‘activated’ proto-onc genes, like proto-ra.v or proto-nzyc, are not autonomous one genes like their viral derivatives, but are at least necessary for the kind of carcinogenesis that requires multiple cooperating oncogenes (Leder et al. 1983; Land et al. 1983; Klein & Klein, 1984; Balmain, 1985; Diamond et al. 1983; Varmus, 1984). Thus activated proto-onc genes are proposed to be functionally different, yet structurally equivalent to viral one genes. According to this theory, activated proto-onc genes would not be expected to register in transformation assays that detect single-hit carcinogens like viral one gnes (Duesberg, 1983, 1985).

However, the hypothesis fails to provide even a speculative explanation for why activated proto-onc genes are no longer to be considered functionally equivalent to viral one genes (Duesberg, 1985). Clearly, until the postulated complementary cancer genes are identified, this hypothesis remains unproven (Duesberg, 1985).

On the basis of genetic and structural analyses of retroviral genes, viral one genes and proto-onc genes and direct comparisons between them, we have found that viral one genes and proto-onc genes differ both structurally and functionally. Therefore we have proposed that viral one genes are indeed new genes, that do not preexist in normal cells, rather than being transduced cellular genes (Duesberg, 1983, 1985, 1979; Watson et al. 1983; Fig. 1).

The original basis for this proposal was the definition of the transforming gene of avian carcinoma virus MC29 (Duesberg et al. 1977) as a genetic hybrid, rather than a transduced cellular oncogene (Mellon et al. 1978). It consists of 5’ regulatory and coding elements (Δgag) from an avian retrovirus linked to 3’ coding elements from cellular proto-myc (Mellon et al. 1978; Fig. 1). Initially this became evident by comparing the structure and map order of MC29 with that of the three essential retrovirus genes, namely 5′ gag-pol-env 3′ (Wang et al. 1975; Wang, 1978; Fig. 1).

Sequence comparison of the viral Δgag-myc gene with the chicken proto-myc gene provided direct proof that only a truncated proto-myc gene was present in MC29. Indeed a complete 5′ proto-myc exon was missing from the viral Δgag-myc gene (Watson et al. 1983). This was apparently not an accident since the same 5′ proto-myc exon was also missing in the three other myc-containing avian carcinoma viruses MH2 (Kan et al. 1984; Zhou et al. 1985), CMII and OKIO (Duesberg, 1985; Hayflick et al. 1985). Thus a viral and a cellular gene functioned as progenitors or proto-onc genes of each of the viral recombinant myc genes (Fig. 1). More recently the four known viral ras genes were each also shown to lack a 5′ proto-ras exon (Cichutek & Duesberg, 1986) (see above; Fig. 1).

Comparisons between the one genes of other retroviruses and the corresponding proto-onc genes proved that all viral one are new genes that are spliced together from proto-owc genes and retroviral genes (Duesberg, 1983, 1985; ‘RNA Tumor Viruses’ 1985; Fig. 1). However, not all viral genes encode hybrid proteins. Based on the origin of their coding elements, the viral one genes can be divided into the four groups illustrated in Fig. 1:

(1) Those with amino and carboxy terminal domains from retroviruses and central domains from proto-owc genes. The one gene avian myeloblastosis virus (AMV) is the prototype (‘RNA Tumor Viruses’ 1985; Hayflick et al. 1985).

(2) Those with amino terminal domains from viral genes and carboxy terminal domains from proto-onc genes. The Δgag-myc gene of MC29 is the original example (see above). The one genes of Fujinami sarcoma virus (Lee et al. 1980) and Abelson leukemia virus (‘RNA Tumor Viruses’ 1985) also have the generic Agng-X structure.

(3) Those that are colinear with a reading frame of a proto-OMC gene. The ras genes of Harvey and Balb murine sarcoma virus (Cichutek & Duesberg, 1986) and the myc gene of avian carcinoma virus MHZ (Zhou et al. 1985) are examples.

(4) Those with an amino terminal domain from a proto-one gene and a carboxy terminal domain from the virus. The src gene of RSV is the prototype (Duesberg, 1983; ‘RNA Tumor Viruses’ 1985).

Since three of the four groups of recombinant viral one genes also encode recombinant proteins, their specific transforming function can be directly related to their specific structure compared to that of proto-one gene products. The transforming function of the recombinant one genes of group 3 which encode transforming proteins that are colinear with proteins encoded by proto-one genes cannot be explained in this fashion. However all viral one genes of this group each lack at least one proto-onc-specific 5′ exon like the avian carcinoma viruses with myc genes (Duesberg, 1985; Watson et al. 1983; Kan et al. 1984; Zhou et al. 1985; Hayflick et al. 1985) or the murine sarcoma viruses with ras genes (Cichutek & Duesberg, 1986). Conceivably elimination of transcribed or untranscribed suppressors or elimination of an upstream proto-ras cistron (Cichutek & Duesberg, 1986) or proto-myc cistron (Bentley & Groudine, 1986) and recombination with viral promoters are the mechanisms that generate transforming function (Fig. 1).

It follows that viral one genes and the corresponding proto-one genes are not isogenic. Viral one genes are hybrid genes that consist of truncated proto-one genes recombined with regulatory and frequently with coding elements from truncated retroviral genes. These consistent structural differences must be the reason why viral one genes inevitably transform and why proto-one genes are not transforming although they are present in all and are active in most normal cells (Duesberg, 1983, 1985).

It may be argued that proto-one gene truncations reflect packaging restrictions of transducing retroviruses, rather than conditions to activate proto-one genes. Such restrictions would have to be mostly sequence-specific, as most retroviruses with one genes can accommodate more RNA, at least 10 kb as in RSV (Duesberg & Vogt, 1970), than they actually contain, namely 3 to 8 kb (‘RNA Tumor Viruses’ 1985). However, there is no evidence that retroviruses discriminate more against certain transduced or artificially introduced sequences (‘RNA Tumor Viruses’ 1985) than against others, because retroviruses can accommodate very heterogeneous sequences, such as the 20 different transformation-specific sequences (Duesberg, 1983, 1985; ‘RNA Tumor Viruses’ 1985; Duesberg, 1979). Yet all nonessential sequences of retroviruses are unstable (Duesberg, 1983, 1985) unless selected for a given function.

The experience that all known viral one genes are new genes that include only truncated proto-onc genes is in itself an argument in favour of truncation as a condition for proto-onc gene activation and against preexisting cellular oncogenes. Clearly if cellular oncogenes preexist in normal cells, it would be much more likfely to find retroviruses with intact cellular oncogenes then retroviruses with new one genes put together from unrelated and truncated viral and cellular genes by illegitimate recombination. Moreover, the same exons were truncated from the same proto-onc genes in completely independent viral transductions. Examples are selective transductions from proto-myc, the precursor of four avian carcinoma viruses (Duesberg, 1985; Watson et al. 1983; Kan et al. 1984) or proto-ras, the precursor of three murine sarcoma viruses (Cichutek & Duesberg, 1986) or proto-myp, the precursor of avian myeloblastosis and erythroblastosis viruses (‘RNA Tumor Viruses’ 1985; Nunn et al. 1983) or proto-erb, the precursor of three avian sarcoma and erythroblastosis viruses (‘RNA Tumor Viruses’ 1985) or proto-fps, the precursor of three feline sarcoma viruses (‘RNA Tumor Viruses’ 1985) or proto-fps, the precursor of three avian sarcoma viruses (‘RNA Tumor Viruses’ 1985) or proto-abl, the precursor of Abelson murine leukemia and a feline sarcoma virus (‘RNA Tumor Viruses’ 1985), or protomos, the precursor of several Moloney sarcoma virus (‘RNA Tumor Viruses’ 1985; van der Hoorn et al. 1986) or proto-src, the precursor of RSV and two other avian sarcoma viruses (Ikawa et al. 1986). In the process of generating viral one genes from these proto-onc genes, the same exons were in some cases selectively truncated at the same breakpoints, as for example the one genes of two sarcoma viruses generated from proto-fps (Pfaff et al. 1985).

The existence of at least five retroviruses that have transduced proto-orac sequences that had already been truncated and rearranged with other cellular genes prior to transduction, lends further independent support to this view. Examples are the one genes of RSV (Duesberg, 1983; ‘RNA Tumor Viruses’ 1985) of avian carcinoma virus MHZ (Duesberg, 1985; Zhou et al. 1985; Hayflick et al. 1985), of avian erythroblastosis and sarcoma virus AEV (‘RNA Tumor Viruses’ 1985) of avian erythro- and myeloblastosis virus E26 (Nunn et al. 1983) and of the feline sarcoma virus GR-FeSV (‘RNA Tumor Viruses’ 1982; Naharro et al. 1984). Certainly the odds against transduction of rare rearranged proto-onc genes instead of normal proto-onc genes are overwhelming. Yet five out of the less than 50 known isolates of retroviruses with one genes (‘RNA Tumor Viruses’ 1985) contain previously rearranged proto-onc sequences, most likely because truncation is necessary for transforming function. Indeed, it may be argued that these viruses have transduced these rearranged proto-onc genes from a preexisting tumour that was generated by these rearrangements. Thus, the rearranged proto-onc genes of these five oncogenic retroviruses may be ‘transduced cellular oncogenes’ after all.

A definitive assessment of why viral one genes transform and cellular proto-onc genes don’t, requires more than comparisons of prijnary structures and transforming tests with DNAs. It will be necessary to know what proto-onc genes do and whether they encode proteins that function alone or as complexes with other proteins.

It is proposed then that proto-onc genes that are transcriptionally activated or have undergone point mutations, but retain a germline structure, are not cellular cancer genes. The hypothesis that proto-onc genes are latent cellular cancer genes that can be converted to active transforming genes by increasing dosage or function, is suggested to be an overinterpretation of sequence homology to structural and functional homology with viral one genes.

This proposal readily resolves the paradoxes posed by the hypothesis that proto-onc genes are latent cellular cancer genes that can be activated by enhanced expression or point mutation. The proposal accounts for the frequent expression of proto-onc genes in normal cells (Duesberg, 1985). The proposal is also entirely consistent with the lack of transforming function of ‘activated’ proto-onc genes from tumours. The fact that mutated proto-ras changes the morphology and enhances tumorigenicity of aneuploid and tumorigenic 3T3 cells is an important observation, but not an exception to the experience that native proto-onc genes from tumours analysed to date do not transform diploid cells. The proposal also provides a rationale for the chromosome abnormalities of tumour cells, as these appear to be microscopic evidence for cancer genes (see below), instead of the ‘activated’ proto-onc genes identified to date.

The proposal that proto-onc genes derive transforming function by truncation and recombination with retroviral genes predicts that illegitimate recombination among cellular genes could also generate transforming genes from proto-onc genes. The generation of retroviral onc genes from viral genes and proto-onc genes could indeed be a model for this process. The view that cellular cancer genes are rare illegitimate recombinants of normal cellular genes is in accord with the fact that rearranged and abnormal chromosomes are the only consistent, transformation-specific markers of tumour cells (Wolman, 1983; Rowley, 1984; Trent, 1984; Duesberg, 1987; Levan, 1956). Further the clonality of chromosome alterations, e.g., the marker chromosomes of tumours (Wolman, 1983; Rowley, 1984; Trent, 1984; Duesberg, 1987; Levan, 1956), indicates that tumours are initiated with and possibly caused by such abnormalities as originally proposed by T. Boveri in 1914.

As yet less than 50 isolates of retroviruses with one genes have been recorded in history (Duesberg, 1985; ‘RNA Tumor Viruses’ 1985), although both potential parents of retroviral one genes are available in many animal or human cells because retroviruses are widespread in all vertebrates (Duesberg, 1987; ‘RNA Tumor Viruses’ 1985). This extremely low birth rate of retroviruses with one genes must then reflect the low probability of generating de now an oncogenic retrovirus from a proto-onc gene and a retrovirus by truncating and recombining viral and cellular genes via illegitimate recombinations (Duesberg, 1983, 1985, 1979). Clearly, two illegitimate recombinations are required (Fig. 1). One to link a 3 ′ truncated retrovirus with a 5 ′ truncated proto-onc gene; the other to break and then splice the resulting hybrid one gene to the 3 ′ part of the retroviral vector.

The first of these steps would already generate a ‘cellular’ cancer gene that ought to be sufficient for carcinogenesis. The birth of such a gene would be more probable than that of an oncogenic retrovirus that requires two illegitimate recombinations, but it would be harder to detect than a complete replicating retrovirus with an one gene. Nevertheless even this would be a rare event. Given that such a recombination would have to take place within the 8 to 9 kb of a retrovirus (Fig. 1) integrated into the 10⁶-kb genome of a eukaryotic cell and also within an estimated 1 to 2 kb of a proto-owc gene (Fig. 1) and assuming that translocation or rearrangement occurs with a probability of 10∼⁴ (see above); the probability of such a recombination per mitosis would be 8×10⁻⁶ 2×10∼⁶ 10⁻⁴ or 10⁻¹⁵. That a second illegitimate recombination is required to generate a retrovirus with an one gene would explain why the occurrence of these viruses is much less frequent than spontaneous transformation. This probability may, nevertheless, be higher than the square of 10⁻¹³, since the two events may be linked and since multiple integrated and unintegrated proviruses exist in most infected cells.

The probability that rearrangements would generate cancer genes from normal cellular genes would also be very low, since most rearrangements would inactivate genes. The above estimates for the probability of spontaneous transformation of 2× 10⁻¹⁷ per mitosis and of translocation of 10⁴, which would be a minimal estimate for rearrangements, suggest that 10¹³ translocations or rearrangements are needed to generate a transforming gene that causes a monoclonal tumour. This could either be a single autonomous transforming gene that is like a viral one gene, or it could be a series of mutually dependent transforming genes that would each arise with a higher probability than an autonomous one gene. The facts that multiple chromosome alterations are typically seen in tumours (Wolman, 1983; Rowley, 1984; Trent, 1984; Levan, 1956) and that as yet no DNAs have been isolated from tumours that transform diploid cells with single-hit kinetics, suggest that most cellular cancer genes are indeed not autonomous carcinogens like viral one genes. It is consistent with this experience that most cellular genes are also not converted to autonomous cancer genes by retroviral transduction and truncation. Only about 20 cellular genes, the proto-one genes, have been converted to autonomous viral one genes, although viral transduction is a random event that does not benefit from sequence homology between retroviruses and cells (Duesberg, 1983, 1985, 1979).

Thus viral one genes have not as yet fingered preexisting cellular cancer genes. But they appear to be models for how cancer genes may arise from normal cellular genes by rare truncations and recombination.

I would like to thank S. A. Aaronson, M. Kraus, M. Pech, K. Robbins, S. Tronick and others from the Laboratory of Cellular and Molecular Biology, National Cancer Insitute, Bethesda, Maryland for critical and amusing discussions and generous support during a sabbatical leave and B. Witkop, National Institute of Arthritis, Diabetes and Digestive and Kidney Diseases, Bethesda, Maryland for asking many of the basic questions that I try to answer in this manuscript. I also thank my colleagues H. Rubin for encouragement and K. Cichutek, R.-P. Zhou, D. Goodrich, S. Pfaff and W. Phares, University of California, Berkeley, California for inspiring comments and their work. Supported by (OIG) National Cancer Institute Grant CA-39915A-01 and Council for Tobacco Research Grant 1547 and by a Scholarship-in-Residence of the Fogarty International Center, NIH, Bethesda, Maryland.