eral laboratories (Cavenee et al. 1983; Friend et al. 1986; Lee et al. 1987; Bookstein et al. 1988). Technically, then, inherited retinoblastoma is a recessive trait. In some affected children (in fact, the majority), on the other hand, both of the necessary events occur in a somatic cell (e.g., retinoblast, osteocyte), as a consequence of which the child is now homozygous for an abnormality at this locus in this cell; but, although the child develops a malignant tumor, he or she does not transmit an altered allele to the next generation because the germ line is not affected. Parental radiation would not be expected to increase the frequency of this type of retinoblastoma.

Wilms tumor also rather clearly meets the specifications of this model (reviewed in Matsunaga 1981; Dao et al. 1987), as, probably, does neuroblastoma (Knudson and Meadows 1978; Bundey and Evans 1982). Although the evidence is less extensive, childhood gonadal dysgerminoma, pheochromocytoma, hepatoblastoma, and, more problematically, rhabdomyosarcoma and the central nervous system tumors may have a similar but less pronounced genetic basis (reviewed in Knudson and Meadows 1978; Koufos et al. 1985; Schimke 1985; Knudson 1986). A further argument for the genetic basis of some of these tumors is the reportedly more frequent occurrence of other malignancies in the parents of affected children (Hartley et al. 1986; Li et al. 1988).

The leukemias and the malignant lymphomas of childhood have not yet been shown to exhibit a pattern comparable to retinoblastoma or Wilms tumor (Videbaek 1947; Steinberg 1960), but, because of the poor survival of affected children in the past, the appropriate systematic genetic data to support an opinion concerning the proportion of the leukemias consistent with the above-described model are not yet available. There are, however, anecdotal data concerning familial clusterings of childhood leukemia (see Gunz et al. 1978), and the risk of leukemia in siblings of children with leukemia is increased about two- to fourfold (Miller 1971; Draper et al. 1977). Furthermore, concordance rates in identical twins are about 20% (MacMahon and Levy 1964). The observation that in more than half the concordant twin pairs described in the literature the leukemia was diagnosed during the first year of life, whereas overall the peak incidence is between 3 and 5 years of age (Zuelzer and Cox 1969), is consistent with (but does not demand) the hypothesis that the very-early-onset cases reflect a mutational event in one of the parents.

In recent years a variety of approaches has led to the recognition of a second class of genes concerned with carcinogenesis, generically termed proto-oncogenes, which as a result of a mutational event in a somatic cell become the basis for a clonally derived malignancy. A related phenomenon is the implication of precise chromosomal regions, including fragile sites, in the chromosomal breakpoint associated with various clonally derived leukemias, lymphomas, and other malignancies. The number of such proto-oncogenes is thought to be of the order of 50 (reviewed in Bishop 1987; Human Gene Mapping 9 1987), and the list continues to grow, although hard data relating many of those protooncogenes to the malignant transformation are still lacking, and there is evidence that, with respect to the childhood malignancies, the number could be relatively limited (Koufos et al. 1985). We are not aware of any clear evidence that a transmitted (germ-line) mutation in a proto-oncogene can be the basis for a childhood malignancy, but, if there are such mutations, in principle they should be no less responsive to radiation than are mutations in tumor-suppressor genes.

Finally, Batra and Sridharan (1964) reported an increase in leukemia persisting over 4 generations in the offspring of radiated mice, but Kohn et al. (1965) saw no increase in leukemia or other tumors under comparable experimental circumstances. More recently, Nomura (1982, 1983, 1986) has reported an increase in the frequency of a variety of adult-onset-type tumors in the offspring of irradiated male mice, an increased frequency which persisted for the several subsequent generations during which the strain was observed. This latter observation created an experimental precedent which in part motivated the present study. Moreover, Shiono et al. (1980) have reported a relative risk of childhood malignancy of 2.61 in the offspring of women who received preconception diagnostic X-ray exposures to the ovaries, the risk being computed from a comparison with matched controls (P=.021). An estimated mean ovarian exposure was not given for these women, but it may be presumed to be <0.01 gray (Gy).

Our hypothesis in investigating the possibility of an increase in malignancy in the children of survivors of the atomic bombings was that the proto-oncogenes and tumor-suppressor genes collectively constitute a sufficient target such that the frequency of cancer of relatively early onset might be a suitable indicator of the genetic effects of the atomic bombs, a thesis reinforced by the experimental findings just cited.