RESULTS AND DISCUSSION
Table 1 shows the segregation analyses results when all relatives with Major Affective Disorders and Affective Spectrum Disorders are included. These results thus represent the maximum likelihood estimates for each group of families according to the different transmission hypotheses. The likelihood values within each group can be compared, but it is not possible to make intergroup comparisons because the absolute values of the likelihoods are related to the number of subjects in the three groups, all of which vary in this respect. The discussed hypotheses are those for which the program could reach a maximum likelihood estimate. The estimated likelihood values obtained allow us to rule out the hypothesis of no genetic transmission. For the group of tricyclic drug responders, the Mendelian model appears to be the most suitable. A good fit also is obtained under the SML additive transmission hypothesis. The model that best fits the group of poor responders is the polygenic one, whereas for the nonresponders, the best fit appears to be the SML hypothesis with transmission probability estimate (for heterozygotic genotype). The incorporated population prevalence values were calculated from the data 
of a previous study (9) on the presence of Major Affective Disorders and Affective Spectrum Disorders among relatives of a control group. Our segregation analyses were carried out with a range of general population prevalence estimates, but the magnitude of the likelihood values and the selection of the best fit solution did not substantially change. These findings confirm that the pharmacological criterion of antidepressant tricyclic drug treatment outcome really allows us to identify different familial segregation structures of Affective Disorders. Looking at the parametric set estimated by the best fitting hypothesis for each group we have seen that the Mendelian model was the best suited to the families of good responders to tricyclics (Table 2). The parameter d represents the position of the heterozygous mean relative to the means of the two homozygous classes. In other words, this parameter expresses the degree of dominance. The model chooses a d value consistent with a codominant SML hypothesis (the only affected genotype is the homozygous one). The parameter t expresses the displacement between the two homozygous genotype means (in standard deviation units), whereas the q parameter is the gene frequency of the allele for the disorder. The value observed here is a very low one. A possible explanation is that a bias occurred subdividing our sample into three groups. It is possible that the group of poor responders, whose best fitting model was the polygenic transmission hypothesis, actually could have been heterogeneous and included patients who really belonged to the tricyclics responder group. Another possible explanation is an underestimation of the population prevalence values because only families of patients with Major Affective Disorders were included in the sample, although we did not take probands with Affective Spectrum Disorders into account. For the group of nonresponders, the most suitable solution was the SML dominant hypothesis with penetrance defect (Table 3). The value of d indicates the complete dominance: both homozygotics and heterozygotics are affected genotypes. The gene frequency q is rather low again and T2, the transmission probability, expresses the penetrance defect. However, we must remember that the polygenic hypothesis of transmission could not be completely ruled out. For the group of poor responders to tricyclics the best fitting model was the polygenic one (Table 4). The parameter H denotes the proportion of the total phenotypic variation belonging to the multifactorial transmissible component reflecting both genetic and environmental effects. Heritability reflects genetic transmission not ascribed to the major gene as well as cultural transmission. The
parent/offspring correlation for the multifactorial transmissible component is assumed to be equal to 0.5. The high estimated heritability could be due to the presence of assortative mating or fertility defect. We can also infer that this group of families is still heterogeneous. The application of segregation analysis appears to confirm the existence of a genetic heterogeneity that cannot be completely detected when the clinical criterion of tricyclic antidepressant drug treatment outcome is employed alone. On the other hand, experimental data have suggested that there are links between the genetic mechanisms underlying Affective Disorder and those at play in determining responses to lithium. Patients who are helped by lithium therapy are much more likely to have a family history of mania or depression (2,8) and their relatives have a higher morbidity risk for Affective Disorders than those of patients who do not benefit from lithium. In our sample, 73 patients had been under long-term lithium treatment for at least 3 years. Forty-eight had had no relapses. Twenty-five patients, who had one or more definite major affective episodes while on therapy, were classified as relapsed or nonresponders. Table 5 shows that patients who are good tricyclics responders also have good lithium treatment outcomes, whereas those who do not respond to tricyclics also have poor responses to lithium. The group of poor tricyclics responders occupies an intermediate position as regards lithium outcome. In this group polarity seems to be a good discriminator since bipolar patients have good lithium outcomes, as do good tricyclics responders, although unipolar patients behave like tricyclics nonresponders. 
The data on outcome on tricyclics and lithium appear to be overlapping, and both are useful for identifying homogeneous subgroups of Affective Disorders from the genetic viewpoint. However, we must remember that negative outcomes are very complex events in susceptible patients and are related to biological or psychic conditions as well as to life events. Finally, it is interesting to observe the distribution of the Affective Spectrum Disorders in the three different family groups (Table 6). These diagnoses are not homogeneously distributed in the three groups. We found an excess of Dysthymic Disorders, Cyclothymic Disorders, and Atypical Depressions among the relatives of the good tricyclics responders. We can suppose that homozygous relatives show the complete phenotypical manifestation of Major Affective Disorders whereas the heterozygous ones manifest the intermediate phenotype, i.e., an Affective Spectrum Disorder. To test this hypothesis and to better understand the mode of inheritance of 
the disease it will be useful to compare the results of the present segregation analyses with those obtained excluding Affective Spectrum Disorders diagnoses from the definition of the affected phenotype.