RESULTS
Looking at our present data we can see the total lod-score values as a function of THETA for the hypotheses of a linkage with HLA В locus and A and В loci together (haplotypes). Figure 1 shows the findings that we observed for all our families tested. The 
two curves each represent the sum of the lod-scores at the given THETA values obtained by the various families represented in the sample, so that the two curves are a graphical representation of total mean values. It should be noted that both the hypotheses of random assortment of the two loci (THETA = 0.5) and complete identity (THETA = 0.0) are unlikely, whereas the best lod-scores are comprised in a range of THETA values from 1.5 to 3.0, therefore indicating a mild to moderate physical genetic distance of the two loci (MDI and HLA) on the same chromosome. Even though numerical analysis does not offer exciting results (we lie in a sort of uncertainty zone where linkage could neither be strongly confirmed nor be ruled out), it can be seen that the hypothesis of a linkage with the В locus is better than with the haplotypic structure. Nevertheless, peaks of the two curves, corresponding to a kind of maximization of the test, are not too far from each other, thus corroborating a biological resemblance of the two different hypotheses. We are still using a model of para-metrical analysis in which parameters are the genetic determinants of the marker locus and of the susceptibility (=MDI) locus. Consequently, when we look at the results obtainable for single families, we can observe a more complex pattern: The different behavior of the three families presented in Fig. 2 reveals the presence of heterogeneity that in this particular case is mainly, or exclusively, genetic in origin since it is based only on genetic parameters (such as gene frequency, penetrance variability). 
The exclusion of families with the worst results (for example, the dotted line family) may improve the total lod-score because it is likely that such families behave as having no linkage with the HLA complex. Moreover, among the various parameters we used, penetrance reveals the higher influence on linkage estimates. In fact, we know that a widely accepted way of measuring the penetrance defect of a complex character is to correlate the penetrance defect itself with the variability in age at onset distribution and to use this distribution for deriving, in a function too complex for description here, an age-corrected weight of every subject involved in the analysis (4). Figure 3 describes the effect of varying the ranges of age at onset distribution over the THETA estimates and also over the Log-Likelihoods values that represent the absolute probabilistic value for a given family. Thus, we must pay careful attention to treating and modifying the genetic parameters, and it should be remembered that the linkage models we used were originally built to study the simplest Mendelian all-or-none traits. When we modify these models to analyze non-Mendelian characters, methodological problems, as we pointed out earlier, may lower the reliability of the lod-score method. Finally, we used only data and genetic parameters strictly derived by experimental approaches without any modified simplification in order to obtain better results. In spite of all this, a comparison of our experimental data with the simulation data of Kruger et al. (6) shows a higher level of absolute probabilistic value that can be seen as an indirect proof of the good reliability of our results (Fig. 4). 
In conclusion a strong support of our findings comes from the results obtained by Smeraldi and Bellodi (8) who, using the Generalized Sib Pair Identity by Descent distribution as an alternative model to study linkage, were able to detect an excess of HLA similarities in doubly affected sibs and a lack of similarities in discordant sibs, thus again showing that HLA may be truly linked with a Primary Affective Disorders susceptibility locus.