Anticipation in bipolar affective disorder: Is age at onset a valid criterion?

INTRODUCTION

Anticipation, a clinical phenomenon characterized by a significantly earlier age at onset (AO) and increasingly severe phenotype in younger generations for certain diseases, has been a topic of controversy since the earliest studies of familial inheritance in psychiatric disorders [Mott, 1910]. After much debate, it was later dismissed as a statistical artifact and a result of several ascertainment biases for which there was no biological basis [Penrose, 1948]. The discovery of unstable DNA trinucleotide repeat (TNR) sequences has, however, renewed interest in the subject. Specifically, TNR sequences appear to expand in length in successive generations of affected family members. In several disorders, these expansions correlate with an increase in disease severity and earlier age at onset as predicted by the anticipation model [see Ross et al., 1993, for a review]. More recently, it has been suggested that anticipation and TNR expansions might play a role in the inheritance of psychiatric disorders such as schizophrenia and bipolar disorder (BD) [McInnis et al., 1993; Bassett and Honer, 1994].

The hypothesis of anticipation in BD has been tested in a number of genomic searches for unstable TNR expansions. The evidence from studies thus far has proven to be conflicting at times, and has failed to resolve the issue conclusively. Initial studies using the repeat expansion detection (RED) technique [Schalling et al., 1993] found an association between longer CAG repeats and bipolar disorder [Lindblad et al., 1995; O'Donovan et al., 1995]. Further studies have also found that an increase in mean trinucleotide CAG repeat length between parental and offspring generations is associated with both an intergenerational increase in phenotype severity and a decrease in AO [Grigoriou-Serbanescu et al., 1997; Mendlewicz et al., 1997; Ohara et al., 1998]. These studies have been countered by a number of others that have both failed to find an association between a long CAG/CTG repeat and BD, and failed to detect a correlation between TNRs, increased illness severity, and earlier AO [Craddock et al., 1997; Li et al., 1998; Zander et al., 1998]. It is also interesting to note that a majority of large CAG/CTG repeats occur at chromosomes 18q.21.1 and 17q21.3, two regions that have previously not been associated with the disorder [Vincent et al., 1999].

In the absence of a definitive explanation for the anticipation effect from molecular genetic studies, it is important to consider once again the contribution of biases to findings of anticipation, especially with regard to the age at onset. These include: 1) right truncation of the AO distribution [Hodge and Wickramaratne, 1995; Heiman et al., 1996]; 2) lower marriage and fertility rates in affected patients [Bassett and Honer, 1994]; 3) a bias against inclusion of early onset parent–late onset child given the time lapse between the onsets and limited time in which to conduct the study; and 4) multiplex ascertainment—the preferential inclusion of families selected for linkage analyses based on the high density of the illness often associated with early onset [McInnis et al., 1993; Vieland and Huang, 1998]. Other confounding factors include the cohort effect [Klerman et al., 1985], bilineality [McInnis et al., 1993; McMahon et al., 1994], and inaccurate recall of illness onset by older subjects [Rice et al., 1987]. Although many of these biases may have been taken into account in earlier studies of anticipation in psychiatric disorders, they were usually adjusted for one at a time in the analyses. Consequently, the combined effect of multiple sources of bias in determining AO in psychiatric disorders is, as yet, unknown. This investigation was conducted to assess whether AO by itself was a valid criterion to determine the presence of anticipation in BD.

In order to compare the true effect of anticipation with that produced by one or more of the confounding factors, we examined and compared data both from families and from unrelated individuals. We assumed that the effects of the confounding factors would be similar when applied to either of the two groups, but anticipation, as measured by a significantly earlier AO in successive generations, would be present only in the family data.

METHODS

RESULTS

The results of the analysis are shown in Table II. Although there was a significant difference in the AO between successive generations, this difference was of similar magnitude in the familial and “unrelated” samples. This is further supported by a nonsignificant “sample × generation” interaction. Thus, the difference in AO between G1 and G2 in the family sample was not significantly different from the difference in AO in the unrelated subjects.

DISCUSSION

The results of this study argue against a strong effect of anticipation in the observed intergenerational difference in AO of BD. We found that the AOs were not significantly different between the familial sample and the sample of unrelated patients. This finding was further confirmed by a nonsignificant interaction between sample and generation, indicating that the intergenerational AOs for each sample were not significantly different from each other. In fact, the difference in AO between G1 and G2 of the unrelated individuals is marginally greater than in the family sample (9.2 vs. 6.2 years). These results are not in accordance with the prediction of the anticipation model or with previous findings [McInnis et al., 1993; Nylander et al., 1994].

Both samples were drawn from very similar populations and were therefore subject to similar biases, such as the right truncation of AO distributions and a relative lack of late onset G1 – early onset G2 pairs. Although previous studies on anticipation have examined some of the confounding factors in post-hoc analyses, they did not consider the compound effect of several such biases acting together. The advantage of the method used is that it enabled us to control for the combined effect of the ascertainment biases. Specifically, the method of comparing family data with epidemiological data from unrelated subjects allowed for other confounding effects to be held constant. As such, the independent variable, familial inheritance, could be examined more accurately.

The results do, however, show a significant difference in AO between G1 and G2 for both related and unrelated samples. Although the statistical analysis shows that this result is not due to anticipation, the exact contribution of the other factors cannot be determined. We do know that the anticipation effect can be mimicked by a number of factors. Cohort and period effect [Klerman et al., 1985], inaccurate recall of onset (where older individuals tend to report a later than actual age at onset due to forgetting of earlier and possibly milder episodes) [Rice et al., 1987], and age-at-interview bias [Heiman et al., 1996] may all have contributed in some way to the significant difference in AO between G1 and G2 for both related and unrelated samples. It is impossible, however, in this method to isolate the contribution of one such factor from another.

In conclusion, our study does not disprove the existence of anticipation in BD. Rather, it suggests that for positive proof of anticipation AO alone is not a sufficient criterion. Other indicators, such as disease severity, also need to be examined.