Anticipation describes a phenomenon in which an inherited disease is diagnosed at an earlier age in successive generations of a family. In addition, increased severity of the disease may be observed in the affected individuals of the younger generation. Initially, the concept of anticipation was dismissed as an artifact of ascertainment bias ( Penrose, 1948). However, a genetic explanation for anticipation was recently provided by the identification of the gene responsible for fragile X syndrome and the discovery that an uncontrolled expansion of a CCG trinucleotide repeat in the gene led to the disorder ( Kremer et al, 1991 ; Verkerk et al, 1991 ; Yu et al, 1992 ). Shortly afterwards, additional discoveries of such novel genetic mutations were made in other inherited diseases. Trinucleotide repeats CAG, CTG and CAG were found in genes responsible for spinal cerebella ataxia ( Orr et al, 1993 ), myotonic dystrophy ( Brook et al, 1992 ; Fu et al, 1992 ) and Huntington's disease ( Koide et al, 1994 ) respectively. In other diseases, such as familial schizophrenia ( Sasaki et al, 1996 ), truncal heart defects ( Bleyl et al, 1995 ), familial myeloma ( Deshpande et al, 1998 ) and familial leukaemia ( Horwitz et al, 1996 ; Yuille et al, 1998 ), anticipation has been demonstrated, but the exact molecular mechanisms have not been determined. We have collected a large series of patients with familial haematological malignancies, and this report demonstrates that familial non-Hodgkin's lymphoma is characterized by anticipation.
Anticipation, a phenomenon in which an inherited disease is diagnosed at an earlier age in each successive generation of a family, has been demonstrated in certain neurological and haematological disorders. This study was conducted to determine whether anticipation occurs in familial non-Hodgkin's lymphoma (NHL). Eleven published reports of multigenerational familial NHL were analysed for evidence of anticipation, together with 18 previously unreported families with familial NHL. Differences in disease-free survival between generations were determined. The difference between age at onset for each affected parent–child pair was tested against the null hypothesis that there was no difference in age at onset. These analyses were also performed separately using only parent–child pairs with age of onset > 25 years to avoid ascertainment bias. In addition, the age at onset distribution of the studied cases was compared with that of the Surveillance Epidemiology and End Results (SEER) Program using data for 1973–98. The median ages at onset in the child and parent generations of all families (48.5 and 71.3 years respectively) and in the selected pairs (52.5 and 71.5 years respectively) were significantly different (P < 0.000002 and P < 0.000001 respectively). The null hypothesis was rejected for all (P < 0.000001) as well as selected pairs (P < 0.000003). A significant difference was observed between the ages of onset of the child generation and the SEER population (P < 0.009), but not between the parent generation and the SEER population. Anticipation occurs in familial NHL, which suggests a genetic basis for it.
PATIENTS AND METHODS
Familial non-Hodgkin's lymphoma (NHL) cases were collected from two sources: the published medical literature ( Camino, 1975; Greene & Miller, 1978; Shelley, 1980; Greene et al, 1982 ; Haim et al, 1982 ; Veltri et al, 1983 ; Kato et al, 1989 ; Lynch et al, 1992 ; Hayoz et al, 1993 ; Donadieu et al, 1996 ; James et al, 1998 ) from 1926 to the present; and our series of unpublished families with multigenerational NHL. Additional information on one of the published families was obtained by personal communication with the author ( James et al, 1998 ). Our series consists of all multigenerational familial cases seen by the authors and all responses to several worldwide web sites in which details on familial haematological malignancies were solicited from physicians and patients. Families included in this study all met the following three criteria: (1) there were two or more affected individuals with NHL in the family; (2) there was one parent–child pair in the family; and (3) to avoid bias towards anticipation, all the affected individuals had been actually diagnosed and ‘obligate’ carriers were considered unaffected.
A total of 39 parent–child pairs were included in this study, 19 from the literature and 20 previously unpublished. The pedigrees of the previously unpublished families are shown in Fig 1.
Of the 18 previously unpublished families in this study, six were identified at the diagnosis of the parent and 12 were identified at the diagnosis of the offspring. The differences in calendar years of diagnosis and ages at diagnosis for all previously unpublished and published patients (when known) in each pair are shown in 12 Tables I and II.
The pathology slides of all previously unpublished families were reviewed, except for those of three parents whose slides were unavailable. The diagnosis was accepted for those three, as a confirming pathology report was available for each. While the histological subtype of NHL was frequently changed during the review, the diagnosis of NHL was confirmed in all material reviewed.
Three analyses were used to evaluate possible anticipation in the study population. Disease-free survival was estimated by the Kaplan–Meier method ( Lawless, 1982). Differences in disease-free survival between generations were then tested by the log rank test. For each affected parent–child pair, the difference in age at onset was tested against the null hypothesis that there was no difference in the age at onset, using a signed rank test ( Conover, 1980, pp. 280–288). These analyses were also performed separately for the selected parent–child pairs with age at onset > 25 years. Finally, the age at onset distribution was compared between the observed cases in this study and those in the general US population using 1990–94 data from the Surveillance Epidemiology and End Results (SEER) Program of the National Cancer Institute USA and 1990 US census data. The difference between the observed age distribution and that of the general population was tested by the Kolmogorov test ( Conover, 1980, pp. 334–357).
These statistical methods assume that the individuals in each pair are independent from individuals in other pairs. Therefore, in a family with multiple affected children, their parent is only counted once. To test the soundness of this assumption, one parent–child pair was selected randomly from each family. Furthermore, in the cumulative disease-free survival analysis, parental age at diagnosis data were censored after the age of diagnosis of the offspring. This was done to avoid the potential bias of longer length of follow-up for the older generation.
Three potential biases cited by Penrose (1948) were considered in this study. First, selection of parents with later onset of disease is favoured, as those with early onset would have reproductive difficulties. To avoid this pitfall, we removed all individuals under the age of 25 years and repeated the first two statistical analyses. Secondly, selection of children with early onset is favoured because of highly manifested symptoms. This bias may not be a factor when studying anticipation in non-Hodgkin's lymphoma, because the time interval between disease onset and the onset of signs and symptoms is usually short. Furthermore, this bias can also be interpreted as evidence for anticipation, as children may have a more aggressive onset of the disease. Thirdly, selection of cases with simultaneous onset in parent and child generations is preferred. To avoid this, a prospectively obtained case study on an extremely large population over a period of several generations is needed. However, such an ideal study would not be feasible for many reasons. Furthermore, Penrose's postulated ‘complementary pairs’, an important basis that supports this bias, have not been reported to date ( Harper et al, 1992 ).
Cumulative disease-free survival analysis for all pairs showed the median age at onset for individuals in the child and parent generations to be 48.5 and 71.3 years, respectively, and for selected pairs to be 52.5 and 71.5 years respectively. The log rank test rejected equality of the age at onset between the generations for both all (P < 0.000002) and selected pairs (P < 0.000001). Survival analysis curves for both all and selected pairs are shown in Fig 2. Mean age at onset was calculated for parent and child generations in both all and selected pairs. The signed rank test rejected the null hypothesis that there was no difference in the age at onset for both all (P < 0.000001) and selected pairs (P < 0.000003). A comparison of the age at onset distribution between the observed cases and the SEER population is shown in Fig 3; a significant difference was observed for the child (P < 0.009) but not for the parent generation.
All statistical analyses for the pairs in which only one parent–child pair was selected randomly from each family were repeated, and the parental information on cancer occurrence was censored after the offspring's age at diagnosis in the disease-free survival analyses. The log rank test still showed that parental age at diagnosis was significantly greater than that of the child for all and selected pairs (P < 0.004).
Potential biases in the study and approaches to mitigate these biases have already been discussed. As the results of this study were unlikely to be caused by ascertainment or other bias alone, they suggest that anticipation exists in familial NHL.
A possible non-genetic explanation for anticipation is that the parents and children have been simultaneously exposed to a causative environmental agent ( Rigby et al, 1968 ). Consequently, the difference in age at onset in individuals in the child and parent generations may only reflect the generational age difference. However, this fails to explain cases of multiple affected individuals in families who did not live together and those cases where parents were diagnosed with the illness decades after their children. Furthermore, there are no known environmental factors proven to be associated with the development of most lymphomas.
Possible explanations for the clinical phenomenon of anticipation have been suggested. Horwitz (1997) suggested that intergenerational inheritance of multiple downstream mutations in proto-oncogenes and tumour suppressor genes that are initiated by a primary defect in a DNA repair gene may be responsible and, more recently, he hypothesized that inherited abnormalities of telomeres may explain anticipation (M. Horwitz, personal communication, 1998).
Consistent with anticipation, substantial median (22.8 years) and mean (20.9 years) age at onset difference was seen between the generations in this study. Some interesting observations were noted when studying the distribution patterns of the SEER population and the familial cases observed here. The distribution pattern for the SEER population rose exponentially as age increased. The observed distribution in the parent generation was similar to that of the SEER population, suggesting that individuals in the parent generation represented patients seen in the general population. However, the distribution pattern in the child generation was different. This difference could be explained by anticipation, as the distribution pattern was not the same and the age at onset peaks had shifted to the earlier ranges. More importantly, this difference suggested that the individuals in the child generation were not representative of the general population, and individuals in the child generation may reflect only a small proportion of patients in the general population.
This realization has some implications when one considers why there was no observed bimodal pattern in the NHL distribution for the SEER population if anticipation truly exists. One explanation is that there are multiple factors involved in the aetiology of NHL. Mechanisms that work to produce anticipation may only be a small factor in a pool of assorted causal factors. However, this does not explain why the individuals in the parent generation of this study have an age of onset distribution pattern that matches that of the SEER population.
The results of this study support those of Zhu et al (1998 ) who, in an examination of the relationship between NHL and family history of a variety of cancers, found that the risk of NHL was highly associated with a history of NHL and other haematological neoplasms in first-degree relatives.
In conclusion, this study has demonstrated that anticipation may exist in familial NHL, and a genetic mechanism is suggested as the cause. Investigation of a prospective series is warranted.
This study was supported in part by Cancer Center Support Grant P30CA13330 awarded by the National Cancer Institute, NIH, DHHS. The contents of this paper are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute. Mr Wang was a recipient of an American Society of Hematology Student Summer Fellowship Award. The authors are grateful to Dr Marshall Horwitz, University of Washington, Seattle, for his thoughtful critique of the manuscript.