Genetic Epidemiology 10:455-459 (1993)

Two New Approaches Toward Linkage

Heterogeneity of FAD: Two-locus
Models and Age of Onset as a
Discriminator
Chantal Merette, Thomas Lehner, and Jurg Ott

Columbia University, Department of Psychiatry, New York, New York

We present two new approaches to the problem of genetic heterogeneity
encountered in linkage analysis of familial Alzheimers disease. We used twolocus models to represent the possible existence of two disease genes while
allowing for intrafamilial heterogeneity, and modeled the occurrence of the early
onset form of the disease with epistasis. We developed a mixture model of
heterogeneity where the early and late onset family types can be either linked to
chromosome 19, 21, or unlinked, and where it is not necessary to arbitrarily
preclassify a family into an early or late onset family type. cb 1993 Wiley-Liss, Inc.
Key words: oligogenic models, age of onset, semi-epistasis, mixture models, Alzheimers disease

INTRODUCTION
Alzheimers disease is the most common form of dementia among the elderly.
Although its etiology is unknown, genetic components play a role in the pathogenesis of
the disease. Linkage studies have suggested the existence of a gene on chromosome 21
that segregates in some early onset families (mean age of onset S 65 years). The families
not showing linkage to chromosome 21 are often of the more common late onset type
(mean age of onset > 65 years) and some of them seem to show linkage to chromosome
19 [Wijsman, this volume]. These results indicate genetic heterogeneity for FAD with
possible linkage to chromosome 2 1 for early onset families, linkage to chromosome 19
for late onset families, and unlinked families of either age of onset.
We explored two new approaches as an alternative explanation for the inharmonious
nature of these findings : A two-locus model and the use of age of onset as a discriminator
in a linkage heterogeneity model.
Address reprint requests to Dr. Chantal MBrette, Department of Psychiatry, Unit 58, Columbia University,
722 West 168th Street, New York, NY 10032.
8 1993 Wiley-Liss, Inc.

456

Merette et al.

METHODS
Two-locus Models
Two-locus models were developed to analyze complex diseases under oligogenic
models and intrafamilial heterogeneity. These models are not necessarily more powerful
than single-locus models in the case where only one disease locus is linked to a given
marker and the other disease locus is unlinked [Vieland et al., 19921. However, the
opposite can be true when two markers unlinked to each other, and each of them linked
to a disease locus, are used [Schork et al., submitted]. We used the two-locus models to
test for linkage with Alzheimer's disease, thus dealing with the presence of the two
putative disease genes on chromosomes 19 and 21. This also makes it possible to model
the presence of both the early and late onset forms of the disease in a single family. The
analyses were performed using the programs TUNK and TMLINK [Lathrop and Ott,
19901.
We first defined a "semi-model of epistasis" which assumes that only the early onset
expression of FAD is due to the interaction of the two disease causing entities. When only
one of the two disease alleles is present, an individual expresses a late-onset form of
Alzheimer's disease. Therefore, the possible genotypes of early age of onset individuals
are restricted to those where each of the two disease alleles is present, while late onset
individuals have one of the four genotypes with only one disease allele present. This
model is shown in Table I. Two liability classes were created for individuals with known
age of onset: Class 1 for early onset cases (I
65) and class 2 for late onset cases (> 65).
Individuals with unknown age of onset were assigned liability class 3 in which all
genotypes showing the presence of at least one disease allele have the same probability
of occurrence. The lifetime penetrance value of 0.94 and the penetrance for noncarriers
of 0.002 were obtained using the mean age of onset M = 66 and variance s2 = 10.3
reported by Pericak-Vance [1991], assuming a normal distribution, and the reported
phenocopy rate of 10%. The gene frequency for the disease genes on chromosomes 2 1 and

TABLE I. Penetrance Values Defining the Two-locus Models

Liability
class

DD19

Dd19
dd19

DD19

Dd19
dd19

Model 2

Model 1

DD19
Dd19
dd19

0.94
0.94
0.002
0.002
0.002
0.94
0.94
0.94
0.94

0.94
0.94
0.002
0.002
0.002
0.94
0.94
0.94
0.94

0.002
0.002
0.002
0.94
0.94
0.002
0.94
0.94
0.002

Model 3

DD21

Dd21

dd21

DD21

Dd21

dd21

0.94
0.94
0.002
0.94
0.94
0.94

0.94
0.94
0.002
0.94
0.94
0.94

0.002
0.002

0.94
0.94
0.002

0.94
0.94
0.002

0.002
0.002
0.002

0.002
0.94
0.94
0.002

New Approaches to Linkage Analysis of Alzheimer's

457

19 were respectively set at 0.001 and 0.01, corresponding to a prevalence of 0.002 for
early onset and 0.02 for late onset cases.
There is no strong evidence that age 65 is the "true" cut off point that distinguishes
early from late onset cases. Therefore, we considered an alternative model in which late
onset individuals are assigned the liability class where there is an equal probability of each
bivariate genotype showing the presence of at least one disease allele. This constitutes our
model 2, also shown in Table I.
To test for a possible model of "complete epistasis," we constructed a single
penetrance class that allows for epistasis without differentiating the individuals by age
(Model 3 in Table I). Therefore every occurrence of the disease, in an early or late form,
is due to the interaction at the two disease loci.
The analysis was performed using the 32 Duke families. To exclude possible
recombinants due to false negatives, all our two-locus linkage analyses assume that
unaffected individuals are unknown. We used markers BCL3/Eco and D21Sl/MspI on
chromosomes 19 and 21, respectively. Table I1 shows the results.
The results of the two-locus analysis do not favor any of the hypothesized twolocus models. None of the lod scores calculated for different hypotheses even reached the
single-locus threshold value for significance of Z,,
2 3. The tested hypotheses within
each model consistently favored linkage of a putative gene to chromosome 19 (HI) over
linkage to chromosome 21 (H2), while the highest lod score values were always found for
the hypothesis of linkage to 19 and 21 (H3). In light of the recent discovery of linkage to
chromosome 14 in early onset FAD families [Schellenberg et al., 19921, it would be
interesting to incorporate the relevant markers on chromosomes 14 and 19 into these
models.

TABLE 11. Recombination Fractions and Maximum Lod Scores for Two-locus Models

Recombination fractions

BCL~ECO-D~,
Model 1
HO: No Linkage
HI: Linkage to 19 only
H2: Linkage to 21 only
H3: Linkage to 19 and 21
Model 2
HO: No Linkage
HI: Linkage to 19 only
H2: Linkage to 21 only
H3: Linkage to 19 and 21
Model 3
Ho: No Linkage
H1: Linkage to 19 only
H2: Linkage to 21 only
H3: Linkage to 19 and 21
aValues in parentheses are fixed.

D~ISIMS~I-D~~

zmax

(0.5)

0
1.20
0.30

0.00

1.44

(0.5)'

0.00

0.00

(0.5)
(0.5)

(0.5)

0.04
0.06

(0.5)

0.00
(0.5)

0.06
(0.5)

0.06

0
1.29
0.41
1.72
0
1.20
0.31
1.48

458

MCrette et al.

Age of Onset as a Discriminator

We have developed a model of linkage heterogeneity that translates the findings
reported in the literature into statistical parameters. The heterogeneity model, shown in
Table 111, assumes a mixture of six family types representing the early and late onset
families within which a family can be either linked to 21, linked to 19, or unlinked. The
proportion of each family type (al to a6,in Table HI), the mean age at onset in a family
(pl and p2) and the recombination fraction between the disease and chromosome 21 or
19 (el and 02)are parameters of unknown values and will be estimated. The Simplex
method of Nelder and Mead [1965] is the numerical method used to maximize this
multidimensional function.
Several specific hypotheses can be tested from the model represented in Table 111.
For example, we may want to assume that there are no late onset families linked to
chromosome 21, as linkage analyses of FAD families seem to indicate so far [PericakVance et al., 19911. Then we would simply give the corresponding parameter, a4,a value
of 0. Similarly, we can test the hypothesis that no early onset families are linked to
chromosome 19 by fixing a2 at 0.
Using markers BCL3IBanI-2 and BCL3/Eco-Mlu-2 on chromosome 19, and
markers D2 1S 1/MspI and D2 1S 1l/EcoRI on chromosome 2 1, we analyzed the 32 FAD
families from Duke pooled with the 21 families from Boston. Table IV shows the
resulting parameter values and the likelihoods of the hypotheses of linkage heterogeneity
and no linkage.
The parameter values that maximized the likelihood under the hypothesis of linkage
heterogeneity indicate that the data separate into a proportion of 16% of early onset
families (mean age of onset of 54.6) linked to chromosome 21,1% of unlinked early onset
families, 8% of late onset families (mean age of onset of 68.5) linked to chromosome 2 1,
39% of late onset families linked to chromosome 19, and 36% of unlinked late onset
families. These results are compatible with those summarized in Wijsman [this volume].
However, the difference in natural log-likelihood between the hypotheses of linkage
heterogeneity and no linkage to either chromosome 19 or 21 is only 3.94 (or 1.71 in log

TABLE 111. Parameters of Heterogeneity Model Based on a Mixture of Families Linked to

Either Chromosome 19 or 21, or Unlinked, Which Uses Age of Onset as a Discriminator
chr. 21
Early onset

Late onset

Proportion
Mean age
Recomb. fract.
Proportion
Mean age
Recomb. fract.

PI
8,

Chr. 19

Unlinked
a3
Pl
a6'h

a4

0.

'h

where:
ai is the proportion of families of type i, i = 1 ...6,
are the mean age of onset in early and late onset families, respectively,
pl and
8, and 8, are the recombination fractions between the first disease gene and a marker on chromosome
21, and the second disease gene and a marker on chromosome 19, respectively.

New Approaches to Linkage Analysis of Alzheimers

459

TABLE IV. Parameter Values of Heterogeneity Model Applied to 53 Familial Alzheimers

Disease Families from Duke and Boston
Chr. 21

chr. 19

Unlinked

Lnlikelihood + 3000 = 46.29

Hypothesis of linkage heterogeneity:

Early onset

Proportion
Mean age
Recomb. fract.

0.16
54.6
0.08

0.00
54.6
0.06

0.01
54.6
0.5

Late onset

Proportion
Mean age
Recomb. fract.

0.08
68.5
0.08

0.39
68.5
0.06

0.36
68.5
0.5

Lnlikelihood + 3000 = 42.35

Hypothesis of no linkage:
Early onset

Proportion
Mean age
Recomb. fract.

n.a.
n.a.
n.a.

n.a.
ma.
n.a.

0.17
54.6
0.5

Late onset

Proportion
Mean age
Recomb. fract.

n.a.
n.a.
ma.

n.a.
ma.
n.a.

0.83
68.4
0.5

n.a. = not applicable

to base lo), corresponding to an odds ratio of only 5 1:1 for linkage. We cannot draw any
conclusion as to the location of the putative FAD disease gene from the results of these
analyses.
ACKNOWLEDGMENTS

This research was supported by grant HG00008 from the National Center for
Human Genome Research.
REFERENCES
Lathrop GM, Ott J (1990): Analysis of complex diseases under oligogenic models and intrafamilial
heterogeneity by the LINKAGE programs. Am J Hum Genet 47:a188.
Nelder JA, Mead R (1965): A simplex method for function minimization. Comput J 7:308-313.
Pericak-Vance MA, Bebout JL, Gaskell PC, Jr., Yamaoka LH, Hung W-Y, Alberts MJ, Walker AP,
Bartlett RJ,Haynes CA, Welsh KA, Earl NL, Heyman A, Clark CM, Roses AD (1991): Linkage
studies in familial Alzheimer disease: evidence for chromosome 19 linkage. Am J Hum Genet
48:1034-1050.
Schellenberg GD, Bird TD, Wijsman EM, Orr HT, Anderson L, Nemens E, White JA, Bonnycastle L,
Weber JL, Alonso ME, Potter H, Heston LL, Martin GM (1992): Genetic linkage evidence for a
familial Alzheimers disease locus on chromosome 14. Science 258:668-671.
Schork NJ, Boehnke M, Tenvilliger JD, Ott J (1993): Two trait locus linkage analysis: a powerful
strategy for mapping complex genetic traits (submitted).
Vieland JV, Hodge SE, Greenberg DA (1992): Adequacy of single-locus approximations for linkage
analyses of oligogenic traits. Genet Epidemiol 9:45-59.