Are linkage analysis and the collection of family data dead? Prospects for family studies in the age of genome-wide association.

Full text

Page 1

Invited editorial

Is linkage and the collection of family data dead? Prospects for
family studies in the age of genome-wide association


Françoise Clerget-Darpoux1,2 and Robert C. Elston3

1INSERM U535, IFR 69, Villejuif, France
2Université Paris-Sud, Villejuif, France
3Case Western Reserve University, Cleveland, Ohio, U.S.A.



The HapMap project and the discovery of millions of single nucleotide
polymorphisms (SNPs) throughout the human genome, together with the
development of bead and chip technology, has made it feasible to type hundreds
of thousands of markers on an individual for a few hundred dollars. Because of
the extensive linkage disequilibrium (LD) that exists, genome-wide association
studies are currently being conducted on samples of unrelated persons in the
belief by some that this is now the design of choice to discover genetic variants
underlying relatively common complex diseases. Huge tissue repositories have
been set up for the purpose of conducting large-sample case-control studies.
Should we therefore stop collecting family data, forget that we inherit our genes
from our parents and ignore the fundamental laws of genetic transmission as
being unnecessary for gene discovery? Can we hope to anticipate, understand and
treat the many diseases to which humans are prone, simply by finding genomic
locations that differ between those who have and those who do not have disease?
Below we argue that, although they have a useful role to play, it should not be
naively assumed that the time has come for genome-wide association studies to
replace genome-wide linkage studies, or that the collection of family data is no
longer necessary.


inserm-00129091, version 1 - 4 Sep 2009
Author manuscript, published in "Human heredity 2007;64(2):91-6"
DOI : 10.1159/000101960

Page 2

The power to detect a gene

Risch and Merikangas (1996) wrote “We argue that {the linkage} method
has limited power to detect genes of modest effect”[1].
What is meant by a gene of modest effect? A classical measure of the
genetic effect of a diallelic locus on a binary trait is the odds ratio (OR), either for
one versus no copies of the susceptibility variant allele, or for two versus one
copy of it, these two measures being equal under a multiplicative model. It is
important to realize that, in an association study with equal numbers of cases and
controls, the power to detect such a variant depends only on the ORs; whereas in a
linkage study it depends on both the ORs and the variant allele frequencies in the
sample. From the data of Bell et al. [2] on type I diabetes in the North American
population , for example, the class 1 allele (M1) of the variable number of tandem
repeats (VNTR) flanking the Insulin gene has a frequency of 0.70 and we can
estimate under a multiplicative model an OR of 3.4 for this allele. There is a good
power to detect this effect in a case-control association study, but little power to
detect it by linkage in affected sib-pairs. On the other hand, had the allele
frequency been 0.10, the same OR would lead to an affected sib pair (ASP)
linkage study with very good power.
When a genome-wide scan is performed and only the most significant
results noted, allowance must be made for multiple testing, and the effect this has
on power is quite different for linkage and association studies because of the
enormous difference in the effective number of independent tests that each entails.
If, on the other hand, there is good motivation for examining a candidate gene or
candidate region (detected by a previous linkage analysis), an association study
becomes the second stage of a multiple sampling strategy, which may lead to a
much more economical study to obtain a given amount of power [3] Linkage
analysis can similarly detect what interactions may exist [4] and in this way also
limit the number of interaction models to examine in an association analysis.
inserm-00129091, version 1 - 4 Sep 2009

Page 3

The problem of replication
“Replication has a vital role in showing that associations that are identified
reflect interesting biological processes rather methodological quirks” [5].
Once an initial study has detected a susceptibility variant, heterogeneity
from population to population with respect to either environmental or genetic
factors can make it difficult to obtain an appropriate sample to perform a
replication study. First, the replication sample will need to be as large or even
larger [6;7], and second, the difference in both variant frequencies and LD
patterns among populations may make the signal undetectable in other
populations. The difficulty of replicating an association study is well illustrated in
the work of Reich et al [8] on Multiple Sclerosis, where an association on
chromosome 1 in African-Americans was not replicated in a different sample of
Afro-Caribbeans. Similarly, an association between Myosin IXB (MYO9B) gene
variants and celiac disease (CD) was recently detected in a Dutch case-control
association study [9], but not confirmed in the British [10], Swedish/Norwegian
[11], Italian [12] or Spanish [13] populations. In these studies, we cannot
determine whether the first reported association was driven by a causal variant in
the one population and not in the others, or whether it was just a false positive
result. Whereas Lohmueller et al [14] argue that a sizable fraction of reported
genetic associations have strong evidence of replication, it is well established that
association studies in general have a high tendency not to be replicated, the first
report being simply a chance occurrence [15]

What are the factors determining disease susceptibility?
Common versus rare variants
The poor success of genome-wide linkage searches for complex diseases
has imposed in a hegemonic way the “common disease-common variant”
(CDCV) hypothesis, which is the major motivation for the HAPMAP project. By
their very construction, common tagging SNPs do not identify rare variants, yet
we know that multiple rare variants in the same gene can play an important role in
disease susceptibility [16-18]. The total of all the susceptibility allele frequencies
inserm-00129091, version 1 - 4 Sep 2009

Page 4

may be quite high, but with extensive allelic heterogeneity. This is the case, for
example, in Crohn’s disease - for which risk alleles in the NOD2 gene have
frequencies each smaller than 5% in populations of European descent [19;20], but
with great variation of the total risk allele frequencies among those populations.
Models of the expected genetic variation at disease susceptibility loci based on
population genetics have suggested that when selection is weak, as seems likely
for many complex disease mutations, genetic drift becomes important and the
total frequency of susceptibility mutations is expected to vary widely among loci.
Equilibrium models and empirical studies suggest a role for both rare and
common variants [18]. Indeed, for classic monogenic diseases allelic
heterogeneity is the rule rather than the exception: is there any reason to think this
should be different for complex disease?

Gene-gene and gene-environment interaction
In the case of a quantitative trait, the presence of interaction (non-
additivity) can depend on the scale of measurement. Tukey’s test for non-
additivity [21] specifically detects removable non-additivity, i.e. non-additivity
that can be removed by application of a monotonic transformation to the data
[22]. Recently, Chatterjee et al [23] have used a quantitative score based on SNPs
as an indictor of a binary trait and hence have been able to use Tukey’s idea to
increase the power of testing main effects in the case of a binary trait when
removable interaction exists; but non-removable interaction is not detected. The
interactive effect of several functional genetic variants may play a larger role than
the marginal effect of each variant in determining a trait, especially in the case of
multifactorial diseases. Non-removable interactions of this sort cannot be easily
found in genome-wide studies without very large samples. On the other hand, if
we restrict our search for interactions by only considering simultaneously genes
that belong to a relevant pathway, or to the a priori most likely types of joint gene
action, correspondingly restricted association studies become feasible. Of course,
it must always be remembered that “a gene association study that begins by
inserm-00129091, version 1 - 4 Sep 2009

Page 5

looking at a priori functional candidates is looking on average for a different class
of variants than is a positional cloning study that started from a significant linkage
signal”[16]. Nevertheless, as we noted above, prior information obtained from
linkage studies can be used to motivate a more limited number of interactions to
consider when conducting an association analysis.
In autoimmune diseases, the HLA component acts through complex
interactions of numerous variants. A deeper understanding of these diseases can
only be obtained through more detailed knowledge of gene diversity, patient
genotype distributions, IBD sharing in affected siblings conditional on patient
genotype, and a biological understanding of the function of the genes in the HLA
region. Detecting an association is gives us very limited information, especially
when we wish to estimate risk, to which we now turn.
Estimation of risk
It is a far cry from detecting a signal indicating the presence of a causative
factor in a genomic region to its identification and the clinically important task of
estimating the disease risk due to it. In the first place, the same population
distribution of a marker can be coupled with different modes of inheritance of the
trait, and hence different risk estimates. For example, the observed distribution of
the VNTR flanking the Insulin gene (coded as a diallelic marker: M1 for less than
1000 repeats, M2 for a larger repeat number) in Type I Diabetes [2] may be the
result of either
- a direct role of the VNTR (situation 1),
in which case the allele frequencies are (0.67, 0.33) and the genotypic relative
risks of M1M2 and M2M2 versus the risk of M1M1 are GRR1= 0.33 and GRR2=
0.06, respectively; or
- a causal diallelic variant (S1, S2) in LD with the VNTR (M1,M2), with
P(S1/M1) = 0.2, P (S2/M1) = 0.8 and P (S1/M1) = 1 (situation 2).
In this case the causal variant has very different allele frequencies (0.13, 0.87) and
the genotypic relative risks of S1S2 and S2S2 versus the risk of S1S1 are GRR1=
0.10 and GRR2= 0.004, respectively.
inserm-00129091, version 1 - 4 Sep 2009

Page 6

We can distinguish between these two situations if we know the allele
sharing probabilities for the insulin gene among ASPs, which for these two cases,
letting zi denote the probability that they share i alleles IBD, are respectively:
(z2, z1, z0) = (0.29, 0.50, 0.21) in situation 1
and (z2, z1, z0) = (0.56, 0.39, 0.05) in situation 2.
In fact, on the basis of 95 such sib pairs [24], we know that the former is the true
situation, i.e. the VNTR has a direct role, as has been confirmed by functional
studies. Thus linkage information is crucial in order to discriminate between these
two situations that lead to the same association signal. It is important to identify
models that can explain both the association and the linkage information that is
found, and not just one of these [25], if we are to obtain good estimates of risk It
was also the IBD sharing information on sibs affected with Rheumatoid Arthritis
which showed that the usual classification [26] of the DRB1 risk alleles was
incorrect because it neglected numerous infrequent risk alleles [27].
Moreover, note that linkage information, far from being redundant once an
association is found, can be confirmatory. If we have a sample of affected sib
pairs, information derives from the IBD distribution of an affected sibling
conditional on that of the index sib [28]. When the index patients have a high risk
genotype, their affected sibs share two alleles IBD with a probability greater than
0.25 and, conversely, when the index patients have a low risk genotype, their
affected sibs share two alleles IBD with probability less than 0.25. One example
of this is the above cited Insulin VNTR, for which the overall IBD sharing is close
to the null hypothesis values (0.25, 0.50, 0.25). When the index case is
homozygous for the susceptibility allele, the proportions of affected sibs sharing
two, one and zero alleles IBD are 0.27, 0.60 and 0.13, respectively. When the
index cases have another genotype, the proportions are 0.07, 0.53, 0.40,
respectively [29]. A similar situation holds for the PTP22 gene in Rheumatoid
Arthritis. Four different genotypic risk groups can be defined and the proportions
of ASPs sharing 2 alleles IBD are 0.30, 0.24, 0.22, 0.09 according to the risk
group (from the highest to the lowest) that the index sib belongs to (Bourgey et al,
personal communication). Thus linkage data contain considerable information
inserm-00129091, version 1 - 4 Sep 2009

Page 7

capable of aiding in the discrimination of causal variation from merely nearby
variation that is in LD with a causal variant.
In a recent study, Lincoln et al [30] used a large-scale SNP association
strategy to study the involvement of the major histocompatiblity complex in
Multiple Sclerosis (MS). The strongest association was observed with haplotype
blocks in the HLA class II region and, conditional on this association, there was
no evidence for other HLA-region risk factors for MS. Despite the use of a large
number of simplex and multiplex families with dense SNP typing, the study
simply concluded with a result that has been known for many years. This is
another illustration of the limitations of strategies that rely solely on association.

The further importance of family information
As we have just seen in the above paragraph, collecting family information
is important to refine our genetic model and risk estimates. For both clinicians and
the relatives of a patient, we need estimates of familial risk. For example, parents
of a child affected with celiac disease (CD) often ask for the risk to a future child.
Thanks to good epidemiological familial surveys conducted in Italy, the risk that
will be given to Italian parents is about 10%. It is now possible to provide even
more accurate information by performing HLA-DQ typing in the family. The risk,
in fact, ranges from 1 to 30%: in some situations one will be able to reassure the
parents with a risk of less than 1%, whereas if a baby is at high risk (about 30%)
specific follow-up may be offered [31].
A major source of complexity in disease etiology is underlying
heterogeneity. In particular, it is possible for a disease that initially appears to be
multifactorial to have monogenic sub-entities. Detection of such heterogeneity
will be impossible through association studies, but is enabled by family studies.
This is well illustrated by breast cancer, for which the existence of monogenic
forms (crucial information in terms of public health) was revealed by segregation
analysis in a very large sample of families containing affected women [32]. The
location of BRCAI [33], as well as its identification [34] was only possible
through familial studies. By taking into account the family history of a woman a
inserm-00129091, version 1 - 4 Sep 2009

Page 8

search of a particular mutant allele may be indicated for her medical follow-up.
Similar examples could be given for other types of cancer, for neurological
diseases (e.g. Alzheimer disease, frontal dementia, Parkinson disease), for
cardiovascular diseases (cardiomyopathy), etc.

Note, too, that the well-established phenomena imprinting and anticipation can
only be detected if family data are collected. Similarly, there have been reports of
maternal genotype effects for neural tube defects [35;36] and autism [37]. These
and epigenetic phenomena leading to transgenerational effects [38;39] can only be
detected if family data are collected. Finally, one could note that it is much easier
to detect typing errors in family data – though this may be of less importance in
view of the high accuracy of the current platforms available for genome-wide
association studies provided appropriate steps are taken to detect errors that are
known to occur, such as because of abnormal copy number.


Conclusion

Finally, let us return to the quotation above from Risch and Merikangas
[1], which has often been quoted to justify genome-wide case-control association
studies (though we note in passing that the comparison they made was between
an ASP linkage study and a study jointly testing linkage and association, i.e.
performing genome-wide transmission disequilibrium tests - TDTs) and take it
two sentences further: “We argue that {the linkage} method has limited power to
detect genes of modest effect. A different approach that utilizes candidate genes
has far greater power, even if one needs to test every gene in the genome. Thus,
the future of the genetics of complex diseases is likely to require large-scale
testing by association analysis” [emphasis ours].
Note carefully it is erroneous to assume that a candidate gene study cannot
be conducted using linkage analysis, and we have seen above that association
information should not be considered as being in some way superior to linkage
inserm-00129091, version 1 - 4 Sep 2009

Page 9

information. We firmly believe that the current efforts being put into the
construction of huge databases for case-control studies should not be done to the
detriment of continued collection of family data. So many different biological
mechanisms are possible that it would be foolhardy to restrict all human genetic
research to a single strategy. This could well lead to the same frustrations that
followed the wholesale adoption of genome-wide linkage scans (Google counts of
these genome-wide studies so far: 183,000 for linkage, 181,000 for association).
One of the greatest current challenges facing human genetics is that of how best to
gather and synthesize the many lines of evidence possible in order to discover the
genetic determinants underlying complex diseases. Just as previously, before we
had so many genetic markers available, it was well recognized that it was
necessary to use many different lines of evidence to confirm a genetic component
in the etiology of a disease – population studies, adoption studies, twin and family
studies - so now, with the availability of all these genetic markers, must we
diversify the kinds of studies we perform when using this new tool for the purpose
of dissecting these genetic components for the betterment of mankind. A wise
investigator, just like a wise investor, should hedge all bets.

Acknowledgements
This work was supported in part by a U.S. Public Health Research grant
(GM28356) from the National Institute of General Medical Sciences and Cancer
Center Support Grant (P30CAD43703) from the National Cancer Institute.



References

1 Risch N, Merikangas K: The future of genetic studies of complex human
diseases. Science 1996;273:1516-1517.
2 Bell GI, Horita S, Karam JH: A polymorphic locus near the human insulin
gene is associated with insulin-dependent diabetes mellitus. Diabetes
1984;33:176-183.
inserm-00129091, version 1 - 4 Sep 2009

Page 10

3 Elston RC, Lin D, Zheng G: Multi-stage sampling for genetic studies. Annu
Rev Genomics Hum Genet 2007;in press.
4 Gauderman WJ, Morrison JL, Siegmund KD: Should we consider gene x
environment interaction in the hunt for quantitative trait loci? Genet
Epidemiol 2001;21 Suppl 1:S831-S836.
5 Hattersley AT, McCarthy MI: What makes a good genetic association
study? Lancet 2005;366:1315-1323.
6 Terwilliger JD, Haghighi F, Hiekkalinna TS, Goring HH: A bias-ed
assessment of the use of SNPs in human complex traits. Curr Opin Genet
Dev 2002;12:726-734.
7 Sohn D: Statistical significance and replicability: Why the former does not
presage the latter. Theory Psychology 1998;8:291-311.
8 Reich D, Patterson N, De Jager PL, McDonald GJ, Waliszewska A, Tandon
A, Lincoln RR, DeLoa C, Fruhan SA, Cabre P, Bera O, Semana G, Kelly
MA, Francis DA, Ardlie K, Khan O, Cree BA, Hauser SL, Oksenberg JR,
Hafler DA: A whole-genome admixture scan finds a candidate locus for
multiple sclerosis susceptibility. Nat Genet 2005;37:1113-1118.
9 Monsuur AJ, de Bakker PI, Alizadeh BZ, Zhernakova A, Bevova MR,
Strengman E, Franke L, van't SR, van Belzen MJ, Lavrijsen IC, Diosdado B,
Daly MJ, Mulder CJ, Mearin ML, Meijer JW, Meijer GA, van OE,
Wapenaar MC, Koeleman BP, Wijmenga C: Myosin IXB variant increases
the risk of celiac disease and points toward a primary intestinal barrier
defect. Nat Genet 2005;37:1341-1344.
10 Hunt KA, Monsuur AJ, McArdle WL, Kumar PJ, Travis SP, Walters JR,
Jewell DP, Strachan DP, Playford RJ, Wijmenga C, van Heel DA: Lack of
association of MYO9B genetic variants with coeliac disease in a British
cohort. Gut 2006;55:969-972.
11 Amundsen SS, Monsuur AJ, Wapenaar MC, Lie BA, Ek J, Gudjonsdottir
AH, Ascher H, Wijmenga C, Sollid LM: Association analysis of MYO9B
gene polymorphisms with celiac disease in a Swedish/Norwegian cohort.
Hum Immunol 2006;67:341-345.
12 Giordano M, Marano C, Mellai M, Limongelli MG, Bolognesi E, Clerget-
Darpoux F, Momigliano-Richiardi P, Greco L: A family-based study does
not confirm the association of MYO9B with celiac disease in the Italian
population. Genes Immun 2006;7:606-608.
13 Nunez C, Marquez A, Varade J, Martinez A, Polanco I, Maluenda C,
Fernandez-Arquero M, de la Concha EG, Urcelay E: No evidence of
inserm-00129091, version 1 - 4 Sep 2009

Page 11

association of the MYO9B polymorphisms with celiac disease in the
Spanish population. Tissue Antigens 2006;68:489-492.
14 Lohmueller KE, Pearce CL, Pike M, Lander ES, Hirschhorn JN: Meta-
analysis of genetic association studies supports a contribution of common
variants to susceptibility to common disease. Nat Genet 2003;33:177-182.
15 Ioannidis JP, Ntzani EE, Trikalinos TA, Contopoulos-Ioannidis DG:
Replication validity of genetic association studies. Nat Genet 2001;29:306-
309.
16 Pritchard JK, Cox NJ: The allelic architecture of human disease genes:
common disease-common variant...or not? Hum Mol Genet 2002;11:2417-
2423.
17 Pritchard JK: Are rare variants responsible for susceptibility to complex
diseases? Am J Hum Genet 2001;69:124-137.
18 di Rienzo A: Population genetics models of common diseases. Curr Opin
Genet Dev 2006;16:630-636.
19 Lesage S, Zouali H, Cezard JP, Colombel JF, Belaiche J, Almer S, Tysk C,
O'Morain C, Gassull M, Binder V, Finkel Y, Modigliani R, Gower-
Rousseau C, Macry J, Merlin F, Chamaillard M, Jannot AS, Thomas G,
Hugot JP: CARD15/NOD2 mutational analysis and genotype-phenotype
correlation in 612 patients with inflammatory bowel disease. Am J Hum
Genet 2002;70:845-857.
20 Vermeire S, Wild G, Kocher K, Cousineau J, Dufresne L, Bitton A,
Langelier D, Pare P, Lapointe G, Cohen A, Daly MJ, Rioux JD: CARD15
genetic variation in a Quebec population: prevalence, genotype-phenotype
relationship, and haplotype structure. Am J Hum Genet 2002;71:74-83.
21 Tukey JW: One degree of freedom for non-additivity. Biometrics
1949;5:232-242.
22 Elston RC: Addiditivity in the analysis of variance. Biometrics
1961;17:209-219.
23 Chatterjee N, Kalaylioglu Z, Moslehi R, Peters U, Wacholder S: Powerful
multilocus tests of genetic association in the presence of gene-gene and
gene-environment interactions. Am J Hum Genet 2006;79:1002-1016.
24 Spielman RS, Baur MP, Clerget-Darpoux F: Genetic analysis of IDDM:
summary of GAW5 IDDM results. Genet Epidemiol 1989;6:43-58.
inserm-00129091, version 1 - 4 Sep 2009

Page 12

25 Clerget-Darpoux F, Babron MC, Prum B, Lathrop GM, Deschamps I, Hors
J: A new method to test genetic models in HLA associated diseases: the
MASC method. Ann Hum Genet 1988;52:247-258.
26 Gao X, Gazit E, Livneh A, Stastny P: Rheumatoid arthritis in Israeli Jews:
shared sequences in the third hypervariable region of DRB1 alleles are
associated with susceptibility. J Rheumatol 1991;18:801-803.
27 du Montcel ST, Michou L, Petit-Teixeira E, Osorio J, Lemaire I, Lasbleiz S,
Pierlot C, Quillet P, Bardin T, Prum B, Cornelis F, Clerget-Darpoux F: New
classification of HLA-DRB1 alleles supports the shared epitope hypothesis
of rheumatoid arthritis susceptibility. Arthritis Rheum 2005;52:1063-1068.
28 Clerget-Darpoux F, Babron MC, Bickeboller H: Comparing the power of
linkage detection by the transmission disequilibrium test and the identity-by-
descent test. Genet Epidemiol 1995;12:583-588.
29 Dizier MH, Babron MC, Clerget-Darpoux F: Interactive effect of two
candidate genes in a disease: extension of the marker-association-
segregation chi(2) method. Am J Hum Genet 1994;55:1042-1049.
30 Lincoln MR, Montpetit A, Cader MZ, Saarela J, Dyment DA, Tiislar M,
Ferretti V, Tienari PJ, Sadovnick AD, Peltonen L, Ebers GC, Hudson TJ: A
predominant role for the HLA class II region in the association of the MHC
region with multiple sclerosis. Nat Genet 2005;37:1108-1112.
31 Bourgey M, Calcagno G, Tinto N, Gennarelli D, Margaritte-Jeannin P,
Greco L, Limongelli M-G, Esposito O, Marano C, Troncone R, Spampanato
A, Clerget-Darpoux F, Sacchetti L: Genetic Counseling in Celiac Disease.
Gut. In press.
32 Newman B, Austin MA, Lee M, King MC: Inheritance of human breast
cancer: evidence for autosomal dominant transmission in high-risk families.
Proc Natl Acad Sci U S A 1988;85:3044-3048.
33 Hall JM, Lee MK, Newman B, Morrow JE, Anderson LA, Huey B, King
MC: Linkage of early-onset familial breast cancer to chromosome 17q21.
Science 1990;250:1684-1689.
34 Miki Y, Swensen J, Shattuck-Eidens D, Futreal PA, Harshman K, Tavtigian
S, Liu Q, Cochran C, Bennett LM, Ding W, .: A strong candidate for the
breast and ovarian cancer susceptibility gene BRCA1. Science 1994;266:66-
71.
35 Doolin MT, Barbaux S, McDonnell M, Hoess K, Whitehead AS, Mitchell
LE: Maternal genetic effects, exerted by genes involved in homocysteine
remethylation, influence the risk of spina bifida. Am J Hum Genet
2002;71:1222-1226.
inserm-00129091, version 1 - 4 Sep 2009

Page 13

36 Brody LC, Conley M, Cox C, Kirke PN, McKeever MP, Mills JL, Molloy
AM, O'Leary VB, Parle-McDermott A, Scott JM, Swanson DA: A
polymorphism, R653Q, in the trifunctional enzyme
methylenetetrahydrofolate dehydrogenase/methenyltetrahydrofolate
cyclohydrolase/formyltetrahydrofolate synthetase is a maternal genetic risk
factor for neural tube defects: report of the Birth Defects Research Group.
Am J Hum Genet 2002;71:1207-1215.
37 Jones MB, Palmour RM, Zwaigenbaum L, Szatmari P: Modifier effects in
autism at the MAO-A and DBH loci. Am J Med Genet B Neuropsychiatr
Genet 2004;126:58-65.
38 Li S, Hursting SD, Davis BJ, McLachlan JA, Barrett JC: Environmental
exposure, DNA methylation, and gene regulation: lessons from
diethylstilbesterol-induced cancers. Ann N Y Acad Sci 2003;983:161-169.
39 Ruden DM, Xiao L, Garfinkel MD, Lu X: Hsp90 and environmental
impacts on epigenetic states: a model for the trans-generational effects of
diethylstibesterol on uterine development and cancer. Hum Mol Genet
2005;14 Spec No 1:R149-R155.


inserm-00129091, version 1 - 4 Sep 2009