www.elsevier.com/locate/jneuroim

Journal of Neuroimmunology 143 (2003) 74–78

Genetic analysis of multiple sclerosis in Europeans: French data

Mehdi Alizadeha,b,1, Emmanuelle Geninc,1,Marie-Claude Babronc,1, Brigitte Birebenta,b, Isabelle Cournu-Rebeixd, Jacqueline Yaouanqe,f,Stephane Dreanoa,b, Stephen Sawcerg, Alastair Compstong, Michel Claneth, Gilles Edane,f,

Bertrand Fontained, Franc�oise Clerget-Darpouxc, Gilbert Semana a,b,*

the French Multiple Sclerosis Genetics GroupaLaboratoire d’Immunologie, UPRES EA 1257 (IFR97), Faculte de Medecine, 2 Avenue du Pr Leon Bernard CS 34317, 35043 Rennes Cedex, France

bLaboratoire d’Immuno-Genetique, Etablissement Franc�ais du Sang Bretagne, Rue Pierre Jean Gineste, 35000 Rennes, Francec INSERM U535, Genetique epidemiologique et structure des populations humaines, Batiment Gregory Pincus,

80 Rue du General Leclerc, 94276 Le Kremlin Bicetre Cedex, Franced INSERM U546, Faculte de Medecine Pitie-Salpetriere, 105 Boulevard de l’Hopital, 75013 Paris, France

eService de Neurologie, CHU Pontchaillou, Rue Henri Leguillou, 35043 Rennes, FrancefLaboratoire Universitaire d’Immunologie, UPRES EA 1257 (IFR97), Faculte de Medecine, 2 Avenue du Pr Leon Bernard CS 34317,

35043 Rennes Cedex, FrancegNeurology Unit, Addenbrooke’s Hospital, University of Cambridge, Box 165, Hills Road, Cambridge CB22 QQ, UK

hDepartement de Neurologie, CHU Purpan, Pavillon Riser, Place du Docteur Baylac, 31059 Toulouse, France

Abstract

We report the results of a genome-wide screen for linkage disequilibrium (LD) in multiple sclerosis (MS) performed on 200 cases, 200

controls and 200 case–parent trios from France employing pooled DNA methodology. A total of 3510 microsatellite markers supplied

through the GAMES collaborative were analysed and ranked according to their evidence for association. The most promising 117 markers

were then followed up in a two-step validation process. In the first step, additional PCR of the DNA pools was performed in order to refine

the ranking order. In the second step, markers were genotyped in individual cases and parents from the trio families. Seven markers showing

nominally significant allele frequency differences between affected and unaffected emerged—D6S265, D12S1064, TNFa, D7S1824,

D14S1426, D14S605 and D21S2051. These potential associations will require confirmation in further studies.

D 2003 Elsevier B.V. All rights reserved.

Keywords: Genome screen; Linkage disequilibrium; French; Multiple sclerosis; DNA pooling

1. Introduction the genetic risk factors involved may be difficult to find by

Multiple sclerosis (MS) is a complex, multifactorial dis-

ease of the central nervous system. The prevalence of the

disease varies with latitude, from about 120/100,000 in

Scandinavia to less than 20/100,000 in the Mediterranean

area. In France, MS prevalence reaches about 40/100,000.

Apart from the HLA region (Yaouanq et al., 1997), no other

genetic risk factor has so far been identified. Whole genome

linkage studies have remained inconclusive and suggest that

0165-5728/$ - see front matter D 2003 Elsevier B.V. All rights reserved.

doi:10.1016/j.jneuroim.2003.08.015

* Corresponding author. Laboratoire d’Immunologie, UPRES EA 1257

(IFR97), Faculte de Medecine, 2 Avenue du Pr Leon Bernard CS 34317,

35043 Rennes Cedex, France. Tel.: +33-2-99-54-83-41; fax: +33-2-99-54-

83-20.

E-mail address: gilbert.semana@univ-rennes1.fr (G. Semana).1 These authors contributed equally.

linkage strategies and that association studies could be more

powerful.

Here, we report the results of a whole genome screen

for linkage disequilibrium (LD) performed using a dense

map of microsatellite markers in pooled DNA from

French cases, controls and family trios. Compared to

other studies performed as part of the GAMES project,

our analysis presents two original aspects. First, DNAs

were pooled on the basis of Taq Man results and second,

we employed our own statistical method for ranking the

markers in terms of their evidence for association. The

most promising markers identified in this screening

experiment were then followed up in a two-step valida-

tion process: Firstly by performing additional PCR of the

pooled samples and then by individually typing cases and

parents in the trios.

M. Alizadeh et al. / Journal of Neuroimmunology 143 (2003) 74–78 75

2. Patients and controls

MS French family trios and individual cases were pro-

spectively recruited as part of a survey of MS patients

identified from throughout France by the French Genetics

Group. Trios were ascertained through one patient per

family and two parents available for typing. All patients

included in the study were examined by one of us and

fulfilled diagnosis criteria for definite MS (Poser et al.,

1983). A total of 200 family trios, 200 unrelated MS cases

and 200 blood donor unrelated healthy controls were

included in this study. Unrelated cases and controls were

matched for sex. Informed consent was given by each

individual participation in the study according to the Hel-

sinki convention (1964) and French law relating to biomed-

ical research.

3. Method

3.1. DNA pooling

DNA pools were constructed using results from Taq Man

PCR of each sample. The effective amount of amplifiable

DNA input in each sample was assessed using a pair of

primers and a Taq ManR probe specific for the gene

(GAPD). Probe and primers sequences were as previously

described (Alizadeh et al., 2002). Four pools were constitut-

ed: (i) pool 1: 200 cases, (ii) pool 2: 200 controls, (iii) pool 3:

200 index cases, (iv) pool 4: parents of 200 index cases.

3.2. Markers

The microsatellite markers used to screen the whole

genome were provided by the GAMES collaborative and

are described elsewhere (Sawcer et al., 2002). Each marker

was amplified for each pool and resulting PCR products

were electrophoresed twice. Electropherograms with at least

one peak >10,000, or with all peaks < 200 were excluded.

3.3. Statistical analysis

The idea of the analysis is to rank the markers according

the difference they show in the patient and control samples.

It is already well established that stutter artefacts and

preferential amplification of the shorter alleles confound

the relationship between the allele image pattern (AIP)

obtained in the analysis of pooled DNA and the underlying

allele frequency distribution. However, Barcellos et al.

(1997) have argued that ‘‘artefacts produced by stutter or

preferential amplification act without bias on both patient

and control samples. If a marker association is strong

enough, significant differences will be observed between

pooled patient and control allele frequencies, even without

adjustment for stutter artefact’’. For each marker, allele

frequencies in the different pools were deduced from the

AIP by dividing peak heights for the different alleles by the

sum of the peak heights. The 5% of markers showing the

greatest difference between the two replicated electrophore-

sis were excluded. In all following analyses, the allele

frequencies used were the average over the two experiments

on the 3510 remaining markers.

For the case–parent trios, non-transmitted allele frequen-

cies were inferred from observed allele frequencies in the

parent and case pools. The frequency fNT(i,j) of a given

allele i at marker j in non-transmitted is simply obtained

from the frequencies fPAR(i,j) and fCAS(i,j) of this allele in

parents and cases:

fNTði; jÞ ¼ 2fPARði; jÞ � fCASði; jÞ:

To test for association between the trait and the different

markers, three comparisons of allele frequency distributions

were performed:

– test C1: cases (pool P1) versus controls (pool P2),

– test C2: index (pool P3) versus non-transmitted,

– test C3: combined test : cases + index versus controls +

non-transmitted.

Test C1 was performed on all 3510 markers. Tests C2 and

C3 were only performed on 3210 autosomal markers: 160

markers on the X chromosome and 140 autosomal markers

for which one of the non-transmitted allele frequencies was

negative, were excluded. The rationale for excluding these

markers from tests C2 and C3 was

– for markers on the X chromosome, in trios, only the

transmissions from the mothers are informative but

mothers and fathers were pooled together in pool 4.

– for autosomal markers, when inferred non-transmitted

allele frequencies are negative, the comparability of

parent and case allele image profiles could be questioned.

To compare these distributions, we used the following

strategies.

(1) We calculated a chi-square distance Tk for each marker:

Tk ¼ 2NXk

a¼1

ðf1a � f2aÞ2

f1a þ f2a

where k is the total number of alleles for the marker and

f1a and f2b the frequencies of allele a in the two groups to

be compared and N is the total sample size (i.e., 400).

(2) We standardised Tk, to obtain TkV:

TkV ¼ Tk � Tk

rTk

:

The mean Tk and standard deviation rTkof Tk were

computed for each test on the 95% smallest Tk values for

each number of alleles k.

Fig. 1. Probability plot of T3 against a theoretical chi-square distribution

with 2 degrees of freedom. If T3 was chi-square, the plots should be lined

with the bisector (shown as a plain line).

Table 1

List of the 25 best-ranked markers after the genome-wide screen

Marker Screening Validation 1: homogeneity Validation 2

Rank Within 6

PCRs

With initial

PCRs

AFBAC

p-value

D19S429 1 yes no

D9S172 2 yes no

D11S4178 3 no no

D9S147E 4 yes no

D1S2882 5 yes no

D21S2049 6 yes yes 0.2011

DXS8019 7 yes no

D6S265 8 yes yes 0.0041

D10S572 9 yes yes 0.4283

D2S1776 10 yes yes 0.1276

D15S101 11 yes no

D1S2777 12 n.t. n.t.

D5S2115 13 yes no

D8S1142 14 yes no

D10S213 15 n.t. n.t.

DXS8054 16 n.t. n.t.

D2S2361 17 yes no

D11S1362 18 n.t. n.t.

D3S3557 19 n.t. n.t.

D12S1064 20 yes yes 0.0297

D1S2756 21 yes no

D13S1243 22 n.t. n.t.

GGAA30B06 23 yes yes Typing problem

D12S92 24 n.t. n.t.

D13S217 25 n.t. n.t.

The respective ranks of the different markers are in the second column. In

the next columns are the results of the first validation steps. The markers for

which the six PCRs were homogeneous are indicated in the column ‘‘within

6 PCRs’’ and the markers showing consistent patterns of differences

between affecteds and unaffecteds in this step as compared to the initial

screen are highlighted in column ‘‘with initial PCRs’’ (n.t. indicates not

tested). The last column shows the p-value obtained for the AFBAC test

after individual genotyping.

M. Alizadeh et al. / Journal of Neuroimmunology 143 (2003) 74–7876

This transformation was used because pooling introdu-

ces additional sources of variance leading to a deviation

of Tk from the theoretical chi-square distribution with

k� 1 degrees of freedom. This is illustrated in Fig. 1 for

the test C1. A probability plot (Chambers et al., 1983) is

presented for the case of markers with three alleles. The

deviation of the T3 from the chi-square distribution with

2 degrees of freedom is evident.

(3) We ranked the different markers based on the TkV values

for test C1 and tests C2 and C3 when available. Note

that TkV values can be ranked irrespective of the actual

number of allele k.

(4) For markers that were considered in all three tests, we

compared the ranks obtained for tests C1 and C2 and

excluded those that show a difference of ranks greater

than 200.

(5) We define a list of the best 25 markers based on the C3

ranks or C1 ranks.

3.4. Follow-up study

For a subset of markers selected after the genome-wide

screen, a follow-up study was performed in two steps.

3.4.1. Validation step 1

Since one important source of variance with pooling is

the variance introduced by the PCR on the pools, we

decided to test again the markers in the same four pools.

For each marker and for each pool, six additional PCRs

were performed. Homogeneity of the allele image profiles

obtained from these six PCRs was tested by chi-square tests

and markers showing significantly different profiles among

the six PCRs at the 5% significance level were excluded.

The average profiles of the two initial PCRs (performed in

Cambridge) and of the six new PCRs (performed in Rennes)

were compared. Only markers showing consistent differ-

ences between affected and unaffected profiles were

retained for the next step.

3.4.2. Validation step 2

The markers selected after the first validation step were

individually genotyped in cases and parents of the trios.

Direct estimates of allele frequencies among cases and

among non-transmitted alleles could thus be obtained. These

allele frequencies were compared to the ones obtained from

the pool. To test for association, an AFBAC analysis was

performed (Thomson, 1995), where allele frequencies trans-

mitted and non-transmitted to affected offspring were com-

pared by a chi-square test. Markers with a nominal p-value

lower than 0.05 were considered as having passed validation

step 2.

4. Results

The 25 markers with the highest ranking after the

genome-wide screen are shown in Table 1. These include

M. Alizadeh et al. / Journal of Neuroimmunology 143 (2003) 74–78 77

one marker from the HLA region, marker D6S265, which

has rank 8. The raw data from these 25 markers was re-

inspected and the 17 with the best quality AIPs were

retested in validation step 1. One marker (D11S4178) was

excluded because of heterogeneity among the six repeat

PCR profiles. A further 10 markers were excluded as they

no longer showed a difference between affecteds and

unaffecteds in the repeat PCR profiles. Therefore, 6/17

markers were considered to have passed validation step 1.

These six markers were individually genotyped in the

trio families and association was tested by comparing the

frequencies of alleles transmitted and non-transmitted to the

affected offspring. The results of this AFBAC analysis are

presented in the last column of Table 1. Two markers,

D6S265 in the HLA region, and D12S1064 gave a result

nominally significant at the 5% level. For three of the other

markers, allele frequencies were not significantly different

among transmitted and non-transmitted alleles. Finally, for

marker GGAA30B06, no individual genotyping was possi-

ble, as there was a problem of competition in the amplifi-

cation of the two alleles. Interestingly, for this marker, the

image allele profiles obtained on the pools were very clear

and apparently unambiguous.

In Table 2, allele frequencies for the associated markers

(D6S265 and D12S1064) as inferred from the pools are

Table 2

Comparison of allele frequencies estimated from the pools and individual

genotyping

Size Estimated allele frequencies

1 2 3 4 5 6 7

D6S265 transmitted

Pooled

typing

400 0.04 0.20 0.07 0.30 0.12 0.15 0.11

Individual

typing

293 NF 0.19 NF 0.34 0.11 0.17 0.16

nontransmitted

Pooled

typing

400 0.02 0.13 0.07 0.27 0.12 0.24 0.14

Individual

typing

293 NF 0.11 0.00a 0.30 0.08 0.27 0.20

D12S1064 transmitted

Pooled

typing

400 0.07 0.10 0.11 0.23 0.20 0.21 0.07

Individual

typing

272 0.04 0.08 0.10 0.23 0.19 0.25 0.05

nontransmitted

Pooled

typing

400 0.13 0.13 0.11 0.19 0.21 0.15 0.08

Individual

typing

272 0.10 0.11 0.07 0.17 0.24 0.19 0.11

Transmitted allele frequencies are estimated in index cases from the trios.

Non-transmitted allele frequencies are deduced from observations in

parents and index cases. Only alleles with frequency greater than 0.02 in

at least one of the samples are shown. The frequencies presented for the

pools are those obtained from the AIPs obtained after validation step 1

(average over the six repeat PCRs). NF stands for not found, indicating that

the allele was not found in the sample.a Allele was present but at a frequency smaller than 1%.

compared to these frequencies as measured after individual

genotyping. As expected the difference in allele frequencies

observed in the pooled analysis closely reflects that seen

after individual genotyping. For marker D6S265, the in-

crease of allele 2 and the decrease of allele 1 can be

observed in both the pools and the individual genotyping

allele distributions. Similarly, for marker D12S1064, one

observes with both methods, an increase of alleles 4 and 6

and a decrease of alleles 1 and 2.

We then extended our validation analysis to those

markers (n = 100) in which at least one allele in both

patients/siblings and controls/cases pool comparisons

showed a frequency difference of greater than 2.5%. Both

validation steps were performed for these 100 markers.

Out of these 100, 33 were validated by step 1 and 5 of

these 33 in step 2 (TNFa, D7S1824, D14S1426, D14S605

and D21S2051). For these five markers, we found signif-

icant AFBAC p-values (0.01, 0.02, 0.01, 0.04, 0.02,

respectively).

5. Discussion

We typed 3510 microsatellite markers provided through

the GAMES collaborative in pooled DNAs from French

cases, controls and case–parent trios in order to search for

evidence of linkage disequilibrium with MS. Using a

statistical analysis, we developed ourselves we identified

117 markers requiring further analysis. After more extensive

repeat typing of these 117 markers in the pooled DNA

samples, only 39 markers continued to show evidence of

association. The fact that of two thirds (n = 78) of the

markers identified in our initial analysis of the screening

data failed to replicate their evidence for association when

the pools were retyped is not unexpected, and illustrates the

important variance due to PCR that is introduced by the

pooling methodology. The 39 markers showing a reproduc-

ible difference in case control allele frequencies in the

analysis of pooled DNA were then individually genotyped

in the trio family samples. Of 39 markers, 7 of these (18%)

showed evidence of association at the 5% level. However,

given the large number of tests performed, this result must

be interpreted with caution. Involvement of these markers

can only be confirmed or refuted by the analysis of further

samples.

In conclusion, this work has identified five non-HLA

microsatellite markers showing nominally significant as-

sociation with multiple sclerosis. Our validation work

highlights the importance of the additional sources of

variance introduced by the pooling methodology and

confirms that results from the analysis of pooled DNA

must be interpreted with caution, replicated and confirmed

by individual genotyping. In our experience, up to two

thirds of the markers identified in each GAMES screen

are likely to be the result of PCR artifact. However, these

markers should be easily and efficiently separated after

M. Alizadeh et al. / Journal of Neuroimmunology 143 (2003) 74–7878

repeat typing of the pooled DNA samples. Analysis of the

remaining markers by individual genotyping is expected

to reveal approximately one fifth as nominally associated.

Given the large number of markers tested, it is expected

that many of the markers identified in a GAMES screen

like ours will be the result of sampling variance. How-

ever, whether these nominal associations are the result of

genuine population frequency differences (real associa-

tion) or simply the result of sampling variation can only

be determined by the analysis of additional sample

cohorts.

Acknowledgements

This study was supported by grants from Fondation pour

la Recherche Medicale (FRM action 2000), by Association

pour la Recherche sur la Sclerose en Plaque (ARSEP). The

study represents one component of the GAMES project

designed and coordinated by Alastair Compston and

Stephen Sawcer (Cambridge, UK) with funding from the

Wellcome Trust (grant 057097) supplemented by the

Multiple Sclerosis Societies of the United Stated and Great

Britain, and MS International Federation.

References

Alizadeh, M., Bernard, M., Danic, B., Dauriac, C., Lapart, C., Lamy, T., Le

Prise, P.Y., Beauplet, A., Semana, G., Quelvennec, E., 2002. Quantita-

tive assessment of hematopoietic chimerism after bone marrow trans-

plantation by real time quantitative PCR. Blood 99, 4618–4625.

Barcellos, L.F., Klizt, W., Field, L.L., Tobias, R., Bowcock, A.M., Wilson,

R., Nelson, M.P., Nagatomi, J., Thomson, G., 1997. Association map-

ping of disease loci, by use of a pooled DNA genomic screen. Am. J.

Hum. Genet. 61, 734–747.

Chambers, J., Cleveland, W., Kleiner, B., Tukey, P., 1983. Graphical

Methods for Data Analysis. Wadsworth, Belmont, CA. 35 pp.

Poser, C.M., Paty, D.Q., Scheinberg, L., McDonald, W.I., Davis, F.A.,

Ebers, G.C., et al., 1983. New diagnostic criteria for multiple sclerosis:

guidelines for research protocol. Ann. Neurol. 52, 1336–1632.

Sawcer, S., Maranian, M., Setakis, E., Curwen, V., Akesson, E., Hensiek,

A., Coraddu, F., Roxburgh, R., Sawcer, D., Gray, J., Deans, J., Good-

fellow, P.N., Walker, N., Clayton, D., Compston, A., 2002. A whole

genome screen for linkage disequilibrium in multiple sclerosis confirms

disease associations with regions previously linked to susceptibility.

Brain 125, 1337–1347.

Thomson, G., 1995. Mapping disease genes: family-based association

studies. Am. J. Hum. Genet. 57, 487–498.

Yaouanq, J., Semana, G., Eichenbaum, S., Quelvennec, E., Clanet, M.,

Clerget-Darpoux, F., 1997. Evidence for linkage disequilibrium be-

tween HLA-DRB1 gene and multiple sclerosis. The French Research

Group on Genetic Susceptibility to MS. Science 276 (5313), 664–665.