In October 2003, in a special issue of the Journal of Neuroimmunology, 17 research groups from 17 countries published the results of a multi-center genotyping initiative called GAMES. Using a methodology first proposed by Barcellos et al. (57), the groups genotyped pools of DNA samples, collected in their respective countries, from MS cases and controls (and, in six studies, from familial "trios" of patients and their parents) for the same 6000 microsatellite markers located throughout the genome. The theoretical rationale for the initiative was threefold: genome-wide association studies (or ''LD screens'') are a more powerful tool than linkage screens for locating genes with small or modest effects in complex disorders (58); sporadic MS cases are more numerous than familial cases and thus easier to ascertain and recruit; and the signals generated by LD screens are topographically more precise than those identified in linkage screens, since the chromosomal segments shared by members of the general population are shorter than those shared by members of the same immediate family (59). The goal of the pan-European design was to identify both ''ubiquitous'' genes, important for MS susceptibility in all populations, and "domestic" genes, important solely in a single population. Below we give a brief outline of the datasets examined and the chromosomal regions identified in each of the GAMES screens.
U.K. GAMES (cases and controls, familial trios): In an initial report (60), of the ten microsatellite markers exhibiting the greatest evidence of association, three were located in the HLA region ("providing a positive control for the method''), four in regions identified in the first U.K. linkage screen (two on chromosome 17q, and one each on chromosomes 1p and 19q); and three in novel regions (on chromosomes 1q, 2p, and 4q). In a second "refined" analysis (61), in which patients and parents in a subset of trios were individually genotyped for the 529 most promising markers, only the three HLA-region markers were confirmed as associated with MS.
Australia GAMES (62) (HLA-DRB1* 1501-positive cases andunselected controls): Evidence of association was uncovered for a total of seven markers-four located in regions identified in earlier linkage screens (on chromosomes 12q15, 16p13, 18p11, and 19p13), and three in novel regions (on chromosomes 11q12, 11q23, and 14q21)— suggesting the possibility of interactions between these loci and the HLA locus. An interaction of this kind, between HLA-DR15 and an allele in the promoter of the gene encoding CTLA-4 on chromosome 2q33, was recently described (63).
Belgium GAMES (64) (cases and controls): The 20 most promising markers included three located in the HLA class II region and one in the HLA class I region. In addition, the regions identified by the remaining markers contained a number of attractive candidate genes, including the gene encoding the integrin ligand EDIL3 (on chromosome 5q14) and the gene encoding the B-cell-specific transcription factor POU2AF1 (on chromosome 11q23).
Finland GAMES (65) (cases and controls): A total of 108 markers displayed evidence of association. Five chromosomal regions (1q43, 2p16, 4p15, 4q34, and the HLA region on 6p21) contained two or more markers within a 1-Mb interval. In addition, evidence of association was found for a marker located on chromosome 19p13.3, in proximity to the gene encoding ICAM-1. Earlier studies have reported an association between a nonsynonymous SNP in exon 6 of ICAM1 and the risk of MS in case-control datasets from Poland (66) and Finland and Spain (67), but not in datasets from Holland (68) or Sweden (55).
France GAMES (69) (cases and controls, familial trios): After a two-step validation process, involving re-typing of both pooled and individual samples for the 117 most promising markers, two HLA-region markers and five markers from non-HLA regions (two on chromosome 14q32, and one each on chromosomes 7q34, 12q21, and 21q21) displayed evidence of association.
German GAMES [HLA-DRB1* 1501-positive cases and unselected controls (first screen); cases and controls, familial trios (second screen)]: In the first screen (70), association with seven markers (two located on chromosome 1p36, and one each on chromosomes 2q34, 3p25, 4q28, 5q14, and 10q21) was confirmed by individual typing. In the second screen (71), evidence of association was found for two HLA-region markers and nine markers from non-HLA regions. Five of the non-HLA markers were located in regions identified in earlier linkage screens (on chromosomes 2q24, 6p25, 11q23, 12q13, and 19q13), while the remaining four were located in novel regions (on chromosomes 2q33, 15q24, 17p13, and Xq13). Six genes mapping to the region of the most promising marker (on 11q23) encode molecules involved either in the activity and protection of neurons (SCN2B and UBE4A) or in immune homeostasis (CD3D, CD3E, CD3G, and IL10RA).
Hungary GAMES (72) (cases and controls): Of the 33 markers exhibiting evidence of association, six were located in non-HLA regions identified in earlier linkage screens (two on chromosome Xp, and one each on chromosomes 3p14, 5p15, 7p13, and 7q21), and the rest in novel non-HLA regions.
Iceland GAMES (73) (cases and controls): Of the six 2-Mb regions harboring at least two associated markers, three (3q25, 19q13, and the HLA region on 6p21) contained more than one of the 20 most strongly associated markers.
Ireland GAMES (74) (cases and controls): Of the 22 markers displaying evidence of association, three were located in the HLA region. Association with one of the remaining markers, D11S1998, was confirmed by individual typing. The marker maps to a region on chromosome 11q23—the most promising region in the German GAMES screen—which contains the candidate genes IL10RA and CD3E.
Italy GAMES (75) (cases and controls, familial trios): None of the 14 most promising markers mapped to the HLA region. After refined laboratory and statistical analysis, only one of these markers retained evidence of association. This marker, D2S367, maps to a region on chromosome 2p22 that contains several candidate genes encoding molecules involved in apoptotic pathways, including CARD12. It has been reproducibly demonstrated that allelic variants of a gene encoding another member of the CARD family, CARD15, are associated with susceptibility to another putatively autoimmune disorder, Crohn's disease (76).
Scandinavia GAMES (77) (cases and controls): In two independent screens of pooled samples from Danish, Norwegian, and Swedish cases and controls, nine markers from eight chromosomal regions (1p33, 3q13, 6q14, 7p22, 9p21, 9q21, Xq22, and the HLA region on 6p21) were associated with MS in both screens. Chromosome 1p33 was positive for linkage in the British and Canadian linkage screens.
Poland GAMES (78) (cases and controls, familial trios): The screen identified five associated markers from five different chromosomal regions (2p16, 3p13, 7p22, 15q26, and the HLA region on 6p21). The region on 7p22 contains a candidate gene encoding the apoptosis-related protein CARD11.
Portugal GAMES (cases and controls): In the first of two separate screens (79), evidence of association was found for ten markers from seven chromosomal regions. Three of these regions (5q13, 7q21, and the HLA region on 6p21) were identified in earlier linkage screens and two in earlier GAMES screens (4q35 in the British screen, and 11p15 in the first German screen). The remaining two regions (10p13 and 11q14-24) were novel. In the second screen (80), 46 markers displayed evidence of association. Three chromosomal regions (6q14, 7q34, and the HLA region on 6p21) contained at least two associated markers within a 1.5-Mb interval.
Sardinia GAMES (81) (cases and controls, familial trios): Five markers (from regions on chromosomes 2q36, 6p25, 6p21, 7p12, and 16p12) displayed evidence of association in both cases and controls and familial trios. The marker on 6p21 (D6S271) is located at more than 10 cM from the HLA region.
Spain GAMES (82) (cases and controls): After repeated typing of the 1269 most promising markers, clusters of associated markers were identified on virtually every chromosome. Of the 25 markers with the lowest probability values, seven mapped to the HLA region, while five (on chromosomes 5p15, 5q14, 12q23, 16p13, and 17q23) were identified in earlier linkage screens.
Turkey GAMES (83) (cases and controls): Evidence of association was demonstrated for 12 markers, one of which was located in a region (on chromosome 5p15) identified in the Turkish linkage screen. This region is also homologous with a murine susceptibility locus in experimental allergic encephalomyelitis, an animal model of MS.
In summary, over 80% of the GAMES screens uncovered associations with one or more markers located in the HLA region. In an editorial in the same special issue of the Journal of Neuroimmunology (84), Barcellos and Thomson conclude that the GAMES results "further underscore the universality" of the HLA association in MS. They also point out that a region on chromosome 19q13 was identified by no fewer than seven of the GAMES groups. This region harbors the APOE gene (see below) and was identified as the most promising non-HLA locus in an early meta-analysis of the first four MS linkage screens (85).
Yeo et al. (61), reporting the results of the refined analysis of the British GAMES screen, offer a critical re-appraisal of the basic design of the GAMES project. The power of the GAMES screens to identify MS susceptibility genes, the authors write, was limited by three important factors:
First, the sample sizes used in GAMES are far too small. The pools in each GAMES screen contained DNA from approximately 200 individuals (MS patients, healthy controls, or unaffected parents). Samples of this size provide only modest power to detect strong genetic signals, such as those emanating from markers in the HLA region, and virtually no power at all to detect any weak signals emanating from non-HLA regions.
Second, pooling methodology further reduces the effective size of the samples. Error in estimating allele-frequency differences between affected and unaffected subjects can be divided into sampling error (random noise in a finite sample) and measurement error (noise related to the precision of the method). Sampling error decreases with increasing sample size, but measurement error does not. As Carlson et al. (86) have recently pointed out, for an allele conferring a 1.5-fold increase in the disease risk with a frequency of 10%, the expected difference in allele frequency between cases and controls is only 4.3%. As the measurement error introduced by pooling is about ±2%, differences of this size could easily be missed, particularly in a genome-wide LD screen, in which corrections for multiple testing must also be performed.
Third, the number of markers used in the GAMES initiative is far too low. The issue of marker density is of course related to the extent of LD throughout the genome, as markers are chosen on account of their presumed proximity to functional polymorphisms. At the time the project was designed, it was believed that LD in the European population was far greater than we now know it to be, and that the entire genome could be screened for association through the use of 6000 microsatellite markers. It turns out, however, based on current estimates of LD, that each GAMES screen tested no more than about 1% of the genome. According to Yeo et al. this last factor—the overestimation of LD and the resulting miscalculation of the required number of markers—represents the greatest shortcoming of the ambitious GAMES initiative.
It is certainly unfortunate that 99% of the genome was left unexplored by the GAMES project. But it is equally problematic that nearly all of the dozens of non-HLA markers ''displaying evidence of association'' in the 1% of the genome that was explored—markers that now ''require confirmation in further studies,'' in the words of one of the GAMES groups (69)—are, in light of the great number of statistical tests performed in each screen, presumably false positives (87).
Was this article helpful?