Subjects and families
In the first-stage mapping, we studied 96 families with at least two individuals with AD, which included 490 family members. The demographic and clinical characteristics of the participating family members are presented in . In addition to probands, approximately four relatives were examined per family, which included affected as well as unaffected individuals. The mean age at onset for affected individuals was 73.6 years. The majority of the families (85.7%) were from the Dominican Republic, 10.2% were from Puerto Rico, and 4.1% came from elsewhere in the Caribbean. Previously, we have shown that the study population is relatively homogeneous.15
Description of Caribbean Hispanic families
Overall, 237 individuals met the criteria for probable AD, 157 were unaffected. The study included families with affected sibpairs as well as families with multiple individuals with AD who were distantly related (eg, avunculars, cousins). There were 12 families with five or more affected family members and 12 families with four affected members. Approximately one-half of the families included two affected members. Eight families had predominantly earlyonset (defined as the majority of affected individuals having onset before age 65 years) disease and 88 families had late-onset AD.
In the second-stage mapping, we added eight families for a total of 104 families and 521 family members. We found the clinical characteristics for the additional families to be similar to those of the families used in the first-stage mapping ().
The frequency of APOE-ε4 among affected individuals was significantly higher than expected under the null hypothesis of no association (Z = 4.26, P= 1.01 × 10-5, allele frequencies in ). We were unable to compare allele frequencies for highly polymorphic microsatellite markers between individuals from the Puerto Rico and Dominican Republic because there were too few individuals from the Puerto Rico to make such comparisons meaningful.
Two-point analysis The positive results of two-point analyses for 340 autosomal microsatellite markers with an average inter-marker distance of 10.2 cM under the genetic homogeneity and affected sibpair models are shown in (see supplementary graphs). We observed 17 markers exceeding an LOD score of 1.0 using either model. The three highest two-point LOD scores were 3.26 for D18S541 in the affected sibpair model, 2.44 for D18S844 in the homogeneity model, and 2.0 for D14S587 in the affected sibpair model. When restricted to late-onset families, the LOD score increased to 3.62 for D18S541 and to 2.98 for D18S844 under the homogeneity model, but decreased to 1.69 for D14S587 under the affected sibpair model.
Summary table for the two-point linkage analysisa results
Multipoint analysis Two of the highest multipoint NPL scores were observed on 18q and 10q (see supplementary material). On 18q, the strongest support was observed for D18S541 (NPL=2.70, P=0.0022; 112.3 cM), and the NPL scores for the adjacent markers were 1.42 (P=0.0646) for D18S1270 and 1.82 (P=0.026; 96.5 cM) for D18S844 (119.7 cM). The peak spanned over 20 cM. For 10q, the major peak was at D10S1230 with an NPL score of 3.15 (P=0.000487), and a minor peak was observed at D10S677 with an NPL score of 1.54 (P=0.048502; 117.4 cM). Although the major peak spans approximately 35 cM, the two peaks combined span as much as 60 cM.
18q We re-examined the entire region between D18S1270 (96.5 cM) and D18S70 (126.0 cM) in detail with nine additional markers, because the initial peak was broad and the distal end of the peak remained elevated. Two-point analysis supported linkage for D18S541 (LOD=3.37) and D18S844 (LOD=2.06). With additional markers, the peak came downto the right of D18S844. The two markers to the distal end of D18S844, namely D18S1122 (122.8 cM) and D18S70 (126.0 cM), had LOD scores of 0.76 and 0.18, respectively, suggesting a boundary of the peak. In the multipoint analysis, the support for linkage was strongest at D18S1106 (NPL=3.65, P=0.000177; 111.7 cM), which is located less than 1cM away from D18S541; further, the peak was much narrower than in the first-stage scan (). As the peak was quite narrow, we conducted an additional analysis to assess the likelihood of genotyping errors leading to a sharp false-positive peak, by repeating the multipoint linkage analysis of the peak region excluding one marker at a time. If the peak were due to genotyping errors, it is expected that the linkage at the peak region would diminish; however, the magnitude of the NPL score change for each marker was minimal (data not shown).
Figure 1 The multipoint nonparametric analysis for 18q from the second-stage fine mapping using all families. The heavy solid line represents the linkage analysis using all families; the light solid line the Apolipoprotein E ε4-positive conditional linkage (more ...)
Using the same set of 104 families, we then conducted a family-based association analysis using Sib-TDT. Multiple markers on 18q met minimal criteria for statistically significant (P<0.05) allelic association as shown in , namely D18S465, D18S1092, D18S1106, D18S541, D18S1358, D18S870, and D18S58. For D18S541 and D18S870, two alleles were associated with AD (P<0.05 uncorrected for multiple testing). Joint linkage and association analysis using PSEUDOMARKER indicated that, for D18S541, linkage in the presence of linkage disequilibrium did not differ significantly from the linkage model alone (P=4.4 × 10-5 vs 3.1 × 10-5).
Fine-mapping family-based association analysis of 18q and 10q
Lastly, we conducted a conditional linkage analysis based on the APOE genotype (). This analysis revealed that the peak on the 18q largely derived from individuals who were APOE4-positive, and the peak essentially disappeared when we considered APOE4-negatives.
10q With 12 additional markers near D10S1230, the location narrowed somewhat. The two markers that gave the strongest signals from the two-point analysis were D10S190 (LOD=1.81) and D10S1230 (LOD=1.34). The multipoint analysis revealed the strongest peak at 138 cM near D10S190 (NPL=2.02; P=0.0157) (). With saturation of 10q, our peak shifted slightly proximal compared with our initial mapping, and yet the candidate region remains broad, spanning over 50 cM.
Figure 2 The multipoint nonparametric analysis for 10q from the second-stage fine mapping using all families. The locations and markers of reported suggestive linkage findings by other groups on 10q are shown below. To estimate confidence intervals, we subtracted (more ...)
Our subsequent family-based association analysis further supported the findings from the linkage analysis. Although multiple markers in this region supported allelic association as shown in , the strongest allelic association was observed with D10S1230 (Z=3.23, P=0.000623). This is the most likely cause of the linkage from our sample. In addition, D10S610 and rs4925 at around 130 cM also showed modest allelic association.
We examined 10q using the two-point joint linkage and association approach. In this set of analyses, D10S190 (at 138 cM; LOD=1.95 under the autosomal recessive model) and D10S1230 (at 143 cM; LOD=1.28 under the autosomal dominant model) showed only modest linkage as well as linkage in the presence of linkage disequilibrium. The autosomal recessive model showed stronger support than the autosomal dominant model. In the smaller peak at around 115 cM, D10S583 showed a significant result for linkage disequilibrium given linkage under the autosomal recessive model (P=0.003869), as well as for a joint linkage and linkage disequilibrium test under the autosomal recessive model (P=0.004188). No other markers showed support for this region.
As our peak for the AD phenotype centered on D10S190 and Li et al35
reported allelic association between glutathione S-transferase omega-1 (GSTO1) and GSTO2 with age at onset of AD and Parkinson’s disease (PD), we examined allelic association for seven markers (D10S583, D10S1671, D10S670, rs4925, D10S1429, D10S610, D10S187) near GSTO1 using the age at onset as the phenotype. We did not find a significant association with rs4925 (P
=0.4); further, the age at onset was slightly lower for the individuals with more common allele compared with those with rare allele (73.4 vs 75.6 for common vs rare allele, respectively). However, we did observe a weak association with D10S610 (F=5.51, P 0.02), which is approximately 3 cM away from the rs4925 SNP. The strongest association with AD was with marker D10S1230 (P
=0.0006), approximately 12 cM distal.
For both 18q22 and 10q, we reanalyzed the two-point and multipoint linkage analyses previously described using the allele frequencies estimated from one randomly selected individual from each family to reduce the influence that large extended pedigrees may have on allele frequency estimation. However, the allele frequencies were comparable and the LOD and NPL scores remained unchanged.
In this unique group of Caribbean Hispanics with familial AD, we observed a modest peak at chromosome 10q near 138 cM, which spans approximately 15 cM and a weaker peak at 115 cM. The major peak on 10q was more distal than those reported by some groups.6-8
A new peak was identified at 18q22, and with additional fine mapping the region surrounding the peak narrowed considerably, and there existed evidence of an epistatic effect with APOE ε4.
Overall, 17 markers had an LOD score of greater than or equal to one in the first stage of the genome scan reported here. Other genome-wide searches3,9,37-39
have reported many of these loci, for example, 1q, 10q, 12p, 14q, while some are new. For example, Kehoe et al3
and Blacker et al38
reported a modest peak on 1q23 (178 cM). We do observe weak linkage in the region, but observed a stronger finding in a more proximal location at ~114 cM. For 14q, we observed a peak on D14S587 (55 cM) as Blacker et al38
did in ‘early/mixed’ families; however, our samples, based predominantly on late-onset AD individuals, showed linkage to the same region. For 10q, 12p, and 18q, we chose to fine map these candidate regions because we observed strong evidence for linkage, and others reported strong linkage.
The observed variability in linkage findings is not surprising given the genetic complexity of AD. Specifically, there are likely to be multiple susceptibility genes, incomplete penetrance, and putative environmental risk factors. In addition, these studies use different genetic models, multiple testing of markers, and different study samples from different ethnic backgrounds. The primary goal of our study was to identify candidate regions that may contain AD susceptibility genes. According to the criteria suggested by Lander and Kruglyak,40
loci on chromosomes 10q26, 12p12, and 18q21 linked to AD would fulfill the criteria for suggestive of linkage. Some of the loci we observed from our genome-wide search will likely be false-positives. Thus, the findings here, as with any other genome scan, will need independent confirmation.
At least four loci have been linked to AD on chromosomes 9p, 10q, 12p, and 20p. However, though several putative loci have been linked to familial AD, only a few reports of allelic variation have been reported, but these reports need to be confirmed in independent samples. Pericak-Vance et al9
identified a locus on chromosome 12p conferring susceptibility to AD. Subsequent work by her group, ourselves, and others has suggested that there may be two separate loci on chromosome 12 depending on the presence of APOE-ε4 and clinical heterogeneity related to the presence of Lewy bodies.5,10,16,36,38,39,41
In the initial scan, Kehoe et al3
observed a peak at 12p only for the APOE-ε4 negative families, not for the APOE-ε4-positive families. However, when Blacker et al38
followed up the earlier genome scan using a larger sample, the support for linkage on 12p weakened. Previously, we reported our fine-mapping results that showed two peaks near D12S1623 (16 cM) and D12S1042 (49 cM).16
When we restricted the analysis to late-onset AD families, both markers showed strong linkage. When we conducted the analysis stratified by the APOE status, the linkage for D12S1623 was observed only in the APOE ε4-negative families. Two candidate genes in the region, alpha-2-macro-globulin (A2M [MIM 103950]) and the low-density lipoprotein receptor-related protein (LRP1 [MIM 107770]), have been the target of investigations, but without consistent results.16,36,42-45
Saunders et al
using the larger NIMH samples, identified several haplotypes that showed an association with AD, suggesting that there may be a gene—either A2M or one nearby—that may confer a modest effect toward AD.
Linkage findings on 10q also provide an illustration of the difficulties in replicating linkage and in estimating the location of gene underlying complex disorders.47,48
Six studies based on three different but correlated phenotypes (familial AD;3,6,8,38
variation in plasma amyloid β levels;7
and age at onset of AD34
) showed evidence of linkage to 10q. The peaks in these studies cover a wide chromosomal region from 81 to 135 cM; however, it is not possible to determine whether these represent one or multiple loci.48
found the strongest support for linkage at ~81 cM (D10S1227-D10S1211), while Bertram et al8
observed a peak between 115 and 127 cM, and later Blacker et al38
reported a peak at 135 cM (D10S1237) in a two-point analysis (θ=0.25). Using age at onset as the phenotype, Li et al34
reported a peak at 135 cM using the same marker (D10S1237) as Blacker et al
, and subsequently reported an association with GSTO1. As we observed a peak nearby at 138 cM with the AD phenotype, we examined allelic association using both AD and age at onset phenotypes. For the age at onset phenotype, we did not find an association with rs4925 (P
=0.4), but we did observe a weak association with D10S610 (F=5.51, P
=0.02), which is approximately 3 cM away from the rs4925 SNP. For the AD phenotype, however, we observed weak allelic associations for rs4925 (Z=1.692; P
=0.045323) as well as D10S610 (Z=1.718; P
=0.042898). Although we did observe a weak association in the region, our finding is somewhat different from that of Li et al
. The mean age at onset observed in our samples was comparable to that reported by Li et al
. Yet, our association with rs4925 is weak for the AD phenotype and not significant for the age at onset phenotype. Further, while Li et al
found the less common allele to be associated with later age at onset, we did not observe any difference in the mean age at onset between the two alleles. Further studies are needed to understand the allelic association since our finding is based on a relatively small sample of Caribbean Hispanics, while that of Li et al
’s findings is based on a larger set of US Caucasians. Thus, at this point, it is not possible to conclude whether GSTO1 or some other gene is responsible for our linkage and association finding in this region.
In contrast, Bertram et al8
reported significant peaks at 115 cM (D10S583) and 127 cM (D10S1671), and observed allelic association at D10S583, located near the gene insulin degrading enzyme (IDE [MIM 146680]). Ait-Ghezala et al49
supported this finding in a follow-up case-control study. However, Abraham et al50
could not confirm the association with D10S583 or with SNPs flanking IDE. In a subsequent genome scan, Blacker et al38
reported two modest and broad peaks on 10q: one more proximal (80-100 cM) than the previously reported peaks (115 and 127 cM) and another at 135 cM. In our data, we did observe a modest peak at 138 cM. Given that multiple data sets using multiple correlated phenotypes show modest support for linkage on 10q covering a wide region, a susceptibility gene, if present, may only confer a weak effect. Alternatively, there may be more than one gene involved and they may interact. In our conditional analysis of APOE 4, we did find a slightly stronger evidence of linkage at the D10S190, but it was not statistically significant. Lastly, Ertekin-Taner et al7
reported linkage to 10q (~80 cM) for plasma amyloid β protein levels as an intermediate risk factor phenotype for AD, then they subsequently reported the QTL locus to be the α-T catenin gene.51
In our data set, we did not observe any signal in this region when we used the AD phenotype.
An additional locus on chromosome 9p22 has been linked to AD from different samples: two studies in the US Caucasians;4,38
and in inbred Israeli Arabs.39
On chromosome 20p, AD8 [MIM607116]), a locus that includes the gene encoding cystatin-C (CST3 [MIM 604312]), was linked to very late-onset AD,37
and appeared to interact with the amyloid precursor protein. In contrast, we did not observe any support for linkage in these two regions to the AD phenotype.
It is important to note that, with the exception of the studies by three groups (Ertekin-Taner et al
Farrer et al
and the current study), all other studies include some subsets of the NIMH samples; thus, overlapping linkage peaks from different studies need to be interpreted with caution.
Our most robust linkage finding on 18q was previously reported by Pericak-Vance et al4
to have an LOD score of 1.1. With additional markers, the linkage peak sharpened, and the peak centers around D18S541, since both the joint linkage and association analysis and family-based association analysis strongly support D18S541. As with any genome scan, it is difficult to exclude the possibility of a false-positive finding. However, the addition of more markers narrowed the chromosomal region considerably, and the NPL score attained greater statistical significance. Further, the distal peak tapered off, arguing against the possibility of a spurious finding that occurs at the distal ends of the chromosome. Subsequent re-analyses of the region—first by sequentially excluding one marker at a time and second by using the allele frequencies based on a single randomly selected individual per family, rather than using all family members—did not change the LOD or NPL scores, suggesting that the observed linkage is less likely to be due to either genotyping errors or due to inaccurate allele frequency estimates.
This region on 18q contains approximately 4 000 000 base pairs including several candidate genes for which mRNA expression in brain has been observed. These include NETO1, a neuropilin (NRP) and tolloid (TLL)-like protein, and the predicted RefSeq genes for HSPC154 and MGC39671. There are also approximately 50 genes predicted using the Genscan Gene prediction track of the UCSC Genome Browser in the region.
We did not observe significant linkage to APOE in either the two-point or multipoint analyses. Similarly, we did not observe an association between APOE-ε4 and sporadic Alzheimer disease in the randomly selected Caribbean Hispanics, suggesting that APOE confers a weak effect in this population.12,13
Among familial AD, however, we reported14
the APOE-ε4 allele to be strongly associated (Z=4.26, P
=1.01 × 10-5
). In the current study, the joint linkage and association analysis of APOE yielded a P
-value of 0.00802 for linkage disequilibrium under the dominant model, a P
-value of 0.04863 for linkage, and a P
-value of 0.00408 for a joint test of linkage and linkage disequilibrium. As expected, the linkage to APOE was weak. This illustrates the difficulties in identifying relatively common alleles, with weak to modest effects when studying late-onset disorders.
In sum, this was the first genome scan focused on familial AD in Caribbean Hispanics. We observed suggestive evidence for linkage to chromosome 18q and continue to support evidence for linkage to 10q.