Abstract
Background
Genome-wide association studies (GWAS) of late-onset Alzheimer's disease (LOAD) identified risk variants. We assessed the association of nine variants with memory and progression to mild cognitive impairment (MCI) or LOAD (MCI/LOAD).
Methods
Older Caucasians, cognitively normal at baseline and longitudinally evaluated at Mayo Clinic Rochester and Jacksonville, were assessed for associations of genetic variants with memory decline (n=2,262) using linear mixed models and for incident MCI/LOAD (n=2,674) with Cox proportional hazards models. Each variant was tested both individually and collectively using a single weighted risk score.
Results
APOE-ε4 was significantly associated with worse memory at baseline (β=-0.88, p=2.78E-03) and increased rate of 5-year decline (β=-1.43, p=3.71E-06) with highly significant overall effect on memory (p=3.88E-09). CLU-locus risk allele rs11136000-G was associated with worse memory at baseline (β=-0.51, p=0.012), but not with increased rate of decline. CLU allele was also associated with incident MCI/LOAD (hazard ratio=HR=1.14, p=0.049) in sensitivity analysis. MS4A6A-locus risk allele rs610932-C was associated with increased incident MCI/LOAD in primary analysis (HR=1.17, p=0.016) and had suggestive association with lower baseline memory (β=-0.35, p=0.08). PICALM-locus risk allele rs3851179-G had nominally significant HR in both primary and sensitivity analysis, but with a protective estimate. LOAD risk alleles ABCA7-rs3764650-C and EPHA1-rs11767557-A associated with increased rates of memory decline in the subset of subjects with a final diagnosis of MCI/LOAD. Risk scores excluding APOE were not significant, whereas APOE-inclusive risk scores associated with worse memory and incident MCI/LOAD.
Conclusions
The collective influence of the nine top LOAD GWAS variants on memory decline and progression to MCI/LOAD appears limited. Given the significant associations observed with APOE-ε4, discovery of the biologically functional variants at these loci may uncover stronger effects on memory and incident disease.
Keywords: Alzheimer's disease, memory, mild cognitive impairment, genetic risk, association, cognitive decline
1. Introduction
Genome-wide association studies (GWAS) identified single nucleotide polymorphisms (SNPs) at 20 genetic loci in addition to APOE ε4, that are associated with late-onset Alzheimer's disease (LOAD) risk in large case-control series(1-6). These twenty SNPs are unlikely to be functional variants, but are rather markers that tag the biologically functional genetic variation at these loci(7). In addition, although the LOAD risk GWAS loci are identified by the names of the nearest genes, the identities of the LOAD risk genes remain to be established. Although uncovering the pathophysiologic mechanisms that underlie the LOAD risk conferred by the GWAS loci awaits discovery of the functional variants and the disease genes, the GWAS variants can nonetheless be evaluated for their effects on biological quantitative phenotypes of AD. This endophenotype approach offers an opportunity to investigate these variants for their influence on key functional outcomes associated with this complex disease, thereby providing not only additional support for their role in AD risk, but potentially also information on their mechanistic effects.
Cognitive phenotypes constitute an important category of endophenotypes for AD. Current conceptualization of the dynamic changes in AD biomarkers posits that subtle cognitive decline begins prior to the clinical diagnosis of mild cognitive impairment (MCI) and certainly AD (8, 9). Genetic variants that influence cognitive decline in these preclinical stages of AD may serve as predictive factors for this disease. Indeed, APOE ε4, which is the strongest, common genetic risk factor for LOAD, associates with cognitive decline prior to the diagnosis of MCI/AD(10-13). Consistent with the model of clinical progression of AD from preclinical cognitive decline to MCI, and then AD(9), APOE ε4 is also associated with increased incidence of MCI(14), AD(15) or dementia(16).
Studies that evaluate the influence of the LOAD risk GWAS loci variants on cognitive endophenotypes are emerging. CR1 locus variant rs6656401(17), and a coding variant in LD with it(18) were associated with episodic memory decline in a longitudinal cohort of >1,600 elderly subjects. In another study that evaluated CLU, CR1 and PICALM loci SNPs, CLU and CR1 variants associated with more rapid global cognitive decline and PICALM with earlier age at midpoint of cognitive decline(19) in 1,831 subjects. In a relatively small cohort of 95 cognitively normal subjects who developed MCI or AD, those with the risky CLU allele had a more rapid cognitive decline(20).
These studies are informative; however to date there are no reports that investigate the rate of memory decline for association with the larger number of published LOAD GWAS risk loci either individually or as a single weighted risk score. At the time of our study, nine loci were reported from LOAD GWAS(2-5). In our study, we evaluate a longitudinally followed cohort of >2,000 elderly, Caucasian subjects, who were cognitively normal at baseline, for association of memory decline and with incident MCI/LOAD with these nine LOAD GWAS variants. We also investigate their ability to discriminate between subjects that develop MCI/LOAD from those that do not. Our findings provide a paradigm for the assessment of LOAD risk variants for their effects on memory decline and progression to MCI/LOAD both individually and collectively as weighted risk scores.
2. Methods and Subjects
2.1 Subjects
We assessed an elderly, Caucasian cohort of subjects all of whom were clinically normal at baseline and followed by Behavioral Neurologists either at Mayo Clinic Rochester, Minnesota (MCR) or Mayo Clinic Jacksonville, Florida (MCJ). Incident MCI was diagnosed according to Petersen criteria(21) and clinically possible or probable AD was determined according to NINCDS-ADRDA criteria(22). All subjects underwent 2 or more clinical evaluations. The 30-minute delayed recall scores (LMDR) from the Wechsler Memory Scale-Revised(23) Logical Memory subtest were used as the cognitive endophenotypes. For the analyses assessing progression to MCI/LOAD, a total of 2,674 subjects were evaluated. For the memory analysis (n=2,262), patients were excluded if they had fewer than 2 LMDR scores or if their LMDR score at their initial assessment was 0. The demographics of all subjects are shown in Table 1. All studies were approved by Mayo Clinic's Institutional Review Board.
Table 1. Characteristics of subjects included in the analyses.
a. Median follow-up refers to that for LMDR assessments for subjects in the cognitive decline analyses and is the time to first MCI/AD or last follow-up for those in the time to MCI/AD analyses. MCR: Mayo Clinic Rochester in Minnesota, MCJ: Mayo Clinic Jacksonville in Florida. LMDR: Logical Memory Delayed Recall scores from the Wechsler Memory Scale-revised.
| Variable | Cognitive Decline Analysis | Time to MCI/AD Analysis | ||||
|---|---|---|---|---|---|---|
|
| ||||||
| Overall (N=2262) |
MCR (N=1800) |
MCJ (N=462) |
Overall (N=2674) |
MCR (N=2228) |
MCJ (N=446) |
|
|
| ||||||
|
Median age at 1st assessment (range), years |
77 (49-98) | 78 (55-98) | 72 (49-91) | 77 (48-98) | 78 (55-98) | 74 (48-94) |
|
| ||||||
| Male gender - n (%) | 991 (44) | 842 (47) | 149 (32) | 1187 (44) | 1034 (46) | 153 (34) |
|
| ||||||
|
Median years of education (range) |
14 (4-20) | 13 (5-20) | 16 (4-20) | 14 (4-20) | 13 (5-20) | 16 (4-20) |
|
| ||||||
| Last diagnosis - n (%) | ||||||
| Normal | 1881 (83) | 1467 (82) | 414 (90) | 2166 (81) | 1778 (80) | 388 (87) |
| MCI | 252 (11) | 236 (13) | 16 (3) | 347 (13) | 332 (15) | 15 (3) |
| AD | 129 (6) | 97 (5) | 32 (7) | 132 (5) | 101 (5) | 31 (7) |
| Other | - | - | - | 29 (1) | 17 (1) | 12 (3) |
|
| ||||||
|
Median follow-up (range), monthsa |
45.9 (8.9-214.1) | 44.6 (8.9-214.1) | 69.1 (10.2-190.4) | 61.7 (0.1-269.3) | 60.7 (8.0-269.3) | 66.8 (0.1-245.3) |
|
| ||||||
| APOE ε4 alleles – n (%) | ||||||
| 0 | 1681 (74) | 1361 (76) | 320 (69) | 1991 (74) | 1679 (75) | 312 (70) |
| 1 | 544 (24) | 412 (23) | 132 (29) | 636 (24) | 512 (23) | 124 (28) |
| 2 | 37 (2) | 27 (2) | 10 (2) | 47 (2) | 37 (2) | 10 (2) |
|
| ||||||
| Median LMDR at 1st assessment (range) | 17 (1-42) | 16 (1-40) | 20 (1-42) | - | - | - |
2.2 Genotyping
The most significant LOAD risk GWAS SNPs from 9 loci(1-5) near CLU, PICALM, CR1, ABCA7, BIN1, MS4A6A, EPHA1, CD2AP, and CD33,in addition to two SNPs defining APOE alleles (rs429358 and rs7412) were genotyped using TaqMan® assays. The genotype frequencies of the SNPs are depicted in Supplementary Table 1.
2.3 Statistical Analysis
All analyses were conducted with each genetic variant tested individually, as well as with weighted risk scores. Two weighted risk scores, both of which included all 9 LOAD risk GWAS SNPs, but one with and one without APOE ε4, were calculated based on previously reported odds ratio (OR) estimates from large AD risk GWAS(4) or their follow-up studies(24-26), according to the following formula: Scorei = Σ (nij * log(ORj)) for the ith patient, where: nij= number of risk alleles for the ith patient and jth SNP; ORj=odds ratio for the jth SNP. The contribution of each variant to the risk score is shown in Supplementary Table 2. Those subjects missing two or more SNPs were excluded from the risk score analyses. If a subject was missing only one SNP, the mean number of risk alleles for that SNP across all other subjects was used in the calculation of the risk score for that subjects.
Linear mixed-effects models with subject-specific random slopes and intercepts were used to evaluate associations of each variant and the risk scores with LMDR. The models included time from the initial LMDR assessment as the time scale in 5-year increments with site (MCJ=1, MCR=0), age at baseline, gender (male=1, female=0), and years of education as covariates. Unless APOE ε4 was being evaluated for association, number of APOE ε4 alleles was also included as a covariate in all models. The impact of each covariate in the model on trends in LMDR over time was evaluated through the inclusion of a time interaction term for each variable. Coefficients (β) for the intercept are interpreted as the effect of each additional risk allele on the baseline LMDR score, where the risk allele was identified from LOAD risk GWAS(1-6). Coefficients for the slope are interpreted as changes in the 5-year rate of LMDR for each additional risk allele or 1 standard deviation increase in the risk score. For each genetic variant, we performed a likelihood ratio test to compare the fit of the full model with a reduced model omitting the genetic variant and its time interaction to evaluate whether there was an overall effect of the genetic variant on LMDR.
Primary analysis for memory associations were conducted on all subjects without discrimination for last diagnosis of MCI/LOAD vs. clinically normal. We also performed secondary analyses where changes in the 5-year rate of LMDR by genotype were estimated separately for subjects with a last diagnosis of MCI/LOAD and those with a last diagnosis of normal. These secondary analyses differed from the primary analyses only in their inclusion of separate time interaction terms by genotype for the two “last diagnoses” categories and another time interaction variable for last diagnosis of MCI/LOAD.
Associations with risk of progression to LOAD or MCI (MCI/LOAD) were evaluated using Cox proportional hazard regression models that included time from baseline as the time scale and adjusted for site, gender, age, years of education, with or without adjustment for the number of APOE ε4 alleles, as described above. The primary endpoint for these analyses was the time to first diagnosis of MCI/LOAD; and those subjects who did not develop MCI/LOAD were censored at the time of their last assessment. In sensitivity analyses, those patients who were diagnosed with MCI/LOAD during their longitudinal assessment but whose last diagnosis was neither MCI nor LOAD were censored at their last assessment. There were 77 such patients, 3 of whom had a diagnosis of LOAD and 74 with a diagnosis of MCI at some point during their longitudinal assessment. Seventy of these subjects had a final diagnosis of “normal” and 7 had a final diagnosis of “other”, and were consequently censored at their last diagnosis in the sensitivity analysis. Hazard ratios (HR) and 95% CI were given for the time-to-MCI/LOAD and for the sensitivity analyses.
Concordance index (c-index) for the prediction of conversion to MCI/LOAD was calculated for the subset of patients who had genotypes for each of the 9 genetic variants (n=2,240). Interpretation of the c-index is similar to the interpretation of the area under the receiver operating characteristic (ROC) curve: a value of 1.0 indicates that the variables in the Cox model perfectly discriminate patients with different outcomes, while a value of 0.5 indicates that the variables contain no predictive information. All analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, North Carolina). Since 10 variants were tested (APOE+9 LOAD GWAS SNPs), P-values <0.005 were considered to be study-wide significant after Bonferroni correction. P-values <0.05 and ≥0.005 were nominally significant, and those ≤0.1 and ≥0.05 were considered to be suggestive .
3. Results
We evaluated elderly, Caucasian subjects who were cognitively normal at baseline and followed longitudinally for association of their memory scores or incident MCI/LOAD with 9 LOAD risk GWAS loci SNPs. Given the restrictions for the cognitive decline analysis based on available LMDR scores, there were less subjects for this analysis compared to those for the time to MCI/LOAD analyses, but the subjects from these analytic groups were overlapping and had similar characteristics (Table 1). Subjects from MCR were older, had fewer years of education, greater male frequency and lower median LMDR scores compared to MCJ subjects. Increased baseline age and male gender were associated with lower baseline LMDR scores, whereas increased years of education were associated with higher scores (Table 2). The MCJ site, where subjects have higher education and are younger, also associated with higher baseline LMDR scores. The MCR site and higher baseline age were associated with faster rate of decline in memory. Male gender had an estimated effect consistent with greater memory decline, but did not reach significance. APOE ε4 was strongly associated with lower baseline memory scores and faster decline, with a highly significant overall effect on memory (p=3.88×10-9).
Table 2. Effect of APOE on logical memory (LMDR).
Coefficients for the estimated change in LMDR, 95% confidence intervals (CI) and P-values are shown for all tested variables. A likelihood ratio test comparing the full model to a reduced model omitting APOE and the interaction of APOE with time was performed to evaluate the overall effect of APOE on LMDR.
| Variables | Coefficient (95% CI) | P-value |
|---|---|---|
|
| ||
| Baseline (intercept) | 17.66 (17.17 to 18.15) | |
|
| ||
| APOE (no. ε4 alleles) | -0.88 (-1.45 to -0.30) | 2.78E-03 |
| MCJ site | 1.60 (0.86 to 2.33) | 2.00E-05 |
| Baseline age (5 yrs) | -1.35 (-1.58 to -1.12) | 3.84E-30 |
| Male gender | -1.07 (-1.63 to -0.50) | 2.30E-04 |
| Yrs. of education (5 yrs) | 2.73 (2.24 to 3.21) | 6.75E-28 |
|
| ||
| 5-Year Change (slope) | 0.79 (0.26 to 1.31) | 3.18E-03 |
|
| ||
| APOE (no. ε4 alleles) | -1.43 (-2.04 to -0.83) | 3.71E-06 |
| MCJ site | 4.17 (3.44 to 4.89) | 6.91E-29 |
| Baseline age (5 yrs) | -0.65 (-0.88 to -0.41) | 1.01E-07 |
| Male gender | -0.28 (-0.89 to 0.33) | 0.37 |
| Yrs. of education (5 yrs) | 0.29 (-0.24 to 0.82) | 0.28 |
|
| ||
| Overall effect of APOE | 3.88E-09 | |
When the nine LOAD GWAS risk loci variants were tested for association with longitudinal memory scores while adjusting for APOE and the other variables shown in Table 2, only CLU locus SNP rs11136000 had nominally significant results (Table 3). The risk allele of this SNP associated with lower baseline LMDR scores. Although this allele was associated with a more negative slope of memory decline, it did not reach significance. CLU-rs11136000 had an overall nominally significant effect on memory, though this would not hold up to adjustment for multiple testing. MS4A6A-rs610932 risk allele had suggestive lower baseline memory scores, but did not have significant association with either slope of memory decline or a significant overall effect.
Table 3. Effect of LOAD risk GWAS loci SNPs on logical memory.
Coefficients for the estimated change in LMDR, 95% confidence intervals (CI) and P-values are shown for all tested variables and are interpreted as either the estimated change in baseline LMDR or the estimated change in LMDR over a 5-year period for a. each additional copy of the risk allele or b. 1 standard deviation increase in the risk score. Each model was adjusted for site, baseline age, gender, years of education and APOE ε4, c. except for the model testing the risk score which includes APOE ε4. Interactions with time were included for each variable in the model. Likelihood ratio tests were performed to evaluate the overall effect of each SNP or risk score on LMDR.
| Variable Type | Risk Variable | Tested Effects | Coefficient (95% CI) |
P-value | Overall p-value |
|---|---|---|---|---|---|
|
Nearest gene SNP ID (risk allele)a |
CLU rs11136000 (G) |
Baseline 5-year change |
-0.51 (-0.92 to -0.110) -0.23 (-0.68 to 0.22) |
0.012 0.32 |
0.012 |
|
PICALM rs3851179 (G) |
Baseline 5-year change |
0.30 (-0.12 to 0.71) 0.05 (-0.41 to 0.50) |
0.16 0.83 |
0.32 | |
|
CR1 rs3818361 (A) |
Baseline 5-year change |
0.31 (-0.18 to 0.80) 0.04 (-0.50 to 0.58) |
0.22 0.88 |
0.42 | |
|
ABCA7 rs3764650 (C) |
Baseline 5-year change |
-0.42 (-1.13 to 0.28) -0.12 (-0.85 to 0.62) |
0.24 0.76 |
0.41 | |
|
BIN1 rs744373 (G) |
Baseline 5-year change |
0.14 (-0.31 to 0.59) -0.23 (-0.72 to 0.26) |
0.55 0.36 |
0.60 | |
|
MS4A6A rs610932 (C) |
Baseline 5-year change |
-0.35 (-0.75 to 0.05) 0.11 (-0.33 to 0.56) |
0.08 0.63 |
0.22 | |
|
EPHA1 rs11767557 (A) |
Baseline 5-year change |
-0.07 (-0.56 to 0.42) -0.15 (-0.67 to 0.38) |
0.77 0.58 |
0.78 | |
|
CD2AP rs9349407 (C) |
Baseline 5-year change |
-0.34 (-0.78 to 0.11) -0.01 (-0.49 to 0.47) |
0.14 0.97 |
0.31 | |
|
CD33 rs3865444 (C) |
Baseline 5-year change |
-0.20 (-0.63 to 0.23) 0.32 (-0.15 to 0.78) |
0.37 0.18 |
0.34 | |
| Risk scoreb | Risk score 1 with 9 SNPs |
Baseline 5-year change |
-0.19 (-0.48 to 0.10) -0.08 (-0.39 to 0.23) |
s0.20 0.63 |
0.31 |
| Risk score 2c with 9 SNPs & APOE |
Baseline 5-year change |
-0.50 (-0.78 to -0.21) -0.64 (-0.95 to -0.34) |
6.70E-04 4.00E-05 |
1.21E-08 |
Given that rate of cognitive decline and its interaction with risk variants may differ between subjects who eventually develop MCI/LOAD vs. those who remain clinically normal, we also performed secondary analysis, which assessed the association of risk genotypes or scores with rate of change in LMDR separately for these two groups of subjects (Supplementary Tables 3 and 4). The baseline and overall effects of APOE on memory remain significant in this secondary analysis (p=1.79E-03 and 3.10E-4, respectively) (Supplementary Table 3). Although APOE ε4 has trends for faster memory decline in both subjects with last diagnosis of MCI/LOAD (β=-0.74, p=0.15) and those that remain normal (β=-0.44, p=0.17), these terms fail to reach significance.
Assessment of the LOAD GWAS loci on memory decline separately for subjects who eventually developed MCI/LOAD vs. those who remained as normal identified association of two LOAD risk alleles, ABCA7-rs3764650-C (p=0.013) and EPHA1-rs11767557-A (p=0.050) with faster rates of decline (Supplementary Table 4). ABCA7 locus also had nominally significant overall association with memory (p=0.018) in these analyses. The effects of the CLU and MS4A6A loci on baseline memory detected in the primary analyses (Table 3) remain essentially unchanged in these secondary analyses. Risk score including APOE retains its significance for effects on baseline memory and overall association with LMDR, however no longer has significant association with faster memory decline in either diagnostic category. CD33-locus risk allele has suggestive, though lower rate of memory decline in subjects who remain cognitively normal.
We next evaluated the influence of a single weighted genetic risk score variable, obtained from the 9 LOAD risk GWAS SNPs (Supplementary Table 2), on memory. Although the risk score excluding APOE had a negative estimated effect on baseline LMDR, this did not achieve significance. There was no effect of this 9-SNP risk score on the slope of memory change. When APOE was included in the risk score, as expected there was significant association with both lower baseline LMDR scores and a negative slope of memory change, with highly significant overall effect (p=1.21×10-8), similar to that seen for APOE alone.
We evaluated the association of the 9 LOAD GWAS SNPs with the risk of progression to MCI/LOAD in our longitudinally followed cohort (Table 4). The primary analysis assessed the time to first diagnosis of MCI/LOAD, regardless of last diagnosis, whereas the sensitivity analysis was censored at last visit for those subjects who had a diagnosis of MCI/LOAD at any point during their assessment but whose last diagnosis was something other than MCI/LOAD. Variants that showed evidence of an association with the risk of progression to MCI/LOAD included the CLU locus SNP rs11136000 risk allele (HR=1.10, p=0.13; Sensitivity HR=1.14, p=0.049) and the MS4A6A locus SNP rs610932 risk allele (HR=1.17, p=0.016; Sensitivity HR=1.11, p=0.11). The only other variant with a nominally significant result was the PICALM locus variant rs3851179, which had a protective HR in both analyses (HR=0.85, p=0.010; Sensitivity HR=0.82, p=0.0045), which is in the opposite direction than would be expected. The weighted risk score for the 9 SNPs did not have a significant HR for progression to MCI/LOAD, in either analysis. When APOE ε4 was included in the risk, there was significant association with risk of progression to MCI/LOAD (HR=1.29, p=1.14×10-9; Sensitivity HR=1.32, p=5.73×10-10).
Table 4. Effect of LOAD risk GWAS loci SNPs on risk of progression to MCI or LOAD.
Hazard ratios (HR) and 95% confidence intervals (CI) were obtained for the genetic variants. Hazard ratios correspond to a. an additional risk allele or b. 1 standard deviation increase in the risk score. All models were adjusted for site, gender, age, years of education and APOE ε4, c. except for the models testing the effect of APOE ε4.
| Variable Type | Nearest gene SNP ID (risk allele) or risk score |
Time-to-MCI/LOAD analysis | Sensitivity analysis | ||||
|---|---|---|---|---|---|---|---|
| HR | 95% CI | P-value | HR | 95% CI | P-value | ||
|
Nearest gene - SNP ID (risk allele)a |
APOE rs429358 (ε4)c |
1.71 | 1.46-1.99 | 1.09E-10 | 1.80 | 1.52-2.11 | 4.98E-11 |
|
CLU rs11136000 (G) |
1.10 | 0.97-1.24 | 0.13 | 1.14 | 1.00-1.30 | 0.049 | |
|
PICALM rs3851179 (G) |
0.85 | 0.75-0.96 | 0.010 | 0.82 | 0.72-0.94 | 4.54E-03 | |
|
CR1 rs3818361 (A) |
1.11 | 0.95-1.28 | 0.19 | 1.12 | 0.95-1.32 | 0.16 | |
| ABCA7 rs3764650 (C) |
1.05 | 0.85-1.27 | 0.67 | 1.07 | 0.85-1.31 | 0.57 | |
|
BIN1 rs744373 (G) |
1.08 | 0.94-1.24 | 0.27 | 1.05 | 0.90-1.22 | 0.54 | |
|
MS4A6A rs610932 (C) |
1.17 | 1.03-1.32 | 0.016 | 1.11 | 0.98-1.28 | 0.11 | |
|
EPHA1 rs11767557 (A) |
0.99 | 0.86-1.14 | 0.84 | 0.97 | 0.83-1.13 | 0.68 | |
|
CD2AP rs9349407 (C) |
1.02 | 0.89-1.17 | 0.76 | 1.04 | 0.90-1.21 | 0.57 | |
|
CD33 rs3865444 (C) |
1.00 | 0.88-1.14 | 0.98 | 0.95 | 0.83-1.10 | 0.51 | |
| Risk scoreb | Risk score with 9 SNPs |
1.03 | 0.95-1.13 | 0.43 | 1.02 | 0.93-1.11 | 0.75 |
| Risk score with 9 SNPs & APOEc |
1.29 | 1.19-1.39 | 1.14E-09 | 1.32 | 1.21-1.43 | 5.73E-10 | |
C-index estimates which are indicative of the ability of variables to discriminate between subjects who develop MCI/LOAD and those who do not are displayed in Supplementary Table 5. The c-index was only slightly improved with addition of APOE (0.685) relative to a reduced model with only age, gender, education and site (0.674). Similarly, addition of the risk score that included APOE (0.684) only barely increased the c-index compared to the reduced model. Inclusion of the other 9 variants individually or the combined risk score excluding APOE did not improve the c-index over the reduced model.
4. Discussion
Understanding the influence of the novel LOAD risk GWAS loci variants on cognitive endophenotypes can provide additional information about their plausible mechanism of action. Genetic variants that associate with cognition in a non-time-dependent fashion may suggest static effects, whereas those that associate with rate of cognitive decline may imply a dynamic influence on biological mechanisms underlying cognitive outcomes. We previously evaluated three LOAD risk GWAS SNPs at CLU, CR1 and PICALM loci for association with verbal and non-verbal episodic memory scores at last evaluation and identified better memory scores in Caucasian subjects with the protective CLU rs11136000 SNP alleles and worse scores in African-American subjects with the risky CR1 SNP (rs6656401, rs3818361) alleles(27). In the current study, we evaluated the association of nine LOAD risk GWAS loci SNPs for association with memory scores in a longitudinally assessed cohort and identified strong association between APOE ε4 with both baseline logical memory and increased rate of memory decline, as expected(10-13). As in our prior study(27), we also identified lower memory scores at baseline associated with the risk allele of the CLU locus SNP rs11136000. Although this consistency may be expected given that the cohort from our prior study and the current one largely overlap, the approach used in this study is distinct because we evaluated baseline and longitudinal cognitive change instead of the last memory score as in our prior work. We did not find significant association with rate of memory decline and rs11136000, although the LOAD risk allele had an estimated negative slope of decline. Both our prior(27) and current findings suggest that the CLU locus influences cognition, and this effect appears to be static rather than a dynamic effect on the rate of decline. Importantly, the direction of the cognitive associations with the CLU locus SNP are congruent with the allelic effects on LOAD risk.
CLU locus SNP rs11136000 was found to associate with cognitive endophenotypes in other studies(19, 20). Fitting a Bayesian model in an initial cohort of 802 subjects, Sweet et al.(19) tested the model in a second cohort of 1,831 subjects who were dementia-free at baseline for association of SNPs at CLU, PICALM and CR1 with cognitive endophenotypes. In that study, they found association between CLU rs11136000 with rate of global cognitive decline, although the LOAD protective allele was associated with faster decline, which is biologically incongruent. Thambisetty et al.(20) evaluated CLU rs11136000 in a cohort of 599 subjects who remained cognitively normal and 95 subjects who converted to MCI or AD in a longitudinal study. There were no significant associations with cognitive decline in the non-converters but significantly greater rate of memory decline was observed in CLU rs11136000 risk allele carriers amongst the smaller cohort of converters. Collectively, these studies and our findings support a role for CLU locus variants in memory endophenotypes, although the effects on static vs. dynamic memory outcomes and the associating allele need to be firmly established.
Since rates of cognitive decline may differ in subjects who eventually convert to MCI/LOAD vs. those who remain clinically normal, we also assessed the effect of LOAD risk variants on memory decline separately in subjects that pertain to these two “last diagnosis” categories. We determined associations with faster memory decline for the risk alleles ABCA7-rs3764650-C and EPHA1-rs11767557-A in those subjects who eventually convert to MCI/LOAD. These findings suggest that ABCA7 and EPHA1 may have an effect on dynamic memory outcomes. That we can only detect an effect on memory decline in subjects who eventually develop MCI/LOAD may be due to the fact that these subjects are at a later stage in their underlying disease process and therefore are more susceptible to the effects of genetic risk factors. Another possibility is that subjects who are “destined for” MCI/LOAD may have other genetic or environmental risk factors that enhance the effects of the tested genetic risk variants on cognitive decline through interactions.
While separate analysis of different diagnostic categories can be informative, as discussed above, this can also lead to loss of power and therefore significance as we observed for APOE ε4 and the risk score including APOE for rate of memory decline, although associations with baseline memory and overall effects remained significant. This is likely due to analysis of smaller number of subjects, as well as overlap between effects of APOE ε4 allele and MCI/LOAD diagnosis on rate of decline. These results highlight the importance of both total group analyses as well as separate assessments for those with incident MCI/LOAD vs. those that remain clinically normal.
In addition to evaluating memory endophenotypes, we also tested the association of these nine LOAD risk GWAS SNPs for their effects on progression to MCI or LOAD. Genetic variants may influence risk of LOAD by accelerating progression to clinically detectable cognitive decline. Further, genetic risk factors may also continue to influence rate of disease progression after clinical diagnosis of AD or MCI. Alternatively, genetic factors may underlie biological processes that confer a static cognitive disadvantage. Thus, testing the effects of LOAD risk GWAS SNPs for incident LOAD or MCI may yield further information about their mechanism of action. In our study, CLU rs11136000 risk allele has an increased hazard ratio for progression to LOAD or MCI. The number of studies evaluating the role of LOAD GWAS SNPs in the rate of progression to MCI/LOAD are yet limited. Rodriguez-Rodriguez et al.(28) investigated the association of eight LOAD risk GWAS loci SNPs with both risk and rate of progression to AD in 297 MCI subjects, 118 of whom were converters. In that study, CLU rs11136000 was associated with risk but not with rate of progression from MCI to AD, unlike in our study. That we are not detecting association with rate of memory decline, but find association with rate of progression to disease may seem inconsistent, however may be due to limited power of alleles with modest effect sizes on these tested outcomes. It is also possible that there may be decline in non-memory cognitive domains, which we did not evaluate.
MS4A6A-locus rs610932-C risk allele was associated with progression to MCI/LOAD in primary analysis with suggestive results for the sensitivity analysis. This allele also had suggestively lower baseline memory estimates. PICALM locus risk allele rs3851179-G had nominally significant association with protective HR estimates in both the primary and sensitivity analyses, which is biologically inconsistent with expected effects based on LOAD risk estimates. Carriers of this risk allele were previously found by others to have a more rapid rate of clinical decline as determined by changes in the Clinical Dementia Rating-sum of boxes as a quantitative trait in 822 Caucasian subjects with amnestic MCI(29), consistent with effects on LOAD risk. The biologically incongruent protective effect in our cohort of the PICALM locus risk allele may be a false positive finding. Alternatively, these opposing effects may be due to the “flip-flop” phenomenon(30) that may ensue when the LD structure between the tested variant and the functional variant(s) differ between studies. Resolution of such findings awaits discovery and testing of putative functional variants at disease risk loci.
To assess the combined influence of genetic risk variants on memory decline and progression to AD or MCI, we utilized a single risk score weighted by the estimated ORs for each of the nine LOAD GWAS variants plus or minus APOE. The risk scores which included APOE had strong associations with lower baseline memory, increased rate of memory decline and faster rate of progression to MCI/LOAD. The risk scores which lacked APOE did not achieve significance for any of the tested outcomes, although they had a negative estimate for baseline memory. These findings are similar to those identified by Verhaaren et al.(31), in a population-based cohort of 5,171 subjects who were nondemented at baseline. That study found only marginal influence of risk scores using 10 GWAS loci variants, on baseline memory and risk of developing AD, despite robust associations when APOE was included in the risk score. Likewise, genetic risk scores based on 8 variants did not associate with risk of conversion from 288 MCI subjects to AD in the Rodriguez-Rodriguez et al.(28) study, although faster rate of progression was identified for the second and third tertile of risk score carriers vs. the first tertile.
The GWAS loci variants or the risk scores had no added value over the non-genetic variables of site, gender, age and years of education for discriminating MCI/LOAD converters. Importantly, even APOE offered very little additional predictive value. This is similar to findings in two population-based cohorts (Rotterdam and Cardiovascular Health Study), in which the value of adding APOE to age and sex variables for predicting progression to AD was minimal and addition of CLU and PICALM loci SNPs provided essentially no further improvement in prediction(3). Similarly, a 10-SNP risk score had essentially no value for predicting conversion to AD in the Rotterdam cohort(31). It has been suggested that non-genetic variants provide significant predictive power(3), therefore even APOE that has strong associations with memory and rate of progression to LOAD or MCI does not contribute significantly to prediction of incident AD above and beyond these non-genetic variables. Despite these findings, it is important to continue to explore these predictive models by including new genetic risk variants as they are uncovered.
Our study has a number of strengths, including a sizable cohort without dementia or MCI at baseline, evaluation of baseline and longitudinal memory associations, in addition to progression to MCI/LOAD with 9 LOAD risk variants both individually and as a single risk score, as well as exploration of their predictive value for MCI/LOAD. The strong associations of APOE with memory at baseline, as well as rate of memory decline and progression to MCI/LOAD provide proof of principle for our approach. The biologically congruent memory and incident disease associations with the CLU locus SNP provide additional support for this locus in influencing memory and disease progression in pre-clinical stages of AD. Likewise, the incident MCI/LOAD risk associations with the MS4A6A-locus risk allele are supportive for this locus. PICALM locus associations for progression to AD needs further assessment due to inconsistent direction of effect compared with the AD risk associating allele.
The lack of associations with the other variants and the genetic risk score may be a reflection of small effect sizes of the tested SNPs, assessment of marker polymorphisms rather than the functional variants, the sensitivity of the memory measure used to the earliest effect of neuropathology, potential survival bias in our elderly cohort where some of the variants may have stronger effects on cognitive or clinical decline earlier on, or the incompleteness of the tested models as there are clearly additional genetic and non-genetic variables to be uncovered that may improve the models, including those for prediction. The approaches used in our study can generalize to other genetic variants that are expected to emerge from AD risk variant discovery efforts and therefore provide valuable information.
Supplementary Material
Supplementary Table 1. Genotype counts: Percent frequencies are shown in parentheses.
Supplementary Table 2. Risk score calculations. Risk score was calculated based on previously reported odds ratio estimates from large AD risk GWAS. Scorei = Σ (nij * log(ORj)) for the ith patient, where: nij= number of risk alleles for the ith patient and jth SNP; ORj=odds ratio for the jth SNP. For purposes of calculating the risk score, the average number of risk alleles for a given SNP was used for any subject missing only 1 SNP. Two separate risk scores were computed, 1 with and 1 without APOE.
Supplementary Table 3: Effect of APOE on logical memory decline according to last diagnosis: Coefficients for the estimated change in LMDR, 95% confidence intervals (CIs) and P-values are shown for all tested variables. A likelihood ratio test comparing the full model to a reduced model omitting APOE, the interaction of APOE with time among subjects with either MCI/LOAD or normal as a last diagnosis was performed to evaluate the overall effect of APOE on LMDR.
Supplementary Table 4: Effect of LOAD risk GWAS loci on logical memory decline according to last diagnosis: Coefficients for the estimated change in LMDR, 95% confidence intervals (CIs) and P-values are shown for all tested variables and are interpreted as either the estimated change in baseline LMDR, or the estimated change in LMDR over a 5-year period among subjects with a last diagnosis of MCI/LOAD or normal for a. each additional copy of the risk allele or b. 1 standard deviation increase in the risk score. Each model was adjusted for site, baseline age, gender, years of education and APOE ε4, c. except for the model testing the risk score which includes APOE ε4. Interactions with time were included for each variable in the model. An interaction of genotype with time was included in each model separately for those with a last diagnosis of MCI/LOAD and those with a last diagnosis of normal. Likelihood ratio tests were performed to evaluate the overall effect of each SNP or risk score on LMDR by comparison of the full model to a reduced model omitting the genotype and its interactions with time for both diagnostic groups.
Supplementary Table 5. Concordance index (C-index) for the time to LOAD or MCI analysis in a subset of 2240 patients who had all genetic variables. Interpretation of the c-index is similar to the interpretation of the area under the receiver operating characteristic (ROC) curve. A value of 1.0 indicates that the variables in the Cox model perfectly discriminate patients with different outcomes, while a value of 0.50 indicates that the variables contain no predictive information.
Acknowledgments
Support for this research was provided by the National Institutes of Health grants: National Institute on Aging (R01 AG032990 to NET and R01 AG018023 to NRG-R and SGY); National Institutes on Neurologic Diseases and Stroke (R01 NS080820 to NET), Mayo Alzheimer's Disease Research Center: (P50 AG0016574 to RCP, DWD, NRG-R, SGY, and NET); Mayo Alzheimer's Disease Patient Registry: (U01 AG006576 to RCP); National Institute on Aging (AG025711, AG017216, AG003949 to DWD). This project was also generously supported by the Robert and Clarice Smith and Abigail Van Buren Alzheimer's Disease Research Program (to RCP, DWD, NRG-R, and SGY), and by the Palumbo Professorship in Alzheimer's Disease Research (to SGY). MMC and NET are supported partly by GHR Foundation grants. We thank Dr. Richard J. Caselli for useful discussion of the manuscript. We are grateful to our patients and their families for their participation, without whom these studies would not have been possible.
Footnotes
Disclosures: R.C. Petersen, M.D., Ph.D. has been a consultant to GE Healthcare and Elan Pharmaceuticals, has served on a data safety monitoring board in clinical trials sponsored by Pfizer Incorporated and Janssen Alzheimer Immunotherapy and gave a CME lecture at Novartis Incorporated. N. Graff-Radford, M.D. has served as a consultant to Codman and received grant support from Elan Pharmaceutical Research, Pfizer Pharmaceuticals, Medivation, and Forrest.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Lambert JC, Heath S, Even G, Campion D, Sleegers K, Hiltunen M, et al. Genome-wide association study identifies variants at CLU and CR1 associated with Alzheimer's disease. Nature genetics. 2009;41(10):1094–9. doi: 10.1038/ng.439. [DOI] [PubMed] [Google Scholar]
- 2.Harold D, Abraham R, Hollingworth P, Sims R, Gerrish A, Hamshere ML, et al. Genome-wide association study identifies variants at CLU and PICALM associated with Alzheimer's disease. Nature genetics. 2009 doi: 10.1038/ng.440. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Seshadri S, Fitzpatrick AL, Ikram MA, DeStefano AL, Gudnason V, Boada M, et al. Genome-wide analysis of genetic loci associated with Alzheimer disease. JAMA. 2010;303(18):1832–40. doi: 10.1001/jama.2010.574. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Hollingworth P, Harold D, Sims R, Gerrish A, Lambert JC, Carrasquillo MM, et al. Common variants at ABCA7, MS4A6A/MS4A4E, EPHA1, CD33 and CD2AP are associated with Alzheimer's disease. Nature genetics. 2011;43(5):429–435. doi: 10.1038/ng.803. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Naj AC, Jun G, Beecham GW, Wang LS, Vardarajan BN, Buros J, et al. Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 are associated with late-onset Alzheimer's disease. Nature genetics. 2011;43(5):436–41. doi: 10.1038/ng.801. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Lambert JC, Ibrahim-Verbaas CA, Harold D, Naj AC, Sims R, Bellenguez C, et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer's disease. Nature genetics. 2013;45(12):1452–8. doi: 10.1038/ng.2802. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Ferrari R, Moreno JH, Minhajuddin AT, O'Bryant SE, Reisch JS, Barber RC, et al. Implication of common and disease specific variants in CLU, CR1, and PICALM. Neurobiology of aging. 2012;33(8):1846, e7–18. doi: 10.1016/j.neurobiolaging.2012.01.110. [DOI] [PubMed] [Google Scholar]
- 8.Jack CR, Jr, Knopman DS, Jagust WJ, Petersen RC, Weiner MW, Aisen PS, et al. Tracking pathophysiological processes in Alzheimer's disease: an updated hypothetical model of dynamic biomarkers. Lancet neurology. 2013;12(2):207–16. doi: 10.1016/S1474-4422(12)70291-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, Fagan AM, et al. Toward defining the preclinical stages of Alzheimer's disease: Recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement. 2011;7(3):280–92. doi: 10.1016/j.jalz.2011.03.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Caselli RJ, Dueck AC, Osborne D, Sabbagh MN, Connor DJ, Ahern GL, et al. Longitudinal modeling of age-related memory decline and the APOE epsilon4 effect. The New England journal of medicine. 2009;361(3):255–63. doi: 10.1056/NEJMoa0809437. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Caselli RJ, Reiman EM, Locke DE, Hutton ML, Hentz JG, Hoffman-Snyder C, et al. Cognitive domain decline in healthy apolipoprotein E epsilon4 homozygotes before the diagnosis of mild cognitive impairment. Archives of neurology. 2007;64(9):1306–11. doi: 10.1001/archneur.64.9.1306. [DOI] [PubMed] [Google Scholar]
- 12.Caselli RJ, Reiman EM, Osborne D, Hentz JG, Baxter LC, Hernandez JL, et al. Longitudinal changes in cognition and behavior in asymptomatic carriers of the APOE e4 allele. Neurology. 2004;62(11):1990–5. doi: 10.1212/01.wnl.0000129533.26544.bf. [DOI] [PubMed] [Google Scholar]
- 13.Bennett DA, De Jager PL, Leurgans SE, Schneider JA. Neuropathologic intermediate phenotypes enhance association to Alzheimer susceptibility alleles. Neurology. 2009;72(17):1495–503. doi: 10.1212/WNL.0b013e3181a2e87d. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Luck T, Riedel-Heller SG, Luppa M, Wiese B, Wollny A, Wagner M, et al. Risk factors for incident mild cognitive impairment--results from the German Study on Ageing, Cognition and Dementia in Primary Care Patients (AgeCoDe) Acta psychiatrica Scandinavica. 2010;121(4):260–72. doi: 10.1111/j.1600-0447.2009.01481.x. [DOI] [PubMed] [Google Scholar]
- 15.Aggarwal NT, Wilson RS, Beck TL, Bienias JL, Berry-Kravis E, Bennett DA. The apolipoprotein E epsilon4 allele and incident Alzheimer's disease in persons with mild cognitive impairment. Neurocase. 2005;11(1):3–7. doi: 10.1080/13554790490903038. [DOI] [PubMed] [Google Scholar]
- 16.Fitzpatrick AL, Kuller LH, Ives DG, Lopez OL, Jagust W, Breitner JC, et al. Incidence and prevalence of dementia in the Cardiovascular Health Study. Journal of the American Geriatrics Society. 2004;52(2):195–204. doi: 10.1111/j.1532-5415.2004.52058.x. [DOI] [PubMed] [Google Scholar]
- 17.Chibnik LB, Shulman JM, Leurgans SE, Schneider JA, Wilson RS, Tran D, et al. CR1 is associated with amyloid plaque burden and age-related cognitive decline. Ann Neurol. 2011;69(3):560–9. doi: 10.1002/ana.22277. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Keenan BT, Shulman JM, Chibnik LB, Raj T, Tran D, Sabuncu MR, et al. A coding variant in CR1 interacts with APOE-epsilon4 to influence cognitive decline. Human molecular genetics. 2012;21(10):2377–88. doi: 10.1093/hmg/dds054. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Sweet RA, Seltman H, Emanuel JE, Lopez OL, Becker JT, Bis JC, et al. Effect of Alzheimer's disease risk genes on trajectories of cognitive function in the Cardiovascular Health Study. The American journal of psychiatry. 2012;169(9):954–62. doi: 10.1176/appi.ajp.2012.11121815. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Thambisetty M, Beason-Held LL, An Y, Kraut M, Nalls M, Hernandez DG, et al. Alzheimer risk variant CLU and brain function during aging. Biological psychiatry. 2013;73(5):399–405. doi: 10.1016/j.biopsych.2012.05.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Petersen RC. Clinical practice. Mild cognitive impairment. The New England journal of medicine. 2011;364(23):2227–34. doi: 10.1056/NEJMcp0910237. [DOI] [PubMed] [Google Scholar]
- 22.McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology. 1984;34(7):939–44. doi: 10.1212/wnl.34.7.939. [DOI] [PubMed] [Google Scholar]
- 23.Wechsler DA. Wechsler Memory Scale-Revised. New York: Psychological Corporation; 1987. [Google Scholar]
- 24.Carrasquillo MM, Belbin O, Hunter TA, Ma L, Bisceglio GD, Zou F, et al. Replication of CLU, CR1, and PICALM associations with alzheimer disease. Archives of neurology. 2010;67(8):961–4. doi: 10.1001/archneurol.2010.147. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Carrasquillo MM, Belbin O, Hunter TA, Ma L, Bisceglio GD, Zou F, et al. Replication of BIN1 association with Alzheimer's disease and evaluation of genetic interactions. Journal of Alzheimer's disease : JAD. 2011;24(4):751–8. doi: 10.3233/JAD-2011-101932. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Carrasquillo MM, Belbin O, Hunter TA, Ma L, Bisceglio GD, Zou F, et al. Replication of EPHA1 and CD33 associations with late-onset Alzheimer's disease: a multi-centre case-control study. Molecular neurodegeneration. 2011;6(1):54. doi: 10.1186/1750-1326-6-54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Pedraza O, Allen M, Jennette K, Carrasquillo M, Crook J, Serie D, et al. Evaluation of memory endophenotypes for association with CLU, CR1, and PICALM variants in black and white subjects. Alzheimer's & dementia : the journal of the Alzheimer's Association. 2013 doi: 10.1016/j.jalz.2013.01.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Rodriguez-Rodriguez E, Sanchez-Juan P, Vazquez-Higuera JL, Mateo I, Pozueta A, Berciano J, et al. Genetic risk score predicting accelerated progression from mild cognitive impairment to Alzheimer's disease. Journal of neural transmission. 2013;120(5):807–12. doi: 10.1007/s00702-012-0920-x. [DOI] [PubMed] [Google Scholar]
- 29.Hu X, Pickering EH, Hall SK, Naik S, Liu YC, Soares H, et al. Genome-wide association study identifies multiple novel loci associated with disease progression in subjects with mild cognitive impairment. Translational psychiatry. 2011;1:e54. doi: 10.1038/tp.2011.50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Lin PI, Vance JM, Pericak-Vance MA, Martin ER. No gene is an island: the flip-flop phenomenon. American journal of human genetics. 2007;80(3):531–8. doi: 10.1086/512133. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Verhaaren BF, Vernooij MW, Koudstaal PJ, Uitterlinden AG, Duijn CM, Hofman A, et al. Alzheimer's Disease Genes and Cognition in the Nondemented General Population. Biological psychiatry. 2012 doi: 10.1016/j.biopsych.2012.04.009. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Table 1. Genotype counts: Percent frequencies are shown in parentheses.
Supplementary Table 2. Risk score calculations. Risk score was calculated based on previously reported odds ratio estimates from large AD risk GWAS. Scorei = Σ (nij * log(ORj)) for the ith patient, where: nij= number of risk alleles for the ith patient and jth SNP; ORj=odds ratio for the jth SNP. For purposes of calculating the risk score, the average number of risk alleles for a given SNP was used for any subject missing only 1 SNP. Two separate risk scores were computed, 1 with and 1 without APOE.
Supplementary Table 3: Effect of APOE on logical memory decline according to last diagnosis: Coefficients for the estimated change in LMDR, 95% confidence intervals (CIs) and P-values are shown for all tested variables. A likelihood ratio test comparing the full model to a reduced model omitting APOE, the interaction of APOE with time among subjects with either MCI/LOAD or normal as a last diagnosis was performed to evaluate the overall effect of APOE on LMDR.
Supplementary Table 4: Effect of LOAD risk GWAS loci on logical memory decline according to last diagnosis: Coefficients for the estimated change in LMDR, 95% confidence intervals (CIs) and P-values are shown for all tested variables and are interpreted as either the estimated change in baseline LMDR, or the estimated change in LMDR over a 5-year period among subjects with a last diagnosis of MCI/LOAD or normal for a. each additional copy of the risk allele or b. 1 standard deviation increase in the risk score. Each model was adjusted for site, baseline age, gender, years of education and APOE ε4, c. except for the model testing the risk score which includes APOE ε4. Interactions with time were included for each variable in the model. An interaction of genotype with time was included in each model separately for those with a last diagnosis of MCI/LOAD and those with a last diagnosis of normal. Likelihood ratio tests were performed to evaluate the overall effect of each SNP or risk score on LMDR by comparison of the full model to a reduced model omitting the genotype and its interactions with time for both diagnostic groups.
Supplementary Table 5. Concordance index (C-index) for the time to LOAD or MCI analysis in a subset of 2240 patients who had all genetic variables. Interpretation of the c-index is similar to the interpretation of the area under the receiver operating characteristic (ROC) curve. A value of 1.0 indicates that the variables in the Cox model perfectly discriminate patients with different outcomes, while a value of 0.50 indicates that the variables contain no predictive information.