SNPs Associated with Cerebrospinal Fluid Phospho-Tau Levels Influence Rate of Decline in Alzheimer's Disease

Alzheimer's Disease (AD) is a complex and multifactorial disease. While large genome-wide association studies have had some success in identifying novel genetic risk factors for AD, case-control studies are less likely to uncover genetic factors that influence progression of disease. An alternative approach to identifying genetic risk for AD is the use of quantitative traits or endophenotypes. The use of endophenotypes has proven to be an effective strategy, implicating genetic risk factors in several diseases, including anemia, osteoporosis and heart disease. In this study we identify a genetic factor associated with the rate of decline in AD patients and present a methodology for identification of other such factors. We have used an established biomarker for AD, cerebrospinal fluid (CSF) tau phosphorylated at threonine 181 (ptau181) levels as an endophenotype for AD, identifying a SNP, rs1868402, in the gene encoding the regulatory sub-unit of protein phosphatase B, associated with CSF ptau181 levels in two independent CSF series . We show no association of rs1868402 with risk for AD or age at onset, but detected a very significant association with rate of progression of disease that is consistent in two independent series . Our analyses suggest that genetic variants associated with CSF ptau181 levels may have a greater impact on rate of progression, while genetic variants such as APOE4, that are associated with CSF Aβ42 levels influence risk and onset but not the rate of progression. Our results also suggest that drugs that inhibit or decrease tau phosphorylation may slow cognitive decline in individuals with very mild dementia or delay the appearance of memory problems in elderly individuals with low CSF Aβ42 levels. Finally, we believe genome-wide association studies of CSF tau/ptau181 levels should identify novel genetic variants which will likely influence rate of progression of AD.


Introduction
Genetic studies have helped to further our understanding of the pathogenic mechanism of several diseases, including AD. To date only the e4 allele of apolipoprotein E (APOE4), present in 50% of late onset AD (LOAD) cases, has been convincingly demonstrated to influence risk for LOAD. The traditional method for searching for genetic risk factors involves the comparison of genes in AD cases and non-demented elderly controls. AD is a complex and multifactorial disease, and as a result very large datasets have been necessary to identify these genetic risk factors [1][2]. An alternative to the standard case-control study design is to use quantitative traits or endophenotypes. Quantitative traits have been used to successfully identify new genetic factors implicated in anemia [3][4][5][6], osteoporosis [7] and heart disease [8][9][10][11]. The advantages of quantitative traits are that they provide higher power than regular case-control analyses, a biological model of disease and the possible effects of the associated genetic variation and may decrease the clinical heterogeneity of the samples. This is likely to be true for Alzheimer's Disease (AD) because up to 30% of individuals in screened elderly non-demented control samples show evidence of AD pathology at autopsy [12], and a similar number have biomarker profiles consistent with preclinical AD [13][14][15], thus reducing the power of a case-control design.
Both Ab and tau protein play an important role in AD, are detectable in cerebrospinal fluid (CSF) in all individuals, and have been used as biomarkers for diagnosis [16][17][18]. Patients with AD show lower CSF Ab 42 levels [19] that inversely correlate with the presence of fibrillar Ab in the brain (as measured by Pittsburgh Compound B (PET-PIB) retention) in demented individuals [14] and plaque counts in brain samples [20]. Several studies suggest that PET-PIB retention and CSF Ab 42 levels could help to identify individuals with AD pathology before the onset of clinically detectable disease (preclinical AD) [14,21]. The CSF levels of total tau and tau phosphorylated at threonine 181 (ptau 181 ) are increased in AD [12,14]. Elevated CSF tau levels are associated with neuronal damage and are also observed in stroke [22] and traumatic brain injury immediately after injury [23], however increases in CSF ptau 181 levels appear to be specific to AD [24][25][26]. In several previous studies we have successfully applied this endophenotype-based approach, leveraging the information from both CSF Ab and tau to identify genetic polymorphisms implicated in AD risk [27][28][29].
Tau activity depends on its state of phosphorylation [30], which is regulated by several kinases, phosphatases and other tau-related proteins [31]. Hyperphosphorylation of tau destabilizes the microtubule network, leading to impaired axonal transport and ultimately to neurofibrillary tangle formation and neuronal death (For review see [32]). In the present study we have evaluated 355 single nucleotide polymorphisms (SNPs) in 34 genes involved in tau modification or metabolism for association with CSF levels of ptau 181 , then determined the effects of those variants on AD risk, onset and rate of progression.

Association with CSF ptau 181 levels: Initial screening
Based on bibliographic data we selected 384 SNPs localized in 34 genes related to tau metabolism (tau kinases, phosphatases, tau O-glcNAcylation or tau degradation (Table S1). 355 SNPs passed quality control (Hardy-Weinberg equilibrium and call rate .95%). Association of SNPs with CSF ptau 181 levels was evaluated by ANCOVA in 353 CSF samples from the Washington University Alzheimer Disease Research Center (WU-ADRC-CSF) (Tables 1 and 2). Clinical Dementia Rating (CDR), age and APOE e4 genotype were included as covariates in the analyses. Eighteen SNPs, located in 7 different genes showed significant association with CSF ptau 181 levels in the WU-ADRC-CSF series after multiple test correction (Table 3). The SNP with the most significant p-value, rs1868402, is located in intron 5 of the regulatory subunit of the protein phosphatase B gene, also known as calcineurin B (PPP3R1; MIM#: 601302). The association of rs1868402 with CSF ptau 181 levels showed the best fit in the dominant model, with minor alleles carriers showing significantly higher CSF ptau 181 levels (P = 5.90610 204 , Figure 1A and Figure  S1). All subsequent analyses for rs1868402 used the dominant model. Six other SNPs in PPP3R1, which are in high linkage disequilibrium (LD) with rs1868402 ( Figure S2, and Table S2), also showed association with CSF ptau 181 . Based on the linkage disequilibrium (LD) in PPP3R1, we selected rs1868402 and rs6546366 for replication. The remaining eleven SNPs that were significant after multiple test correction were also selected for replication (Table 3).

Association with CSF ptau 181 levels: Replication in an independent CSF series and combined analyses
In a replication series of 493 independent CSF samples from ADNI (ADNI-CSF) and University of Washington (UW) ( Tables 1  and 2) only the SNP located in calcineurin B, rs1868402, replicated and passed the FDR filter (P = 0.005 ptau 181 , Table 3, Figure S1). In this series rs1868402 also showed the best fit in the dominant model and minor allele carriers have higher CSF ptau 181 levels. In the replication CSF series, rs6546366 (PPP3R1), showed no association with CSF ptau 181 levels ( Table 3). The lack of association of rs6546366 with CSF ptau 181 levels is probably due to a lower level of LD with rs1868402 in the replication series (r 2 = 0.65) compared with the LD between these two SNPs in the WU-ADRC-CSF series (r 2 = 0.80). This result indicates that rs1868402, or another unknown variant in LD with rs1868402, is the variant that drives the association with CSF ptau 181 levels.
We also performed a combined analysis by combining the residuals for CSF ptau 181 after correcting for the covariates (see Materials and Methods). In this analysis we did not included site or platform as a covariate because there were no significant differences between datasets and/or platform for the ptau 181 residuals. Inclusion of site/platform as a covariate did not significantly change the p-value. In the combined data, rs1868402 showed the most significant association with CSF ptau 181 levels (combined analysis P~1:71|10 {05 ), with a pvalue that was significant after Bonferroni correction for the entire study (a; 0:05=384~1:30|10 {04 ). Minor allele carriers have a 2.4 fold increased risk of being in the highest quartile of the CSF ptau 181 distribution compared to the lowest quartile (odds ratio = 2.37, 95% confidence interval 1.59-3.54). None of the other SNPs were significant after Bonferroni correction (Table 3).

Association with CSF ptau 181 levels: Context-dependent effects
It has been demonstrated in our longitudinal data and that of others that the increase in CSF tau and ptau 181 levels seen in mild AD is preceded by decreases in CSF Ab 42 levels [14,21]. This likely reflects deposition of Ab in the brain [14]. Individuals with CSF Ab 42 levels less than 500 pg/ml in the WU-ADRC-CSF, and less than 192 pg/ml in the ADNI-CSF series, have evidence of Ab deposition in the brain, as detected by PET-PIB [14,21]. We used these CSF Ab 42 thresholds to stratify the WU-ADRC-CSF and ADNI-CSF samples into individuals with low or high CSF Ab 42 levels (with and without likely Ab deposition in the brain). The difference in CSF Ab 42 threshold levels between the WU-ADRC-CSF and ADNI-CSF series is due to different antibodies and procedures used to measure the CSF levels (see Materials and Methods). The data necessary to examine the correlation between the CSF Ab 42 levels and PET-PIB signal is not available in the UW CSF series and therefore this dataset could not be included in this analysis. In these analyses we calculated the p-value and the Odds ratios for rs1868402 with CSF ptau 181 levels by comparing the frequency of this SNP in the lowest versus the highest quartile of the CSF ptau 181 levels after correcting for the covariates. When we stratified the WU-ADRC-CSF and ADNI-CSF series by CSF Ab 42 levels, we observed very significant association for rs1868402 with CSF ptau 181 in the low Ab 42 stratum (combined analysis P~1:13|10 {04 , Odds Ratio 3.48; 95% confidence interval 1.8-6.7) and a nominally significant association in the high Ab 42 stratum (combined analysis P = 0.023; Odds Ratio 2.54; 95% confidence interval 1.0-6.75, Table 4). . Both the high and low Ab 42 level strata have sufficient power to detect the association between rs1868402 with CSF ptau 181 levels (high Ab 42 0.985; low Ab 42 = 0.975; a = 0.05). The nominally significant p-value in the high Ab stratum may indicate a moderate effect on CSF ptau 181 levels in healthy individuals, but it is clear that when AD pathology is present the effect of this SNP is more marked.
Implication of SNPs associated with CSF ptau 181 levels in AD: Association with rate of progression of AD but not risk for AD or age at onset The premise of this endophenotype-based approach is that a SNP, such as rs1868402 that shows strong, replicable association with an important AD biomarker should also modulate risk, onset and/or progression of AD. We tested whether rs1868402 influences risk for AD, age at onset and disease progression. We found no association between rs1868402 and risk for AD (P = 0.10, Table 5) or age at onset (P = 0.19 Figure 2) in 1106 cases and 1216 controls of European descent.
To examine disease progression we used two longitudinal datasets: 109 subjects from WU-ADRC-CSF (399 observations) and 150 subjects from ADNI-CSF (620 observations). Association with rate of progression was evaluated by comparing the change in sum of boxes of the CDR (SB-CDR) per year (slope) by genotype including age at the first visit, gender, APOE genotype and initial CDR as covariates. CSF ptau 181 and Ab 42 levels were also included in the model to correct for the potential association between these

Author Summary
Alzheimer's disease (AD) is the most common neurodegenerative disease affecting more than 4.5 million people in the US. Genetic studies of AD have previously identified pathogenic mutations in three genes (APP, PSEN1 and PSEN2) and polymorphisms in APOE as risk factors. These findings have led to a better understanding of the underlying disease mechanisms. However, half of all AD cases have no known genetic risk factors for disease. Most studies are designed to identify variants associated with risk or age at onset, but rarely cover other important facets of AD, such as disease progression or duration. In this study we have used an established AD biomarker (cerebrospinal fluid tau phosphorylated at threonine 181, ptau 181 ) to find genetic variants that influence levels of ptau 181 in the cerebrospinal fluid. This novel and powerful approach has allowed us to identify a genetic factor located in the regulatory subunit of the calcineurin that is also strongly associated with rate of progression of AD. This study is important because it defines a strategy to find novel genetic factors influencing different facets of AD pathobiology including risk, onset and progression. phenotypes with progression [33]. Because the association of rs1868402 with CSF ptau 181 was mainly in individuals with low CSF Ab 42 levels, association with progression was analyzed in individuals with low CSF Ab 42 levels (less than 500 pg/ml in the WU-ADRC-CSF, and less than 192 pg/ml in the ADNI-CSF series), and an initial CDR of 0 or 0.5. Individuals with CDR 0 and 0.5 were selected to maximize the amount of information on progression per individual and to avoid possible ceiling effects from individuals who began the study with advanced levels of dementia. Carriers of the rs1868402 allele associated with higher CSF ptau 181 levels showed an increase of 0.58 SB-CDR per year, which is six-fold faster than the rate seen in individuals homozygous for the allele associated with low CSF ptau 181 levels (P = 0.0026; Table 6, Figure 3A and Figure S3), and almost two times faster than the average change for the entire series (SB-CDR per year for the entire series 0.31). The association of rs1868402 with progression replicated in the ADNI-CSF series (P = 0.014) with a P combined = 1.96610 205 . In addition, we also used the ADNI samples with no CSF data (ADNI-CC, Table 6) to replicate the association with rate of progression in an independent sample. In this dataset rs1868402 also showed a significant association with rate of progression (P = 0.018, Table 6) and in the same direction as in the previous analyses.
We tested whether rs1868402 interacts with rs3785883, a SNP located in MAPT, which is also associated with CSF ptau 181 levels [27]. Rs3785883 also showed a p-value for association with CSF ptau 181 levels of 0.008 in the combined series of this study.We found significant epistasis between these SNPs. Individuals carrying alleles associated with higher CSF ptau 181 for rs1868402 and rs3785883 showed an increase of 1.02 SB-CDR per year on average, whereas those carrying the alleles associated with lower CSF ptau 181 at each SNP showed essentially no change in the SB-CDR per year (combined analyses P~1:43|10 {05 ; Figure 3B; Table 6). The interaction between rs1868402 and rs3785883 also replicated in the ADNI-CSF and ADNI-CC series ( Table 6).

Gene expression
Since rs1868402 is located in a region that encodes a tau phosphatase, we tested next whether the SNP is associated with PPP3R1 mRNA levels and tau pathology in brain. We extracted total RNA from the parietal lobe of 82 AD cases and 39 nondemented elderly individuals. The allele of rs1868402, associated with higher CSF ptau 181 levels, showed significantly lower PPP3R1 mRNA levels (P = 0.010; Figure 1B), and higher tangle pathology as measured by Braak stage (P = 0.005; Figure 1C) in brain samples with Ab pathology, but not in neuropathologically normal samples. We also found that there was a correlation between PPP3R1 mRNA levels and Braak stage in these samples (P = 0.018; Figure 1D). Rs12713636, in LD with rs1868402 (D9 = 1, R 2 = 0.75) also shows association with PPP3R1 mRNA levels and in the same direction in the publicly available GEO GSE8919 [34] dataset (P = 0.015).

Discussion
In the present study we have used a novel and powerful endophenotype-based approach to identify a novel genetic factor implicated in AD. Most genetic studies in AD have focused on identifying genetic factors that modulate risk or age at onset of  Table 3. SNPs associated with CSF ptau 181 levels in the initial series, the replication series and the combined dataset.   disease [1][2]. The genetic factors influence other important facets of AD, such as rate of progression or disease duration remain poorly understood. Here we report that the minor allele of rs1868402 shows significant, replicable association with higher CSF ptau 181 levels and faster rate of progression of AD. We failed to detect evidence for association with risk for disease or age at onset. This is consistent with the known pathobiology of AD, in which Ab aggregation and deposition is an early preclinical event, followed by increased CSF tau and ptau 181 levels and tau pathology during the clinical phase of the disease. Under this model, factors that affect tau phosphorylation and aggregation may be expected to modify disease progression but not risk for disease, while genes that influence Ab aggregation, such as APOE4 [35], would be expected to influence CSF Ab 42 levels and risk/age at onset of disease but not CSF tau/ptau 181 levels or rate of progression of disease. Table 7 illustrates that this is the case in our data. Genetic analyses of CSF Ab 42 and tau variation therefore allow identification of genetic factors that influence different components of the disease process.
Our results suggest that rs1868402, or another variant in LD with it, may reduce calcineurin expression/activity leading to an increase in tau phosphorylation increasing tau pathology and neurodegeneration in individuals with Ab deposition. The resulting increase in tau-related pathology would then increase the rate of progression of AD. Several studies provide support for a role of calcineurin in AD pathogenesis. Inhibition of calcineurin in mouse brains by cyclosporin A or FK506 or rat brain by antisense oligonucleotides led to enhanced tau phosphorylation [36][37][38]. In mice the increase in tau phosphorylation was accompanied by impaired spatial memory, a characteristic feature of AD [36]. Finally, in AD patients, calcineurin activity is decreased and correlates with neuropathologic changes [39].
Our analyses suggest that genetic variants associated with CSF Ab 42 levels also influence risk and age at onset (e.g. APOE) but variants associated with CSF ptau 181 levels have a greater impact on rate of progression (Table 7). Genome-wide association studies of CSF tau/ptau 181 levels should identify novel genetic variants which will likely influence rate of progression of AD. Variants that influence disease progression may have significant clinical benefit. For example, these variants have the potential to predict more accurately the time from diagnosis to functional impairment that may require nursing home placement. Stratification of samples by such SNPs will enable cheaper and more efficient clinical trials by selecting individuals expected to have faster rates of progression. By targeting different facets of AD biology this approach can identify a broader range of potential therapeutic targets than a conventional case-control design. Drugs that inhibit or decrease tau phosphorylation would be expected to decrease cognitive decline in individuals with very mild dementia or delay the appearance of memory problems in elderly individuals with low CSF Ab 42 levels. Finally, we believe that this approach is applicable to other common neurological and psychiatric disorders, where biomarkers of disease have been identified, and underlines the value and importance of finding such markers in other diseases.

Subjects and endophenotypes
The cerebrospinal fluid discovery series includes 353 individuals enrolled in longitudinal studies at the WU-ADRC. CSF collection and Ab 42 , tau and ptau 181 measurements were performed as described previously [14]. Table 1 shows the demographic data for the CSF series and Table 2 shows a description of the CSF biomarker in each dataset. The CSF replication series consists of 236 individuals from the ADNI dataset and 257 individuals from the University of Washington (UW, Seattle). All CSF samples were from individuals of European descent. Written consent was obtained from all participants. While there are differences in the absolute levels of the biomarker measurements between the two studies that likely reflect differences in the methods used for quantification (regular ELISA vs Luminex), ascertainment, and/or

SNP selection and genotyping
Based on bibliographic data, we selected 384 SNPs in the most relevant tau kinases, phosphatases, and in other genes implicated in other posttranslational modifications of tau, or tau degradation [45][46][47] (Table S1). Tagging SNPs (r 2 .0.8), based on CEU-HapMap data, were selected for each of these genes. We used Pupasuite software [48] to select potentially functional variants in the selected genes and flanking regions. SNPs were genotyped Rs1868402 is associated with progression rate and shows epistasis with rs3785883, which is also associated with CSF tau levels [27]. Rate of progression were analyzed by the change of the Clinical Dementia Rating sum of boxes (CDR-SB) score per year in individuals with low CSF Ab 42 levels (ADRC,500pg/ml; ADNI,192pg/ml) and in samples with no CSF data. Association of SNPs with progression was calculated with mixed linear models (proc mixed) after controlling for age, sex, APOE, CDR and/or CSF ptau 181 and Ab 42 . doi:10.1371/journal.pgen.1001101.t006 using the Illumina Golden Gate, Sequenom and/or Taqman genotyping technologies. Only SNPs with a genotyping call rate higher than 95% and in Hardy-Weinberg equilibrium were used in the analyses.

Gene expression
Expression studies were carried out using cDNA obtained from the parietal lobe of 82 AD cases and 39 non-demented individuals (CDR = 0) obtained through the WU-ADRC Neuropathology Core (Brain samples; Table 1). AD changes were measured using Braak and Braak staging [43]. All AD cases had a Braak and Braak score of 5 or 6. Among the non-demented individuals 24 brains had a Braak and Braak score ranging from 1-4 indicating the presence of some tangle pathology.
Total RNA was extracted from the parietal lobe of 82 AD cases and 39 non-demented individuals, using the RNeasy mini kit (Qiagen) following the manufacturer's protocol. cDNAs were prepared from the total RNA, using the High-Capacity cDNA Archive kit (ABI). Gene expression was analyzed by real-time PCR, using an ABI-7500 real-time PCR system. Real-time PCR assays were used to quantify PPP3R1 cDNA levels. Taqman assays for GAPDH (sequences available on request) PPP3R1 (ABI: C_ 12044272_10) and cyclophilin A (ABI: 4326316E) were used to quantify the gene expression levels. Each real-time PCR run included within-plate duplicates and each experiment was performed, at least twice for each sample. Real-time data were analyzed using the comparative Ct method. The Ct values of each sample were normalized with the Ct value for the housekeeping genes, GADPH and cyclophilin, and were corrected for the PCR efficiency of each assay [44], although the efficiency of all reactions was close to 100%. Only samples with a standard error of ,0.15% were analyzed.

Statistical analyses
CSF ptau 181 values were log-log transformed to approximate a normal distribution. Analysis of the covariance (ANCOVA) was used to test for association between genotypes and CSF ptau 181 levels. In order to identify the covariates that affect CSF ptau 181 levels, we performed a stepwise discriminant analysis including CDR, age, gender and APOE genotype. CDR, age, and APOE genotype were identified as significant covariates in the WU-ADRC-CSF series and, CDR and APOE genotype in the replication series (this series has a narrower age range than the WU-ADRC-CSF series). These covariates were included in the respective ANCOVA. Each SNP was tested using an additive model with minor allele homozygotes coded as 0, heterozygotes coded as 1, and major allele homozygotes coded as 2. When the additive model was significant after multiple test correction, dominant and recessive models were tested to determine whether the y were a better fit. Because the CSF ptau 181 levels in the WU-ADRC-CSF, ADNI-CSF and UW samples were measured using different platforms (Innotest plate ELISA vs AlzBia3 bead-based ELISA, respectively) we were not able to combine the raw data, rather we combined the residual values of the CSF ptau 181 obtained after correcting for the covariates. No significant differences in the residuals from the different series were found, indicating that the differences in the CSF levels due to the different platforms were corrected by using the residuals.
Multiple test correction: Initial tests for association of SNPs with CSF ptau 181 levels were evaluated using a False Discovery Rate (FDR) filter of 0.1 to correct for multiple testing [49]. In the initial screening only p-values more significant than 8:1|10 {3 passed FDR filter of 0.1. In the replication series p-values more significant than 5:0|10 {3 passed FDR filter. To reduce the probability of false positives we also used the more stringent Bonferroni correction to adjust the alpha level in the analysis of association with CSF ptau 181 levels in the combined samples. In this case, the threshold for Bonferroni correction for the combined sample is 1:3|10 {04 (a; 0:05=384~1:3|10 {04 ). No multiple test correction was applied for association with risk for disease, age at onset or progression because only one or two SNPs with specific hypotheses were tested for association. To calculate the impact of rs1868402 on the CSF ptau 181 levels, we calculated the Odds Ratio (OR) of this SNP by comparing its frequency in the highest vs lowest quartile of the residuals for CSF ptau 181 levels.
Allelic association with risk for AD was tested using logistic regression including APOE, gender, age and series as covariates. Association with AAO was carried out using the Kaplan-Meier method and tested for significant differences, using a log-rank test. Association with rate of disease progression was evaluated as described previously [33]. Briefly, progression of disease was measured by the change in sum of boxes on the CDR (clinical dementia rate; SB-CDR) per year. CDR is a global measurement of the severity of symptoms of dementia. CDR evaluates cognitive and functional performance in six areas (memory, orientation, judgment and problem solving, community affairs, home and hobbies and, personal care), each of these areas has a possible score of 0, 0.5, 1, 2 or 3. The sum of boxes can vary between 0 and 18. Higher scores indicate more significant memory problems and correlate with neurodegeneration [50]. The change in SB-CDR per year fitted a linear model in both series and therefore we used a mixed linear model (PROC MIXED; SAS Institute Inc) to determine whether there is a relationship between the slope of the SB-CDR score and time as a function of genotype after controlling for initial age, gender, APOE, initial CDR, CSF ptau 181 and Ab 42 levels. Because the association between rs1868402 and CSF ptau 181 was driven by individuals with low CSF Ab 42 levels, association with progression in the CSF datasets was analyzed only in individuals with low CSF Ab 42 levels (less than 500 pg/ml in the WU-ADRC-CSF series [14] or 192 pg/ml in the ADNI-CSF series [21]). We analyzed whether the combination of genotypes of rs1868402 and rs3785883 predicts the rate of progression better than either of these SNPs alone. To do that we included an interaction term rs1868402*rs3785883 in the ''proc mixed'' SAS program, that combine the genotypes for the two SNPs. We also combined the data from the WU-ADRC-CSF and ADNI-CSF to increase the statistical power. Association between cDNA levels, tau pathology (Braak tangle stage) and genotypes were carried out using ANCOVA. Stepwise discriminant analysis was used to determine the significant covariates (age, gender, postmortem interval, APOE genotype, and CDR). One-tailed P-values were calculated, because a priori predictions were made based on the associations with CSF ptau 181 levels. We also used the GEO dataset GSE8919 [51] to analyze the association between rs12713636, in LD with rs1868402 (D9 = 1, R 2 = 0.75) and PPP3R1 gene expression. In this dataset there are genotype and expression data from 486 Late onset Alzheimer Diseases cases and 279 neuropathologically confirmed controls. We only extracted the normalized PPP3R1 mRNA levels and the genotype data for rs12713636. Genotypes for rs12713636 were used because rs1868402 was not included in this dataset.

ADNI material and methods
Data used in the preparation of this article were obtained from the ADNI database (www.loni.ucla.edu/ADNI). The ADNI was launched in 2003 by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, the Food and Drug Administration, private pharmaceutical companies and nonprofit organizations, as a $60 million, 5-year public-private partnership. The Principal Investigator of this initiative is Michael W. Weiner, M.D. ADNI is the result of efforts of many coinvestigators from a broad range of academic institutions and private corporations, and subjects have been recruited from over 50 sites across the U.S. and Canada. The initial goal of ADNI was to recruit 800 adults, ages 55 to 90, to participate in the researchapproximately 200 cognitively normal older individuals to be followed for 3 years, 400 people with MCI to be followed for 3 years, and 200 people with early AD to be followed for 2 years.'' For up-to-date information see www.adni-info.org. Figure S1 Minor allele carriers of rs1868402 present significantly higher CSF ptau 181 levels. The mean and the standard error of the mean (SEM) for the raw and residuals CSF ptau 181 levels for the WU-ADRC-CSF, ADNI-CSF and UW series is shown. A. Raw CSF ptau 181 levels for the WU-ADRC-CSF series by rs1868402 genotype. CC+CT Figure S2 Linkage disequilibrium among PPP3R1 SNPs significantly associated with CSF tau levels in the ADRC series. Color represents D9 = 1 and numbers correspond to r2. Found at: doi:10.1371/journal.pgen.1001101.s002 (0.16 MB DOC) Figure S3 Genetic variants associated with CSF ptau 181 levels are also associated with rate of progression. Average progression rate by genotype with the 95% confidence interval. Solid lines represent the average progression rate. Dotted lines represent the 95% confidence interval. The lines for the different genotypes are color code. A. Minor allele carriers of rs1868402, are associated with higher CSF ptau 181 levels, and show a 6-fold faster progression than homozygotes for the major allele (CDR-SB/ year: 0.58 vs. 0.09; p = 0.0026) in individuals from the WU-ADRC-CSF with low CSF Ab 42 levels (,500pg/ml). B. rs3785883 genotypes do not have significantly different progression rates P = 0.057. The genotype frequency distribution for rs3785883 with disease progression is most likely not significant due to the low statistical power. AA carriers show a CDR-SB of 1.01, AG of 0.47 and GG 0.26 (p = 0.057).