There are many active research projects accessing and applying shared ADNI data. Use the search above to find specific research focuses on the active ADNI investigations. This information is requested annually as a requirement for data access.
| Principal Investigator | |
| Principal Investigator's Name: | Richard Pither |
| Institution: | Cytox Ltd |
| Department: | Head Office |
| Country: | |
| Proposed Analysis: | Cytox summary proposal for ADNI: SNP-profiling and polygenic risk score analysis as an approach to risk stratification for amyloid status by PET Overall Goal of Study and Hypotheses to be Tested: Cytox is developing two new RUO (Research-Use-Only) genotyping array and software products for early detection of Alzheimer’s Disease (AD). The primary aim of this study will be to test the hypothesis that Polygenic Risk Score (PRS) algorithms can be used to stratify subjects within the ADNI cohort for increased likelihood of AD and amyloid-positivity. In particular, Cytox has developed two prototype PRS algorithms using different modelling approaches which have been used to identify AD risk and amyloid-positive subjects with an accuracy of greater than 80%. This proposed study will aim to further validate and develop these algorithms in a proportion of the ADNI subjects, for whom whole genome sequencing (WGS) data is already available. Background: Cytox is currently developing two important new RUO (Research-Use-Only) products for early detection of Alzheimer’s Disease with our co-development and commercial partners, Affymetrix (now part of Thermo Fisher Scientific). Cytox is conducting this work in partnership with Professor John Hardy, an expert in AD genetics from University College London (UCL), Dr Zsuzsa Nagy, University of Birmingham. The overall goal of Cytox and our partners is to develop a gene array and associated PRS algorithms for the prediction of Alzheimer’s disease risk in individual patients. In doing so, we are working towards the near-term aim of providing a stratification tool for the pharma industry and academic-led registry studies, to reduce the screening failure rate in finding amyloid-positive subjects for recruitment into clinical trials, with huge savings in cost and time. The Cytox variaTECT™ Single Nucleotide Polymorphism (SNP) panel has been derived from a combination of published SNPs, largely identified from GWAS studies and from our own in-house efforts. In the latter case, we have pursed whole exome sequencing (WES) to identify risk-associated and protective SNPs in biomarker-confirmed cases and controls. The variaTECT array is being developed to offer a simple blood-based research test to help users in the pharmaceutical industry and academia reduce the screening failure rate in finding amyloid positive subjects for recruitment into clinical trials, with savings in cost and time, and provides a comprehensive SNP panel to assist research and novel drug discovery studies. The entire variaTECT SNP panel comprises ~130,000 SNPs and we are in the process of building a SNP array in partnership with Affymetrix. The variaTECT array is currently the most wide-ranging research panel available for the detection of AD informative SNPs, configured on the Affymetrix Axiom™ plates and processed on an Affymetrix GeneTitan® scanner. The variaTECT array can be used together with our new SNPfitR software, in which our two prototype PRS algorithms are embedded, to enrich amyloid-positive cohorts and so reduce screening failure rates. Cytox is conducting this work in partnership with Professor John Hardy, a well-known expert in AD genetics from University College London (UCL), Dr Zsuzsa Nagy, University of Birmingham, and other partners. Preliminary Results: The current PRS algorithms were tested in cohorts that included AD patients along with MCI and cognitively-normal groups (AIBL, INSIGHT, and other clinical cohort samples) with PET amyloid imaging results. We have two different predictive algorithms developed by our academic partners:- PRS Model 1: Stratification tool for the pre-screening of clinical trial subjects by predicting the risk of AD and beta-amyloid accumulation in the brain (UCL) The early identification of subjects at high risk of AD is important for early diagnosis and successful treatment of AD, as recently suggested by recent reports from Pharma companies pursuing new treatments for AD, including both Biogen and Eli Lilly7. It is well documented that early symptomatic (MCI) or elderly pre-symptomatic individuals who are amyloid-positive are at increased risk of future cognitive decline and AD. It is thought that the lack of large effect variants, outside of the APOE locus, in sporadic AD can be attributed to the polygenic nature of AD. One study looking at genetic data from the largest AD GWAS to date was able to predict with ca. 76% the incidence of AD in an independent test set using a polygenic risk score approach8. We therefore have used the same approach to detect both AD status as well as amyloid positivity in a cohort of well characterized samples in order to identify a Polygenic Risk Score algorithm comprising of risk or protective variants for amyloid plaque formations. The approach necessitates two independent data sets. The first of these, the Training set, will be used to perform various regressions models, in order to identify association and effects for each SNP with the trait of interest. The second data set, known as the Test set will be used to test the ability of SNPs identified through the Training set, to predict the phenotype of individuals in the set. UCL, in collaboration with Cytox, has thus far obtained approximately 900 Amyloid status known samples, which have been used to derive a model, consisting of 21 SNPs or less, used for prediction of amyloid status in several different Test sets. Preliminary results of the analysis (Table 1) suggests that with as few as 12 SNPs it is possible to predict AD status, using the whole exome sequenced AIBL samples, with an accuracy of 67.5%, positive predictive value (PPV) and negative predictive value (NPV) of 62%, without any demographic data being available (Figure 1 and Table 1a). Upon addition of APOE to the algorithm, the said values can increase to 87.9, 79.8 and 81% respectively. Figure 1: Performance of the model upon ca. 150 AIBL samples Table 1a: Performance statistics of the model on discerning AD status in 150 of the AIBL cohort subjects. N.B The elevated APOE4 status in this cohort can be seen by the large predictability yielded with using APOE SNPs alone. Effect SE p R2 NSNPs Sensitivity Specificity AUC 95% CI PPV NPV model (IGAP) 0.694 0.235 0.003 0.128 12 62.0 62 67.5 57.0-78.1 62 62 model + APOE (IGAP) 1.99 0.188 0 0.536 14 80.3 80.4 87.9 84.3-91.6 79.8 81 APOE alone (IGAP) 1.798 0.172 0 0.495 2 78.1 84.8 84.4 80.5-88.2 77.5 85.2 Using the same model that was used to derive the above algorithm, we then tested the ability to predict Amyloid positivity in a cohort of 300 subjects with various cognitive states, including control, MCI, and AD. The results indicate that the present model is capable of discerning Amyloid positivity with an accuracy of ca. 79%, and PPV and NPV values of 80 and 63.6% respectively (Table 1b). Table 1b: Performance statistics of the model on amyloid positivity in 300 subjects. Effect SE p R2 NSNPs Sensitivity Specificity AUC 95 % CI PPV NPV APOE_genotyped(a&s adjusted) 1.032 0.148 0 0.254 2 72.6 71.7 74.6 68.3-80.8 79.7 63.0 model + APOE genotyped(a&s adjusted) 1.078 0.157 0 0.258 22 72.6 72.2 76.2 70.3-82.1 79.9 63.3 model + APOE genotyped(a&s adjusted)+ age and sex 1.128 0.162 0 0.309 22 72.6 72.7 78.9 73.5-84.4 80.1 63.6 age+sex 0.168 0.249 0.498 0.052 0 56.9 59.1 61.7 55.0-68.4 71.7 42.9 The above model was found to perform consistently to the same standards depending on the cohort analysed. That is the performance of the model alone, is elevated in the absence of APOE4 positive samples, and in the presence of excess APOE4 subjects. Table 1c: Performance statistics of the model on amyloid positivity in ca. 200 APOE4 negative samples. Given the promising results of this array in detection of Amyloid status, in particular in the absence of conventionally used methods such as APOE4 status, we propose a twofold use of the ADNI data. The ca. 800 whole genome sequenced samples will be used as: i) blinded as a test set to detect Amyloid status using the current algorithm from the array, and the results provided to ADNI for unblinding. ii) Upon unblinding of the Amyloid status of the WGSed samples, the cohort will be used as a training set which will allow development of a new/more refined algorithm, to be tested on an imputed data set (currently genotyped on the Cytox array). The results of this will thus be shared with the ADNI team with a direct comparison with previous algorithms also provided. In each step the training set will undergo various regression models for the ascertainment of the algorithm. PRS Model 2: Prediction of Alzheimer’s disease risk in individual patients (UoB) There is mounting evidence that mTOR (mammalian target of Rapamycin) mediated signalling is involved in the pathogenesis of Alzheimer’s disease . Our working hypothesis is that the activation of the mTOR regulated pathways in the ageing brain on the background of diverse combinations of genetic variants on the downstream molecular pathways leads to a catastrophic imbalance of gene expression and activation that drives neurodegeneration (figure 2). Figure 2: mTOR signalling pathway Using the accumulated knowledge on the molecular networks affected by mTOR (Rapamycin sensitive molecular pathways) that are consistently affected in Alzheimer’s disease (expression studies in the brain and genetic studies) we have constructed an algorithm that scores the severity of genetic variation burden on these pathways. Our preliminary studies in the AIBL cohort (CASES: 156; CONTROLS: 173) indicate that this burden score on the different Rapamycin-sensitive pathways is able to discriminate between AD patients and controls with 83% accuracy (Figure 3). Figure 3: mTOR PRS algorithm is able discriminate between AD patients and controls Methodology: In order to develop our predictive algorithm we need further well characterised case-control cohorts to: A) increase the control (reference cohort) and B) validate the efficacy of the algorithms in predicting the risk of AD. This part of the project would require the NGS data on all subjects (814 subjects, excluding those in the temporary sequester). We acknowledge that we will receive blinded samples. The blinded VCF files will be uploaded to the Ingenuity Variant Analysis software to allow the filtering of the dataset for the variants (39,000) included in the genoTOR algorithms. The filtered data will be further analysed using our own software algorithms to calculate the genetic burden scores for the selected molecular pathways. These pathway scores will then be released to the ADNI consortium to trigger the ‘breaking of the blind’. Following the un-blinding of the data, we will assess the diagnostic accuracy of our predictions using ROC curve analysis. Data from the oldest controls (>age 80) will be used to enlarge our existing reference database that will ultimately allow us to predict AD risk in individual patients. Sample and Data Requirements: Access to existing whole genome DNA sequencing data from ADNI database (as described above). The clinical data required to aid our analysis are: demographic variables of the patients, cognitive assessment results (MMSE, geriatric depression scale, all on baseline and follow up), Clinical diagnosis, and ApoE genotype (where available). We would also like to have (if available) plasma (or serum) homocysteine measurements (did not find reference to existing biochemistry on the website). Clinical and demographic data, anonymised, from subjects (where available): • Date of birth (year/month sufficient) • Date the patient/control was last seen • Sex of subject • Amyloid status by PET (1.5 SUVR cut-off using 11C-PIB tracer) • ApoE genotype (if available) • Cognitive score last seen (MMSE or other cognitive scores) at baseline and at follow-up (with dates) • Age of onset (or age when last seen normal) • Family history of dementia or other major illness (diabetes, high BP, dementia, cancer) • Personal History of smoking, high BP, diabetes, cancer • Educational attainment • Plasma homocysteine, cholesterol • Family links with other participants in the project • Medications (chronic use only) Potential impact and future use in the field The early treatment of Alzheimer’s Disease will be required if we are to achieve optimal effects on delaying progression and improving patient outcomes with disease-modifying drugs. In order to achieve this, we will need better tools in order to positively identify subjects at the earliest symptomatic or even pre-symptomatic stages of the disease process. Alzheimer’s disease is a multifactorial polygenic complex disorder and to date we have no algorithm to predict the genetic component of AD risk in individual subjects. We believe that our PRS approach may provide a step forward in understanding and describing the genetic risk in this disease. Genetic analysis through SNP (single nucleotide polymorphism) profiling is a promising route to this important clinical dissection. As such it offers potential in helping to diagnose those within the early MCI population who have early AD and also identifying the variables which determine their rate of conversion to full blown disease. In turn, this would be expected to facilitate AD clinical trials, and ultimately selection of patients for treatment, by reducing both misdiagnosis rates and enabling better prediction of expected disease progression rates. Additionally we may gain further understanding of how these genetic factors interact with environmental risk factors and ageing to bring about Alzheimer’s disease. Estimated Timelines From the point of access to whole genome DNA sequencing data from ADNI database, we anticipate that it will take a minimum of two months to analyse the data and perform the required stringent quality control activities. Data analysis, modelling and further validation studies is likely to take a further 2-3 months. References: 1. The Triangle of Death in Alzheimer's Disease Brain: The Aberrant Cross-Talk Among Energy Metabolism, Mammalian Target of Rapamycin Signaling, and Protein Homeostasis Revealed by Redox Proteomics. Di Domenico F, Barone E, Perluigi M, Butterfield DA. Antioxid Redox Signal. 2016 Oct 20. 2. Targeting neuronal MAPK14/p38α activity to modulate autophagy in the Alzheimer disease brain. Alam J, Scheper W. Autophagy. 2016 Oct 7:0. 3. The protein-interaction network with functional roles in tumorigenesis, neurodegeneration, and aging. Nahálková J. Mol Cell Biochem. 2016 Oct 3. Review. 4. mTOR and neuronal cell cycle reentry: How impaired brain insulin signaling promotes Alzheimer's disease. Norambuena A, Wallrabe H, McMahon L, Silva A, Swanson E, Khan SS, Baerthlein D, Kodis E, Oddo S, Mandell JW, Bloom GS. Alzheimers Dement. 2016 Sep 29. 5. Dysfunction of the mTOR pathway is a risk factor for Alzheimer's disease. Yates SC, Zafar A, Hubbard P, Nagy S, Durant S, Bicknell R, Wilcock G, Christie S, Esiri MM, Smith AD, Nagy Z. Acta Neuropathol Commun. 2013 May 8;1:3. 6. Cell cycle kinesis in lymphocytes in the diagnosis of Alzheimer's disease. Nagy Z, Combrinck M, Budge M, McShane R. Neurosci Lett. 2002 Jan 11;317(2):81-4. 7. AAIC 2015 book of abstracts 8. Common polygenic variation enhances risk prediction for Alzheimer’s Disease. Escott-Price et al, 2015 Dec;138(Pt 12):3673-84 |
| Additional Investigators | |
| Investigator's Name: | Zsuzsa Nagy |
| Proposed Analysis: | PRS Model 2: Prediction of Alzheimer’s disease risk in individual patients (UoB) There is mounting evidence that mTOR (mammalian target of Rapamycin) mediated signalling is involved in the pathogenesis of Alzheimer’s disease . Our working hypothesis is that the activation of the mTOR regulated pathways in the ageing brain on the background of diverse combinations of genetic variants on the downstream molecular pathways leads to a catastrophic imbalance of gene expression and activation that drives neurodegeneration. Using the accumulated knowledge on the molecular networks affected by mTOR (Rapamycin sensitive molecular pathways) that are consistently affected in Alzheimer’s disease (expression studies in the brain and genetic studies) we have constructed an algorithm that scores the severity of genetic variation burden on these pathways. Our preliminary studies in the AIBL cohort (CASES: 156; CONTROLS: 173) indicate that this burden score on the different Rapamycin-sensitive pathways is able to discriminate between AD patients and controls with 83% accuracy. Methodology In order to develop our predictive algorithm we need further well characterised case-control cohorts to: A) increase the control (reference cohort) and B) validate the efficacy of the algorithms in predicting the risk of AD. This part of the project would require the NGS data on all subjects (814 subjects, excluding those in the temporary sequester). We acknowledge that we will receive blinded samples. The blinded VCF files will be uploaded to the Ingenuity Variant Analysis software to allow the filtering of the dataset for the variants (39,000) included in the genoTOR algorithms. The filtered data will be further analysed using our own software algorithms to calculate the genetic burden scores for the selected molecular pathways. These pathway scores will then be released to the ADNI consortium to trigger the ‘breaking of the blind’. Following the un-blinding of the data, we will assess the diagnostic accuracy of our predictions using ROC curve analysis. Data from the oldest controls (>age 80) will be used to enlarge our existing reference database that will ultimately allow us to predict AD risk in individual patients. |
