15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Received: 8 July 2020 Revised: 16 January 2021 Accepted: 13 February 2021 DOI: 10.1002/alz.12330 RESEARCH ARTICLE TMEM106B and CPOX are genetic determinants of cerebrospinal fluid Alzheimer’s disease biomarker levels Shengjun Hong1 Valerija Dobricic1 Olena Ohlei1 Isabelle Bos2,3 Stephanie J. B. Vos2,3 Dmitry Prokopenko4 Betty M. Tijms3 Ulf Andreasson5,6 Kaj Blennow5,6 Rik Vandenberghe7,8 Silvy Gabel7 Philip Scheltens3 Charlotte E. Teunissen9 Sebastiaan Engelborghs10,11 Giovanni Frisoni12,13 Olivier Blin14 Jill C. Richardson15 Regis Bordet16 Alzheimer’s Disease Neuroimaging Initiative (ADNI)* Alberto Lleó17 Daniel Alcolea17 Julius Popp18,19 Christopher Clark18 Gwendoline Peyratout19 Pablo Martinez-Lage20 Mikel Tainta20 Richard J. B. Dobson21,22,23,24,25 Cristina Legido-Quigley26 Kristel Sleegers27,28 Christine Van Broeckhoven27,28 Rudolph E. Tanzi4 Mara ten Kate29,30 Michael Wittig31 Andre Franke31 Christina M. Lill1,32 Frederik Barkhof33 Simon Lovestone34 Johannes Streffer35,36 Henrik Zetterberg5,6,37,38 Pieter Jelle Visser2,3,39 Lars Bertram1,40 1 Lübeck Interdisciplinary Platform for Genome Analytics (LIGA), University of Lübeck, Lübeck, Germany 2 Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Alzheimer Centrum Limburg, Maastricht University, Maastricht, the Netherlands 3 Alzheimer Center Amsterdam, Department of Neurology, Amsterdam Neuroscience, Vrije Universiteit Amsterdam, Amsterdam UMC, the Netherlands 4 Genetics and Aging Unit and McCance Center for Brain Health, Department of Neurology, Massachusetts General Hospital, Boston, Massachusetts, USA 5 Institute of Neuroscience and Physiology, Department of Psychiatry and Neurochemistry, The Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden 6 Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden 7 Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, Leuven, Belgium 8 Neurology Service, University Hospital Leuven, Leuven, Belgium 9 Neurochemistry Laboratory, Department of Clinical Chemistry, Amsterdam Neuroscience, Amsterdam University Medical Centers, Vrije Universiteit, Amsterdam, the Netherlands 10 Department of Biomedical Sciences, Institute Born-Bunge, University of Antwerp, Antwerp, Belgium 11 Department of Neurology and Center for Neurosciences, UZ Brussel and Vrije Universiteit Brussel (VUB), Brussels, Belgium 12 University of Geneva, Geneva, Switzerland 13 IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy 14 AIX Marseille University, INS, Ap-hm, Marseille, France 15 Neurosciences Therapeutic Area, GlaxoSmithKline R&D, Stevenage, UK 16 Inserm, CHU Lille, University of Lille, Lille, France 17 Memory Unit, Neurology Department. Hospital de Sant Pau, Barcelona and Centro de Investigación Biomédica en Red en enfermedades Neurodegenerativas (CIBERNED), Madrid, Spain 18 Centre for Gerontopsychiatric Medicine, Department of Geriatric Psychiatry, University Hospital of Psychiatry Zurich, Zürich, Switzerland This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2021 The Authors. Alzheimer’s & Dementia published by Wiley Periodicals LLC on behalf of Alzheimer’s Association 1628 wileyonlinelibrary.com/journal/alz Alzheimer’s Dement. 2021;17:1628–1640. 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License HONG ET AL . 1629 19 Old Age Psychiatry, Department of Psychiatry, University Hospital of Lausanne, Lausanne, Switzerland 20 Department of Neurology, Center for Research and Advanced Therapies, CITA-Alzheimer Foundation, San Sebastian, Spain 21 Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, UK 22 NIHR BioResource Centre Maudsley, NIHR Maudsley Biomedical Research Centre (BRC) at South London and Maudsley NHS Foundation Trust (SLaM) & Institute of Psychiatry, Psychology and Neuroscience (IoPPN), King’s College London, London, UK 23 Health Data Research UK London, University College London, London, UK 24 Institute of Health Informatics, University College London, London, UK 25 The National Institute for Health Research University College London Hospitals Biomedical Research Centre, University College London, London, UK 26 Steno Diabetes Center, Copenhagen, Denmark and Institute of Pharmaceutical Sciences, King’s College London, London, UK 27 Neurodegenerative Brain Diseases Group, Center for Molecular Neurology, VIB, Antwerp, Belgium 28 Department of Biomedical Sciences, University of Antwerp, Antwerp, Belgium 29 Alzheimer Center and Department of Neurology, Amsterdam University Medical Centers, Amsterdam Neuroscience, Amsterdam, the Netherlands 30 Department of Radiology and Nuclear Medicine, Amsterdam University Medical Centers, Amsterdam Neuroscience, Amsterdam, the Netherlands 31 Institute of Clinical Molecular Biology, Christian-Albrechts-University of Kiel, Kiel, Germany 32 Ageing Epidemiology Research Unit, School of Public Health, Imperial College, London, United Kingdom 33 Institutes of Neurology and Healthcare Engineering, University College London, London, UK 34 Department of Psychiatry, University of Oxford, Oxford, UK 35 Reference Center for Biological Markers of Dementia (BIODEM), Institute Born-Bunge, University of Antwerp, Antwerp, Belgium 36 Translational Medicine Neuroscience, UCB Biopharma SPRL, Braine l’Alleud, Belgium 37 Department of Neurodegenerative Disease, UCL Queen Square Institute of Neurology, Queen Square, London, UK 38 UK Dementia Research Institute at UCL, London, UK 39 Department of Neurobiology, Care Sciences and Society, Division of Neurogeriatrics, Karolinska Instutet, Stockholm, Sweden 40 Department of Psychology, University of Oslo, Oslo, Norway Correspondence Prof. Lars Bertram, Lübeck Interdisciplinary Abstract Platform for Genome Analytics, (LIGA), Uni- versity of Lübeck, Ratzeburger Allee 160, Haus Introduction: Neurofilament light (NfL), chitinase-3-like protein 1 (YKL-40), and neu- V50, 23562 Lübeck, Germany. rogranin (Ng) are biomarkers for Alzheimer’s disease (AD) to monitor axonal damage, E-mail: lars.bertram@uni-luebeck.de astroglial activation, and synaptic degeneration, respectively. Simon Lovestone: currently employed at Methods: We performed genome-wide association studies (GWAS) using DNA and Janssen-Cilag. * cerebrospinal fluid (CSF) samples from the EMIF-AD Multimodal Biomarker Discov- Alzheimer’s Disease Neuroimaging Initiative (ADNI): Data used in preparation of this article ery study for discovery, and the Alzheimer’s Disease Neuroimaging Initiative study for were obtained from the Alzheimer’s Disease validation analyses. GWAS were performed for all three CSF biomarkers using linear Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators regression models adjusting for relevant covariates. within the ADNI contributed to the design and Results: We identify novel genome-wide significant associations between DNA vari- implementation of ADNI and/or provided data but did not participate in analysis or writing ants in TMEM106B and CSF levels of NfL, and between CPOX and YKL-40. We con- of this report. A complete listing of ADNI firm previous work suggesting that YKL-40 levels are associated with DNA variants in investigators can be found at: http://adni.loni. usc.edu/wp-content/uploads/how_to_apply/ CHI3L1. ADNI_Acknowledgement_List.pdf Discussion: Our study provides important new insights into the genetic architecture Funding information underlying interindividual variation in three AD-related CSF biomarkers. In particular, European Union’s Seventh Framework Pro- our data shed light on the sequence of events regarding the initiation and progression gram, Grant/Award Number: EMIF-AD #115372; German Research Foundation, of neuropathological processes relevant in AD. Grant/Award Number: BE2287/7-1 KEYWORDS Alzheimer’s disease, biomarker, cerebrospinal fluid, chitinase-3-like protein 1, genome-wide asso- ciation study, neurofilament light, neurogranin 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1630 HONG ET AL . 1 BACKGROUND RESEARCH IN CONTEXT 1. Systematic Review: Despite >30 independent loci Elucidation of the genetic architecture underlying Alzheimer’s disease reported to show genome-wide significant associa- (AD) susceptibility has recently seen substantial progress using the tion with AD risk from GWAS, the genetic underpin- genome-wide association study (GWAS) approach. Based on results nings determining interindividual variation in levels from the two most recent and largest GWAS in the field,1,2 there are of molecular AD biomarkers are less well known. We now >30 independent loci showing genome-wide significant associa- searched PubMed, Google Scholar, and the GWAS tion with AD risk.3 In contrast, the genetic underpinnings determining catalog for GWAS articles on CSF neurofilament light interindividual variation in levels of molecular AD biomarkers are less chain (NfL), chitinase-3-like protein 1 (YKL-40), neu- well known. Apart from the two “core” AD proteins, that is amyloid rogranin (Ng) and their association with Alzheimer’s beta 42 (Aβ42) and tau, there are currently only a few GWAS shed- disease (AD) with no language restrictions from ding light on the genetic factors determining other blood or cere- database inception up to November 30, 2019, using brospinal fluid (CSF) AD biomarkers levels. In an effort to close this the combination (AND/OR) of the following terms: knowledge gap, we combined CSF and genome-wide single-nucleotide “neurofilament light/NF-L/NfL,” “chitinase-3-like protein polymorphism (SNP) genotyping data generated in the EMIF-AD Multi- 1/CHI3L1/YKL-40,″ “neurogranin,” “cerebrospinal fluid modal Biomarker Discovery dataset4 and performed the first bona fide biomarkers/CSF,” “Alzheimer’s disease/AD,” “genetic GWAS on CSF levels of neurofilament light chain (NfL), chitinase-3- association,” “genome-wide association study/GWAS,” like protein 1 (YKL-40), and neurogranin (Ng). Specifically, our analyses considering only studies reporting the results of genetic assessed the hypothesis that variation in these three biomarker out- association analyses for these three CSF biomarkers. For comes is (at least partially) genetically determined and that the iden- CSF levels of NfL, YKL-40, or Ng we only identified two tification of the underlying genetic factors will eventually yield new small previous GWAS, one investigating CSF YKL-40 in insights into the pathophysiology of cognitive decline and AD. 133 Japanese individuals and one investigating CSF NfL NfL is one type of four different neurofilament subunits which func- in 265 non-demented elderly. tion as structural components of the neural cytoskeleton5 performing 2. Interpretation: Our analyses yielded novel genome-wide essential roles in axon development6 and synaptic function.6 As such, significant associations with markers in TMEM106B and NfL is considered one of several “core” biomarkers of axonal injury and CSF levels of NfL in GWAS analyses using the EMIF- neurodegeneration across neurological diseases.7,8 In addition, other AD Multimodal Biomarker Discovery dataset, a finding recent data suggest that changes in NfL serum levels can predict dis- that was independently replicated in the ADNI study. ease onset and progression of brain neurodegeneration at very early, We also uncovered novel genetic association signals with pre-symptomatic stages of familial AD.9 YKL-40 is a glycoprotein pro- DNA variants in CPOX and cerebrospinal fluid (CSF) lev- duced in several inflammatory conditions and cancers,10 and was clas- els of YKL-40. Moreover, we confirmed previous studies sified as an “emerging” AD biomarker in a recent meta-analysis.7 While reporting that variants in CHI3L1 act as cis protein quan- its precise physiological role remains elusive, it appears that in AD, titative trait loci (pQTL) for CSF YKL-40. YKL-40 is predominantly expressed in astrocytes and likely plays a role 3. Future directions: Our study sheds light on the sequence in the inflammatory response occurring near Aβ plaques.10,11 Finally, of events regarding the initiation and progression of neu- Ng is a neuron-specific protein, mainly expressed in the cortex and hip- ropathological processes relevant in AD. Further inde- pocampus, where it is involved in synaptic long-term potentiation and pendent studies of comparable or larger sample size are learning.12–14 In AD, CSF Ng was proposed to represent a marker of needed to validate our results and to elucidate the molec- synaptic degeneration and was recently reported to correlate with cog- ular mechanisms underlying the observed associations. nitive decline.15 To the best of our knowledge, our study represents the first bona fide GWAS on these biomarker traits with the exception of two small (n = 133 and n = 265, respectively) CSF protein quantita- tive trait loci (pQTL) GWAS on YKL-40 in people of Asian descent34 and a GWAS on NfL in the subset of non-demented elderly from the Alzheimer Disease Neuroimaging Initiative (ADNI) cohort.35 mean = 67.9, standard deviation [SD] = 8.3; 667 females, 554 males) with different cognitive diagnoses at baseline (NC = normal cogni- 2 METHODS tion; MCI = mild cognitive impairment; AD = AD-type dementia). Clinical follow-up data were available for 759 individuals. Indepen- 2.1 Sample description dent CSF biomarker and genotyping data used in the replication phase of this study were obtained from the ADNI database (URL: The ascertainment procedures for the EMIF-AD Multimodal http://adni.loni.usc.edu). ADNI was launched in 2003 as a public– Biomarker Discovery (MBD) dataset are described elsewhere.16 private partnership, led by Principal Investigator Michael W. Weiner, In brief, the dataset includes 1221 elderly individuals (years of age: MD. The primary goal of ADNI has been to test whether serial 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License HONG ET AL . 1631 TA B L E 1 Demographic information and summary of CSF traits used in GWAS analyses of the EMIF-AD MBD and ADNI datasets EMIF-AD MBD dataset (discovery) CSF biomarker CSF biomarker description Detection method N N male/female Mean age ± SD (range) N Controls N MCI N AD NfL Neurofilament ELISA (UmanDiagnostics) 671 319/352 69.49 ± 8.35(45.94-92.29) 122 395 154 light YKL-40 Chitinase-3-like ELISA (R&D systems) 677 323/354 69.42 ± 8.31(45.94-92.29) 122 401 154 protein 1 Ng Neurogranin Immunoassay (in-house) 672 319/353 69.50 ± 8.34(45.94-92.29) 122 398 152 ADNI dataset (replication) CSF biomarker CSF biomarker description Detection method N N male/female Mean age ± SD (range) N Controls N MCI N AD NfL Neurofilament ELISA (UmanDiagnostics) 308 185/123 74.80 ± 6.98(58.40-89.60) 87 151 70 light YKL-40 Chitinase-3-like Electrochemiluminescence 131 85/46 74.88 ± 6.09(58.40-89.60) 49 70 12 protein 1 (Meso Scale Discovery) Ng Neurogranin ELISA (Quidel Corp) 308 185/123 74.80 ± 6.98(58.40-89.60) 87 151 70 Abbreviations: AD, Alzheimer’s disease; ADNI, Alzheimer’s Disease Neuroimaging Initiative; CSF, cerebrospinal fluid; ELISA, enzyme-linked immunosorbent assay; GWAS, genome-wide association studies; MBD, multimodal biomarker discovery; MCI, mild cognitive impairment; NfL, neurofilament light chain; Ng, neurogranin; SD, standard deviation; YKL-40, chitinase-3-like protein 1. magnetic resonance imaging (MRI), positron emission tomography Ng levels were measured using an in-house immunoassay for Ng.19 (PET), other biological markers, and clinical and neuropsychological YKL-40 levels were measured using a human chitinase-3 quantikine assessment can be combined to measure the progression of MCI and ELISA kit (R&D Systems, Inc.; Olsson et al.20 ). To reduce the skew- early AD. The demographic information for the three quantitative ness of phenotype distributions, data for all three CSF variables CSF phenotypes in the EMIF-AD MBD and ADNI datasets used in this were log-transformed prior to analysis (Figure S1 in supporting study is summarized in Table 1. information). 2.2 DNA extraction, genotype imputation, and 2.4 GWAS and post-GWAS analyses quality control Linear regression (using mach2qtl21,22 ) was used to perform SNP- A detailed account of the genotyping procedures in EMIF-AD MBD based association analyses using imputation-derived allele dosages as and subsequent bioinformatic workflows can be found in the support- independent variables and the log-transformed concentrations of CSF ing information and Hong et al.17 In brief, a total of 936 DNA sam- NfL, CSF Ng, and CSF YKL-40 as dependent outcome variables. Covari- ples were sent for genome-wide SNP genotyping using the Infinium ates included into the regression models were sex, age at examina- Global Screening Array (GSA) with Shared Custom Content (Illumina tion, diagnostic groups (coded as AD = 3, MCI = 2, controls = 1), Inc.). After extensive quality control (QC) and imputation, a total of and ancestry-specific principal components (PCs; here the first five 7,778,465 autosomal SNPs with minor allele frequency (MAF) ≥0.01 were used). The genomic inflation factor was calculated in R using the were retained in 898 individuals of European ancestry for the down- “GenABEL” package.23 The FUMA platform (http://fuma.ctglab.nl/)24 stream genome-wide association analyses. For more details see sup- was used for post-GWAS analyses, including gene-based analyses.25 porting information and Hong et al.17 Genome-wide significance was defined as follows for the two types of genome-wide analyses performed: 5E-08, a widely accepted threshold 2.3 CSF biomarkers for the SNP-based analyses,1,2 and 2.651E-6 for the gene-based anal- yses, corresponding to testing 18,862 protein coding genes (as sug- Details of the CSF biomarker measurements can be found in Bos et al.4 gested on FUMA). Overall, the effective sample sizes for the GWAS In brief, the CSF specimens were collected individually at each of the analyses were n = 671, n = 677, and n = 672 in EMIF-AD MBD for 11 EMIF-AD MDB participating sites. CSF samples were shipped to CSF NfL, YKL-40, and Ng, respectively (Table 1; see below for corre- the Department of Psychiatry and Neurochemistry at University of sponding numbers in ADNI). In addition to running separate GWAS for Gothenburg, Sweden. Relevant to the analyses presented here, NfL lev- EMIF-AD and ADNI, we combined results from both datasets by meta- els were measured using a commercial enzyme-linked immunosorbent analysis using inverse variance weighted fixed-effect models as imple- assay (ELISA; NF-light ELISA, UmanDiagnostics; Zetterberg et al.18 ). mented in METAL (http://csg.sph.umich.edu/abecasis/Metal/). 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1632 HONG ET AL . 2.5 Polygenic risk score analysis In contrast to the SNP-based results, the gene-based analyses using MAGMA elicited a third locus on chromosome 7 (7p21.3) at trans- Summary statistics of the two largest and most recent AD case-control membrane protein 106B (TMEM106B) showing genome-wide signifi- GWAS1,2 were used for calculating polygenic risk score (PRS) for each cant association with CSF NfL levels (Table 2, Figure S2A, and Table S1). individual in EMIF-AD MDB. PRS were constructed for 11 different This gene, which is an established genetic risk modifier for frontotem- P-value thresholds using PLINK v1.926 after additional QC (supporting poral lobar degeneration (FTLD),27,28 contains 187 SNPs of which the information). The resulting PRSs were used as independent variables majority (n = 124 SNPs) are in strong linkage disequilibrium (LD; in the regression models adjusting for sex, age, diagnosis, and PC1-5 as r2 > 0.6) with the lead variant in TMEM106B, that is, rs1548884 show- covariates as in the primary GWAS analyses. ing evidence for genome-wide suggestive association in EMIF-AD MBD (P = 2.62E-07, Table 2 and Table S1). Given its MAF of ≈0.42, this 2.6 Replication analyses in ADNI SNP could be assessed in the independent ADNI dataset where it showed nominal evidence of association with consistent effect direc- For replication purposes, we used CSF biomarker and genotyping data tion (P = .0026, Table 2, Figure S2B, Table S1). Combining the EMIF-AD from the ADNI study to assess replicability of the EMIF-AD MBD MBD and ADNI GWAS results for this SNP by meta-analysis yielded findings. The ADNI sample used here comprises n = 131 (CSF YKL-40) association evidence that exceeded the threshold for genome-wide or n = 308 (CSF NfL and Ng) subjects with available microarray-based significance (Pmeta = 3.85E-09; Figure 1A, Table 2). In addition to genotyping and relevant biomarker phenotype data (Table 1; for more rs1548884, >80 variants in the TMEM106B region elicited genome- details on ADNI see http://adni.loni.usc.edu/). In brief, CSF NfL levels wide significant association in the meta-analyses (best Pmeta = 2.27E- in ADNI were measured with the same ELISA as for EMIF-AD MBD 09 at rs7797705; Figure 1A, Table S1). Gene-based analyses using (NF-light ELISA, UmanDiagnostics). Ng levels were measured using MAGMA in ADNI also showed independent association between a monoclonal antibody specific for neurogranin (Ng7) with electro- TMEM106B and CSF NfL (P = .00128; Table 2, Figure S2B), increasing chemiluminescence technology (Meso Scale Discovery). YKL-40 levels the overall gene-based evidence of meta-analyses across both EMIF- were measured with the MicroVue YKL-40 ELISA (Quidel Corp). ADNI AD MBD and ADNI datasets (Pmeta = 1.32E-08; Figure 1B, Table 2, Fig- genotyping data was processed using the same workflow as for the ure S2C and D, Table S1). EMIF-AD MBD analyses (i.e., same QC and imputation pipeline, same FUMA-based functional annotations show one non-synonymous regression models) with one exception: we applied a slightly more variant (rs3173615; in nearly perfect LD [r2 > 0.99] with rs1548884, conservative MAF threshold (i.e., 0.03 instead of 0.01), as variants Tables S2 & S3 in supporting information) eliciting a Thr185Ser change in the lower allele spectrum tend to produce less reliable results in with a CADD score of 21.4, and a predicted “moderate” impact by small datasets. Accordingly, this led to a lower overall number of SNPs ENSEMBL’s variant effect predictor (VEP) algorithm. In addition to available for GWAS (5,858,348) in this dataset. possibly exerting an effect on protein function by directly altering the A detailed description of all methods and procedure applied in this amino-acid sequence of TMEM106B, the same variant is also reported study can be found in the supporting information. as a modest expression quantitative trait locus (eQTL) in cortical brain samples of the Genotype Tissue Expression (GTEx, v8) project (P = 3.8E-05; Table S3). Interestingly, the lead variant in our CSF NfL 3 RESULTS GWAS (rs1548884) is in strong LD (r2 = 0.98; Table S2) with the SNP 3.1 GWAS analyses using CSF neurofilament light originally identified and subsequently replicated to be associated with levels FTLD (rs1990622, Pmeta = 4.21E-09, Table S1; Van Deerlin et al.27 ). The GWAS analyses using CSF NfL as outcome yielded five SNPs show- 3.2 GWAS analyses using CSF YKL-40 levels ing genome-wide significant association in the EMIF-AD MBD dataset (Table 2, Figure S2A, and Table S1 in supporting information). Three of The SNP-based GWAS using CSF YKL-40 levels yielded one genome- these SNPs are located in two distinct loci on chromosome 7 (i.e., on wide significantly associated locus on chromosome 1q32.1. This signal chr. 7q36.1 [rs111748411, rs3094407] and on 7p15.3[rs77589784]), was driven by three independent SNPs (i.e., rs7551263, rs1417152, while the other two are located on chromosomes 1p36.12 (rs4654961) and rs10399931; Table 2, Figure S3A, Table S4 in supporting informa- and 10q26.3 (rs138898705; Table S1 and Figure S2A). MAFs for SNPs tion) and also represents the single most significant GWAS signal in the in all but the 7q36.1 locus were ≈1% complicating any inferences and EMIF-AD MBD dataset of this study (P = 4.79E-11 for rs10399931). functional interpretations of these variants given the limited size of Unlike the GWAS results for the other two CSF markers analyzed here, our dataset. For the two common SNPs in the chr. 7q36.1 locus (i.e., the strongest results were observed with relatively common variants rs111748411, rs3094407), post-GWAS variant annotation in FUMA showing allele frequencies between ≈16% and 21% in people of Euro- suggested no obvious functional consequences (Table S2 in support- pean ancestry (Table S5a in supporting information). Association anal- ing information). Owing to their low allele frequency, none of these yses in the ADNI dataset showed independent evidence of association variants were assessed in the ADNI dataset, so these CSF NfL findings with the same direction of effect for two of the three SNPs (P = .041, should be viewed with caution until independent replication in suffi- P = .5523, P = 9.19E-07, for rs7551263, rs1417152, and rs10399931, ciently sized samples is provided. respectively; Table 2, Figure S3B, Table S4). Accordingly, the HONG ET AL . TA B L E 2 Top results from GWAS analyses of three CSF traits measured in the EMIF-AD MBD and ADNI datasets Results for EMIF-AD MDB dataset Results for ADNI dataset Results for meta-analysis Trait ID Chr Pos Nearest gene Alleles MAF Beta P (SNP) P (Gene) Beta P (SNP) P (Gene) Beta P (SNP) P (Gene) CSF NfL rs4654961 1p36.12 21865495 ALPL A/G 0.01 −0.359 3.143E-08 0.20497 n.a n.a 0.13704 n.a n.a 0.069226 rs77589784 7p15.3 25481515 n.a. G/A 0.0118 −0.367 1.643E-08 n.a. n.a n.a n.a n.a n.a n.a rs1548884 7p21.3 12279761 TMEM106B C/A 0.4283 0.067 2.622E-07 2.6072E-07 0.101 0.002621 0.0012809 0.0715 3.85E-09 1.3381E-08 rs111748411 7q36.1 152254171 AC104843.4 A/C 0.0167 −0.403 2.285E-09 n.a. n.a n.a n.a n.a n.a n.a rs3094407 7q36.1 152373705 XRCC2 T/G 0.027 0.236 4.071E-08 0.21394 n.a n.a 0.70739 n.a n.a 0.211 rs138898705 10q26.3 132563077 n.a. T/C 0.0106 −0.389 5.238E-09 n.a. n.a n.a n.a n.a n.a n.a CSF YKL-40 rs7551263 1q32.1 203150756 CHI3L1 C/T 0.1726 −0.166 1.567E-10 2.5151E-08 −0.111 0.041 0.28397 −0.1559 3.09E-11 4.1383E-09 rs10399931 1q32.1 203156080 CHI3L1 C/T 0.2381 −0.143 4.789E-11 2.5151E-08 −0.224 9.19E-07 0.28397 −0.1583 1.42E-15 4.1383E-09 rs58943879 3q11.2-12.1 98280429 CPOX C/T 0.4967 0.096 6.102E-07 8.7456E-09 −0.014 0.7385 0.69852 0.0781 7.09E-06 1.2043E-06 CSF Ng rs10052776 5p15.2 11178648 CTNND2 T/C 0.393 0.157 1.022E-07 0.041368 0.122 0.2516 0.058017 −0.1366 2.25E-06 0.0091132 rs8111341 19q13.33 48889558 KDELR1 G/A 0.156 −0.167 1.17E-05 2.29E-05 −0.27 0.06313 0.16703 −0.1736 2.34E-06 8.0186E-06 Abbreviations: AD, Alzheimer’s disease; ADNI, Alzheimer’s Disease Neuroimaging Initiative; CSF, cerebrospinal fluid; ELISA, enzyme-linked immunosorbent assay; GWAS, genome-wide association studies; MBD, multimodal biomarker discovery; MCI, mild cognitive impairment; n.a., not available; NfL, neurofilament light chain; Ng, neurogranin; SD, standard deviation; SNP, single-nucleotide polymorphism; YKL- 40, chitinase-3-like protein 1. Notes: More results from these analyses can be found in the supporting information tables. Bold font indicates genome-wide significant (on SNP- or gene-level) results (see Methods section for details). Chr and Pos according to GRCh37/hg19. Top results from these GWAS analyses can be found in the supporting information tables (except for KDELR1). 1633 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1634 HONG ET AL . F I G U R E 1 Manhattan plots of (A) single-nucleotide polymorphism (SNP) level and (B) gene-level genome-wide association results using log-transformed cerebrospinal fluid neurofilament light chain levels as outcome trait in meta-analyses combining EMIF-AD MBD and ADNI datasets (n = 979). Gene assignments are according to FUMA.24 Dotted red lines represent the threshold for genome-wide significance, that is α = 5.0E-08 for SNP-based (A) and α = 2.651E-6 for gene-based (B) analyses (see Methods section). Manhattan plots of dataset specific results can be found in supporting information meta-analyses across both datasets yielded strong genome-wide verge on CHI3L1 as the most strongly and most significantly associated significant support for an association between two of these mark- gene when using the YKL-40–associated SNPs or their proxies as ers and CSF levels of YKL-40 (Pmeta = 3.09E-11, Pmeta = 5.87E-08, input (see also Table S6 in supporting information). Interestingly, this Pmeta = 1.42E-15, for rs7551263, rs1417152, and rs10399931, SNP is also listed as methylation QTL (mQTL) on the mQTL database respectively; Figure 2A, Table 2, Figure S3C and D, Table S4). While (http://www.mqtldb.org/).29 FUMA-based gene annotations (Table S5b) highlight up to 26 different Gene-based GWAS analyses confirmed the association with CHI3L1 gene symbols in the implicated region, the most obvious candidate (P = 2.52E-08; Figure S3A, Table S4), and revealed a second, indepen- of likely biological relevance is CHI3L1 (chitinase-3-like 1), that is, dent locus, CPOX (coproporphyrinogen oxidase; chromosome 3q11.2- the gene encoding YKL-40 protein. In other words, this GWAS result 12.1), showing even more significant gene-based association with CSF represent a bona fide cis pQTL result. Furthermore, and corresponding YKL-40 levels (P = 8.75E-09; Figure S3A, Table S4). The most signifi- to these pQTL results, eQTL annotations summarized by FUMA con- cantly associated single variant in CPOX was rs58943879 (P = 6.1E-07; 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License HONG ET AL . 1635 F I G U R E 2 Manhattan plots of (A) single-nucleotide polymorphism (SNP) level and (B) gene-level genome-wide association results using log-transformed cerebrospinal fluid CHITINASE-3-LIKE PROTEIN 1levels as outcome trait in meta-analyses combining EMIF-AD MBD and ADNI datasets (n = 808). Gene assignments are according to FUMA.24 Dotted red lines represent the threshold for genome-wide significance, that is, α = 5.0E-08 for SNP-based (A) and α = 2.651E-6 for gene-based (B) analyses (see Methods section). Manhattan plots of dataset specific results can be found in supporting information Table S4). While the ADNI analyses did not provide independent sup- 3.3 GWAS analyses using CSF neurogranin levels port for an association between CPOX and CSF YKL-40 (Table 2, Figure S3B, Table S4), the gene-based GWAS across both datasets still elicited Last, the GWAS using CSF Ng levels yielded no genome-wide sig- gene-wide significance, albeit with reduced evidence compared to the nificant association in the SNP-based analyses (Table 2, Figure S4A, EMIF-AD MBD dataset alone (Pmeta = 1.2043E-06; Figure 2B, Table 2, and Table S7 in supporting information). The top-ranking SNP-based Table S3C and D, Table S4). Manual lookup on the GTEx portal (v8) finding was elicited by rs10052776 (P = 1.0E-07, Table S7), located revealed no previously observed eQTLs in the brain, despite CPOX’s in CTNND2 mapping to chromosome 5p15.2. Interestingly, SNPs relatively pronounced expression in all brain tissues sampled in GTEx in this gene were previously associated with both late-onset AD (https://www.gtexportal.org/home/gene/CPOX). and cognitive performance by GWAS30,31 according to the “GWAS 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1636 HONG ET AL . F I G U R E 3 Manhattan plots of (A) single-nucleotide polymorphism (SNP) level and (B) gene-level genome-wide association results using log-transformed cerebrospinal fluid neurogranin levels as outcome trait in meta-analyses combining EMIF-AD MBD and ADNI datasets (n = 980). Gene assignments are according to FUMA.24 Dotted red lines represent the threshold for genome-wide significance, that is, α = 5.0E-08 for SNP-based (A) and α = 2.651E-6 for gene-based (B) analyses (see Methods section). Manhattan plots of dataset specific results can be found in supporting information catalog” (https://www.ebi.ac.uk/gwas/genes/CTNND2). However, to chromosome 19q13.33, a gene hitherto not associated with the replication analyses in ADNI did not support these findings, result- traits listed in the GWAS catalog. ing in less pronounced association evidence in the meta-analyses (Pmeta = 0.2516 for rs10052776; Table 2, Figure 3A, Table S7). Gene- based association analyses using MAGMA also did not reveal any 3.4 Genetic correlation analyses genome-wide significant signals with variants annotated to the 18,862 genes used in these analyses, neither in EMIF-AD MBD alone (Figure To assess whether and to which degree variation of the three AD CSF S4A) nor in the meta-analyses with ADNI (Figure 3B; Figure S4C and biomarkers analyzed in this study show association with “AD-related” D, Table S7). The top-ranking gene-based finding with 21 SNPs was variants we calculated PRSs as predictors of CSF biomarker varia- observed with KDELR1 (PEMIF = 2.29E-05, Pmeta = 8.21E-06) mapping tion (see supporting information for justification and details). Similar 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License HONG ET AL . 1637 analyses in earlier work17 had revealed a comparatively strong asso- with TMEM106B mRNA expression in the brain (cortex and cerebel- ciation with Aβ-derived but not tau-derived biomarkers. Similar to lum), albeit at lesser significance (https://www.gtexportal.org/home/ CSF-tau measures, the CSF biomarker phenotype variance in EMIF- snp/rs10950398). Of note, this pQTL/eQTL variant (rs10950398) is in AD MBD explained by AD PRS was collectively minor, reaching nomi- nearly perfect LD with the lead SNP identified here to show associ- nal significance for some CSF phenotypes and P-value thresholds using ation with CSF NfL levels (i.e., rs1548884, r2 = 0.97; Tables S1 & 2). Kunkle et al.,2 but none from the Jansen et al.1 data (Table S8 in sup- In summary, there is now convincing evidence converging from mul- porting information). This is in contrast to applying AD GWAS-based tiple lines of independent data and datasets that DNA variants in PRS to Aβ-related CSF phenotypes in the same EMIF-AD MDB dataset: TMEM106B not only show association with several neuropsychiatric here, the strongest associations explained 6% to 12% of the pheno- and non-neurological phenotypes, but also cis (TMEM106B, previous typic variance (P = 9E-09 and 1E-07, respectively),17 suggesting that work) and trans (CSF NfL, this study) pQTL associations with proteins the genetic architectures underlying AD risk and variation at CSF NfL, relevant for neuronal function. In the AD field, NfL recently (re)gained YKL-40, and Ng in general do not show any substantial overlap. interest based on data suggesting that NfL protein dynamics in serum may predict progression and brain neurodegeneration at early pre- symptomatic stages of familial AD,9 and help track neurodegeneration 4 DISCUSSION in sporadic AD.38 The novel results from our GWAS indicate that DNA sequence variants in TMEM106B may be involved in regulating CSF NfL We performed GWAS analyses for three CSF AD-related biomarkers protein levels. Because the same variant(s) are also cis eQTLs/pQTLs of in the EMIF-AD MDB dataset and identified novel genome-wide sig- TMEM106B mRNA/protein levels, it is tempting to speculate that the nificant association with genetic markers in the established FTLD risk observed effect on CSF NfL may be mediated by TMEM106B mRNA gene TMEM106B and NfL protein concentrations in the CSF. This find- or protein. In line with this hypothesis is the observation that the lead ing showed consistent independent replication in comparable CSF data genetic variant in TMEM106B highlighted in our analyses (rs1548884) from ADNI. In addition, we detected very strong and genome-wide sig- shows some, albeit sub–genome-wide, evidence for association with nificant association between markers in CHI3L1 and CSF levels of YKL- AD risk in the two largest and most recent GWAS in the field, that is, 40, representing the only cis pQTL finding in our analyses. To the best P = 0.00005 and P = 0.005 in Jansen et al.1 and Kunkle et al.,2 respec- of our knowledge, our study is the first bona fide GWAS for all three tively. In conclusion, our novel data now provide the genetic foundation of these CSF biomarkers with the exception of two small (n = 133 and for future work aimed at elucidating whether the observed increase in n = 265) CSF pQTL GWAS on YKL-40 in people of Asian descent34 and CSF NfL levels represents a “cause” or “effect” of the neurodegenera- a GWAS on NfL in a subset of non-demented elderly from the ADNI tive processes underlying symptomatic and pre-symptomatic AD. The cohort.35 Of note, the former study also identified strong and genome- observation that the same TMEM106B variants show association with wide significant cis pQTL effects at the CHI3L1/YKL-40 locus, corrob- both AD risk and CSF NfL levels in independent datasets provides a first orating our findings. Other noteworthy results from our study include indication that the recently proposed change in “NfL dynamics” may, evidence for several rare-variant associations with CSF NfL levels and indeed, contribute to AD neuropathology rather than simply reflect an an overall lack of AD-related genetic association signals with the CSF effect of the same. However, additional work is needed, for example biomarkers analyzed here. This latter point explicitly includes genetic replication of the original NfL dynamics result and approaches apply- variants in or near the apolipoprotein E gene, the strongest currently ing Mendelian randomization, to address this question more formally. known genetic AD risk factor,3 which did not show association with any The only other GWAS investigating CSF NfL was performed on of the CSF biomarkers analyzed here. 265 non-demented individuals from the ADNI cohort.35 These authors Possibly the most noteworthy novel signal observed in our study highlighted two SNPs (i.e., rs465401 and rs460420) in ADAMTS1 to relates to the association between TMEM106B and CSF NfL. DNA show genome-wide significant association with CSF NfL. While, as variants in TMEM106B have first been implicated in neuropsychiatric expected, both markers showed some level of association in the ADNI research by a GWAS on FTLD with TAR DNA-binding protein (TDP- analyses of our study as well (both with P-values = .001), there was no 43) inclusions (FTLD-TDP; Van Deerlin et al.27 ), a finding that was sub- evidence of association in the EMIF-AD MBD dataset (P-values = .5485 sequently confirmed in independent datasets (see Pottier et al.28 for and .5391, respectively) nor in the meta-analyses across all sam- recent GWAS results). Furthermore, a recent meta-analysis revealed ples (Pmeta = 0.56 and 0.55, respectively). Conversely, variants in that CSF NfL levels are significantly increased in FTLD.8 In addi- TMEM106B were not highlighted as “peak results” in that study. How- tion to FTLD, the “GWAS catalog” database lists a number of other, ever, apart from the difference in sample size, there are several addi- mostly neuropsychiatric (e.g., depression, differential brain aging, neu- tional noteworthy differences in the analysis approach used by us and roticism) but also non-neurological (e.g., leukemia, height, high-density Niu et al.35 (e.g., they did not use genotype imputations to increase lipoprotein levels) traits showing genome-wide association with their coverage of untyped portions of the genome). Future work in TMEM106B (https://www.ebi.ac.uk/gwas/genes/TMEM106B; Buniello more individuals needs to determine whether the association between et al.36 ) There is also strong cis pQTL GWAS evidence for SNPs in TMEM106B and CSF NfL observed here will continue to prove genuine. this region (in particular: rs10950398) and TMEM106B protein lev- The second novel result emerging from our analyses implies els in blood.37 In GTEx, this same SNP is also reported to correlate that variants in CPOX (encoding coproporphyrinogen oxidase) are 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1638 HONG ET AL . associated with YKL-40 levels in CSF in gene-based analyses of the AD MBD dataset is the first and/or largest to allow GWAS analyses EMIF-AD MBD dataset. While these findings were not independently on all three CSF variables covered, the sample size available for anal- replicated in ADNI, we note that data for CSF YKL-40 currently remain ysis is still relatively modest (range: 671 to 677), limiting our power to particularly scarce (n = 131; Table 1) in this dataset as opposed to CSF detect genetic variants of moderate to small effects. Thus, the results NfL and Ng (n = 308). Notwithstanding, gene-based results across both of our GWAS likely only represent the “lowest hanging fruit” of the datasets elicited gene-wide significant association between markers genetic factors underlying the analyzed traits. Third, we note that while in CPOX and CSF YKL-40, albeit at reduced significance compared to both the genome-wide SNP genotypes as well as CSF biomarker con- EMIF-AD MBD alone. CPOX is ubiquitously expressed (based on GTEx centrations were generated in one run of consecutive experiments in release [v8]) and encodes an enzyme involved in the heme biosynthetic two dedicated laboratories (one for genotyping, one for CSF markers; pathway. Intracellularly, it localizes to the mitochondria and catalyzes likely reducing the possibility of batch effects), the CSF specimens were the two-step oxidative decarboxylation of the heme precursor copro- collected individually at each of the 11 EMIF-AD MDB participating porphyrinogen III to protoporphyrinogen IX (https://omim.org/entry/ sites, sometimes using different CSF collection procedures. While this 612732). While common variants in this gene have hitherto not been sampling heterogeneity could have affected our results (although YKL- associated with any human trait recorded in the “GWAS catalog,” rare 40 and Ng were recently shown to be quite stable across a range of mutations in CPOX can cause coproporphyria and harderoporphyria conditions44 ), we note that CSF drawing was performed independent (OMIM phenotype ID # 121300), hereditary forms of porphyrias char- of genotype, so any batch effect in this particular setting should be min- acterized by enzyme deficiencies in the heme biosynthetic pathway. imal. Fourth, to clarify the role of the observed SNP/CSF protein asso- Previous work has suggested that heme has a strong affinity for bind- ciations for use as potential biomarkers for an early recognition of the ing Aβ42 peptide in vitro,39 leading to speculations that porphyrias development of AD, it would have been useful to separately analyze could potentially alter the risk and/or course of AD.40 However, given subjects with other non-AD neurodegenerative disorders. However, that these potential links have not hitherto been directly proven in these types of diagnoses are currently not available within this dataset. experimental or other work and owing to the current lack of indepen- Finally, we note that the EMIF-AD MBD dataset was not designed to be dent validation in ADNI, the potential link between CPOX and CSF NfL “representative” of the general population but was assembled with the must be considered preliminary until further replication evidence is aim to achieve approximately equal proportions of “amyloid-positive” accrued. versus “amyloid–negative” individuals across individuals with normal Finally, the third main finding worth discussing is the cis pQTL result cognition and MCI although this was only achieved for MCI.16 While linking markers in CHI3L1 to CSF levels of YKL-40, which represents this ascertainment strategy does not invalidate our GWAS results, they the strongest and most significant of all association signals in our may not be generalizable to the underlying population as a whole. How- GWAS. Several prior publications have implicated genetic variants in ever, this limitation can affect any study with clinically ascertained par- CHI3L1 to represent cis pQTLs of YKL-40 levels in blood37,41–43 (for ticipants and, thus, applies to most previously published GWAS in the more details, see: https://www.ebi.ac.uk/gwas/genes/CHI3L1). How- field, including those performed in ADNI. ever, to the best of our knowledge, only one prior study has investi- In conclusion, our GWAS on CSF NfL, YKL-40, and Ng pro- gated YKL-40 levels in the human CSF.34 Notwithstanding that study’s vides important new insights into the genetic architecture underlying relatively small sample size (n = 133) and different ethnicity (Japanese), interindividual variation in these traits. Together with recent GWAS this GWAS also reported very pronounced cis pQTL effects of genetic results on AD risk from case-control studies, especially the TMEM106B variants in CHI3L1. Taken together, there is now compelling converg- effects on CSF NfL shed important new light on the sequence of events ing evidence that expression of YKL-40 in both blood and CSF is reg- in relation to the initiation and progression of neuropathological pro- ulated by DNA sequence variants located in the very gene encoding cesses relevant in AD. Additional work is needed to set our results this protein. These same variants are also found as eQTL and mQTLs in onto a broader evidence-based foundation and to clarify the molecu- independent datasets. However, unlike the situation observed for our lar mechanisms underlying the observed associations. GWAS results for NfL, the YKL-40 regulatory SNPs do not show any evidence of association with AD risk in the GWAS by Jansen et al.1 and ACKNOWLEDGMENTS Kunkle et al.2 Thus, owing to this general absence of genetic associa- The present study was conducted as part of the EMIF-AD MBD tion with AD risk it appears that the observed association between CSF project, which has received support from the Innovative Medicines Ini- YKL-40 and AD status7 probably lies downstream of the initiation of tiative Joint Undertaking under EMIF grant agreement No. 115372, AD neuropathology. the resources of which are composed of financial contribution from As with all GWAS, our study is subject to some limitations. First, the European Union’s Seventh Framework Program (FP7/2007-2013) while we successfully provided a first line of replication evidence of our and EFPIA companies’ in kind contribution. Parts of this study were main EMIF-AD MBD findings in data from the ADNI project, we note made possible through support from the German Research Founda- that the currently available sample size for the biomarkers in ques- tion (DFG grant FOR2488: Main support by subproject “INF-GDAC” tion in ADNI is comparatively small, especially for YKL-40 (Table 1). BE2287/7-1 to LB) and the Cure Alzheimer’s Fund (to LB and RET). Thus, these analyses will need to be repeated when more extensive RV acknowledges support by the Stichting Alzheimer Onderzoek biomarker assessments become available. Second, although the EMIF- (#13007, #11020, #2017-032) and the Flemish Government (VIND 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License HONG ET AL . 1639 IWT 135043). KB is supported by the Swedish Research Council CONFLICTS OF INTEREST (#2017-00915); the Alzheimer Drug Discovery Foundation (ADDF), KB has served as a consultant, on advisory boards, or on data monitor- USA (#RDAPB-201809-2016615); the Swedish Alzheimer Foundation ing committees for Abcam, Axon, Biogen, Julius Clinical, Lilly, MagQu, (#AF-742881), Hjärnfonden, Sweden (#FO2017-0243); the Swedish Novartis, Roche Diagnostics, and Siemens Healthineers, and is a co- state under the agreement between the Swedish government and founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which the County Councils; the ALF-agreement (#ALFGBG-715986); and is a part of the GU Ventures Incubator Program, all unrelated to the European Union Joint Program for Neurodegenerative Disorders work presented in this paper. HZ has served on scientific advisory (JPND2019-466-236). HZ is a Wallenberg Scholar supported by grants boards for Denali, Roche Diagnostics, Wave, Samumed and CogRx; has from the Swedish Research Council (#2018-02532), the European given lectures in symposia sponsored by Fujirebio, Alzecure, and Bio- Research Council (#681712), and Swedish State Support for Clini- gen; and is a co-founder of Brain Biomarker Solutions in Gothenburg cal Research (#ALFGBG-720931). SJBV received funding from the AB (BBS), which is a part of the GU Ventures Incubator Program. The Innovative Medicines Initiative 2 Joint Undertaking under ROADMAP other authors declare no conflicts of interests. grant agreement No. 116020 and from ZonMw during the conduct of this study. No conflict of interest exists. Research at VIB-UAntwerp AUTHOR CONTRIBUTIONS was in part supported by the University of Antwerp Research Fund. Shengjun Hong performed all the analyses and interpretation of data The authors acknowledge the assistance of Ellen De Roeck, Naomi De and co-wrote all drafts of the manuscript. Valerija Dobricic extracted Roeck, and Hanne Struyfs (UAntwerp) with data collection. The Lau- EMIF-AD MBD DNA sample. Isabelle Bos, Stephanie J. B. Vos, Betty sanne study was funded by a grant from the Swiss National Research M. Tijms, and Pieter Jelle Visser coordinated the collection and harmo- Foundation (SNF 320030_141179) to JP. nization of phenotypes and biosamples in EMIF-AD MBD and helped We thank Mrs. Tanja Wesse and Mrs. Sanaz Sedghpour Sabet at the identify equivalent phenotypes from the ADNI catalog. Olena Ohlei, Institute of Clinical Molecular Biology, Christian-Albrechts-University Christina Maria Lill, Dmitry Prokopenko, and Rudolph E. Tanzi con- of Kiel, Kiel, Germany for technical assistance with the GSA geno- tributed to replication analyses in ADNI. Andre Franke and Michael typing. We thank Dr. Fabian Kilpert for his assistance with the QC Wittig supervised the genotyping experiments. Ulf Andreasson, Kaj and genotype imputations. The LIGA team acknowledges computa- Blennow, and Henrik Zetterberg performed CSF biomarker measure- tional support from the OMICS compute cluster at the University of ments and took part in cut-point determinations. Kristel Sleegers and Lübeck. Christine Van Broeckhoven contributed to genetic characterization of For ADNI: Data was used for this project of which collection and samples and design of the genomics studies in EMIF-AD MBD. Rik sharing was funded by the Alzheimer’s Disease Neuroimaging Initia- Vandenberghe and Silvy Gabel contributed to sample and data col- tive (ADNI) (National Institutes of Health Grant U01 AG024904) and lection. Johannes Streffer, Pieter Jelle Visser, and Simon Lovestone DOD ADNI (Department of Defense award number W81XWH-12-2- are leads for the EMIF-AD MBD; as such they designed and man- 0012). ADNI is funded by the National Institute on Aging, the National aged the platform. Lars Bertram designed and supervised the genomics Institute of Biomedical Imaging and Bioengineering, and through portion of the EMIF-AD MBD project and co-wrote all drafts of the generous contributions from the following: AbbVie; Alzheimer’s manuscript. All authors critically revised all manuscripts drafts, and Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; read and approved the final manuscript. BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; REFERENCES Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; 1. Jansen IE, Savage JE, Watanabe K, et al. Genome-wide meta-analysis EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company identifies new loci and functional pathways influencing Alzheimer’s Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen disease risk. Nat Genet. 2019;51(3):404-413. Alzheimer Immunotherapy Research & Development, LLC; Johnson 2. Kunkle BW, Grenier-Boley B, Sims R, et al. Genetic meta-analysis of & Johnson Pharmaceutical Research & Development LLC; Lumosity; diagnosed Alzheimer’s disease identifies new risk loci and implicates Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC; NeuroRx Aβ, tau, immunity and lipid processing. Nat Genet. 2019;51:414-430. 3. Bertram L, Tanzi RE. Genomic mechanisms in Alzheimer’s Disease. Research; Neurotrack Technologies; Novartis Pharmaceuticals Cor- Brain Pathol. 2020;30(5):966-977. poration; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical 4. Bos I, Vos S, Verhey F, et al. Cerebrospinal fluid biomarkers of neurode- Company; and Transition Therapeutics. The Canadian Institutes of generation, synaptic integrity, and astroglial activation across the clin- Health Research is providing funds to support ADNI clinical sites in ical Alzheimer’s disease spectrum. Alzheimer’s. Dement. 2019;15:644- 654. Canada. Private sector contributions are facilitated by the Foundation 5. Yuan A, Nixon RA. Specialized roles of neurofilament proteins in for the National Institutes of Health (www.fnih.org). The grantee synapses: relevance to neuropsychiatric disorders. Brain Res Bull. organization is the Northern California Institute for Research and 2016;126:334-346. Education, and the study is coordinated by the Alzheimer’s Therapeu- 6. Yuan A, Rao MV, Veeranna NixonRA. Neurofilaments at a glance. J Cell Sci. 2012;125:3257-3263. tic Research Institute at the University of Southern California. ADNI 7. Olsson B, Lautner R, Andreasson U, et al. CSF and blood biomarkers data are disseminated by the Laboratory for Neuro Imaging at the for the diagnosis of Alzheimer’s disease: a systematic review and meta- University of Southern California. analysis. Lancet Neurol. 2016;15:673-684. 15525279, 2021, 10, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12330 by Readcube (Labtiva Inc.), Wiley Online Library on [03/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1640 HONG ET AL . 8. Bridel C, Van Wieringen WN, Zetterberg H, et al. Diagnostic value of 29. Gaunt TR, Shihab HA, Hemani G, et al. Systematic identification cerebrospinal fluid neurofilament light protein in neurology: a system- of genetic influences on methylation across the human life course. atic review and meta-analysis. JAMA Neurol. 2019;76:1035-1048. Genome Biol. 2016;17:61. 9. Preische O, Schultz SA, Apel A, et al. Serum neurofilament dynam- 30. Lee JJ, Wedow R, Okbay A, et al. Gene discovery and polygenic predic- ics predicts neurodegeneration and clinical progression in presymp- tion from a genome-wide association study of educational attainment tomatic Alzheimer’s disease. Nat Med. 2019;25:277-283. in 1.1 million individuals. Nat Genet. 2018;50:1112-1121. 10. Molinuevo JL, Ayton S, Batrla R, et al. Current state of Alzheimer’s fluid 31. Mez J, Chung J, Jun G, et al. Two novel loci, COBL and SLC10A2, biomarkers. Acta Neuropathol. 2018;136:821-853. for Alzheimer’s disease in African Americans. Alzheimer’s. Dement. 11. Querol-Vilaseca M, Colom-Cadena M, Pegueroles J, et al. YKL-40 2017;13:119-129. (Chitinase 3-like I) is expressed in a subset of astrocytes in Alzheimer’s 32. Bulik-Sullivan B, Finucane HK, Anttila V, et al. An atlas of genetic cor- disease and other tauopathies. J Neuroinflammation. 2017;14: relations across human diseases and traits. Nat Genet. 2015;47:1236- 118. 1241. 12. Chang JW, Schumacher E, Coulter PM, Vinters HV, Watson JB. 33. Byers SA, Price JP, Cooper JJ, Li Q, Price DH. HEXIM2, a HEXIM1- Dendritic translocation of RC3/neurogranin mRNA in normal aging, related protein, regulates positive transcription elongation factor b Alzheimer disease and fronto-temporal dementia. J Neuropathol Exp through association with 7SK. J Biol Chem. 2005;280:16360-16367. Neurol. 1997;56:1105-1118. 34. Sasayama D, Hattori K, Ogawa S, et al. Genome-wide quantitative trait 13. Scheff SW, Price DA, Schmitt FA, DeKosky ST, Mufson EJ. Synap- loci mapping of the human cerebrospinal fluid proteome. Hum Mol tic alterations in CA1 in mild Alzheimer’s disease and mild cognitive Genet. 2017;26:44-51. impairment. Neurology. 2007;68:1501-1508. 35. Niu L-D, Xu W, Li J-Q, et al. Genome-wide association study of cere- 14. Slemmon JR, Feng B, Erhardt JA. Small proteins that modu- brospinal fluid neurofilament light levels in non-demented elders. Ann late calmodulin-dependent signal transduction. Mol Neurobiol. Transl Med. 2019;7:657-657. 2000;22:099-114. 36. Buniello A, Macarthur JAL, Cerezo M, et al. The NHGRI-EBI 15. Tarawneh R, D’Angelo G, Crimmins D, et al. Diagnostic and prognos- GWAS Catalog of published genome-wide association studies, tic utility of the synaptic marker neurogranin in Alzheimer’s disease. targeted arrays and summary statistics 2019. Nucleic Acids Res. JAMA Neurol. 2016;73:561-571. 2019;47:D1005-D1012. 16. Bos I, Vos S, Vandenberghe R, et al. The EMIF-AD multimodal 37. Emilsson V, Ilkov M, Lamb JR, et al. Co-regulatory networks of human biomarker discovery study: design, methods and cohort characteris- serum proteins link genetics to disease. Science (80-). 2018;36:769- tics. Alzheimers Res Ther. 2018;10:64. 773. 17. Hong S, Prokopenko D, Dobricic V, et al. Genome-wide association 38. Mattsson N, Cullen NC, Andreasson U, Zetterberg H, Blennow K. study of Alzheimer’s disease CSF biomarkers in the EMIF-AD mul- Association between longitudinal plasma neurofilament light and neu- timodal biomarker discovery dataset. Transl Psychiatry. 2020;10(1): rodegeneration in patients with Alzheimer’s disease. JAMA Neurol. 403. 2019;76:791. 18. Zetterberg H, Skillbäck T, Mattsson N, et al. Association of cere- 39. Atamna H, Frey WH. A role for heme in Alzheimer’s disease: heme brospinal fluid neurofilament light concentration with Alzheimer’s dis- binds amyloid β and has altered metabolism. Proc Natl Acad Sci U S A. ease progression. JAMA Neurol. 2016;73:60-67. 2004;101:11153-11158. 19. Portelius E, Zetterberg H, Skillbäck T, et al. Cerebrospinal fluid neu- 40. Dwyer BE, Stone M, Zhu X, Perry G, Smith MA. Heme deficiency rogranin: relation to cognition and neurodegeneration in Alzheimer’s in Alzheimer’s disease: a possible connection to porphyria. J Biomed disease. Brain. 2015;138:3373-3385. Biotechnol. 2006;2006:24038. 20. Olsson B, Hertze J, Lautner R, et al. Microglial markers are elevated in 41. Sun BB, Maranville JC, Peters JE, et al. Genomic atlas of the human the prodromal phase of Alzheimer’s disease and vascular dementia. J plasma proteome. Nature. 2018;558(7708):73-79. Alzheimer’s Dis. 2013;33:45-53. 42. Suhre K, Arnold M, Bhagwat AM, et al. Connecting genetic risk to dis- 21. Li Y, Willer CJ, Ding J, Scheet P, Abecasis GR. MaCH: using sequence ease end points through the human blood plasma proteome. Nat Com- and genotype data to estimate haplotypes and unobserved genotypes. mun. 2017;8:14357. Genet Epidemiol. 2010;34:816-834. 43. Ober C, Tan Z, Sun Y, et al. Effect of variation in CHI3L1 on 22. Li Y, Willer C, Sanna S, Abecasis G. Genotype Imputation. Annu Rev serum YKL-40 level, risk of asthma, and lung function. N Engl J Med. Genomics Hum Genet. 2009;10:387-406. 2008;358(16):1682-1691. 23. Aulchenko YS, Ripke S, Isaacs A, van Duijn CM. GenABEL: an R library 44. Willemse EAJ, Vermeiren Y, Garcia-Ayllon MS, et al. Pre-analytical for genome-wide association analysis. Bioinformatics. 2007;23:1294- stability of novel cerebrospinal fluid biomarkers. Clin Chim Acta. 1296. 2019;497:204-211. 24. Watanabe K, Taskesen E, van Bochoven A, Posthuma D. Functional mapping and annotation of genetic associations with FUMA. Nat Com- mun. 2017;8:1826. SUPPORTING INFORMATION 25. de Leeuw CA, Mooij JM, Heskes T, Posthuma D. MAGMA: gen- Additional supporting information may be found online in the Support- eralized gene-set analysis of GWAS data. PLOS Comput Biol. 2015;11:e1004219. ing Information section at the end of the article. 26. Purcell S, Neale B, Todd-Brown K, et al. PLINK: a tool set for whole- genome association and population-based linkage analyses. Am J Hum Genet. 2007;81:559-575. 27. Van Deerlin VM, Sleiman PMA, Martinez-Lage M, et al. Common vari- How to cite this article: Hong S, Dobricic V, Ohlei O, et al. ants at 7p21 are associated with frontotemporal lobar degeneration TMEM106B and CPOX are genetic determinants of with TDP-43 inclusions. Nat Genet. 2010;42:234-239. cerebrospinal fluid Alzheimer’s disease biomarker levels. 28. Pottier C, Zhou X, Perkerson RB, et al. Potential genetic modifiers of Alzheimer’s Dement. 2021;17:1628–1640. disease risk and age at onset in patients with frontotemporal lobar degeneration and GRN mutations: a genome-wide association study. https://doi.org/10.1002/alz.12330 Lancet Neurol. 2018;17:548-558.