ARTICLE OPEN ACCESS

Integrative Network-Based Analysis RevealsGene Networks and Novel Drug RepositioningCandidates for Alzheimer DiseaseZachary F Gerring PhD Eric R Gamazon PhD Anthony White PhD and Eske M Derks PhD

Neurol Genet 20217e622 doi101212NXG0000000000000622

Correspondence

Dr Gerring

zacharygerring

qimrberghofereduau

AbstractBackground and ObjectivesTo integrate genome-wide association study data with tissue-specific gene expression in-formation to identify coexpression networks biological pathways and drug repositioningcandidates for Alzheimer disease

MethodsWe integrated genome-wide association summary statistics for Alzheimer disease with tissue-specific gene coexpression networks from brain tissue samples in the Genotype-Tissue Ex-pression study We identified gene coexpression networks enriched with genetic signals forAlzheimer disease and characterized the associated networks using biological pathway analysisThe disease-implicated modules were subsequently used as a molecular substrate for a com-putational drug repositioning analysis in which we (1) imputed genetically regulated geneexpression within Alzheimer disease implicated modules (2) integrated the imputed geneexpression levels with drug-gene signatures from the connectivity map to identify compoundsthat normalize dysregulated gene expression underlying Alzheimer disease and (3) prioritizeddrug compounds and mechanisms of action based on the extent to which they normalizedysregulated expression signatures

ResultsGenetic factors for Alzheimer disease are enriched in brain gene coexpression networks in-volved in the immune response Computational drug repositioning analyses of expressionchanges within the disease-associated networks retrieved known Alzheimer disease drugs (egmemantine) as well as biologically meaningful drug categories (eg glutamate receptorantagonists)

DiscussionOur results improve the biological interpretation of genetic data for Alzheimer disease andprovide a list of potential antidementia drug repositioning candidates for which the efficacyshould be investigated in functional validation studies

From the Translational Neurogenomics Laboratory (ZFG EMD) QIMR Berghofer Medical Research Institute Brisbane Queensland Australia Division of Genetic Medicine(ERG) Vanderbilt Genetics Institute Vanderbilt University Medical Center Nashville TN and Cellular and Molecular Neurodegeneration (AW) QIMR Berghofer Medical ResearchInstitute Brisbane Queensland Australia

Go to NeurologyorgNG for full disclosures Funding information is provided at the end of the article

The Article Processing Charge was funded by the authors

This is an open access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License 40 (CC BY-NC-ND) which permits downloadingand sharing the work provided it is properly cited The work cannot be changed in any way or used commercially without permission from the journal

Copyright copy 2021 The Author(s) Published by Wolters Kluwer Health Inc on behalf of the American Academy of Neurology 1

Alzheimer disease is a common neurodegenerative disordercharacterized in its early stages by mild memory loss andprogressing to severe impairment of broad executive and cog-nitive functions The most common form of Alzheimer disease(late-onset Alzheimer disease) typically affects individuals olderthan 65 years and has an oligogenic architecture with 1 major(APOE) and around 100 smaller genetic risk factors1 A recentgenome-wide association study (GWAS) meta-analysis of71880 Alzheimer cases and proxy cases and 383378 controlsidentified 20 disease-associated loci2 Detailed functionalstudies showed that these loci harbor common (minor allelefrequency gt001) single nucleotide polymorphisms (SNPs)that regulate the activity of genes in immune-related peripheraltissues (whole blood liver and spleen) as well as microglialcellsmdashthe primary immune cells of the brain Biological path-way analysis of the implicated genes shows enrichment ofdysfunctional lipoprotein clearance3 highlighting a potentiallink between dysfunctional lipid metabolism and immune re-sponses in the brain4

The integration of these genetic data with large-scale drug-response databases provides an avenue to identify existing drugsthat may alleviate the signs and symptoms of Alzheimer diseaseThis approach to drug discovery known as drug repositioningoften circumvents expensive and time-consuming phase I andphase II clinical trials and may double the success rate in drugapproval5We therefore aimed to develop an analytical pipeline tointegrate genetic risk factors with drug-response data to identifynovel compounds for the treatment of Alzheimer disease

Genetic risk factors for Alzheimer disease may converge onhighly correlated groups of genes that interact with one an-other to alter the activity of multiple biological pathways andcellular processes in a disease relevant tissue6 Gene expres-sion is an intermediate molecular phenotype that is directlymodified by DNA sequence variation (expression quantitativetrait loci [eQTLs]) epigenetic marks such as DNA methyl-ation and the environment as well as the expression of othergenes7 Gene expression analyses of postmortem brain tissuehave identified distinct cell types and biological pathwaysunderlying Alzheimer pathogenesis89 These studies arelargely based on tests of association with individual genes orgroups of curated genes with a common biological functionAn alternative agnostic approach is to model gene interac-tions using gene coexpression analysis which takes the cor-relation between every gene pair expressed in a particular(tissue) sample to generate a molecular substrate for associ-ation testing with a disease state10 We recently built gene

coexpression networks using expression data from 14 humantissues (13 from brain) obtained from healthy donors fromthe Genotype-Tissue Expression (GTEx) study11 We usedthese data to test for the enrichment of GWAS signals withingene coexpression modules (or groups of highly correlatedgenes) under the biologically relevant assumption that con-nectivity among genes may be leveraged to identify genes notdirectly implicated in disease The use of gene expression datafrom healthy participants rather than diseased cases to buildcoexpression networks has a number of advantages First itremoves the effect of ascertainment bias when collecting caseand control samples where common factors (such as medi-cation use) underlie both the exposure of interest and thedisease1213 Second it mitigates the effect of reverse causa-tion where the disease process leads to changes in gene ex-pression rather than the other way around Third for manybrain-related diseases including Alzheimer disease the dis-ease process is likely to start early before the manifestation ofsymptoms for case ascertainment Successful interventionsare required before irreversible neuronal dysfunction and losshave occurred Finally expression data from nondiseased in-dividuals are easier to collect and uniformly process in thenumbers required to characterize and model complex mo-lecular interactions These advantages ensure the construc-tion of a robust molecular substrate for the subsequentintegration of disease associations from independent samples

Recent studies exploring the role of gene coexpression net-works have been performed using postmortem brain tissues ofpatients with Alzheimer disease and non-Alzheimer controlsand recapitulate a role of immune and microglial biologicalpathways identified in GWASs14 Groups of highly connected(ie correlated) immune-related genes contain central regu-lators (or hub genes) that are highly correlated with theirneighboring genes and whose expression changes with cog-nitive impairment in Alzheimer disease14 More recent in-tegrated approaches that incorporate knowledge of networkcoexpression with other genomic elements (such as epige-netic modifications) identified gene targets and drug com-pounds for a range of immune-related disorders15 These datasuggest that characterizing the interaction and dynamic re-lationship between genes within implicated modules within agene coexpression network-based paradigm can identify andprioritize genes that may serve as effective targets for thera-peutic intervention

Coexpression networks can also be used to model the effect ofa drug compound on a group of functionally related genes

GlossaryCMap =ConnectivityMap eQTL = expression quantitative trait lociGTEx =Genotype-Tissue ExpressionGWAS = genome-wide association study LD = linkage disequilibriumMOA = mechanism of action NSAID = nonsteroidal anti-inflammatorydrug RPKM = Reads Per Kilobase of transcript per Million mapped reads SNP = single nucleotide polymorphism UKBB =UK Biobank WGCNA = weighted gene coexpression network analysis

2 Neurology Genetics | Volume 7 Number 5 | October 2021 NeurologyorgNG

For example the ConnectivityMap (CMap)16 contains geneexpression signatures resulting from genetic and pharma-cologic perturbagens measured across multiple cell typesDrug-gene signaturesmdashthat is gene expression changes

following a genetic or pharmacologic perturbagenmdashcan beintegrated with disease-associated gene expression changesto identify existing compounds that might normalize geneexpression Characterizing the complex interactions

Figure 1 Overview of the Computational Drug Repositioning Pipeline

(A) For each tissue perform aweighted gene co-expression analysis to identify modules of highly correlated genes (B) Test for the enrichment of Alzheimerrsquosdisease association signals within tissue-specific gene co-expression networks Assess the functional relevance of enriched modules by performing abiological pathway analysis of themodular genes (C) Use S-PrediXcan to impute genetically regulated levels of gene expression for modular genes All geneswithin the enriched module are separated into up-regulated and down-regulated gene sets (D) Integrate the Alzheimer disease gene sets with the CMapdatabase to identify compounds that are predicated to ldquonormaliserdquo the Alzheimer disease-associated signature (E) Generate a ranked list of drug reposi-tioning candidates and mechanism of action categories

NeurologyorgNG Neurology Genetics | Volume 7 Number 5 | October 2021 3

between genes in a network-based framework may identifytargets for potential treatments through computational drugrepositioning In the present study we have developed anovel computational drug repositioning approach that in-tegrates tissue-specific gene coexpression networks withAlzheimer association signals and drug-gene signature datato identify and prioritize drug compounds that target diseaseprocesses

MethodsAlzheimer Disease GWAS Summary StatisticsDetailed methods including a description of population co-horts quality control of raw SNP genotype data and associ-ation analyses for the Alzheimer disease GWAS are describedin detail elsewhere2 The Alzheimer disease GWAS was per-formed in a three-stage meta-analysis The first phase con-sisted of 24087 Alzheimer cases and 55058 controls collectedby the Alzheimer disease working group of the PsychiatricGenomics Consortium the International Genomics of Alz-heimerrsquos Project and the Alzheimerrsquos Disease SequencingProject All cases in phase 1 received clinical confirmation oflate-onset Alzheimer disease The second phase included47793 proxy cases and 328320 proxy controls from the UKBiobank (UKBB) proxy cases were defined as individualswith one or both parents diagnosed with Alzheimer diseasewhereas proxy controls were defined as individuals with

parents who do not have Alzheimer disease Phase 3 involvedthe meta-analysis of phase 1 and phase 2 cohorts the results ofwhich were tested for replication in an additional independentcase-control sample from deCODE (6593 Alzheimer casesand 174289 controls) Raw genotype data for each cohortwere processed according to a standardized quality controlpipeline2 Logistic regression association tests were per-formed on imputed marker dosages and binary phenotypes inphase 1 and linear regression for variables treated as con-tinuous outcomes (the number of parents with Alzheimerdisease) in phase 2 For phase 1 phenotypes the associationtests were adjusted for sex batch and the first 4 principalcomponents with age also included as a covariation in theAlzheimer-PGC cohort For phase 2 (UKBB) data age sexbatch and assessment center were included as covariatesSummary statistics for 13367301 autosomal SNPs fromphase 3 of the analyses described in reference 2 (N samples =455258) were made available by the Complex Trait GeneticsLaboratory at VU University and VU Medical CentreAmsterdam and were used in our study

Identification of Gene Coexpression ModulesWe downloaded preprocessed and normalized gene ex-pression data for 13 brain tissues collected by the GTExproject (gtexportalorg) (version 7) The expression datawere filtered to include genes with 10 or more tissue donorswith expression estimates gt01 Reads Per Kilobase of tran-script per Million mapped reads (RPKM) and an aligned

Table 1 Gene-Set Enrichment Analysis and Biological Pathway Analysis of Alzheimer Disease Modules Across 13 BrainTissues in GTEx

Tissue

Gene-set enrichment analysis Biological pathway analysis

N genes Beta SE p Value p Adjust Term Term name p Cor

Amygdala 630 0157 00351 380E-06 560E-05 GO0002376 Immune system process 248E-73

Anterior cingulate cortex BA24 353 0164 0047 234E-04 450E-03 GO0006955 Immune response 454E-78

Caudate basal ganglia 393 0143 00442 628E-04 101E-02 GO0002376 Immune system process 784E-73

Cerebellar hemisphere 197 0248 00635 478E-05 700E-04 GO0006955 Immune response 208E-50

Cerebellum 103 0425 00868 478E-07 158E-05 GO0002376 Immune system process 160E-32

Cortex 233 0226 00567 332E-05 300E-04 GO0006955 Immune response 164E-67

Frontal cortex BA9 420 013 00423 102E-03 190E-02 GO0006955 Immune response 114E-76

Hippocampus 485 0212 00395 398E-08 760E-07 GO0002376 Immune system process 112E-95

Hypothalamus 768 0191 00315 753E-10 668E-08 GO0006955 Immune response 183E-96

Nucleus accumbens basal ganglia 463 0171 00411 159E-05 400E-04 GO0006955 Immune response 969E-81

Putamen basal ganglia 352 0175 00469 961E-05 140E-03 GO0006955 Immune response 595E-72

Spinal cord cervical c-1 1205 00915 00248 110E-04 170E-03 GO0002376 Immune system process 586E-77

Substantia nigra 888 0133 00291 244E-06 394E-05 GO0002376 Immune system process 314E-88

Abbreviations Beta = test statistic from the gene-set enrichment analysis in MAGMA GO = Gene Ontology GTEx = Genotype-Tissue Expression N Genes =number of genes inmodule p adjust = p value corrected for gene size gene density and gene correlation p cor = p value corrected for correlated structure ofGO terms and corresponds to an experiment-wide threshold of α = 005 SE standard error of the test statisticSee eTable 2 linkslwwcomNXGA452 for a complete list of enriched biological pathways for the immune systemndashrelated modules

4 Neurology Genetics | Volume 7 Number 5 | October 2021 NeurologyorgNG

read count of a least 6 within each tissue The distribution ofRPKMs in each tissue sample was quantile transformed us-ing the average empirical distribution observed across allsamples Finally the gene expression values for each gene ineach tissue were transformed to the quantiles of the standardnormal distribution

The construction of gene coexpression modules followed aprotocol previously described by our laboratory11We used theweighted gene coexpression network analysis (WGCNA)

package in R v35110 to build gene coexpression networks for13 GTEx brain tissues First we computed an unsigned pair-wise correlation matrix from the GTEx gene expression valuesusing Pearson product-moment correlation coefficient Foreach correlation matrix we selected an appropriate soft-thresholding value in WGCNA by plotting the strength ofcorrelation against a range (2ndash20) of soft threshold powersWecalculated network adjacency for each correlation matrix usingthe appropriate soft-threshold power and normalized eachadjacency to generate a topological overlap matrix Hierarchical

Figure 2 Gene Overlap (A) and Coexpression Preservation (B) Between Modules Enriched With Alzheimer Disease Asso-ciation Signals

(A) The number of overlapping genes across tissue-specific modules is represented by the legend colorscale Blue represents the lowest gene overlap andred represents the highest gene overlap (B) Thepreservation of modular connectivity across tissuesis represented by the legend color scale A Z scoregreater than 10 (light blue-green) suggests that thereis strong evidence that a module is preserved be-tween the reference and test network moduleswhereas a value between 2 and 10 indicates weak tomoderate preservation and a value less than 2 indi-cates no preservation (dark blue) All preservation Zscores were above 10 (minimum Z = 1396 [betweenspinal cord cervical c-1 and anterior cingulate cor-tex]) indicating strong preservation of modular con-nectivity across brain tissuesregions

NeurologyorgNG Neurology Genetics | Volume 7 Number 5 | October 2021 5

clustering was performed on each topological overlap matrixusing average linkage with one minus the topological overlapmatrix as the distancemeasure The hierarchical cluster tree wascut into gene modules using the dynamic tree cut algorithm17

with a minimummodule size of 50 genes For each module wecalculated the first principal component the gene expressionvalues (known as an eigengene) and merged modules if thecorrelation between their eigengenes was greater or equalto 08

To assess the comparability of our module assignments topostmortemAlzheimer disease brain samples we cross-tabulatedour module assignments with those generated by Morabitoet al18 Morabito and colleagues used WGCNA to cluster 1268postmortem Alzheimer disease brain tissue gene expression datafrom 3 different studiesmdashtheMayoClinic Brain Bank (temporalcortex) Religious Orders Study andMemory and Aging Project(prefrontal cortex) and the Mount Sinai School of Medicinestudy (parahippocampal gyrus inferior frontal gyrus superiortemporal gyrus and frontal pole)

Gene-Set Analysis of GeneCoexpression ModulesThe gene-set enrichment analysis of gene coexpressionmodules followed a protocol previously described by ourlaboratory11 First we used MAGMA v10719 to (1) identifyAlzheimer disease risk genes and (2) identify coexpressionmodules enriched with Alzheimer risk genes using gene-setanalysis The gene-based test of MAGMA assigns SNPs togenes within a predefined genomic window (35 kb upstreamor 10 kb downstream of a gene body) and calculates a gene-based statistic based on the sum of the assigned SNP ndashlog(10)p values while accounting for linkage disequilibrium (LD) Toidentify coexpression modules enriched with Alzheimer dis-ease risk genes we performed a competitive gene-set analysisin MAGMA The gene-set analysis tests whether the genes

in a gene coexpression module have a greater number ofAlzheimer risk genes compared with other modules thanexpected by chance while accounting for gene size and genedensity We used an adaptive permutation procedure (N =10000 permutations) to correct for multiple testing (falsediscovery rate lt 005) The 1000 Genomes European refer-ence panel (phase 3) was used to account for LD betweenSNPs

Biological Characterization of Alzheimer-Associated Gene Expression ModulesWe performed functional enrichment analysis for each mod-ule using gProfiler (biitcsuteegprofiler)20 Gene symbolswithin tissue-specific modules were used as input and thegene universe was restricted to annotated genes We tested forthe overrepresentation of modular genes in Gene Ontologybiological processes and Reactome biological pathways ThegProfiler algorithm uses a Fisher 1-tailed test for gene path-way enrichment which tests the probability a given gene isboth a member of a coexpression module and particular bi-ological pathway or process Multiple testing correction wasperformed using gSCS to account for the correlated structureof biological annotation terms corresponding to anexperiment-wide threshold of α = 005

Overlap and Preservation of GeneCoexpression Networks Across TissuesWe assessed the module overlap and preservation of coex-pression patterns across tissue-specific coexpression modulesModule overlap was calculated as the number and proportionof genes present in each pairwise tissue comparison (N = 78)Module preservation was calculated using the mod-ulePreservation function implemented in WGCNA21 Themodule preservation function assesses the similarity of coex-pression patterns across tissue-specific modules We used theZsummary statistic to represent preservation a Zsummary

Table 2 Number (Proportion) of Drug Compounds With Significant (le-90) Connectivity Scores Across 13 Brain Tissues

Tissue

Gene input (N) Number (proportion) of significant connectivity scores by cell type

Total Up Dn A375 A549 HA1E HCC515 HEPG2 HT29 MCF7 PC3 VCAP

Amygdala 34 18 16 93 (003) 53 (002) 62 (002) 100 (004) 43 (002) 108 (004) 37 (001) 74 (003) 55 (002)

Caudate 31 18 13 51 (002) 22 (001) 77 (003) 94 (004) 46 (002) 58 (002) 20 (001) 31 (001) 28 (001)

Frontal cortex 27 18 9 93 (003) 51 (002) 59 (002) 71 (003) 60 (003) 55 (002) 36 (001) 48 (002) 28 (001)

Hippocampus 32 19 13 58 (002) 38 (001) 44 (002) 51 (002) 55 (003) 64 (002) 20 (001) 40 (001) 22 (001)

Hypothalamus 49 25 24 82 (003) 26 (001) 52 (002) 43 (002) 43 (002) 63 (002) 35 (001) 34 (001) 18 (001)

Nucleus accumbens 26 11 15 64 (002) 43 (002) 47 (002) 52 (002) 47 (002) 53 (002) 24 (001) 31 (001) 19 (001)

Spinal cord cervical 81 41 40 41 (001) 38 (001) 28 (001) 67 (003) 35 (002) 53 (002) 25 (001) 19 (001) 51 (002)

Substantia nigra 36 18 18 89 (003) 33 (001) 76 (003) 74 (003) 52 (003) 52 (002) 15 (001) 29 (001) 27 (001)

Abbreviations CMap = Connectivity Map Dn = number of downregulated genes N = total number of genes uploaded to CMap Up = number of upregulatedgenesOnly genes with reliably imputed genetically regulated gene expression were included in the CMap analysis

6 Neurology Genetics | Volume 7 Number 5 | October 2021 NeurologyorgNG

statistic greater than 10 indicates strong modular preservationacross tissues whereas a statistic between 2 and 10 indicatesweak to moderate modular preservation and a statistic lessthan 2 suggests no preservation

Computational Drug RepositioningOur computational drug repositioning analysis tests the pre-dicted effect of a drug compound on dysregulated gene ex-pression modules underlying Alzheimer disease First weused S-PrediXcan (version 0610) to estimate the magnitudeand direction of gene expression changes associated withAlzheimer disease This approach integrates eQTL in-formation with GWAS summary statistics to estimate theeffect of genetic variation underlying a disease or trait on geneexpression12 We used eQTL information (expressionweights) from 13 tissues generated by the GTEx project (v7)12 and LD information from the 1000 Genomes Project Phase322 These data were processed with beta values and standarderrors from the GWAS of Alzheimer disease to estimate theexpression-GWAS association statistic For each GTEx tissuewe extracted the S-PrediXcan Z scores for genes withinmodules enriched with Alzheimer disease association signalsand created 2 lists containing genes with either upregulated ordownregulated expression Second the gene lists were used asthe basis of drug repositioning using drug gene signaturesdownloaded from the CMap16 For each gene list and uniquecompound in CMap we calculated a connectivity score basedon a modified Kolmogorov-Smirnov score which summa-rizes the transcriptional relationship to the Alzheimer dis-ease module genes The connectivity score is a standardizedstatistic ranging from minus100 to 100 where a highly nega-tive score indicates predicted expression effect fromS-PrediXcan and the drug-gene signatures are opposing(ie genes that are upregulated in disease cases are down-regulated by the compound and vice versa) Third we se-lected compounds with connectivity scores of minus90 andlower (indicating a significant effect of a compound on theAlzheimer disease expression signature) The selected

Table 3 Number and Distribution of Connectivity Scoresby Mechanism of Action

MOADrugs(N)

Genes(N)

Score quantile

25th 50th 75th

Acetylcholine receptorantagonist

35 46 minus5150 minus1527 2848

Adrenergic receptoragonist

41 49 minus4459 minus477 3522

Adrenergic receptorantagonist

52 117 minus4484 minus718 3342

ATPase inhibitor 13 56 minus5454 minus1965 2070

Bacterial wallsynthesis inhibitor

22 40 minus4514 minus031 4156

Calcium channelblocker

24 131 minus4651 minus1290 2833

Cyclooxygenaseinhibitor

51 152 minus3642 minus384 3063

Dopamine receptoragonist

23 53 minus4596 minus1081 3112

Dopamine receptorantagonist

59 163 minus3393 minus257 2998

EGFR inhibitor 22 71 minus4357 365 4291

Estrogen receptoragonist

18 87 minus5111 minus2308 2689

Estrogen receptorantagonist

9 102 minus6286 minus3098 1564

FLT3 inhibitor 13 98 minus6699 minus2026 4339

GABA receptorantagonist

10 33 minus5237 minus1413 3292

GABA receptormodulator

8 40 minus4925 670 4138

Glucocorticoidreceptor agonist

31 160 minus5086 minus1613 3240

Glutamate receptorantagonist

31 103 minus4973 minus1092 3869

HDAC inhibitor 19 41 minus2895 1962 5130

Histamine receptorantagonist

39 96 minus4075 minus757 3032

JAK inhibitor 11 26 minus4671 minus944 4231

MEK inhibitor 12 38 minus3830 minus779 3077

Phosphodiesteraseinhibitor

31 104 minus3848 minus479 2922

PPAR receptor agonist 19 85 minus4877 640 4645

Progesterone receptoragonist

13 81 minus5321 minus2236 3557

Protein synthesisinhibitor

11 47 minus6555 minus2779 3574

Retinoid receptoragonist

10 107 minus4862 minus1289 2672

Serotonin receptoragonist

29 67 minus3608 107 3778

Table 3 Number andDistribution of Connectivity Scores byMechanism of Action (continued)

MOADrugs(N)

Genes(N)

Score quantile

25th 50th 75th

Serotonin receptorantagonist

60 157 minus4209 minus489 3409

Sodium channelblocker

23 98 minus3884 minus634 2414

Tyrosine kinaseinhibitor

14 87 minus4491 minus1065 4680

VEGFR inhibitor 18 91 minus5458 662 4119

Abbreviations Drugs (N) = number of drugs within each MOA categoryGenes (N) = number of target genes from drug bank and the drug-geneinteraction database within each MOA MOA = mechanism of actionScore quantile shows the 25th 50th and 75th quantiles for connectivityscores within each MOA

NeurologyorgNG Neurology Genetics | Volume 7 Number 5 | October 2021 7

compounds were mapped to their target genes using drug-target information curated from DrugBank23 and the Drug-Gene Interaction database24 and assign to mechanism ofaction (MOA) categories to identify chemogenomic trendsTo assess the disease specificity of the CMap enrichmentswe performed our pipeline using GWAS summary statisticsfor schizophrenia25 a brain-related neuropsychiatric disor-der with an immune component Top-ranked compoundsfrom schizophrenia and Alzheimer disease were placed in acontingency table and assessed for significant overlap(ie significant in both schizophrenia and Alzheimer dis-ease) using the hypergeometric test

To assess how drugs that have undergone a clinical trial forAlzheimer disease or its associated symptoms were prioritizedby our repositioning pipeline we downloaded all availableclinical trial data for Alzheimer disease from ClinicalTrialsgov resulting in 2565 records We subset these data to drugcompound interventions leaving 1707 records Finally weintersected the clinical trial drug list with repositioning in-formation from our study and summarized the connectivityscores for each drug

To test the significance of the Alzheimer disease perturba-tional enrichments (ie ensuring that significant results are

Figure 3 Proportion of Drug Target Genes With Significant p Values for Alzheimer Disease by MOA Category

A single asterisk indicates nominal(Fisher exact test p lt 005) enrichmentof smaller than expected (p lt 005) pvalues within a drug category A doubleasterisk indicates significant enrich-ment of smaller than expected p valueswithin a drug category after multipletesting correction for the number ofMOA categories (p lt 00016) MOA =mechanism of action

8 Neurology Genetics | Volume 7 Number 5 | October 2021 NeurologyorgNG

not due to random chance) we grouped the observed coex-pression values for pairs of genes from a single tissue(amygdala) into 100 bins We randomly sampled genes acrossbins selecting the same number of gene coexpression valuesfrom each bin as the observed data This stratified method ofsampling was performed to ensure that the observed andpermuted data were matched on connectivity The permutedcoexpression modules were uploaded to CMap using the clueAPI client and connectivity scores for each compound wereextracted from the output files We calculated empirical pvalues for observed compounds with connectivity scoressmaller than minus90 by counting the number of times the samecompound from the permuted data had a connectivity scoresmaller than minus90 A compound with an empirical p valuelt005 is unlikely to be prioritized by random chance

Standard Protocol Approvals Registrationsand Patient ConsentsThe GTEx v7 data were downloaded from the publiclyavailable GTEx data portal (gtexportalorghomedatasets)Data access followed the guidelines in the Data Use Certifi-cation Agreement All study protocols regarding humansubjects have been approved by their local institutional reviewboard and were compliant with Health Insurance Portabilityand Accountability Act (HIPAA) regulations Informedwritten consent was given by family decision makers of de-ceased donors26

Data AvailabilityAll data generated during this study are included in thispublished article and its supplementary information filesOriginal raw data can be made available on request

ResultsFigure 1 provides an overview of our analytical pipeline Step1 For 13 brain tissues in GTEx we performed a weightedgene coexpression analysis to identify modules of highlycorrelated genes Step 2 We test for the enrichment of Alz-heimer disease association signals within tissue-specific genecoexpression networks and assess the functional relevance ofenriched modules by performing a biological pathway analysisof the modular genes Step 3 We use S-PrediXcan to imputegenetically regulated levels of gene expression for modulargenes which provides a direction of effect of dysregulatedgene expression in Alzheimer disease (Alzheimer diseasesignature) All genes within the enriched module are sepa-rated into upregulated and downregulated gene sets Step 4We interrogate the CMap database for drug compounds withnegative connectivity scores that are predicated to normalizethe Alzheimer diseasendashassociated signature Step 5 We gen-erate a ranked list of drug repositioning candidates and MOAcategories that show enrichment of nominally significant (p lt005) gene-based associations (MAGMA and S-PrediXcan)across compounds within each category

Alzheimer Disease Risk Genes Are Enriched inGene Coexpression Modules Associated Withthe Immune SystemOur MAGMA gene-based analysis of existing GWAS datarevealed 74 genes significantly associated with Alzheimerdisease after multiple testing correction (p lt 278 times 10minus6)(eTable 1 linkslwwcomNXGA451 and eFigure 1 linkslwwcomNXGA463) We tested for the enrichment of gene-based association in gene coexpression modules built from 13

Figure 4 Top 3 Drugs Within MOA Categories Enriched With Gene-Based Association Signals for Alzheimer Disease

Thedensity plots andhistograms show the distribution of connectivity scores across cell types and tissues for each top-ranked compound MOA=mechanismof action

NeurologyorgNG Neurology Genetics | Volume 7 Number 5 | October 2021 9

GTEx brain tissues11 In each of the 13 tissues a single genemodule was enriched with Alzheimer disease risk genes(henceforth referred to as Alzheimer disease modules)(Table 1) (eTable 2 linkslwwcomNXGA452) Therewas substantial sharing of genes across the tissue-specificAlzheimer disease modules The modules comprised 103(cerebellum) to 1205 (spinal cord) genes 87 genes werecommon across all modules and the proportion of sharedgenes in pairwise tissue comparisons ranged from 16(amygdala and cerebellum) to 100 (cerebellum and hy-pothalamus) (Figure 2A eTable 3 linkslwwcomNXGA453) Gene pathway analyses found the enrichment ofimmune system pathways (eg immune system process inthe brain amygdala p = 248 times 10minus73 and immune responsein the brain substantia nigra p = 248 times 10minus88) within allAlzheimer modules (Table 1) (see eTable 4 linkslwwcomNXGA454 for a full list of biological pathway enrichments inthe Alzheimer disease modules) We assessed the preservation(ie reproducibility) of the connectivity structure across brain

tissues using the WGCNA modulePreservation algorithmStrong modular preservation (Z score ge 10) was observedacross all brain tissues (Figure 2B) suggesting that the con-nectivity structure (ie correlation) of genes is similar acrossbrain tissues

We also found nonenriched modules that harbored significant(p lt 278 times 10minus6) Alzheimer disease risk genes For examplemodule M1 in the amygdala contained 2 Alzheimer diseaserisk genes PICALM and BIN1 and was enriched for bi-ological pathways related to myelination and neurogenesisFurthermore module M2 in the amygdala contained 8 riskgenes (including APOC1) and was enriched with pathwaysrelated to synaptic signaling (eTable 5 linkslwwcomNXGA455) However these modules were not statisticallyenriched with Alzheimer disease risk genes on a global leveland did not replicate well across GTEx brain tissues Wetherefore focused on the immune system-related modules forthe computational repositioning analysis

Table 4 Top-Ranked Glutamate and Dopamine Receptor Antagonists for Repositioning in Alzheimer Disease

Compound MOA

Connectivity score

Mean Min Max N le 290

Methylergometrine Dopamine receptor antagonist minus5065 minus9762 minus1086 1

L-689560 Glutamate receptor antagonist minus4796 minus9576 3817 6

Memantinea Glutamate receptor antagonist minus4388 minus9474 6557 2

Ziprasidone Dopamine receptor antagonist minus3964 minus9886 4544 1

Thiothixene Dopamine receptor antagonist minus3843 minus9271 2385 1

Fluphenazine Dopamine receptor antagonist minus3093 minus9507 6263 1

Dextromethorphan Glutamate receptor antagonist minus3092 minus9936 4845 2

YM-298198 Glutamate receptor antagonist minus2114 minus9432 8377 1

N20C Glutamate receptor antagonist minus2032 minus9404 7600 2

Ifenprodil Glutamate receptor antagonist minus1650 minus9286 6801 2

Flupirtine Glutamate receptor antagonist minus1646 minus8888 9251 0

Felbamate Glutamate receptor antagonist minus1613 minus9340 6999 1

SCH-23390 Dopamine receptor antagonist minus1405 minus9008 7737 1

Clozapine Dopamine receptor antagonist minus1222 minus9672 8741 2

Droperidol Dopamine receptor antagonist minus1060 minus9977 8629 5

GR-103691 Dopamine receptor antagonist minus1045 minus9667 8793 3

SIB-1893 Glutamate receptor antagonist minus939 minus9726 8433 3

NBQX Glutamate receptor antagonist minus938 minus9964 9374 3

Iloperidone Dopamine receptor antagonist minus906 minus9764 8481 1

Triflupromazine Dopamine receptor antagonist minus849 minus9929 6473 1

Abbreviations CMap = Connectivity Map MOA = mechanism of actionN le minus90 shows the number of times a given drug compound had a connectivity score le minus90 across tissuesCMap cell typesa Memantine is currently used for the treatment of moderate-to-severe Alzheimer disease

10 Neurology Genetics | Volume 7 Number 5 | October 2021 NeurologyorgNG

We compared our gene coexpression modules which werederived from brain tissue samples from healthy donors withmodules reported by Morabito et al who derived theirmodules from postmortem Alzheimer disease brain tissuesamples We found an average of 54 of genes overlappingbetween our modules and a single postmortem module fromMorabito et al with nonoverlapping genes falling within anunassigned (gray) module (eTable 6 linkslwwcomNXGA456 and eTable 7 linkslwwcomNXGA457) Impor-tantly the overlapping postmortem module was enrichedwith Alzheimer disease GWAS association signals as well asmicroglial cell markers and immune-related biological path-ways These data suggest that gene expression from non-diseased brain tissues coupled with imputation of geneticallyregulated gene expression may capture gene coexpressionnetworks underlying Alzheimer disease

A Computational Drug Repurposing AnalysisIdentifies Drug Compounds forFurther AnalysisOur tissue-specific gene coexpression modules provide auseful substrate for the identification and prioritization ofdrugs that may normalize altered gene coexpression in Alz-heimer disease We used S-PrediXcan to identify genes whosegenetically regulated expression is associated with geneticvariation underlying Alzheimer disease (eTable 8 linkslwwcomNXGA471) The use of genetically regulated geneexpression removed unwanted environmental effects on geneexpression (eg medication use) and thereby mitigates re-verse causation We assigned the S-PrediXcan Z score for thedirection and magnitude of effect to all genes within Alz-heimer disease risk modules and generated lists of upregulatedand downregulated genes The gene lists were used as input tothe CMap which computes a connectivity score based on thetranscriptional relationship between the gene lists and ob-served drug-gene signatures across multiple cell types Table 2provides a summary of the number of upregulated anddownregulated genes uploaded to CMap and the number andproportion of drug compounds with significant connectivityscores by brain tissue (a full list of compounds with significantconnectivity scores [score le -90] is provided in eTable 9linkslwwcomNXGA458)

To identify drug categories associated with Alzheimer diseasewe first assigned each drug compound to a MOA category(Table 3 eTable 10 linkslwwcomNXGA459) We thentested for the enrichment of nominally significant (p lt 005)gene-based associations (MAGMA and S-PrediXcan) acrosscompounds within each category Two categoriesmdashglutamate receptor antagonists and dopamine receptorantagonistsmdashharbored a larger proportion of nominally sig-nificant p values for Alzheimer disease than expected bychance (Figure 3) We extracted top-ranked compounds (bymean connectivity score across all tissues and cell types)within significant MOAs and plotted the distribution of theirconnectivity scores (Figure 4) The top 3 glutamate receptorantagonists included memantine commonly used to treat the

symptoms of moderate-to-severe Alzheimer disease anddextromethorphan a compound with clinical efficacy for thetreatment of agitation associated with Alzheimer diseasehighlighting the utility of our approach Top dopamine re-ceptor antagonists included a number of antipsychotics (egziprasidone) that are used to treat aggression and behavioralissues in Alzheimer disease (Table 4)27 We also examinedhow drugs that have undergone a clinical trial for Alzheimerdisease andor its associated symptoms performed in therepositioning pipeline Although their respective MOA cate-gories were not enriched with Alzheimer disease associationsignals top-ranked drug compounds that have undergone aclinical trial included naproxen (cyclooxygenase inhibitor)mirtazapine (serotonin receptor antagonist) and caffeine(phosphodiesterase inhibitor) (eTable 11 linkslwwcomNXGA460)

To assess the significance of drug-gene level results we ap-plied a permutation procedure (methods) using expressiondata from the amygdalamdashthe tissue with the largest numberof drug-gene associations The results show that top-rankedcompounds are unlikely to be due to correlated expression(eTable 12 linkslwwcomNXGA461) We also ran ournetwork-based pipeline with GWAS summary statistics forschizophrenia a brain-related disorder with an immunecomponent that is not genetically correlated with Alzheimerdisease as a negative control and found no significant overlap(hypergeometric test) with our observed results for Alzheimerdisease across cell types (eTable 13 linkslwwcomNXGA462) These observations strengthen the candidacy of po-tential Alzheimer therapeutics and illustrate the potential ofCMap within a gene coexpression network framework togenerate novel unbiased hypotheses on the pharmacologicmodulation of disease states

DiscussionWe developed a novel computational drug repositioning ap-proach based on the integration of SNP genotype tissue-specific gene coexpression patterns and drug perturbationdata Computational drug repositioning provides a bi-ologically valid approach to evaluate the predicted effect ofdrug compounds on cellular activity We applied a tissue-specific network-based gene coexpression method to identifygroups of highly correlated functionally related genes asso-ciated with Alzheimer disease Gene-based analyses of GWASsummary statistics were enriched in a single gene module in13 brain tissues each of which contained genes involved in theimmune system and immune response A computational drugrepositioning analysis of genes within these tissue-specificAlzheimer modules identified drugs and broader mechanismsof action categories Some of the identified compounds havebeen approved to treat Alzheimer disease and its associatedsymptoms (eg memantine) We also provide a list of plau-sible novel drug candidates for functional validation studiesOur results demonstrate that a tissue-specific approach to

NeurologyorgNG Neurology Genetics | Volume 7 Number 5 | October 2021 11

gene discovery in Alzheimer disease may not only identifycandidate causal genes tissues and biological pathways butalso targets for therapeutic intervention

Neuroinflammation has an important role in the onset andprogression of the pathologic changes underlying Alzheimerdisease Independent studies have identified immune-relatedproteins and cells in the proximity of β-amyloid plaques28 forexample and epidemiologic reports suggested that anti-inflammatory agents used to treat immune disorders such asrheumatoid arthritis decrease the risk of Alzheimer disease29

It was not until the publication of a large-scale GWAS onAlzheimer that the first robust evidence for a causal associa-tion between neuroinflammation and disease onset wasestablished2 We performed a secondary analysis of theGWAS results and the study of tissue-specific gene coex-pression patterns allowed us to investigate a larger set of genesthat might be implicated in disease based on network con-nectivity We identified immune systemndashrelated tissue-specific modules (ie groups) of coexpressed genes that areboth enriched with Alzheimer disease association signals andstrongly preserved (ie replicated) across tissues

Our gene coexpression networks built frombrain gene expressiondata from healthy individuals showed good overlap with post-mortemAlzheimer disease samples whichwere also enrichedwithGWAS signals and immune systemndashrelated biological pathwaysThis suggests that coexpression patterns from nondiseased braintissue followed by the integration of disease-associated geneticallyregulated gene expressionmay be used to identify groups of geneswhose activity drives Alzheimer disease onset and progressionFurthermore the diversity of sampled brain regions betweenGTEx and the data sets used byMorabito which included samplesfrom the temporal cortex30 prefrontal cortex31 parahippocampalgyrus inferior frontal gyrus superior temporal gyrus and frontalpole32 suggests that genetically regulated changes underlyingAlzheimer disease may converge on neuroimmune networksoperating across brain regions Therefore meaningful biologicalinsights for Alzheimer diseasemay be derived from the integrationof disease-associated genetic data with large numbers of non-diseased and nonndashtissue-specific brain samples which are rela-tively easy to collect for a sufficiently powered study

We further demonstrated the versatility of coexpressionnetwork-based methods with the application of a novel com-putational drug repositioning approach where the imputed ef-fect direction in Alzheimer disease for all genes within disease-implicated modules was used as input to CMap This approachwas taken under the biologically valid assumption that a drugcompound not only alters the activity of a single target gene butthe activity of multiple related genes through coregulation33

Furthermore by imputing gene expression effects from GWASsummary statistics we focused only on genetically regulatedgene expression effects thereby removing unwanted variation ongene expression from environmental effects (eg medicationuse) as well as controlling for reverse causation12 This approachidentified drug compounds within disease relevant mechanisms

of action that are predicted to normalize the expression of can-didate causal genes in Alzheimer disease

We identified 2 drugMOA categories enriched with smaller thanexpected Alzheimer disease association signals glutamate re-ceptor antagonists and dopamine receptor antagonists Gluta-mate is present in higher levels in patients with Alzheimerdisease34 Increased glutamate in the brain is widely thought topromote neurotoxicity and neurodegeneration andmay also mayalso trigger neuroinflammation in (genetically) susceptible indi-viduals35 Genetic studies strongly support a role of the immunesystem in Alzheimer disease pathophysiology36 and risk genesare highly expressed in microgliamdashthe primary immune cells ofthe brain3738 Therefore glutamate may be an important linkbetween the nervous and immune systems in Alzheimer diseaseonset and progression with a central role for microglial cellshowever a functional mechanism has yet to be established Do-pamine receptor modulators are also used to treat some of thesymptoms of Alzheimer disease such as agitation and psycho-sis39 however a clear role of dopamine in Alzheimer patho-physiology has yet to be established Nonetheless dopamine is animportant regulator of immune function and response in thebrain40 and microglial cells express functional dopamine recep-tors Therefore dopamine may also play a role in the crosstalkbetween immune and nervous systems in Alzheimer disease

Top-ranked glutamate receptor antagonists included mem-antine which is approved to treat the symptoms of moderate-to-severe Alzheimer disease41 and dextromethorphan which isused to treat agitation associated with Alzheimer disease39

Other highly ranked selective antagonists of glutamate recep-tors with potential neuroprotective effects included ifenprodilshown to ameliorate amyloid-β induced inhibition of synaptictransmission and hippocampal dysfunction42 and flupirtine awell-tolerated nonopioid analgesic drug with potential neuro-protective effects43 Highly ranked dopamine receptor antag-onists included methylergometrine which has been shown toinhibit inflammasome degranulation under proinflammatoryconditions44 with potential therapeutic benefits for diseaseswith an inflammatory component such as Alzheimer diseaseWhile these drugs alleviate the symptoms of Alzheimer diseaseit is not known if they target a causal gene or biologicalmechanism Our results suggest that these drugs may indeedtarget a causal mechanism given that the drug-gene signatureswere derived from genetic data from Alzheimer disease casesand controls The use of drug perturbation data may thereforerefine our understanding of gene mechanisms underlyingAlzheimer disease in addition to potential therapeutic targets

Genetic associations for Alzheimer disease are enriched ingenomic regions that encode druggable gene targets45

Computational repositioning is therefore a promising avenuefor the translation of genetic associations to drug targets Soet al46 used GWAS-imputed transcriptome profiles and theCMap algorithm to identify candidates for drug repositioningin neuropsychiatric disorders and identified several non-steroidal anti-inflammatory drugs (NSAIDs) also known as

12 Neurology Genetics | Volume 7 Number 5 | October 2021 NeurologyorgNG

cyclooxygenase inhibitors with possible benefits in Alzheimerdisease We identified several NSAIDs with promising con-nectivity score distributions across cell types and tissues in-cluding (for example) naproxen which has shown mixedresults for their protective effect from Alzheimer disease4748

However we did not find enrichment of significant gene-based association signals for Alzheimer disease within knowngene targets of NSAIDs We extend the analysis of So et alwith the use of a larger more highly powered GWAS andnetwork-based methods to implicate additional genes withpotential relevance in Alzheimer disease pathology A recentlypublished study used genetic information and network-basedmethods to develop a priority index for drug target validationin immune-mediated traits15 The priority index incorporatedfunctional genomic information with protein-protein networkconnectivity information and was shown to successfullyidentify current therapeutics and prioritize alternative com-pounds for early-stage testing Their network annotationshowever do not directly integrate genetic coexpression butinstead rely on disparate sources of protein interaction data tocharacterize gene connectivity Our gene coexpression-basedapproach on the other hand directly anchors changes ingenetically regulated gene expression to observed levels ofcoexpression between genes and arguably more closely rep-resents underlying biological relationships

The results of this study should be interpreted in view of thefollowing limitations First we used GWAS summary statisticsthat included Alzheimer diseasendashby-proxy cases Although proxycases based on parental diagnoses show a high genetic corre-lationwith clinically diagnosed cases follow-up studies should beconducted using GWAS summary statistics from diagnosedAlzheimer disease studies36 Second we used gene expressiondata from bulk human brain tissue as single-cell expression dataare not available in GTEx Bulk brain tissue is not homogeneouswith respect to individual cell types As a result true Alzheimerdisease association signals may be diluted by nonspecific ex-pression or expression differences may simply reflect mosaiceffects of different cell types This is especially problematic forAlzheimer disease where many of the risk genes are highlyexpressed inmicroglia cells which only account for around 3 ofthe total brain cell population49-51 RNA sequencing of in-dividualized cells (known as single-cell RNA sequencing) maypartition genetic signals to causal cell types and improve powerto identify functional genes and mechanisms underlying Alz-heimer disease and in turn improve the accuracy of drug po-sitioning9 Third the CMap database does not contain drug-gene signatures for every approved and experimental drug Im-portantly 3 FDA acetylcholinesterase inhibitors used to treat thesymptoms of Alzheimer disease (galantamine rivastigmine anddonepezil) were missing from CMap The absence of thesecompounds may explain why we did not observe significantenrichment of Alzheimer disease association signals within theacetylcholinesterase inhibitors MOA category Fourth our ob-served drug compound associations are derived from the com-bined effect of gene expression patterns within Alzheimer diseasemodules Therefore our analytical pipeline cannot identify

individual drug-gene pairs that drive the observed drug com-pound enrichments in coexpression networks Finally CMaponly contains drug-gene signatures from cultured human cancercell lines A comparison of drug compounds profiled in neuronalprogenitor cells and differentiated neurons with 9 cancer celllines profiled inCMap suggested that neuronal cell lines producedifferent connectivity patterns in a subset of compounds52 Ourpipeline would therefore benefit from the inclusion of drug-genesignatures derived from neuronal and glial (astrocytes microgliaand oligodendrocytes) cell lines as they become available

In summary we developed a novel drug repositioning approachthat first identifies gene networks associated with Alzheimerdisease before using the implicated networks as a molecularsubstrate for a drug repositioning analysis Our approachidentified drug mechanisms of action categories containingdrug compounds with both proven and potential therapeuticbenefit in Alzheimer disease We identified the drug mem-antine which is 1 of only 4 FDA-approved drugs for Alzheimerdisease supporting the validity of our approach Follow-upmolecular studies will seek to validate prioritized drug candi-dates in relevant human cell models such as monocyte-inducedmicroglia Our approach will help researchers leverage geneticdata for drug discovery and development in Alzheimer disease

AcknowledgmentThe authors thank the International Genomics of AlzheimerrsquosProject (IGAP) for providing Alzheimer disease meta-analysissummary results data for these analyses The investigatorswithin IGAP contributed to the design and implementation ofIGAP andor provided data but did not participate in analysisor writing of this report IGAP was made possible by thegenerous participation of the controls the patients and theirfamilies ERG is grateful to the President and Fellows of ClareHall University of Cambridge for the stimulating intellectualenvironment during his fellowship in the college ERG issupported by the National Human Genome Research Instituteof the National Institutes of Health under Award NumbersR35HG010718 and R01HG011138 The content is solely theresponsibility of the authors and does not necessarily representthe official views of the National Institutes of Health

Study FundingNo targeted funding reported

DisclosureZF Gerring A White and EM Derks received seed fundingfrom the Takeda Pharmaceutical Company related to this pro-ject The Takeda Pharmaceutical Company was not involved instudy design or data analysis or in the writing and preparation ofthe manuscript ER Gamazon reports no disclosures relevant tothe manuscript Go to NeurologyorgNG for full disclosures

Publication HistoryReceived byNeurology Genetics January 31 2021 Accepted in final formJuly 13 2021 This manuscript was prepublished in doiorg101101853580

NeurologyorgNG Neurology Genetics | Volume 7 Number 5 | October 2021 13

References1 Zhang Q Sidorenko J Couvy-Duchesne B et al Risk prediction of late-onset Alz-

heimerrsquos disease implies an oligogenic architecture Nat Commun 20201147992 Jansen IE Savage JE Watanabe K et al Genome-wide meta-analysis identifies new

loci and functional pathways influencing Alzheimerrsquos disease risk Nat Genet 201951(3)404-413

3 Gerring ZF Lupton MK Edey D Gamazon ER Derks EM An analysis of geneticallyregulated gene expression across multiple tissues implicates novel gene candidates inAlzheimerrsquos disease Alzheimers Res Ther 202012(1)43

4 Di Paolo G Kim T-W Linking lipids to Alzheimerrsquos disease cholesterol and beyondNat Rev Neurosci 201112(5)284-296

5 Nelson MR Tipney H Painter JL et al The support of human genetic evidence forapproved drug indications Nat Genet 201547(8)856-860

6 Wang D Liu S Warrell J et al Comprehensive functional genomic resource andintegrative model for the human brain Science 2018362(6420)eaat8464

7 Mele M Ferreira PG Reverter F et al The human transcriptome across tissues andindividuals Science 2015348(6235)660-665

8 Mathys H Davila-Velderrain J Peng Z et al Single-cell transcriptomic analysis ofAlzheimerrsquos disease Nature 2019570(7761)332-337

9 Grubman A Chew G Ouyang JF et al A single cell brain atlas in human Alzheimerrsquosdisease Biorxiv 2019 doi 101101628347

10 Langfelder P Horvath S WGCNA an R package for weighted correlation networkanalysis BMC Bioinformatics 20089559

11 Gerring ZF Gamazon ER Derks EM A gene co-expression network-based analysis ofmultiple brain tissues reveals novel genes and molecular pathways underlying majordepression PLoS Genet 201915(7)e1008245

12 Gamazon ER Wheeler HE Shah KP et al A gene-based association method formapping traits using reference transcriptome data Nat Genet 201547(9)1091-1098

13 Gusev A Ko A Shi H et al Integrative approaches for large-scale transcriptome-wideassociation studies Nat Genet 201648(3)245-252

14 Zhang B Gaiteri C Bodea L-G et al Integrated systems approach identifies geneticnodes and networks in late-onset Alzheimerrsquos disease Cell 2013153(3)707-720

15 Fang H Beckmann G Bountra C et al A genetics-led approach defines the drugtarget landscape of 30 immune-related traits Nat Genet 201951(7)1082-1091

16 Lamb J Crawford ED Peck D et al The connectivity map using gene-expressionsignatures to connect small molecules genes and disease Science 2006313(5795)1929-1935

17 Langfelder P Zhang B Horvath S Defining clusters from a hierarchical cluster treethe Dynamic Tree Cut package for R Bioinformatics 200824(5)719-720

18 Morabito S Miyoshi E Michael N Swarup V Integrative genomics approach iden-tifies conserved transcriptomic networks in Alzheimerrsquos disease Hum Mol Genet202029(17)2899-2919

19 de Leeuw CA Mooij JM Heskes T Posthuma D MAGMA generalized gene-setanalysis of GWAS data PLoS Comput Biol Public Libr Sci 201511(4)e1004219

20 Reimand J Arak T Adler P et al gProfilermdasha web server for functional interpretationof gene lists (2016 update) Nucleic Acids Res 201644(W1)W83ndashW89

21 Langfelder P Luo R Oldham MC Horvath S Is my network module preserved andreproducible PLoS Comput Biol 20117(1)e1001057

22 Delaneau O Marchini J 1000 Genomes Project Consortium Integrating sequenceand array data to create an improved 1000 Genomes Project haplotype referencepanel Nat Commun 201453934

23 Wishart DS Feunang YD Guo AC et al DrugBank 50 a major update to theDrugBank database for 2018 Nucleic Acids Res 201746(D1)D1074-D1082

24 Cotto KC Wagner AH Feng Y-Y et al DGIdb 30 a redesign and expansion of thedrugndashgene interaction database Nucleic Acids Res 201746(D1)D1068-D1073

25 Pardintildeas AF Holmans P Pocklington AJ et al Common schizophrenia alleles areenriched in mutation-intolerant genes and in regions under strong background se-lection Nat Genet 201850(3)381-389

26 Siminoff LAWilson-GendersonM Gardiner HMMosavel M Barker KL Consent to apostmortem tissue procurement study distinguishing family decision makersrsquo knowl-edge of the genotype-tissue expression project Biopreserv Biobank 201816(3)200-206

27 Tible OP Riese F Savaskan E von Gunten A Best practice in the management ofbehavioural and psychological symptoms of dementia Ther Adv Neurol Disord 201710(8)297-309

28 Minter MR Taylor JM Crack PJ The contribution of neuroinflammation to amyloidtoxicity in Alzheimerrsquos disease J Neurochem 2016136(3)457-474

29 Chou RC Kane M Ghimire S Gautam S Gui J Treatment for rheumatoid arthritisand risk of Alzheimerrsquos disease a nested case-control analysis CNS Drugs 201630(11)1111-1120

30 Allen M Carrasquillo MM Funk C et al Human whole genome genotype andtranscriptome data for Alzheimerrsquos and other neurodegenerative diseases Sci Data20163160089

31 Mostafavi S Gaiteri C Sullivan SE et al A molecular network of the aging humanbrain provides insights into the pathology and cognitive decline of Alzheimerrsquos dis-ease Nat Neurosci 201821(6)811-819

32 WangMBeckmannND Roussos P et al TheMount Sinai cohort of large-scale genomictranscriptomic and proteomic data in Alzheimerrsquos disease Sci Data 20185180185

33 Plenge RM Priority index for human genetics and drug discovery Nat Genet 201951(7)1073-1075

34 Wang R Reddy H Role of glutamate and NMDA in Alzheimerrsquos disease J AlzheimersDesese 201757(4)1041-1048

35 Heneka MT Carson MJ Khoury J El et al Neuroinflammation in Alzheimerrsquos dis-ease Lancet Neurol 201514(4)388-405

36 Kunkle BW Grenier-Boley B Sims R et al Genetic meta-analysis of diagnosedAlzheimerrsquos disease identifies new risk loci and implicates Aβ tau immunity and lipidprocessing Nat Genet 201951(3)414-430

37 Spangenberg E Severson PL Hohsfield LA et al Sustained microglial depletion withCSF1R inhibitor impairs parenchymal plaque development in an Alzheimerrsquos diseasemodel Nat Commun 201910(1)3758

38 Ulland TK Colonna M TREM2mdasha key player in microglial biology and Alzheimerdisease Nat Rev Neurol 201814(11)667-675

39 Cummings JL Lyketsos CG Peskind ER et al Effect of dextromethorphan-quinidineon agitation in patients with Alzheimer disease dementia a randomized clinical trialJAMA 2015314(12)1242-1254

40 Matt SM Gaskill PJ Where is dopamine and how do immune cells see it dopamine-mediated immune cell function in Health and disease J Neuroimmune Pharmacol202015(1)114-164

41 Reisberg B Doody R Stoffler A Schmitt F Ferris S Mobius HJ Memantine inmoderate-to-severe Alzheimerrsquos disease N Engl J Med 2003348(14)1333-1341

42 Costa RO Lacor PN Ferreira IL et al Endoplasmic reticulum stress occurs down-stream of GluN2B subunit of N-methyl-d-aspartate receptor in mature hippocampalcultures treated with amyloid-β oligomers Aging Cell 201211(5)823-833

43 Klawe C Maschke M Flupirtine pharmacology and clinical applications of a non-opioid analgesic and potentially neuroprotective compound Expert Opin Pharmac-other 200910(9)1495-1500

44 Garcıa-Laınez G SanchoM Garcıa-Bayarri V Orzaez M Identification and validationof uterine stimulant methylergometrine as a potential inhibitor of caspase-1 activationApoptosis 201722(10)1310-1318

45 Finan C Gaulton A Kruger FA et al The druggable genome and support for targetidentification and validation in drug development Sci Transl Med 20179(383)eaag1166

46 So H-C Chau CK-L Chiu W-T et al Analysis of genome-wide association datahighlights candidates for drug repositioning in psychiatry Nat Neurosci 201720(10)1342-1349

47 Aisen PS Schafer KA Grundman M et al Effects of rofecoxib or naproxen vs placeboon Alzheimer disease progression a randomized controlled trial JAMA 2003289(21)2819-2826

48 Lyketsos CG Breitner JCS Green RC et al Naproxen and celecoxib do not preventAD in early results from a randomized controlled trial Neurology 200768(21)1800-1808

49 Kinney JW Bemiller SM Murtishaw AS Leisgang AM Salazar AM Lamb BT In-flammation as a central mechanism in Alzheimerrsquos disease Alzheimers Dement (NewYork N Y) 20184575-590

50 Pimenova AA Raj T Goate AM Untangling genetic risk for Alzheimerrsquos disease BiolPsychiatry 201883(4)300-310

51 Efthymiou AG Goate AM Late onset Alzheimerrsquos disease genetics implicatesmicroglial pathways in disease risk Mol Neurodegener Biomed Cent 201712(1)43

52 Subramanian A Narayan R Corsello SM et al A next generation connectivity mapL1000 platform and the first 1000000 profiles Cell 2017171(6)1437-1452e17

Appendix Authors

Name Location Contribution

Zachary FGerringPhD

QIMR Berghofer MedicalResearch Institute BrisbaneQueensland Australia

Designed and conceptualizedthe study analyzed the datainterpreted the data anddrafted the manuscript forintellectual content

Eric RGamazonPhD

Vanderbilt UniversityMedical Centre NashvilleTN

Interpreted the data andrevised the manuscript forintellectual content

AnthonyWhitePhD

QIMR Berghofer MedicalResearch InstituteQueensland Australia

Interpreted the data andrevised the manuscript forintellectual content

Eske MDerksPhD

QIMR Berghofer MedicalResearch InstituteQueensland Australia

Designed and conceptualizedthe study and revised themanuscript for intellectualcontent

14 Neurology Genetics | Volume 7 Number 5 | October 2021 NeurologyorgNG

DOI 101212NXG000000000000062220217 Neurol Genet

Zachary F Gerring Eric R Gamazon Anthony White et al Repositioning Candidates for Alzheimer Disease

Integrative Network-Based Analysis Reveals Gene Networks and Novel Drug

This information is current as of September 13 2021

ServicesUpdated Information amp

httpngneurologyorgcontent75e622fullhtmlincluding high resolution figures can be found at

References httpngneurologyorgcontent75e622fullhtmlref-list-1

This article cites 52 articles 5 of which you can access for free at

Subspecialty Collections

httpngneurologyorgcgicollectiongene_expression_studiesGene expression studies

httpngneurologyorgcgicollectionassociation_studies_in_geneticsAssociation studies in genetics

httpngneurologyorgcgicollectionalzheimers_diseaseAlzheimers diseasefollowing collection(s) This article along with others on similar topics appears in the

Permissions amp Licensing

httpngneurologyorgmiscaboutxhtmlpermissionsits entirety can be found online atInformation about reproducing this article in parts (figurestables) or in

Reprints

httpngneurologyorgmiscaddirxhtmlreprintsusInformation about ordering reprints can be found online

reserved Online ISSN 2376-7839Published by Wolters Kluwer Health Inc on behalf of the American Academy of Neurology All rightsan open-access online-only continuous publication journal Copyright Copyright copy 2021 The Author(s)

is an official journal of the American Academy of Neurology Published since April 2015 it isNeurol Genet