Human Mutation

Department of Epidemiology, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands Department of Genetics, School of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran Department of Psychiatry, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands Department of Gastroenterology, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands Department of Clinical Genetics, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands Department of Radiology and Nuclear Medicine, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands Department of Lung Matrix Remodeling, Excellence Cluster Cardio‐Pulmonary System (ECCPS), University Justus Liebig Giessen, Giessen, Germany Center for Biomics, Department of Cell Biology, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands Department of Immunology, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands

include TGFBR1 and PICALM, of which their derepression in the brain due to reduced expression levels of miR-142-3p may reduce the risk of AD.

| INTRODUCTION
Alzheimer disease (AD) is the most common age-related neurodegenerative disease worldwide manifested by the progressive loss of memory and cognitive decline (Brookmeyer, Johnson, Ziegler-Graham, & Arrighi, 2007). Enormous efforts have been made over the past decades to discover risk factors for developing AD and to identify biomarkers for early diagnosis of the disease (Crews & Masliah, 2010;Hampel et al., 2010;Schmechel et al., 1993). The determinants of early-onset AD have been primarily associated with mutations in one of three genes: APP, PSEN1, and PSEN2 (Ertekin-Taner, 2007). In contrast, late-onset AD (after 65 years of age), the most common form of AD with a heritability of 60-80%, is a genetically heterogeneous disease (Gatz et al., 2006). In addition to apolipoprotein E (APOE) polymorphisms that explain approximately 25% of the heritability, more than 30 genetic loci have so far been established as contributing to late-onset AD risk (Jansen et al., 2019;Jun et al., 2017;Lambert et al., 2013). However, they explain only a fraction of the estimated heritability and the genetics of AD are yet to be fully understood (Zhou et al., 2018). To fully grasp the contribution of genetic factors to AD, we must go beyond classical genetics, and explore the multiple interacting layers that regulate the genome. This includes the analysis of not only the protein-coding sequences but the vast noncoding regions as well.
Recent developments in omics technologies have revealed the complexity of the human genome, displaying that protein-coding RNAs constitute only approximately 2% of the human transcriptome, highlighting the distinct possibility that noncoding RNAs (ncRNAs) might meaningfully contribute to human disease (Carninci et al., 2005;Consortium, 2012). Noncoding RNAs are functional RNA molecules that are transcribed from DNA but not translated into proteins. They  (Ma, Bajic, & Zhang, 2013). It has become increasingly evident that lncRNAs impact disease pathogenesis primarily through posttranscriptional regulation of gene expression (Rinn & Chang, 2012). Despite constituting the majority of noncoding transcriptome, few lncRNAs most notably BACE1-AS and BC200 have been so far characterized to play a role in the pathogenesis of AD to date (Modarresi et al., 2011;Mus, Hof, & Tiedge, 2007).
In the present study, we conducted a genome-wide scan to identify miRNAs and lncRNAs associated with AD by leveraging data from the largest available GWAS of late-onset AD (Jansen et al., 2019). We found several ncRNA loci significantly associated with AD, including a newly identified susceptibility locus on 17q22. We performed various in silico and in vitro studies to determine the functionality of ncRNA variant in this locus and to gain insight into the role of associated ncRNA in AD pathogenesis.

| Genome-wide association study on AD
Summary statistics data were retrieved from a recent large-scale GWAS meta-analysis of late-onset AD including 455,258 individuals of European ancestry, meta-analyzed in three phases (Jansen et al., 2019 (N proxy cases = 47,793; N proxy controls = 328,320). Phase 3 was the meta-analysis of phase 1 and 2, including 71,880 (proxy) AD cases and 383,378 (proxy) controls. More details about the consortia and participants are described elsewhere (Jansen et al., 2019). All participating studies in the AD GWAS had provided informed consent for participation in genetics studies and were approved by their local ethical committees.

| Genetic variants in noncoding RNAs
Genetic variants in human lncRNA transcripts were extracted using lncRNASNP, a comprehensive database including 495,729 SNPs in 32,108 lncRNA transcripts of 17,436 lncRNAs (Gong et al., 2014).
Moreover, as primary transcript of miRNAs has been suggested to be 3-4 kb in length (Saini, Griffiths-Jones, & Enright, 2007), we used dbSNP database (https://www.ncbi.nlm.nih.gov/SNP/) to extract 16,178 SNPs located in +/−2 kb of 1,318 mature miRNA sequences reported in miRBase v21 (http://www.mirbase.org/). We excluded SNPs with minor allele frequency (MAF) <0.01. Of the remaining SNPs, we analyzed the association with AD of 96,950 SNPs in 14,790 lncRNA transcripts and 12,404 SNPs in 1,237 primary miRNA transcripts that were present in the GWAS summary statistics data (Jansen et al., 2019). To obtain the number of independent SNPs, we used the LD based SNP pruning in PLINK (http://pngu.mgh.harvard. edu/~purcell/plink/), where we excluded the SNPs with R 2 > 0.7. The Bonferroni correction was used to adjust p-value for the number of tests (0.05/49,323 independent SNPs) and the significance threshold was set at 1.02 × 0 -6 . Regional plots showing the association of ncRNA SNPs and flanking variants in the corresponding loci with AD were generated by the LocusZoom web tool (Pruim et al., 2010).

| Assessing the biological functionality of noncoding RNA variants
For the ncRNA SNPs associated with AD, the LD region (R 2 > 0.7) was determined using the 1000 Genomes Phase 3. We investigated whether known protein-coding variants were in strong LD with the associated ncRNA SNPs. Further, we examined whether the associated SNPs in ncRNAs are annotated to regulatory features, including promoter and enhancer regulatory motifs, DNase footprinting sites, and conserved sequences using HaploReg (v4.1; Ward & Kellis, 2016). For each set of variants in strong LD with a given ncRNA SNP, we also investigated whether the SNP was located in a potential regulatory region using the Roadmap consortium reference epigenomes data set (Romanoski, Glass, Stunnenberg, Wilson, & Almouzni, 2015). To test the association of ncRNA SNPs with gene expression, we used expression quantitative trait loci (eQTL) data provided by GTEx (https://www.gtexportal.org/home/) and BBMRI-NL atlas (http://atlas.bbmrirp3-lumc. surf-hosted.nl/#query ;Bonder et al., 2017). The UCSC genome browser was used for visualization of the ncRNA SNP location in the genome. The ncRNA secondary structure and the effect of an SNP on the minimum free energy (MFE) of the predicted ncRNA structure were investigated using the Vienna RNA Package 2.0 (Lorenz et al., 2011).

| Plasmids, miRNA promoter constructs, and cell culture transfection
To compare the activity of miR-142 promoter containing either rs2526377 alleles, the full-length 589-bp fragment corresponding to the upstream region of the pri-miR-142 transcript was synthesized by Integrated DNA Technologies (IDT) and cloned into pGreenFire-CMV-EF1-puro (System Biosciences) digested with EcoR I and Spe I.
DNA sequencing verified all constructs. HEK293 cells were then used to generate the lentivirus with cotransfection of reporter gene vectors, HIV gag-pol and VSV-G in a ratio of 1:0.8:0.2. For transduction assay, cells were seeded into 24-well plates and transduced with lentiviral particles. With selection by puromycin at a concentration of 2 ug/ml, cells were calculated and seeded into 96well. After incubation for 24-hr, the cell supernatant was harvested and the luciferase activity was then measured on a luminometer (LB960; Berthold) using the Dual-Luciferase Reporter Assay System (Promega). The ratio of firefly luciferase to Renilla luciferase was calculated for each well. The experiments were performed five times.

| Quantitative real-time polymerase chain reaction
Total RNA from a human induced pluripotent stem cell (iPSC)derived neural progenitor cells (NPCs) and human brain cryopreserved sections were isolated using Trizol LS reagent (Invitrogen, Carlsbad, CA) according to the manufacturer's protocols. The concentration of total RNA was determined using a NanoDrop ND-1000 spectrophotometer (NanoDrop, Wilmington, DE). TaqMan qPCR Assays were performed according to the manufacturer's protocols (Applied Biosystems, Foster City, CA) to determine the expression levels of miR-142-3p, miR-142-5p, and BZRAP1-AS1. The assays were run using the Applied Biosystems 7900HT Real-Time PCR system. RNU6B was used as an internal control for miRNA expression analysis. All the experiments were performed in triplicates. The human frozen brain tissues (three gray matter and three white matter) were obtained from the Netherlands Brain Bank (Amsterdam, The Netherlands). All samples were free of neurological disease.

| Putative target genes of miR-142
TargetScan V7.1 (http://www.targetscan.org/; Agarwal, Bell, Nam, & Bartel, 2015) was used to identify the putative targets of miR-142-3p and -5p in human and mouse. This program predicts biological targets of miRNAs by searching for conserved 7/8-mer sites that match the miRNA seed region. The predictions are ranked based on the putative efficacy of targeting as calculated using context scores of the sites, the higher context score, the greater the probability that a miRNA could target a particular gene. For our analysis, we used the predicted targets that had a recommended context score <−0.01.
Further, we retrieved the list of putative target genes of miR-142 (3p and 5p) from two other widely-used online miRNA target prediction databases, miRtarget2 (Wang & El Naqa, 2008) and DIANA-microT (Paraskevopoulou et al., 2013). Then, RNA-Seq data from the Human Body Map 2.0 (Kawamoto et al., 2000) was used to check which of the miR-142 putative target genes are expressed in the human brain (fragments per kilobase million, FPKM ≥1), target genes not expressed in the brain were excluded.
Pathway analysis was performed using KEGG and IPA databases.
KEGG incorporates knowledge of known gene networks and identifies significantly enrichment of miRNA putative targets in these networks according to a t test (Subramanian et al., 2005). IPA is a knowledge database generated from peer-reviewed scientific publications that enable the discovery of highly represented biological mechanisms, pathways or functions most relevant to the genes of interest from large, quantitative datasets. We uploaded the list of the miR-142 target genes and performed a core analysis with GHANBARI ET AL.

| 2133
the default settings in IPA. We mapped the miRNA target genes to biological functions or canonical pathways to see whether they are enriched in specific networks. The p-values are calculated using the right-tailed Fisher's Exact Test and a p-value of less than .05 indicates a statistically significant, nonrandom association.

| RNA-Seq analysis in human iPSC-derived neural progenitor cells (NPCs)
NPCs derived from human control iPSCs (Sigma-Aldrich line iPSC0028) were cultured to 70% confluency in six-well plates (Corning) according to standard protocols (Shi, Kirwan, & Livesey, 2012). NPCs were transfected with 10 nM miRNA mimics (mirVanaTM Mimics, Thermo Fisher Scientific) including miR-142-3p, miR-142-5p, and the standard negative control #1 (Catalog nr. 4464060), or without any mimic (untreated). Transfections were performed using X-treme GENE ™ transfection reagent (Merck) according to the manufacturer's instructions. The experiment was run in triplicate. Total RNA was isolated 72 hr after transfection from the four groups of NPC samples using the RNeasy mini kit (74104; Qiagen). The RNA quality was checked by Agilent's 2100 Bioanalyzer (using Eukaryote Total RNA Nano kit). RNA-Seq analysis was performed at ErasmusMC Center for Biomics to test the changes in gene expression pattern in NPCs after overexpression of either miR-142-3p or -5p compared to controls.

| RNA-Seq analysis in the hippocampus of miR-142 KO mice and wt littermates
MiR-142 in mice is located on chromosome 11 and in the vicinity of the second exon belonging to Mir142hg (ENSMUSG00000084796).
The miR-142 −/− knockout mouse is a model with complete deletion of miR-142, with a significant decrease in the expression levels of both miR-142-3p and -5p isoforms (Shrestha et al., 2015). The expression of Bzrap1, a gene immediately flanking miR-142 is not altered in the miR-142-null mice, whereas the expression of a long Noncoding RNA (Mir142hg) embedded within the miR-142 gene is decreased. We confirmed the homo-and heterozygosity of miR-142 KO mice by qPCR. The hippocampi of miR-142 −/− , miR-142 −/+ and miR-142 +/+ littermates (n = 4 for each group, age 13-14 weeks, gender-balanced in all groups) were collected and total RNA was extracted using the RNeasy lipid tissue kit (74804; Qiagen). RNA quality was checked by Agilent's 2100 Bioanalyzer (using Eukaryote Total RNA Nano kit).

RNA-Seq analysis was performed at ErasmusMC Center for Biomics
to identify target genes of mmu-miR-142a-3p and mmu-miR-142a-5p

| Data analysis and statistics
The nonparametric Mann-Whitney test was used to compare miR-142 expression between the genotype groups, and an unpaired/ independent t test was used to compare reporter gene activities. In vitro experiments were repeated at least three times and histograms represent mean ± standard deviation. Statistical differences were measured using unpaired two-sided Student's t-test. p < .05 was considered statistically significant. Data analysis was performed using Excel Software Version 14.4.5.

RNA-Seq was performed with the Illumina TruSeq Stranded mRNA
Library prep kit. The resulting DNA libraries were sequenced on the HiSeq. 2500, for single-end reads of 50 bp length. Reads were generated of 50 base-pairs in length. Reads were mapped against the GRCm38 reference genome using HiSat2 (version 2.0.4; Kim, Langmead, & Salzberg, 2015). We called gene expression values (reads per gene) using htseq-count (version 0.6.1; Anders, Pyl, & Huber, 2015). We took only expressed genes into account, genes with at least five reads in at least seven samples (half of the samples plus one). This filtering in iPSCderived NPCs resulted in 17,181 genes. Differential expression analysis of the RNA-Seq data in human iPSC-derived NPCs was performed using R (version 3.3.2) and DESeq. 2 (version 1.14.1; Cuypers et al., 2017;Love, Huber, & Anders, 2014;Tian et al., 2011). Briefly, DESeq. 2 generated three values for each gene that were used for subsequent analysis: (a) Log2 fold change (Log2FC), (b) p-value based on the Wald test, and (c) Corrected p-value controlling the false discovery rate to 5%.
Genes were considered differentially expressed if the corrected p-value was lower than .05. For the biological interpretation of the results, we placed an additional cut-off of fold change ≥1.2.

| Noncoding RNAs associated with AD
In total, we examined the associations of 108,862 unique SNPs in primary miRNA sequences and lncRNAs with AD. Of these, SNPs annotated to 5 miRNAs and 10 lncRNAs, located in 7 distinct loci (each locus defined as 1 Mb), exceeded the significance threshold (p < 1.02 × 10 −6 ; Table 1). We assessed whether the associated ncRNA SNPs are likely to be functional in their corresponding loci based on a set of criteria recommended by previous studies to assess the potential functionality of ncRNA SNPs in GWAS results (Ghanbari et al., 2018;Ryan, Robles, & Harris, 2010). These criteria include an established association between SNP and the trait, the correlation of SNP with an expression of the host ncRNA, the localization of SNP in the ncRNA regulatory regions and the potential of SNP for structural perturbations in the host ncRNA.
As shown in the regional association plots ( Figure S1), in six of the seven identified noncoding RNA loci were coding variants in strong LD (R 2 > 0.7) with the ncRNA SNPs more significantly associated with AD (Table S1) (Table S2 and Figure S2), which their bindings to the miR-142 promoter might be perturbed by the SNP. Moreover, the eQTL data from the BBMRI-NL consortium showed that the rs2632577 minor allele is associated with lower miR-142 expression levels in the blood (p = 4.84 × 10 -11 ; Z-score = −6.58).
As an additional analysis, we tested whether rs2526377 is connected to cognitive functioning before the clinical manifestations of AD so that the variant can be used as early marker of disease. To this end, we examined the association of rs2526377 with cognitive ability and educational attainment using the publicly available GWAS data (Davies et al., 2018;Okbay et al., 2016). The SNP minor allele (G) was positively associated with cognitive function (p = .046; β = 0.011) and educational attainment (p = .005; β = 0.01), which is consistent with the protective effect of the G allele for AD risk.
Together, these data indicate that rs2526377 fulfilled predefined criteria for being a functional variant in the 17q22 locus, which possibly functions by altering the expression of miR-142. We  Note: Shown are the top variants in 5 miRNAs and 10 lncRNAs (located in seven distinct loci), exceeding the significance threshold (p < 1.02 × 10 −6 ) to be associated with AD. The associations are based on the data from a meta-analysis of phase 1 and 2 of the recent AD GWAS (Jansen et al., 2019

| Potential miR-142 target genes implicated in AD
A miRNA and its target genes should be expressed in the target tissue for any biological function to be exerted. Thus, we first measured the expression of miR-142 in the human brain. Both strands of miR-142 were expressed in the brain, miR-142-3p with an average Ct-value of 26, and miR-142-5p with an average Ct-value of 30, relative to the endogenous control RNU6B with an average Ct-value of 21.5 (Table   S3). To identify target genes that could mediate the function of miR-142 in the brain, we compiled a list of all putative target genes of miR-142-3p and -5p from three miRNA target prediction databases (TargetsScan, miRDB, and DIANA-microT). This resulted in 885 putative target genes for miR-142-3p and 1,541 putative target genes for miR-142-5p (Figure 4). We filtered these target genes on the basis of human brain expression, using the Illumina's Human Body Map RNA-Seq data set. We focused our analysis on the 814 brain-expressed putative target genes of miR-142-3p and 1,393 of miR-142-5p. To examine the regulatory effect of miR-142-3p and -5p on the expression levels of their putative target genes, we used human iPSC-derived neural progenitor cells (NPCs). We overexpressed either miR-142-3p or -5p in NPCs using mirVana™ miRNA Mimics and performed differential expression analysis by RNA-Seq. To elucidate miR-142 target genes implicated in the AD pathogenesis, we applied the two most commonly used methods for detecting miRNA targets.
First, we conducted a hypothesis-free differential expression analysis considering all brain-expressed target genes of miR-142. We sought to identify target genes that were significantly downregulated after overexpression of the mature miRNA (3p or 5p) with FDRadjusted p < .05 and fold change ≥ 1.2. Of the 814 brain-expressed putative target genes of miR-142-3p, 280 genes were significantly downregulated by the miR-142-3p mimic versus untreated, and 74 genes were significantly downregulated in NPCs transfected with miR-142-3p mimic versus negative control (Table S4 and Table 2).
We performed the KEGG pathway analysis for the 74 identified target genes of miR-142-3p and observed significant enrichment in Second, we examined the association of the 814 putative target genes of miR-142-3p and 1,393 putative target genes of miR-142-5p with AD using a candidate gene approach. To this end, we extracted genetic variants located in these target genes and tested their associations with AD using the GWAS data (Jansen et al., 2019).
Moreover, PICALM, the target gene found to be significantly associated with AD in the GWAS data, was upregulated in the hippocampus of miR-142 KO mice versus Wt littermates (p = 1.5 × 10 −2 , fold change = 1.13) and heterozygous KO mice versus Wt littermates (p = 7.2 × 10 −4 , fold change = 1.2; Figure 5). Luciferase reporter assay was performed to determine the effect of rs2526377 on the miR-142 promoter activity. The reporter gene constructs containing either the SNP alleles were generated and HEK293 cells were transfected with the reporter plasmids. The construct carrying the major allele A of rs2526377 had 20% higher basal activity in HEK293 cells than the construct carrying the minor allele G (p = .037). Error bars represent standard deviation (SD). *p < .05 compared with the control group (Student's t-test). NS, nonsignificant; SNP, single nucleotide polymorphism associations with human disease. In this study, we performed a genome-wide scan to systematically investigate the association of miRNAs and lncRNAs with AD by leveraging publicly available GWAS summary statistics (Jansen et al., 2019). We found seven distinct ncRNA loci significantly associated with AD including a newly identified susceptibility locus on 17q22, in which the ncRNA variant leads the signal and fulfills predefined criteria for being functional.

| DISCUSSION
The locus has not been reported as significant in the original GWAS, because the p-value of the top SNP in the meat-analysis of phase 1 (AD case/control) and phase 2 (AD-by-proxy) was above the GWAS threshold (Jansen et al., 2019). However, the SNP exceeds the GWAS threshold in phase 1 of this GWAS meta-analysis (p = 1.42 × 10 −9 ), combining data from the two large-scale AD case/control consortia, IGAP and PGC-ALZ. In phase 2, using the AD-by-proxy phenotype from the UK biobank cohort, the association between rs2632516 and AD is less significant (p-value = 5.0 × 10 −3 ), but still in the same direction. The lower association signal for the 17q22 locus in the UK biobank cohort could be explained by differences in case ascertainment of AD. In the UK biobank, Alzheimer dementia is ascertained via self-report information from family history (parent or a firstdegree relative with AD or dementia) as a proxy-phenotype for the participants (Marioni et al., 2018). This method relies on people to F I G U R E 4 Identification of potential target genes of miR-142 in the brain implicated in AD. We compiled a list of all putative targets of miR-142-3p and -5p from three widely used miRNA target prediction databases (miRDb, TargetScan, and DIANA-micro T). The target genes found to be expressed in the human brain were included. We overexpressed miR-142 (3p or 5p) in human iPS-derived NPCs and performed RNA-Seq to examine the changes in gene expression pattern. Target genes significantly downregulated (FDR-adjusted p < .05 and fold change ≥ 1.2) in NPCs transfected with mature miR-142 mimic (3p or 5p) versus untreated and negative control were retrieved. Subsequently, RNA-Seq was performed on the hippocampus of miR-142 KO mice and their wild-type littermates to confirm miR-142-mediated regulation of the identified target genes in human iPS-derived NPCs. AD, Alzheimer's disease; iPS, induced pluripotent stem cell; KO, knockout; NPC, neuronal progenitor cell GHANBARI ET AL.

| 2139
provide accurate information about whether their parents developed AD, for which misclassification of case status is of greater concern than consortia relying upon clinician-reported diagnoses. In addition, a transethnic GWAS, by adding more samples to the IGAP GWAS data, recently reported the significant association of 17q22 with AD (Jun et al., 2017). In this trans-ethnic GWAS, however, the leading ncRNA variant in the 17q22 locus was annotated to the closest protein-coding gene (BZRAP1), and the potential impact of miR-142 has been overlooked. While a more recent GWAS, investigating the association of rare coding variants with AD, with even larger sample size, did not find any significant association between rare variants in BZRAP1 gene and AD (Sims et al., 2017). In this line, our results demonstrated that miR-142 is the most likely functional target in the 17q22 locus implicated in AD pathogenesis.
Genetic variants in miRNA-encoding sequences have been shown previously to affect miRNAs expression and subsequently influence gene regulation in complex diseases (Dorn, Matkovich, Eschenbacher, & Zhang, 2012;Ghanbari, et al., 2016;Ghanbari et al., 2014). Moreover, the functional impact of variants on the promoter activity of miRNAs has been revealed, most notably for rs57095329 located in miR-146a, by altering the miRNA processing and expression level (Luo et al., 2011).
Here, we demonstrated that rs2526377 affects promoter activity and reduces the expression levels of miR-142. Previously, Skarn et al. (2013) characterized the miR-142 promoter region and demonstrated that DNA methylation of specific CpG sites in the region represses the promoter activity and reduces the expression level of miR-142 in mesenchymal stem cells. Moreover, an independent study by Mor, Nardone, Sams, and Elliott (2015) revealed that hypomethylation of the CpGs in the miR-142 promoter region increases the miRNA expression level in the prefrontal cortex of patients with autism. These data may indicate that rs2526377 attenuates the risk of AD via reducing the miR-142 expression levels in the brain.
MiR-142 is a highly conserved miRNA amongst multiple invertebrate and vertebrate species. The role of miR-142 has extensively been studied in the hematopoietic system, lung development, and cardiac hypertrophy (Shrestha, et al., 2017).  brain lesions than normal white matter, and suggested miR-142 to be involved in the brain inflammatory and degenerative diseases.
Similarly, Mandolesi et al. (2017) observed that miR-142 is increased in the CSF of patients with active MS. Moreover, Sorensen, Nygaard, and Christensen (2016) performed miRNA expression profiles in CSF and blood of patients with AD and found a number of differentially expressed miRNAs, in which miR-142 is one of the significantly upregulated miRNAs in patients with AD compared to controls. Two independent studies also revealed that the expression of miR-142 is increased by age (Huan et al., 2018;Zhang, Azhar, & Wei, 2012).
Here, our expression data confirmed that both mature miR-142-3p and -5p are expressed at relatively high levels in the brain; though, our RNA-Seq analysis proposed miR-142-3p, which is the guide strand of miR-142, to be more active in the regulation of its target genes in the brain. Consistent with this notion, Lau et al. (2013) have performed miRNA expression profiling of the hippocampus of a cohort of 41 AD patients and 23 age-matched controls and found miR-142-3p among the 15 significantly upregulated miRNAs in the AD group. Moreover, miR-142-3p has been reported as one of the eight miRNAs upregulated in synaptoneurosomes from forebrains and hippocampus of mice during prion disease (Boese et al., 2016).
Together, these data endorse that alterations in the expression of miR-142 in the brain could confer AD risk, where higher levels of miR-142-3p increase a person's risk of developing Alzheimer's.
Upregulation of miR-142 in the brain may influence AD risk through different mechanisms. Gene ontology analysis on the putative target genes of miR-142-3p and -5p has shown enrichment in categories related to synaptic transmission (dopaminergic synapse, neurotrophin signaling, axon guidance) and signal transduction (TGFbet signaling, MAPK signaling, ErbB signaling), reported by Mor et al., 2015;Shrestha et al., 2017. In addition, Mandolesi et al. (2017 proposed miR-142 to be related to neuroinflammatory changes in the brain occurring during MS by regulating the expression of IL-1β. Also, Chaudhuri, Yelamanchili, Marcondes, and Fox (2013) suggested the involvement of miR-142 in autoimmune and neuroinflammation in the brain, via miR-142-mediated repression of SIRT1 in primary F I G U R E 5 The interaction and regulatory effect between miR-142-3p and its three target genes. The figure illustrates the binding of miR-142-3p to its three highlighted target genes (TGFBR1, CLF2, and PICALM). The expression of these target genes were significantly downregulated in human iPS-NPCs transfected with miR-142-3p mimic versus untreated, and in iPS-NPCs transfected with miR-142-3p mimic versus negative control. In contrast, the expression of these target genes were upregulated in the hippocampus of miR-142 KO mice versus Wt littermates. Error bars represent standard deviation (SD). *p < .05, **p < .01, and ***p < .001 compared with the control group (Wald-test). KO, knockout; NPC, neuronal progenitor cells human neurons. In an independent study, Chaudhuri et al., 2013 verified that miR-142 indirectly reduces MAOA protein level via regulating SIRT1 expression. Since MAOA is a neurotransmittermetabolizing enzyme and delaminates serotonin, melanin, epinephrine, and norepinephrine, they postulated that miR-142 upregulation might contribute to change the dopaminergic neurotransmission by lowering MAOA expression and activity. In this study, we further demonstrated the miR-142-3p-mediated regulation of multiple target genes in the brain that are likely involved in the pathways underlying AD. TGFBR1 and PICALM, among others, are of particular interest ( Figure 6). TGFBR1 has been shown in several studies to be implicated in AD pathogenesis (Caraci et al., 2012;Chen, Ke, Lu, Qiu, & Peng, 2015;Flanders, Lippa, Smith, Pollen, & Sporn, 1995;Lippa, Flanders, Kim, & Croul, 1998;Tesseur et al., 2006;Wyss-Coray et al., 2002). The regulation of TGFBR1 expression by miR-142-3p has been experimentally confirmed at mRNA and protein levels in previous studies (Talebi et al., 2017;Yang et al., 2017). Our differential expression analysis for all miR-142-3p target genes demonstrated that TGFBR1 was significantly downregulated in miR-142-3p overexpressing human iPS-derived NPCs and the top target gene upregulated in the hippocampus of miR-142 KO mice. Locating at the intersection of anti-inflammatory, antiaging and neuroprotective pathways, makes TGFBR1 a promising molecule for mediating the function of miR-142-3p in AD.
PICALM is ubiquitously expressed in all tissue types with prominent expression in neurons and is nonselectively distributed in pre-and postsynaptic terminals, where it plays an essential role in the fusion of synaptic vesicles to the presynaptic membrane in neurotransmitter release (Ando et al., 2013). Several GWA studies have independently confirmed the association of PICALM with AD (Lambert et al., 2013;Naj et al., 2014;Sims et al., 2017). Recent studies have also shown that PICALM level is reduced in the AD brain endothelium and postulated that it can potentially lead to Aβ accumulation in the brain by hindering LRP1-mediated Aβ transport (Parikh, Fardo, & Estus, 2014;Zhao et al., 2015). These data suggest that derepression of PICALM in response to the reduced miR-142 expression may decrease AD risk that deserves further and deeper investigation in future experimental work.

| CONCLUSIONS
In this study, we endorse 17q22 as a susceptibility locus for AD and provide evidence demonstrating that miR-142 is the most likely functional target in the locus involved in AD pathogenesis.
Furthermore, we revealed miR-142-3p-mediated regulation of multiple target genes in the brain that are implicated in the inflammatory and neurodegenerative manifestations of AD. These include two well-validated AD genes, TGFBR1 and PICALM, of which their derepression in the brain due to reduced expression levels of miR-142-3p may decrease the risk of AD. Our findings may also suggest the therapeutic potential of miR-142 inhibition for AD, which warrants further investigations in the future. F I G U R E 6 Rs2526377 in the promoter of miR-142 modulating its expression and conferring risk of AD. The SNP rs2526377 occurring within the promoter region of miR-142 alters the promoter activity and reduces the expression level of miR-142. Downregulation of miR-142-3p in the brain results in derepression of multiple target genes (e.g., TGFBR1 and PICALM) that contribute to the pathogenesis of AD. AD, Alzheimer's disease; GWAS, genome-wide association studies; SNP, single nucleotide polymorphism