Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Analysis of long noncoding RNAs highlights region-specific altered expression patterns and diagnostic roles in Alzheimer’s disease

Analysis of long noncoding RNAs highlights region-specific altered expression patterns and... Abstract Increasing evidence has revealed the multiple roles of long noncoding RNAs (lncRNAs) in neurodevelopment, brain function and aging, and their dysregulation was implicated in many types of neurological diseases. However, expression pattern and diagnostic role of lncRNAs in Alzheimer’s disease (AD) remain largely unknown and has gained significant attention. In this study, we performed a comparative analysis for lncRNA expression profiles in four brain regions in brain aging and AD. Our analysis revealed age- and disease-dependent region-specific lncRNA expression patterns in aging and AD. Moreover, we identified a panel of nine lncRNAs (termed LncSigAD9) in a discovery cohort of 114 samples using supervised machine learning and stepwise selection method. The LncSigAD9 was able to differentiate between AD and healthy controls with high diagnostic sensitivity and specificity both in the discovery cohort (86.3 and 89.5%) and the additional independent AD cohort (90.8 and 83.8%). The receiver operating characteristic curves for the LncSigAD9 were 0.863 and 0.939 for discovery and independent cohorts, respectively. Furthermore, the LncSigAD9 demonstrated higher diagnostic performance than nine-minus-one lncRNA signature and mRNA-based signature with a similar number of genes. In silico functional analysis indicated the involvement of lncRNA expression variation in brain development- and metabolism-related biological processes. Taken together, our study highlights the importance of lncRNAs in brain aging and AD, and demonstrated the utility of lncRNAs as a promising biomarker for early AD diagnosis and treatment. Alzheimer’s disease, brain aging, biomarkers, long noncoding RNAs Introduction Alzheimer’s disease (AD) is a chronic neurodegenerative disease and remains the sixth leading cause of all deaths worldwide [1]. With the arrival of the aging of the world, we are facing tremendous challenges of the looming global epidemic of AD around the world. With the predicted demographic shift to increasingly elderly populations, it is estimated that the worldwide prevalence of AD will be approximately 100 million in 2050, which means 1 in 85 persons worldwide will be living with the disease [2, 3]. Patients with AD require resource-intensive care, and the cost of AD on society and health-care systems will be unendurable. Early diagnosis of AD and subsequent intervention to slow or halt disease progression become increasingly important. Thus, there is an increasing focus on the need to identify molecular biomarkers for accelerating early clinical diagnosis and treatment at early prodromal stages of AD. A growing body of evidence from large-scale transcriptomic studies has suggested that most of the human genome sequence is actively transcribed into large mRNA-like noncoding transcripts of >200 nucleotides in length, namely, long noncoding RNAs (lncRNAs) [4, 5]. There is increasing realization that lncRNAs are not transcriptional noise but play varied and essential roles in diverse biological phenomena by acting as a regulator of gene expression at epigenetic, transcriptional and posttranscriptional levels [6, 7]. Recent advancement in cancer genomics demonstrated the emerging roles as the hallmarks of cancer [8]. LncRNAs can act as an oncogene or/and tumor-suppressor involved in cancer initiation and progression [9, 10]. Increasing evidence also revealed the multiple roles of lncRNAs in neurodevelopment, brain function and aging, and the dysregulation of lncRNA expression was implicated in many types of neurological diseases, including AD [11–13]. Several lncRNAs, such as BACE1-AS, 51A, BC200 and NDM29, have been found to be aberrantly expressed in AD compared with healthy controls and were involved in AD pathogenesis by the low-throughput experiments [14–17]. Subsequent expression profiles analysis also identified hundreds of differentially expressed lncRNAs between AD samples and healthy controls in human and rat models [18–21], which gives strong evidence of the associations between altered lncRNA expression pattern and AD, and make them attractive as promising potential biomarkers and therapeutic targets for early AD diagnosis and treatment. In the present study, we reviewed the emerging roles and association of lncRNAs in AD and conducted a comprehensive analysis of lncRNAs expression profiles in four brain regions from a large cohort of >400 samples to characterize the expression pattern in aging and AD. In addition, we further assessed and analyzed the utility of lncRNA expression pattern as novel biomarkers for differentiating AD cases and healthy controls. Overall, our study highlights the importance of lncRNAs in brain aging and AD, and indicates that lncRNA may provide additional valuable information for early AD diagnosis and treatment. Materials and methods Sample data sets and study design In this study, two independent large data sets comprising of microarray data and clinical information of AD patients and healthy control samples were obtained from the publicly available Gene Expression Omnibus (GEO) database (GEO accession numbers are GSE48350 and GSE5281, respectively). The GSE48350 data set containing a large sample size with 253 tissue samples from four brain regions [entorhinal cortex (EC), hippocampus (HC), post-central gyrus (PCG) and superior frontal gyrus (SFG)] of 57 non-AD controls and 28 AD cases was adapted as primary data source to study dynamic expression pattern of lncRNA across multiple brain regions in aging and AD. For the purpose of lncRNA biomarkers discovery, 114 samples (57 pairs including 57 AD samples and 57 healthy control samples) were selected from 253 tissue samples of GSE48350 cohort as discovery cohort according to the paired design principle in terms of the following selection criteria: (1) the same gender; (2) the same detected brain region; and (3) the similar age (age difference <5 years). In addition, 161 samples (87 AD patient samples and 74 healthy control samples) from GSE5281 cohort were used as an additional independent AD cohort for biomarker validation. Acquisition and analysis of lncRNA expression profiles Raw microarray data (.CEL) of GSE48350 and GSE5281 measured with Affymetrix HG-U133_Plus_2.0 platform were downloaded from GEO database. Raw probe-level intensity values are background corrected, log2 transformed and quantile normalized using the Robust Multi-array Average (RMA) algorithm. LncRNA expression data were obtained by repurposing the probes from Affymetrix microarray based on the NetAffx annotation of the probe sets and the Refseq and GENCODE annotations as follows [22]: (1) for the probe sets with Refseq IDs, we only extracted those labeled as ‘NR_’ and annotated with ‘long non-coding RNA’ in Refseq database; (2) for the probe sets with Ensembl gene IDs, we only retained those annotated with ‘long non-coding RNA’ in GENCODE project (release 25); (3) finally, expression profiles of 2466 unique lncRNAs corresponding to 3431 probe sets were used for further analysis. A combination of significance analysis of microarrays (SAM) and variance filtering was used to evaluate differences in the expression of lncRNAs between different sample sets. Only those lncRNAs with a false discovery rate (FDR)-adjusted P-value < 0.2 and coefficient of variation (CV) value > 5% were considered as differentially expressed lncRNAs. Hierarchical clustering of lncRNA expression data was performed using R software with Euclidean distance and complete linkage, and the Pearson's chi-squared test was used to assess the significance of the association between sample clusters and samples’ status. Aging-AD continuum genes were identified as previously described [23]: (1) CV value > 5%; (2) have significant expression trend (P< 0.01) across young, aged and AD groups by linear regression analysis and (3) have a consistent directional change based on 80% confidence interval for the difference in expression means for each pairwise comparison. Statistics for classification and prediction A supervised machine learning method, support vector machine (SVM) with radial basis function kernel, was used to construct the lncRNA-based classifier for the classification of AD and healthy control samples using the R package e1071. To identify optimal lncRNA biomarkers form differentially expressed lncRNAs, we performed feature selection procedure using a stepwise selection method by a combination of forward selection and backward elimination approaches as follows: (1) each lncRNA was selected as the initial set; (2) searching through the remaining n-minus-1 lncRNAs to find out which two variables should be added to the current set to best improve the accuracy of classifier; (3) removing the least useful lncRNA from current classifier, one-at-a-time and (4) repeating Step 2 and Step 3 until all candidate lncRNAs have been picked by classifier. Binary classification performances were evaluated by estimating the accuracy, sensitivity and specificity of the lncRNA-based classifier with 5-fold cross-validation. In addition, a receiver operating characteristic (ROC) curve was plotted and the area under the ROC (AUC) was calculated to illustrate the discriminative power of the lncRNA-based classifier. In silico functional analysis of lncRNAs Co-expression relationships between lncRNAs and mRNAs were measured by calculating the Pearson correlation coefficient using the paired lncRNA and mRNA expression profiles in 253 samples of GSE48350 cohort. To infer the biological roles of lncRNAs, we performed functional enrichment analysis of Gene Ontology (GO) terms for co-expressed mRNAs using ClueGO plugin (version 2.3.3) in Cytoscape [24] limited in biological processes. Enriched GO terms and functional clusters were obtained and visualized based on a two-sided hypergeometric test with Bonferroni step down correction and kappa score threshold of 0.4, and limited in the GO-level intervals 3–8 with minimum gene 20 and P-value ≤ 0.05. In addition, we also conducted functional enrichment analysis of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway using DAVID Bioinformatics Resources (https://david.ncifcrf.gov/, version 6.8) [25]. Significantly enriched KEGG pathways were obtained using functional annotation chart options with the whole human genome as background when P-value <0.05. Results Region-specific altered lncRNA expression patterns in brain aging or AD To obtain a global view of lncRNA expression pattern in brain aging, lncRNA expression levels were compared in four brain regions (EC, HC, PCG and SFG) between young and aged healthy controls. A total of 383 lncRNAs reveal significant expression changes in four brain regions in aging (Supplementary File S1). Of them, only eight lncRNAs are dysregulated expressed in all the four brain regions, whereas others (375 lncRNAs, 97.9%) revealed a region-specific altered expression pattern in the course of aging (Figure 1A). Specifically, two brain regions, the PCG and SFG, showed most extensive lncRNA expression changes between the two age control groups (258 lncRNAs in the PCG and 247 lncRNAs in the SFG), whereas relative few lncRNA expression change was observed in the EC (70 lncRNAs) and HC (58 lncRNAs) (Figure 1A). Moreover, the majority of dysregulated lncRNAs would undergo increased expression in aging in three of four brain regions, particularly in the HC, whereas approximately 88.6% of dysregulated lncRNAs revealed downregulated expression with increasing age in the EC (Figure 1B). Then, we performed hierarchical clustering analysis for all samples in each of four brain regions according to expression levels of differentially expressed lncRNAs and found that altered expression pattern of lncRNA can distinguish the young and aged healthy controls for different brain regions (Figure 1C). Figure 1. View largeDownload slide Differential lncRNA expression pattern across four brain regions in aging. (A) Venn diagram for the number of dysregulated lncRNAs across four brain regions. (B) Number of downregulated and upregulated lncRNAs with increasing age in each brain region. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in each brain region. Figure 1. View largeDownload slide Differential lncRNA expression pattern across four brain regions in aging. (A) Venn diagram for the number of dysregulated lncRNAs across four brain regions. (B) Number of downregulated and upregulated lncRNAs with increasing age in each brain region. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in each brain region. To characterize AD-associated lncRNA expression pattern, lncRNA expression profiles were next compared in four brain regions between AD cases and aged healthy controls. Overall, 146 lncRNAs were detected as differentially expressed mainly in three brain regions (the EC, HC and SFG) except the PCG between AD cases and aged healthy controls (Supplementary File S2). Moreover, most of them (133 lncRNAs) undergo region-specific expression changes in AD cases, and only fewer lncRNA (13 lncRNAs) revealed expression changes in more than two regions (Figure 2A). Notably, all dysregulated lncRNAs revealed upregulated expression in the EC in AD cases, whereas the HC showed the largest number of downregulated lncRNAs. In addition, the SFG demonstrated the balanced distribution of upregulated and downregulated lncRNAs in AD cases (Figure 2B). Also, unsupervised hierarchical clustering of these dysregulated lncRNAs revealed AD-associated expression pattern differentiating AD cases from healthy controls (Figure 2C), which were further validated in another independent GSE5281 data set (Supplementary File S3). Figure 2. View largeDownload slide Differential lncRNA expression pattern across three brain regions in AD. (A) Venn diagram for the number of dysregulated lncRNAs across three brain regions. (B) Number of downregulated and upregulated lncRNAs in each brain region in AD. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in three brain regions. Figure 2. View largeDownload slide Differential lncRNA expression pattern across three brain regions in AD. (A) Venn diagram for the number of dysregulated lncRNAs across three brain regions. (B) Number of downregulated and upregulated lncRNAs in each brain region in AD. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in three brain regions. Because our above analysis revealed differences in lncRNA expression between young and aged healthy controls, and between AD cases and aged healthy controls, we next examined whether expression changes in lncRNA have a consistent directional trend from aging to AD. Using linear regression and pairwise comparison analysis, we identified 155 Aging-AD continuum lncRNAs in four brain regions (Supplementary File S4). These Aging-AD continuum lncRNAs also showed region-specific progressive expression changes, with fewer or no overlap for progressively upregulated or downregulated lncRNAs over aging and AD (Figure 3A). Although the SFG showed a large number of Aging-AD continuum lncRNAs, the distribution of progressively increased or decreased lncRNAs from aging to AD is nearly balanced in four brain regions (Figure 3B and C). Figure 3. View largeDownload slide Region-specific progressive expression changes of Aging-AD continuum lncRNAs. (A) Venn diagram for the number of lncRNAs undergoing increased or decreased expression across aging and AD. (B) Distribution of progressively increased or decreased lncRNAs from aging to AD in four brain regions. (C) Profiles of Aging-AD continuum lncRNAs in four brain regions. Green lines represent lncRNAs undergoing progressively decreased expression across aging and AD, and red lines represent lncRNAs undergoing progressively increased expression across aging and AD. Figure 3. View largeDownload slide Region-specific progressive expression changes of Aging-AD continuum lncRNAs. (A) Venn diagram for the number of lncRNAs undergoing increased or decreased expression across aging and AD. (B) Distribution of progressively increased or decreased lncRNAs from aging to AD in four brain regions. (C) Profiles of Aging-AD continuum lncRNAs in four brain regions. Green lines represent lncRNAs undergoing progressively decreased expression across aging and AD, and red lines represent lncRNAs undergoing progressively increased expression across aging and AD. Initial identification and validation of lncRNA biomarkers for early detection of AD In view of the region-specific altered lncRNA expression changes in aging or AD, a total of 114 samples (57 sample pairs) from GSE48350 was selected as discovery cohort using the paired design principle to detect potential lncRNA biomarkers, and the remaining 139 samples of GSE48350 (including 23 AD patient samples and 116 healthy control samples) were used as internal testing cohort. We first performed differentially expression analysis for lncRNAs between 57 AD patient samples and 57 healthy control samples in the discovery cohort. Overall, 47 lncRNAs were detected as differentially expressed between AD patient samples and healthy control samples (FDR-adjusted P-value < 0.01 and CV >5%) (Supplementary File S5). Of them, 11 lncRNAs were upregulated and 36 lncRNAs were downregulated in AD patient samples compared with healthy control samples. Then, we conducted unsupervised hierarchical clustering analysis on 114 samples of discovery cohort using the set of 47 differentially expressed lncRNAs. As expected, we observed that differentially expressed lncRNAs exhibited distinctive expression pattern, which clearly separated AD patient samples from healthy control samples (Fisher’s exact test P = 1.94E-09) (Figure 4A), suggesting that these 47 differentially expressed lncRNAs could be as candidate predictive biomarkers for early detection of AD. Figure 4. View largeDownload slide Identification and validation of lncRNA biomarkers for early detection of AD in the discovery cohort. (A) Unsupervised hierarchical clustering of 114 samples based on 47 differentially expressed lncRNAs. (B) The highest predicted accuracy of each lncRNA combination constructed by a specific number of lncRNAs (k = 1, 2,…, 47). (C) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the discovery cohort. (D) Performance comparison of 9-lncRNA diagnostic model and 1000 random 9-lncRNA combinations. Figure 4. View largeDownload slide Identification and validation of lncRNA biomarkers for early detection of AD in the discovery cohort. (A) Unsupervised hierarchical clustering of 114 samples based on 47 differentially expressed lncRNAs. (B) The highest predicted accuracy of each lncRNA combination constructed by a specific number of lncRNAs (k = 1, 2,…, 47). (C) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the discovery cohort. (D) Performance comparison of 9-lncRNA diagnostic model and 1000 random 9-lncRNA combinations. To identify optimal lncRNA biomarkers, we performed a classification and feature selection for 47 differentially expressed lncRNAs on 114 samples of discovery cohort using the SVM and stepwise selection model. The abovementioned analysis uncovered a combination of nine lncRNAs (MIR7-3HG, AL109615.3, NEBL-AS1, ATP6V0E2-AS1, PDXDC2P-NPIPB14P, LOC441204, A2M-AS1, TGFB2-OT1 and LINC00672), which yields a greatest discriminative ability with a predictive accuracy of 87.7% (Figure 4B). Therefore, this combination of nine lncRNAs was defined as an AD-related predictive signature (hereafter referred to as LncSigAD9). When tested in the discovery cohort using SVM and 5-fold cross-validation, the LncSigAD9 correctly classified 49 of 57 AD patient samples and 51 of 57 healthy control samples, achieving an AUC of 0.863 with a predictive accuracy of 87.7%, the sensitivity of 86.3% and the specificity of 89.5% (Figure 4C). We then generated 1000 random 9-lncRNA combinations by randomly selected nine lncRNAs and evaluated their performance using the same protocol as used for the LncSigAD9. As shown in the kernel density plots, the LncSigAD9 demonstrated superior performance compared with all random 9-lncRNA combinations (Figure 4D). Further validation of the LncSigAD9 with an additional independent AD cohort To evaluate the robustness of the LncSigAD9, further validation of the predictive ability of the LncSigAD9 was conducted using an additional independent AD cohort of 161 samples from GEO accession GSE5281. We first clustered 116 samples in the GSE5281 cohort according to the expression levels of nine lncRNAs in the LncSigAD9 by hierarchical clustering analysis and observed two distinct sample groups significantly correlated with associated their AD status (Fisher’s exact test P = 6.52E-12) (Figure 5A). In addition, we also performed classification analysis for 161 samples by LncSigAD9 using the SVM and 5-fold cross-validation as in the discovery cohort. As shown in Figure 5B, the LncSigAD9 showed a well predictive performance comparable with the discovery cohort. The validation result showed that the LncSigAD9 correctly classified 79 of 87 AD patient samples and 62 of 74 healthy control samples, achieving an AUC of 0.939 with a predictive accuracy of 87.6%, the sensitivity of 90.8% and the specificity of 83.8% (Figure 5B). Figure 5. View largeDownload slide Evaluation of robustness and reproducibility of the 9-lncRNA diagnostic model an additional independent AD cohort. (A) Unsupervised hierarchical clustering of 116 samples based on expression levels of nine lncRNAs in the LncSigAD9. (B) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the independent AD cohort. Figure 5. View largeDownload slide Evaluation of robustness and reproducibility of the 9-lncRNA diagnostic model an additional independent AD cohort. (A) Unsupervised hierarchical clustering of 116 samples based on expression levels of nine lncRNAs in the LncSigAD9. (B) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the independent AD cohort. To further investigate whether all of the nine lncRNAs in the LncSigAD9 are essential for its predictive value, we constructed all possible nine-minus-one lncRNA signatures by the removal of one lncRNA at a time and performed comparison analysis of predictive power for LncSigAD9 and other nine-minus-one lncRNA signatures using the SVM and 5-fold cross-validation in the discovery cohort and independent AD cohort. The comparison showed that none of the nine-minus-one lncRNA signatures yields better distinction than the LncSigAD9 between AD samples and healthy controls both in the discovery cohort (Figure 4C) and independent AD cohort (Figure 5B). This indicates that all nine lncRNAs are essential for the predictive power of the LncSigAD9. Comparison of the LncSigAD9 with existing mRNA-related biomarkers Finally, we compared the predictive value of the LncSigAD9 to mRNA-related biomarkers. Ten most predictive mRNAs from AclarusDx™ (ZNF267, CSPG2, KIAA1009, GALNT3, JAK2, TNNI3K, ANKRD49, BCL2A1, ROCK1 and CLEC2B) were selected to constitute a 10-mRNA signature (hereafter referred to as mRSigAD10) [26]. Hierarchical clustering analysis was first performed for samples of discovery cohort and independent AD cohort using expression pattern of 10 mRNAs in mRSigAD10. The expression pattern of 10 mRNAs revealed relatively poor sample classification performance than that of the LncSigAD9 both in the discovery cohort (Fisher’s exact test P = 0.0093 versus P = 1.94E-09) and independent AD cohort (Fisher’s exact test P = 0.0863 versus P = 6.52E-12) (Supplementary File S6). Then, we compared the sensitivity and specificity in the diagnosis of AD between the LncSigAD9 and mRSigAD10. Using the same SVM and 5-fold cross-validation, the mRSigAD10 obtained an AUC value of 0.678 with the sensitivity of 57.9% and the specificity of 70.2% in the discovery cohort (Figure 4C) and an AUC value of 0.897 with the sensitivity of 86.2% and the specificity of 85.1% in the independent AD cohort (Figure 5B), which are all lower than that of the LncSigAD9. Functional implication of the LncSigAD9 To gain a first insight into the biological function of the LncSigAD9, we first measured the association between expression levels of each lncRNA in the LncSigAD9 and that of mRNAs in 253 samples of GSE48350 cohort and identified positively co-expressed mRNAs with the LncSigAD9 (top 0.5%). Then, we performed GO and KEGG pathway function enrichment analysis for positively co-expressed mRNAs using clueGO and DAVID. GO analysis revealed that mRNAs positively correlated at least one of the nine signature lncRNAs were significantly enriched in some GO biological processes most of which can be categorized into five functional groups including ATP metabolic process, brain development, neurogenesis, neuron differentiation and development (Figure 6A). KEGG pathway analysis revealed 15 significant enriched biological pathways most of which are already known to be related to AD pathogenesis, including oxidative phosphorylation (OXPHOS), peroxisome, MAPK signaling pathway, FoxO signaling pathway, pyruvate metabolism, proteoglycans in cancer, metabolic pathways, melanoma, hypertrophic cardiomyopathy, AD, Parkinson’s disease, Huntington’s disease, regulation of actin cytoskeleton, carbon metabolism and arginine and proline metabolism, which have been reported to be important biological processes and pathways involved in AD pathogenesis. Figure 6. View largeDownload slide Functions enriched by the protein-coding genes co-expressed with nine lncRNAs in the LncSigAD9. (A) Functionally grouped network with enriched GO terms as nodes linked based on their kappa score level (≥0.4). Node size represents the term enrichment significance. (B) The most significantly enriched KEGG pathways. The node size represents the number of genes in the pathways, and the color represents the pathway enrichment significance. Figure 6. View largeDownload slide Functions enriched by the protein-coding genes co-expressed with nine lncRNAs in the LncSigAD9. (A) Functionally grouped network with enriched GO terms as nodes linked based on their kappa score level (≥0.4). Node size represents the term enrichment significance. (B) The most significantly enriched KEGG pathways. The node size represents the number of genes in the pathways, and the color represents the pathway enrichment significance. Discussion Previous microarray studies have characterized the expression pattern and functional roles of protein-coding genes in brain aging and AD [23, 27–30]. Recent advances in transcriptomic studies have discovered a new class of noncoding RNA transcripts (termed lncRNAs) and highlighted their critical roles in genome regulatory network and disease phenotypes [7, 31]. Despite the fact that several studies have reported lncRNA expression profiles and provided preliminary evidence for the involvement of lncRNAs in brain development and AD [19, 21, 32–34], the dynamic expression pattern of lncRNA across multiple brain regions in aging and AD remains largely unknown. In this study, we first compared differential lncRNA expression profiles of four brain regions in young and aged healthy controls and reported region-specific and age-dependent lncRNA expression pattern in brain aging. Our results demonstrated that the most extensive lncRNA expression changes were observed in the PCG and SFG, although all four brain regions showed differential responsiveness with increasing age, which is similar to protein-coding genes [23, 27, 28]. However, our analysis observed that lncRNA revealed opposite trend when compared with protein-coding genes. It has been reported that most of responsive protein-coding genes underwent decreased expression in the PCG and SFG [23, 27]. In contrast, extensive upregulated expression was observed with increased aging in the PCG, SFG and HC. These findings demonstrated that lncRNA plays important roles in transcriptional regulation in aging of brain and maybe a new molecular hallmark of brain aging. With the advances of genomics and transcriptomics in human complex disease, molecular profiles (such as mRNA profiling and miRNA profiling) are increasingly used as an alternative as diagnostic biomarkers for AD because of their noninvasion and easy accessibility compared with those from cerebrospinal fluid (SCF) [26, 35–37]. Until now, some efforts have been made to investigate the role of lncRNA in AD by focusing on some specific lncRNAs, as summarized by Luo and Chen [38]. Recently, several publications dealing with lncRNA expression profiles in AD reported a large number of dysregulated lncRNA. Despite these previously investigation, the expression pattern and diagnostic role of lncRNA in AD remain largely unknown. Therefore, to characterize the expression pattern of lncRNA in AD, we first compared differential lncRNA profiles of AD cases and aged healthy controls in four brain regions. This analysis demonstrated that extensive expression changes occur in 146 lncRNAs, most of which showing decreased expression in AD, which is in line with previous studies [18]. However, a notable feature of lncRNA expression pattern we found is that their alterations are regionally specific in AD, which has not been previously reported. In particular, the HC revealed the largest number of downregulated lncRNAs, and the EC revealed the largest number of upregulated lncRNAs in AD, and this could be because the HC and EC are not only affected in the early stages of AD but also are highly interactive components of a functionally connected network [39, 40]. Berchtold et al. [23] found that apparent expression changes of some synaptic genes in AD have been initiated to some degree already in normal aging. Therefore, we further examined whether some lncRNAs undergo progressive change across aging and AD and identified 155 Aging-AD continuum lncRNAs in four brain regions, suggesting that altered expression changes of these Aging-AD continuum lncRNAs in the AD brain have occurred to soma degree in the aged brain. Overall, these results suggested that lncRNA revealed age- and disease-dependent expression patterns in aging and AD, which could be served as molecular biomarkers for early detection of AD. As the first step toward identifying diagnostic lncRNA biomarkers, we conducted differential expression analysis of lncRNAs in 57 AD-control sample pairs using the paired design principle and identified 47 lncRNAs as potential biomarker candidates. To develop a clinically applicable lncRNA-based diagnostic marker, 47 candidate lncRNAs were subjected to the SVM and stepwise selection model to further narrow down the number of candidate lncRNAs. Finally, a panel of nine lncRNAs (termed LncSigAD9), including MIR7-3HG, AL109615.3, NEBL-AS1, ATP6V0E2-AS1, PDXDC2P-NPIPB14P, LOC441204, A2M-AS1, TGFB2-OT1 and LINC00672, with the highest accuracy was selected as the final signature. The LncSigAD9 achieved an AUC value of 0.863 for distinguishing AD patient samples and healthy controls in the discovery cohort. We further tested the performance of the LncSigAD9 by using an independent AD data set from GEO, in which the LncSigAD9 could differentiate AD patient samples from healthy controls with high specificity and sensitivity. These results demonstrated the reproducible predictive power and general applicability of the LncSigAD9 in determining the risk of AD. Furthermore, we also constructed all nine-minus-one lncRNA signatures and compared their predictive performance with LncSigAD9 in the discovery cohort and independent AD cohort. The comparison showed the necessity of all nine lncRNAs in the LncSigAD9 for the diagnosis of AD. Moreover, the LncSigAD9 revealed superior sensitivity and specificity than mRNA-based signature with a similar number of genes for AD diagnosis. To better understanding possible biological roles of the LncSigAD9 in the pathophysiology of AD, we tried to infer the potential functions of the identified LncSigAD9 by performing functional enrichment analysis for mRNAs co-expressed with nine lncRNAs in the LncSigAD9. GO analysis revealed that these lncRNAs are involved in brain development- and ATP metabolism-related biological processes. KEGG analysis indicated that genes whose expression correlated with the LncSigAD9 were enriched significantly in 15 KEGG pathways. Of the 15 KEGG pathways, 3 pathways are directly neurodegenerative disease-related KEGG pathways, including Alzheimer's disease, Parkinson’s disease and Huntington’s disease. Most of remaining enriched pathways are related to metabolism processes, including OXPHOS, metabolic pathways, arginine and proline metabolism, pyruvate metabolism, carbon metabolism, peroxisome and FoxO signaling pathway. Aberrations in metabolic processes have been reported to be involved in AD [41]. For example, OXPHOS is the classic role of mitochondria and is a vital part of metabolism that provides energy (ATP) for basal metabolism. Many previous studies have indicated abnormalities in OXPHOS in the pathogenesis of AD [42, 43]. In a mouse model, Isopi et al. [44] found that pyruvate could prevent the development and progression of AD-related cognitive deficits by reducing lipid peroxidation and oxidative stress. In the brain, peroxisomes are present in all neural cell types and peroxisomal alterations contributed to the progression of AD [45, 46]. Apart from abovementioned neurodegenerative disease-related and ATP metabolism-related pathways, other several enriched KEGG pathways, such as MAPK signaling pathway and regulation of actin cytoskeleton, have also been involved in the pathophysiology and pathogenesis of AD [47–49]. Overall, in silico functional analysis suggests that expression variation in the lncRNAs of the LncSigAD9 perhaps perturbs gene regulatory network, which may have profound effects on important biological processes and pathways involved in AD pathogenesis. In summary, our study represented a comprehensive analysis of lncRNAs in four brain regions and uncovered region-specific altered lncRNA expression pattern in aging and AD. In addition, our study investigated the diagnostic potential of lncRNA expression profiles and constructed a 9-lncRNA classifier, which could differentiate AD from normal controls with high diagnostic specificity and sensitivity in multiple cohorts. Functional analysis indicated the involvement of lncRNA expression variation in brain development- and metabolism-related biological processes. These findings highlight the importance of lncRNAs in the pathogenesis of AD and demonstrated the utility of lncRNAs as a promising biomarker to aid early AD diagnosis. However, further experimental studies or independent validation will be needed to fully elucidate the molecular mechanism and clinical value of lncRNAs in AD. Key Points We have highlighted the emerging roles and association of lncRNA in brain aging and AD. Our study demonstrated age- and disease-dependent expression patterns of lncRNAs in aging and AD, which could be served as molecular biomarkers for early detection of AD. We identified a robust and reproducible 9-lncRNA diagnostic model with high sensitivity and specificity in differentiating between AD and healthy controls. Supplementary Data Supplementary data are available online at https://academic.oup.com/bib. Funding This study was supported by the National Natural Science Foundation of China (grant number 61602134). Meng Zhou is an associate professor at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University. His research interests include bioinformatics and computational RNomics. Hengqiang Zhao is a student at the College of Bioinformatics Science and Technology, Harbin Medical University. His research interests include bioinformatics and disease systems biology. Xinyu Wang is a bioinformatics engineer at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University. His research interests include computational epigenomics and NGS. Jie Sun is an associate professor at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University, and College of Bioinformatics Science and Technology, Harbin Medical University. Her research interests include cancer bioinformatics and translational medicine. Jianzhong Su is a professor at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University. His research interests in bioinformatics and cancer epigenetics. References 1 Reitz C. Alzheimer's disease and the amyloid cascade hypothesis: a critical review . Int J Alzheimers Dis 2012 ; 2012 : 369808. Google Scholar PubMed 2 Brookmeyer R , Johnson E , Ziegler-Graham K , et al. Forecasting the global burden of Alzheimer's disease . Alzheimers Dement 2007 ; 3 ( 3 ): 186 – 91 . Google Scholar CrossRef Search ADS PubMed 3 Prince M , Bryce R , Albanese E , et al. The global prevalence of dementia: a systematic review and metaanalysis . Alzheimers Dement 2013 ; 9 ( 1 ): 63 – 75.e62 . Google Scholar CrossRef Search ADS PubMed 4 Consortium EP , Birney E , Stamatoyannopoulos JA , et al. Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project . Nature 2007 ; 447 : 799 – 816 . Google Scholar CrossRef Search ADS PubMed 5 Clark MB , Amaral PP , Schlesinger FJ , et al. The reality of pervasive transcription . PLoS Biol 2011 ; 9 ( 7 ): e1000625 . Google Scholar CrossRef Search ADS PubMed 6 Quinn JJ , Chang HY. Unique features of long non-coding RNA biogenesis and function . Nat Rev Genet 2016 ; 17 ( 1 ): 47 – 62 . Google Scholar CrossRef Search ADS PubMed 7 Rinn JL , Chang HY. Genome regulation by long noncoding RNAs . Annu Rev Biochem 2012 ; 81 : 145 – 66 . Google Scholar CrossRef Search ADS PubMed 8 Gutschner T , Diederichs S. The hallmarks of cancer: a long non-coding RNA point of view . RNA Biol 2012 ; 9 ( 6 ): 703 – 19 . Google Scholar CrossRef Search ADS PubMed 9 Tsai MC , Spitale RC , Chang HY. Long intergenic noncoding RNAs: new links in cancer progression . Cancer Res 2011 ; 71 ( 1 ): 3 – 7 . Google Scholar CrossRef Search ADS PubMed 10 Bartonicek N , Maag JL , Dinger ME. Long noncoding RNAs in cancer: mechanisms of action and technological advancements . Mol Cancer 2016 ; 15 ( 1 ): 43. Google Scholar CrossRef Search ADS PubMed 11 Barry G , Guennewig B , Fung S , et al. Long Non-Coding RNA Expression during aging in the human subependymal zone . Front Neurol 2015 ; 6 : 45. Google Scholar CrossRef Search ADS PubMed 12 Roberts TC , Morris KV , Wood MJ. The role of long non-coding RNAs in neurodevelopment, brain function and neurological disease . Philos Trans R Soc Lond B Biol Sci 2014 ; 369 ( 1652 ): 20130507. Google Scholar CrossRef Search ADS PubMed 13 Kim J , Kim KM , Noh JH , et al. Long noncoding RNAs in diseases of aging . Biochim Biophys Acta 2016 ; 1859 ( 1 ): 209 – 21 . Google Scholar CrossRef Search ADS PubMed 14 Faghihi MA , Modarresi F , Khalil AM , et al. Expression of a noncoding RNA is elevated in Alzheimer's disease and drives rapid feed-forward regulation of beta-secretase . Nat Med 2008 ; 14 ( 7 ): 723 – 30 . Google Scholar CrossRef Search ADS PubMed 15 Ciarlo E , Massone S , Penna I , et al. An intronic ncRNA-dependent regulation of SORL1 expression affecting Abeta formation is upregulated in post-mortem Alzheimer's disease brain samples . Dis Model Mech 2013 ; 6 ( 2 ): 424 – 33 . Google Scholar CrossRef Search ADS PubMed 16 Lin D , Pestova TV , Hellen CU , et al. Translational control by a small RNA: dendritic BC1 RNA targets the eukaryotic initiation factor 4A helicase mechanism . Mol Cell Biol 2008 ; 28 ( 9 ): 3008 – 19 . Google Scholar CrossRef Search ADS PubMed 17 Massone S , Ciarlo E , Vella S , et al. NDM29, a RNA polymerase III-dependent non coding RNA, promotes amyloidogenic processing of APP and amyloid beta secretion . Biochim Biophys Acta 2012 ; 1823 ( 7 ): 1170 – 7 . Google Scholar CrossRef Search ADS PubMed 18 Zhou X , Xu J. Identification of Alzheimer's disease-associated long noncoding RNAs . Neurobiol Aging 2015 ; 36 ( 11 ): 2925 – 31 . Google Scholar CrossRef Search ADS PubMed 19 Yang B , Xia ZA , Zhong B , et al. Distinct hippocampal expression profiles of long non-coding RNAs in an Alzheimer's disease model . Mol Neurobiol 2017 ; 54 ( 7 ): 4833 – 46 . Google Scholar CrossRef Search ADS PubMed 20 Magistri M , Velmeshev D , Makhmutova M , et al. Transcriptomics profiling of Alzheimer's disease reveal neurovascular defects, altered amyloid-beta homeostasis, and deregulated expression of long noncoding RNAs . J Alzheimers Dis 2015 ; 48 ( 3 ): 647 – 65 . Google Scholar CrossRef Search ADS PubMed 21 Lee DY , Moon J , Lee ST , et al. Distinct expression of long non-coding RNAs in an Alzheimer's disease model . J Alzheimers Dis 2015 ; 45 ( 3 ): 837 – 49 . Google Scholar PubMed 22 Zhang X , Sun S , Pu JK , et al. Long non-coding RNA expression profiles predict clinical phenotypes in glioma . Neurobiol Dis 2012 ; 48 ( 1 ): 1 – 8 . Google Scholar CrossRef Search ADS PubMed 23 Berchtold NC , Coleman PD , Cribbs DH , et al. Synaptic genes are extensively downregulated across multiple brain regions in normal human aging and Alzheimer's disease . Neurobiol Aging 2013 ; 34 ( 6 ): 1653 – 61 . Google Scholar CrossRef Search ADS PubMed 24 Bindea G , Mlecnik B , Hackl H , et al. ClueGO: a Cytoscape plug-in to decipher functionally grouped gene ontology and pathway annotation networks . Bioinformatics 2009 ; 25 ( 8 ): 1091 – 3 . Google Scholar CrossRef Search ADS PubMed 25 Huang da W , Sherman BT , Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources . Nat Protoc 2009 ; 4 ( 1 ): 44 – 57 . Google Scholar CrossRef Search ADS PubMed 26 Fehlbaum-Beurdeley P , Sol O , Desire L , et al. Validation of AclarusDx, a blood-based transcriptomic signature for the diagnosis of Alzheimer's disease . J Alzheimers Dis 2012 ; 32 : 169 – 81 . Google Scholar PubMed 27 Berchtold NC , Cribbs DH , Coleman PD , et al. Gene expression changes in the course of normal brain aging are sexually dimorphic . Proc Natl Acad Sci USA 2008 ; 105 ( 40 ): 15605 – 10 . Google Scholar CrossRef Search ADS PubMed 28 Cribbs DH , Berchtold NC , Perreau V , et al. Extensive innate immune gene activation accompanies brain aging, increasing vulnerability to cognitive decline and neurodegeneration: a microarray study . J Neuroinflammation 2012 ; 9 ( 1 ): 179 . Google Scholar CrossRef Search ADS PubMed 29 Liu G , Bao X , Wang R. Expression quantitative trait loci regulate HNF4A and PTBP1 expression in human brains . Proc Natl Acad Sci USA 2015 ; 112 ( 30 ): E3975. Google Scholar CrossRef Search ADS PubMed 30 Xiang Z , Xu M , Liao M , et al. Integrating genome-wide association study and brain expression data highlights cell adhesion molecules and purine metabolism in Alzheimer's disease . Mol Neurobiol 2015 ; 52 ( 1 ): 514 – 21 . Google Scholar CrossRef Search ADS PubMed 31 Schmitz SU , Grote P , Herrmann BG. Mechanisms of long noncoding RNA function in development and disease . Cell Mol Life Sci 2016 ; 73 ( 13 ): 2491 – 509 . Google Scholar CrossRef Search ADS PubMed 32 Aprea J , Prenninger S , Dori M , et al. Transcriptome sequencing during mouse brain development identifies long non-coding RNAs functionally involved in neurogenic commitment . EMBO J 2013 ; 32 ( 24 ): 3145 – 60 . Google Scholar CrossRef Search ADS PubMed 33 Lv J , Cui W , Liu H , et al. Identification and characterization of long non-coding RNAs related to mouse embryonic brain development from available transcriptomic data . PLoS One 2013 ; 8 ( 8 ): e71152 . Google Scholar CrossRef Search ADS PubMed 34 Shi C , Zhang L , Qin C. Long non-coding RNAs in brain development, synaptic biology, and Alzheimer's disease . Brain Res Bull 2017 ; 132 : 160 – 9 . Google Scholar CrossRef Search ADS PubMed 35 Sood S , Gallagher IJ , Lunnon K , et al. A novel multi-tissue RNA diagnostic of healthy ageing relates to cognitive health status . Genome Biol 2015 ; 16 ( 1 ): 185 . Google Scholar CrossRef Search ADS PubMed 36 Leidinger P , Backes C , Deutscher S , et al. A blood based 12-miRNA signature of Alzheimer disease patients . Genome Biol 2013 ; 14 ( 7 ): R78 . Google Scholar CrossRef Search ADS PubMed 37 Kumar P , Dezso Z , MacKenzie C , et al. Circulating miRNA biomarkers for Alzheimer's disease . PLoS One 2013 ; 8 ( 7 ): e69807 . Google Scholar CrossRef Search ADS PubMed 38 Luo Q , Chen Y. Long noncoding RNAs and Alzheimer’s disease . Clin Interv Aging 2016 ; 11 : 867. Google Scholar CrossRef Search ADS PubMed 39 van Strien NM , Cappaert NL , Witter MP. The anatomy of memory: an interactive overview of the parahippocampal-hippocampal network . Nat Rev Neurosci 2009 ; 10 ( 4 ): 272 – 82 . Google Scholar CrossRef Search ADS PubMed 40 Braak H , Braak E. Neuropathological stageing of Alzheimer-related changes . Acta Neuropathol 1991 ; 82 ( 4 ): 239 – 59 . Google Scholar CrossRef Search ADS PubMed 41 Chen J , Xie C , Zhao Y , et al. Gene expression analysis reveals the dysregulation of immune and metabolic pathways in Alzheimer's disease . Oncotarget 2016 ; 7 ( 45 ): 72469 – 74 . Google Scholar PubMed 42 Biffi A , Sabuncu MR , Desikan RS , et al. Genetic variation of oxidative phosphorylation genes in stroke and Alzheimer's disease . Neurobiol Aging 2014 ; 35 ( 8 ): 1956.e1 – 8 . Google Scholar CrossRef Search ADS 43 Shoffner JM. Oxidative phosphorylation defects and Alzheimer's disease . Neurogenetics 1997 ; 1 ( 1 ): 13 – 9 . Google Scholar CrossRef Search ADS PubMed 44 Isopi E , Granzotto A , Corona C , et al. Pyruvate prevents the development of age-dependent cognitive deficits in a mouse model of Alzheimer's disease without reducing amyloid and tau pathology . Neurobiol Dis 2015 ; 81 : 214 – 24 . Google Scholar CrossRef Search ADS PubMed 45 Berger J , Dorninger F , Forss-Petter S , et al. Peroxisomes in brain development and function . Biochim Biophys Acta 2016 ; 1863 ( 5 ): 934 – 55 . Google Scholar CrossRef Search ADS PubMed 46 Kou J , Kovacs GG , Hoftberger R , et al. Peroxisomal alterations in Alzheimer's disease . Acta Neuropathol 2011 ; 122 ( 3 ): 271 – 83 . Google Scholar CrossRef Search ADS PubMed 47 Kim EK , Choi EJ. Pathological roles of MAPK signaling pathways in human diseases . Biochim Biophys Acta 2010 ; 1802 ( 4 ): 396 – 405 . Google Scholar CrossRef Search ADS PubMed 48 Zhu X , Lee HG , Raina AK , et al. The role of mitogen-activated protein kinase pathways in Alzheimer's disease . Neurosignals 2002 ; 11 ( 5 ): 270 – 81 . Google Scholar CrossRef Search ADS PubMed 49 Penzes P , Vanleeuwen JE. Impaired regulation of synaptic actin cytoskeleton in Alzheimer's disease . Brain Res Rev 2011 ; 67 ( 1–2 ): 184 – 92 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Briefings in Bioinformatics Oxford University Press

Analysis of long noncoding RNAs highlights region-specific altered expression patterns and diagnostic roles in Alzheimer’s disease

Download PDF
11 pages

Loading next page...
 
/lp/ou_press/analysis-of-long-noncoding-rnas-highlights-region-specific-altered-JrY0jA0094

References (52)

Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected]
ISSN
1467-5463
eISSN
1477-4054
DOI
10.1093/bib/bby021
Publisher site
See Article on Publisher Site

Abstract

Abstract Increasing evidence has revealed the multiple roles of long noncoding RNAs (lncRNAs) in neurodevelopment, brain function and aging, and their dysregulation was implicated in many types of neurological diseases. However, expression pattern and diagnostic role of lncRNAs in Alzheimer’s disease (AD) remain largely unknown and has gained significant attention. In this study, we performed a comparative analysis for lncRNA expression profiles in four brain regions in brain aging and AD. Our analysis revealed age- and disease-dependent region-specific lncRNA expression patterns in aging and AD. Moreover, we identified a panel of nine lncRNAs (termed LncSigAD9) in a discovery cohort of 114 samples using supervised machine learning and stepwise selection method. The LncSigAD9 was able to differentiate between AD and healthy controls with high diagnostic sensitivity and specificity both in the discovery cohort (86.3 and 89.5%) and the additional independent AD cohort (90.8 and 83.8%). The receiver operating characteristic curves for the LncSigAD9 were 0.863 and 0.939 for discovery and independent cohorts, respectively. Furthermore, the LncSigAD9 demonstrated higher diagnostic performance than nine-minus-one lncRNA signature and mRNA-based signature with a similar number of genes. In silico functional analysis indicated the involvement of lncRNA expression variation in brain development- and metabolism-related biological processes. Taken together, our study highlights the importance of lncRNAs in brain aging and AD, and demonstrated the utility of lncRNAs as a promising biomarker for early AD diagnosis and treatment. Alzheimer’s disease, brain aging, biomarkers, long noncoding RNAs Introduction Alzheimer’s disease (AD) is a chronic neurodegenerative disease and remains the sixth leading cause of all deaths worldwide [1]. With the arrival of the aging of the world, we are facing tremendous challenges of the looming global epidemic of AD around the world. With the predicted demographic shift to increasingly elderly populations, it is estimated that the worldwide prevalence of AD will be approximately 100 million in 2050, which means 1 in 85 persons worldwide will be living with the disease [2, 3]. Patients with AD require resource-intensive care, and the cost of AD on society and health-care systems will be unendurable. Early diagnosis of AD and subsequent intervention to slow or halt disease progression become increasingly important. Thus, there is an increasing focus on the need to identify molecular biomarkers for accelerating early clinical diagnosis and treatment at early prodromal stages of AD. A growing body of evidence from large-scale transcriptomic studies has suggested that most of the human genome sequence is actively transcribed into large mRNA-like noncoding transcripts of >200 nucleotides in length, namely, long noncoding RNAs (lncRNAs) [4, 5]. There is increasing realization that lncRNAs are not transcriptional noise but play varied and essential roles in diverse biological phenomena by acting as a regulator of gene expression at epigenetic, transcriptional and posttranscriptional levels [6, 7]. Recent advancement in cancer genomics demonstrated the emerging roles as the hallmarks of cancer [8]. LncRNAs can act as an oncogene or/and tumor-suppressor involved in cancer initiation and progression [9, 10]. Increasing evidence also revealed the multiple roles of lncRNAs in neurodevelopment, brain function and aging, and the dysregulation of lncRNA expression was implicated in many types of neurological diseases, including AD [11–13]. Several lncRNAs, such as BACE1-AS, 51A, BC200 and NDM29, have been found to be aberrantly expressed in AD compared with healthy controls and were involved in AD pathogenesis by the low-throughput experiments [14–17]. Subsequent expression profiles analysis also identified hundreds of differentially expressed lncRNAs between AD samples and healthy controls in human and rat models [18–21], which gives strong evidence of the associations between altered lncRNA expression pattern and AD, and make them attractive as promising potential biomarkers and therapeutic targets for early AD diagnosis and treatment. In the present study, we reviewed the emerging roles and association of lncRNAs in AD and conducted a comprehensive analysis of lncRNAs expression profiles in four brain regions from a large cohort of >400 samples to characterize the expression pattern in aging and AD. In addition, we further assessed and analyzed the utility of lncRNA expression pattern as novel biomarkers for differentiating AD cases and healthy controls. Overall, our study highlights the importance of lncRNAs in brain aging and AD, and indicates that lncRNA may provide additional valuable information for early AD diagnosis and treatment. Materials and methods Sample data sets and study design In this study, two independent large data sets comprising of microarray data and clinical information of AD patients and healthy control samples were obtained from the publicly available Gene Expression Omnibus (GEO) database (GEO accession numbers are GSE48350 and GSE5281, respectively). The GSE48350 data set containing a large sample size with 253 tissue samples from four brain regions [entorhinal cortex (EC), hippocampus (HC), post-central gyrus (PCG) and superior frontal gyrus (SFG)] of 57 non-AD controls and 28 AD cases was adapted as primary data source to study dynamic expression pattern of lncRNA across multiple brain regions in aging and AD. For the purpose of lncRNA biomarkers discovery, 114 samples (57 pairs including 57 AD samples and 57 healthy control samples) were selected from 253 tissue samples of GSE48350 cohort as discovery cohort according to the paired design principle in terms of the following selection criteria: (1) the same gender; (2) the same detected brain region; and (3) the similar age (age difference <5 years). In addition, 161 samples (87 AD patient samples and 74 healthy control samples) from GSE5281 cohort were used as an additional independent AD cohort for biomarker validation. Acquisition and analysis of lncRNA expression profiles Raw microarray data (.CEL) of GSE48350 and GSE5281 measured with Affymetrix HG-U133_Plus_2.0 platform were downloaded from GEO database. Raw probe-level intensity values are background corrected, log2 transformed and quantile normalized using the Robust Multi-array Average (RMA) algorithm. LncRNA expression data were obtained by repurposing the probes from Affymetrix microarray based on the NetAffx annotation of the probe sets and the Refseq and GENCODE annotations as follows [22]: (1) for the probe sets with Refseq IDs, we only extracted those labeled as ‘NR_’ and annotated with ‘long non-coding RNA’ in Refseq database; (2) for the probe sets with Ensembl gene IDs, we only retained those annotated with ‘long non-coding RNA’ in GENCODE project (release 25); (3) finally, expression profiles of 2466 unique lncRNAs corresponding to 3431 probe sets were used for further analysis. A combination of significance analysis of microarrays (SAM) and variance filtering was used to evaluate differences in the expression of lncRNAs between different sample sets. Only those lncRNAs with a false discovery rate (FDR)-adjusted P-value < 0.2 and coefficient of variation (CV) value > 5% were considered as differentially expressed lncRNAs. Hierarchical clustering of lncRNA expression data was performed using R software with Euclidean distance and complete linkage, and the Pearson's chi-squared test was used to assess the significance of the association between sample clusters and samples’ status. Aging-AD continuum genes were identified as previously described [23]: (1) CV value > 5%; (2) have significant expression trend (P< 0.01) across young, aged and AD groups by linear regression analysis and (3) have a consistent directional change based on 80% confidence interval for the difference in expression means for each pairwise comparison. Statistics for classification and prediction A supervised machine learning method, support vector machine (SVM) with radial basis function kernel, was used to construct the lncRNA-based classifier for the classification of AD and healthy control samples using the R package e1071. To identify optimal lncRNA biomarkers form differentially expressed lncRNAs, we performed feature selection procedure using a stepwise selection method by a combination of forward selection and backward elimination approaches as follows: (1) each lncRNA was selected as the initial set; (2) searching through the remaining n-minus-1 lncRNAs to find out which two variables should be added to the current set to best improve the accuracy of classifier; (3) removing the least useful lncRNA from current classifier, one-at-a-time and (4) repeating Step 2 and Step 3 until all candidate lncRNAs have been picked by classifier. Binary classification performances were evaluated by estimating the accuracy, sensitivity and specificity of the lncRNA-based classifier with 5-fold cross-validation. In addition, a receiver operating characteristic (ROC) curve was plotted and the area under the ROC (AUC) was calculated to illustrate the discriminative power of the lncRNA-based classifier. In silico functional analysis of lncRNAs Co-expression relationships between lncRNAs and mRNAs were measured by calculating the Pearson correlation coefficient using the paired lncRNA and mRNA expression profiles in 253 samples of GSE48350 cohort. To infer the biological roles of lncRNAs, we performed functional enrichment analysis of Gene Ontology (GO) terms for co-expressed mRNAs using ClueGO plugin (version 2.3.3) in Cytoscape [24] limited in biological processes. Enriched GO terms and functional clusters were obtained and visualized based on a two-sided hypergeometric test with Bonferroni step down correction and kappa score threshold of 0.4, and limited in the GO-level intervals 3–8 with minimum gene 20 and P-value ≤ 0.05. In addition, we also conducted functional enrichment analysis of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway using DAVID Bioinformatics Resources (https://david.ncifcrf.gov/, version 6.8) [25]. Significantly enriched KEGG pathways were obtained using functional annotation chart options with the whole human genome as background when P-value <0.05. Results Region-specific altered lncRNA expression patterns in brain aging or AD To obtain a global view of lncRNA expression pattern in brain aging, lncRNA expression levels were compared in four brain regions (EC, HC, PCG and SFG) between young and aged healthy controls. A total of 383 lncRNAs reveal significant expression changes in four brain regions in aging (Supplementary File S1). Of them, only eight lncRNAs are dysregulated expressed in all the four brain regions, whereas others (375 lncRNAs, 97.9%) revealed a region-specific altered expression pattern in the course of aging (Figure 1A). Specifically, two brain regions, the PCG and SFG, showed most extensive lncRNA expression changes between the two age control groups (258 lncRNAs in the PCG and 247 lncRNAs in the SFG), whereas relative few lncRNA expression change was observed in the EC (70 lncRNAs) and HC (58 lncRNAs) (Figure 1A). Moreover, the majority of dysregulated lncRNAs would undergo increased expression in aging in three of four brain regions, particularly in the HC, whereas approximately 88.6% of dysregulated lncRNAs revealed downregulated expression with increasing age in the EC (Figure 1B). Then, we performed hierarchical clustering analysis for all samples in each of four brain regions according to expression levels of differentially expressed lncRNAs and found that altered expression pattern of lncRNA can distinguish the young and aged healthy controls for different brain regions (Figure 1C). Figure 1. View largeDownload slide Differential lncRNA expression pattern across four brain regions in aging. (A) Venn diagram for the number of dysregulated lncRNAs across four brain regions. (B) Number of downregulated and upregulated lncRNAs with increasing age in each brain region. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in each brain region. Figure 1. View largeDownload slide Differential lncRNA expression pattern across four brain regions in aging. (A) Venn diagram for the number of dysregulated lncRNAs across four brain regions. (B) Number of downregulated and upregulated lncRNAs with increasing age in each brain region. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in each brain region. To characterize AD-associated lncRNA expression pattern, lncRNA expression profiles were next compared in four brain regions between AD cases and aged healthy controls. Overall, 146 lncRNAs were detected as differentially expressed mainly in three brain regions (the EC, HC and SFG) except the PCG between AD cases and aged healthy controls (Supplementary File S2). Moreover, most of them (133 lncRNAs) undergo region-specific expression changes in AD cases, and only fewer lncRNA (13 lncRNAs) revealed expression changes in more than two regions (Figure 2A). Notably, all dysregulated lncRNAs revealed upregulated expression in the EC in AD cases, whereas the HC showed the largest number of downregulated lncRNAs. In addition, the SFG demonstrated the balanced distribution of upregulated and downregulated lncRNAs in AD cases (Figure 2B). Also, unsupervised hierarchical clustering of these dysregulated lncRNAs revealed AD-associated expression pattern differentiating AD cases from healthy controls (Figure 2C), which were further validated in another independent GSE5281 data set (Supplementary File S3). Figure 2. View largeDownload slide Differential lncRNA expression pattern across three brain regions in AD. (A) Venn diagram for the number of dysregulated lncRNAs across three brain regions. (B) Number of downregulated and upregulated lncRNAs in each brain region in AD. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in three brain regions. Figure 2. View largeDownload slide Differential lncRNA expression pattern across three brain regions in AD. (A) Venn diagram for the number of dysregulated lncRNAs across three brain regions. (B) Number of downregulated and upregulated lncRNAs in each brain region in AD. (C) Unsupervised hierarchical clustering of samples based on expression data of dysregulated lncRNAs in three brain regions. Because our above analysis revealed differences in lncRNA expression between young and aged healthy controls, and between AD cases and aged healthy controls, we next examined whether expression changes in lncRNA have a consistent directional trend from aging to AD. Using linear regression and pairwise comparison analysis, we identified 155 Aging-AD continuum lncRNAs in four brain regions (Supplementary File S4). These Aging-AD continuum lncRNAs also showed region-specific progressive expression changes, with fewer or no overlap for progressively upregulated or downregulated lncRNAs over aging and AD (Figure 3A). Although the SFG showed a large number of Aging-AD continuum lncRNAs, the distribution of progressively increased or decreased lncRNAs from aging to AD is nearly balanced in four brain regions (Figure 3B and C). Figure 3. View largeDownload slide Region-specific progressive expression changes of Aging-AD continuum lncRNAs. (A) Venn diagram for the number of lncRNAs undergoing increased or decreased expression across aging and AD. (B) Distribution of progressively increased or decreased lncRNAs from aging to AD in four brain regions. (C) Profiles of Aging-AD continuum lncRNAs in four brain regions. Green lines represent lncRNAs undergoing progressively decreased expression across aging and AD, and red lines represent lncRNAs undergoing progressively increased expression across aging and AD. Figure 3. View largeDownload slide Region-specific progressive expression changes of Aging-AD continuum lncRNAs. (A) Venn diagram for the number of lncRNAs undergoing increased or decreased expression across aging and AD. (B) Distribution of progressively increased or decreased lncRNAs from aging to AD in four brain regions. (C) Profiles of Aging-AD continuum lncRNAs in four brain regions. Green lines represent lncRNAs undergoing progressively decreased expression across aging and AD, and red lines represent lncRNAs undergoing progressively increased expression across aging and AD. Initial identification and validation of lncRNA biomarkers for early detection of AD In view of the region-specific altered lncRNA expression changes in aging or AD, a total of 114 samples (57 sample pairs) from GSE48350 was selected as discovery cohort using the paired design principle to detect potential lncRNA biomarkers, and the remaining 139 samples of GSE48350 (including 23 AD patient samples and 116 healthy control samples) were used as internal testing cohort. We first performed differentially expression analysis for lncRNAs between 57 AD patient samples and 57 healthy control samples in the discovery cohort. Overall, 47 lncRNAs were detected as differentially expressed between AD patient samples and healthy control samples (FDR-adjusted P-value < 0.01 and CV >5%) (Supplementary File S5). Of them, 11 lncRNAs were upregulated and 36 lncRNAs were downregulated in AD patient samples compared with healthy control samples. Then, we conducted unsupervised hierarchical clustering analysis on 114 samples of discovery cohort using the set of 47 differentially expressed lncRNAs. As expected, we observed that differentially expressed lncRNAs exhibited distinctive expression pattern, which clearly separated AD patient samples from healthy control samples (Fisher’s exact test P = 1.94E-09) (Figure 4A), suggesting that these 47 differentially expressed lncRNAs could be as candidate predictive biomarkers for early detection of AD. Figure 4. View largeDownload slide Identification and validation of lncRNA biomarkers for early detection of AD in the discovery cohort. (A) Unsupervised hierarchical clustering of 114 samples based on 47 differentially expressed lncRNAs. (B) The highest predicted accuracy of each lncRNA combination constructed by a specific number of lncRNAs (k = 1, 2,…, 47). (C) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the discovery cohort. (D) Performance comparison of 9-lncRNA diagnostic model and 1000 random 9-lncRNA combinations. Figure 4. View largeDownload slide Identification and validation of lncRNA biomarkers for early detection of AD in the discovery cohort. (A) Unsupervised hierarchical clustering of 114 samples based on 47 differentially expressed lncRNAs. (B) The highest predicted accuracy of each lncRNA combination constructed by a specific number of lncRNAs (k = 1, 2,…, 47). (C) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the discovery cohort. (D) Performance comparison of 9-lncRNA diagnostic model and 1000 random 9-lncRNA combinations. To identify optimal lncRNA biomarkers, we performed a classification and feature selection for 47 differentially expressed lncRNAs on 114 samples of discovery cohort using the SVM and stepwise selection model. The abovementioned analysis uncovered a combination of nine lncRNAs (MIR7-3HG, AL109615.3, NEBL-AS1, ATP6V0E2-AS1, PDXDC2P-NPIPB14P, LOC441204, A2M-AS1, TGFB2-OT1 and LINC00672), which yields a greatest discriminative ability with a predictive accuracy of 87.7% (Figure 4B). Therefore, this combination of nine lncRNAs was defined as an AD-related predictive signature (hereafter referred to as LncSigAD9). When tested in the discovery cohort using SVM and 5-fold cross-validation, the LncSigAD9 correctly classified 49 of 57 AD patient samples and 51 of 57 healthy control samples, achieving an AUC of 0.863 with a predictive accuracy of 87.7%, the sensitivity of 86.3% and the specificity of 89.5% (Figure 4C). We then generated 1000 random 9-lncRNA combinations by randomly selected nine lncRNAs and evaluated their performance using the same protocol as used for the LncSigAD9. As shown in the kernel density plots, the LncSigAD9 demonstrated superior performance compared with all random 9-lncRNA combinations (Figure 4D). Further validation of the LncSigAD9 with an additional independent AD cohort To evaluate the robustness of the LncSigAD9, further validation of the predictive ability of the LncSigAD9 was conducted using an additional independent AD cohort of 161 samples from GEO accession GSE5281. We first clustered 116 samples in the GSE5281 cohort according to the expression levels of nine lncRNAs in the LncSigAD9 by hierarchical clustering analysis and observed two distinct sample groups significantly correlated with associated their AD status (Fisher’s exact test P = 6.52E-12) (Figure 5A). In addition, we also performed classification analysis for 161 samples by LncSigAD9 using the SVM and 5-fold cross-validation as in the discovery cohort. As shown in Figure 5B, the LncSigAD9 showed a well predictive performance comparable with the discovery cohort. The validation result showed that the LncSigAD9 correctly classified 79 of 87 AD patient samples and 62 of 74 healthy control samples, achieving an AUC of 0.939 with a predictive accuracy of 87.6%, the sensitivity of 90.8% and the specificity of 83.8% (Figure 5B). Figure 5. View largeDownload slide Evaluation of robustness and reproducibility of the 9-lncRNA diagnostic model an additional independent AD cohort. (A) Unsupervised hierarchical clustering of 116 samples based on expression levels of nine lncRNAs in the LncSigAD9. (B) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the independent AD cohort. Figure 5. View largeDownload slide Evaluation of robustness and reproducibility of the 9-lncRNA diagnostic model an additional independent AD cohort. (A) Unsupervised hierarchical clustering of 116 samples based on expression levels of nine lncRNAs in the LncSigAD9. (B) ROC curves of the 9-lncRNA diagnostic model, other nine-minus-one lncRNA diagnostic model and mRNA-related diagnostic model in the independent AD cohort. To further investigate whether all of the nine lncRNAs in the LncSigAD9 are essential for its predictive value, we constructed all possible nine-minus-one lncRNA signatures by the removal of one lncRNA at a time and performed comparison analysis of predictive power for LncSigAD9 and other nine-minus-one lncRNA signatures using the SVM and 5-fold cross-validation in the discovery cohort and independent AD cohort. The comparison showed that none of the nine-minus-one lncRNA signatures yields better distinction than the LncSigAD9 between AD samples and healthy controls both in the discovery cohort (Figure 4C) and independent AD cohort (Figure 5B). This indicates that all nine lncRNAs are essential for the predictive power of the LncSigAD9. Comparison of the LncSigAD9 with existing mRNA-related biomarkers Finally, we compared the predictive value of the LncSigAD9 to mRNA-related biomarkers. Ten most predictive mRNAs from AclarusDx™ (ZNF267, CSPG2, KIAA1009, GALNT3, JAK2, TNNI3K, ANKRD49, BCL2A1, ROCK1 and CLEC2B) were selected to constitute a 10-mRNA signature (hereafter referred to as mRSigAD10) [26]. Hierarchical clustering analysis was first performed for samples of discovery cohort and independent AD cohort using expression pattern of 10 mRNAs in mRSigAD10. The expression pattern of 10 mRNAs revealed relatively poor sample classification performance than that of the LncSigAD9 both in the discovery cohort (Fisher’s exact test P = 0.0093 versus P = 1.94E-09) and independent AD cohort (Fisher’s exact test P = 0.0863 versus P = 6.52E-12) (Supplementary File S6). Then, we compared the sensitivity and specificity in the diagnosis of AD between the LncSigAD9 and mRSigAD10. Using the same SVM and 5-fold cross-validation, the mRSigAD10 obtained an AUC value of 0.678 with the sensitivity of 57.9% and the specificity of 70.2% in the discovery cohort (Figure 4C) and an AUC value of 0.897 with the sensitivity of 86.2% and the specificity of 85.1% in the independent AD cohort (Figure 5B), which are all lower than that of the LncSigAD9. Functional implication of the LncSigAD9 To gain a first insight into the biological function of the LncSigAD9, we first measured the association between expression levels of each lncRNA in the LncSigAD9 and that of mRNAs in 253 samples of GSE48350 cohort and identified positively co-expressed mRNAs with the LncSigAD9 (top 0.5%). Then, we performed GO and KEGG pathway function enrichment analysis for positively co-expressed mRNAs using clueGO and DAVID. GO analysis revealed that mRNAs positively correlated at least one of the nine signature lncRNAs were significantly enriched in some GO biological processes most of which can be categorized into five functional groups including ATP metabolic process, brain development, neurogenesis, neuron differentiation and development (Figure 6A). KEGG pathway analysis revealed 15 significant enriched biological pathways most of which are already known to be related to AD pathogenesis, including oxidative phosphorylation (OXPHOS), peroxisome, MAPK signaling pathway, FoxO signaling pathway, pyruvate metabolism, proteoglycans in cancer, metabolic pathways, melanoma, hypertrophic cardiomyopathy, AD, Parkinson’s disease, Huntington’s disease, regulation of actin cytoskeleton, carbon metabolism and arginine and proline metabolism, which have been reported to be important biological processes and pathways involved in AD pathogenesis. Figure 6. View largeDownload slide Functions enriched by the protein-coding genes co-expressed with nine lncRNAs in the LncSigAD9. (A) Functionally grouped network with enriched GO terms as nodes linked based on their kappa score level (≥0.4). Node size represents the term enrichment significance. (B) The most significantly enriched KEGG pathways. The node size represents the number of genes in the pathways, and the color represents the pathway enrichment significance. Figure 6. View largeDownload slide Functions enriched by the protein-coding genes co-expressed with nine lncRNAs in the LncSigAD9. (A) Functionally grouped network with enriched GO terms as nodes linked based on their kappa score level (≥0.4). Node size represents the term enrichment significance. (B) The most significantly enriched KEGG pathways. The node size represents the number of genes in the pathways, and the color represents the pathway enrichment significance. Discussion Previous microarray studies have characterized the expression pattern and functional roles of protein-coding genes in brain aging and AD [23, 27–30]. Recent advances in transcriptomic studies have discovered a new class of noncoding RNA transcripts (termed lncRNAs) and highlighted their critical roles in genome regulatory network and disease phenotypes [7, 31]. Despite the fact that several studies have reported lncRNA expression profiles and provided preliminary evidence for the involvement of lncRNAs in brain development and AD [19, 21, 32–34], the dynamic expression pattern of lncRNA across multiple brain regions in aging and AD remains largely unknown. In this study, we first compared differential lncRNA expression profiles of four brain regions in young and aged healthy controls and reported region-specific and age-dependent lncRNA expression pattern in brain aging. Our results demonstrated that the most extensive lncRNA expression changes were observed in the PCG and SFG, although all four brain regions showed differential responsiveness with increasing age, which is similar to protein-coding genes [23, 27, 28]. However, our analysis observed that lncRNA revealed opposite trend when compared with protein-coding genes. It has been reported that most of responsive protein-coding genes underwent decreased expression in the PCG and SFG [23, 27]. In contrast, extensive upregulated expression was observed with increased aging in the PCG, SFG and HC. These findings demonstrated that lncRNA plays important roles in transcriptional regulation in aging of brain and maybe a new molecular hallmark of brain aging. With the advances of genomics and transcriptomics in human complex disease, molecular profiles (such as mRNA profiling and miRNA profiling) are increasingly used as an alternative as diagnostic biomarkers for AD because of their noninvasion and easy accessibility compared with those from cerebrospinal fluid (SCF) [26, 35–37]. Until now, some efforts have been made to investigate the role of lncRNA in AD by focusing on some specific lncRNAs, as summarized by Luo and Chen [38]. Recently, several publications dealing with lncRNA expression profiles in AD reported a large number of dysregulated lncRNA. Despite these previously investigation, the expression pattern and diagnostic role of lncRNA in AD remain largely unknown. Therefore, to characterize the expression pattern of lncRNA in AD, we first compared differential lncRNA profiles of AD cases and aged healthy controls in four brain regions. This analysis demonstrated that extensive expression changes occur in 146 lncRNAs, most of which showing decreased expression in AD, which is in line with previous studies [18]. However, a notable feature of lncRNA expression pattern we found is that their alterations are regionally specific in AD, which has not been previously reported. In particular, the HC revealed the largest number of downregulated lncRNAs, and the EC revealed the largest number of upregulated lncRNAs in AD, and this could be because the HC and EC are not only affected in the early stages of AD but also are highly interactive components of a functionally connected network [39, 40]. Berchtold et al. [23] found that apparent expression changes of some synaptic genes in AD have been initiated to some degree already in normal aging. Therefore, we further examined whether some lncRNAs undergo progressive change across aging and AD and identified 155 Aging-AD continuum lncRNAs in four brain regions, suggesting that altered expression changes of these Aging-AD continuum lncRNAs in the AD brain have occurred to soma degree in the aged brain. Overall, these results suggested that lncRNA revealed age- and disease-dependent expression patterns in aging and AD, which could be served as molecular biomarkers for early detection of AD. As the first step toward identifying diagnostic lncRNA biomarkers, we conducted differential expression analysis of lncRNAs in 57 AD-control sample pairs using the paired design principle and identified 47 lncRNAs as potential biomarker candidates. To develop a clinically applicable lncRNA-based diagnostic marker, 47 candidate lncRNAs were subjected to the SVM and stepwise selection model to further narrow down the number of candidate lncRNAs. Finally, a panel of nine lncRNAs (termed LncSigAD9), including MIR7-3HG, AL109615.3, NEBL-AS1, ATP6V0E2-AS1, PDXDC2P-NPIPB14P, LOC441204, A2M-AS1, TGFB2-OT1 and LINC00672, with the highest accuracy was selected as the final signature. The LncSigAD9 achieved an AUC value of 0.863 for distinguishing AD patient samples and healthy controls in the discovery cohort. We further tested the performance of the LncSigAD9 by using an independent AD data set from GEO, in which the LncSigAD9 could differentiate AD patient samples from healthy controls with high specificity and sensitivity. These results demonstrated the reproducible predictive power and general applicability of the LncSigAD9 in determining the risk of AD. Furthermore, we also constructed all nine-minus-one lncRNA signatures and compared their predictive performance with LncSigAD9 in the discovery cohort and independent AD cohort. The comparison showed the necessity of all nine lncRNAs in the LncSigAD9 for the diagnosis of AD. Moreover, the LncSigAD9 revealed superior sensitivity and specificity than mRNA-based signature with a similar number of genes for AD diagnosis. To better understanding possible biological roles of the LncSigAD9 in the pathophysiology of AD, we tried to infer the potential functions of the identified LncSigAD9 by performing functional enrichment analysis for mRNAs co-expressed with nine lncRNAs in the LncSigAD9. GO analysis revealed that these lncRNAs are involved in brain development- and ATP metabolism-related biological processes. KEGG analysis indicated that genes whose expression correlated with the LncSigAD9 were enriched significantly in 15 KEGG pathways. Of the 15 KEGG pathways, 3 pathways are directly neurodegenerative disease-related KEGG pathways, including Alzheimer's disease, Parkinson’s disease and Huntington’s disease. Most of remaining enriched pathways are related to metabolism processes, including OXPHOS, metabolic pathways, arginine and proline metabolism, pyruvate metabolism, carbon metabolism, peroxisome and FoxO signaling pathway. Aberrations in metabolic processes have been reported to be involved in AD [41]. For example, OXPHOS is the classic role of mitochondria and is a vital part of metabolism that provides energy (ATP) for basal metabolism. Many previous studies have indicated abnormalities in OXPHOS in the pathogenesis of AD [42, 43]. In a mouse model, Isopi et al. [44] found that pyruvate could prevent the development and progression of AD-related cognitive deficits by reducing lipid peroxidation and oxidative stress. In the brain, peroxisomes are present in all neural cell types and peroxisomal alterations contributed to the progression of AD [45, 46]. Apart from abovementioned neurodegenerative disease-related and ATP metabolism-related pathways, other several enriched KEGG pathways, such as MAPK signaling pathway and regulation of actin cytoskeleton, have also been involved in the pathophysiology and pathogenesis of AD [47–49]. Overall, in silico functional analysis suggests that expression variation in the lncRNAs of the LncSigAD9 perhaps perturbs gene regulatory network, which may have profound effects on important biological processes and pathways involved in AD pathogenesis. In summary, our study represented a comprehensive analysis of lncRNAs in four brain regions and uncovered region-specific altered lncRNA expression pattern in aging and AD. In addition, our study investigated the diagnostic potential of lncRNA expression profiles and constructed a 9-lncRNA classifier, which could differentiate AD from normal controls with high diagnostic specificity and sensitivity in multiple cohorts. Functional analysis indicated the involvement of lncRNA expression variation in brain development- and metabolism-related biological processes. These findings highlight the importance of lncRNAs in the pathogenesis of AD and demonstrated the utility of lncRNAs as a promising biomarker to aid early AD diagnosis. However, further experimental studies or independent validation will be needed to fully elucidate the molecular mechanism and clinical value of lncRNAs in AD. Key Points We have highlighted the emerging roles and association of lncRNA in brain aging and AD. Our study demonstrated age- and disease-dependent expression patterns of lncRNAs in aging and AD, which could be served as molecular biomarkers for early detection of AD. We identified a robust and reproducible 9-lncRNA diagnostic model with high sensitivity and specificity in differentiating between AD and healthy controls. Supplementary Data Supplementary data are available online at https://academic.oup.com/bib. Funding This study was supported by the National Natural Science Foundation of China (grant number 61602134). Meng Zhou is an associate professor at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University. His research interests include bioinformatics and computational RNomics. Hengqiang Zhao is a student at the College of Bioinformatics Science and Technology, Harbin Medical University. His research interests include bioinformatics and disease systems biology. Xinyu Wang is a bioinformatics engineer at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University. His research interests include computational epigenomics and NGS. Jie Sun is an associate professor at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University, and College of Bioinformatics Science and Technology, Harbin Medical University. Her research interests include cancer bioinformatics and translational medicine. Jianzhong Su is a professor at the School of Ophthalmology and Optometry and Eye Hospital, School of Biomedical Engineering, Wenzhou Medical University. His research interests in bioinformatics and cancer epigenetics. References 1 Reitz C. Alzheimer's disease and the amyloid cascade hypothesis: a critical review . Int J Alzheimers Dis 2012 ; 2012 : 369808. Google Scholar PubMed 2 Brookmeyer R , Johnson E , Ziegler-Graham K , et al. Forecasting the global burden of Alzheimer's disease . Alzheimers Dement 2007 ; 3 ( 3 ): 186 – 91 . Google Scholar CrossRef Search ADS PubMed 3 Prince M , Bryce R , Albanese E , et al. The global prevalence of dementia: a systematic review and metaanalysis . Alzheimers Dement 2013 ; 9 ( 1 ): 63 – 75.e62 . Google Scholar CrossRef Search ADS PubMed 4 Consortium EP , Birney E , Stamatoyannopoulos JA , et al. Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project . Nature 2007 ; 447 : 799 – 816 . Google Scholar CrossRef Search ADS PubMed 5 Clark MB , Amaral PP , Schlesinger FJ , et al. The reality of pervasive transcription . PLoS Biol 2011 ; 9 ( 7 ): e1000625 . Google Scholar CrossRef Search ADS PubMed 6 Quinn JJ , Chang HY. Unique features of long non-coding RNA biogenesis and function . Nat Rev Genet 2016 ; 17 ( 1 ): 47 – 62 . Google Scholar CrossRef Search ADS PubMed 7 Rinn JL , Chang HY. Genome regulation by long noncoding RNAs . Annu Rev Biochem 2012 ; 81 : 145 – 66 . Google Scholar CrossRef Search ADS PubMed 8 Gutschner T , Diederichs S. The hallmarks of cancer: a long non-coding RNA point of view . RNA Biol 2012 ; 9 ( 6 ): 703 – 19 . Google Scholar CrossRef Search ADS PubMed 9 Tsai MC , Spitale RC , Chang HY. Long intergenic noncoding RNAs: new links in cancer progression . Cancer Res 2011 ; 71 ( 1 ): 3 – 7 . Google Scholar CrossRef Search ADS PubMed 10 Bartonicek N , Maag JL , Dinger ME. Long noncoding RNAs in cancer: mechanisms of action and technological advancements . Mol Cancer 2016 ; 15 ( 1 ): 43. Google Scholar CrossRef Search ADS PubMed 11 Barry G , Guennewig B , Fung S , et al. Long Non-Coding RNA Expression during aging in the human subependymal zone . Front Neurol 2015 ; 6 : 45. Google Scholar CrossRef Search ADS PubMed 12 Roberts TC , Morris KV , Wood MJ. The role of long non-coding RNAs in neurodevelopment, brain function and neurological disease . Philos Trans R Soc Lond B Biol Sci 2014 ; 369 ( 1652 ): 20130507. Google Scholar CrossRef Search ADS PubMed 13 Kim J , Kim KM , Noh JH , et al. Long noncoding RNAs in diseases of aging . Biochim Biophys Acta 2016 ; 1859 ( 1 ): 209 – 21 . Google Scholar CrossRef Search ADS PubMed 14 Faghihi MA , Modarresi F , Khalil AM , et al. Expression of a noncoding RNA is elevated in Alzheimer's disease and drives rapid feed-forward regulation of beta-secretase . Nat Med 2008 ; 14 ( 7 ): 723 – 30 . Google Scholar CrossRef Search ADS PubMed 15 Ciarlo E , Massone S , Penna I , et al. An intronic ncRNA-dependent regulation of SORL1 expression affecting Abeta formation is upregulated in post-mortem Alzheimer's disease brain samples . Dis Model Mech 2013 ; 6 ( 2 ): 424 – 33 . Google Scholar CrossRef Search ADS PubMed 16 Lin D , Pestova TV , Hellen CU , et al. Translational control by a small RNA: dendritic BC1 RNA targets the eukaryotic initiation factor 4A helicase mechanism . Mol Cell Biol 2008 ; 28 ( 9 ): 3008 – 19 . Google Scholar CrossRef Search ADS PubMed 17 Massone S , Ciarlo E , Vella S , et al. NDM29, a RNA polymerase III-dependent non coding RNA, promotes amyloidogenic processing of APP and amyloid beta secretion . Biochim Biophys Acta 2012 ; 1823 ( 7 ): 1170 – 7 . Google Scholar CrossRef Search ADS PubMed 18 Zhou X , Xu J. Identification of Alzheimer's disease-associated long noncoding RNAs . Neurobiol Aging 2015 ; 36 ( 11 ): 2925 – 31 . Google Scholar CrossRef Search ADS PubMed 19 Yang B , Xia ZA , Zhong B , et al. Distinct hippocampal expression profiles of long non-coding RNAs in an Alzheimer's disease model . Mol Neurobiol 2017 ; 54 ( 7 ): 4833 – 46 . Google Scholar CrossRef Search ADS PubMed 20 Magistri M , Velmeshev D , Makhmutova M , et al. Transcriptomics profiling of Alzheimer's disease reveal neurovascular defects, altered amyloid-beta homeostasis, and deregulated expression of long noncoding RNAs . J Alzheimers Dis 2015 ; 48 ( 3 ): 647 – 65 . Google Scholar CrossRef Search ADS PubMed 21 Lee DY , Moon J , Lee ST , et al. Distinct expression of long non-coding RNAs in an Alzheimer's disease model . J Alzheimers Dis 2015 ; 45 ( 3 ): 837 – 49 . Google Scholar PubMed 22 Zhang X , Sun S , Pu JK , et al. Long non-coding RNA expression profiles predict clinical phenotypes in glioma . Neurobiol Dis 2012 ; 48 ( 1 ): 1 – 8 . Google Scholar CrossRef Search ADS PubMed 23 Berchtold NC , Coleman PD , Cribbs DH , et al. Synaptic genes are extensively downregulated across multiple brain regions in normal human aging and Alzheimer's disease . Neurobiol Aging 2013 ; 34 ( 6 ): 1653 – 61 . Google Scholar CrossRef Search ADS PubMed 24 Bindea G , Mlecnik B , Hackl H , et al. ClueGO: a Cytoscape plug-in to decipher functionally grouped gene ontology and pathway annotation networks . Bioinformatics 2009 ; 25 ( 8 ): 1091 – 3 . Google Scholar CrossRef Search ADS PubMed 25 Huang da W , Sherman BT , Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources . Nat Protoc 2009 ; 4 ( 1 ): 44 – 57 . Google Scholar CrossRef Search ADS PubMed 26 Fehlbaum-Beurdeley P , Sol O , Desire L , et al. Validation of AclarusDx, a blood-based transcriptomic signature for the diagnosis of Alzheimer's disease . J Alzheimers Dis 2012 ; 32 : 169 – 81 . Google Scholar PubMed 27 Berchtold NC , Cribbs DH , Coleman PD , et al. Gene expression changes in the course of normal brain aging are sexually dimorphic . Proc Natl Acad Sci USA 2008 ; 105 ( 40 ): 15605 – 10 . Google Scholar CrossRef Search ADS PubMed 28 Cribbs DH , Berchtold NC , Perreau V , et al. Extensive innate immune gene activation accompanies brain aging, increasing vulnerability to cognitive decline and neurodegeneration: a microarray study . J Neuroinflammation 2012 ; 9 ( 1 ): 179 . Google Scholar CrossRef Search ADS PubMed 29 Liu G , Bao X , Wang R. Expression quantitative trait loci regulate HNF4A and PTBP1 expression in human brains . Proc Natl Acad Sci USA 2015 ; 112 ( 30 ): E3975. Google Scholar CrossRef Search ADS PubMed 30 Xiang Z , Xu M , Liao M , et al. Integrating genome-wide association study and brain expression data highlights cell adhesion molecules and purine metabolism in Alzheimer's disease . Mol Neurobiol 2015 ; 52 ( 1 ): 514 – 21 . Google Scholar CrossRef Search ADS PubMed 31 Schmitz SU , Grote P , Herrmann BG. Mechanisms of long noncoding RNA function in development and disease . Cell Mol Life Sci 2016 ; 73 ( 13 ): 2491 – 509 . Google Scholar CrossRef Search ADS PubMed 32 Aprea J , Prenninger S , Dori M , et al. Transcriptome sequencing during mouse brain development identifies long non-coding RNAs functionally involved in neurogenic commitment . EMBO J 2013 ; 32 ( 24 ): 3145 – 60 . Google Scholar CrossRef Search ADS PubMed 33 Lv J , Cui W , Liu H , et al. Identification and characterization of long non-coding RNAs related to mouse embryonic brain development from available transcriptomic data . PLoS One 2013 ; 8 ( 8 ): e71152 . Google Scholar CrossRef Search ADS PubMed 34 Shi C , Zhang L , Qin C. Long non-coding RNAs in brain development, synaptic biology, and Alzheimer's disease . Brain Res Bull 2017 ; 132 : 160 – 9 . Google Scholar CrossRef Search ADS PubMed 35 Sood S , Gallagher IJ , Lunnon K , et al. A novel multi-tissue RNA diagnostic of healthy ageing relates to cognitive health status . Genome Biol 2015 ; 16 ( 1 ): 185 . Google Scholar CrossRef Search ADS PubMed 36 Leidinger P , Backes C , Deutscher S , et al. A blood based 12-miRNA signature of Alzheimer disease patients . Genome Biol 2013 ; 14 ( 7 ): R78 . Google Scholar CrossRef Search ADS PubMed 37 Kumar P , Dezso Z , MacKenzie C , et al. Circulating miRNA biomarkers for Alzheimer's disease . PLoS One 2013 ; 8 ( 7 ): e69807 . Google Scholar CrossRef Search ADS PubMed 38 Luo Q , Chen Y. Long noncoding RNAs and Alzheimer’s disease . Clin Interv Aging 2016 ; 11 : 867. Google Scholar CrossRef Search ADS PubMed 39 van Strien NM , Cappaert NL , Witter MP. The anatomy of memory: an interactive overview of the parahippocampal-hippocampal network . Nat Rev Neurosci 2009 ; 10 ( 4 ): 272 – 82 . Google Scholar CrossRef Search ADS PubMed 40 Braak H , Braak E. Neuropathological stageing of Alzheimer-related changes . Acta Neuropathol 1991 ; 82 ( 4 ): 239 – 59 . Google Scholar CrossRef Search ADS PubMed 41 Chen J , Xie C , Zhao Y , et al. Gene expression analysis reveals the dysregulation of immune and metabolic pathways in Alzheimer's disease . Oncotarget 2016 ; 7 ( 45 ): 72469 – 74 . Google Scholar PubMed 42 Biffi A , Sabuncu MR , Desikan RS , et al. Genetic variation of oxidative phosphorylation genes in stroke and Alzheimer's disease . Neurobiol Aging 2014 ; 35 ( 8 ): 1956.e1 – 8 . Google Scholar CrossRef Search ADS 43 Shoffner JM. Oxidative phosphorylation defects and Alzheimer's disease . Neurogenetics 1997 ; 1 ( 1 ): 13 – 9 . Google Scholar CrossRef Search ADS PubMed 44 Isopi E , Granzotto A , Corona C , et al. Pyruvate prevents the development of age-dependent cognitive deficits in a mouse model of Alzheimer's disease without reducing amyloid and tau pathology . Neurobiol Dis 2015 ; 81 : 214 – 24 . Google Scholar CrossRef Search ADS PubMed 45 Berger J , Dorninger F , Forss-Petter S , et al. Peroxisomes in brain development and function . Biochim Biophys Acta 2016 ; 1863 ( 5 ): 934 – 55 . Google Scholar CrossRef Search ADS PubMed 46 Kou J , Kovacs GG , Hoftberger R , et al. Peroxisomal alterations in Alzheimer's disease . Acta Neuropathol 2011 ; 122 ( 3 ): 271 – 83 . Google Scholar CrossRef Search ADS PubMed 47 Kim EK , Choi EJ. Pathological roles of MAPK signaling pathways in human diseases . Biochim Biophys Acta 2010 ; 1802 ( 4 ): 396 – 405 . Google Scholar CrossRef Search ADS PubMed 48 Zhu X , Lee HG , Raina AK , et al. The role of mitogen-activated protein kinase pathways in Alzheimer's disease . Neurosignals 2002 ; 11 ( 5 ): 270 – 81 . Google Scholar CrossRef Search ADS PubMed 49 Penzes P , Vanleeuwen JE. Impaired regulation of synaptic actin cytoskeleton in Alzheimer's disease . Brain Res Rev 2011 ; 67 ( 1–2 ): 184 – 92 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Briefings in Bioinformatics – Oxford University Press

Published: Apr 17, 2018

There are no references for this article.