Advances in transcriptome analysis of human brain aging

Aging is associated with gradual deterioration of physiological and biochemical functions, including cognitive decline. Transcriptome profiling of brain samples from individuals of varying ages has identified the whole-transcriptome changes that underlie age-associated cognitive declines. In this review, we discuss transcriptome-based research on human brain aging performed by using microarray and RNA sequencing analyses. Overall, decreased synaptic function and increased immune function are prevalent in most regions of the aged brain. Age-associated gene expression changes are also cell dependent and region dependent and are affected by genotype. In addition, the transcriptome changes that occur during brain aging include different splicing events, intersample heterogeneity, and altered levels of various types of noncoding RNAs. Establishing transcriptome-based hallmarks of human brain aging will improve the understanding of cognitive aging and neurodegenerative diseases and eventually lead to interventions that delay or prevent brain aging.


Introduction
Aging is a nearly universal biological process associated with gradual deterioration in physiological and biochemical functions, including cognitive decline 1,2 . Aging is also a major risk factor for many neurodegenerative diseases, including Alzheimer's disease and Parkinson's disease 2,3 . This close relationship between aging and neurodegenerative diseases suggests the existence of common transcriptional and posttranscriptional gene regulation mechanisms in the brain.
Age-associated cognitive declines usually correlate with corresponding age-related structural changes in the brain, such as neuronal loss and synaptic changes 1,4 . However, the correlation is not universal. For example, changes in gene expression profiles in the brain during aging can precede apparent histopathological degeneration of the brain and appear to underlie age-associated cognitive declines at the molecular level [5][6][7][8] . Therefore, transcriptome analysis of brain aging offers a systematic molecular approach to understand the causes of neurocognitive aging and neurodegenerative diseases.
The transcriptome is defined as "the complete complement of mRNA molecules generated by a cell or population of cells" 9,10 . The information of cell populations is stored in the genome; therefore, transcriptome profiling is important for interpreting functional genomic elements 11 . The objectives of transcriptomics, or the study of the transcriptome, are to categorize all transcripts, to analyze the variants of expressed genes and to quantify the levels of transcripts under different conditions 12 . The subsequent functional characterization of the genes and transcripts of interest obtained from transcriptome analyses also helps validate their biological relevance.
Transcriptome profiling of brain samples from individuals of varying ages is a useful strategy for studying human brain aging by allowing the detection of overall age-associated transcriptome changes. Although transcriptome profiling alone generates descriptive information, assessment and interpretation of this information allows researchers to effectively design subsequent functional analyses of candidate genes and transcripts 11,13 . Table 1 Representative transcriptomic studies of human brain aging discussed in this manuscript. Because human postmortem tissues are difficult to obtain, progress in transcriptome profiling of human brain aging has been very slow. However, international collaboration for the collection of human brain samples and subsequent generation of various sets of experimental data has paved the way for molecular biological research of human brain aging [14][15][16] . This collaboration has thus contributed to the comprehensive understanding of aging-related transcriptomic changes in the human brain.
In this article, we review and discuss studies based on transcriptome analyses of human brain aging (Table 1). We categorize the subtopics of transcriptome analysis into global gene expression changes, cell type-specific expression changes, gene coexpression networks, expression quantitative trait loci, and alternative splicing.
Decreased synaptic function and increased immune function are prevalent in most analyses, in addition to other features that have been detected in some studies. Increased intersample heterogeneity is also an indispensable factor in the understanding of human brain aging. In addition, we address age-associated expression changes in various types of noncoding RNAs, including microRNAs (miRNAs), long noncoding RNAs (lncRNAs), and circular RNAs (circRNAs). Transcriptome analysis of human brain aging will pave the way for the early diagnosis and prevention of cognitive aging and ageassociated brain diseases.
Age-dependent transcriptome changes in the brain include decreased synaptic function and increased immune function One of the most extensive transcriptome analyses on human brain aging with a large sample size and diverse ethnicity provided insights into global changes across the whole brain, as well as in specific regions 17 . Generally, temporal regulation of protein-coding genes is more pronounced than spatial or regional regulation (Fig. 1). The expression of 57.7% of genes is changed during prenatal development. In addition, 9.1% of the gene transcripts that are expressed in the neocortex are changed during postnatal development (an age of under 20 years), but only 0.7% of them are altered during adulthood (an age of over 20 years). These data indicate that the majority of differentially expressed genes display temporal changes during development and that the effects of aging on gene expression changes are relatively minor.
However, research using adult brains has indicated noticeable transcriptional changes during adult aging ( Fig.  1). Another comprehensive transcriptome study analyzing the prefrontal cortex of 269 human individuals indicates that reduced rates of gene expression changes in the early postnatal brain are maintained throughout early childhood and through the first four decades of life 18 . Thereafter, substantial gene expression changes across multiple brain regions are detected in the sixth and seventh decades of human life, based on an analysis mostly including people of Caucasian ancestry 6,18 , indicating the existence of a qualitative transition phase (Fig. 1). The majority of the gene expression changes observed during adult aging are similar to those detected during postnatal development, which includes infancy, childhood, and puberty; interestingly, the directions of these changes are opposite those observed during fetal development 18,19 (Fig. 1). In addition, males display a larger number of gene expression changes across multiple brain regions with aging than females 6,17 . In contrast to this sexual dimorphism, a study that analyzed two large data sets containing over 1800 human samples shows that the expression of genes exhibits similar changes with aging across brain regions 20 .

REST TP73
Microglial genes C1QA CX3CR1 Fig. 1 Global age-dependent transcriptomic changes in human brains. Expression changes are mainly temporal by relative preservation of their spatial identity. Substantial expression changes are detected in the sixth and seventh decades of human life. The directions of gene expression changes during adult aging tend to be opposite those occurring during fetal development. Throughout multiple brain regions, age-dependent transcriptomic changes include decreased synaptic function and increased immunity. It is worth noting that synaptic aging occurs even prior to neuronal loss. Decreased synaptic function underlies increased expression of repressor element-1-silencing transcription factor (REST) and decreased expression of the tumor protein 73 (TP73) gene. Increased immune responses are accompanied by upregulation of microglial genes and complement component 1q A (C1QA) and downregulation of genes encoding immunosuppressive factors, including C-X3-C motif chemokine receptor 1 (CX3CR1).
Notably, many independent transcriptome analyses have commonly indicated that genes that are upregulated during aging in most brain regions are enriched in inflammatory or immune-responsive genes annotated in the Gene Ontology (GO) database 5,6,[20][21][22][23][24][25] (Fig. 1). Although appropriate immune activation is neuroprotective, chronic immune activation appears to increase the vulnerability of the brain to neurocognitive aging and neurodegenerative diseases 22,23,26 . In contrast, multiple studies have repeatedly reported that genes that are downregulated across multiple brain regions during aging are enriched with genes related to synaptic transmission and plasticity [5][6][7]21,24,25,27,28 (Fig. 1). These findings suggest that age-associated cognitive decline is driven by synaptic aging without neuronal loss 1,5 . Consistent with these observations, accumulating evidence indicates that dysregulation of intracellular calcium ion homeostasis is linked to age-related cognitive declines 29,30 . Genes that exhibit age-dependent expression changes in brain regions are enriched in other biological processes as well; these processes include neurotransmitter transport, energy metabolism, chaperone response, apoptosis, RNA metabolism, amino acid biosynthesis, DNA repair, mitochondrial function, and reactive oxygen species metabolism 5,19,21,27,[31][32][33] . Many of these processes appear to be neuroprotective.
Recent transcriptomic studies in combination with thorough molecular genetic analysis indicate that repressor element-1-silencing transcription factor (REST), a neuronal repressor that is crucial for brain development, functions as a neuroprotective factor against brain aging 34 . The expression levels of REST are increased in normal aged brains but not in the brains of individuals with mild cognitive impairment or Alzheimer's disease 25,34 . REST downregulates excitatory synaptic transmission and upregulates the longevity-associated transcription factor forkhead box O1 (FOXO1) in the mammalian brain, as well as its homolog DAF-16 in Caenorhabditis elegans to promote longevity 25 . These studies regarding REST suggest the existence of active protective mechanisms in the brain against neurodegeneration and the contribution of reduced excitatory synaptic transmission to longevity (Fig. 1).
Age-dependent gene expression changes in specific cell types have also been detected in diverse brain regions. Multiple papers have commonly reported that neuronspecific genes are predominantly downregulated in all brain regions during aging 7,19,21,33,35 , but these genes display region-specific, age-dependent expression changes by preserving their regional identity 20 . In contrast, gliaspecific genes with age-dependent alterations in their expression lose their regional identity, and their expression is shifted to multiple regions 20 . Microglia act as innate immune cells of the central nervous system 36 .
During aging, microglia-specific genes are generally upregulated in all brain regions, with the exception of the cerebellum 20,21,23,26 (Fig. 1). In addition, endothelial cellspecific genes are transcriptionally upregulated across multiple brain regions during aging 20 . Oligodendrocytes act as supporting cells for neurons in the central nervous system by forming myelin sheaths around axons 37 . In contrast to the expression level changes in microgliaspecific and endothelial cell-specific genes, the expression levels of oligodendrocyte-specific genes tend to decrease in various brain regions during aging 20,35 . These findings provide evidence supporting the occurrence of global and cell-or region-specific transcriptional changes during human brain aging.

Gene coexpression network analysis supports decreased synaptic function and increased immune function during brain aging
Aging is regulated by complex interplay among various components; 2,3,8 therefore, the transcriptomes of aged samples usually exhibit high intersample variation. Gene coexpression network analysis is utilized for detecting modules or clusters of highly correlated genes associated with specific functions and for obtaining key information embedded in heterogeneous transcriptomes 38 (Fig. 2a). A network analysis of published transcriptomic data obtained from adult hippocampus samples has indicated that a module of genes that display age-dependent upregulation is enriched in myelination 39 . This study also confirmed an immune process-enriched module that correlated positively with age and a synaptic transmissionenriched module that correlated negatively with age 39 .
Although decreased synaptic function has been reproducibly reported in aged prefrontal cortices 40 , the increase in immune function in aged brains has been disputed by one study that analyzed three data sets obtained from the frontal cortices of 381 individuals 41 . Specifically, the expression levels of multiple genes in a microglia-specific module, including microglia surface receptors and Tolllike receptors, decrease in the prefrontal cortex with age 41 . Thus, some functions of microglia appear to decline during brain aging. However, the microglia module also contains genes encoding immunosuppressive factors, including C-X3-C motif chemokine receptor 1 (CX 3 CR1); 41 therefore, the balance between immune activation and suppression is important for appropriate immune responses 6,22 . For instance, the microglia-specific neuroprotective receptor for fractalkine (FKN), CX 3 CR1, suppresses excessive immune responses via microglia activation in mice and rats 42,43 . Signaling through the FKN ligand and its receptor, CX 3 CR1, in microglia is reduced in aged brains 22,41,43 , and this reduction likely contributes to neurotoxic inflammation and results in neurodegeneration (Fig. 1). However, we cannot rule out the possibility that age-dependent downregulation of a subset of microglia-specific genes may reflect the complex regulation of these genes, as well as the altered number of microglia during aging. Collectively, these findings based on experimental data from multiple studies show that decreased synaptic function and increased immune function are predominant during human brain aging.
Glia-specific genes lose their regional identity during brain aging Examination of the age-dependent changes in the expression of cell-type-specific genes reveals that genes specific to glia, including astrocytes and oligodendrocytes, exhibit altered regional expression patterns during brain aging 20 (Fig. 2b). These changes in the regional expression of glia-specific genes is most prominent in the hippocampus in the forebrain and in the substantia nigra in the midbrain, both of which are susceptible to Alzheimer's disease and Parkinson's disease pathogenesis. For example, the expression of astrocyte-specific genes in the hippocampus is similar to that observed in the cortex in young brains; however, the expression of these genes shifts toward that observed in intralobular white matter and the putamen in aged brains. Oligodendrocyte-specific genes exhibit differential expression in the hippocampus and in the substantia nigra in young brains; however, their expression in these structures becomes similar in aged brains. In contrast, neuron-specific genes preserve their regional identity during brain aging (Fig. 2b). Thus, during human brain aging, the expression of genes in specific cell types, including astrocytes and oligodendrocytes, changes dramatically and regional identities are lost.

Analysis of expression quantitative trait loci (eQTLs) reveals that genetic backgrounds affect gene expression in an age-dependent manner
Quantitative trait locus (QTL) analysis is a method that is used to infer the effects of genetic background on different traits. Expression QTLs (eQTLs) are genomic loci that are highly related to the expression levels of target genes 44,45 . Many eQTLs show consistent effects on the expression levels of their targets independent of age 17,18 . Interestingly, however, several studies using less stringent thresholds or a particular type of cell, such as microglia, have reported that some of these loci display differential effects on the brain at different ages 26,32 . One example is the single-nucleotide polymorphism (SNP) rs55675298 32 ; individuals with the GG allele show an age-dependent increase in the expression of the tumor protein 73 (TP73) gene, which encodes a member of the p53 protein family, whereas individuals with the GT or TT alleles do not exhibit this increase (Fig. 3a). TP73 is critical for normal neuronal development and survival in mice [46][47][48] ; therefore, age-dependent changes in the expression of TP73 potentially affect brain aging and age-related diseases (Fig.  1). Indeed, the allele frequency of TP73 is implicated in human Alzheimer's disease 49 , and haploinsufficiency of TP73 confers susceptibility to Alzheimer's disease in model mice 50 . Similarly, the GG allele of the SNP rs72788737 is associated with age-dependent increases in the expression of complement component 1q A (C1QA), which encodes the C1q protein responsible for innate immune responses, whereas the GT or TT allele is not 32 . The expression levels of C1QA increase during normal brain aging (Fig. 1), and the C1q protein accumulates in several brain regions, including the hippocampus, Targeted analysis of transcriptomic changes during human brain aging. a Age-dependent upregulation of myelination-related genes, which is undetectable in global analysis, is detected via gene coexpression network analysis. Genes that exhibit highly positive correlations among themselves are connected to one another. Different colors represent different groups of genes related to specific functions. b Glia-specific genes lose their regional identity during brain aging, whereas neuron-specific genes preserve their regional identity. Glial markers include genes that are specifically expressed in astrocytes and oligodendrocytes. The dimensions of the transcriptome are reduced for visualization by principal component analysis or multidimensional scaling.
substantia nigra, and piriform cortex, which are vulnerable to neurodegeneration 51 . In addition, the number of neuroprotective apolipoprotein E (APOE) gene ε2 haplotypes but not APOE ε4 haplotypes is associated with reduced expression levels of microglia-enriched genes in aged brains 26 . This pattern suggests that the APOE ε2 haplotypes moderate the excessive microglial activation that occurs during aging. Overall, these studies show the interaction between the genetic background and agedependent gene expression changes, which may in turn affect human brain aging and age-associated diseases.
Analysis of alternative splicing variants suggests the role of RNA-binding proteins in brain aging In addition to gene expression changes, two notable papers that analyzed splicing changes have indicated that human brain aging is associated with alternative splicing 17,52 . Variations in alternative splicing are predominant during prenatal development 17 . After birth, 0.9% of the genes expressed in the neocortex exhibit differential splicing during postnatal development, whereas 1.4% of them do so during adulthood 17 . As the brain exhibits the highest level of splicing and preferential expression of genes encoding RNA-binding proteins among all organs of the body 16 , the 1.4% rate of agedependent splicing changes is substantial compared with the 0.9% rate.
Remarkably, age-dependent splicing changes do not correlate with changes in the expression of the corresponding genes 52 . The splicing changes that occur during postnatal development and during subsequent aging processes display opposite trends; 52 75% of segments (exons or introns) show decreased inclusion in transcripts during postnatal development, whereas 77% of segments show increased inclusion during aging. The majority of segments that follow this "down-up" inclusion pattern are retained introns 52 . Transcripts with retained introns can be targets for nonsense-mediated decay (NMD), which is a crucial system for RNA quality control that degrades various mRNAs, including those with premature stop codons, and regulates alternative splicing 53 . Among the 15 of the 21 genes encoding NMD pathway components that are expressed in the brain, five [up-frameshift suppressor 3 homolog B (UPF3b), suppressor with morphological effect on genitalia homolog 1 (SMG1), SMG5, SMG6, and cancer susceptibility candidate 3 (CASC3)] follow the "down-up" pattern 52 . These findings suggest that components acting in the NMD pathway display characteristic age-dependent expression changes. NMD plays crucial roles in organismal longevity in invertebrates 54,55 . Thus, it will be interesting to evaluate whether the age-dependent expression changes in NMD pathway componentencoding genes contribute to brain aging or ageassociated neurodegenerative diseases.
The expression levels of six genes encoding splicing factors, i.e., polypyrimidine tract-binding proteins 1 and 2 (PTBP1 and PTBP2) and heterogeneous nuclear ribonucleoproteins A1, H1, H3, and F (hnRNPA1, hnRNPH1, hnRNPH3, and hnRNPF), follow the age-dependent "down-up" expression pattern 52 . The expression levels of these genes correlate positively with the "down-up" splicing pattern of particular genes and correlate negatively with the "up-up" or "up-down" splicing patterns of other genes. These relationships may reflect the idea that PTBP1 and PTBP2, which are located in the vicinity of spliced segments 56 , positively or negatively affect segment inclusion 57 . Indeed, a large proportion of PTB targets are segments that exhibit age-related splicing changes 35 .
Analysis of splicing QTLs (sQTLs), also called splicingisoform QTLs (isoQTLs), identified 9,044 genomic loci associated with splicing events in 3,006 genes in the aged prefrontal cortex, including the SNP rs2439540 58 (Fig.  3b). Although most of these sQTLs do not appear to affect gene expression, 42 loci correlate with the expression and splicing of the genes that contain the loci 58 . In addition, the 9,044 age-associated sQTLs are enriched with potential binding of 18 RNA-binding proteins (RBPs); 58 PTBP1 is the most prevalently bound RBP, followed by HNRNPC, cleavage and polyadenylation specific factor 7 (CPSF7), and ELAVL1 [embryonic lethal, abnormal vision (ELAV)-like RNA-binding protein 1]. The age-associated sQTLs are also enriched with other hnRNPs 58 . For example, the frequency of the G allele of the SNP rs2439540, which correlates negatively with intron usage in the TBC1 domain family member 7 (TBC1D7) gene, correlates positively with binding of HNRNPA2B1 (Fig.  3b). Moreover, the mRNA levels of hnRNP splicing factors, including HNRNPA2B1 and HNRNPC, correlate negatively with the intron usage levels of hundreds of genes (Fig. 3b), some of which are associated with Alzheimer's disease 58 . These findings suggest that these RBPs, as well as PTBP1 and PTBP2, play roles in the agedependent changes in splicing.
The increased variability of gene expression indicates impairment of genome maintenance and repair with age As described above, the transcriptome of the aged brain after the sixth and seventh decades of human life becomes similar to that of the fetal brain 18,19 . However, these changes are not consistently detected between cells or tissues, because the gene expression profiles of brains become heterogeneous during aging 5,27 . This increased variability of gene expression appears to be caused by accumulation of cellular damage and somatic mutations rather than by differences in germline mutations or aging rates between individuals 2,59-61 . Indeed, a pioneering paper that analyzed microarray data obtained from humans of various ages reported that DNA damage is highly enriched in the promoters of genes that are downregulated in aged brains 5 . This finding is consistent with reports showing that gene expression variability is more pronounced among age-dependently downregulated genes than age-dependently upregulated genes 33,60 . The expression of many age-dependently downregulated genes correlates with increased DNA damage 5 , indicating accelerated genome instability during brain aging. The susceptibility of specific brain regions, such as the hippocampus, substantia nigra, and basal ganglia, to DNA damage may be involved in the increased vulnerability of these brain regions to aging compared with that of other regions 62 . Prolonged accumulation of cellular defects often results in immune activation, which causes additional damage 63 , consistent with the increased immune responses observed in aged brains. In summary, agedependent downregulation of genes associated with increased genome instability may underlie the accumulation of stochastic events in aged brain cells. Old cells subsequently become heterogeneous and inflammatory, and these changes may eventually lead to gradual declines in the physiological integrity of the brain.
Noncoding RNAs that participate in human brain aging Transcriptomic studies have identified many novel classes of noncoding RNAs (ncRNAs) that are crucial for brain physiology 64,65 . Moreover, functional analyses have established ncRNAs as key regulators of brain aging 31,66,67 . In this section, we highlight the major classes of ncRNAs, i.e., microRNAs (miRNAs), long noncoding RNAs (lncRNAs), and circular RNAs (circRNAs), that participate in human brain aging (Fig. 4). Function in aging?

Inflammation
Changes in noncoding RNAs during human brain aging Fig. 4 Noncoding RNAs that participate in human brain aging. a Several microRNAs (miRNAs) enhance neuroprotection in the elderly at the cost of inflammation. The expression of miR-29a/b, miR-34a, and miR-144 increases with age, whereas that of miR-222 decreases. b Long noncoding RNAs (lncRNAs) display age-dependent expression changes with preserved regional specificity. In the prefrontal cortex, the expression levels of the antisense RNA of the gene encoding proline-rich protein (PRR34-AS1), LINC01094, and LINC00844 correlate positively with age. In contrast, the expression levels of the antisense RNA of the gene encoding opacity-associated (Opa) interacting protein 1 (OIP5-AS1), MIR7-3 host gene (MIR7-3HG), LINC00643, LINC00507, and brain cytoplasmic 200 (BC200) correlate negatively with age. In the subependymal zone, GOMAFU, metastasis-associated lung adenocarcinoma transcript 1 (MALAT1), nuclear paraspeckle assembly transcript 1 (NEAT1), and taurine-upregulated gene 1 (TUG1) display age-dependent upregulation, whereas the expression of LINC00657 and small nucleolar RNA, C/D box 3A (SNORD3A) is downregulated with age. c Circular RNAs (circRNAs) tend to accumulate during brain aging, but the function of their accumulation in aging remains unclear.

Several miRNAs enhance adaptive neuroprotection in the elderly at the cost of microglial inflammation
Two independent studies that reported in-depth analysis of miRNAs indicated that expression changes in miRNAs are prominent during human brain aging 19,68 . Overall, the expression of miRNAs in human brains decreases with age after adolescence; in contrast, a subset of miRNAs display age-dependent upregulation 19,68 . In some cases, age-dependent expression changes in miR-NAs appear to be protective against neurodegenerative diseases and brain aging. The expression of miR-144 increases with age in the cerebellum and suppresses the expression of the ataxin 1 (ATXN1) gene 69 (Fig. 4a). Because trinucleotide repeat expansion mutations in ATXN1, which augment the glutamine repeats in the encoded protein, underlie the development of spinocerebellar ataxia 1 (SCA1) 70 , miR-144 may play a protective role against this neurodegenerative disease. In addition, an age-dependent decrease in miR-222 expression appears to result in derepression of its target genes 19 , including REV1/deoxycytidyl transferase, which regulates the DNA damage response 71 (Fig. 4a). This process may eventually reduce the excessive DNA damage in aged brains.
However, the adaptive neuroprotection afforded by agedependent expression changes in several miRNAs also seems to be associated with health complications, including excessive microglial inflammation 31 . For example, miR-34a, which acts as a tumor suppressor in neuroblastoma 72 , exhibits an age-dependent increase, which in turn downregulates neuronal genes, thus possibly participating in brain aging 19 (Fig. 4a). In addition, the expression of miR-29a/b is dramatically increased with age in the superior frontal gyrus and the postcentral parietal area 19,73 and exerts neuroprotective functions 74,75 , as well as induces excessive microglial inflammation 73 (Fig. 4a). Low expression of miR-29a/b is observed in the brains of patients with sporadic Alzheimer's disease and correlates with high BACE1/β-secretase levels 74 and a high amyloid beta load 76 , again suggesting the neuroprotective role of miR-29a/b. Conversely, miR-29a/b expression correlates negatively with the mRNA levels of insulin-like growth factor 1 (IGF1) and CX 3 CR1 73 , both of which inhibit excessive microglial activation 42,43,77 . Thus, the age-dependent increases in miR-29a/b levels seem to protect against neurodegeneration in the elderly at the cost of excessive microglial inflammation (Fig. 4a).
lncRNAs display high regional specificity in aging human brains Multiple studies focusing on analysis of lncRNAs have reported that aged human brains exhibit expression changes in various lncRNAs with regional specificity [78][79][80][81] ( Fig. 4b). Among the 383 lncRNAs that are differentially expressed between young and aged samples from four brain regions, only eight exhibit consistent changes in all brain regions 80 , indicating the high regional specificity of lncRNAs.
In the prefrontal cortex, the expression levels of the antisense RNA of the gene encoding proline-rich protein (PRR34-AS1), LINC01094, and LINC00844 correlate positively with age, whereas the expression levels of the antisense RNA of the gene encoding opacity-associated (Opa) interacting protein 1 (OIP5-AS1), MIR7-3 host gene (MIR7-3HG), LINC00643, LINC00507, and brain cytoplasmic 200 (BC200) correlate negatively with age 78,81 (Fig. 4b). These age-dependent changes in the expression of lncRNAs in the prefrontal cortex reflect cognitive declines with age. LINC00507, which has ribosomebinding activity and micropeptide-coding potential, is specific to cortical regions in primates 82 , suggesting that its downregulation with age is associated with cognitive decline. Another lncRNA, BC200, inhibits local mRNA translation in dendrites and synapses 83,84 . The expression level of BC200 in the prefrontal cortex decreases by 65% between the ages of 49 and 86 years 78 and correlates with age-dependent declines in synaptic functions 85 .
In the subependymal zone of the adult brain, which contains neural stem cells 86 , GOMAFU [also called myocardial infarction associated transcript (MIAT) and retinal noncoding RNA 2 (RNCR2)], metastasis-associated lung adenocarcinoma transcript 1 (MALAT1), nuclear paraspeckle assembly transcript 1 (NEAT1), and taurineupregulated gene 1 (TUG1) exhibit age-dependent upregulation, whereas the expression of LINC00657 and small nucleolar RNA, C/D box 3A (SNORD3A) is downregulated with age 79 (Fig. 4b). GOMAFU acts as a scaffold for the sequestration of splicing factors and modulates the splicing of transcripts that are required for synaptic plasticity 87 . Neuronal activation decreases the expression of GOMAFU and releases splicing factors into the nucleoplasm 87 . In addition to GOMAFU, MALAT1 and NEAT1 are located in nuclear speckles and paraspeckles, respectively, where they regulate alternative splicing in neurons by sequestering splicing factors 88 . However, the mechanisms by which MALAT1 and NEAT1 participate in brain aging remain poorly understood. TUG1 sequesters miR-9, and its upregulation leads to release of BCL2 like 11 (BCL2L11) mRNA from miR-9 to promote apoptosis in rats with ischemia 89 . Thus, the agedependent increase in the expression of TUG1 in the subependymal zone may indicate the acceleration of neuronal apoptosis in aged brains 79 .
The age-dependent expression changes in a subset of lncRNAs also display specificity for other brain regions. The postcentral gyrus and superior frontal gyrus exhibit a larger number of differentially expressed lncRNAs than does the entorhinal cortex or hippocampus 80 . The majority of the age-dependent changes in lncRNA expression manifest as upregulation in all regions except the entorhinal cortex 80 . Overall, lncRNAs appear to be more region-specific than mRNAs in aging human brains.
circRNA levels tend to increase during human brain aging circRNAs are mostly noncoding RNAs that are formed by backsplicing events; 90 however, the biological functions of circRNAs remain largely unexplored. Unlike linear RNAs, circRNAs are covalently linked at their termini and lack 3′ poly(A) tails; therefore, circRNAs display high resistance against exonucleases [91][92][93] . This property endows circRNAs with enhanced stability and raises the possibility that the levels of circRNAs remain high in nonproliferating cells. The levels of circRNAs do not correlate positively with those of their cognate mRNA isoforms 94,95 . circRNAs are highly expressed in brain tissues compared with nonbrain tissues in mice and humans 94 . Importantly, circRNAs display agedependent accumulation in the brains of mice and a nonhuman primate, the rhesus macaque 95,96 (Fig. 4c). In addition, certain circRNAs are specifically detected in patients with Alzheimer's disease 97 . Therefore, although circRNAs have not been directly characterized in the aging human brain, it seems likely that the levels of circRNAs increase during human brain aging. It will be important to determine by future research whether the accumulation of circRNAs plays a role in aging and agerelated diseases.

Conclusions and perspectives
Transcriptome analysis of human brain aging offers a comprehensive view of neurocognitive aging and neurodegenerative diseases. Here, we reviewed studies regarding the transcriptome changes associated with human brain aging. Decreased synaptic function and increased immune function are prevalent in most brain regions during aging, but other age-associated transcriptome changes display cellular or regional specificities. Gene expression in glial cells, including astrocytes and oligodendrocytes, is dramatically altered during brain aging and is accompanied by loss of the regional identities of these genes. Age-dependent gene expression changes are also affected by genetic backgrounds and splicing events. The role of RBPs, including splicing factors, is important for the splicing events in many transcripts during brain aging. Increased intersample heterogeneity is a property that implies the impairment of genome maintenance and repair in aged brains. The role of age-associated expression changes in noncoding RNAs, including miRNAs, lncRNAs, and circRNAs, remains less well understood than those in mRNAs. Thus, future studies of these noncoding RNAs will lead to a deeper understanding of brain aging at the molecular level.
The emergence of single-cell RNA-seq technology has enhanced the potential of transcriptome analysis and will help improve the understanding of human brain aging at the single-cell level. Specifically, investigation of singlecell transcriptomes provides unbiased insights into the subtypes and the diversities of brain cells 98,99 . Single-cell RNA-seq overcomes the problem of bulk RNA-seq, i.e., that the transcriptional contribution from minor cell types is masked by major cell types in tissues. Currently, examination of single nuclei is usually preferred 98 . However, this method cannot estimate the expression of cytosolic transcripts. Therefore, optimization of singlecell isolation is required for brain transcriptome analysis at the single-cell level. A recent study reported successful analysis of the single-cell transcriptome in the aging mouse brain by introducing sophisticated brain tissue dissociation steps 100 . Thus, future research employing single-cell RNA-seq technology with human brain samples will pave the way for providing new insights into human brain aging. In addition, by combining this singlecell technique with in situ hybridization 101 , transcriptome analysis will allow spatial separation for visualizing gene expression in human brain cells during aging.
One of the limitations of previous research on transcriptome analysis of the human brain is that the majority of participants in postmortem brain studies are of Caucasian ethnicity. Among 1007 individuals with specified ethnicity in previous studies (Table 1), 733 (72.8%) and 32 (3.2%) were Caucasian and European, respectively. African-American (198, 19.7%) and African (16, 1.6%), Hispanic (18, 1.8%) and Asian (8, 0.8%) populations comprised a minor fraction. Thus, our understanding of human brain aging may be effective for only a subset of the human population at best. Future investigation of transcriptomes obtained from a wide spectrum of ethnic groups will be crucial for a comprehensive understanding of human brain aging.
In conclusion, understanding the transcriptome in aged human brains will lead to the early detection of cognitive aging and the development of interventions that delay brain aging at the molecular level. Furthermore, given the close relationship between brain aging and neurodegenerative diseases, the establishment of transcriptome-based hallmarks of human brain aging will facilitate the diagnosis and treatment of these disorders.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.