Genetic Biomarkers on Age-Related Cognitive Decline

With ever-increasing elder populations, age-related cognitive decline, which is characterized as a gradual decline in cognitive capacity in the aging process, has turned out to be a mammoth public health concern. Since genetic information has become increasingly important to explore the biological mechanisms of cognitive decline, the search for genetic biomarkers of cognitive aging has received much attention. There is growing evidence that single-nucleotide polymorphisms (SNPs) within the ADAMTS9, BDNF, CASS4, COMT, CR1, DNMT3A, DTNBP1, REST, SRR, TOMM40, circadian clock, and Alzheimer’s diseases-associated genes may contribute to susceptibility to cognitive aging. In this review, we first illustrated evidence of the genetic contribution to disease susceptibility to age-related cognitive decline in recent studies ranging from approaches of candidate genes to genome-wide association studies. We then surveyed a variety of association studies regarding age-related cognitive decline with consideration of gene–gene and gene–environment interactions. Finally, we highlighted their limitations and future directions. In light of advances in precision medicine and multi-omics technologies, future research in genomic medicine promises to lead to innovative ideas that are relevant to disease prevention and novel drugs for cognitive aging.

diseases and neurodegenerative disorders has become a huge public health concern owing to high social and economic burdens (3). What is more, cognitive aging processes implicate multiple complex pathogeneses including genetic and environmental factors (4). In this light, the identification of genetic biomarkers has become an important area of research that aims to preclude the advancement of cognitive aging and to grasp the biology of cognitive aging in an increasingly aging society (5). It is noteworthy that untangling genetic biomarkers for cognitive aging has been at the center of major investigations in the field of precision medicine, and the relevant biomarkers for AD are generally utilized in cognitive aging research as well because of the increased risk for AD in the elderly individuals (6). More recent research in genome-wide association studies (GWAS) has implicated that single-nucleotide polymorphisms (SNPs) among 11 genes appear to affect the risk of AD, encompassing the PICALM, MS4A6E, MS4A4E, MS4A4A, EPHA1, CR1, CLU, CD33, CD2AP, BIN1, APOE, and ABCA7 gene (7)(8)(9)(10)(11). Consequently, a meta-analysis of GWAS studies (n = 74,046) identified 14 risk genes in AD, encompassing the ZCWPW1, SORL1, SLC24A4, RIN3, PTK2B, NME8, MEF2C, INPP5D, HLA-DRB4, HLA-DRB1, FERMT2, DSG2, CELF1, and CASS4 genes (12). The succeeding whole-exome sequencing analysis further tracked down the PLD3 gene to likely be a risk gene for AD (13). Moreover, it has been found that cognitive decline was linked with the CR1 rs6656401 SNP by using the established AD-associated genes (14). Additionally, recent epistasis studies suggested that the CLU-MS4A4E (15,16) and CD33-MS4A4E (16) gene-gene interactions might have a considerable influence on the susceptibility of AD. By utilizing the known AD-associated genes, it has also been demonstrated that cognitive decline was related with the ABCA7 rs3764650 and CD33 rs3865444 SNPs in the elderly Caucasian women (17).
In this review, we first surveyed some genetic biomarkers that were linked with age-related cognitive decline in several recent association studies (Table 1). Furthermore, we assessed some potential gene-gene and gene-environment interactions on age-related cognitive decline. This review does not intend to comprehensively survey all literature. We mainly focused on the most recent developments for biomarker research in cognitive aging. Finally, the limitations and future perspectives associated with cognitive aging in terms of genetic biomarkers were summarized. Future replication studies in larger samples with longitudinal follow-up are required to confirm the findings of the biomarkers for cognitive aging discovered in the association studies.
Consistent with the findings by Lin et al. (18), two preceding GWAS studies (8,10) have reported that the EPHA1 rs11767557 SNP may affect the vulnerability to AD. In addition, the FERMT2 rs17125944, MEF2C rs190982, and SLC24A4 rs10498633 SNPs were prone to AD in a meta-analysis study (12). Moreover, a whole-exome sequencing study pinpointed a rare rs145999145 (Val232Met) variant in the PLD3 gene was liable to AD (13). The PLD3 gene may play a central role in dealing with amyloid-beta precursor protein (13).
The CASS4 protein is proposed to better characterize the functions of cell growing, spreading, adhesion, and other activities (19). Moreover, it was speculated that CASS4 plays a vital part in the hallmarks of AD such as the amyloid precursor protein (APP) and Tau protein (20). More recent studies also indicated an association between AD and the CASS4 rs7274581, rs6024870, and rs16979934 SNPs (12,(21)(22)(23). In contrast, a replication study indicated that the CASS4 rs7274581 SNP did not affect the risk of AD in a Spanish population (24).
Gene-gene interaction study inferred that SLC24A4, MEF2C, FERMT2, EPHA1, and CASS4 synergistically raised the propensity of cognitive aging by using the GMDR approach (18). It was further speculated that these five genes are comprised in the relevant pathology and pathways (18). EPHA1 is implicated in regulating neurodevelopment (20). FERMT2 contributes to Tau neurotoxicity and cell adhesion (25). MEF2C may influence hippocampal synaptic connectivity and thereby mediate hippocampal-dependent memory and learning (26). SLC24A4, in the vicinity of the RIN3 gene, is linked with neurodevelopment (27) as well as Tau pathology and APP (28).

APOE
The APOE gene, located on chromosome 19q13.32, encodes a major protein which is crucial for the regular catabolism of triglyceride-rich lipoprotein constituents (29). Conducting a GWAS study with the longitudinal cognitive testing data such as memory and perceptual speed, De Jager et al. discovered that APOE was genome-wide significantly correlated with age-related cognitive decline (mean age: 72.0~80.8 years) (30). A meta-analysis of 77 studies (n = 40,942) also suggested that carriers of the APOE ɛ4 allele, which were linked with late-onset AD, performed worse on several domains of cognitive functions including overall global cognitive ability, episodic memory, and executive functioning (31). Further, the APOE rs405509 and APOE rs440446 SNPs were more likely to develop non-pathological cognitive aging, independent of APOE major isoforms, in a Finnish population (32). In contrast, a replication study indicated that the APOE gene did not affect the risk of age-related cognitive decline in older Taiwanese adults (18).

BDNF
Another potential candidate gene such as BDNF was reported to be implicated in age-related cognitive decline among the elderly (33). The BDNF gene, encoding proteins of the nerve growth factor family, was demonstrated to alter cognitive deficits in healthy subjects (mean age: 72.7 years) for normal aging by using the mean Z-scores based on three cognitive domains including motor function, memory, and perceptual speed (34). Ward et al. also disclosed an interaction between APOE and BDNF that predicted a cognitive effect in healthy older adults (35).

COMT
The COMT gene, located on chromosome 22q11.21, is essential in the metabolic degradation of prefrontal dopamine (36). A functional COMT Val158Met polymorphism has been shown to contribute to individual differences in cognitive aging in Taiwanese (mean age: 78.7 years) and German (mean age: 64.9~65.3 years) populations using MMSE scores (36,37). Based on a 4-year longitudinal study, Dixon et al. also revealed that the APOE and COMT genes are complementary biomarkers in normal cognitive aging and early MCI for older adults (38).

CR1
The CR1 gene encodes a protein that plays a role in cellular binding and immune complexes (14). A replication study reported that the rs6656401 SNP in the CR1 gene, which is one of the susceptibility loci for AD, was significantly associated with cognitive deficits (mean age: 75.5~84.4 years) for normal aging (in terms of global cognitive decline) as well as for pathological aging (in terms of global AD pathology) using the cognitive testing data such as memory and global cognition (14). A subsequent replication study also indicated that the rs4844609 SNP in the CR1 gene modulates episodic memory decline and an interaction between APOE and CR1 influences cognitive decline for normal aging and for pathological aging (39). In contrast, a replication study indicated that the CR1 gene did not affect the risk of age-related cognitive decline in older Taiwanese adults (18).

Circadian Clock Genes
Circadian rhythms are instinctively recurring cycles that determine the timing of biological events such as energy metabolism, hormone release, and sleep-wake cycles (40). Additionally, dysregulation of circadian rhythms is characteristic of the natural process of ongoing aging and cognitive decline (40). The circadian rhythms are maintained and triggered by a composition of core circadian clock genes, which can be classified as vital genes on provoking circadian rhythms in specific cells (41).
Circadian clock genes, encompassing RORB, RORA, PER3, PER2, PER1, NR1D1, NPAS2, CRY2, CRY1, CLOCK, and ARNT, may be also involved in cognitive impairment (40,41). To find out potential genes implicated in the regulation of age-related cognitive decline, a recent study analyzed whether the 11 aforementioned core circadian clock genes as well as complex gene-gene and gene-environment interactions contributed to cognitive aging in more than 634 elderly individuals (mean age: 64.2 years) in Taiwan (42). As a result, four SNPs, including the RORA-rs13329238, NPAS2-rs17655330, CLOCK-rs3749473, and RORB-rs10781247 SNPs individually and interactively alters the hazard of cognitive deficits in terms of MMSE scores for normal aging (42). Finally, environmental factors such as smoking status, alcohol consumption, social support, and physical activity also interacted with these SNPs in regulating the liability of age-related cognitive decline (42).
The RORA and RORB proteins create a family of nuclear hormone receptors (47). These two genes were reported to have a key part in a wide variety of regulations; for instance, cellular metabolism, circadian rhythm, embryonic development, immunity, and inflammatory responses (48,49).
It should be noted that NPAS2 encodes transcription factors which are a component belonging to the basic helix-loop-helix family, and the ARNTL/NPAS2 and ARNTL/CLOCK heterodimeric proteins can bind to chromatin, bringing about the upregulation in gene expression for PER1, PER2, PER3, CRY1, and CRY2 (50). In this manner, products of these CRY1, CRY2, PER1, PER2, and PER3 genes build a complex that inhibits the action of the ARNTL-containing complexes and in turn forms a negative feedback loop (50).

insulin Resistance-Associated Genes
To identify possible genes implicated in the process of agerelated cognitive decline, a recent association study of the Taiwan Biobank has hypothesized that SNPs in insulin resistanceassociated genes, including the PPARG, GCKR, and ADAMTS9 genes, might be associated with cognitive aging individually or collectively in a cohort of Taiwanese individuals (n = 547; age > 60 years; mean age: 64.1 years) (51). The results indicated that 4 SNPs (including rs9831846, rs4317088, rs9985304, and rs73832338) within ADAMTS9 were linked with MMSE scores (that is, cognitive aging) after performing Bonferroni correction (51). In addition, Lin et al. identified a SNP-SNP interaction between ADAMTS9-rs76346246 and ADAMTS9-rs9985304 that may influence cognitive aging by using the GMDR approach (51).
ADAMTS9 is demonstrated in the regulation of a wide variety of processes, such as the control of organ shape during growth, the inhibition of angiogenesis, and the cleavage of proteoglycans (51). Since one of the risk factors for cognitive aging and AD is insulin resistance, insulin resistance-related genes might be linked with cognitive aging (52,53). Insulin abnormalities raise the uncertainty for neurodegenerative disorders including MCI, AD, and cognitive aging (54,55). Insulin is also a fundamental factor for normal brain functioning. It should be noted that ADAMTS9 is associated with insulin resistance, insulin sensitivity, and type 2 diabetes (56,57). There is growing evidence that the ADAMTS-9 protein might participate in the processes of brain disease states including spinal cord injury, ischemic stroke, and transient middle cerebral artery occlusion in animal studies (58)(59)(60). In accordance, a GWAS study has identified the ADAMTS9 rs6795735 SNP as a candidate biomarker for age-related macular degeneration, which commonly occurs in elder adults (61).

Re1-Silencing Transcription Factor (REST) Gene
A growing body of evidence suggests that REST may be involved with AD and cognitive aging (62). A recent replication study has assessed whether REST SNPs are associated with cognitive aging as well as via SNP-SNP interactions for normal aging in elder Taiwanese subjects (n = 634; mean age: 64.2 years) (62). Their analysis results demonstrated that REST-rs1277306 was linked with cognitive aging, which was measured by MMSE scores (62). This prediction is further supported by evidence that the association remained significant for individuals without APOE ε4 allele after Bonferroni correction (62). On the other hand, the REST rs1277306 SNP was not a predicting factor for cognitive aging among individuals with at least one APOE ε4 allele (62). In addition, Lin et al. tracked down an SNP-SNP interaction between the REST rs1713985 and REST rs1277306 SNPs on cognitive aging by using the GMDR. The REST protein is indicated in the modulation of synaptic plasticity, ion channels, vesicular transport, axonal growth, and neuronal differentiation (63,64). In addition, REST is associated with amyloid β-protein toxicity, protection from oxidative stress, AD pathology, MCI, brain aging, and slow hippocampal loss (65)(66)(67). Additionally, REST might act as both a primary protector against neurodegeneration and an essential repressor for normal neurogenesis (65)(66)(67). Although APOE is well-established regarding its major role in cognitive decline in elder adults (68,69), the biologically synergistic effects between the APOE and REST genes on cognitive aging are still unknown. It was speculated that APOE and REST might involve in a comparable pathway relevant to cognitive aging (70).

DNA Methylation
Recent studies indicate that DNA methylation, one of main epigenetic mechanisms, plays a crucial role in cognitive aging (74,75). DNA methylation involves the inclusion of a methyl group to the DNA molecule, especially when a cytosine is followed by a guanine (76). DNA methylation is regularly associated with reduced transcriptional activity and is triggered by a family of DNA methyl-transferase proteins (77). Using repeated measures of composite scores for annual cognitive testing, Chouliaras et al. investigated associations among common SNPs in genes modulating DNA methylation and cognitive aging. They found that the rs11887120 SNP in the DNMT3A gene was associated with annual decline in cognitive deficits for normal aging in a Dutch population (78). In contrast, this finding was not replicated in German subjects (79). DNMT3A, encoding a DNA methyltransferase, is located in the cytoplasm and nucleus in de novo methylation (80).

TOMM40
With longitudinal cognitive ability data, a GWAS study reported that the rs2075650 SNP in the TOMM40 gene, which is adjacent to the APOE gene, was significantly associated with age-related cognitive decline (mean age: 64.6~79.1 years) (81). After fine SNP mapping of the TOMM40/APOE region, both APOE rs429358 and TOMM40 rs11556505 were correlated with cognitive aging (81). Furthermore, SNPs within the TOMM40/APOE zone possessed a non-protein-coding regulatory and functional effect in a functional genomic analysis, indicating that the TOMM40/APOE zone may be linked with nonpathological cognitive aging (81). The TOMM40 gene, located on chromosome 19q13.32, encodes the mitochondrial outer membrane complex relevant to the channel-forming subunit of the translocase, which is indispensable for construction of protein precursors to mitochondria (82).

Other Potential Genes
The DTNBP1 gene encodes a protein that plays a role in the biogenesis of organelle linked with lysosomes, platelet dense granules, and melanosomes (83). It has been suggested that the DTNBP1 gene modulates general cognitive abilities both in schizophrenia patients and in healthy subjects in Japanese (mean age: 34.1~39.2) and German (mean age: 24.8) populations (84)(85)(86)(87). In addition, Burdick et al. found that the CTCTAC risk haplotype of 6 SNPs including rs909706, rs1018381, rs2619522, rs760761, rs2619528, and rs1011313 in the DTNBP1 gene was associated with general cognitive ability and cognitive decline in schizophrenia patients (84).
In a recent systems genetics study applying a genetically diverse population of mice, Neuner et al. pinpointed Hp1bp3 gene to be a novel modulator of cognitive aging (88). Their findings also confirmed that as compared to cognitively healthy individuals, levels of HP1BP3 protein were significantly decreased in the hippocampi of elderly subjects with cognitive impairment, suggesting that reduced expression of Hp1bp3 may contribute to cognitive aging in both mice and humans (88). The HP1BP3 gene is located on chromosome 1p36.12 and encodes a histone H1 related protein with non-redundant and specific roles vital for viability and gain in humans (89).
It is worth mentioning a potential gene called SRR although, to our knowledge, there are only animal studies but no population studies for this gene in cognitive aging. SRR encodes the serine racemase enzyme which converts L-serine to d-serine. d-serine is an endogenous co-agonist for N-methyl-D-aspartate receptors (90). The SRR rs408067 SNP, located in the promoter region, may affect the transcription activity of the SRR gene (91). Reduced SRR expression impaired hippocampal age-related cognitive function in an animal study, suggesting that the SRR-dependent pathway might be one possible target of the hippocampus-related cognitive decline in aging (92). Another expression profiling study reported that various genes that are associated with cognitive ability were influenced by the SRR mutation (93).

liMiTATiONS iN CURReNT STUDieS
Notwithstanding, there were several limitations with respect to the aforementioned studies. First, there is certainly room for development of much further research and comprehensive evaluation to reassess whether the current results remain in other ethnic populations for the investigated genetic variants with cognitive aging in terms of the association and interactions (62).
Second, given the relatively young mean age of the sample in several studies mentioned previously, the current results are unable to be extrapolated to much older populations that have higher risk for developing age-associated diseases and neurodegenerative disorders, such as MCI, AD, and other dementias (18).
Because of logistical and ethical matters, it is challenging to assess homogeneous genetic backgrounds and recruit a large enough cohort of participants at the same time (94). Furthermore, some findings were not replicated, and the discordant results found among these studies may be due to issues in the sample size, ethnicities, study design, and phenotype definitions. Moreover, confounding factors may not be fully handled, and thereby considerable bias may not be excluded.
In order to reinforce the statistical findings, it is appealing to seek more supplementary biologically relevant evidences owing to the fact that the investigated SNPs might be greatly enhanced in association studies (95).
Besides, the aforementioned studies utilized various methods to assess cognitive function. A major challenge is to ensure a proper approach for evaluating cognitive function. The wellestablished MMSE approach is chosen to evaluate cognitive function in several aforementioned studies because it is the most widely used screening test of cognition (96). Nevertheless, the ceiling effect of MMSE in healthy young subjects as well as its floor effect in the oldest subjects diminishes the variability (96). Similar to MMSE, another strategy is the General Practitioner Assessment of Cognition (GPCOG) with psychometric properties. However, using the GPCOG is required to further examine for its possible language or cultural tendency (97,98). Another more recent language independent method is the CANTAB, a visual and cognitive assessment tool used on computers (99). Nonetheless, because the correlations between CANTAB and commonly used cognitive tools such as GPCOG and MMSE are only modest, the application of CANTAB should be justified in future studies (100). Moreover, an alternative is the ADAS-Cog which achieves higher sensitivity with a change of four-point (101). Although the ADAS-Cog is a well-validated scale in cognitive performance, the drawback is that about 40 min is needed to complete the task and this fact causes it unacceptable in most large-scale studies (96).
Cross-sectional design for cognitive aging studies has been adopted by many researchers because it is less feasible to examine aging trajectories for individual participants with longitudinal studies due to the high cost and long follow-up time (102). Nevertheless, it is always important to recognize the limitations of transversal studies on aging. Most importantly, we are unable to make longitudinal or causal inferences about changes in cognitive function by using cross-sectional data.

FUTURe OUTlOOK
Over the past decade, advances in genome science have spawned numerous lines of research into precision medicine and multiomics (103). In spite of spectacular progress in precision medicine and multi-omics technology, which can assemble a mammoth amount of multi-omics data, there are no established approaches to take advantage of that data in a predictive fashion (103). Therefore, we face a challenge of developing a fundamental, personalizable, mathematical model, which is calibrated on a broad range of multi-omics and clinical data (103). To conquer this challenge, a key component of future projects is to be able to advance the aforementioned predictive capability by facilitating machine learning and predictive approaches (103).
Building up a set of genetic biomarkers which are immensely dependable as a benchmark of disease status or drug response for cognitive aging will be considerably indispensable in the future (104). At this juncture, no genetic biomarkers found in the aforementioned studies would be unquestionably qualified to be incorporated in the panel owing to the aforementioned limitations (104).
Moreover, machine learning and predictive techniques such as Bayesian networks may present a conceivable approach to forecast novel drug efficacy and establish statistical models for predicting disease status (103). In future research, we will be able to help physicians in the prescription by creating predictive models which forecast the likelihood of diseases or treatment response (103,104). In addition, predictive and machine learning approaches such as Bayesian networks might be important in weighing correlations in RNA-RNA molecule, correlations between miRNA and mRNA, as well as interactions between gene and environment (105). Moreover, the statistical modeling such as meta-analysis, pathway analysis, and gene-gene expression correlations is intrinsic to eliminate the false positive biomarkers observed during the association analyses of current precision medicine studies (105).
Essentially, evidence shows that multi-omics data and biomarkers such as genetic, epigenetic, metabolomic, transcriptomic, and proteomic profiles are important in assorted pathophysiology for a certain disease and novel drug treatment (103,105). Subsequently, the systematic and integrative analyses of different profiles with apparently cooperative functions might have a big impact on the disclosing for the mysterious pathogenic processes of a certain disease and novel drug treatment (103,(105)(106)(107). Finally, in order to unquestionably carry out disease pathogenesis as well as novel drug therapy, future studies will have to accomplish an integrative and systematic way of using clinical information, biomarkers, and multi-omics data (103,105).

SUMMARY
In this review, we have focused primarily on recent findings and relevant studies for age-related cognitive decline. The current review also highlighted the merit of association studies with relatively large sample size to incorporate a wide variety of populations for cognitive aging. In order to advance personalized treatment and prevention strategies worldwide, a main challenge is how best to integrate these findings with other pieces until the picture of cognitive aging is adequately apparent. Similarly, these findings have indicated that machine learning and predictive tools might be beneficial for clinical decision making by integrating multi-omics data and biomarkers.
In light of recent developments, novel machine learning and predictive algorithms will be the new frontier in the decades to come for establishing prognostic and diagnostic assessments by using huge data technologies for precision medicine (103,105). Future research using machine learning and predictive approaches is warranted in the matter of managing the interactions of biomarkers and foretelling the relationship between biomarkers and drug response in precision medicine studies (103,105). In our opinion, yet a number of challenges remain and a host of deeply key and crucial research issues must be ironed out. As we enter a period of the new envisioned science of precision medicine, personalized therapy for individuals would undoubtedly become a reality.

AUTHOR CONTRiBUTiONS
EL, H-YL, and C-HL involved in conception and design; EL and C-HL involved in literature review and interpretation, and manuscript writing; EL, H-YL, and C-HL involved in financial support and final approval of manuscript.

FUNDiNG
The authors extend their sincere thanks to Vita Genomics, Inc. for funding this research. This work was supported by the Ministry of Economic Affairs in Taiwan (SBIR Grant S099000280249