microRNAs as Early Biomarkers of Alzheimer’s Disease: A Synaptic Perspective

Pathogenic processes underlying Alzheimer’s disease (AD) affect synaptic function from initial asymptomatic stages, long time before the onset of cognitive decline and neurodegeneration. Therefore, reliable biomarkers enabling early AD diagnosis and prognosis are needed to maximize the time window for therapeutic interventions. MicroRNAs (miRNAs) have recently emerged as promising cost-effective and non-invasive biomarkers for AD, since they can be readily detected in different biofluids, including cerebrospinal fluid (CSF) and blood. Moreover, a growing body of evidence indicates that miRNAs regulate synaptic homeostasis and plasticity processes, suggesting that they may be involved in early synaptic dysfunction during AD. Here, we review the current literature supporting a role of miRNAs during early synaptic deficits in AD, including recent studies evaluating their potential as AD biomarkers. Besides targeting genes related to Aβ and tau metabolism, several miRNAs also regulate synaptic-related proteins and transcription factors implicated in early synaptic deficits during AD. Furthermore, individual miRNAs and molecular signatures have been found to distinguish between prodromal AD and healthy controls. Overall, these studies highlight the relevance of considering synaptic-related miRNAs as potential biomarkers of early AD stages. However, further validation studies in large cohorts, including longitudinal studies, as well as implementation of standardized protocols, are needed to establish miRNA-based biomarkers as reliable diagnostic and prognostic tools.


Introduction
Alzheimer's disease (AD) is a progressive and irreversible brain disorder and the most frequent form of dementia among the elderly, reaching nearly 70% of cases [1,2]. Although the development of the disease is variable between patients, three phases can be distinguished as part of a continuous process of degeneration ( Figure 1): (1) The preclinical phase, usually lasting more than 10 years, characterized by early changes in biomarkers in the absence of clinical symptoms; (2) the prodromal phase comprising the earliest symptomatic stage of the disease when cognitive decline starts to be evident while biomarker levels do not reach the cutoff criteria for diagnosing dementia [3], and (3) a dementia stage in which the pathology and symptomatology is fully developed [4]. The duration of each phase is not constant but rather depends on the age of onset, gender, and genetic risk factors [5].
The development of these stages is related to the spread of pathological changes across different brain regions. AD pathological hallmarks, comprising accumulation of amyloid-β peptide (Aβ) aggregates forming extracellular plaques, and hyperphosphorylated species function evaluated by tests such as the Mini-Mental State Exam (MMSE) or the delayed word list recall tests that evaluate hippocampus-dependent tasks exhibit a positive correlation with the number of synapses in brain related areas as the hippocampus and frontal cortex of early AD patients [14,29].
Interestingly, synaptic density constitutes a better correlation with clinical symptoms than classical AD histopathological markers such as amyloid plaques and NFT [14,26,29,30]. In this regard, solid evidence indicates that oligomeric forms of Aβ (oAβ), instead of amyloid plaques, contribute to synaptic alterations and correlate with synaptic loss [12,[31][32][33]. In the same line, it has been described that Aβ aggregation is enriched at synapses even before the formation of amyloid plaques or tau NFT [34].
Accordingly, oAβ are thought to initiate the pathological events described as the amyloid cascade hypothesis [35], according to which changes in Aβ metabolism result in its oligomerization that initially triggers disruption and loss of synaptic connections follow by inflammatory response involving microglial and astrocytic activation, alteration of calcium homeostasis and oxidative stress resulting in neurodegeneration. Indeed, oAβ were shown to alter pre-synaptic functions such as axonal transport, synaptic vesicles trafficking and recycling, and neurotransmitter release [36]. Moreover, evidence indicate that oAβ also affect post-synaptic function and are especially toxic to glutamatergic synapses. Dysregulation of glutamatergic transmission is the main described mechanism by which oAβ could alter spines shape and number [33,[37][38][39][40] leading to synaptic failure [37]. For instance, the cascade triggered by oAβ affect the regulation of downstream kinases and phosphatases that increase NMDAR and AMPAR internalization [28,37,40], creating an imbalance between LTP and LTD. While these structural and functional alterations have been shown to contribute to cognitive dysfunction present in AD patients [41], several studies have also linked AMPAR decrease in the cell surface with synaptic alterations in AD experimental models, including primary neuronal cultures and transgenic mice models [28,42] in which, interestingly, a correlation also exists with learning and memory deficits [27,28]. Moreover, loss of synapses in the CA1 of transgenic mice models of AD are consistent with the synapse loss described in AD [15].
Furthermore, oAβ would facilitate tau phosphorylation, which in turn decreases its affinity for microtubules and facilitates its aggregation and NFTs formation contributing to signaling deficiency and neurodegeneration [43]. On the other hand, besides the wellknown role of tau in microtubule assembly, studies in transgenic mice highlighted a dendritic role for tau during synaptic pathology [44][45][46]. During pathological processes, tau is able to bind to scaffolding proteins and glutamate receptors at the PSD affecting synaptic function by altering LTP-LTD balance [32]. In addition, tau has been shown to be necessary for Src-family tyrosine kinase Fyn recruitment to post-synaptic NMDAR complexes in a mechanism that mediates Aβ excitotoxicity [46]. Thus, both Aβ and tau constitute key pathogenic players in early neurodegenerative processes linked to AD [32], and they are thought to interact locally at synapses leading to synaptic failure and cognitive impairment [30,47,48].

miRNAs as Mediators of Synaptic Dysfunction in AD
Increasing evidence indicates that the alteration of protein functionality in the synapse could be involved in early synaptic alterations in AD [26,27]. Given the regulatory role of miRNAs, those miRNAs targeting synaptic-related proteins may be an important mechanism underlying the synaptic dysfunction present in early stages of the disease [49]. Thus, alterations in the levels of specific miRNAs could be important in the development of synaptic pathology that leads to neurodegeneration in AD by modifying the synaptic structure and function that underlie synaptic plasticity. In addition, miRNAs are capable of regulating the mechanisms of toxicity mediated by central factors of the pathology such as tau and Aβ. Moreover; miRNAs can be enveloped in membranous vesicles that can be released to the peripheral circulation in the form of extracellular vesicles such as exosomes, favoring miRNAs conservation [50]. In this context, it is reasonable to think that the possibility of detecting alteration in the levels of specific miRNAs during pathology constitutes a valuable tool for AD detection. Indeed, these facts have opened a very prolific field with an increasing number of candidates to be used as biomarkers for AD, among which, synaptic-related miRNAs seem the most promising candidates as biomarkers for the detection of early stages of AD since synaptic dysfunction precedes neurodegeneration and the clinical symptoms.

Synaptic Role of miRNAs
miRNAs are the most studied small noncoding RNAs (ncRNAs), a class of functional RNA molecules of approximately 22 nucleotides (nt) of length, lacking protein-coding properties. Over 2600 human mature miRNAs have been annotated so far (miRBase.org) and conserved miRNA-binding sites have been detected in more than 60 percent of proteincoding genes, suggesting a wide presence of miRNA-mediated gene expression regulation [51]. miRNAs post-transcriptionally regulate gene expression by associating with Argonaute (Ago) proteins to form the RNA-induced silencing complex (RISC). Binding by partial sequence complementarity usually to the 3-untranslated region (3 -UTR) of target messenger RNAs (mRNAs), a single miRNA could repress mRNA translation of hundreds of different targets [51,52] constituting a fine regulatory mechanism of protein expression. Although a repression of translation is the main mechanism of miRNAs regulation, other mechanisms including an increase in translation, have also been reported [53].
The enrichment of specific miRNAs in axons and dendrites indicates a potential role for miRNAs locally regulating protein levels, thus synaptic structure and function. In this regard, an elegant study by Erin M. Schuman and colleagues has demonstrated that synaptic activity can increase the processing of pre-miRNAs locally in dendrites. In particular, local maturation of miR-181a and subsequent decrease in its target CaMKIIα has been described [61]. Local regulation of Ca 2+ signaling cascades related to learning and memory processes following neuronal activity through this key kinase, would constitute a fast local mechanism of synaptic function regulation by miRNAs [61].
Of particular interest, several transcription factors and co-activators such as the cyclic adenosine monophosphate (cAMP) response element-binding protein (CREB), CREB-regulated transcription coactivator-1 (CRTC1), and nuclear factor-kappa-light-chain-enhancer of activated B cells (NF-κB), have been related to AD. Interestingly, some of these transcriptional regulators are synaptonuclear factors implicated in regulating transcriptional programs related to synaptic function, often in an activity-dependent manner and through integration of signals mediated by CREB [62]. This transcription factor plays an essential role in activating transcriptional programs underlying synaptic plasticity [19]. Thereby, miRNAs binding to transcription factors would be able to regulate synaptic function. In this regard, Gerhard M. Schratt and colleagues have reported that CREB and myocyte enhancing factor 2 (MEF2) (a transcription factor that negatively regulates excitatory synapses' number) regulate the expression of miR-212/132 family where miR-132 and miR-134 are included. Conversely, these miRNAs and other members of the family regulate CREB, SIRT, and the methyl CpG-binding protein 2 (MeCP2) expressions that can in turn regulate BDNF levels, triggering the induction of miR-212/132 expression. Indeed, miR-132 can bind directly to BDFN and regulate its expression. This complex regulatory loop constitutes a mechanism by which miRNAs can control their own levels in response to neural activity changes [63]. The specific case of miR-134, a brain specific miRNA and activity regulated member of this family, has been largely studied by Schratt et al. [64,65]. They have observed that miR-134 can be transported to dendrites as a mature miRNA, but it also exists as pre-miRNA in the synapto-dendritic compartment. Locally, miR-134 can mature when needed and can negatively regulate dendritic spines size targeting LIM domain kinase 1 (Limk1), involved in spine maturation. Interestingly, following synaptic activity, the repression that miR-134 exert over Limk1 translation is released, and dendritogenesis is promoted [64,65]. On the other hand, miR-134 can also facilitate homeostatic synaptic depression in response to chronic activity, targeting the local translational repressor Pumilio-2 [66]. Recently, miR-134 localization in dendrites has been supported using atomic force microscopy showing a negative correlation between the amount of miRNA and the maturity and function of synapses [49]. Other members of the family, miR-132 and miR-138, were also linked to synaptic formation and function [63].
Interestingly, miR-29a/b are known to bind to Arpc3 subunit of the ARP2/3 complex, involved in regulation of actin filament branching and dendritic spine morphogenesis [67], whereas miR-191 is known to target tropomodulin-2, a neuron-specific regulator of actin dynamics that decrease in response to increased miR-191 expression following NMDAR activity during LTD [68]. Similarly, miR-9 has been reported to target transcriptional repressor REST, thereby, acting as a mediator in mRNA translation activation, necessary for functions such as dendritic growth [69].
Since glutamatergic transmission plays a key role in synaptic function and plasticity processes, the study of miRNAs targeting excitatory synapses and specially glutamate receptors has caught special attention. For instance, the brain-enriched miR-9-3p above mentioned, targets SAP97 (a member of PSD-95-like membrane associated guanylate kinases-PSD-MAGUKs-) that binds to mRNA coding for the GluA1 AMPAR subunit, thus, modulating AMPAR trafficking [70]. Moreover, miR-92, miR-137, and miR-501 are able to selectively regulate GluA1 trafficking, so that their overexpression in vitro reduces AMPAR insertion in cell surface during homeostatic scaling [71][72][73]. In addition, other studies have indicated the direct regulation of GluA2 subunit by miR-124 [74], miR-218 [75] and miR-186 [76] in an activity-dependent manner. Of interest, Olde Loohuis and colleagues have reported that in addition to the regulation that miR-137 exerts on AMPAR-mediated synaptic transmission and the turnover between silent and active synapse, an activity dependent mechanism is in turn regulating miR-137 levels. Upregulation of miR-137 levels was observed following metabotropic glutamate receptor 5 (mGluR5) activation, leading to a negative feedback mechanism controlling mGluR5-dependent synaptic plasticity [73]. miRNAs regulating NMDAR expression have also been reported, for instance, miR-139-5p [77] and miR-125b [78] have been involved in synaptic plasticity regulation by binding to mRNA coding for NR2A subunit while miR-539 and miR-34a regulate NR2B subunit of NMDAR [79,80]. Moreover, indirect regulation of NMDAR through miR-128 binding to STIM2 (stromal interaction molecule 2) has also been suggested as a miRNAmediated mechanism of synaptic plasticity modulation [81].
Regarding receptors regulation, the modulation of proteins related to receptor trafficking and function is also of great interest. For instance, miR-181c is predicted to target neuronal pentraxin 1 (NPTX1) and neuronal pentraxin receptor (NPTXR) while miR-210 has been shown to regulate the levels of both proteins [82] implicated in the modulation of glutamatergic transmission through modulation of AMPAR recruitment and clustering [83]. Interestingly, whereas miR-210 at the post-synaptic terminals is involved in the regulation of receptor trafficking and synaptic plasticity, at the pre-synaptic terminal, it has been shown to target SNAP25, participating in neurotransmitter release [84]. Regarding pre-synaptic function, miR-137 over-expression has been reported to alter synaptic vesicle docking accompanied by a decrease in active zone size and in synaptic vesicles number in hippocampal cultures, human-induced neurons, and in vivo experiments performed in mice [85]. Moreover, miR-485 has been associated with altered functional synapses number, spine maturation, PSD-95 clustering, and surface GluA2 expression by targeting the synaptic vesicle glycoprotein 2A (SV2A) involved in neurotransmitters release [86].
Given the central role of glutamatergic transmission in synaptic plasticity processes, the study of miRNAs targeting excitatory synapses, and especially glutamate receptors, has particularly grown during the last years. However, Katharine R. Smith and colleagues' recent work has put the focus on GABAA receptor regulation by miR-376. In this study, a mechanism by which miR-376 locally inhibit GABRA1 and GABRG2 translation in dendrites has been reported [87]. In this context, dendrite local de novo synthesis of synaptic GABAAR has been described, constituting a mechanism of long-term surface GABAAR clustering conservation during iLTP expression. Additionally, they described another level of regulation of this receptor, by which miR-376 is repressed through a calcineurin-NFAT-HDAC signaling pathway following NMDA-induced inhibitory long-term potentiation (iLTP), relieving miRNA-inhibition on GABAAR subunits translation.
Data reviewed in this section (summarized in Table 1 and Figure 2) strongly support an essential role of miRNAs in synaptic plasticity modulation and consequently suggested their implication also in synaptic dysfunction related to neurodegenerative diseases. Certainly, understanding how specific miRNAs regulate synaptic function at different levels is essential to complete the full landscape of mechanisms underlying AD pathology, especially regarding synaptic failure present in the early stages of the disease. been shown to target SNAP25, participating in neurotransmitter release [84]. Regarding pre-synaptic function, miR-137 over-expression has been reported to alter synaptic vesicle docking accompanied by a decrease in active zone size and in synaptic vesicles number in hippocampal cultures, human-induced neurons, and in vivo experiments performed in mice [85]. Moreover, miR-485 has been associated with altered functional synapses number, spine maturation, PSD-95 clustering, and surface GluA2 expression by targeting the synaptic vesicle glycoprotein 2A (SV2A) involved in neurotransmitters release [86].
Given the central role of glutamatergic transmission in synaptic plasticity processes, the study of miRNAs targeting excitatory synapses, and especially glutamate receptors, has particularly grown during the last years. However, Katharine R. Smith and colleagues' recent work has put the focus on GABAA receptor regulation by miR-376. In this study, a mechanism by which miR-376 locally inhibit GABRA1 and GABRG2 translation in dendrites has been reported [87]. In this context, dendrite local de novo synthesis of synaptic GABAAR has been described, constituting a mechanism of long-term surface GABAAR clustering conservation during iLTP expression. Additionally, they described another level of regulation of this receptor, by which miR-376 is repressed through a calcineurin-NFAT-HDAC signaling pathway following NMDA-induced inhibitory longterm potentiation (iLTP), relieving miRNA-inhibition on GABAAR subunits translation.
Data reviewed in this section (summarized in Table 1 and Figure 2) strongly support an essential role of miRNAs in synaptic plasticity modulation and consequently suggested their implication also in synaptic dysfunction related to neurodegenerative diseases. Certainly, understanding how specific miRNAs regulate synaptic function at different levels is essential to complete the full landscape of mechanisms underlying AD pathology, especially regarding synaptic failure present in the early stages of the disease.  Synaptic transmission miR-34a [80,109] Key synaptic-related targets and their functions are indicated. miRNAs targeting them that will be further discussed as potential MCI-AD biomarkers ( Table 2) are included.

miRNAs and Synaptic Dysfunction in AD
A large number of miRNAs have been shown to be dysregulated during AD progression [57,111] and in AD experimental models [89] expanding the role of miRNAs from physiology to pathology. Considering the central role that synaptic dysfunction plays in AD onset [30,47] the study of miRNAs particularly regulating synaptic function/dysfunction is of special interest and highlighting them constitutes the aim of this review.
Several of the miRNAs mentioned above, known to regulate synaptic proteins, have been suggested as important mediators of AD pathological processes (Table 1). For instance, miR-34a has been proposed as a potential contributor to AD pathology since its increased expression has been observed in APP/PS1 transgenic mice prior to Aβ increase, deposition into plaques, and cognitive deficits (appearing later in this model). In contrast, synaptic plasticity was shown to be enhanced in miR-34a-KO/APP/PS1 mice [109]. Besides multiple synaptic targets already described for this miRNA, including AMPAR and NMDAR, overexpression of miR-34a has been associated with a consistent reduction of sirtuin1 (SIRT1) [80]. Interestingly, other miRNAs, including miR-9 and miR-181c, have also been proposed to target SIRT1 [93] constituting interesting candidates of synaptic regulation in the context of AD pathology. Whereas miR-34a overexpression has also been reported to block tau synthesis by binding directly to the 3 -UTR of human tau mRNA [112], another member of the same family, miR-34c, has been reported to be upregulated in a transgenic AD model and in hippocampal neurons exposed to Aβ. Aβ toxicity was able to alter synaptic vesicle exocytosis through a miR-34c-mediated reduction of VAMP2, a protein component of the SNARE complex. These changes were associated with synaptic failure and learning and memory deficits, recovered after miR-34c increase blockade [105]. miR-134-5p, a well-studied brain-specific miRNA in synaptic function, has been reported to be upregulated after hippocampal Aβ treatment in rats. Interestingly, these results were linked to miR-134-mediated post-transcriptional regulation of CREB and BDNF since knockdown of miR-134-5p abolished CREB and BDNF miR-134-mediated decrease and rescues Aβ induced synaptic deficits [113]. In this way, miR-134-5p could be involved in a main molecular mechanism underlying synaptic plasticity alterations associated to AD.
In addition, miR-206 increased in both Tg2576 mice, and AD brain has also been shown to regulate memory function in AD mice models by targeting BDNF [98]. miR-206 is also able to regulate insulin like growth factor 1 (IGF1) [114]; likewise, miR-26b was associated with an increase in Aβ production through inhibition of IGF1 translation in vitro [97] regulate neprilysin (NEP), a type II transmembrane glycoprotein found on the pre-synaptic membrane and which downregulation has been linked to Aβ increased levels in mice AD model [99].
Related to AMPAR, different subunits have been validated as targets for several miRNAs in vitro and linked to cognitive deficits in experimental rodent models of AD. This redundant targeting of different glutamate receptor subunits suggests a complex miRNA-mediated post-transcriptional regulation of AMPAR trafficking during homeostatic scaling and subsequent role in synaptic plasticity alterations present in AD. For instance, the downregulation of miR-181a in 3xTg-AD mice rescued memory impairment through restoration of GluA2 levels [110]. By regulating GluA2 transcripts, miR-181a participate in Aβ-induced synaptic alterations while in the hippocampus it has been shown to regulate soluble tau levels, constituting a negative regulator of synaptic plasticity [110]. Furthermore, miR-30b, miR-186-5p, and miR-218 have also been reported to bind Gria2 3 -UTR and their overexpression have been associated with a decrease in GluA2 levels regulating synaptic function and cognitive decline in AD experimental models [76,115]. Other miRNA targeting GluA2, miR-124, has been reported to be strongly increased in the hippocampus of AD patients and also in AD mice model (Tg2576) accompanied by deficits in synaptic transmission, plasticity, and memory impairment [116]. Mechanisms underlying these synaptic deficits were further studied in vitro using hippocampal neurons, where oAβ treatment produce a miR-124-mediated decrease of PTPN1 (protein tyrosine phosphatase non-receptor type 1), impairing GluA2 membrane insertion [116]. Moreover, studies in P301S mice indicate that miR-124/PTPN1 alterations could also be modulating tau phosphorylation state through kinase/phosphatase activity imbalance [112].
Taken together, these studies showed that levels of miRNAs regulating glutamate receptors and synaptic-related proteins are altered in AD patients and mouse models, suggesting that they may play key pathological roles, besides informing about early synaptic deficits during the disease progression.

Role of miRNAs in Aβ/Tau Mediated Synaptic Dysfunction and Neuroinflammation
An increasing amount of studies in the area have already provided evidence of miRNAs dysregulated in AD, which are implicated in regulating key genes involved in the disease onset, such as PSEN1, APP, or BACE1 [111,117].
Studies using experimental models of AD have allowed the identification of miRNAs directly linked to tau neuropathology in AD. Regulating tau phosphorylation, specific miRNAs can modulate tau affinity for microtubule, maintenance of microtubule network, and tau aggregation/deposition in NFTs [43,118,119]. For instance, miR-125b and miR-138, both frequently upregulated in AD, have been shown to increase tau hyperphosphorylation and aggregation in neuronal cultures [120] and to impair associative learning in fear conditioning test in mice model of AD [121]. Moreover, miR-132, consistently decreased in AD studies [57,122], has also been associated with tau phosphorylation state regulation [111]. Increased tau hyperphosphorylation and aggregation was observed in 3xTg-AD mice expressing lower levels of miR-132 accompanied by long-term memory deficits [123].
Furthermore, miRNAs targeting directly APP have also been described supporting the role of miRNAs in AD pathogenesis. Specifically, miR-101 levels downregulation observed in AD brain [88] is consistent with in vitro studies where inhibition of miR-101 increase APP levels [127]. Moreover, miR-16 and miR-147 have also been described to target APP in experimental models of AD pathology [128][129][130]. Interestingly, miR-16 expression has also been related to tau phosphorylation in primary cortical neurons [131].
Moreover, several miRNAs have been suggested to regulate metalloproteinase ADAM10 levels, including miR-23a, miR-34a, miR-107, and miR-451 [80,132,133]. Since APP is one of the well-known substrates of ADAM10, these miRNAs would be involved in Aβ metabolism. Additionally, other substrates with potential synaptic effect including cell adhesion molecules such as Neuroligin-1, N-Cadherin or NrCAM, have also been established for ADAM10 [134] relating these miRNAs to synaptic deficits linked to AD.
Otherwise, it is known that reactive astrocytes and inflammatory microglia are part of the pathological AD environment, although outside the scope of this review, it is interesting to point out that both cell populations produce and secrete extracellular vesicles (EVs) that are able to fuse with target cells to which they transfer their content (containing protein, lipids, and RNA species among which miRNAs are the most abundant type). This physiological communication channel between different cell types is known to be altered in pathological conditions [135]. In this regard, EVs secreted from reactive microglia were shown to be enriched in specific miRNAs, including miR-146a-5p that can be transferred to neurons, thereby downregulating neuronal synaptic targets such as synaptotagmin1 (Syt1) and neuroligin1 (Nlg1), negatively affecting dendritic spine density and excitatory synapses reduction [106]. Of interest, miR-26a, enriched in astrocyte-derived EVs, can be transferred to hippocampal neurons modulating their dendritic complexity [50].
Therefore, besides regulating synaptic-related proteins and transcription factors implicated in synaptic plasticity, miRNAs targeting genes related to Aβ and tau metabolism and activity can also contribute to synaptic dysfunction during AD, which may be further modulated by glial cells during neuroinflammation.

Current Biomarkers for AD
The increasing incidence of AD project devastating numbers: 152 million people will be living with dementia by 2050 [136] while the annual cost of dementia will reach 1 trillion dollars. The massive impact of the disease on the health and economic systems, as well as on families and caregivers, raises the pressure to accelerate the search for effective treatments able to, at least, delay the disease progression, but also to validate diagnostic tools that enable earlier detection of the disease [136]. Since MCI patients converted to AD at an annual rate of 17.2 percent [137], an earlier diagnosis would allow a better scenario for potential therapies to succeed, avoiding the limitation of an environment where neuronal death is already widespread and allowing the inclusion in clinical studies of patients in early stages of the disease, increasing the possibility of achieving better therapeutic results.
Currently, the use of biomarkers that can identify AD during its asymptomatic phase [138][139][140] is limited to Aβ 1-42 /Tau levels in cerebrospinal fluid (CSF) and imaging techniques such as the positron emission tomography (PET) that allow the observation of Aβ accumulation in the brain or 18F-fluorodeoxyglucose-PET (FDG-PET) to detect the decrease in glucose metabolic rate. Even combination of imaging approaches at different stages of the diagnosis are being evaluated to improve the available diagnostic tools [141]. However, due to their invasiveness and/or cost, they are not able to be included in routine clinical screenings, needed for early detection. In addition, recent studies discourage the use of CSF Aβ levels as a tool for detecting MCI cases since the accuracy in these cases is lower than recommended [142].
A promising candidate for monitoring disease progression and treatment response in pre-clinical research is the Neurofilament Light Chain (NfL). Remarkably, work done on the Dominantly Inherited Alzheimer Network (DIAN) demonstrated that the increased NfL levels in serum and CSF predict disease progression and brain neurodegeneration at asymptomatic stages of familial AD [143]. Of importance, NfL would constitute a general marker of neurodegeneration regardless of the subjacent cause [144].

Synaptic-Related miRNAs as Early Biomarkers for AD
The potential value of other molecules present in biological fluids (including CSF, blood, urine, and tears) to differentiate between AD and control subjects are also under evaluation. In this regard, miRNAs, close modulators of the specific pathogenic processes underlying AD and present in all these circulating fluids, are presented as promising candidates. Easily measured by simple and affordable techniques, including quantitative real-time PCR (RT-qPCR), miRNAs detection would allow a cost-effective and non-invasive method that could certainly be included in screening studies and patient follow-up over time.
miRNAs dysregulated either in brain or circulating fluids during AD pathology have been described over the last years and are the subject of recent review articles [145,146]. Some miRNAs already commented in this review (including miR-26b, miR-34a/c, miR-125b, miR-146a, and miR-210) have been described to be altered in brain and blood from AD patients, although the direction of changes is not always consistent between both miRNA sources [145,146]. Moreover, miRNAs isolated from AD plasma and serum (including miR-107, miR342-3p, miR-15b-5p, miR-545-3p, and miR-191-5p) have been proposed as potential AD biomarkers [145]. In addition, miR-455-3p, related to Aβ toxicity modulation, has been proposed as potential AD biomarker since increased levels observed in serum are consistent with levels in AD brains, fibroblasts, lymphocytes, and even AD transgenic models. Furthermore, a panel of miRNAs involved in pathological processes underlying AD, such as neuroinflammation, has been proposed as a diagnostic tool to predict AD onset, since the ability of a single miRNA to detect prodromal AD has not been fully demonstrated [146].
While intensive research is being done on miRNA-based biomarkers for AD, the relationship between synaptic function regulation and AD biomarkers is not present in many studies in the field. Most relevant findings in synaptic-related miRNAs obtained from circulating biofluids of MCI and AD patients and their potential value as biomarkers are summarized in Table 2, where it is evident that most studies have been done in blood samples, including serum and plasma, indicating an interest to explore less invasive biomarkers.   miRNAs also included in a molecular signature in Table 3  For instance, miR-132 decrease in serum from MCI and AD patients [151,161] has been reproduced in plasma samples [160], however an increase has also been reported in MCI subjects by Sheinerman and colleagues [159], evidencing that reproducibility between studies can be difficult even when obtaining miRNAs from the same sample source. In the same study, they have also reported the increase of another member of the same family, miR-134, and remarkably, both miRNAs showed notable performance with an area under the curve (AUC) over 0.9 (close to the ideal value of 1, which represents the maximum accuracy of a classifier distinguishing between two groups [169] and sensitivity between 82% and 86% for detection of MCI cases. Interestingly, miR-132 has also been proposed as a member of a serum-based signature for MCI detection, together with miR-206, which is also downregulated in MCI serum [161]. With an outstanding AUC of 0.98, a sensitivity of 85%, and specificity of 98%, this synaptic-related miRNA signature seems certainly promising.  The use of miRNA-based signatures, which consider the alteration of more than one miRNA at the same time, can give higher accuracy, sensitivity, and specificity values that could be an advantage for potential diagnostic tools as exposed in Table 3. In this regard, another signature based in four serum-miRNA levels, including synaptic-related miR-29a, miR-125b, and miR-23a has shown an encouraging performance for differentiating AD cases from healthy cognitive controls (HCC) [149]. The diagnostic value of miR-29a/b family has been tested both in serum and CSF [148,149,151,154,155] although results are not consistent between studies, the alteration of these miRNAs in biological fluids during AD pathology seems to be clear. On the other hand, miR-125b and miR-23a increase has been consistently observed in serum [149,158], displaying good capacity of differentiation between AD and control subjects specifically for miR-125b alone. The potential of miR-125b has also been tested in CSF [152,154], where it has been proposed as a specific tool since, in addition, it is able to distinguish AD from FTD patients [152]. An increase in related miR-125a levels has also been described in CSF from AD patients [151,155], constituting a potential biomarker as previously reported, with special value differentiating also AD from FTD patients [151].
Altered levels of miR-34a were observed both in plasma and CSF [94,154,156], and values near 0.8 were reported for the AUC for plasma-based miRNA performance as AD biomarker. Another member of miR-34 family, miR-34c, was reported to be increased in plasma samples from AD patients, and more interestingly, in mild AD patients [94]. Furthermore, both family members have been included in a signature combining miRNA and piRNAs (PIWI-interacting RNAs) that can distinguish AD cases with 83% accuracy. Although the great value of this signature is predicting the conversion of MCI to AD, this capacity is due to piRNAs levels [172].
Regarding longitudinal studies, to our knowledge, only a few articles have evaluated miRNA levels over time as a predictive tool of MCI progression to AD. Indeed, miR-206 (mentioned above for its potential to distinguish MCI patients) has also shown a good prognostic capacity to classify MCI progression to AD over five years [165]. In another study, miR-181a and miR-146a levels in blood from MCI patients have been documented as a useful tool to distinguish between MCI patients that progress to AD (pMCI) and those who remain stable in MCI state (sMCI) two years after the first measure [162]. The levels in serum of another member of the family, miR-181c, have also been reported to distinguish MCI and AD from control subjects, although the direction of level changes is inconsistent between studies [148,150,158]. Interestingly, an increase in miR-181c in plasma samples from MCI and AD patients has been reported to be specific compared with an FTD cohort [163]. Moreover, both miR-181a and miR-181c have been included in individual plasma-based miRNA signatures. The first one, composed of six miRNAs [170], also include miR-9, which levels have been reported to similarly be altered in wholeblood [173], serum [148], and exosome-enriched CSF [147] from AD patients compared to HCC. With a high AUC, sensitivity, and specificity, this signature can differentiate AD, PD, and FTD from HCC, constituting a useful tool for detecting general neurodegeneration. Interestingly, the signature including miR-181c, is composed of two other synaptic-related miRNAs: miR-92a-3p and miR-210-3p, and besides the potential for distinguishing MCI and AD patients from HCC, promising preliminary results were reported in the signature capacity for evaluating MCI progression to AD after a follow-up of the patients from 1 to 11 years [163]. The potential value of miR-92a as biomarker for AD has also been evaluated in serum, where its levels were shown to be decreased during AD pathology [151]. miR-26b consistent upregulation has been reported in serum and whole blood samples from AD patients [152,153,174], whereas downregulation of miR-26a [150] has been reported in a study that comprised one of the biggest cohorts of AD patients included in this review. While miR-26b levels in CSF samples exhibited specificity for AD detection with an AUC value of 0.82 and the capacity to distinguish AD from FTD patients [152]; differentiation between AD and FTD patients was not possible in serum despite a higher AUC (0.97) for AD cohort [151]. Interestingly, both members of the family were comprised in a blood-based miRNA signature with other 10 miRNAs [171] including miR-107, also proposed itself as plasma biomarker for MCI and AD detection. Remarkably, this signature was tested in other neurodegenerative diseases, such as MS and PD, exhibiting specificity for MCI and AD detection. Furthermore, while miR-26b was also included in a CSF-based signature together with miR-125b [152]; miR-26a was included in serum-based miRNA signature together with miR-181c and other four miRNAs. This last signature's AUC, sensitivity, and specificity is the highest included in this review for detecting AD cases, with 0.99 accuracy, 93% sensitivity and 99% specificity [150]. miR-137 and miR-501, both known to target GRIA1, have been proposed as potential AD biomarkers after decreased levels identification in serum [148,168], while miR-128 [159] and miR-191 [164] have been proposed as plasma-based biomarkers for AD with an AUC performance over 0.95. miR-191 has been considered together with miR-15 for diagnostic value improvement, and slightly better results were achieved, especially for specificity [164].
Consistent downregulation of miR-146a levels was reported in serum, plasma, and CSF from AD patients [151,154]. Despite the small cohort size, downregulation of miR-451 was also observed in plasma EVs with high AUC value distinguishing AD cases from HCC and specific results compared to Dementia with Lewy bodies (DLB) plasma samples [166]. Increased levels of miR-485-3p were reported in AD serum compared to healthy controls, with an AUC value of 0.93 differentiating both groups [167].

Potential and Limitations of Synaptic-Related miRNAs as Early AD Biomarkers
As mentioned above, miRNAs could also be isolated from exosome-enriched or EVs fraction, though the best source for obtaining miRNAs-enriched samples is still unclear. Some studies propose that exosome-related miRNAs could represent a stable source of miR-NAs over time and differentially expressed during pathological states [175,176], whereas other reports suggest that miRNAs expression within exosomes could be certainly low [177]. Even miRNAs included in vesicles and whole plasma cell-free miRNA profiles have been reported to be different [178]. The possibility of diverse miRNA sources, including CSF, plasma, and serum, is an advantage but also raises a problem when looking for reproducibility between studies. However, limitation goes beyond the above mentioned, since reported changes are often inconsistent within the same sample type, as observable in Table 2. The cohort size the parameters used for groups classification, subjects' inclusionexclusion criteria, miRNAs extraction and quantification methods, and statistical analysis performed, could make intricate the comparison between studies. Moreover, although the ROC curve analysis is not the only method available for testing diagnostic value, it is certainly the most used. However, the information given in some studies is limited to the significant or not significant differential expression between groups, which beyond the importance of this information; it does not determine the diagnostic value of a molecule itself.
The advantages and potential value of miRNAs as early biomarkers for AD highlight the urgent need for protocols standardization as an essential tool that would allow a faster progress in obtaining more reliable results in order to bring the advances to the clinics. Remarkably, molecular diagnostics companies, such as DiamiR, are already developing and commercializing miRNA-based technologies, constituting a reflection of the progress made in the area and the real possibilities of incorporating miRNAs as biomarkers for AD not only in screening and diagnosis but also as a valuable tool for improving definition of clinical trials participants.

miRNAs as Therapeutic Targets
In addition to the potential of miRNAs as biomarkers for AD, their capacity to regulate several targets related to the pathology suggest that they may be also considered as therapeutic targets. miRNAs can be modulated pharmacologically by administration of specific compounds (such as anti-inflammatory drugs) or by using antisense oligonucleotides or double-stranded synthetic oligonucleotide miRNAs that inhibit or mimic endogenous miRNAs function, respectively. Recent studies using in vitro approaches and AD animal models have shown that regulation of specific miRNAs may impact on neuroprotection, cognitive function, and neuronal regeneration (reviewed in [179]). Interestingly, some synaptic-related miRNAs discussed above, including miR-30a, miR-124, miR-128, and miR-146a, which regulate targets related to Aβ accumulation and synaptic dysfunction, have been suggested as potential therapeutic targets for AD [179]. In addition, miR-485, related to neurotransmitter release [86] and neuroinflammation [167], has also been proposed both as potential AD biomarker and therapeutic target [167]. Furthermore, multitargeted therapeutic strategies such as the combination of acetylcholinesterase (AChE) inhibitors and modulation of specific miRNAs are also under evaluation [180].
Two key limitations of approaching miRNAs as therapeutic targets are 1) their capacity to regulate several transcripts (up to hundreds of them) simultaneously, and 2) the difficulty to achieve efficient miRNA delivery. Regarding the latter, a recent study using engineered exosomes to deliver miR-29 in a rat model of induced Aβ pathology, showed a rescue of memory deficits [181].

Concluding Remarks
Substantial progress has been made in the study of miRNAs as biomarkers of AD, which is especially important in finding miRNAs capable of detecting asymptomatic stages of the disease. In this review, we highlighted the huge potential of synaptic-related miRNAs in this regard; nonetheless, much remains to be explored. For instance, the potential of miRNAs for predicting MCI conversion into AD dementia has only been explored in a few studies. Therefore, it is necessary to widely perform longitudinal studies in future research. Screening of pre-existing blood banks could be a valuable source of material for studies that aim to elucidate the alteration of specific miRNAs during the pathology progression. This kind of studies would allow the acquisition of an immense amount of data on the evolution of specific miRNAs levels over the years in healthy controls and MCI and AD patients, which would allow not only to confirm the significance of miRNAs in AD diagnosis but also to probe their potential as a valuable prognostic tool.

Data Availability Statement:
No new data were created or analyzed in this study. Data sharing is not applicable to this article.