Plasma miRNAs as Biomarkers for Amnestic Mild Cognitive Impairment and Modulation on BACE1

Background: Amnestic mild cognitive impairment (aMCI) is a prodromal stage of Alzheimer’s disease with the beta-site amyloid precursor protein cleaving enzyme 1 (BACE1) imbalance. Altered microRNAs (miRNA) associated with BACE1 may be potential biomarkers for aMCI diagnosis. The aims of this study were to nd an effective diagnostic model for aMCI, and to explore the modulation of candidate miRNAs on BACE1 expression. Methods: Plasma RNA was extracted from participants enrolled in China longitudinal aging study (CLAS). MiRNA proling was performed using microarray sequencing on a discovery set, compromising ve aMCI subjects that conrmed by 18 F-Flutemetamol PET scan and ve normal controls (NC). Quantitative reverse transcription PCR was used to validate the expression levels of differently expressed miRNAs on an analysis set (20 aMCI subjects and 10 NC). Diagnostic capability of candidate miRNAs was assessed in a validation set (40 aMCI subjects and 40 NC). Modulation of candidate miRNAs on BACE1 was explored in rat and human hippocampal neurons in vitro through transfection of miRNA mimics or inhibitor lentivirus. Results: In discovery set, we veried 46 signicantly differentially expressed miRNAs between aMCI and NC groups (P<0.05). Among these, 33 miRNAs were down-regulated and 13 miRNAs were up-regulated. In analysis set, miR-1185-2-3p, miR-1909-3p, miR-22-5p, and miR-134-3p also signicantly decreased in aMCI group. Four miRNAs above and miR-107 which was found to decline in previous study were selected as potential biomarkers. A diagnostic model consists of ve miRNAs above had an outstanding diagnostic accuracy (81.25%) to diagnose aMCI. Except for miR-134-3p, other four miRNAs modulated BACE1 expression and C-terminal fragments-beta production effectively in rat hippocampal neurons in vitro. Similar modulation of miR-1185-2-3p and miR-1909-3p were conrmed in human hippocampal neurons in vitro. Conclusion: A diagnostic model consists of ve plasma miRNAs

bind to Aβ would have only marginal effects on preexisting amyloid 8 . Thus, many failures remind us that the gene regulation on the production of Aβ in prodromal stage of AD should be concern.
Recent studies advanced the view that cognitive decline in AD is a continuum, while mild cognitive impairment (MCI) is de ned as an intermediate phase between normal cognition and dementia 9 .
Amnestic MCI (aMCI) may have a higher risk of developing AD 10,11 . Clinical characteristics, neurocognitive tests, positron emission tomography (PET) and magnetic resonance imaging (MRI) can diagnose aMCI but they are time consuming 12 . Amyloid PET with different probes ( 18 F-utemetamol and 18 F-orbetapir, et al.) are available to precisely detect and monitor brain amyloid deposition in MCI patients but costliness 13,14 . Longitudinal studies showed that amyloid-positive MCI subjects have greater cognitive deterioration than amyloid-negative subjects 15,16 . In addition to technologies, body uids biomarkers should also be used for accurate diagnosis of MCI. However, CSF assay requires a lumbar puncture which is an invasive procedure, and neuroimaging requires high costs and sophisticated equipment, making their universal utility impractical. Thus, blood biomarker could be an attractive option, since getting blood sample is an economical, non-invasive and simple procedure.
MicroRNA (miRNA) is a class of noncoding RNA, ne-tuning the protein-encoding genes at a posttranscriptional level 17,18 . Several miRNAs have been shown to play signi cant role in brain and neuronal development 19 . Aberrant expression of miRNAs may predict onset of cognitive dysfunctions 20 . Recently, the role of altered miRNAs in the diagnosis for AD has been extensively studied. Several miRNAs have been shown to regulate AD-relative genes, such as amyloid precursor protein (APP), beta-site APP cleaving enzyme 1 (BACE1), and brain-derived neurotrophic factor (BDNF) [21][22][23] . This ability of selected miRNAs to target mRNAs which are altered in disease conditions makes them potential candidate as therapeutics or as targets of therapeutics. However, few studies have distinguished different subtypes of MCI. Since not every MCI-affected subject will develop AD, it is important to improve the accurate diagnostic criteria for aMCI. In this study, we aimed to nd an effective diagnostic model composed of plasma miRNAs for early diagnosis of aMCI, and to investigate the modulation mechanism of miRNAs on BACE1 expression in vitro.

Study Population
The work ow of this study is shown in Fig 1a. All plasma samples were collected from subjects enrolled in the China longitudinal aging study (CLAS, ClinicalTrials.gov Identi er: NCT03672448) 24  Petersen's criteria 25 was used to diagnose aMCI: i) memory complaint, preferably corroborated by a spouse or relative; ii) objective memory impairment; iii) normal general cognitive function; iv) intact activities of daily living; v) absence of dementia. In discovery set, clinical diagnosis of aMCI was con rmed by 18 F-Flutemetamol PET scan according to the 2011 criteria of National Institute on Aging and Alzheimer's Association. Controls were age-, gender-, and educational level-matched elderly. Subjects with other mental disorders, nervous system diseases, and history of psychotropic medicines were excluded.
Peripheral blood was collected in every subject after fasting for 12 hours. Then plasma was separated by centrifugation at 3,000 rpm for 20 minutes at 4℃ and stored at -80℃ until required. miRNA microarray sequencing in discovery set RNA was isolated using TRIzol reagent (Invitrogen life technologies) and puri ed with RNeasy mini kit (Qiagen, Germany) according to manufacturer's instructions. RNA quality and quantity were determined by 260/280nm absorbance using a Nanodrop ND-1000 spectrophotometer (Nanodrop Technologies).
MiRNA expression pro les were carried out by applying the miRCURYTMTM LNA Array (v.19.0) (Exiqon, Denmark). Replicated miRNAs were averaged and miRNAs that intensities ≥30 in all samples were chosen for calculating normalization factor. Expressed data were normalized using the Median normalization. After normalization, signi cant differentially expressed miRNAs between two groups were identi ed through fold-change and P value of two-sample independent t-test.

Replication of results in an analysis set
The expression levels of miRNAs were con rmed by real-time quantitative reverse transcription PCR (qRT-PCR) in an analysis set. RNA was extracted with TRIzol LS Reagent (Invitrogen, USA). MiRNA quality and quantity were determined by 260/280nm absorbance using a Nanodrop ND-1000 spectrophotometer (Nanodrop Technologies). Total of 300ng RNA was used to prepare cDNA samples on Gene Amp PCR System 9700 (Applied Biosystems, USA). PCR reaction was set in the QuantStudio5 Real-time PCR System (Applied Biosystems, USA). Primers were designed by Primer 5.0 (Supplementary Table 1). The relative levels of miRNAs were determined in terms of their fold change, using the formula (2 -∆∆CT ) 26 . Hsa-miR-93 was used as endogenous control. Real time qRT-PCR was performed in triplicate.

Construction of Diagnostic Model
To identify independent predictive parameters of aMCI, univariable and multivariable logistic regression analyses were performed in each miRNA plasma levels. P<0.05 was considered statistically signi cant. Parameters with P<0.05 based on the univariate analysis were further included in the multivariable logistic regression analysis. According to the multivariable logistic regression results, we performed a diagnostic model. The diagnostic accuracy of this model was examined using a receiver operating characteristic (ROC) curve analysis. The best cut-off value was selected according to Youden-index.
Rat primary hippocampal neurons were isolated from E18 embryos according to the following protocol. The protocol has received approval by the Institution's Ethical Committee of Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine. Euthanize the pregnant rat by inhalation of CO 2 . Remove the uterine and place it in cold phosphate buffer saline (PBS) as soon as possible. Then remove the embryos from uterine segments and place in a new petri dish containing cold PBS. Remove the brain from head and dissect the embryonic rat hippocampal tissues under surgical microscope after decapitating embryos. Digest the tissues with 0.15% trypsin for 20 min at 37℃. To lter the hippocampal neuronal suspensions with 70-μm cell strainers. Neurons were plated on poly-L-lysine coated 6-well plates at ~100,000 cells/cm2, then cultured in neurobasal media supplemented with 2% B27, 0.25% Glutamax, 0.25% Glutamate, and 0.5% Penicillin/Streptomycin. The media were replaced with another neurobasal media supplemented with 2% B27, 0.25% Glutamax, and 0.5% Penicillin/Streptomycin next day.
All cells were cultured at 37℃ in humidi ed incubator with 5% CO 2 environment.

Lentivirus transduction
Lentiviral particles containing the GV358 expression vector encoding BACE1 were purchased from GeneChem (Shanghai, China) and were transduced into SH-SY5Y cells following the manufacturer's instruction. The virus-infected SH-SY5Y BACE1 cells, stably overexpressing BACE1, were ltrated by puromycin.
Real-time qRT-PCR For cultured cells and neurons, total RNAs were extracted with TRIzol LS Reagent (Invitrogen, USA). RNA quality and quantity were measured by Nanodrop-2000 (Thermo Fisher Scienti c, USA). The cDNAs were synthesized using HiScript II 1st Strand cDNA Synthesis Kit (Vazyme, China) according to manufacturer's instructions. PCR reactions was set in the Fast-7000 Real-time PCR system (Thermo Fisher Scienti c).
Threshold cycles (C T value) were generated automatically, and the relative expressions were shown as (2 -∆∆CT ) 26 . Real-time qRT-PCR was performed in triplicate. Relative mRNA levels of target genes were normalized to the indicated reference genes. Primers were synthesized by Sangon Biotech (Shanghai, China. Supplementary Table 1).

Western blotting analysis
Cells or neurons were lysed in RIPA Lysis Buffer (Beyotime, Shanghai) which contained 1mM of Phenylmethanesulfonyl uoride (Beyotime, Shanghai) according to manufacturer's instructions. Cell lysate samples were electrophoresed on SDS-polyacrylamide gels in Tris-glycine buffer containing SDS.
Proteins were transferred to 0.22μm NC-membranes. Then membranes were blocked in 3% Bovine Serum Albumin (BSA) diluted in TBST buffer at room temperature, and were incubated overnight at 4℃ with primary antibodies (

Screening of miRNAs as potential biomarkers
Total of 3,100 mature miRNAs were detected in a discovery set, which consists of ve aMCI subjects con rmed by 18 F-Flumetamol PET (amyloid positive) and ve normal controls (NC, Table 1). There were 46 differently expressed miRNAs between aMCI group and NC group (P<0.05, Fig. 1b). As compared with normal controls, 13 miRNAs were up-regulated and 33 were down-regulated in aMCI group (Fig. 1b). Among down-regulated miRNAs, 31 miRNAs are human miRNAs. Besides miRNAs above, we were also interested in miR-134-3p, the mature sequence of miR-134. Reason are as followed: i) an outlier value in miR-134-3p was found in microarray sequencing data (Data are openly available in GEO database, reference number [GSE147232]), which may cause a low fold-change value with high P value (Fold-change=0.099, P=0.362); ii) miR-134 is abundantly expressed in brain and was demonstrated to regulate synaptic plasticity and memory formation 27,28 ; iii) few previous studies have focused on miR-134-3p.
Thus, miR-134-3p together with thirty-one signi cantly decreased miRNAs above were selected for secondary screening.  Figure 2. Mature sequence of miR-2-3p has three putative binding sites. The remaining three miRNAs has one target site on BACE1 mRNA.
To validate microarray results, an analysis set consists of 20 aMCI subjects and 10 normal controls (Table 1), was utilized to quantitate the above selected miRNAs through real-time qRT-PCR. Except miR-5691, the remaining four downregulated miRNAs in aMCI subjects which were consistent with microarray results (Fig. 3a). Our previous study demonstrated a high capability of plasma miR-107 to discriminate aMCI subjects from normal controls 32 . Furthermore, miR-107 has a potential target site in the BACE1 mRNA 21,32,33 . Hence, combining the results above, miR-107 together with miR-1185-2-3p, miR-1909-3p, miR-22-5p, and miR-134-3p were selected as candidates for potential biomarkers.
The predictive effects of ve altered miRNAs were performed using logistic regression. That lower levels of each miRNA increased the risk of aMCI according to univariate logistic regression (all P<0.05, Table 2). All factors were subsequently included in the multivariable logistic regression (Enter) analysis. The results showed that miR-1909-3p (P=0.001) and miR-107 (P=0.023) might be independent factors for diagnosis of aMCI (Table 2). According to multivariable logistic regression results, we performed a diagnostic model:  Fig. 3d). According to Youden-index, the optimal cut-off value for this model was 0.174, the sensitivity and speci city were 80.0% and 82.5% separately. protein levels in SH-SY5Y BACE1 cells (Fig. 4b and c). It demonstrated that these miRNAs suppress BACE1 expression by both inhibiting translation and destabilizing mRNA. Inhibition of endogenous miR-107 and miR-1185-2-3p signi cantly increased BACE1 protein levels separately (Fig. 4e and f). Nevertheless, inhibition of other three miRNAs did not increase BACE1 protein levels, which might be due to the competition with other miRNAs for the same targeting site. One of the potential multiple reasons that miR-1185-2-3p regulated the target gene more strongly than other selected miRNAs is that it has more binding positions than the others (Fig. 2).

Modulation on BACE1 expression in hippocampal neurons
The elevated expression of BACE1 may exacerbate endoproteolysis of APP and promote subsequent production of CTF-β and toxic soluble Aβ peptide, thereby contributing to the pathophysiology of AD. Therefore, we next investigated effects of selected miRNAs on modulating BACE1 expression and CTF-β production.
In rat hippocampal neurons, overexpression of miR-1185-2-3p, miR-1909-3p, miR-22-5p, and miR-107 separately suppressed BACE1 expression as compared with control ( Fig. 5c and d). Downregulation of subsequent CTF-β production was observed in parallel with BACE1 protein levels ( Fig. 5c to e). Conversely, inhibition of these four miRNAs signi cantly increased the BACE1 protein levels ( Fig. 5f and g), indicating that BACE1 levels are regulated by these miRNAs in hippocampal neurons. Upregulation of BACE1 levels led to the signi cant increase in production of CTF-β ( Fig. 5f to h).
In human hippocampal neurons, inhibition of miR-1185-2-3p/miR-1909-3p signi cantly increased the BACE1 protein levels and promoted subsequent CTF-β production (Fig. 6), in line with the results found in rat hippocampal neurons. Interestingly, downregulation of miR-134-3p only elevated BACE1 protein and CTF-β protein levels in human hippocampal neurons, but not in rat. It may due to the low conservativeness of miR-134-3p between human and rat. Meanwhile, overexpression of BACE1 levels increased APP expression may through positive regulation mechanism.

Discussion
In the face of continuous failure of new drugs, searching for effective diagnostic biomarkers for prodromal AD or aMCI is currently urgent. Although existing neuroimaging techniques are available to detect Aβ deposition very early, the high costs make it unavailable for screening. Tau level in plasma is too low to detection, while Aβ peptides themselves may not serve as effective biomarkers. Circulating miRNAs offer a rapid, convenient, non-invasive and cost-effective method for early diagnosis. Currently, may efforts have been made to discover altered circulating miRNAs in AD patients, which providing many potential biomarkers. However, few studies have focused on aMCI subjects. In this study, we used microarray sequencing to provide altered miRNA pro les in plasma of aMCI subjects and narrow the eld of potential peripheral biomarkers.

Altered miRNA pro les between aMCI and controls
Using microarray sequencing, we detected 3,100 mature miRNAs in plasma of aMCI and normal controls. In a discovery set, 46 differentially expressed miRNAs were found between aMCI and normal controls, 13 miRNAs were up-regulated and 33 miRNAs were down-regulated. In an analysis set, applying bioinformatics methods and real-time qRT-PCR technique, four of the original 31 down-regulated human miRNAs were replicated (miR-1185-2-3p, miR-1909-3p, miR-22-5p, and miR-134-3p). As compared with previous study, current study is the rst to report dysregulation of miR-1185-2-3p, miR-1909-3p, miR-22-5p, and miR-134-3p in plasma of aMCI subjects. We hypothesized these four miRNAs above and miR-107 32 as candidate biomarkers for aMCI for further validation. In a validation set, we veri ed that miR-1909-3p and miR-107 are potent biomarkers separately. The high diagnostic accuracy of miR-107 observed here is consistent with our previous study 32 . Several studies have revealed that miR-107 is decreased in brain and bio uids in AD patients, and indicated that miR-107 may be involved in AD pathogenesis through regulating BACE1 and CDK5 activity 21,[32][33][34] . This consistency in different tissues and different stages of cognitive decline makes it a stable biomarker. Combination of all ve altered miRNAs in a diagnostic model had high diagnostic accuracy with 81.25%, which considered to be promising biomarkers. This
BACE1 is the rate-limiting enzyme in APP process. Aβ production is owing to sequentially cleaved by βsecretase (BACE1) and γ-secretase from APP 35 . Therapeutic strategies targeting BACE1 may be useful to normalize BACE1 and reduce Aβ production. However, recent clinical studies have shown that cognitive decline is worsen in participants who received BACE1 inhibitor than those received placebo [36][37][38] . BACE1 has multiple substrates that are involved in various processed in brain [39][40][41] . These clinical trials indicated that BACE1 inhibitor lower CSF Aβ levels up to ~90% 42 . The strongly inhibition might cause risk mechanism-based side effects [36][37][38] . Furthermore, several BACE1 inhibitor even elevated BACE1 level by extending the protein's half-life 43 . Thus, how to decrease BACE1 effectively and moderately remains a challenge. Our study demonstrated that over-expression of ve altered miRNAs decrease BACE1 expression approximately 20-50% separately. These miRNAs may be potential treatment target to modulate BACE1. Several miRNAs have been identi ed as regulators of BACE1, suggesting that an extensive network of multiple miRNAs may regulate BACE1 expression in a coordinated manner. The ability of ve selected miRNAs to target BACE1 which are associated with sporadic AD makes these molecules interesting candidates as therapeutics (in the form of miRNA mimics) to reduce brain Aβ in vivo. Several miRNAs mimics are currently under investigation in clinical trials in some cancers 44,45 .
Since miRNAs are important in neuronal biological functions, their role may extend to AD as potential therapeutic targets. Analysis of sequencing data enables us to understand miRNA-target networks and to identify key miRNAs involved in cognitive decline processes.
Inhibition of four miRNAs (miR-1185-2-3p, miR-1909-3p, miR-22-5p, and miR-107) markedly increased expression of BACE1 and CTF-β, which suggesting a tantalizing possibility that certain lower miRNAs levels may initiate Aβ deposition in vivo through weakened inhibition on BACE1 expression. Cells likely have homeostatic modulation mechanism for regulating gene expression, reducing the effects of dysregulation of a single gene regulator dysregulation. Therefore, it seems that multiple altered miRNAs as well as other gene regulators synergize to regulate BACE1 expression in vivo. Our results supported that these four decreased miRNAs play a role in regulation network and may promote AD pathogenesis.

Limitations of study and future direction
There are some limitations in current study. We believe that current study provides preliminary clinical and pathology evidence for the inclusion of plasma-based biomarkers toward aMCI. However, this plasma miRNAs battery may have somewhat limited value in differentiating Alzheimer's disease from other types of dementia. This plasma miRNAs battery may be not speci c to Alzheimer's disease, as one of the miRNAs in the battery, miR-107, is also hypothesized to regulate progranulin/granulin, a molecule known to be important in non-tau frontotemporal dementia 46 . The possibility of differentiating aMCI from other types of MCI (MCI caused by frontotemporal dementia or Lewy body dementia) by this plasma miRNAs battery should be examined in separate future studies. Secondly, effects of miRNAs on modulation of BACE1 were only explored in hippocampal neurons in vitro. Thus, we need more extensive research requires validation in vivo to demonstrate the particular regulatory mechanism of these miRNAs.

Conclusion
Overall, we uncovered a differential expression pro le of both previously identi ed and novel miRNAs battery altered in aMCI subjects. The plasma levels of all ve miRNAs effectively discriminate aMCI from normal controls, suggesting the potential diagnostic value of combinatorial miRNAs. Furthermore, to discover the miRNAs regulation process may help us to deeply understand the fundamental cause of BACE1 imbalance. Last, since these miRNAs are relevant to molecular processes underlying AD by targeting BACE1, further functional characterization of these miRNAs may offer new therapeutic avenues for patients with AD. Availability of data and materials The microarray sequencing data generated during the current study was available in GEO dataset (reference number: GSE147232), https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE147232.
Other datasets generated during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.