Alzheimer’s genetic risk factor FERMT2 (Kindlin-2) is regulated by risk allele-binding of miR-4504 and leads to synaptic dysfunction in an APP-dependent manner

We combined genome-wide high-content screenings to identify miRNAs (n = 2,555) and target genes (n = 18,107) involved in the APP metabolism. This approach highlighted FERMT2 (or Kindlin-2), a genetic risk factor of Alzheimer’s disease (AD). FERMT2 down-regulation by multiple miRNAs modulated APP metabolism and increased Aβ secretion. Among these, miR-4504, which is over-expressed in AD brains, decreased FERMT2 expression depending on the presence of the deleterious rs7143400 variant. Decreased FERMT2 expression altered both axonal growth and long-term potentiation in an APP-dependent manner. Overall, this work provides important mechanistic insight on the roles of FERMT2 and miRNAs in APP biology.


Immunoblotting and Aβ quantification
Equal amounts ( Live-cell microscopy for axon elongation and actin dynamics author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint After DIV5, once the axons reached the axonal chamber of microfluidic devices, the culture medium was replaced with Neurobasal A without phenol red, supplemented with GlutaMax, 2% B 27 , and 25 mM HEPES. Phase-contrast images of growing axons were acquired every 10 min for 110 min using Zeiss AxioObserver Z1 microscope equipped with a Prime 95B Scientific CMOS (Photometrics, Tucson,AZ) camera and 32× objective. Movies were analyzed using Fiji MTrack J Plugin (Meijering et al. 2012) to determine the axon growth speed.
To visualize filamentous actin (F-actin) dynamics in the growth cones of elongating axons, neurons were co-transducted with LifeAct-Ruby at DIV1. At DIV5, growth cones expressing LifeAct-Ruby were imaged using a Nikon microscope equipped with Yokogawa spinning disk system and a Nikon CFI Apochromat 100× TIRF objective (NA 1.49), in live superresolution mode (66 nm/px). Processed movies were analyzed using Imaris (Bitplane, Zurich, Switzerland) surface tracking tool which reports the speed and direction of F-actin puncta undergoing actin retrograde flow.

Visualisation of miRNA expression at the single-cell level
To visualize RNA molecules by fluorescence at single-cell resolution and quantify gene expression, we used ViewRNA Cell Plus Assay kit (ThermoFischer Scientific) according to the manufacturer's instructions. Briefly, after fixation and permeabilization, cells were washed 3× with PBS containing RNAse inhibitor and were incubated with probes directed against specific mRNA or miRNA for 2 h at 40°C. After washes, probes were amplified, first, in a pre-amplifier solution and secondly, in an amplifier solution, both for 1 h at 40°C. Then, cells were incubated with nucleotide probes stained with different fluorophores allowing the detection of mRNA or miRNA puncta. This approach was coupled with immunofluorescence experiments described. author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint miR quantification in human brain samples This study was approved by CHU de Québec -Université Laval Research Ethics Committee (#2017-3017). Frozen human brain tissue (0.5-1.2 g per sample) was obtained from the Harvard Brain Tissue Resource Center in Belmont, USA, the Brain Endowment Bank in Florida, USA, and the Human Brain and Spinal Fluid Resource Center in Los Angeles, USA, via NIH Neurobiobank. The cohort of patients included non-dementia controls (N = 30) and AD cases (N = 52), based on neuropathological diagnosis. Upon receipt of the specimens, frozen post-mortem parietal coxtex (BA39) was crushed using a biopulverizer prior to RNA extraction and analysis.
Then, cells were lysed and immunoprecipitated with streptavidin beads. Precipitated proteins were eluted from the beads with loading buffer containing 50 mM DTT, heated for 10 min at 95°C and analyzed by WB.

Co-Immunoprecipitation
Equal amounts of protein were collected in co-immunoprecipitation buffer (10 mM HEPES, 140 mM NaCl, 0.5% NP-40, pH = 7.4) containing protease inhibitors (Complete mini, Roche Applied Science) and phosphatase inhibitor (100 mM sodium orthovanadate) and incubated with the primary β-amyloid antibody clone 4G8 (RRID:AB_662812) overnight, with gentle rocking. Production of recombinant C100 fragment was performed as previously described (Sevalle et al. 2009). Co-immunoprecipitation was carried out using Pierce Protein A/G magnetic beads kit (Thermo Scientific, 88802) according to the manufacturer's instructions. Samples with proteins and antibody complexes were incubated with 25 µL (0.25 mg) of A/G magnetic beads previously washed with co-immunoprecipitation buffer. After 1 h of incubation at 4°C, the magnetic beads were washed 3×, resuspended with loading buffer (LDS and reducing agent) for 10 min at RT, and analyzed by WB. author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint

Synaptosome extraction
To verify the presence of proteins at the synaptic level we did a subcellular fractionation as previously described (Frandemiche et al. 2014). Briefly, cortical neurons were resuspended in a solution (0.32 M sucrose and 10 mM HEPES, pH = 7.4) and were centrifuged at 1,000× g for 10 min to remove nuclei and debris. The supernatant was centrifuged at 12,000× g for 20 min to remove the cytosolic fraction.
The pellet was resuspended in a second solution (4 mM HEPES, 1 mM EDTA, pH = 7.4) and was centrifuged 2× at 12,000× g for 20 min. The new pellet was resuspended in a third solution (20 mM HEPES, 100 mM NaCl, 0.5% Triton X-100, pH = 7.2) for 1 h at 4°C and centrifuged at 12,000× g for 20 min. The supernatant collected corresponds to the non-PSD fraction (Triton-soluble). The remaining pellet was resuspended in a fourth solution (20 mM HEPES, 0.15 mM NaCl, 1% Triton X-100, 1% deoxycholicacid, 1% SDS, pH = 7.5) for 1 h at 4°C and was centrifuged at 10,000× g for 15 min to obtain a supernatant containing the PSD fraction (Triton-insoluble). The different fractions were then analyzed by WB.

Lentivirus injection
For stereotactic injections, C57Bl6/J mice (RRID:IMSR_JAX:000664) were anesthetized with 4% isoflurane (2 L/min) and placed in a stereotaxic frame (68528,RWD Life Science,Shenzhen,China) in which the head of the animal was fixed with a pair of ear bars and a perpendicular tooth bar. During surgical procedures 1.5% isoflurane (2 L/min) was delivered through a facial mask via spontaneous respiration. Their body temperature was maintained between 36.5 and 37.5°C with a homeothermic blanket. Head was shaved and Vetedine was applied. Wounds and pressure points were infiltrated with lidocaine. A skin incision was made along the sagittal midline of the scalp. Craniotomy was made to target our structures of interest. Lentiviruses were injected in right and left hippocampus (1.5 µL per hemisphere; 0.2 μL/min). After injections, wound clips were used for skin closure. For the sham group, surgical procedures were performed without any injection. During the surgery, the level of anesthesia was regularly verified by testing the nociceptive hind limb withdrawal reflex. Subjects were then allowed to recover in their home cages for at least 7 d before sacrifice for ex-vivo electrophysiological recordings.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint holding chamber filled with aCSF. Slices were allowed to recover in these conditions at least 1 h before recording.

Electrophysiological recordings
For electrophysiological recordings, a single slice was placed in the recording chamber, submerged and continuously superfused with gassed (95% O 2 , 5% CO 2 ) aCSF at a constant rate (2 mL/min).
Extracellular fEPSPs were recorded in the CA1 stratum radiatum using a glass micropipette filled with aCSF. fEPSPs were evoked by the electrical stimulation of Schaffer collaterals/commissural pathway at 0.1 Hz with a glass stimulating electrode placed in the stratum radiatum (100 µsec duration).
To test the effect of miRNA-expressing lentiviruses on basal synaptic transmission, Input/Output (I/V) curves were constructed at the beginning of the experiment. The slope of fEPSPs was measured and plotted against different intensities of stimulation (from 0 to 100 µA).
Stable baseline fEPSPs were recorded by stimulating at 30% maximal field amplitude for 10 min prior to beginning experiments (single stimulation every 10 s (0.1 Hz). The same intensity of stimulation was kept for the reminder of the experiment. After a 10 min stable baseline period, long-term potentiation (LTP) was induced by the following stimulation protocol: 3 trains of 100 stimulations at 100 Hz at the same stimulus intensity, with a 20 s interval between trains. Following this conditioning stimulus, a 1 h test period was recorded where responses were again elicited by a single stimulation every 10 s (0.1 Hz) at the same stimulus intensity. Signals were amplified with an Axopatch 200B amplifier (Molecular Devices, Union City, CA) digitized by a Digidata 1550 interface (Axon Instruments, Molecular Devices, US) and sampled at 10 kHz. Recordings were acquired using Clampex (Molecular Devices) and analyzed with Clampfit (Molecular Devices). Experimenters were blinded to treatment for all experiments.
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint

Screening for miRNAs modulating APP metabolism
We used an unbiased screening approach to identify miRNAs that modulate APP metabolism in our HCS system (Chapuis et al. 2017). Specifically, this model is based on HEK293 cells stably overexpressing an APP fusion protein (mCherry-APP 695WT -YFP) that allows for the quantification of intracellular APP fragments. We screened a total of 2,555 mature human miRNAs in a 384-well plate format. As positive control, a siRNA directed against APP was used to calculate the standardized mean difference (denoted as β in Supplementary Fig. 1). The complete screen was performed in triplicate and plates with β > 3 were analyzed according to HCS guidelines (Bray et al. 2013). This procedure led us to identify 50 miRNAs (top and bottom 1%) with the strongest impact on APP metabolism ( Fig. 1a, Supplementary Fig. 1 and Table 1). Pathway enrichment analysis revealed that these 50 candidate miRNAs are predicted to regulate pathways already identified to involve APP, such as axon guidance (Sosa et al. 2013) and glutamatergic synapse (Martinsson et al. 2019).
To determine which genes were potentially regulated by these top 50 hits, we selected the intersection of predictions resulting from at least four different algorithms (see methods) and thus identified 6,009 putative miRNA-target genes. To further refine the list of predicted genes, we cross-checked them against the 832 genes identified to have a major impact on APP metabolism in our previous genomewide siRNA screen (Fig. 1c). This resulted in 180 common genes that are putative targets of 41 miRNAs.
To determine if any of these 180 genes were preferentially regulated by this pool of 41 miRNAs, we performed 1 million drawing lots of 41 miRNAs among the 2,555 tested and compared them against the list of miRNAs predicted to bind in the 3'-UTR of each of the 180 genes (see Supplementary Fig.   2). Distribution of the obtained p-values is shown in Fig. 1d. FERMT2 (encoding Kindlin-2), an AD genetic risk factor, was among the 7 most significant genes (p-value < 2.7×10 -4 after Bonferroni correction, Fig. 1e), i.e., genes that strongly modulate APP metabolism and whose expression are potentially regulated by miRNAs that also strongly modulate APP metabolism. We therefore focused on this gene.

Endogenous FERMT2 expression is down-regulated by miRNAs
Four (out of 41) miRNAs were predicted to target FERMT2 3'UTR: miR-582-5p, miR-200b-3p, miR-221-3p and miR-222-3p. Of note, we previously identified a variant (rs7143400) within the 3'-UTR of FERMT2, where the minor T allele creates an 8-mer binding site for miR-4504 (Delay et al. 2016) ( Fig. 2a and Supplementary Fig. 3). However, miR-4504 had limited effects on APP metabolism in our HCS (mCherry fold change = 1.18±0.09, rank 688 in the top 30%). Since the potential effects of this miRNA on FERMT2 expression would depend on the presence of the rs7143400 minor T allele, we genotyped the HEK293 cell line (used in our HCS) for this variant. We observed that this cell line was homozygous (GG) for rs7143400 and, as a consequence, FERMT2 expression in this cell line author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint could not be modulated by miR-4504, potentially limiting its impact on APP metabolism. However, since rs7143400 exhibited genome-wide significance in its association with AD risk, we decided (i) to add miR-4504 to the list of four miRNAs resulting from the HCS screening and (ii) to consider the rs7143400 variant in our subsequent analyses.
We first developed a luciferase activity assay in HEK293 cells to determine if the 5 miRNAs of interest could indeed modulate FERMT2 expression. The capacity of each miRNA to induce a downregulation of the reporter gene expression (luciferase activity) in the presence of FERMT2 3'UTR sequence was confirmed (Fig. 2b). As hypothesized, miR-4504 was able to down-regulate reporter gene expression only in the presence of the rs7143400 T allele. We then assessed whether these miRNAs were also able to modulate endogenous FERMT2. For this purpose, we transfected the 5 selected miRNAs in either wild-type HEK293 (HEK293 wt ) or rs7143400-mutated HEK293 cell lines (HEK293 rs7143400-G/T ) generated via the Crispr-Cas9 technology (Supplementary Fig. 4). All 5 miRNAs were able to decrease the endogenous FERMT2 expression; however, miR-4504 only in the HEK293 rs7143400-G/T cell line ( Fig. 2c and 2d). Notably, all 5 miRNAs could down-regulate endogenous FERMT2 expression in HeLa cells that are heterozygous for rs7143400 (Fig. 2e).
Collectively, these data suggest that endogenous FERMT2 expression is dependent on miRNAs expression and that a functional variant is able to modulate the miRNA-dependent FERMT2 expression by creating a binding site for miRNAs.

APP metabolism is regulated by miRNAs
According to our hypothesis, the 5 miRNAs are potential candidates to modulate APP metabolism through a direct down-regulation of FERMT2. However, it is also possible that these miRNAs modulate the APP metabolism through targeting additional genes, making it difficult to determine the true effect of FERMT2 on the read-outs observed. To assess this question, we reasoned that if a candidate miRNA affects the APP metabolism mainly through down-regulating FERMT2, this candidate miRNA would have similar effects on APP metabolism as the direct FERTM2 downregulation by siRNAs we assessed previously (Chapuis et al. 2017). For this purpose, we evaluated the effects of miR-582-5p, miR-200b-3p, miR-221-3p and miR-222-3p over-expression on the APP metabolism in HEK293 WT cells. Only miR-222-3p was associated with an increase in endogenous APP levels (Fig. 3a). We further assessed the impact of miR-222-3p on Aβ secretion by cotransfecting this miRNA with APP cDNA in HEK293 WT . Results showed that miR-222-3p decreased FERMT2 expression and increased Aβ secretion in a dose-dependent manner (Fig. 3b). Of note, miR-582-5p and miR-200b-3p were also predicted to directly target APP, thus making it impossible to assess whether these miRNAs affected APP metabolism directly or through FERMT2.
We also addressed the same question for miR-4504 by taking advantage of the HEK293 rs7143400-G/T and HeLa cell lines. As for miR-222-3p, we observed that miR-4504 transfection was associated with an increase in total APP in both cell lines (Fig. 3c). Nonetheless, our findings indicate that miR-222-3p author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint and miR-4504 have the same impact on APP metabolism as seen by FERMT2 silencing alone. To note, the impact of miR-4504 was only observed in HEK293 rs7143400-G/T cells.

miR-4504 is over-expressed in AD brains
To provide further physiological relevance to our findings, we combined classical immunocytochemisty with RNA hybridisation allowing miRNAs detection at single-copy sensitivity.
We first observed that both miR-222-3p and miR-4504 were expressed in primary hippocampal neurons (Fig. 4a). We next measured the expression levels of miR-222-3p and miR-4504 in the postmortem brain samples from AD patients and control subjects. We observed that the expression levels of miR-4504, but not of miR-222-3p, were significantly increased in AD brains compared to controls (Fig. 4b). Taken together, these data suggest that several miRNAs (some of which are over-expressed in AD brains) are able to down-regulate endogenous FERMT2 expression and modulate APP metabolism. In addition, this miRNA-dependent regulation is modulated by a functional genetic variant associated with an increased AD risk in GWAS. This implies that under-expression of FERMT2 is potentially deleterious in an AD pathophysiological context in neurons.

Interaction between FERMT2 and APP controls APP metabolism
How FERMT2 modulates APP metabolism is not fully established. To address this, we tested the possibility that APP and FERMT2 form a protein-protein complex. This hypothesis is supported by proximity ligation assay (PLA) signals in hippocampal primary neuronal cultures (Fig. 5a). We then aimed to validate this complex formation via three complementary approaches: (i) Pull-down of endogenous APP from hippocampal primary neuronal culture extracts co-immunoprecipitated endogenous FERMT2 (Fig. 5b). (ii) Over-expression of wild-type FERMT2 (FERMT2 WT ) was also able to pull-down the recombinant intracellular domain of APP (C100) (Fig. 5c and d). (iii) In addition, we generated a Q 621 W 622 AA FERMT2 mutant (FERMT2 QW ) which was previously shown to abolish the interaction between the FERMT2 F3 domain and the NxTY motif of Integrin-β3 (which is also present within the intracellular domain of APP) (Ma et al. 2008). Remarkably, when overexpressed in HEK293 cells, FERMT2 QW was not able to pull-down the recombinant intracellular domain of APP. Cumulatively, these findings support a direct interaction between FERMT2 and APP.
Based on the recently solved crystal structure of FERMT2 in complex with the integrin-β3-tail (H. Li et al. 2017), we built a structural model of the FERMT2/APP complex (Fig. 5e), supporting our hypothesis that a protein-protein interaction exists between FERMT2 and APP.
We next assessed the biological impact of the FERMT2/APP interaction on APP metabolism. By performing extracellular biotinylation experiments, we observed that FERMT2 over-expression in HEK293-APP 695wt cell line decreased the levels of APP at the cell surface, an effect that was abolished by the presence of the QW mutation (Fig. 5f). Further, a dominant negative effect of the FERMT2 QW mutant was observed: its over-expression impacted APP metabolism similarly to FERMT2 silencing, author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint i.e., resulting in increased mature APP at the cell surface and increased Aβ production, as previously reported (Chapuis et al. 2017). Altogether, our data suggest that a FERMT2/APP interaction is necessary for FERMT2 to have an impact on APP metabolism. We decided to further characterize the interaction between FERMT2 and APP at the molecular level and to determine how FERMT2 underexpression may modulate neuronal phenotypes in which APP is known to be involved.

FERMT2/APP interaction is involved in axonal growth
During brain development, APP is enriched in axonal growth cones and acts as a co-receptor for axon guidance and cell migration cues through its interaction with the extracellular matrix (Soldano & Hassan 2014) (Sosa et al. 2013). Interestingly, our pathway analysis highlighted axon guidance as the most enriched pathway (Fig. 1b). For this reason, we decided to investigate the potential involvement of FERMT2/APP interaction in axon growth. Using primary neurons cultured in microfluidic devices that fluidically isolate axons from their cell bodies, we first observed the co-localization of endogenous FERMT2 with APP, as well as with PLA spots in axonal growth cones (Fig. 6a). These data suggest a potential function of the FERMT2/APP complex in axon growth behavior. We then addressed the impact of FERMT2 silencing on axonal growth cone morphology after transduction with lentiviral vectors expressing either shRNA against FERMT2 (shFERMT2) or a non-targeting shRNA (shNT). Actin staining revealed than FERMT2 under-expression led to a significant decrease in growth cone area (9.13±0.71 vs 12.79±1.10 μm²) as well as in the angular dispersion of growth cone filopodia (0.67±0.04 vs 0.84±0.02) (Fig. 6b). Of note, no significant impact on actin retrograde flow rate was observed (0.166±0.003 vs 0.157±0.002 μm/s). These observations suggest a potential impairment of the exploration behavior of the growth cones due to FERMT2 silencing, but not an effect on actin dynamics per se. FERMT2 under-expression was also associated with an accumulation of endogenous APP in the growth cones (1.38±0.11 vs 0.85±0.08, after normalization by the growth cone area).
In order to characterize in-depth the impact of FERMT2 and/or APP expression on axonal growth, we conducted time-lapse microscopy and measured axon growth speed at DIV5 following lentiviral transduction (shNT, shFERMT2, or shAPP) of neurons in microfluidic devices at DIV1. FERMT2 silencing led to 31.7% increase in axon growth speed (Fig. 6c). Conversely, APP under-expression led to 16.7% decrease in axon growth speed. Remarkably, silencing of APP was able to fully abolish the effect of FERMT2 under-expression on axon growth speed, suggesting that APP was required for the molecular mechanism by which FERMT2 controls the axon growth speed. In addition, we observed that FERMT2 QW mutant over-expression was able to induce 15.9% increase in axon growth speed (Fig. 6c). Since over-expression of FERMT2 WT did not show any impact, these data also suggested a potential dominant negative effect of the FERMT2 QW mutant and further supported the involvement of FERMT2/APP complex in axonal growth. author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint

FERMT2 expression regulates synaptic plasticity in APP dependent manner
Next, we decided to investigate the impact of FERMT2 silencing on neuronal maturation at DIV14.
First, co-staining between FERMT2 and synaptic markers (Synaptophysin and Homer) suggested the localization of FERMT2 at the synapse (Fig. 7a). The presence of FERMT2 in both pre-and postynaptic compartments was confirmed by synaptosomal purification (Fig. 7b). PLA-FERMT2/APP signals were also co-localized with Synaptophysin and Homer puncta (Fig. 7c) supporting the possibility of the involvement of FERMT2/APP complex in synapses.
We sought to establish the functional impact of FERMT2 and/or APP silencing on CA1 basal synaptic transmission and long-term potentiation (LTP) in ex vivo mouse hippocampal slices, after stereotactic lentivirus injection allowing expression of shNT, shFERMT2, shAPP or shFERMT2+shAPP. No significant difference was observed for the normalized average slope of the evoked field excitatory postsynaptic potential (fEPSP), indicating no alteration of the CA1 basal synaptic transmission in each group analyzed (Fig. 7d, Supplementary Fig. 5). Then, in the same slice, tetanic stimulation was delivered to the Shaffer collaterals (SC) in order to induce LTP (Fig. 7e). Tetanic stimulation of the SC resulted in a robust, long-lasting potentiation of the fEPSP slope in slices from mice infected with shNT and in shAPP, whereas LTP was impaired in slice form shFERMT2-infected mice. An LTP deficit was observed in hippocampal slices infected with shFERMT2, but not in those also infected with shAPP ( Fig. 7e 7f and 7g). Remarkably, this deficit was abolished when both APP and FERMT2 were silenced, suggesting that APP was required for the molecular mechanism by which FERMT2 impacts LTP.

Discussion
As in other multifactorial diseases, GWAS in AD are agnostic approaches, and how a genetic risk factor is implicated in pathophysiological processes is typically unknown. Sometimes, even the physiological functions of an AD genetic risk factor in the brain are not known. Understanding the role of these genes is thus a challenge that requires several key questions to be addressed: (i) Does the corresponding protein interact with other key players and pathways known to be involved in AD? (ii) What is (are) the functional variant(s) responsible for the GWAS signal and does this (do these) variant(s) impact the biological function of the corresponding protein and its interaction with key players of AD? (iii) How does the protein (taking into account how the functional variant(s) impact the protein) modulate the physiological and/or pathophysiological functions of key players/pathways? With these questions in mind, we developed systematic approaches to determine the genes that are involved in APP metabolism, a major player in AD development. To this end, we previously developed an HCS, based on the quantification of intracellular APP fragments, to measure the impact of under-expression of 18,107 genes (via siRNA pools) on APP metabolism (Chapuis et al. 2017). In the current study, we screened the impact of the over-expression of 2,555 miRNAs on APP metabolism with the hypothesis that genes (i) that modulate APP metabolism (ii) whose expression author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint levels are regulated by miRNAs that also modulate the APP metabolism are likely some of the key actors controlling APP metabolism and functions. The convergence of these two agnostic screens highlighted FERMT2, a GWAS-defined genetic risk factor of AD, for which almost nothing is known in the cerebral and AD contexts. We thus proposed that FERMT2 could be down-regulated by several miRNAs and involved in AD through regulating APP metabolism and/or function (Fig. 8).
Of note, we previously identified the rs7143400 variant located in FERMT2 3'UTR as susceptible to alter a binding site for miR-4504 (Delay et al. 2016). Here, in addition to in silico prediction supporting the impact of this variant on miRNA binding (Supplementary Fig. 3), we demonstrate that this variant is functional: the AD-associated rs7143400 T allele down-regulates FERMT2 and modulates APP metabolism via its interaction with miR-4504. Remarkably, we observed that miR-4504 is over-expressed in the brains of AD cases compared to controls, and is mainly expressed in neurons in primary mixed hippocampal cultures.
Together, our data indicate that a deleterious over-expression of miR-4504 can lead to a decrease in FERMT2 expression in individuals bearing the rs7143400 minor T allele, which subsequently modulates APP metabolism (Fig. 8). Supporting a link between FERMT2 and APP metabolism, studies from cohorts of patients have reported an association between variants in FERMT2 gene and Aβ in CSF (Chapuis et al. 2017) and brain amyloidosis (Apostolova et al. 2018).
Of note, in publicly-available RNA-seq analyses (Mayo Clinic Brain Bank), an over-expression of FERMT2 mRNA is observed in post-mortem human temporal cortex of AD patients relative to controls (Sullivan et al. 2018). Even if we did not observe such a variation of the FERMT2 mRNA level in a smaller number of brains, we nevertheless detected an increase in the protein levels as a function of Braak stages, especially at the latter stages (Supplementary Fig. 6). This point is of particular importance since in the Genotype-Tissue Expression Database (GTEx Consortium et al. 2015), FERMT2 variants associated with an increase in AD risk at the genome-wide significance level are also part of an expression quantitative trait locus, significantly associated with decreased brain expression of FERMT2 mRNA (sentinel variant in GWAS rs17125924; -18% p-value = 2×10 -6 ). This observation strongly supports our conclusion that FERMT2 under-expression is deleterious and suggests that FERMT2 down-regulation would be deleterious at the earliest stage of the disease, whereas, FERMT2 over-expression may occur as a pathological consequence at a very late stage.
Here, we propose that FERMT2 down-regulation at the earliest stage of AD would depend in part on (i) the miR-4504 expression, (ii) cerebral cell type (i.e., neurons), and (iii) the presence of the rs7143400 minor T allele (observed in 9% of Caucasians). Unfortunately, it is important to keep in mind that all these constraints will make difficult, if not impossible, to detect such a miRNAdependent decrease in FERMT2 mRNA levels. Of note, this point may also underline the limitation of expression databases in deciphering the mechanisms underlying the functional effects of GWAS variants, for they do not allow capturing (even hide) subtle mechanisms. author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint Our current hypothesis holds that FERMT2 controls APP internalization and degradation (Chapuis et al. 2017), such that FERMT2 silencing leads to a greater amount of full-length APP and by-products and FERMT2 over-expression leads to a reduction of these metabolites. In the current paper, we demonstrate that a direct interaction between FERMT2 and APP -through the F3 domain of FERMT2 and the NxTY motif within APP's intracellular domainis necessary for FERMT2 to have an impact on APP metabolism. Moreover, we observed than the FERMT2/APP interaction could be involved in the regulation of axonal growth in line with APP's function within the growth cone (Sosa et al. 2013) (data we replicated in this study). Because APP silencing abolishes the impact of FERMT2 silencing on axon growth speed, we could postulate that the effect of FERMT2 might be -at least in part-due to the accumulation of full-length APP or its by-products within the growth cone, impairing vesicle trafficking and/or cell adhesion.
Apart from this developmental read-out, we also analyzed synaptic plasticity, a read-out more relevant to AD, where synaptic dysfunction/loss is one of the earliest events observed. FERMT2 underexpression leads to a harmful effect on LTP in an APP-dependent manner. As FERMT2 silencing leads to an accumulation of full-length APP and all by-products (including Aβ peptides), we can hypothesize that these accumulations could be involved in the synaptic dysfunction observed when FERMT2 is under-expressed. Moreover, it could explain why APP expression seems to be required (for FERMT2 to impact developmental and synaptic read-outs), although further experiments are needed to decipher the consequences of the causal link between FERMT2 and APP, i.e., to determine whether it impacts full-length APP function or invokes Aβ synaptotoxicity. This is of particular interest since, APP shedding strongly enhances its cell adhesion and synaptogenic activity (Stahl et al. 2014). Moreover, the APP intracellular domain is required for normal synaptic morphology and synaptic plasticity, suggesting that intracellular interactors could be required for proper synaptic function (Klevanski et al. 2015). Remarkably, we have recently proposed a circular model of AD pathogenesis, where the core of the focal adhesion pathway -which FERMT2 and APP are part of-may participate in the dysfunction of synaptic plasticity in AD (Dourlen et al. 2019).
In conclusion, we propose that FERMT2 under-expression through miRNAs and/or genetic regulation leads to synaptic dysfunction in an APP-dependent manner. Our hypothesis may thus call for new therapeutic approaches in AD targeting FERMT2 and/or APP function, rather than Aβ peptide production/clearance. Once therapeutic approaches that target FERMT2-dependent pathways are eventually developed, one can argue that it might be an example of personalized medicine for AD, preferentially targeting patients who have high levels of miR-4504 and who bear the rs7143400 minor T allele. author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint showing a downregulation). b. The 10 most likely canonical pathways identified after pathway enrichment analysis using DIANA Tools mirPath (v3.0). c. A graphical representation of the workflow developed to identify miRNA-targeted genes that are also able to modulate APP metabolism according to our genome-wide, high-content siRNA screening. Predictions of target sites for genes expressed in the brain were determined by at least 4 out of 7 software programs used (see methods). d. p-value distribution for the 180 genes targeted by the 41 miRNAs identified by HCS (see Fig. S2). e. Genes significantly enriched to be targeted by these 41 miRNAs after Bonferroni correction (0.05/180=2.7×10 -4 ). Bold text indicates the only AD genetic risk factor in the list and its predicted miRNAs.
author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint   Expression of miR-4504 is increased in AD brains. Relative miR-222 and miR-4504 expression levels in temporal lobes of non-demented (CTL) and AD groups. Mann-Whitney test; **** p < 0.0001. author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/767194 doi: bioRxiv preprint